202
333
commit to be valid, deletes against the basis MUST be recorded via
203
334
builder.record_delete().
205
raise NotImplementedError(self.will_record_deletes)
207
def record_iter_changes(self, tree, basis_revision_id, iter_changes):
336
self._recording_deletes = True
338
basis_id = self.parents[0]
340
basis_id = _mod_revision.NULL_REVISION
341
self.basis_delta_revision = basis_id
343
def record_entry_contents(self, ie, parent_invs, path, tree,
345
"""Record the content of ie from tree into the commit if needed.
347
Side effect: sets ie.revision when unchanged
349
:param ie: An inventory entry present in the commit.
350
:param parent_invs: The inventories of the parent revisions of the
352
:param path: The path the entry is at in the tree.
353
:param tree: The tree which contains this entry and should be used to
355
:param content_summary: Summary data from the tree about the paths
356
content - stat, length, exec, sha/link target. This is only
357
accessed when the entry has a revision of None - that is when it is
358
a candidate to commit.
359
:return: A tuple (change_delta, version_recorded, fs_hash).
360
change_delta is an inventory_delta change for this entry against
361
the basis tree of the commit, or None if no change occured against
363
version_recorded is True if a new version of the entry has been
364
recorded. For instance, committing a merge where a file was only
365
changed on the other side will return (delta, False).
366
fs_hash is either None, or the hash details for the path (currently
367
a tuple of the contents sha1 and the statvalue returned by
368
tree.get_file_with_stat()).
370
if self.new_inventory.root is None:
371
if ie.parent_id is not None:
372
raise errors.RootMissing()
373
self._check_root(ie, parent_invs, tree)
374
if ie.revision is None:
375
kind = content_summary[0]
377
# ie is carried over from a prior commit
379
# XXX: repository specific check for nested tree support goes here - if
380
# the repo doesn't want nested trees we skip it ?
381
if (kind == 'tree-reference' and
382
not self.repository._format.supports_tree_reference):
383
# mismatch between commit builder logic and repository:
384
# this needs the entry creation pushed down into the builder.
385
raise NotImplementedError('Missing repository subtree support.')
386
self.new_inventory.add(ie)
388
# TODO: slow, take it out of the inner loop.
390
basis_inv = parent_invs[0]
392
basis_inv = Inventory(root_id=None)
394
# ie.revision is always None if the InventoryEntry is considered
395
# for committing. We may record the previous parents revision if the
396
# content is actually unchanged against a sole head.
397
if ie.revision is not None:
398
if not self._versioned_root and path == '':
399
# repositories that do not version the root set the root's
400
# revision to the new commit even when no change occurs (more
401
# specifically, they do not record a revision on the root; and
402
# the rev id is assigned to the root during deserialisation -
403
# this masks when a change may have occurred against the basis.
404
# To match this we always issue a delta, because the revision
405
# of the root will always be changing.
406
if ie.file_id in basis_inv:
407
delta = (basis_inv.id2path(ie.file_id), path,
411
delta = (None, path, ie.file_id, ie)
412
self._basis_delta.append(delta)
413
return delta, False, None
415
# we don't need to commit this, because the caller already
416
# determined that an existing revision of this file is
417
# appropriate. If its not being considered for committing then
418
# it and all its parents to the root must be unaltered so
419
# no-change against the basis.
420
if ie.revision == self._new_revision_id:
421
raise AssertionError("Impossible situation, a skipped "
422
"inventory entry (%r) claims to be modified in this "
423
"commit (%r).", (ie, self._new_revision_id))
424
return None, False, None
425
# XXX: Friction: parent_candidates should return a list not a dict
426
# so that we don't have to walk the inventories again.
427
parent_candiate_entries = ie.parent_candidates(parent_invs)
428
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
430
for inv in parent_invs:
431
if ie.file_id in inv:
432
old_rev = inv[ie.file_id].revision
433
if old_rev in head_set:
434
heads.append(inv[ie.file_id].revision)
435
head_set.remove(inv[ie.file_id].revision)
438
# now we check to see if we need to write a new record to the
440
# We write a new entry unless there is one head to the ancestors, and
441
# the kind-derived content is unchanged.
443
# Cheapest check first: no ancestors, or more the one head in the
444
# ancestors, we write a new node.
448
# There is a single head, look it up for comparison
449
parent_entry = parent_candiate_entries[heads[0]]
450
# if the non-content specific data has changed, we'll be writing a
452
if (parent_entry.parent_id != ie.parent_id or
453
parent_entry.name != ie.name):
455
# now we need to do content specific checks:
457
# if the kind changed the content obviously has
458
if kind != parent_entry.kind:
460
# Stat cache fingerprint feedback for the caller - None as we usually
461
# don't generate one.
464
if content_summary[2] is None:
465
raise ValueError("Files must not have executable = None")
467
if (# if the file length changed we have to store:
468
parent_entry.text_size != content_summary[1] or
469
# if the exec bit has changed we have to store:
470
parent_entry.executable != content_summary[2]):
472
elif parent_entry.text_sha1 == content_summary[3]:
473
# all meta and content is unchanged (using a hash cache
474
# hit to check the sha)
475
ie.revision = parent_entry.revision
476
ie.text_size = parent_entry.text_size
477
ie.text_sha1 = parent_entry.text_sha1
478
ie.executable = parent_entry.executable
479
return self._get_delta(ie, basis_inv, path), False, None
481
# Either there is only a hash change(no hash cache entry,
482
# or same size content change), or there is no change on
484
# Provide the parent's hash to the store layer, so that the
485
# content is unchanged we will not store a new node.
486
nostore_sha = parent_entry.text_sha1
488
# We want to record a new node regardless of the presence or
489
# absence of a content change in the file.
491
ie.executable = content_summary[2]
492
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
494
text = file_obj.read()
498
ie.text_sha1, ie.text_size = self._add_text_to_weave(
499
ie.file_id, text, heads, nostore_sha)
500
# Let the caller know we generated a stat fingerprint.
501
fingerprint = (ie.text_sha1, stat_value)
502
except errors.ExistingContent:
503
# Turns out that the file content was unchanged, and we were
504
# only going to store a new node if it was changed. Carry over
506
ie.revision = parent_entry.revision
507
ie.text_size = parent_entry.text_size
508
ie.text_sha1 = parent_entry.text_sha1
509
ie.executable = parent_entry.executable
510
return self._get_delta(ie, basis_inv, path), False, None
511
elif kind == 'directory':
513
# all data is meta here, nothing specific to directory, so
515
ie.revision = parent_entry.revision
516
return self._get_delta(ie, basis_inv, path), False, None
517
self._add_text_to_weave(ie.file_id, '', heads, None)
518
elif kind == 'symlink':
519
current_link_target = content_summary[3]
521
# symlink target is not generic metadata, check if it has
523
if current_link_target != parent_entry.symlink_target:
526
# unchanged, carry over.
527
ie.revision = parent_entry.revision
528
ie.symlink_target = parent_entry.symlink_target
529
return self._get_delta(ie, basis_inv, path), False, None
530
ie.symlink_target = current_link_target
531
self._add_text_to_weave(ie.file_id, '', heads, None)
532
elif kind == 'tree-reference':
534
if content_summary[3] != parent_entry.reference_revision:
537
# unchanged, carry over.
538
ie.reference_revision = parent_entry.reference_revision
539
ie.revision = parent_entry.revision
540
return self._get_delta(ie, basis_inv, path), False, None
541
ie.reference_revision = content_summary[3]
542
self._add_text_to_weave(ie.file_id, '', heads, None)
544
raise NotImplementedError('unknown kind')
545
ie.revision = self._new_revision_id
546
self._any_changes = True
547
return self._get_delta(ie, basis_inv, path), True, fingerprint
549
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
550
_entry_factory=entry_factory):
208
551
"""Record a new tree via iter_changes.
210
553
:param tree: The tree to obtain text contents from for changed objects.
215
558
to basis_revision_id. The iterator must not include any items with
216
559
a current kind of None - missing items must be either filtered out
217
560
or errored-on beefore record_iter_changes sees the item.
561
:param _entry_factory: Private method to bind entry_factory locally for
218
563
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
219
564
tree._observed_sha1.
221
raise NotImplementedError(self.record_iter_changes)
224
class RepositoryWriteLockResult(LogicalLockResult):
225
"""The result of write locking a repository.
227
:ivar repository_token: The token obtained from the underlying lock, or
229
:ivar unlock: A callable which will unlock the lock.
232
def __init__(self, unlock, repository_token):
233
LogicalLockResult.__init__(self, unlock)
234
self.repository_token = repository_token
237
return "RepositoryWriteLockResult(%s, %s)" % (self.repository_token,
566
# Create an inventory delta based on deltas between all the parents and
567
# deltas between all the parent inventories. We use inventory delta's
568
# between the inventory objects because iter_changes masks
569
# last-changed-field only changes.
571
# file_id -> change map, change is fileid, paths, changed, versioneds,
572
# parents, names, kinds, executables
574
# {file_id -> revision_id -> inventory entry, for entries in parent
575
# trees that are not parents[0]
579
revtrees = list(self.repository.revision_trees(self.parents))
580
except errors.NoSuchRevision:
581
# one or more ghosts, slow path.
583
for revision_id in self.parents:
585
revtrees.append(self.repository.revision_tree(revision_id))
586
except errors.NoSuchRevision:
588
basis_revision_id = _mod_revision.NULL_REVISION
590
revtrees.append(self.repository.revision_tree(
591
_mod_revision.NULL_REVISION))
592
# The basis inventory from a repository
594
basis_inv = revtrees[0].inventory
596
basis_inv = self.repository.revision_tree(
597
_mod_revision.NULL_REVISION).inventory
598
if len(self.parents) > 0:
599
if basis_revision_id != self.parents[0] and not ghost_basis:
601
"arbitrary basis parents not yet supported with merges")
602
for revtree in revtrees[1:]:
603
for change in revtree.inventory._make_delta(basis_inv):
604
if change[1] is None:
605
# Not present in this parent.
607
if change[2] not in merged_ids:
608
if change[0] is not None:
609
basis_entry = basis_inv[change[2]]
610
merged_ids[change[2]] = [
612
basis_entry.revision,
615
parent_entries[change[2]] = {
617
basis_entry.revision:basis_entry,
619
change[3].revision:change[3],
622
merged_ids[change[2]] = [change[3].revision]
623
parent_entries[change[2]] = {change[3].revision:change[3]}
625
merged_ids[change[2]].append(change[3].revision)
626
parent_entries[change[2]][change[3].revision] = change[3]
629
# Setup the changes from the tree:
630
# changes maps file_id -> (change, [parent revision_ids])
632
for change in iter_changes:
633
# This probably looks up in basis_inv way to much.
634
if change[1][0] is not None:
635
head_candidate = [basis_inv[change[0]].revision]
638
changes[change[0]] = change, merged_ids.get(change[0],
640
unchanged_merged = set(merged_ids) - set(changes)
641
# Extend the changes dict with synthetic changes to record merges of
643
for file_id in unchanged_merged:
644
# Record a merged version of these items that did not change vs the
645
# basis. This can be either identical parallel changes, or a revert
646
# of a specific file after a merge. The recorded content will be
647
# that of the current tree (which is the same as the basis), but
648
# the per-file graph will reflect a merge.
649
# NB:XXX: We are reconstructing path information we had, this
650
# should be preserved instead.
651
# inv delta change: (file_id, (path_in_source, path_in_target),
652
# changed_content, versioned, parent, name, kind,
655
basis_entry = basis_inv[file_id]
656
except errors.NoSuchId:
657
# a change from basis->some_parents but file_id isn't in basis
658
# so was new in the merge, which means it must have changed
659
# from basis -> current, and as it hasn't the add was reverted
660
# by the user. So we discard this change.
664
(basis_inv.id2path(file_id), tree.id2path(file_id)),
666
(basis_entry.parent_id, basis_entry.parent_id),
667
(basis_entry.name, basis_entry.name),
668
(basis_entry.kind, basis_entry.kind),
669
(basis_entry.executable, basis_entry.executable))
670
changes[file_id] = (change, merged_ids[file_id])
671
# changes contains tuples with the change and a set of inventory
672
# candidates for the file.
674
# old_path, new_path, file_id, new_inventory_entry
675
seen_root = False # Is the root in the basis delta?
676
inv_delta = self._basis_delta
677
modified_rev = self._new_revision_id
678
for change, head_candidates in changes.values():
679
if change[3][1]: # versioned in target.
680
# Several things may be happening here:
681
# We may have a fork in the per-file graph
682
# - record a change with the content from tree
683
# We may have a change against < all trees
684
# - carry over the tree that hasn't changed
685
# We may have a change against all trees
686
# - record the change with the content from tree
689
entry = _entry_factory[kind](file_id, change[5][1],
691
head_set = self._heads(change[0], set(head_candidates))
694
for head_candidate in head_candidates:
695
if head_candidate in head_set:
696
heads.append(head_candidate)
697
head_set.remove(head_candidate)
700
# Could be a carry-over situation:
701
parent_entry_revs = parent_entries.get(file_id, None)
702
if parent_entry_revs:
703
parent_entry = parent_entry_revs.get(heads[0], None)
706
if parent_entry is None:
707
# The parent iter_changes was called against is the one
708
# that is the per-file head, so any change is relevant
709
# iter_changes is valid.
710
carry_over_possible = False
712
# could be a carry over situation
713
# A change against the basis may just indicate a merge,
714
# we need to check the content against the source of the
715
# merge to determine if it was changed after the merge
717
if (parent_entry.kind != entry.kind or
718
parent_entry.parent_id != entry.parent_id or
719
parent_entry.name != entry.name):
720
# Metadata common to all entries has changed
721
# against per-file parent
722
carry_over_possible = False
724
carry_over_possible = True
725
# per-type checks for changes against the parent_entry
728
# Cannot be a carry-over situation
729
carry_over_possible = False
730
# Populate the entry in the delta
732
# XXX: There is still a small race here: If someone reverts the content of a file
733
# after iter_changes examines and decides it has changed,
734
# we will unconditionally record a new version even if some
735
# other process reverts it while commit is running (with
736
# the revert happening after iter_changes did it's
739
entry.executable = True
741
entry.executable = False
742
if (carry_over_possible and
743
parent_entry.executable == entry.executable):
744
# Check the file length, content hash after reading
746
nostore_sha = parent_entry.text_sha1
749
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
751
text = file_obj.read()
755
entry.text_sha1, entry.text_size = self._add_text_to_weave(
756
file_id, text, heads, nostore_sha)
757
yield file_id, change[1][1], (entry.text_sha1, stat_value)
758
except errors.ExistingContent:
759
# No content change against a carry_over parent
760
# Perhaps this should also yield a fs hash update?
762
entry.text_size = parent_entry.text_size
763
entry.text_sha1 = parent_entry.text_sha1
764
elif kind == 'symlink':
766
entry.symlink_target = tree.get_symlink_target(file_id)
767
if (carry_over_possible and
768
parent_entry.symlink_target == entry.symlink_target):
771
self._add_text_to_weave(change[0], '', heads, None)
772
elif kind == 'directory':
773
if carry_over_possible:
776
# Nothing to set on the entry.
777
# XXX: split into the Root and nonRoot versions.
778
if change[1][1] != '' or self.repository.supports_rich_root():
779
self._add_text_to_weave(change[0], '', heads, None)
780
elif kind == 'tree-reference':
781
if not self.repository._format.supports_tree_reference:
782
# This isn't quite sane as an error, but we shouldn't
783
# ever see this code path in practice: tree's don't
784
# permit references when the repo doesn't support tree
786
raise errors.UnsupportedOperation(tree.add_reference,
788
reference_revision = tree.get_reference_revision(change[0])
789
entry.reference_revision = reference_revision
790
if (carry_over_possible and
791
parent_entry.reference_revision == reference_revision):
794
self._add_text_to_weave(change[0], '', heads, None)
796
raise AssertionError('unknown kind %r' % kind)
798
entry.revision = modified_rev
800
entry.revision = parent_entry.revision
803
new_path = change[1][1]
804
inv_delta.append((change[1][0], new_path, change[0], entry))
807
self.new_inventory = None
809
self._any_changes = True
811
# housekeeping root entry changes do not affect no-change commits.
812
self._require_root_change(tree)
813
self.basis_delta_revision = basis_revision_id
815
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
816
parent_keys = tuple([(file_id, parent) for parent in parents])
817
return self.repository.texts._add_text(
818
(file_id, self._new_revision_id), parent_keys, new_text,
819
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
822
class RootCommitBuilder(CommitBuilder):
823
"""This commitbuilder actually records the root id"""
825
# the root entry gets versioned properly by this builder.
826
_versioned_root = True
828
def _check_root(self, ie, parent_invs, tree):
829
"""Helper for record_entry_contents.
831
:param ie: An entry being added.
832
:param parent_invs: The inventories of the parent revisions of the
834
:param tree: The tree that is being committed.
837
def _require_root_change(self, tree):
838
"""Enforce an appropriate root object change.
840
This is called once when record_iter_changes is called, if and only if
841
the root was not in the delta calculated by record_iter_changes.
843
:param tree: The tree which is being committed.
845
# versioned roots do not change unless the tree found a change.
241
848
######################################################################
245
class Repository(_RelockDebugMixin, controldir.ControlComponent):
851
class Repository(object):
246
852
"""Repository holding history for one or more branches.
248
854
The repository holds and retrieves historical information including
249
855
revisions and file history. It's normally accessed only by the Branch,
250
856
which views a particular line of development through that history.
252
See VersionedFileRepository in bzrlib.vf_repository for the
253
base class for most Bazaar repositories.
858
The Repository builds on top of some byte storage facilies (the revisions,
859
signatures, inventories, texts and chk_bytes attributes) and a Transport,
860
which respectively provide byte storage and a means to access the (possibly
863
The byte storage facilities are addressed via tuples, which we refer to
864
as 'keys' throughout the code base. Revision_keys, inventory_keys and
865
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
866
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
867
byte string made up of a hash identifier and a hash value.
868
We use this interface because it allows low friction with the underlying
869
code that implements disk indices, network encoding and other parts of
872
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
873
the serialised revisions for the repository. This can be used to obtain
874
revision graph information or to access raw serialised revisions.
875
The result of trying to insert data into the repository via this store
876
is undefined: it should be considered read-only except for implementors
878
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
879
the serialised signatures for the repository. This can be used to
880
obtain access to raw serialised signatures. The result of trying to
881
insert data into the repository via this store is undefined: it should
882
be considered read-only except for implementors of repositories.
883
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
884
the serialised inventories for the repository. This can be used to
885
obtain unserialised inventories. The result of trying to insert data
886
into the repository via this store is undefined: it should be
887
considered read-only except for implementors of repositories.
888
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
889
texts of files and directories for the repository. This can be used to
890
obtain file texts or file graphs. Note that Repository.iter_file_bytes
891
is usually a better interface for accessing file texts.
892
The result of trying to insert data into the repository via this store
893
is undefined: it should be considered read-only except for implementors
895
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
896
any data the repository chooses to store or have indexed by its hash.
897
The result of trying to insert data into the repository via this store
898
is undefined: it should be considered read-only except for implementors
900
:ivar _transport: Transport for file access to repository, typically
901
pointing to .bzr/repository.
904
# What class to use for a CommitBuilder. Often its simpler to change this
905
# in a Repository class subclass rather than to override
906
# get_commit_builder.
907
_commit_builder_class = CommitBuilder
908
# The search regex used by xml based repositories to determine what things
909
# where changed in a single commit.
910
_file_ids_altered_regex = lazy_regex.lazy_compile(
911
r'file_id="(?P<file_id>[^"]+)"'
912
r'.* revision="(?P<revision_id>[^"]+)"'
256
915
def abort_write_group(self, suppress_errors=False):
257
916
"""Commit the contents accrued within the current write group.
313
980
return InterRepository._assert_same_model(self, repository)
982
def add_inventory(self, revision_id, inv, parents):
983
"""Add the inventory inv to the repository as revision_id.
985
:param parents: The revision ids of the parents that revision_id
986
is known to have and are in the repository already.
988
:returns: The validator(which is a sha1 digest, though what is sha'd is
989
repository format specific) of the serialized inventory.
991
if not self.is_in_write_group():
992
raise AssertionError("%r not in write group" % (self,))
993
_mod_revision.check_not_reserved_id(revision_id)
994
if not (inv.revision_id is None or inv.revision_id == revision_id):
995
raise AssertionError(
996
"Mismatch between inventory revision"
997
" id and insertion revid (%r, %r)"
998
% (inv.revision_id, revision_id))
1000
raise AssertionError()
1001
return self._add_inventory_checked(revision_id, inv, parents)
1003
def _add_inventory_checked(self, revision_id, inv, parents):
1004
"""Add inv to the repository after checking the inputs.
1006
This function can be overridden to allow different inventory styles.
1008
:seealso: add_inventory, for the contract.
1010
inv_lines = self._serialise_inventory_to_lines(inv)
1011
return self._inventory_add_lines(revision_id, parents,
1012
inv_lines, check_content=False)
1014
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1015
parents, basis_inv=None, propagate_caches=False):
1016
"""Add a new inventory expressed as a delta against another revision.
1018
See the inventory developers documentation for the theory behind
1021
:param basis_revision_id: The inventory id the delta was created
1022
against. (This does not have to be a direct parent.)
1023
:param delta: The inventory delta (see Inventory.apply_delta for
1025
:param new_revision_id: The revision id that the inventory is being
1027
:param parents: The revision ids of the parents that revision_id is
1028
known to have and are in the repository already. These are supplied
1029
for repositories that depend on the inventory graph for revision
1030
graph access, as well as for those that pun ancestry with delta
1032
:param basis_inv: The basis inventory if it is already known,
1034
:param propagate_caches: If True, the caches for this inventory are
1035
copied to and updated for the result if possible.
1037
:returns: (validator, new_inv)
1038
The validator(which is a sha1 digest, though what is sha'd is
1039
repository format specific) of the serialized inventory, and the
1040
resulting inventory.
1042
if not self.is_in_write_group():
1043
raise AssertionError("%r not in write group" % (self,))
1044
_mod_revision.check_not_reserved_id(new_revision_id)
1045
basis_tree = self.revision_tree(basis_revision_id)
1046
basis_tree.lock_read()
1048
# Note that this mutates the inventory of basis_tree, which not all
1049
# inventory implementations may support: A better idiom would be to
1050
# return a new inventory, but as there is no revision tree cache in
1051
# repository this is safe for now - RBC 20081013
1052
if basis_inv is None:
1053
basis_inv = basis_tree.inventory
1054
basis_inv.apply_delta(delta)
1055
basis_inv.revision_id = new_revision_id
1056
return (self.add_inventory(new_revision_id, basis_inv, parents),
1061
def _inventory_add_lines(self, revision_id, parents, lines,
1062
check_content=True):
1063
"""Store lines in inv_vf and return the sha1 of the inventory."""
1064
parents = [(parent,) for parent in parents]
1065
return self.inventories.add_lines((revision_id,), parents, lines,
1066
check_content=check_content)[0]
1068
def add_revision(self, revision_id, rev, inv=None, config=None):
1069
"""Add rev to the revision store as revision_id.
1071
:param revision_id: the revision id to use.
1072
:param rev: The revision object.
1073
:param inv: The inventory for the revision. if None, it will be looked
1074
up in the inventory storer
1075
:param config: If None no digital signature will be created.
1076
If supplied its signature_needed method will be used
1077
to determine if a signature should be made.
1079
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1081
_mod_revision.check_not_reserved_id(revision_id)
1082
if config is not None and config.signature_needed():
1084
inv = self.get_inventory(revision_id)
1085
plaintext = Testament(rev, inv).as_short_text()
1086
self.store_revision_signature(
1087
gpg.GPGStrategy(config), plaintext, revision_id)
1088
# check inventory present
1089
if not self.inventories.get_parent_map([(revision_id,)]):
1091
raise errors.WeaveRevisionNotPresent(revision_id,
1094
# yes, this is not suitable for adding with ghosts.
1095
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1098
key = (revision_id,)
1099
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1100
self._add_revision(rev)
1102
def _add_revision(self, revision):
1103
text = self._serializer.write_revision_to_string(revision)
1104
key = (revision.revision_id,)
1105
parents = tuple((parent,) for parent in revision.parent_ids)
1106
self.revisions.add_lines(key, parents, osutils.split_lines(text))
315
1108
def all_revision_ids(self):
316
1109
"""Returns a list of all the revision ids in the repository.
909
1788
signature = gpg_strategy.sign(plaintext)
910
1789
self.add_signature_text(revision_id, signature)
912
1792
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)
1793
self.signatures.add_lines((revision_id,), (),
1794
osutils.split_lines(signature))
1796
def find_text_key_references(self):
1797
"""Find the text key references within the repository.
1799
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1800
to whether they were referred to by the inventory of the
1801
revision_id that they contain. The inventory texts from all present
1802
revision ids are assessed to generate this report.
1804
revision_keys = self.revisions.keys()
1805
w = self.inventories
1806
pb = ui.ui_factory.nested_progress_bar()
1808
return self._find_text_key_references_from_xml_inventory_lines(
1809
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1813
def _find_text_key_references_from_xml_inventory_lines(self,
1815
"""Core routine for extracting references to texts from inventories.
1817
This performs the translation of xml lines to revision ids.
1819
:param line_iterator: An iterator of lines, origin_version_id
1820
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1821
to whether they were referred to by the inventory of the
1822
revision_id that they contain. Note that if that revision_id was
1823
not part of the line_iterator's output then False will be given -
1824
even though it may actually refer to that key.
1826
if not self._serializer.support_altered_by_hack:
1827
raise AssertionError(
1828
"_find_text_key_references_from_xml_inventory_lines only "
1829
"supported for branches which store inventory as unnested xml"
1830
", not on %r" % self)
1833
# this code needs to read every new line in every inventory for the
1834
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1835
# not present in one of those inventories is unnecessary but not
1836
# harmful because we are filtering by the revision id marker in the
1837
# inventory lines : we only select file ids altered in one of those
1838
# revisions. We don't need to see all lines in the inventory because
1839
# only those added in an inventory in rev X can contain a revision=X
1841
unescape_revid_cache = {}
1842
unescape_fileid_cache = {}
1844
# jam 20061218 In a big fetch, this handles hundreds of thousands
1845
# of lines, so it has had a lot of inlining and optimizing done.
1846
# Sorry that it is a little bit messy.
1847
# Move several functions to be local variables, since this is a long
1849
search = self._file_ids_altered_regex.search
1850
unescape = _unescape_xml
1851
setdefault = result.setdefault
1852
for line, line_key in line_iterator:
1853
match = search(line)
1856
# One call to match.group() returning multiple items is quite a
1857
# bit faster than 2 calls to match.group() each returning 1
1858
file_id, revision_id = match.group('file_id', 'revision_id')
1860
# Inlining the cache lookups helps a lot when you make 170,000
1861
# lines and 350k ids, versus 8.4 unique ids.
1862
# Using a cache helps in 2 ways:
1863
# 1) Avoids unnecessary decoding calls
1864
# 2) Re-uses cached strings, which helps in future set and
1866
# (2) is enough that removing encoding entirely along with
1867
# the cache (so we are using plain strings) results in no
1868
# performance improvement.
1870
revision_id = unescape_revid_cache[revision_id]
1872
unescaped = unescape(revision_id)
1873
unescape_revid_cache[revision_id] = unescaped
1874
revision_id = unescaped
1876
# Note that unconditionally unescaping means that we deserialise
1877
# every fileid, which for general 'pull' is not great, but we don't
1878
# really want to have some many fulltexts that this matters anyway.
1881
file_id = unescape_fileid_cache[file_id]
1883
unescaped = unescape(file_id)
1884
unescape_fileid_cache[file_id] = unescaped
1887
key = (file_id, revision_id)
1888
setdefault(key, False)
1889
if revision_id == line_key[-1]:
1893
def _inventory_xml_lines_for_keys(self, keys):
1894
"""Get a line iterator of the sort needed for findind references.
1896
Not relevant for non-xml inventory repositories.
1898
Ghosts in revision_keys are ignored.
1900
:param revision_keys: The revision keys for the inventories to inspect.
1901
:return: An iterator over (inventory line, revid) for the fulltexts of
1902
all of the xml inventories specified by revision_keys.
1904
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1905
for record in stream:
1906
if record.storage_kind != 'absent':
1907
chunks = record.get_bytes_as('chunked')
1908
revid = record.key[-1]
1909
lines = osutils.chunks_to_lines(chunks)
1913
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1915
"""Helper routine for fileids_altered_by_revision_ids.
1917
This performs the translation of xml lines to revision ids.
1919
:param line_iterator: An iterator of lines, origin_version_id
1920
:param revision_keys: The revision ids to filter for. This should be a
1921
set or other type which supports efficient __contains__ lookups, as
1922
the revision key from each parsed line will be looked up in the
1923
revision_keys filter.
1924
:return: a dictionary mapping altered file-ids to an iterable of
1925
revision_ids. Each altered file-ids has the exact revision_ids that
1926
altered it listed explicitly.
1928
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1929
line_iterator).iterkeys())
1930
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
1931
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1932
self._inventory_xml_lines_for_keys(parent_keys)))
1933
new_keys = seen - parent_seen
1935
setdefault = result.setdefault
1936
for key in new_keys:
1937
setdefault(key[0], set()).add(key[-1])
920
1940
def _find_parent_ids_of_revisions(self, revision_ids):
921
1941
"""Find all parent ids that are mentioned in the revision graph.
951
1996
uniquely identify the file version in the caller's context. (Examples:
952
1997
an index number or a TreeTransform trans_id.)
1999
bytes_iterator is an iterable of bytestrings for the file. The
2000
kind of iterable and length of the bytestrings are unspecified, but for
2001
this implementation, it is a list of bytes produced by
2002
VersionedFile.get_record_stream().
954
2004
:param desired_files: a list of (file_id, revision_id, identifier)
957
raise NotImplementedError(self.iter_files_bytes)
2008
for file_id, revision_id, callable_data in desired_files:
2009
text_keys[(file_id, revision_id)] = callable_data
2010
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2011
if record.storage_kind == 'absent':
2012
raise errors.RevisionNotPresent(record.key, self)
2013
yield text_keys[record.key], record.get_bytes_as('chunked')
2015
def _generate_text_key_index(self, text_key_references=None,
2017
"""Generate a new text key index for the repository.
2019
This is an expensive function that will take considerable time to run.
2021
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2022
list of parents, also text keys. When a given key has no parents,
2023
the parents list will be [NULL_REVISION].
2025
# All revisions, to find inventory parents.
2026
if ancestors is None:
2027
graph = self.get_graph()
2028
ancestors = graph.get_parent_map(self.all_revision_ids())
2029
if text_key_references is None:
2030
text_key_references = self.find_text_key_references()
2031
pb = ui.ui_factory.nested_progress_bar()
2033
return self._do_generate_text_key_index(ancestors,
2034
text_key_references, pb)
2038
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2039
"""Helper for _generate_text_key_index to avoid deep nesting."""
2040
revision_order = tsort.topo_sort(ancestors)
2041
invalid_keys = set()
2043
for revision_id in revision_order:
2044
revision_keys[revision_id] = set()
2045
text_count = len(text_key_references)
2046
# a cache of the text keys to allow reuse; costs a dict of all the
2047
# keys, but saves a 2-tuple for every child of a given key.
2049
for text_key, valid in text_key_references.iteritems():
2051
invalid_keys.add(text_key)
2053
revision_keys[text_key[1]].add(text_key)
2054
text_key_cache[text_key] = text_key
2055
del text_key_references
2057
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2058
NULL_REVISION = _mod_revision.NULL_REVISION
2059
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2060
# too small for large or very branchy trees. However, for 55K path
2061
# trees, it would be easy to use too much memory trivially. Ideally we
2062
# could gauge this by looking at available real memory etc, but this is
2063
# always a tricky proposition.
2064
inventory_cache = lru_cache.LRUCache(10)
2065
batch_size = 10 # should be ~150MB on a 55K path tree
2066
batch_count = len(revision_order) / batch_size + 1
2068
pb.update("Calculating text parents", processed_texts, text_count)
2069
for offset in xrange(batch_count):
2070
to_query = revision_order[offset * batch_size:(offset + 1) *
2074
for rev_tree in self.revision_trees(to_query):
2075
revision_id = rev_tree.get_revision_id()
2076
parent_ids = ancestors[revision_id]
2077
for text_key in revision_keys[revision_id]:
2078
pb.update("Calculating text parents", processed_texts)
2079
processed_texts += 1
2080
candidate_parents = []
2081
for parent_id in parent_ids:
2082
parent_text_key = (text_key[0], parent_id)
2084
check_parent = parent_text_key not in \
2085
revision_keys[parent_id]
2087
# the parent parent_id is a ghost:
2088
check_parent = False
2089
# truncate the derived graph against this ghost.
2090
parent_text_key = None
2092
# look at the parent commit details inventories to
2093
# determine possible candidates in the per file graph.
2096
inv = inventory_cache[parent_id]
2098
inv = self.revision_tree(parent_id).inventory
2099
inventory_cache[parent_id] = inv
2101
parent_entry = inv[text_key[0]]
2102
except (KeyError, errors.NoSuchId):
2104
if parent_entry is not None:
2106
text_key[0], parent_entry.revision)
2108
parent_text_key = None
2109
if parent_text_key is not None:
2110
candidate_parents.append(
2111
text_key_cache[parent_text_key])
2112
parent_heads = text_graph.heads(candidate_parents)
2113
new_parents = list(parent_heads)
2114
new_parents.sort(key=lambda x:candidate_parents.index(x))
2115
if new_parents == []:
2116
new_parents = [NULL_REVISION]
2117
text_index[text_key] = new_parents
2119
for text_key in invalid_keys:
2120
text_index[text_key] = [NULL_REVISION]
2123
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2124
"""Get an iterable listing the keys of all the data introduced by a set
2127
The keys will be ordered so that the corresponding items can be safely
2128
fetched and inserted in that order.
2130
:returns: An iterable producing tuples of (knit-kind, file-id,
2131
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2132
'revisions'. file-id is None unless knit-kind is 'file'.
2134
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2137
for result in self._find_non_file_keys_to_fetch(revision_ids):
2140
def _find_file_keys_to_fetch(self, revision_ids, pb):
2141
# XXX: it's a bit weird to control the inventory weave caching in this
2142
# generator. Ideally the caching would be done in fetch.py I think. Or
2143
# maybe this generator should explicitly have the contract that it
2144
# should not be iterated until the previously yielded item has been
2146
inv_w = self.inventories
2148
# file ids that changed
2149
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2151
num_file_ids = len(file_ids)
2152
for file_id, altered_versions in file_ids.iteritems():
2154
pb.update("fetch texts", count, num_file_ids)
2156
yield ("file", file_id, altered_versions)
2158
def _find_non_file_keys_to_fetch(self, revision_ids):
2160
yield ("inventory", None, revision_ids)
2163
# XXX: Note ATM no callers actually pay attention to this return
2164
# instead they just use the list of revision ids and ignore
2165
# missing sigs. Consider removing this work entirely
2166
revisions_with_signatures = set(self.signatures.get_parent_map(
2167
[(r,) for r in revision_ids]))
2168
revisions_with_signatures = set(
2169
[r for (r,) in revisions_with_signatures])
2170
revisions_with_signatures.intersection_update(revision_ids)
2171
yield ("signatures", None, revisions_with_signatures)
2174
yield ("revisions", None, revision_ids)
2177
def get_inventory(self, revision_id):
2178
"""Get Inventory object by revision id."""
2179
return self.iter_inventories([revision_id]).next()
2181
def iter_inventories(self, revision_ids):
2182
"""Get many inventories by revision_ids.
2184
This will buffer some or all of the texts used in constructing the
2185
inventories in memory, but will only parse a single inventory at a
2188
:param revision_ids: The expected revision ids of the inventories.
2189
:return: An iterator of inventories.
2191
if ((None in revision_ids)
2192
or (_mod_revision.NULL_REVISION in revision_ids)):
2193
raise ValueError('cannot get null revision inventory')
2194
return self._iter_inventories(revision_ids)
2196
def _iter_inventories(self, revision_ids):
2197
"""single-document based inventory iteration."""
2198
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2199
yield self.deserialise_inventory(revision_id, text)
2201
def _iter_inventory_xmls(self, revision_ids):
2202
keys = [(revision_id,) for revision_id in revision_ids]
2203
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2205
for record in stream:
2206
if record.storage_kind != 'absent':
2207
text_chunks[record.key] = record.get_bytes_as('chunked')
2209
raise errors.NoSuchRevision(self, record.key)
2211
chunks = text_chunks.pop(key)
2212
yield ''.join(chunks), key[-1]
2214
def deserialise_inventory(self, revision_id, xml):
2215
"""Transform the xml into an inventory object.
2217
:param revision_id: The expected revision id of the inventory.
2218
:param xml: A serialised inventory.
2220
result = self._serializer.read_inventory_from_string(xml, revision_id,
2221
entry_cache=self._inventory_entry_cache)
2222
if result.revision_id != revision_id:
2223
raise AssertionError('revision id mismatch %s != %s' % (
2224
result.revision_id, revision_id))
2227
def serialise_inventory(self, inv):
2228
return self._serializer.write_inventory_to_string(inv)
2230
def _serialise_inventory_to_lines(self, inv):
2231
return self._serializer.write_inventory_to_lines(inv)
2233
def get_serializer_format(self):
2234
return self._serializer.format_num
2237
def get_inventory_xml(self, revision_id):
2238
"""Get inventory XML as a file object."""
2239
texts = self._iter_inventory_xmls([revision_id])
2241
text, revision_id = texts.next()
2242
except StopIteration:
2243
raise errors.HistoryMissing(self, 'inventory', revision_id)
2247
def get_inventory_sha1(self, revision_id):
2248
"""Return the sha1 hash of the inventory entry
2250
return self.get_revision(revision_id).inventory_sha1
959
2252
def get_rev_id_for_revno(self, revno, known_pair):
960
2253
"""Return the revision id of a revno, given a later (revno, revid)
1191
2518
@needs_write_lock
1192
2519
def sign_revision(self, revision_id, gpg_strategy):
1193
testament = _mod_testament.Testament.from_revision(self, revision_id)
1194
plaintext = testament.as_short_text()
2520
plaintext = Testament.from_revision(self, revision_id).as_short_text()
1195
2521
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
1197
2523
@needs_read_lock
1198
def verify_revision(self, revision_id, gpg_strategy):
1199
"""Verify the signature on a revision.
1201
:param revision_id: the revision to verify
1202
:gpg_strategy: the GPGStrategy object to used
1204
:return: gpg.SIGNATURE_VALID or a failed SIGNATURE_ value
1206
if not self.has_signature_for_revision_id(revision_id):
1207
return gpg.SIGNATURE_NOT_SIGNED, None
1208
signature = self.get_signature_text(revision_id)
1210
testament = _mod_testament.Testament.from_revision(self, revision_id)
1211
plaintext = testament.as_short_text()
1213
return gpg_strategy.verify(signature, plaintext)
1215
2524
def has_signature_for_revision_id(self, revision_id):
1216
2525
"""Query for a revision signature for revision_id in the repository."""
1217
raise NotImplementedError(self.has_signature_for_revision_id)
2526
if not self.has_revision(revision_id):
2527
raise errors.NoSuchRevision(self, revision_id)
2528
sig_present = (1 == len(
2529
self.signatures.get_parent_map([(revision_id,)])))
1219
2533
def get_signature_text(self, revision_id):
1220
2534
"""Return the text for a signature."""
1221
raise NotImplementedError(self.get_signature_text)
2535
stream = self.signatures.get_record_stream([(revision_id,)],
2537
record = stream.next()
2538
if record.storage_kind == 'absent':
2539
raise errors.NoSuchRevision(self, revision_id)
2540
return record.get_bytes_as('fulltext')
1223
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2543
def check(self, revision_ids=None):
1224
2544
"""Check consistency of all history of given revision_ids.
1226
2546
Different repository implementations should override _check().
1228
2548
:param revision_ids: A non-empty list of revision_ids whose ancestry
1229
2549
will be checked. Typically the last revision_id of a branch.
1230
:param callback_refs: A dict of check-refs to resolve and callback
1231
the check/_check method on the items listed as wanting the ref.
1233
:param check_repo: If False do not check the repository contents, just
1234
calculate the data callback_refs requires and call them back.
1236
return self._check(revision_ids=revision_ids, callback_refs=callback_refs,
1237
check_repo=check_repo)
1239
def _check(self, revision_ids=None, callback_refs=None, check_repo=True):
1240
raise NotImplementedError(self.check)
1242
def _warn_if_deprecated(self, branch=None):
1243
if not self._format.is_deprecated():
2551
return self._check(revision_ids)
2553
def _check(self, revision_ids):
2554
result = check.Check(self)
2558
def _warn_if_deprecated(self):
1245
2559
global _deprecation_warning_done
1246
2560
if _deprecation_warning_done:
1250
conf = config.GlobalConfig()
1252
conf = branch.get_config()
1253
if conf.suppress_warning('format_deprecation'):
1255
warning("Format %s for %s is deprecated -"
1256
" please use 'bzr upgrade' to get better performance"
1257
% (self._format, self.bzrdir.transport.base))
1259
_deprecation_warning_done = True
2562
_deprecation_warning_done = True
2563
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2564
% (self._format, self.bzrdir.transport.base))
1261
2566
def supports_rich_root(self):
1262
2567
return self._format.rich_root_data
1277
2582
except UnicodeDecodeError:
1278
2583
raise errors.NonAsciiRevisionId(method, self)
2585
def revision_graph_can_have_wrong_parents(self):
2586
"""Is it possible for this repository to have a revision graph with
2589
If True, then this repository must also implement
2590
_find_inconsistent_revision_parents so that check and reconcile can
2591
check for inconsistencies before proceeding with other checks that may
2592
depend on the revision index being consistent.
2594
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2597
# remove these delegates a while after bzr 0.15
2598
def __make_delegated(name, from_module):
2599
def _deprecated_repository_forwarder():
2600
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2601
% (name, from_module),
2604
m = __import__(from_module, globals(), locals(), [name])
2606
return getattr(m, name)
2607
except AttributeError:
2608
raise AttributeError('module %s has no name %s'
2610
globals()[name] = _deprecated_repository_forwarder
2613
'AllInOneRepository',
2614
'WeaveMetaDirRepository',
2615
'PreSplitOutRepositoryFormat',
2616
'RepositoryFormat4',
2617
'RepositoryFormat5',
2618
'RepositoryFormat6',
2619
'RepositoryFormat7',
2621
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2625
'RepositoryFormatKnit',
2626
'RepositoryFormatKnit1',
2628
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2631
def install_revision(repository, rev, revision_tree):
2632
"""Install all revision data into a repository."""
2633
install_revisions(repository, [(rev, revision_tree, None)])
2636
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2637
"""Install all revision data into a repository.
2639
Accepts an iterable of revision, tree, signature tuples. The signature
2642
repository.start_write_group()
2644
inventory_cache = lru_cache.LRUCache(10)
2645
for n, (revision, revision_tree, signature) in enumerate(iterable):
2646
_install_revision(repository, revision, revision_tree, signature,
2649
pb.update('Transferring revisions', n + 1, num_revisions)
2651
repository.abort_write_group()
2654
repository.commit_write_group()
2657
def _install_revision(repository, rev, revision_tree, signature,
2659
"""Install all revision data into a repository."""
2660
present_parents = []
2662
for p_id in rev.parent_ids:
2663
if repository.has_revision(p_id):
2664
present_parents.append(p_id)
2665
parent_trees[p_id] = repository.revision_tree(p_id)
2667
parent_trees[p_id] = repository.revision_tree(
2668
_mod_revision.NULL_REVISION)
2670
inv = revision_tree.inventory
2671
entries = inv.iter_entries()
2672
# backwards compatibility hack: skip the root id.
2673
if not repository.supports_rich_root():
2674
path, root = entries.next()
2675
if root.revision != rev.revision_id:
2676
raise errors.IncompatibleRevision(repr(repository))
2678
for path, ie in entries:
2679
text_keys[(ie.file_id, ie.revision)] = ie
2680
text_parent_map = repository.texts.get_parent_map(text_keys)
2681
missing_texts = set(text_keys) - set(text_parent_map)
2682
# Add the texts that are not already present
2683
for text_key in missing_texts:
2684
ie = text_keys[text_key]
2686
# FIXME: TODO: The following loop overlaps/duplicates that done by
2687
# commit to determine parents. There is a latent/real bug here where
2688
# the parents inserted are not those commit would do - in particular
2689
# they are not filtered by heads(). RBC, AB
2690
for revision, tree in parent_trees.iteritems():
2691
if ie.file_id not in tree:
2693
parent_id = tree.inventory[ie.file_id].revision
2694
if parent_id in text_parents:
2696
text_parents.append((ie.file_id, parent_id))
2697
lines = revision_tree.get_file(ie.file_id).readlines()
2698
repository.texts.add_lines(text_key, text_parents, lines)
2700
# install the inventory
2701
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2702
# Cache this inventory
2703
inventory_cache[rev.revision_id] = inv
2705
basis_inv = inventory_cache[rev.parent_ids[0]]
2707
repository.add_inventory(rev.revision_id, inv, present_parents)
2709
delta = inv._make_delta(basis_inv)
2710
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2711
rev.revision_id, present_parents)
2713
repository.add_inventory(rev.revision_id, inv, present_parents)
2714
except errors.RevisionAlreadyPresent:
2716
if signature is not None:
2717
repository.add_signature_text(rev.revision_id, signature)
2718
repository.add_revision(rev.revision_id, rev, inv)
1281
2721
class MetaDirRepository(Repository):
1282
2722
"""Repositories in the new meta-dir layout.
1621
3066
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
1622
3067
format_registry.register_lazy(
1623
3068
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
1624
'bzrlib.repofmt.knitpack_repo',
3069
'bzrlib.repofmt.pack_repo',
1625
3070
'RepositoryFormatKnitPack1',
1627
3072
format_registry.register_lazy(
1628
3073
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
1629
'bzrlib.repofmt.knitpack_repo',
3074
'bzrlib.repofmt.pack_repo',
1630
3075
'RepositoryFormatKnitPack3',
1632
3077
format_registry.register_lazy(
1633
3078
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
1634
'bzrlib.repofmt.knitpack_repo',
3079
'bzrlib.repofmt.pack_repo',
1635
3080
'RepositoryFormatKnitPack4',
1637
3082
format_registry.register_lazy(
1638
3083
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
1639
'bzrlib.repofmt.knitpack_repo',
3084
'bzrlib.repofmt.pack_repo',
1640
3085
'RepositoryFormatKnitPack5',
1642
3087
format_registry.register_lazy(
1643
3088
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
1644
'bzrlib.repofmt.knitpack_repo',
3089
'bzrlib.repofmt.pack_repo',
1645
3090
'RepositoryFormatKnitPack5RichRoot',
1647
3092
format_registry.register_lazy(
1648
3093
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
1649
'bzrlib.repofmt.knitpack_repo',
3094
'bzrlib.repofmt.pack_repo',
1650
3095
'RepositoryFormatKnitPack5RichRootBroken',
1652
3097
format_registry.register_lazy(
1653
3098
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
1654
'bzrlib.repofmt.knitpack_repo',
3099
'bzrlib.repofmt.pack_repo',
1655
3100
'RepositoryFormatKnitPack6',
1657
3102
format_registry.register_lazy(
1658
3103
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
1659
'bzrlib.repofmt.knitpack_repo',
3104
'bzrlib.repofmt.pack_repo',
1660
3105
'RepositoryFormatKnitPack6RichRoot',
1662
format_registry.register_lazy(
1663
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
1664
'bzrlib.repofmt.groupcompress_repo',
1665
'RepositoryFormat2a',
1668
3108
# Development formats.
1669
# Check their docstrings to see if/when they are obsolete.
3109
# Obsolete but kept pending a CHK based subtree format.
1670
3110
format_registry.register_lazy(
1671
3111
("Bazaar development format 2 with subtree support "
1672
3112
"(needs bzr.dev from before 1.8)\n"),
1673
'bzrlib.repofmt.knitpack_repo',
3113
'bzrlib.repofmt.pack_repo',
1674
3114
'RepositoryFormatPackDevelopment2Subtree',
1676
format_registry.register_lazy(
1677
'Bazaar development format 8\n',
1678
'bzrlib.repofmt.groupcompress_repo',
1679
'RepositoryFormat2aSubtree',
3117
# 1.14->1.16 go below here
3118
format_registry.register_lazy(
3119
'Bazaar development format - group compression and chk inventory'
3120
' (needs bzr.dev from 1.14)\n',
3121
'bzrlib.repofmt.groupcompress_repo',
3122
'RepositoryFormatCHK1',
3125
format_registry.register_lazy(
3126
'Bazaar development format - chk repository with bencode revision '
3127
'serialization (needs bzr.dev from 1.16)\n',
3128
'bzrlib.repofmt.groupcompress_repo',
3129
'RepositoryFormatCHK2',
3131
format_registry.register_lazy(
3132
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3133
'bzrlib.repofmt.groupcompress_repo',
3134
'RepositoryFormat2a',
1721
3177
:param revision_id: if None all content is copied, if NULL_REVISION no
1722
3178
content is copied.
3179
:param pb: optional progress bar to use for progress reports. If not
3180
provided a default one will be created.
1725
raise NotImplementedError(self.fetch)
3183
from bzrlib.fetch import RepoFetcher
3184
f = RepoFetcher(to_repository=self.target,
3185
from_repository=self.source,
3186
last_revision=revision_id,
3187
fetch_spec=fetch_spec,
3188
pb=pb, find_ghosts=find_ghosts)
3190
def _walk_to_common_revisions(self, revision_ids):
3191
"""Walk out from revision_ids in source to revisions target has.
3193
:param revision_ids: The start point for the search.
3194
:return: A set of revision ids.
3196
target_graph = self.target.get_graph()
3197
revision_ids = frozenset(revision_ids)
3198
missing_revs = set()
3199
source_graph = self.source.get_graph()
3200
# ensure we don't pay silly lookup costs.
3201
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3202
null_set = frozenset([_mod_revision.NULL_REVISION])
3203
searcher_exhausted = False
3207
# Iterate the searcher until we have enough next_revs
3208
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3210
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3211
next_revs.update(next_revs_part)
3212
ghosts.update(ghosts_part)
3213
except StopIteration:
3214
searcher_exhausted = True
3216
# If there are ghosts in the source graph, and the caller asked for
3217
# them, make sure that they are present in the target.
3218
# We don't care about other ghosts as we can't fetch them and
3219
# haven't been asked to.
3220
ghosts_to_check = set(revision_ids.intersection(ghosts))
3221
revs_to_get = set(next_revs).union(ghosts_to_check)
3223
have_revs = set(target_graph.get_parent_map(revs_to_get))
3224
# we always have NULL_REVISION present.
3225
have_revs = have_revs.union(null_set)
3226
# Check if the target is missing any ghosts we need.
3227
ghosts_to_check.difference_update(have_revs)
3229
# One of the caller's revision_ids is a ghost in both the
3230
# source and the target.
3231
raise errors.NoSuchRevision(
3232
self.source, ghosts_to_check.pop())
3233
missing_revs.update(next_revs - have_revs)
3234
# Because we may have walked past the original stop point, make
3235
# sure everything is stopped
3236
stop_revs = searcher.find_seen_ancestors(have_revs)
3237
searcher.stop_searching_any(stop_revs)
3238
if searcher_exhausted:
3240
return searcher.get_result()
1727
3242
@needs_read_lock
1728
def search_missing_revision_ids(self,
1729
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
1730
find_ghosts=True, revision_ids=None, if_present_ids=None,
3243
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1732
3244
"""Return the revision ids that source has that target does not.
1734
3246
:param revision_id: only return revision ids included by this
1736
:param revision_ids: return revision ids included by these
1737
revision_ids. NoSuchRevision will be raised if any of these
1738
revisions are not present.
1739
:param if_present_ids: like revision_ids, but will not cause
1740
NoSuchRevision if any of these are absent, instead they will simply
1741
not be in the result. This is useful for e.g. finding revisions
1742
to fetch for tags, which may reference absent revisions.
1743
3248
:param find_ghosts: If True find missing revisions in deep history
1744
3249
rather than just finding the surface difference.
1745
:param limit: Maximum number of revisions to return, topologically
1747
3250
:return: A bzrlib.graph.SearchResult.
1749
raise NotImplementedError(self.search_missing_revision_ids)
3252
# stop searching at found target revisions.
3253
if not find_ghosts and revision_id is not None:
3254
return self._walk_to_common_revisions([revision_id])
3255
# generic, possibly worst case, slow code path.
3256
target_ids = set(self.target.all_revision_ids())
3257
if revision_id is not None:
3258
source_ids = self.source.get_ancestry(revision_id)
3259
if source_ids[0] is not None:
3260
raise AssertionError()
3263
source_ids = self.source.all_revision_ids()
3264
result_set = set(source_ids).difference(target_ids)
3265
return self.source.revision_ids_to_search_result(result_set)
1752
3268
def _same_model(source, target):
1773
3289
"different serializers")
3292
class InterSameDataRepository(InterRepository):
3293
"""Code for converting between repositories that represent the same data.
3295
Data format and model must match for this to work.
3299
def _get_repo_format_to_test(self):
3300
"""Repository format for testing with.
3302
InterSameData can pull from subtree to subtree and from non-subtree to
3303
non-subtree, so we test this with the richest repository format.
3305
from bzrlib.repofmt import knitrepo
3306
return knitrepo.RepositoryFormatKnit3()
3309
def is_compatible(source, target):
3310
return InterRepository._same_model(source, target)
3313
class InterWeaveRepo(InterSameDataRepository):
3314
"""Optimised code paths between Weave based repositories.
3316
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3317
implemented lazy inter-object optimisation.
3321
def _get_repo_format_to_test(self):
3322
from bzrlib.repofmt import weaverepo
3323
return weaverepo.RepositoryFormat7()
3326
def is_compatible(source, target):
3327
"""Be compatible with known Weave formats.
3329
We don't test for the stores being of specific types because that
3330
could lead to confusing results, and there is no need to be
3333
from bzrlib.repofmt.weaverepo import (
3339
return (isinstance(source._format, (RepositoryFormat5,
3341
RepositoryFormat7)) and
3342
isinstance(target._format, (RepositoryFormat5,
3344
RepositoryFormat7)))
3345
except AttributeError:
3349
def copy_content(self, revision_id=None):
3350
"""See InterRepository.copy_content()."""
3351
# weave specific optimised path:
3353
self.target.set_make_working_trees(self.source.make_working_trees())
3354
except (errors.RepositoryUpgradeRequired, NotImplemented):
3356
# FIXME do not peek!
3357
if self.source._transport.listable():
3358
pb = ui.ui_factory.nested_progress_bar()
3360
self.target.texts.insert_record_stream(
3361
self.source.texts.get_record_stream(
3362
self.source.texts.keys(), 'topological', False))
3363
pb.update('copying inventory', 0, 1)
3364
self.target.inventories.insert_record_stream(
3365
self.source.inventories.get_record_stream(
3366
self.source.inventories.keys(), 'topological', False))
3367
self.target.signatures.insert_record_stream(
3368
self.source.signatures.get_record_stream(
3369
self.source.signatures.keys(),
3371
self.target.revisions.insert_record_stream(
3372
self.source.revisions.get_record_stream(
3373
self.source.revisions.keys(),
3374
'topological', True))
3378
self.target.fetch(self.source, revision_id=revision_id)
3381
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3382
"""See InterRepository.missing_revision_ids()."""
3383
# we want all revisions to satisfy revision_id in source.
3384
# but we don't want to stat every file here and there.
3385
# we want then, all revisions other needs to satisfy revision_id
3386
# checked, but not those that we have locally.
3387
# so the first thing is to get a subset of the revisions to
3388
# satisfy revision_id in source, and then eliminate those that
3389
# we do already have.
3390
# this is slow on high latency connection to self, but as this
3391
# disk format scales terribly for push anyway due to rewriting
3392
# inventory.weave, this is considered acceptable.
3394
if revision_id is not None:
3395
source_ids = self.source.get_ancestry(revision_id)
3396
if source_ids[0] is not None:
3397
raise AssertionError()
3400
source_ids = self.source._all_possible_ids()
3401
source_ids_set = set(source_ids)
3402
# source_ids is the worst possible case we may need to pull.
3403
# now we want to filter source_ids against what we actually
3404
# have in target, but don't try to check for existence where we know
3405
# we do not have a revision as that would be pointless.
3406
target_ids = set(self.target._all_possible_ids())
3407
possibly_present_revisions = target_ids.intersection(source_ids_set)
3408
actually_present_revisions = set(
3409
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3410
required_revisions = source_ids_set.difference(actually_present_revisions)
3411
if revision_id is not None:
3412
# we used get_ancestry to determine source_ids then we are assured all
3413
# revisions referenced are present as they are installed in topological order.
3414
# and the tip revision was validated by get_ancestry.
3415
result_set = required_revisions
3417
# if we just grabbed the possibly available ids, then
3418
# we only have an estimate of whats available and need to validate
3419
# that against the revision records.
3421
self.source._eliminate_revisions_not_present(required_revisions))
3422
return self.source.revision_ids_to_search_result(result_set)
3425
class InterKnitRepo(InterSameDataRepository):
3426
"""Optimised code paths between Knit based repositories."""
3429
def _get_repo_format_to_test(self):
3430
from bzrlib.repofmt import knitrepo
3431
return knitrepo.RepositoryFormatKnit1()
3434
def is_compatible(source, target):
3435
"""Be compatible with known Knit formats.
3437
We don't test for the stores being of specific types because that
3438
could lead to confusing results, and there is no need to be
3441
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3443
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3444
isinstance(target._format, RepositoryFormatKnit))
3445
except AttributeError:
3447
return are_knits and InterRepository._same_model(source, target)
3450
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3451
"""See InterRepository.missing_revision_ids()."""
3452
if revision_id is not None:
3453
source_ids = self.source.get_ancestry(revision_id)
3454
if source_ids[0] is not None:
3455
raise AssertionError()
3458
source_ids = self.source.all_revision_ids()
3459
source_ids_set = set(source_ids)
3460
# source_ids is the worst possible case we may need to pull.
3461
# now we want to filter source_ids against what we actually
3462
# have in target, but don't try to check for existence where we know
3463
# we do not have a revision as that would be pointless.
3464
target_ids = set(self.target.all_revision_ids())
3465
possibly_present_revisions = target_ids.intersection(source_ids_set)
3466
actually_present_revisions = set(
3467
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3468
required_revisions = source_ids_set.difference(actually_present_revisions)
3469
if revision_id is not None:
3470
# we used get_ancestry to determine source_ids then we are assured all
3471
# revisions referenced are present as they are installed in topological order.
3472
# and the tip revision was validated by get_ancestry.
3473
result_set = required_revisions
3475
# if we just grabbed the possibly available ids, then
3476
# we only have an estimate of whats available and need to validate
3477
# that against the revision records.
3479
self.source._eliminate_revisions_not_present(required_revisions))
3480
return self.source.revision_ids_to_search_result(result_set)
3483
class InterDifferingSerializer(InterRepository):
3486
def _get_repo_format_to_test(self):
3490
def is_compatible(source, target):
3491
"""Be compatible with Knit2 source and Knit3 target"""
3492
# This is redundant with format.check_conversion_target(), however that
3493
# raises an exception, and we just want to say "False" as in we won't
3494
# support converting between these formats.
3495
if source.supports_rich_root() and not target.supports_rich_root():
3497
if (source._format.supports_tree_reference
3498
and not target._format.supports_tree_reference):
3502
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3503
"""Get the best delta and base for this revision.
3505
:return: (basis_id, delta)
3507
possible_trees = [(parent_id, cache[parent_id])
3508
for parent_id in parent_ids
3509
if parent_id in cache]
3510
if len(possible_trees) == 0:
3511
# There either aren't any parents, or the parents aren't in the
3512
# cache, so just use the last converted tree
3513
possible_trees.append((basis_id, cache[basis_id]))
3515
for basis_id, basis_tree in possible_trees:
3516
delta = tree.inventory._make_delta(basis_tree.inventory)
3517
deltas.append((len(delta), basis_id, delta))
3519
return deltas[0][1:]
3521
def _get_parent_keys(self, root_key, parent_map):
3522
"""Get the parent keys for a given root id."""
3523
root_id, rev_id = root_key
3524
# Include direct parents of the revision, but only if they used
3525
# the same root_id and are heads.
3527
for parent_id in parent_map[rev_id]:
3528
if parent_id == _mod_revision.NULL_REVISION:
3530
if parent_id not in self._revision_id_to_root_id:
3531
# We probably didn't read this revision, go spend the
3532
# extra effort to actually check
3534
tree = self.source.revision_tree(parent_id)
3535
except errors.NoSuchRevision:
3536
# Ghost, fill out _revision_id_to_root_id in case we
3537
# encounter this again.
3538
# But set parent_root_id to None since we don't really know
3539
parent_root_id = None
3541
parent_root_id = tree.get_root_id()
3542
self._revision_id_to_root_id[parent_id] = None
3544
parent_root_id = self._revision_id_to_root_id[parent_id]
3545
if root_id == parent_root_id:
3546
# With stacking we _might_ want to refer to a non-local
3547
# revision, but this code path only applies when we have the
3548
# full content available, so ghosts really are ghosts, not just
3549
# the edge of local data.
3550
parent_keys.append((parent_id,))
3552
# root_id may be in the parent anyway.
3554
tree = self.source.revision_tree(parent_id)
3555
except errors.NoSuchRevision:
3556
# ghost, can't refer to it.
3560
parent_keys.append((tree.inventory[root_id].revision,))
3561
except errors.NoSuchId:
3564
g = graph.Graph(self.source.revisions)
3565
heads = g.heads(parent_keys)
3567
for key in parent_keys:
3568
if key in heads and key not in selected_keys:
3569
selected_keys.append(key)
3570
return tuple([(root_id,)+ key for key in selected_keys])
3572
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3573
for root_key in root_keys_to_create:
3574
parent_keys = self._get_parent_keys(root_key, parent_map)
3575
yield versionedfile.FulltextContentFactory(root_key,
3576
parent_keys, None, '')
3578
def _fetch_batch(self, revision_ids, basis_id, cache):
3579
"""Fetch across a few revisions.
3581
:param revision_ids: The revisions to copy
3582
:param basis_id: The revision_id of a tree that must be in cache, used
3583
as a basis for delta when no other base is available
3584
:param cache: A cache of RevisionTrees that we can use.
3585
:return: The revision_id of the last converted tree. The RevisionTree
3586
for it will be in cache
3588
# Walk though all revisions; get inventory deltas, copy referenced
3589
# texts that delta references, insert the delta, revision and
3591
root_keys_to_create = set()
3594
pending_revisions = []
3595
parent_map = self.source.get_parent_map(revision_ids)
3596
for tree in self.source.revision_trees(revision_ids):
3597
current_revision_id = tree.get_revision_id()
3598
parent_ids = parent_map.get(current_revision_id, ())
3599
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3601
if self._converting_to_rich_root:
3602
self._revision_id_to_root_id[current_revision_id] = \
3604
# Find text entries that need to be copied
3605
for old_path, new_path, file_id, entry in delta:
3606
if new_path is not None:
3609
if not self.target.supports_rich_root():
3610
# The target doesn't support rich root, so we don't
3613
if self._converting_to_rich_root:
3614
# This can't be copied normally, we have to insert
3616
root_keys_to_create.add((file_id, entry.revision))
3618
text_keys.add((file_id, entry.revision))
3619
revision = self.source.get_revision(current_revision_id)
3620
pending_deltas.append((basis_id, delta,
3621
current_revision_id, revision.parent_ids))
3622
pending_revisions.append(revision)
3623
cache[current_revision_id] = tree
3624
basis_id = current_revision_id
3626
from_texts = self.source.texts
3627
to_texts = self.target.texts
3628
if root_keys_to_create:
3629
root_stream = self._new_root_data_stream(root_keys_to_create,
3631
to_texts.insert_record_stream(root_stream)
3632
to_texts.insert_record_stream(from_texts.get_record_stream(
3633
text_keys, self.target._format._fetch_order,
3634
not self.target._format._fetch_uses_deltas))
3635
# insert inventory deltas
3636
for delta in pending_deltas:
3637
self.target.add_inventory_by_delta(*delta)
3638
if self.target._fallback_repositories:
3639
# Make sure this stacked repository has all the parent inventories
3640
# for the new revisions that we are about to insert. We do this
3641
# before adding the revisions so that no revision is added until
3642
# all the inventories it may depend on are added.
3644
revision_ids = set()
3645
for revision in pending_revisions:
3646
revision_ids.add(revision.revision_id)
3647
parent_ids.update(revision.parent_ids)
3648
parent_ids.difference_update(revision_ids)
3649
parent_ids.discard(_mod_revision.NULL_REVISION)
3650
parent_map = self.source.get_parent_map(parent_ids)
3651
for parent_tree in self.source.revision_trees(parent_ids):
3652
basis_id, delta = self._get_delta_for_revision(tree, parent_ids, basis_id, cache)
3653
current_revision_id = parent_tree.get_revision_id()
3654
parents_parents = parent_map[current_revision_id]
3655
self.target.add_inventory_by_delta(
3656
basis_id, delta, current_revision_id, parents_parents)
3657
# insert signatures and revisions
3658
for revision in pending_revisions:
3660
signature = self.source.get_signature_text(
3661
revision.revision_id)
3662
self.target.add_signature_text(revision.revision_id,
3664
except errors.NoSuchRevision:
3666
self.target.add_revision(revision.revision_id, revision)
3669
def _fetch_all_revisions(self, revision_ids, pb):
3670
"""Fetch everything for the list of revisions.
3672
:param revision_ids: The list of revisions to fetch. Must be in
3674
:param pb: A ProgressBar
3677
basis_id, basis_tree = self._get_basis(revision_ids[0])
3679
cache = lru_cache.LRUCache(100)
3680
cache[basis_id] = basis_tree
3681
del basis_tree # We don't want to hang on to it here
3683
for offset in range(0, len(revision_ids), batch_size):
3684
self.target.start_write_group()
3686
pb.update('Transferring revisions', offset,
3688
batch = revision_ids[offset:offset+batch_size]
3689
basis_id = self._fetch_batch(batch, basis_id, cache)
3691
self.target.abort_write_group()
3694
hint = self.target.commit_write_group()
3697
if hints and self.target._format.pack_compresses:
3698
self.target.pack(hint=hints)
3699
pb.update('Transferring revisions', len(revision_ids),
3703
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3705
"""See InterRepository.fetch()."""
3706
if fetch_spec is not None:
3707
raise AssertionError("Not implemented yet...")
3708
if (not self.source.supports_rich_root()
3709
and self.target.supports_rich_root()):
3710
self._converting_to_rich_root = True
3711
self._revision_id_to_root_id = {}
3713
self._converting_to_rich_root = False
3714
revision_ids = self.target.search_missing_revision_ids(self.source,
3715
revision_id, find_ghosts=find_ghosts).get_keys()
3716
if not revision_ids:
3718
revision_ids = tsort.topo_sort(
3719
self.source.get_graph().get_parent_map(revision_ids))
3720
if not revision_ids:
3722
# Walk though all revisions; get inventory deltas, copy referenced
3723
# texts that delta references, insert the delta, revision and
3725
first_rev = self.source.get_revision(revision_ids[0])
3727
my_pb = ui.ui_factory.nested_progress_bar()
3730
symbol_versioning.warn(
3731
symbol_versioning.deprecated_in((1, 14, 0))
3732
% "pb parameter to fetch()")
3735
self._fetch_all_revisions(revision_ids, pb)
3737
if my_pb is not None:
3739
return len(revision_ids), 0
3741
def _get_basis(self, first_revision_id):
3742
"""Get a revision and tree which exists in the target.
3744
This assumes that first_revision_id is selected for transmission
3745
because all other ancestors are already present. If we can't find an
3746
ancestor we fall back to NULL_REVISION since we know that is safe.
3748
:return: (basis_id, basis_tree)
3750
first_rev = self.source.get_revision(first_revision_id)
3752
basis_id = first_rev.parent_ids[0]
3753
# only valid as a basis if the target has it
3754
self.target.get_revision(basis_id)
3755
# Try to get a basis tree - if its a ghost it will hit the
3756
# NoSuchRevision case.
3757
basis_tree = self.source.revision_tree(basis_id)
3758
except (IndexError, errors.NoSuchRevision):
3759
basis_id = _mod_revision.NULL_REVISION
3760
basis_tree = self.source.revision_tree(basis_id)
3761
return basis_id, basis_tree
3764
InterRepository.register_optimiser(InterDifferingSerializer)
3765
InterRepository.register_optimiser(InterSameDataRepository)
3766
InterRepository.register_optimiser(InterWeaveRepo)
3767
InterRepository.register_optimiser(InterKnitRepo)
1776
3770
class CopyConverter(object):
1777
3771
"""A repository conversion tool which just performs a copy of the content.
1792
3786
:param to_convert: The disk object to convert.
1793
3787
:param pb: a progress bar to use for progress information.
1795
pb = ui.ui_factory.nested_progress_bar()
1798
3792
# this is only useful with metadir layouts - separated repo content.
1799
3793
# trigger an assertion if not such
1800
3794
repo._format.get_format_string()
1801
3795
self.repo_dir = repo.bzrdir
1802
pb.update('Moving repository to repository.backup')
3796
self.step('Moving repository to repository.backup')
1803
3797
self.repo_dir.transport.move('repository', 'repository.backup')
1804
3798
backup_transport = self.repo_dir.transport.clone('repository.backup')
1805
3799
repo._format.check_conversion_target(self.target_format)
1806
3800
self.source_repo = repo._format.open(self.repo_dir,
1808
3802
_override_transport=backup_transport)
1809
pb.update('Creating new repository')
3803
self.step('Creating new repository')
1810
3804
converted = self.target_format.initialize(self.repo_dir,
1811
3805
self.source_repo.is_shared())
1812
3806
converted.lock_write()
1814
pb.update('Copying content')
3808
self.step('Copying content into repository.')
1815
3809
self.source_repo.copy_content_into(converted)
1817
3811
converted.unlock()
1818
pb.update('Deleting old repository content')
3812
self.step('Deleting old repository content.')
1819
3813
self.repo_dir.transport.delete_tree('repository.backup')
1820
ui.ui_factory.note('repository converted')
3814
self.pb.note('repository converted')
3816
def step(self, message):
3817
"""Update the pb by a step."""
3819
self.pb.update(message, self.count, self.total)
3831
def _unescaper(match, _map=_unescape_map):
3832
code = match.group(1)
3836
if not code.startswith('#'):
3838
return unichr(int(code[1:])).encode('utf8')
3844
def _unescape_xml(data):
3845
"""Unescape predefined XML entities in a string of data."""
3847
if _unescape_re is None:
3848
_unescape_re = re.compile('\&([^;]*);')
3849
return _unescape_re.sub(_unescaper, data)
3852
class _VersionedFileChecker(object):
3854
def __init__(self, repository, text_key_references=None):
3855
self.repository = repository
3856
self.text_index = self.repository._generate_text_key_index(
3857
text_key_references=text_key_references)
3859
def calculate_file_version_parents(self, text_key):
3860
"""Calculate the correct parents for a file version according to
3863
parent_keys = self.text_index[text_key]
3864
if parent_keys == [_mod_revision.NULL_REVISION]:
3866
return tuple(parent_keys)
3868
def check_file_version_parents(self, texts, progress_bar=None):
3869
"""Check the parents stored in a versioned file are correct.
3871
It also detects file versions that are not referenced by their
3872
corresponding revision's inventory.
3874
:returns: A tuple of (wrong_parents, dangling_file_versions).
3875
wrong_parents is a dict mapping {revision_id: (stored_parents,
3876
correct_parents)} for each revision_id where the stored parents
3877
are not correct. dangling_file_versions is a set of (file_id,
3878
revision_id) tuples for versions that are present in this versioned
3879
file, but not used by the corresponding inventory.
3882
self.file_ids = set([file_id for file_id, _ in
3883
self.text_index.iterkeys()])
3884
# text keys is now grouped by file_id
3885
n_weaves = len(self.file_ids)
3886
files_in_revisions = {}
3887
revisions_of_files = {}
3888
n_versions = len(self.text_index)
3889
progress_bar.update('loading text store', 0, n_versions)
3890
parent_map = self.repository.texts.get_parent_map(self.text_index)
3891
# On unlistable transports this could well be empty/error...
3892
text_keys = self.repository.texts.keys()
3893
unused_keys = frozenset(text_keys) - set(self.text_index)
3894
for num, key in enumerate(self.text_index.iterkeys()):
3895
if progress_bar is not None:
3896
progress_bar.update('checking text graph', num, n_versions)
3897
correct_parents = self.calculate_file_version_parents(key)
3899
knit_parents = parent_map[key]
3900
except errors.RevisionNotPresent:
3903
if correct_parents != knit_parents:
3904
wrong_parents[key] = (knit_parents, correct_parents)
3905
return wrong_parents, unused_keys
3908
def _old_get_graph(repository, revision_id):
3909
"""DO NOT USE. That is all. I'm serious."""
3910
graph = repository.get_graph()
3911
revision_graph = dict(((key, value) for key, value in
3912
graph.iter_ancestry([revision_id]) if value is not None))
3913
return _strip_NULL_ghosts(revision_graph)
1824
3916
def _strip_NULL_ghosts(revision_graph):
1832
3924
return revision_graph
3927
class StreamSink(object):
3928
"""An object that can insert a stream into a repository.
3930
This interface handles the complexity of reserialising inventories and
3931
revisions from different formats, and allows unidirectional insertion into
3932
stacked repositories without looking for the missing basis parents
3936
def __init__(self, target_repo):
3937
self.target_repo = target_repo
3939
def insert_stream(self, stream, src_format, resume_tokens):
3940
"""Insert a stream's content into the target repository.
3942
:param src_format: a bzr repository format.
3944
:return: a list of resume tokens and an iterable of keys additional
3945
items required before the insertion can be completed.
3947
self.target_repo.lock_write()
3950
self.target_repo.resume_write_group(resume_tokens)
3953
self.target_repo.start_write_group()
3956
# locked_insert_stream performs a commit|suspend.
3957
return self._locked_insert_stream(stream, src_format, is_resume)
3959
self.target_repo.abort_write_group(suppress_errors=True)
3962
self.target_repo.unlock()
3964
def _locked_insert_stream(self, stream, src_format, is_resume):
3965
to_serializer = self.target_repo._format._serializer
3966
src_serializer = src_format._serializer
3968
if to_serializer == src_serializer:
3969
# If serializers match and the target is a pack repository, set the
3970
# write cache size on the new pack. This avoids poor performance
3971
# on transports where append is unbuffered (such as
3972
# RemoteTransport). This is safe to do because nothing should read
3973
# back from the target repository while a stream with matching
3974
# serialization is being inserted.
3975
# The exception is that a delta record from the source that should
3976
# be a fulltext may need to be expanded by the target (see
3977
# test_fetch_revisions_with_deltas_into_pack); but we take care to
3978
# explicitly flush any buffered writes first in that rare case.
3980
new_pack = self.target_repo._pack_collection._new_pack
3981
except AttributeError:
3982
# Not a pack repository
3985
new_pack.set_write_cache_size(1024*1024)
3986
for substream_type, substream in stream:
3987
if substream_type == 'texts':
3988
self.target_repo.texts.insert_record_stream(substream)
3989
elif substream_type == 'inventories':
3990
if src_serializer == to_serializer:
3991
self.target_repo.inventories.insert_record_stream(
3994
self._extract_and_insert_inventories(
3995
substream, src_serializer)
3996
elif substream_type == 'chk_bytes':
3997
# XXX: This doesn't support conversions, as it assumes the
3998
# conversion was done in the fetch code.
3999
self.target_repo.chk_bytes.insert_record_stream(substream)
4000
elif substream_type == 'revisions':
4001
# This may fallback to extract-and-insert more often than
4002
# required if the serializers are different only in terms of
4004
if src_serializer == to_serializer:
4005
self.target_repo.revisions.insert_record_stream(
4008
self._extract_and_insert_revisions(substream,
4010
elif substream_type == 'signatures':
4011
self.target_repo.signatures.insert_record_stream(substream)
4013
raise AssertionError('kaboom! %s' % (substream_type,))
4014
# Done inserting data, and the missing_keys calculations will try to
4015
# read back from the inserted data, so flush the writes to the new pack
4016
# (if this is pack format).
4017
if new_pack is not None:
4018
new_pack._write_data('', flush=True)
4019
# Find all the new revisions (including ones from resume_tokens)
4020
missing_keys = self.target_repo.get_missing_parent_inventories(
4021
check_for_missing_texts=is_resume)
4023
for prefix, versioned_file in (
4024
('texts', self.target_repo.texts),
4025
('inventories', self.target_repo.inventories),
4026
('revisions', self.target_repo.revisions),
4027
('signatures', self.target_repo.signatures),
4028
('chk_bytes', self.target_repo.chk_bytes),
4030
if versioned_file is None:
4032
missing_keys.update((prefix,) + key for key in
4033
versioned_file.get_missing_compression_parent_keys())
4034
except NotImplementedError:
4035
# cannot even attempt suspending, and missing would have failed
4036
# during stream insertion.
4037
missing_keys = set()
4040
# suspend the write group and tell the caller what we is
4041
# missing. We know we can suspend or else we would not have
4042
# entered this code path. (All repositories that can handle
4043
# missing keys can handle suspending a write group).
4044
write_group_tokens = self.target_repo.suspend_write_group()
4045
return write_group_tokens, missing_keys
4046
hint = self.target_repo.commit_write_group()
4047
if (to_serializer != src_serializer and
4048
self.target_repo._format.pack_compresses):
4049
self.target_repo.pack(hint=hint)
4052
def _extract_and_insert_inventories(self, substream, serializer):
4053
"""Generate a new inventory versionedfile in target, converting data.
4055
The inventory is retrieved from the source, (deserializing it), and
4056
stored in the target (reserializing it in a different format).
4058
for record in substream:
4059
bytes = record.get_bytes_as('fulltext')
4060
revision_id = record.key[0]
4061
inv = serializer.read_inventory_from_string(bytes, revision_id)
4062
parents = [key[0] for key in record.parents]
4063
self.target_repo.add_inventory(revision_id, inv, parents)
4065
def _extract_and_insert_revisions(self, substream, serializer):
4066
for record in substream:
4067
bytes = record.get_bytes_as('fulltext')
4068
revision_id = record.key[0]
4069
rev = serializer.read_revision_from_string(bytes)
4070
if rev.revision_id != revision_id:
4071
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4072
self.target_repo.add_revision(revision_id, rev)
4075
if self.target_repo._format._fetch_reconcile:
4076
self.target_repo.reconcile()
4079
class StreamSource(object):
4080
"""A source of a stream for fetching between repositories."""
4082
def __init__(self, from_repository, to_format):
4083
"""Create a StreamSource streaming from from_repository."""
4084
self.from_repository = from_repository
4085
self.to_format = to_format
4087
def delta_on_metadata(self):
4088
"""Return True if delta's are permitted on metadata streams.
4090
That is on revisions and signatures.
4092
src_serializer = self.from_repository._format._serializer
4093
target_serializer = self.to_format._serializer
4094
return (self.to_format._fetch_uses_deltas and
4095
src_serializer == target_serializer)
4097
def _fetch_revision_texts(self, revs):
4098
# fetch signatures first and then the revision texts
4099
# may need to be a InterRevisionStore call here.
4100
from_sf = self.from_repository.signatures
4101
# A missing signature is just skipped.
4102
keys = [(rev_id,) for rev_id in revs]
4103
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4105
self.to_format._fetch_order,
4106
not self.to_format._fetch_uses_deltas))
4107
# If a revision has a delta, this is actually expanded inside the
4108
# insert_record_stream code now, which is an alternate fix for
4110
from_rf = self.from_repository.revisions
4111
revisions = from_rf.get_record_stream(
4113
self.to_format._fetch_order,
4114
not self.delta_on_metadata())
4115
return [('signatures', signatures), ('revisions', revisions)]
4117
def _generate_root_texts(self, revs):
4118
"""This will be called by __fetch between fetching weave texts and
4119
fetching the inventory weave.
4121
Subclasses should override this if they need to generate root texts
4122
after fetching weave texts.
4124
if self._rich_root_upgrade():
4126
return bzrlib.fetch.Inter1and2Helper(
4127
self.from_repository).generate_root_texts(revs)
4131
def get_stream(self, search):
4133
revs = search.get_keys()
4134
graph = self.from_repository.get_graph()
4135
revs = list(graph.iter_topo_order(revs))
4136
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4138
for knit_kind, file_id, revisions in data_to_fetch:
4139
if knit_kind != phase:
4141
# Make a new progress bar for this phase
4142
if knit_kind == "file":
4143
# Accumulate file texts
4144
text_keys.extend([(file_id, revision) for revision in
4146
elif knit_kind == "inventory":
4147
# Now copy the file texts.
4148
from_texts = self.from_repository.texts
4149
yield ('texts', from_texts.get_record_stream(
4150
text_keys, self.to_format._fetch_order,
4151
not self.to_format._fetch_uses_deltas))
4152
# Cause an error if a text occurs after we have done the
4155
# Before we process the inventory we generate the root
4156
# texts (if necessary) so that the inventories references
4158
for _ in self._generate_root_texts(revs):
4160
# NB: This currently reopens the inventory weave in source;
4161
# using a single stream interface instead would avoid this.
4162
from_weave = self.from_repository.inventories
4163
# we fetch only the referenced inventories because we do not
4164
# know for unselected inventories whether all their required
4165
# texts are present in the other repository - it could be
4167
for info in self._get_inventory_stream(revs):
4169
elif knit_kind == "signatures":
4170
# Nothing to do here; this will be taken care of when
4171
# _fetch_revision_texts happens.
4173
elif knit_kind == "revisions":
4174
for record in self._fetch_revision_texts(revs):
4177
raise AssertionError("Unknown knit kind %r" % knit_kind)
4179
def get_stream_for_missing_keys(self, missing_keys):
4180
# missing keys can only occur when we are byte copying and not
4181
# translating (because translation means we don't send
4182
# unreconstructable deltas ever).
4184
keys['texts'] = set()
4185
keys['revisions'] = set()
4186
keys['inventories'] = set()
4187
keys['chk_bytes'] = set()
4188
keys['signatures'] = set()
4189
for key in missing_keys:
4190
keys[key[0]].add(key[1:])
4191
if len(keys['revisions']):
4192
# If we allowed copying revisions at this point, we could end up
4193
# copying a revision without copying its required texts: a
4194
# violation of the requirements for repository integrity.
4195
raise AssertionError(
4196
'cannot copy revisions to fill in missing deltas %s' % (
4197
keys['revisions'],))
4198
for substream_kind, keys in keys.iteritems():
4199
vf = getattr(self.from_repository, substream_kind)
4200
if vf is None and keys:
4201
raise AssertionError(
4202
"cannot fill in keys for a versioned file we don't"
4203
" have: %s needs %s" % (substream_kind, keys))
4205
# No need to stream something we don't have
4207
# Ask for full texts always so that we don't need more round trips
4208
# after this stream.
4209
# Some of the missing keys are genuinely ghosts, so filter absent
4210
# records. The Sink is responsible for doing another check to
4211
# ensure that ghosts don't introduce missing data for future
4213
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4214
self.to_format._fetch_order, True))
4215
yield substream_kind, stream
4217
def inventory_fetch_order(self):
4218
if self._rich_root_upgrade():
4219
return 'topological'
4221
return self.to_format._fetch_order
4223
def _rich_root_upgrade(self):
4224
return (not self.from_repository._format.rich_root_data and
4225
self.to_format.rich_root_data)
4227
def _get_inventory_stream(self, revision_ids):
4228
from_format = self.from_repository._format
4229
if (from_format.supports_chks and self.to_format.supports_chks
4230
and (from_format._serializer == self.to_format._serializer)):
4231
# Both sides support chks, and they use the same serializer, so it
4232
# is safe to transmit the chk pages and inventory pages across
4234
return self._get_chk_inventory_stream(revision_ids)
4235
elif (not from_format.supports_chks):
4236
# Source repository doesn't support chks. So we can transmit the
4237
# inventories 'as-is' and either they are just accepted on the
4238
# target, or the Sink will properly convert it.
4239
return self._get_simple_inventory_stream(revision_ids)
4241
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4242
# inventories. Note that this should probably be done somehow
4243
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4244
# figure out how a non-chk repository could possibly handle
4245
# deserializing an inventory stream from a chk repo, as it
4246
# doesn't have a way to understand individual pages.
4247
return self._get_convertable_inventory_stream(revision_ids)
4249
def _get_simple_inventory_stream(self, revision_ids):
4250
from_weave = self.from_repository.inventories
4251
yield ('inventories', from_weave.get_record_stream(
4252
[(rev_id,) for rev_id in revision_ids],
4253
self.inventory_fetch_order(),
4254
not self.delta_on_metadata()))
4256
def _get_chk_inventory_stream(self, revision_ids):
4257
"""Fetch the inventory texts, along with the associated chk maps."""
4258
# We want an inventory outside of the search set, so that we can filter
4259
# out uninteresting chk pages. For now we use
4260
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4261
# could use that instead.
4262
start_rev_id = self.from_repository._find_revision_outside_set(
4264
start_rev_key = (start_rev_id,)
4265
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4266
if start_rev_id != _mod_revision.NULL_REVISION:
4267
inv_keys_to_fetch.append((start_rev_id,))
4268
# Any repo that supports chk_bytes must also support out-of-order
4269
# insertion. At least, that is how we expect it to work
4270
# We use get_record_stream instead of iter_inventories because we want
4271
# to be able to insert the stream as well. We could instead fetch
4272
# allowing deltas, and then iter_inventories, but we don't know whether
4273
# source or target is more 'local' anway.
4274
inv_stream = self.from_repository.inventories.get_record_stream(
4275
inv_keys_to_fetch, 'unordered',
4276
True) # We need them as full-texts so we can find their references
4277
uninteresting_chk_roots = set()
4278
interesting_chk_roots = set()
4279
def filter_inv_stream(inv_stream):
4280
for idx, record in enumerate(inv_stream):
4281
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4282
bytes = record.get_bytes_as('fulltext')
4283
chk_inv = inventory.CHKInventory.deserialise(
4284
self.from_repository.chk_bytes, bytes, record.key)
4285
if record.key == start_rev_key:
4286
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4287
p_id_map = chk_inv.parent_id_basename_to_file_id
4288
if p_id_map is not None:
4289
uninteresting_chk_roots.add(p_id_map.key())
4292
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4293
p_id_map = chk_inv.parent_id_basename_to_file_id
4294
if p_id_map is not None:
4295
interesting_chk_roots.add(p_id_map.key())
4296
### pb.update('fetch inventory', 0, 2)
4297
yield ('inventories', filter_inv_stream(inv_stream))
4298
# Now that we have worked out all of the interesting root nodes, grab
4299
# all of the interesting pages and insert them
4300
### pb.update('fetch inventory', 1, 2)
4301
interesting = chk_map.iter_interesting_nodes(
4302
self.from_repository.chk_bytes, interesting_chk_roots,
4303
uninteresting_chk_roots)
4304
def to_stream_adapter():
4305
"""Adapt the iter_interesting_nodes result to a single stream.
4307
iter_interesting_nodes returns records as it processes them, along
4308
with keys. However, we only want to return the records themselves.
4310
for record, items in interesting:
4311
if record is not None:
4313
# XXX: We could instead call get_record_stream(records.keys())
4314
# ATM, this will always insert the records as fulltexts, and
4315
# requires that you can hang on to records once you have gone
4316
# on to the next one. Further, it causes the target to
4317
# recompress the data. Testing shows it to be faster than
4318
# requesting the records again, though.
4319
yield ('chk_bytes', to_stream_adapter())
4320
### pb.update('fetch inventory', 2, 2)
4322
def _get_convertable_inventory_stream(self, revision_ids):
4323
# XXX: One of source or target is using chks, and they don't have
4324
# compatible serializations. The StreamSink code expects to be
4325
# able to convert on the target, so we need to put
4326
# bytes-on-the-wire that can be converted
4327
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4329
def _stream_invs_as_fulltexts(self, revision_ids):
4330
from_repo = self.from_repository
4331
from_serializer = from_repo._format._serializer
4332
revision_keys = [(rev_id,) for rev_id in revision_ids]
4333
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4334
for inv in self.from_repository.iter_inventories(revision_ids):
4335
# XXX: This is a bit hackish, but it works. Basically,
4336
# CHKSerializer 'accidentally' supports
4337
# read/write_inventory_to_string, even though that is never
4338
# the format that is stored on disk. It *does* give us a
4339
# single string representation for an inventory, so live with
4341
# This would be far better if we had a 'serialized inventory
4342
# delta' form. Then we could use 'inventory._make_delta', and
4343
# transmit that. This would both be faster to generate, and
4344
# result in fewer bytes-on-the-wire.
4345
as_bytes = from_serializer.write_inventory_to_string(inv)
4346
key = (inv.revision_id,)
4347
parent_keys = parent_map.get(key, ())
4348
yield versionedfile.FulltextContentFactory(
4349
key, parent_keys, None, as_bytes)
1835
4352
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
1836
4353
stop_revision=None):
1837
4354
"""Extend the partial history to include a given index