202
336
commit to be valid, deletes against the basis MUST be recorded via
203
337
builder.record_delete().
205
raise NotImplementedError(self.will_record_deletes)
207
def record_iter_changes(self, tree, basis_revision_id, iter_changes):
339
self._recording_deletes = True
341
basis_id = self.parents[0]
343
basis_id = _mod_revision.NULL_REVISION
344
self.basis_delta_revision = basis_id
346
def record_entry_contents(self, ie, parent_invs, path, tree,
348
"""Record the content of ie from tree into the commit if needed.
350
Side effect: sets ie.revision when unchanged
352
:param ie: An inventory entry present in the commit.
353
:param parent_invs: The inventories of the parent revisions of the
355
:param path: The path the entry is at in the tree.
356
:param tree: The tree which contains this entry and should be used to
358
:param content_summary: Summary data from the tree about the paths
359
content - stat, length, exec, sha/link target. This is only
360
accessed when the entry has a revision of None - that is when it is
361
a candidate to commit.
362
:return: A tuple (change_delta, version_recorded, fs_hash).
363
change_delta is an inventory_delta change for this entry against
364
the basis tree of the commit, or None if no change occured against
366
version_recorded is True if a new version of the entry has been
367
recorded. For instance, committing a merge where a file was only
368
changed on the other side will return (delta, False).
369
fs_hash is either None, or the hash details for the path (currently
370
a tuple of the contents sha1 and the statvalue returned by
371
tree.get_file_with_stat()).
373
if self.new_inventory.root is None:
374
if ie.parent_id is not None:
375
raise errors.RootMissing()
376
self._check_root(ie, parent_invs, tree)
377
if ie.revision is None:
378
kind = content_summary[0]
380
# ie is carried over from a prior commit
382
# XXX: repository specific check for nested tree support goes here - if
383
# the repo doesn't want nested trees we skip it ?
384
if (kind == 'tree-reference' and
385
not self.repository._format.supports_tree_reference):
386
# mismatch between commit builder logic and repository:
387
# this needs the entry creation pushed down into the builder.
388
raise NotImplementedError('Missing repository subtree support.')
389
self.new_inventory.add(ie)
391
# TODO: slow, take it out of the inner loop.
393
basis_inv = parent_invs[0]
395
basis_inv = Inventory(root_id=None)
397
# ie.revision is always None if the InventoryEntry is considered
398
# for committing. We may record the previous parents revision if the
399
# content is actually unchanged against a sole head.
400
if ie.revision is not None:
401
if not self._versioned_root and path == '':
402
# repositories that do not version the root set the root's
403
# revision to the new commit even when no change occurs (more
404
# specifically, they do not record a revision on the root; and
405
# the rev id is assigned to the root during deserialisation -
406
# this masks when a change may have occurred against the basis.
407
# To match this we always issue a delta, because the revision
408
# of the root will always be changing.
409
if ie.file_id in basis_inv:
410
delta = (basis_inv.id2path(ie.file_id), path,
414
delta = (None, path, ie.file_id, ie)
415
self._basis_delta.append(delta)
416
return delta, False, None
418
# we don't need to commit this, because the caller already
419
# determined that an existing revision of this file is
420
# appropriate. If its not being considered for committing then
421
# it and all its parents to the root must be unaltered so
422
# no-change against the basis.
423
if ie.revision == self._new_revision_id:
424
raise AssertionError("Impossible situation, a skipped "
425
"inventory entry (%r) claims to be modified in this "
426
"commit (%r).", (ie, self._new_revision_id))
427
return None, False, None
428
# XXX: Friction: parent_candidates should return a list not a dict
429
# so that we don't have to walk the inventories again.
430
parent_candiate_entries = ie.parent_candidates(parent_invs)
431
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
433
for inv in parent_invs:
434
if ie.file_id in inv:
435
old_rev = inv[ie.file_id].revision
436
if old_rev in head_set:
437
heads.append(inv[ie.file_id].revision)
438
head_set.remove(inv[ie.file_id].revision)
441
# now we check to see if we need to write a new record to the
443
# We write a new entry unless there is one head to the ancestors, and
444
# the kind-derived content is unchanged.
446
# Cheapest check first: no ancestors, or more the one head in the
447
# ancestors, we write a new node.
451
# There is a single head, look it up for comparison
452
parent_entry = parent_candiate_entries[heads[0]]
453
# if the non-content specific data has changed, we'll be writing a
455
if (parent_entry.parent_id != ie.parent_id or
456
parent_entry.name != ie.name):
458
# now we need to do content specific checks:
460
# if the kind changed the content obviously has
461
if kind != parent_entry.kind:
463
# Stat cache fingerprint feedback for the caller - None as we usually
464
# don't generate one.
467
if content_summary[2] is None:
468
raise ValueError("Files must not have executable = None")
470
if (# if the file length changed we have to store:
471
parent_entry.text_size != content_summary[1] or
472
# if the exec bit has changed we have to store:
473
parent_entry.executable != content_summary[2]):
475
elif parent_entry.text_sha1 == content_summary[3]:
476
# all meta and content is unchanged (using a hash cache
477
# hit to check the sha)
478
ie.revision = parent_entry.revision
479
ie.text_size = parent_entry.text_size
480
ie.text_sha1 = parent_entry.text_sha1
481
ie.executable = parent_entry.executable
482
return self._get_delta(ie, basis_inv, path), False, None
484
# Either there is only a hash change(no hash cache entry,
485
# or same size content change), or there is no change on
487
# Provide the parent's hash to the store layer, so that the
488
# content is unchanged we will not store a new node.
489
nostore_sha = parent_entry.text_sha1
491
# We want to record a new node regardless of the presence or
492
# absence of a content change in the file.
494
ie.executable = content_summary[2]
495
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
497
lines = file_obj.readlines()
501
ie.text_sha1, ie.text_size = self._add_text_to_weave(
502
ie.file_id, lines, heads, nostore_sha)
503
# Let the caller know we generated a stat fingerprint.
504
fingerprint = (ie.text_sha1, stat_value)
505
except errors.ExistingContent:
506
# Turns out that the file content was unchanged, and we were
507
# only going to store a new node if it was changed. Carry over
509
ie.revision = parent_entry.revision
510
ie.text_size = parent_entry.text_size
511
ie.text_sha1 = parent_entry.text_sha1
512
ie.executable = parent_entry.executable
513
return self._get_delta(ie, basis_inv, path), False, None
514
elif kind == 'directory':
516
# all data is meta here, nothing specific to directory, so
518
ie.revision = parent_entry.revision
519
return self._get_delta(ie, basis_inv, path), False, None
521
self._add_text_to_weave(ie.file_id, lines, heads, None)
522
elif kind == 'symlink':
523
current_link_target = content_summary[3]
525
# symlink target is not generic metadata, check if it has
527
if current_link_target != parent_entry.symlink_target:
530
# unchanged, carry over.
531
ie.revision = parent_entry.revision
532
ie.symlink_target = parent_entry.symlink_target
533
return self._get_delta(ie, basis_inv, path), False, None
534
ie.symlink_target = current_link_target
536
self._add_text_to_weave(ie.file_id, lines, heads, None)
537
elif kind == 'tree-reference':
539
if content_summary[3] != parent_entry.reference_revision:
542
# unchanged, carry over.
543
ie.reference_revision = parent_entry.reference_revision
544
ie.revision = parent_entry.revision
545
return self._get_delta(ie, basis_inv, path), False, None
546
ie.reference_revision = content_summary[3]
548
self._add_text_to_weave(ie.file_id, lines, heads, None)
550
raise NotImplementedError('unknown kind')
551
ie.revision = self._new_revision_id
552
self._any_changes = True
553
return self._get_delta(ie, basis_inv, path), True, fingerprint
555
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
556
_entry_factory=entry_factory):
208
557
"""Record a new tree via iter_changes.
210
559
:param tree: The tree to obtain text contents from for changed objects.
215
564
to basis_revision_id. The iterator must not include any items with
216
565
a current kind of None - missing items must be either filtered out
217
566
or errored-on beefore record_iter_changes sees the item.
567
:param _entry_factory: Private method to bind entry_factory locally for
218
569
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
219
570
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,
572
# Create an inventory delta based on deltas between all the parents and
573
# deltas between all the parent inventories. We use inventory delta's
574
# between the inventory objects because iter_changes masks
575
# last-changed-field only changes.
577
# file_id -> change map, change is fileid, paths, changed, versioneds,
578
# parents, names, kinds, executables
580
# {file_id -> revision_id -> inventory entry, for entries in parent
581
# trees that are not parents[0]
585
revtrees = list(self.repository.revision_trees(self.parents))
586
except errors.NoSuchRevision:
587
# one or more ghosts, slow path.
589
for revision_id in self.parents:
591
revtrees.append(self.repository.revision_tree(revision_id))
592
except errors.NoSuchRevision:
594
basis_revision_id = _mod_revision.NULL_REVISION
596
revtrees.append(self.repository.revision_tree(
597
_mod_revision.NULL_REVISION))
598
# The basis inventory from a repository
600
basis_inv = revtrees[0].inventory
602
basis_inv = self.repository.revision_tree(
603
_mod_revision.NULL_REVISION).inventory
604
if len(self.parents) > 0:
605
if basis_revision_id != self.parents[0] and not ghost_basis:
607
"arbitrary basis parents not yet supported with merges")
608
for revtree in revtrees[1:]:
609
for change in revtree.inventory._make_delta(basis_inv):
610
if change[1] is None:
611
# Not present in this parent.
613
if change[2] not in merged_ids:
614
if change[0] is not None:
615
basis_entry = basis_inv[change[2]]
616
merged_ids[change[2]] = [
618
basis_entry.revision,
621
parent_entries[change[2]] = {
623
basis_entry.revision:basis_entry,
625
change[3].revision:change[3],
628
merged_ids[change[2]] = [change[3].revision]
629
parent_entries[change[2]] = {change[3].revision:change[3]}
631
merged_ids[change[2]].append(change[3].revision)
632
parent_entries[change[2]][change[3].revision] = change[3]
635
# Setup the changes from the tree:
636
# changes maps file_id -> (change, [parent revision_ids])
638
for change in iter_changes:
639
# This probably looks up in basis_inv way to much.
640
if change[1][0] is not None:
641
head_candidate = [basis_inv[change[0]].revision]
644
changes[change[0]] = change, merged_ids.get(change[0],
646
unchanged_merged = set(merged_ids) - set(changes)
647
# Extend the changes dict with synthetic changes to record merges of
649
for file_id in unchanged_merged:
650
# Record a merged version of these items that did not change vs the
651
# basis. This can be either identical parallel changes, or a revert
652
# of a specific file after a merge. The recorded content will be
653
# that of the current tree (which is the same as the basis), but
654
# the per-file graph will reflect a merge.
655
# NB:XXX: We are reconstructing path information we had, this
656
# should be preserved instead.
657
# inv delta change: (file_id, (path_in_source, path_in_target),
658
# changed_content, versioned, parent, name, kind,
661
basis_entry = basis_inv[file_id]
662
except errors.NoSuchId:
663
# a change from basis->some_parents but file_id isn't in basis
664
# so was new in the merge, which means it must have changed
665
# from basis -> current, and as it hasn't the add was reverted
666
# by the user. So we discard this change.
670
(basis_inv.id2path(file_id), tree.id2path(file_id)),
672
(basis_entry.parent_id, basis_entry.parent_id),
673
(basis_entry.name, basis_entry.name),
674
(basis_entry.kind, basis_entry.kind),
675
(basis_entry.executable, basis_entry.executable))
676
changes[file_id] = (change, merged_ids[file_id])
677
# changes contains tuples with the change and a set of inventory
678
# candidates for the file.
680
# old_path, new_path, file_id, new_inventory_entry
681
seen_root = False # Is the root in the basis delta?
682
inv_delta = self._basis_delta
683
modified_rev = self._new_revision_id
684
for change, head_candidates in changes.values():
685
if change[3][1]: # versioned in target.
686
# Several things may be happening here:
687
# We may have a fork in the per-file graph
688
# - record a change with the content from tree
689
# We may have a change against < all trees
690
# - carry over the tree that hasn't changed
691
# We may have a change against all trees
692
# - record the change with the content from tree
695
entry = _entry_factory[kind](file_id, change[5][1],
697
head_set = self._heads(change[0], set(head_candidates))
700
for head_candidate in head_candidates:
701
if head_candidate in head_set:
702
heads.append(head_candidate)
703
head_set.remove(head_candidate)
706
# Could be a carry-over situation:
707
parent_entry_revs = parent_entries.get(file_id, None)
708
if parent_entry_revs:
709
parent_entry = parent_entry_revs.get(heads[0], None)
712
if parent_entry is None:
713
# The parent iter_changes was called against is the one
714
# that is the per-file head, so any change is relevant
715
# iter_changes is valid.
716
carry_over_possible = False
718
# could be a carry over situation
719
# A change against the basis may just indicate a merge,
720
# we need to check the content against the source of the
721
# merge to determine if it was changed after the merge
723
if (parent_entry.kind != entry.kind or
724
parent_entry.parent_id != entry.parent_id or
725
parent_entry.name != entry.name):
726
# Metadata common to all entries has changed
727
# against per-file parent
728
carry_over_possible = False
730
carry_over_possible = True
731
# per-type checks for changes against the parent_entry
734
# Cannot be a carry-over situation
735
carry_over_possible = False
736
# Populate the entry in the delta
738
# XXX: There is still a small race here: If someone reverts the content of a file
739
# after iter_changes examines and decides it has changed,
740
# we will unconditionally record a new version even if some
741
# other process reverts it while commit is running (with
742
# the revert happening after iter_changes did it's
745
entry.executable = True
747
entry.executable = False
748
if (carry_over_possible and
749
parent_entry.executable == entry.executable):
750
# Check the file length, content hash after reading
752
nostore_sha = parent_entry.text_sha1
755
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
757
lines = file_obj.readlines()
761
entry.text_sha1, entry.text_size = self._add_text_to_weave(
762
file_id, lines, heads, nostore_sha)
763
yield file_id, change[1][1], (entry.text_sha1, stat_value)
764
except errors.ExistingContent:
765
# No content change against a carry_over parent
766
# Perhaps this should also yield a fs hash update?
768
entry.text_size = parent_entry.text_size
769
entry.text_sha1 = parent_entry.text_sha1
770
elif kind == 'symlink':
772
entry.symlink_target = tree.get_symlink_target(file_id)
773
if (carry_over_possible and
774
parent_entry.symlink_target == entry.symlink_target):
777
self._add_text_to_weave(change[0], [], heads, None)
778
elif kind == 'directory':
779
if carry_over_possible:
782
# Nothing to set on the entry.
783
# XXX: split into the Root and nonRoot versions.
784
if change[1][1] != '' or self.repository.supports_rich_root():
785
self._add_text_to_weave(change[0], [], heads, None)
786
elif kind == 'tree-reference':
787
if not self.repository._format.supports_tree_reference:
788
# This isn't quite sane as an error, but we shouldn't
789
# ever see this code path in practice: tree's don't
790
# permit references when the repo doesn't support tree
792
raise errors.UnsupportedOperation(tree.add_reference,
794
entry.reference_revision = \
795
tree.get_reference_revision(change[0])
796
if (carry_over_possible and
797
parent_entry.reference_revision == reference_revision):
800
self._add_text_to_weave(change[0], [], heads, None)
802
raise AssertionError('unknown kind %r' % kind)
804
entry.revision = modified_rev
806
entry.revision = parent_entry.revision
809
new_path = change[1][1]
810
inv_delta.append((change[1][0], new_path, change[0], entry))
813
self.new_inventory = None
815
self._any_changes = True
817
# housekeeping root entry changes do not affect no-change commits.
818
self._require_root_change(tree)
819
self.basis_delta_revision = basis_revision_id
821
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
822
# Note: as we read the content directly from the tree, we know its not
823
# been turned into unicode or badly split - but a broken tree
824
# implementation could give us bad output from readlines() so this is
825
# not a guarantee of safety. What would be better is always checking
826
# the content during test suite execution. RBC 20070912
827
parent_keys = tuple((file_id, parent) for parent in parents)
828
return self.repository.texts.add_lines(
829
(file_id, self._new_revision_id), parent_keys, new_lines,
830
nostore_sha=nostore_sha, random_id=self.random_revid,
831
check_content=False)[0:2]
834
class RootCommitBuilder(CommitBuilder):
835
"""This commitbuilder actually records the root id"""
837
# the root entry gets versioned properly by this builder.
838
_versioned_root = True
840
def _check_root(self, ie, parent_invs, tree):
841
"""Helper for record_entry_contents.
843
:param ie: An entry being added.
844
:param parent_invs: The inventories of the parent revisions of the
846
:param tree: The tree that is being committed.
849
def _require_root_change(self, tree):
850
"""Enforce an appropriate root object change.
852
This is called once when record_iter_changes is called, if and only if
853
the root was not in the delta calculated by record_iter_changes.
855
:param tree: The tree which is being committed.
857
# versioned roots do not change unless the tree found a change.
241
860
######################################################################
245
class Repository(_RelockDebugMixin, controldir.ControlComponent):
863
class Repository(object):
246
864
"""Repository holding history for one or more branches.
248
866
The repository holds and retrieves historical information including
249
867
revisions and file history. It's normally accessed only by the Branch,
250
868
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.
870
The Repository builds on top of some byte storage facilies (the revisions,
871
signatures, inventories, texts and chk_bytes attributes) and a Transport,
872
which respectively provide byte storage and a means to access the (possibly
875
The byte storage facilities are addressed via tuples, which we refer to
876
as 'keys' throughout the code base. Revision_keys, inventory_keys and
877
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
878
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
879
byte string made up of a hash identifier and a hash value.
880
We use this interface because it allows low friction with the underlying
881
code that implements disk indices, network encoding and other parts of
884
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
885
the serialised revisions for the repository. This can be used to obtain
886
revision graph information or to access raw serialised revisions.
887
The result of trying to insert data into the repository via this store
888
is undefined: it should be considered read-only except for implementors
890
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
891
the serialised signatures for the repository. This can be used to
892
obtain access to raw serialised signatures. The result of trying to
893
insert data into the repository via this store is undefined: it should
894
be considered read-only except for implementors of repositories.
895
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
896
the serialised inventories for the repository. This can be used to
897
obtain unserialised inventories. The result of trying to insert data
898
into the repository via this store is undefined: it should be
899
considered read-only except for implementors of repositories.
900
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
901
texts of files and directories for the repository. This can be used to
902
obtain file texts or file graphs. Note that Repository.iter_file_bytes
903
is usually a better interface for accessing file texts.
904
The result of trying to insert data into the repository via this store
905
is undefined: it should be considered read-only except for implementors
907
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
908
any data the repository chooses to store or have indexed by its hash.
909
The result of trying to insert data into the repository via this store
910
is undefined: it should be considered read-only except for implementors
912
:ivar _transport: Transport for file access to repository, typically
913
pointing to .bzr/repository.
916
# What class to use for a CommitBuilder. Often its simpler to change this
917
# in a Repository class subclass rather than to override
918
# get_commit_builder.
919
_commit_builder_class = CommitBuilder
920
# The search regex used by xml based repositories to determine what things
921
# where changed in a single commit.
922
_file_ids_altered_regex = lazy_regex.lazy_compile(
923
r'file_id="(?P<file_id>[^"]+)"'
924
r'.* revision="(?P<revision_id>[^"]+)"'
256
927
def abort_write_group(self, suppress_errors=False):
257
928
"""Commit the contents accrued within the current write group.
313
988
return InterRepository._assert_same_model(self, repository)
990
def add_inventory(self, revision_id, inv, parents):
991
"""Add the inventory inv to the repository as revision_id.
993
:param parents: The revision ids of the parents that revision_id
994
is known to have and are in the repository already.
996
:returns: The validator(which is a sha1 digest, though what is sha'd is
997
repository format specific) of the serialized inventory.
999
if not self.is_in_write_group():
1000
raise AssertionError("%r not in write group" % (self,))
1001
_mod_revision.check_not_reserved_id(revision_id)
1002
if not (inv.revision_id is None or inv.revision_id == revision_id):
1003
raise AssertionError(
1004
"Mismatch between inventory revision"
1005
" id and insertion revid (%r, %r)"
1006
% (inv.revision_id, revision_id))
1007
if inv.root is None:
1008
raise AssertionError()
1009
return self._add_inventory_checked(revision_id, inv, parents)
1011
def _add_inventory_checked(self, revision_id, inv, parents):
1012
"""Add inv to the repository after checking the inputs.
1014
This function can be overridden to allow different inventory styles.
1016
:seealso: add_inventory, for the contract.
1018
inv_lines = self._serialise_inventory_to_lines(inv)
1019
return self._inventory_add_lines(revision_id, parents,
1020
inv_lines, check_content=False)
1022
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1023
parents, basis_inv=None, propagate_caches=False):
1024
"""Add a new inventory expressed as a delta against another revision.
1026
:param basis_revision_id: The inventory id the delta was created
1027
against. (This does not have to be a direct parent.)
1028
:param delta: The inventory delta (see Inventory.apply_delta for
1030
:param new_revision_id: The revision id that the inventory is being
1032
:param parents: The revision ids of the parents that revision_id is
1033
known to have and are in the repository already. These are supplied
1034
for repositories that depend on the inventory graph for revision
1035
graph access, as well as for those that pun ancestry with delta
1037
:param basis_inv: The basis inventory if it is already known,
1039
:param propagate_caches: If True, the caches for this inventory are
1040
copied to and updated for the result if possible.
1042
:returns: (validator, new_inv)
1043
The validator(which is a sha1 digest, though what is sha'd is
1044
repository format specific) of the serialized inventory, and the
1045
resulting inventory.
1047
if not self.is_in_write_group():
1048
raise AssertionError("%r not in write group" % (self,))
1049
_mod_revision.check_not_reserved_id(new_revision_id)
1050
basis_tree = self.revision_tree(basis_revision_id)
1051
basis_tree.lock_read()
1053
# Note that this mutates the inventory of basis_tree, which not all
1054
# inventory implementations may support: A better idiom would be to
1055
# return a new inventory, but as there is no revision tree cache in
1056
# repository this is safe for now - RBC 20081013
1057
if basis_inv is None:
1058
basis_inv = basis_tree.inventory
1059
basis_inv.apply_delta(delta)
1060
basis_inv.revision_id = new_revision_id
1061
return (self.add_inventory(new_revision_id, basis_inv, parents),
1066
def _inventory_add_lines(self, revision_id, parents, lines,
1067
check_content=True):
1068
"""Store lines in inv_vf and return the sha1 of the inventory."""
1069
parents = [(parent,) for parent in parents]
1070
return self.inventories.add_lines((revision_id,), parents, lines,
1071
check_content=check_content)[0]
1073
def add_revision(self, revision_id, rev, inv=None, config=None):
1074
"""Add rev to the revision store as revision_id.
1076
:param revision_id: the revision id to use.
1077
:param rev: The revision object.
1078
:param inv: The inventory for the revision. if None, it will be looked
1079
up in the inventory storer
1080
:param config: If None no digital signature will be created.
1081
If supplied its signature_needed method will be used
1082
to determine if a signature should be made.
1084
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1086
_mod_revision.check_not_reserved_id(revision_id)
1087
if config is not None and config.signature_needed():
1089
inv = self.get_inventory(revision_id)
1090
plaintext = Testament(rev, inv).as_short_text()
1091
self.store_revision_signature(
1092
gpg.GPGStrategy(config), plaintext, revision_id)
1093
# check inventory present
1094
if not self.inventories.get_parent_map([(revision_id,)]):
1096
raise errors.WeaveRevisionNotPresent(revision_id,
1099
# yes, this is not suitable for adding with ghosts.
1100
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1103
key = (revision_id,)
1104
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1105
self._add_revision(rev)
1107
def _add_revision(self, revision):
1108
text = self._serializer.write_revision_to_string(revision)
1109
key = (revision.revision_id,)
1110
parents = tuple((parent,) for parent in revision.parent_ids)
1111
self.revisions.add_lines(key, parents, osutils.split_lines(text))
315
1113
def all_revision_ids(self):
316
1114
"""Returns a list of all the revision ids in the repository.
921
1737
signature = gpg_strategy.sign(plaintext)
922
1738
self.add_signature_text(revision_id, signature)
924
1741
def add_signature_text(self, revision_id, signature):
925
"""Store a signature text for a revision.
927
:param revision_id: Revision id of the revision
928
:param signature: Signature text.
930
raise NotImplementedError(self.add_signature_text)
932
def _find_parent_ids_of_revisions(self, revision_ids):
933
"""Find all parent ids that are mentioned in the revision graph.
935
:return: set of revisions that are parents of revision_ids which are
936
not part of revision_ids themselves
938
parent_map = self.get_parent_map(revision_ids)
940
map(parent_ids.update, parent_map.itervalues())
941
parent_ids.difference_update(revision_ids)
942
parent_ids.discard(_mod_revision.NULL_REVISION)
945
def fileids_altered_by_revision_ids(self, revision_ids):
1742
self.signatures.add_lines((revision_id,), (),
1743
osutils.split_lines(signature))
1745
def find_text_key_references(self):
1746
"""Find the text key references within the repository.
1748
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1749
to whether they were referred to by the inventory of the
1750
revision_id that they contain. The inventory texts from all present
1751
revision ids are assessed to generate this report.
1753
revision_keys = self.revisions.keys()
1754
w = self.inventories
1755
pb = ui.ui_factory.nested_progress_bar()
1757
return self._find_text_key_references_from_xml_inventory_lines(
1758
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1762
def _find_text_key_references_from_xml_inventory_lines(self,
1764
"""Core routine for extracting references to texts from inventories.
1766
This performs the translation of xml lines to revision ids.
1768
:param line_iterator: An iterator of lines, origin_version_id
1769
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1770
to whether they were referred to by the inventory of the
1771
revision_id that they contain. Note that if that revision_id was
1772
not part of the line_iterator's output then False will be given -
1773
even though it may actually refer to that key.
1775
if not self._serializer.support_altered_by_hack:
1776
raise AssertionError(
1777
"_find_text_key_references_from_xml_inventory_lines only "
1778
"supported for branches which store inventory as unnested xml"
1779
", not on %r" % self)
1782
# this code needs to read every new line in every inventory for the
1783
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1784
# not present in one of those inventories is unnecessary but not
1785
# harmful because we are filtering by the revision id marker in the
1786
# inventory lines : we only select file ids altered in one of those
1787
# revisions. We don't need to see all lines in the inventory because
1788
# only those added in an inventory in rev X can contain a revision=X
1790
unescape_revid_cache = {}
1791
unescape_fileid_cache = {}
1793
# jam 20061218 In a big fetch, this handles hundreds of thousands
1794
# of lines, so it has had a lot of inlining and optimizing done.
1795
# Sorry that it is a little bit messy.
1796
# Move several functions to be local variables, since this is a long
1798
search = self._file_ids_altered_regex.search
1799
unescape = _unescape_xml
1800
setdefault = result.setdefault
1801
for line, line_key in line_iterator:
1802
match = search(line)
1805
# One call to match.group() returning multiple items is quite a
1806
# bit faster than 2 calls to match.group() each returning 1
1807
file_id, revision_id = match.group('file_id', 'revision_id')
1809
# Inlining the cache lookups helps a lot when you make 170,000
1810
# lines and 350k ids, versus 8.4 unique ids.
1811
# Using a cache helps in 2 ways:
1812
# 1) Avoids unnecessary decoding calls
1813
# 2) Re-uses cached strings, which helps in future set and
1815
# (2) is enough that removing encoding entirely along with
1816
# the cache (so we are using plain strings) results in no
1817
# performance improvement.
1819
revision_id = unescape_revid_cache[revision_id]
1821
unescaped = unescape(revision_id)
1822
unescape_revid_cache[revision_id] = unescaped
1823
revision_id = unescaped
1825
# Note that unconditionally unescaping means that we deserialise
1826
# every fileid, which for general 'pull' is not great, but we don't
1827
# really want to have some many fulltexts that this matters anyway.
1830
file_id = unescape_fileid_cache[file_id]
1832
unescaped = unescape(file_id)
1833
unescape_fileid_cache[file_id] = unescaped
1836
key = (file_id, revision_id)
1837
setdefault(key, False)
1838
if revision_id == line_key[-1]:
1842
def _inventory_xml_lines_for_keys(self, keys):
1843
"""Get a line iterator of the sort needed for findind references.
1845
Not relevant for non-xml inventory repositories.
1847
Ghosts in revision_keys are ignored.
1849
:param revision_keys: The revision keys for the inventories to inspect.
1850
:return: An iterator over (inventory line, revid) for the fulltexts of
1851
all of the xml inventories specified by revision_keys.
1853
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1854
for record in stream:
1855
if record.storage_kind != 'absent':
1856
chunks = record.get_bytes_as('chunked')
1857
revid = record.key[-1]
1858
lines = osutils.chunks_to_lines(chunks)
1862
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1864
"""Helper routine for fileids_altered_by_revision_ids.
1866
This performs the translation of xml lines to revision ids.
1868
:param line_iterator: An iterator of lines, origin_version_id
1869
:param revision_ids: The revision ids to filter for. This should be a
1870
set or other type which supports efficient __contains__ lookups, as
1871
the revision id from each parsed line will be looked up in the
1872
revision_ids filter.
1873
:return: a dictionary mapping altered file-ids to an iterable of
1874
revision_ids. Each altered file-ids has the exact revision_ids that
1875
altered it listed explicitly.
1877
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1878
line_iterator).iterkeys())
1879
# Note that revision_ids are revision keys.
1880
parent_maps = self.revisions.get_parent_map(revision_ids)
1882
map(parents.update, parent_maps.itervalues())
1883
parents.difference_update(revision_ids)
1884
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1885
self._inventory_xml_lines_for_keys(parents)))
1886
new_keys = seen - parent_seen
1888
setdefault = result.setdefault
1889
for key in new_keys:
1890
setdefault(key[0], set()).add(key[-1])
1893
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
946
1894
"""Find the file ids and versions affected by revisions.
948
1896
:param revisions: an iterable containing revision ids.
1897
:param _inv_weave: The inventory weave from this repository or None.
1898
If None, the inventory weave will be opened automatically.
949
1899
:return: a dictionary mapping altered file-ids to an iterable of
950
revision_ids. Each altered file-ids has the exact revision_ids
951
that altered it listed explicitly.
1900
revision_ids. Each altered file-ids has the exact revision_ids that
1901
altered it listed explicitly.
953
raise NotImplementedError(self.fileids_altered_by_revision_ids)
1903
selected_keys = set((revid,) for revid in revision_ids)
1904
w = _inv_weave or self.inventories
1905
pb = ui.ui_factory.nested_progress_bar()
1907
return self._find_file_ids_from_xml_inventory_lines(
1908
w.iter_lines_added_or_present_in_keys(
1909
selected_keys, pb=pb),
955
1914
def iter_files_bytes(self, desired_files):
956
1915
"""Iterate through file versions.
963
1922
uniquely identify the file version in the caller's context. (Examples:
964
1923
an index number or a TreeTransform trans_id.)
1925
bytes_iterator is an iterable of bytestrings for the file. The
1926
kind of iterable and length of the bytestrings are unspecified, but for
1927
this implementation, it is a list of bytes produced by
1928
VersionedFile.get_record_stream().
966
1930
:param desired_files: a list of (file_id, revision_id, identifier)
969
raise NotImplementedError(self.iter_files_bytes)
971
def get_rev_id_for_revno(self, revno, known_pair):
972
"""Return the revision id of a revno, given a later (revno, revid)
973
pair in the same history.
975
:return: if found (True, revid). If the available history ran out
976
before reaching the revno, then this returns
977
(False, (closest_revno, closest_revid)).
979
known_revno, known_revid = known_pair
980
partial_history = [known_revid]
981
distance_from_known = known_revno - revno
982
if distance_from_known < 0:
984
'requested revno (%d) is later than given known revno (%d)'
985
% (revno, known_revno))
988
self, partial_history, stop_index=distance_from_known)
989
except errors.RevisionNotPresent, err:
990
if err.revision_id == known_revid:
991
# The start revision (known_revid) wasn't found.
993
# This is a stacked repository with no fallbacks, or a there's a
994
# left-hand ghost. Either way, even though the revision named in
995
# the error isn't in this repo, we know it's the next step in this
997
partial_history.append(err.revision_id)
998
if len(partial_history) <= distance_from_known:
999
# Didn't find enough history to get a revid for the revno.
1000
earliest_revno = known_revno - len(partial_history) + 1
1001
return (False, (earliest_revno, partial_history[-1]))
1002
if len(partial_history) - 1 > distance_from_known:
1003
raise AssertionError('_iter_for_revno returned too much history')
1004
return (True, partial_history[-1])
1006
@symbol_versioning.deprecated_method(symbol_versioning.deprecated_in((2, 4, 0)))
1934
for file_id, revision_id, callable_data in desired_files:
1935
text_keys[(file_id, revision_id)] = callable_data
1936
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1937
if record.storage_kind == 'absent':
1938
raise errors.RevisionNotPresent(record.key, self)
1939
yield text_keys[record.key], record.get_bytes_as('chunked')
1941
def _generate_text_key_index(self, text_key_references=None,
1943
"""Generate a new text key index for the repository.
1945
This is an expensive function that will take considerable time to run.
1947
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1948
list of parents, also text keys. When a given key has no parents,
1949
the parents list will be [NULL_REVISION].
1951
# All revisions, to find inventory parents.
1952
if ancestors is None:
1953
graph = self.get_graph()
1954
ancestors = graph.get_parent_map(self.all_revision_ids())
1955
if text_key_references is None:
1956
text_key_references = self.find_text_key_references()
1957
pb = ui.ui_factory.nested_progress_bar()
1959
return self._do_generate_text_key_index(ancestors,
1960
text_key_references, pb)
1964
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1965
"""Helper for _generate_text_key_index to avoid deep nesting."""
1966
revision_order = tsort.topo_sort(ancestors)
1967
invalid_keys = set()
1969
for revision_id in revision_order:
1970
revision_keys[revision_id] = set()
1971
text_count = len(text_key_references)
1972
# a cache of the text keys to allow reuse; costs a dict of all the
1973
# keys, but saves a 2-tuple for every child of a given key.
1975
for text_key, valid in text_key_references.iteritems():
1977
invalid_keys.add(text_key)
1979
revision_keys[text_key[1]].add(text_key)
1980
text_key_cache[text_key] = text_key
1981
del text_key_references
1983
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1984
NULL_REVISION = _mod_revision.NULL_REVISION
1985
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1986
# too small for large or very branchy trees. However, for 55K path
1987
# trees, it would be easy to use too much memory trivially. Ideally we
1988
# could gauge this by looking at available real memory etc, but this is
1989
# always a tricky proposition.
1990
inventory_cache = lru_cache.LRUCache(10)
1991
batch_size = 10 # should be ~150MB on a 55K path tree
1992
batch_count = len(revision_order) / batch_size + 1
1994
pb.update("Calculating text parents", processed_texts, text_count)
1995
for offset in xrange(batch_count):
1996
to_query = revision_order[offset * batch_size:(offset + 1) *
2000
for rev_tree in self.revision_trees(to_query):
2001
revision_id = rev_tree.get_revision_id()
2002
parent_ids = ancestors[revision_id]
2003
for text_key in revision_keys[revision_id]:
2004
pb.update("Calculating text parents", processed_texts)
2005
processed_texts += 1
2006
candidate_parents = []
2007
for parent_id in parent_ids:
2008
parent_text_key = (text_key[0], parent_id)
2010
check_parent = parent_text_key not in \
2011
revision_keys[parent_id]
2013
# the parent parent_id is a ghost:
2014
check_parent = False
2015
# truncate the derived graph against this ghost.
2016
parent_text_key = None
2018
# look at the parent commit details inventories to
2019
# determine possible candidates in the per file graph.
2022
inv = inventory_cache[parent_id]
2024
inv = self.revision_tree(parent_id).inventory
2025
inventory_cache[parent_id] = inv
2027
parent_entry = inv[text_key[0]]
2028
except (KeyError, errors.NoSuchId):
2030
if parent_entry is not None:
2032
text_key[0], parent_entry.revision)
2034
parent_text_key = None
2035
if parent_text_key is not None:
2036
candidate_parents.append(
2037
text_key_cache[parent_text_key])
2038
parent_heads = text_graph.heads(candidate_parents)
2039
new_parents = list(parent_heads)
2040
new_parents.sort(key=lambda x:candidate_parents.index(x))
2041
if new_parents == []:
2042
new_parents = [NULL_REVISION]
2043
text_index[text_key] = new_parents
2045
for text_key in invalid_keys:
2046
text_index[text_key] = [NULL_REVISION]
2049
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2050
"""Get an iterable listing the keys of all the data introduced by a set
2053
The keys will be ordered so that the corresponding items can be safely
2054
fetched and inserted in that order.
2056
:returns: An iterable producing tuples of (knit-kind, file-id,
2057
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2058
'revisions'. file-id is None unless knit-kind is 'file'.
2060
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2063
for result in self._find_non_file_keys_to_fetch(revision_ids):
2066
def _find_file_keys_to_fetch(self, revision_ids, pb):
2067
# XXX: it's a bit weird to control the inventory weave caching in this
2068
# generator. Ideally the caching would be done in fetch.py I think. Or
2069
# maybe this generator should explicitly have the contract that it
2070
# should not be iterated until the previously yielded item has been
2072
inv_w = self.inventories
2074
# file ids that changed
2075
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2077
num_file_ids = len(file_ids)
2078
for file_id, altered_versions in file_ids.iteritems():
2080
pb.update("fetch texts", count, num_file_ids)
2082
yield ("file", file_id, altered_versions)
2084
def _find_non_file_keys_to_fetch(self, revision_ids):
2086
yield ("inventory", None, revision_ids)
2089
# XXX: Note ATM no callers actually pay attention to this return
2090
# instead they just use the list of revision ids and ignore
2091
# missing sigs. Consider removing this work entirely
2092
revisions_with_signatures = set(self.signatures.get_parent_map(
2093
[(r,) for r in revision_ids]))
2094
revisions_with_signatures = set(
2095
[r for (r,) in revisions_with_signatures])
2096
revisions_with_signatures.intersection_update(revision_ids)
2097
yield ("signatures", None, revisions_with_signatures)
2100
yield ("revisions", None, revision_ids)
2103
def get_inventory(self, revision_id):
2104
"""Get Inventory object by revision id."""
2105
return self.iter_inventories([revision_id]).next()
2107
def iter_inventories(self, revision_ids):
2108
"""Get many inventories by revision_ids.
2110
This will buffer some or all of the texts used in constructing the
2111
inventories in memory, but will only parse a single inventory at a
2114
:param revision_ids: The expected revision ids of the inventories.
2115
:return: An iterator of inventories.
2117
if ((None in revision_ids)
2118
or (_mod_revision.NULL_REVISION in revision_ids)):
2119
raise ValueError('cannot get null revision inventory')
2120
return self._iter_inventories(revision_ids)
2122
def _iter_inventories(self, revision_ids):
2123
"""single-document based inventory iteration."""
2124
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2125
yield self.deserialise_inventory(revision_id, text)
2127
def _iter_inventory_xmls(self, revision_ids):
2128
keys = [(revision_id,) for revision_id in revision_ids]
2129
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2131
for record in stream:
2132
if record.storage_kind != 'absent':
2133
text_chunks[record.key] = record.get_bytes_as('chunked')
2135
raise errors.NoSuchRevision(self, record.key)
2137
chunks = text_chunks.pop(key)
2138
yield ''.join(chunks), key[-1]
2140
def deserialise_inventory(self, revision_id, xml):
2141
"""Transform the xml into an inventory object.
2143
:param revision_id: The expected revision id of the inventory.
2144
:param xml: A serialised inventory.
2146
result = self._serializer.read_inventory_from_string(xml, revision_id,
2147
entry_cache=self._inventory_entry_cache)
2148
if result.revision_id != revision_id:
2149
raise AssertionError('revision id mismatch %s != %s' % (
2150
result.revision_id, revision_id))
2153
def serialise_inventory(self, inv):
2154
return self._serializer.write_inventory_to_string(inv)
2156
def _serialise_inventory_to_lines(self, inv):
2157
return self._serializer.write_inventory_to_lines(inv)
2159
def get_serializer_format(self):
2160
return self._serializer.format_num
2163
def get_inventory_xml(self, revision_id):
2164
"""Get inventory XML as a file object."""
2165
texts = self._iter_inventory_xmls([revision_id])
2167
text, revision_id = texts.next()
2168
except StopIteration:
2169
raise errors.HistoryMissing(self, 'inventory', revision_id)
2173
def get_inventory_sha1(self, revision_id):
2174
"""Return the sha1 hash of the inventory entry
2176
return self.get_revision(revision_id).inventory_sha1
1007
2178
def iter_reverse_revision_history(self, revision_id):
1008
2179
"""Iterate backwards through revision ids in the lefthand history
1289
2468
except UnicodeDecodeError:
1290
2469
raise errors.NonAsciiRevisionId(method, self)
2471
def revision_graph_can_have_wrong_parents(self):
2472
"""Is it possible for this repository to have a revision graph with
2475
If True, then this repository must also implement
2476
_find_inconsistent_revision_parents so that check and reconcile can
2477
check for inconsistencies before proceeding with other checks that may
2478
depend on the revision index being consistent.
2480
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2483
# remove these delegates a while after bzr 0.15
2484
def __make_delegated(name, from_module):
2485
def _deprecated_repository_forwarder():
2486
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2487
% (name, from_module),
2490
m = __import__(from_module, globals(), locals(), [name])
2492
return getattr(m, name)
2493
except AttributeError:
2494
raise AttributeError('module %s has no name %s'
2496
globals()[name] = _deprecated_repository_forwarder
2499
'AllInOneRepository',
2500
'WeaveMetaDirRepository',
2501
'PreSplitOutRepositoryFormat',
2502
'RepositoryFormat4',
2503
'RepositoryFormat5',
2504
'RepositoryFormat6',
2505
'RepositoryFormat7',
2507
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2511
'RepositoryFormatKnit',
2512
'RepositoryFormatKnit1',
2514
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2517
def install_revision(repository, rev, revision_tree):
2518
"""Install all revision data into a repository."""
2519
install_revisions(repository, [(rev, revision_tree, None)])
2522
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2523
"""Install all revision data into a repository.
2525
Accepts an iterable of revision, tree, signature tuples. The signature
2528
repository.start_write_group()
2530
inventory_cache = lru_cache.LRUCache(10)
2531
for n, (revision, revision_tree, signature) in enumerate(iterable):
2532
_install_revision(repository, revision, revision_tree, signature,
2535
pb.update('Transferring revisions', n + 1, num_revisions)
2537
repository.abort_write_group()
2540
repository.commit_write_group()
2543
def _install_revision(repository, rev, revision_tree, signature,
2545
"""Install all revision data into a repository."""
2546
present_parents = []
2548
for p_id in rev.parent_ids:
2549
if repository.has_revision(p_id):
2550
present_parents.append(p_id)
2551
parent_trees[p_id] = repository.revision_tree(p_id)
2553
parent_trees[p_id] = repository.revision_tree(
2554
_mod_revision.NULL_REVISION)
2556
inv = revision_tree.inventory
2557
entries = inv.iter_entries()
2558
# backwards compatibility hack: skip the root id.
2559
if not repository.supports_rich_root():
2560
path, root = entries.next()
2561
if root.revision != rev.revision_id:
2562
raise errors.IncompatibleRevision(repr(repository))
2564
for path, ie in entries:
2565
text_keys[(ie.file_id, ie.revision)] = ie
2566
text_parent_map = repository.texts.get_parent_map(text_keys)
2567
missing_texts = set(text_keys) - set(text_parent_map)
2568
# Add the texts that are not already present
2569
for text_key in missing_texts:
2570
ie = text_keys[text_key]
2572
# FIXME: TODO: The following loop overlaps/duplicates that done by
2573
# commit to determine parents. There is a latent/real bug here where
2574
# the parents inserted are not those commit would do - in particular
2575
# they are not filtered by heads(). RBC, AB
2576
for revision, tree in parent_trees.iteritems():
2577
if ie.file_id not in tree:
2579
parent_id = tree.inventory[ie.file_id].revision
2580
if parent_id in text_parents:
2582
text_parents.append((ie.file_id, parent_id))
2583
lines = revision_tree.get_file(ie.file_id).readlines()
2584
repository.texts.add_lines(text_key, text_parents, lines)
2586
# install the inventory
2587
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2588
# Cache this inventory
2589
inventory_cache[rev.revision_id] = inv
2591
basis_inv = inventory_cache[rev.parent_ids[0]]
2593
repository.add_inventory(rev.revision_id, inv, present_parents)
2595
delta = inv._make_delta(basis_inv)
2596
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2597
rev.revision_id, present_parents)
2599
repository.add_inventory(rev.revision_id, inv, present_parents)
2600
except errors.RevisionAlreadyPresent:
2602
if signature is not None:
2603
repository.add_signature_text(rev.revision_id, signature)
2604
repository.add_revision(rev.revision_id, rev, inv)
1293
2607
class MetaDirRepository(Repository):
1294
2608
"""Repositories in the new meta-dir layout.
1733
3046
:param revision_id: if None all content is copied, if NULL_REVISION no
1734
3047
content is copied.
3048
:param pb: optional progress bar to use for progress reports. If not
3049
provided a default one will be created.
1737
raise NotImplementedError(self.fetch)
3052
from bzrlib.fetch import RepoFetcher
3053
f = RepoFetcher(to_repository=self.target,
3054
from_repository=self.source,
3055
last_revision=revision_id,
3056
fetch_spec=fetch_spec,
3057
pb=pb, find_ghosts=find_ghosts)
3059
def _walk_to_common_revisions(self, revision_ids):
3060
"""Walk out from revision_ids in source to revisions target has.
3062
:param revision_ids: The start point for the search.
3063
:return: A set of revision ids.
3065
target_graph = self.target.get_graph()
3066
revision_ids = frozenset(revision_ids)
3067
# Fast path for the case where all the revisions are already in the
3069
# (Although this does incur an extra round trip for the
3070
# fairly common case where the target doesn't already have the revision
3072
if set(target_graph.get_parent_map(revision_ids)) == revision_ids:
3073
return graph.SearchResult(revision_ids, set(), 0, set())
3074
missing_revs = set()
3075
source_graph = self.source.get_graph()
3076
# ensure we don't pay silly lookup costs.
3077
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3078
null_set = frozenset([_mod_revision.NULL_REVISION])
3079
searcher_exhausted = False
3083
# Iterate the searcher until we have enough next_revs
3084
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3086
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3087
next_revs.update(next_revs_part)
3088
ghosts.update(ghosts_part)
3089
except StopIteration:
3090
searcher_exhausted = True
3092
# If there are ghosts in the source graph, and the caller asked for
3093
# them, make sure that they are present in the target.
3094
# We don't care about other ghosts as we can't fetch them and
3095
# haven't been asked to.
3096
ghosts_to_check = set(revision_ids.intersection(ghosts))
3097
revs_to_get = set(next_revs).union(ghosts_to_check)
3099
have_revs = set(target_graph.get_parent_map(revs_to_get))
3100
# we always have NULL_REVISION present.
3101
have_revs = have_revs.union(null_set)
3102
# Check if the target is missing any ghosts we need.
3103
ghosts_to_check.difference_update(have_revs)
3105
# One of the caller's revision_ids is a ghost in both the
3106
# source and the target.
3107
raise errors.NoSuchRevision(
3108
self.source, ghosts_to_check.pop())
3109
missing_revs.update(next_revs - have_revs)
3110
# Because we may have walked past the original stop point, make
3111
# sure everything is stopped
3112
stop_revs = searcher.find_seen_ancestors(have_revs)
3113
searcher.stop_searching_any(stop_revs)
3114
if searcher_exhausted:
3116
return searcher.get_result()
1739
3118
@needs_read_lock
1740
def search_missing_revision_ids(self,
1741
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
1742
find_ghosts=True, revision_ids=None, if_present_ids=None,
3119
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1744
3120
"""Return the revision ids that source has that target does not.
1746
3122
:param revision_id: only return revision ids included by this
1748
:param revision_ids: return revision ids included by these
1749
revision_ids. NoSuchRevision will be raised if any of these
1750
revisions are not present.
1751
:param if_present_ids: like revision_ids, but will not cause
1752
NoSuchRevision if any of these are absent, instead they will simply
1753
not be in the result. This is useful for e.g. finding revisions
1754
to fetch for tags, which may reference absent revisions.
1755
3124
:param find_ghosts: If True find missing revisions in deep history
1756
3125
rather than just finding the surface difference.
1757
:param limit: Maximum number of revisions to return, topologically
1759
3126
:return: A bzrlib.graph.SearchResult.
1761
raise NotImplementedError(self.search_missing_revision_ids)
3128
# stop searching at found target revisions.
3129
if not find_ghosts and revision_id is not None:
3130
return self._walk_to_common_revisions([revision_id])
3131
# generic, possibly worst case, slow code path.
3132
target_ids = set(self.target.all_revision_ids())
3133
if revision_id is not None:
3134
source_ids = self.source.get_ancestry(revision_id)
3135
if source_ids[0] is not None:
3136
raise AssertionError()
3139
source_ids = self.source.all_revision_ids()
3140
result_set = set(source_ids).difference(target_ids)
3141
return self.source.revision_ids_to_search_result(result_set)
1764
3144
def _same_model(source, target):
1785
3165
"different serializers")
3168
class InterSameDataRepository(InterRepository):
3169
"""Code for converting between repositories that represent the same data.
3171
Data format and model must match for this to work.
3175
def _get_repo_format_to_test(self):
3176
"""Repository format for testing with.
3178
InterSameData can pull from subtree to subtree and from non-subtree to
3179
non-subtree, so we test this with the richest repository format.
3181
from bzrlib.repofmt import knitrepo
3182
return knitrepo.RepositoryFormatKnit3()
3185
def is_compatible(source, target):
3186
return InterRepository._same_model(source, target)
3189
class InterWeaveRepo(InterSameDataRepository):
3190
"""Optimised code paths between Weave based repositories.
3192
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3193
implemented lazy inter-object optimisation.
3197
def _get_repo_format_to_test(self):
3198
from bzrlib.repofmt import weaverepo
3199
return weaverepo.RepositoryFormat7()
3202
def is_compatible(source, target):
3203
"""Be compatible with known Weave formats.
3205
We don't test for the stores being of specific types because that
3206
could lead to confusing results, and there is no need to be
3209
from bzrlib.repofmt.weaverepo import (
3215
return (isinstance(source._format, (RepositoryFormat5,
3217
RepositoryFormat7)) and
3218
isinstance(target._format, (RepositoryFormat5,
3220
RepositoryFormat7)))
3221
except AttributeError:
3225
def copy_content(self, revision_id=None):
3226
"""See InterRepository.copy_content()."""
3227
# weave specific optimised path:
3229
self.target.set_make_working_trees(self.source.make_working_trees())
3230
except (errors.RepositoryUpgradeRequired, NotImplemented):
3232
# FIXME do not peek!
3233
if self.source._transport.listable():
3234
pb = ui.ui_factory.nested_progress_bar()
3236
self.target.texts.insert_record_stream(
3237
self.source.texts.get_record_stream(
3238
self.source.texts.keys(), 'topological', False))
3239
pb.update('copying inventory', 0, 1)
3240
self.target.inventories.insert_record_stream(
3241
self.source.inventories.get_record_stream(
3242
self.source.inventories.keys(), 'topological', False))
3243
self.target.signatures.insert_record_stream(
3244
self.source.signatures.get_record_stream(
3245
self.source.signatures.keys(),
3247
self.target.revisions.insert_record_stream(
3248
self.source.revisions.get_record_stream(
3249
self.source.revisions.keys(),
3250
'topological', True))
3254
self.target.fetch(self.source, revision_id=revision_id)
3257
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3258
"""See InterRepository.missing_revision_ids()."""
3259
# we want all revisions to satisfy revision_id in source.
3260
# but we don't want to stat every file here and there.
3261
# we want then, all revisions other needs to satisfy revision_id
3262
# checked, but not those that we have locally.
3263
# so the first thing is to get a subset of the revisions to
3264
# satisfy revision_id in source, and then eliminate those that
3265
# we do already have.
3266
# this is slow on high latency connection to self, but as this
3267
# disk format scales terribly for push anyway due to rewriting
3268
# inventory.weave, this is considered acceptable.
3270
if revision_id is not None:
3271
source_ids = self.source.get_ancestry(revision_id)
3272
if source_ids[0] is not None:
3273
raise AssertionError()
3276
source_ids = self.source._all_possible_ids()
3277
source_ids_set = set(source_ids)
3278
# source_ids is the worst possible case we may need to pull.
3279
# now we want to filter source_ids against what we actually
3280
# have in target, but don't try to check for existence where we know
3281
# we do not have a revision as that would be pointless.
3282
target_ids = set(self.target._all_possible_ids())
3283
possibly_present_revisions = target_ids.intersection(source_ids_set)
3284
actually_present_revisions = set(
3285
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3286
required_revisions = source_ids_set.difference(actually_present_revisions)
3287
if revision_id is not None:
3288
# we used get_ancestry to determine source_ids then we are assured all
3289
# revisions referenced are present as they are installed in topological order.
3290
# and the tip revision was validated by get_ancestry.
3291
result_set = required_revisions
3293
# if we just grabbed the possibly available ids, then
3294
# we only have an estimate of whats available and need to validate
3295
# that against the revision records.
3297
self.source._eliminate_revisions_not_present(required_revisions))
3298
return self.source.revision_ids_to_search_result(result_set)
3301
class InterKnitRepo(InterSameDataRepository):
3302
"""Optimised code paths between Knit based repositories."""
3305
def _get_repo_format_to_test(self):
3306
from bzrlib.repofmt import knitrepo
3307
return knitrepo.RepositoryFormatKnit1()
3310
def is_compatible(source, target):
3311
"""Be compatible with known Knit formats.
3313
We don't test for the stores being of specific types because that
3314
could lead to confusing results, and there is no need to be
3317
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3319
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3320
isinstance(target._format, RepositoryFormatKnit))
3321
except AttributeError:
3323
return are_knits and InterRepository._same_model(source, target)
3326
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3327
"""See InterRepository.missing_revision_ids()."""
3328
if revision_id is not None:
3329
source_ids = self.source.get_ancestry(revision_id)
3330
if source_ids[0] is not None:
3331
raise AssertionError()
3334
source_ids = self.source.all_revision_ids()
3335
source_ids_set = set(source_ids)
3336
# source_ids is the worst possible case we may need to pull.
3337
# now we want to filter source_ids against what we actually
3338
# have in target, but don't try to check for existence where we know
3339
# we do not have a revision as that would be pointless.
3340
target_ids = set(self.target.all_revision_ids())
3341
possibly_present_revisions = target_ids.intersection(source_ids_set)
3342
actually_present_revisions = set(
3343
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3344
required_revisions = source_ids_set.difference(actually_present_revisions)
3345
if revision_id is not None:
3346
# we used get_ancestry to determine source_ids then we are assured all
3347
# revisions referenced are present as they are installed in topological order.
3348
# and the tip revision was validated by get_ancestry.
3349
result_set = required_revisions
3351
# if we just grabbed the possibly available ids, then
3352
# we only have an estimate of whats available and need to validate
3353
# that against the revision records.
3355
self.source._eliminate_revisions_not_present(required_revisions))
3356
return self.source.revision_ids_to_search_result(result_set)
3359
class InterPackRepo(InterSameDataRepository):
3360
"""Optimised code paths between Pack based repositories."""
3363
def _get_repo_format_to_test(self):
3364
from bzrlib.repofmt import pack_repo
3365
return pack_repo.RepositoryFormatKnitPack6RichRoot()
3368
def is_compatible(source, target):
3369
"""Be compatible with known Pack formats.
3371
We don't test for the stores being of specific types because that
3372
could lead to confusing results, and there is no need to be
3375
InterPackRepo does not support CHK based repositories.
3377
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3378
from bzrlib.repofmt.groupcompress_repo import RepositoryFormatCHK1
3380
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3381
isinstance(target._format, RepositoryFormatPack))
3382
not_packs = (isinstance(source._format, RepositoryFormatCHK1) or
3383
isinstance(target._format, RepositoryFormatCHK1))
3384
except AttributeError:
3386
if not_packs or not are_packs:
3388
return InterRepository._same_model(source, target)
3391
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3393
"""See InterRepository.fetch()."""
3394
if (len(self.source._fallback_repositories) > 0 or
3395
len(self.target._fallback_repositories) > 0):
3396
# The pack layer is not aware of fallback repositories, so when
3397
# fetching from a stacked repository or into a stacked repository
3398
# we use the generic fetch logic which uses the VersionedFiles
3399
# attributes on repository.
3400
from bzrlib.fetch import RepoFetcher
3401
fetcher = RepoFetcher(self.target, self.source, revision_id,
3402
pb, find_ghosts, fetch_spec=fetch_spec)
3403
if fetch_spec is not None:
3404
if len(list(fetch_spec.heads)) != 1:
3405
raise AssertionError(
3406
"InterPackRepo.fetch doesn't support "
3407
"fetching multiple heads yet.")
3408
revision_id = list(fetch_spec.heads)[0]
3410
if revision_id is None:
3412
# everything to do - use pack logic
3413
# to fetch from all packs to one without
3414
# inventory parsing etc, IFF nothing to be copied is in the target.
3416
source_revision_ids = frozenset(self.source.all_revision_ids())
3417
revision_ids = source_revision_ids - \
3418
frozenset(self.target.get_parent_map(source_revision_ids))
3419
revision_keys = [(revid,) for revid in revision_ids]
3420
index = self.target._pack_collection.revision_index.combined_index
3421
present_revision_ids = set(item[1][0] for item in
3422
index.iter_entries(revision_keys))
3423
revision_ids = set(revision_ids) - present_revision_ids
3424
# implementing the TODO will involve:
3425
# - detecting when all of a pack is selected
3426
# - avoiding as much as possible pre-selection, so the
3427
# more-core routines such as create_pack_from_packs can filter in
3428
# a just-in-time fashion. (though having a HEADS list on a
3429
# repository might make this a lot easier, because we could
3430
# sensibly detect 'new revisions' without doing a full index scan.
3431
elif _mod_revision.is_null(revision_id):
3436
revision_ids = self.search_missing_revision_ids(revision_id,
3437
find_ghosts=find_ghosts).get_keys()
3438
except errors.NoSuchRevision:
3439
raise errors.InstallFailed([revision_id])
3440
if len(revision_ids) == 0:
3442
return self._pack(self.source, self.target, revision_ids)
3444
def _pack(self, source, target, revision_ids):
3445
from bzrlib.repofmt.pack_repo import Packer
3446
packs = source._pack_collection.all_packs()
3447
pack = Packer(self.target._pack_collection, packs, '.fetch',
3448
revision_ids).pack()
3449
if pack is not None:
3450
self.target._pack_collection._save_pack_names()
3451
copied_revs = pack.get_revision_count()
3452
# Trigger an autopack. This may duplicate effort as we've just done
3453
# a pack creation, but for now it is simpler to think about as
3454
# 'upload data, then repack if needed'.
3455
self.target._pack_collection.autopack()
3456
return (copied_revs, [])
3461
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3462
"""See InterRepository.missing_revision_ids().
3464
:param find_ghosts: Find ghosts throughout the ancestry of
3467
if not find_ghosts and revision_id is not None:
3468
return self._walk_to_common_revisions([revision_id])
3469
elif revision_id is not None:
3470
# Find ghosts: search for revisions pointing from one repository to
3471
# the other, and vice versa, anywhere in the history of revision_id.
3472
graph = self.target.get_graph(other_repository=self.source)
3473
searcher = graph._make_breadth_first_searcher([revision_id])
3477
next_revs, ghosts = searcher.next_with_ghosts()
3478
except StopIteration:
3480
if revision_id in ghosts:
3481
raise errors.NoSuchRevision(self.source, revision_id)
3482
found_ids.update(next_revs)
3483
found_ids.update(ghosts)
3484
found_ids = frozenset(found_ids)
3485
# Double query here: should be able to avoid this by changing the
3486
# graph api further.
3487
result_set = found_ids - frozenset(
3488
self.target.get_parent_map(found_ids))
3490
source_ids = self.source.all_revision_ids()
3491
# source_ids is the worst possible case we may need to pull.
3492
# now we want to filter source_ids against what we actually
3493
# have in target, but don't try to check for existence where we know
3494
# we do not have a revision as that would be pointless.
3495
target_ids = set(self.target.all_revision_ids())
3496
result_set = set(source_ids).difference(target_ids)
3497
return self.source.revision_ids_to_search_result(result_set)
3500
class InterDifferingSerializer(InterRepository):
3503
def _get_repo_format_to_test(self):
3507
def is_compatible(source, target):
3508
"""Be compatible with Knit2 source and Knit3 target"""
3509
# This is redundant with format.check_conversion_target(), however that
3510
# raises an exception, and we just want to say "False" as in we won't
3511
# support converting between these formats.
3512
if source.supports_rich_root() and not target.supports_rich_root():
3514
if (source._format.supports_tree_reference
3515
and not target._format.supports_tree_reference):
3519
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3520
"""Get the best delta and base for this revision.
3522
:return: (basis_id, delta)
3524
possible_trees = [(parent_id, cache[parent_id])
3525
for parent_id in parent_ids
3526
if parent_id in cache]
3527
if len(possible_trees) == 0:
3528
# There either aren't any parents, or the parents aren't in the
3529
# cache, so just use the last converted tree
3530
possible_trees.append((basis_id, cache[basis_id]))
3532
for basis_id, basis_tree in possible_trees:
3533
delta = tree.inventory._make_delta(basis_tree.inventory)
3534
deltas.append((len(delta), basis_id, delta))
3536
return deltas[0][1:]
3538
def _get_parent_keys(self, root_key, parent_map):
3539
"""Get the parent keys for a given root id."""
3540
root_id, rev_id = root_key
3541
# Include direct parents of the revision, but only if they used
3544
for parent_id in parent_map[rev_id]:
3545
if parent_id == _mod_revision.NULL_REVISION:
3547
if parent_id not in self._revision_id_to_root_id:
3548
# We probably didn't read this revision, go spend the
3549
# extra effort to actually check
3551
tree = self.source.revision_tree(parent_id)
3552
except errors.NoSuchRevision:
3553
# Ghost, fill out _revision_id_to_root_id in case we
3554
# encounter this again.
3555
# But set parent_root_id to None since we don't really know
3556
parent_root_id = None
3558
parent_root_id = tree.get_root_id()
3559
self._revision_id_to_root_id[parent_id] = None
3561
parent_root_id = self._revision_id_to_root_id[parent_id]
3562
if root_id == parent_root_id or parent_root_id is None:
3563
parent_keys.append((root_id, parent_id))
3564
return tuple(parent_keys)
3566
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3567
for root_key in root_keys_to_create:
3568
parent_keys = self._get_parent_keys(root_key, parent_map)
3569
yield versionedfile.FulltextContentFactory(root_key,
3570
parent_keys, None, '')
3572
def _fetch_batch(self, revision_ids, basis_id, cache):
3573
"""Fetch across a few revisions.
3575
:param revision_ids: The revisions to copy
3576
:param basis_id: The revision_id of a tree that must be in cache, used
3577
as a basis for delta when no other base is available
3578
:param cache: A cache of RevisionTrees that we can use.
3579
:return: The revision_id of the last converted tree. The RevisionTree
3580
for it will be in cache
3582
# Walk though all revisions; get inventory deltas, copy referenced
3583
# texts that delta references, insert the delta, revision and
3585
root_keys_to_create = set()
3588
pending_revisions = []
3589
parent_map = self.source.get_parent_map(revision_ids)
3590
for tree in self.source.revision_trees(revision_ids):
3591
current_revision_id = tree.get_revision_id()
3592
parent_ids = parent_map.get(current_revision_id, ())
3593
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3595
if self._converting_to_rich_root:
3596
self._revision_id_to_root_id[current_revision_id] = \
3598
# Find text entries that need to be copied
3599
for old_path, new_path, file_id, entry in delta:
3600
if new_path is not None:
3603
if not self.target.supports_rich_root():
3604
# The target doesn't support rich root, so we don't
3607
if self._converting_to_rich_root:
3608
# This can't be copied normally, we have to insert
3610
root_keys_to_create.add((file_id, entry.revision))
3612
text_keys.add((file_id, entry.revision))
3613
revision = self.source.get_revision(current_revision_id)
3614
pending_deltas.append((basis_id, delta,
3615
current_revision_id, revision.parent_ids))
3616
pending_revisions.append(revision)
3617
cache[current_revision_id] = tree
3618
basis_id = current_revision_id
3620
from_texts = self.source.texts
3621
to_texts = self.target.texts
3622
if root_keys_to_create:
3623
root_stream = self._new_root_data_stream(root_keys_to_create,
3625
to_texts.insert_record_stream(root_stream)
3626
to_texts.insert_record_stream(from_texts.get_record_stream(
3627
text_keys, self.target._format._fetch_order,
3628
not self.target._format._fetch_uses_deltas))
3629
# insert inventory deltas
3630
for delta in pending_deltas:
3631
self.target.add_inventory_by_delta(*delta)
3632
if self.target._fallback_repositories:
3633
# Make sure this stacked repository has all the parent inventories
3634
# for the new revisions that we are about to insert. We do this
3635
# before adding the revisions so that no revision is added until
3636
# all the inventories it may depend on are added.
3638
revision_ids = set()
3639
for revision in pending_revisions:
3640
revision_ids.add(revision.revision_id)
3641
parent_ids.update(revision.parent_ids)
3642
parent_ids.difference_update(revision_ids)
3643
parent_ids.discard(_mod_revision.NULL_REVISION)
3644
parent_map = self.source.get_parent_map(parent_ids)
3645
for parent_tree in self.source.revision_trees(parent_ids):
3646
basis_id, delta = self._get_delta_for_revision(tree, parent_ids, basis_id, cache)
3647
current_revision_id = parent_tree.get_revision_id()
3648
parents_parents = parent_map[current_revision_id]
3649
self.target.add_inventory_by_delta(
3650
basis_id, delta, current_revision_id, parents_parents)
3651
# insert signatures and revisions
3652
for revision in pending_revisions:
3654
signature = self.source.get_signature_text(
3655
revision.revision_id)
3656
self.target.add_signature_text(revision.revision_id,
3658
except errors.NoSuchRevision:
3660
self.target.add_revision(revision.revision_id, revision)
3663
def _fetch_all_revisions(self, revision_ids, pb):
3664
"""Fetch everything for the list of revisions.
3666
:param revision_ids: The list of revisions to fetch. Must be in
3668
:param pb: A ProgressBar
3671
basis_id, basis_tree = self._get_basis(revision_ids[0])
3673
cache = lru_cache.LRUCache(100)
3674
cache[basis_id] = basis_tree
3675
del basis_tree # We don't want to hang on to it here
3676
for offset in range(0, len(revision_ids), batch_size):
3677
self.target.start_write_group()
3679
pb.update('Transferring revisions', offset,
3681
batch = revision_ids[offset:offset+batch_size]
3682
basis_id = self._fetch_batch(batch, basis_id, cache)
3684
self.target.abort_write_group()
3687
self.target.commit_write_group()
3688
pb.update('Transferring revisions', len(revision_ids),
3692
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3694
"""See InterRepository.fetch()."""
3695
if fetch_spec is not None:
3696
raise AssertionError("Not implemented yet...")
3697
if (not self.source.supports_rich_root()
3698
and self.target.supports_rich_root()):
3699
self._converting_to_rich_root = True
3700
self._revision_id_to_root_id = {}
3702
self._converting_to_rich_root = False
3703
revision_ids = self.target.search_missing_revision_ids(self.source,
3704
revision_id, find_ghosts=find_ghosts).get_keys()
3705
if not revision_ids:
3707
revision_ids = tsort.topo_sort(
3708
self.source.get_graph().get_parent_map(revision_ids))
3709
if not revision_ids:
3711
# Walk though all revisions; get inventory deltas, copy referenced
3712
# texts that delta references, insert the delta, revision and
3714
first_rev = self.source.get_revision(revision_ids[0])
3716
my_pb = ui.ui_factory.nested_progress_bar()
3719
symbol_versioning.warn(
3720
symbol_versioning.deprecated_in((1, 14, 0))
3721
% "pb parameter to fetch()")
3724
self._fetch_all_revisions(revision_ids, pb)
3726
if my_pb is not None:
3728
return len(revision_ids), 0
3730
def _get_basis(self, first_revision_id):
3731
"""Get a revision and tree which exists in the target.
3733
This assumes that first_revision_id is selected for transmission
3734
because all other ancestors are already present. If we can't find an
3735
ancestor we fall back to NULL_REVISION since we know that is safe.
3737
:return: (basis_id, basis_tree)
3739
first_rev = self.source.get_revision(first_revision_id)
3741
basis_id = first_rev.parent_ids[0]
3742
# only valid as a basis if the target has it
3743
self.target.get_revision(basis_id)
3744
# Try to get a basis tree - if its a ghost it will hit the
3745
# NoSuchRevision case.
3746
basis_tree = self.source.revision_tree(basis_id)
3747
except (IndexError, errors.NoSuchRevision):
3748
basis_id = _mod_revision.NULL_REVISION
3749
basis_tree = self.source.revision_tree(basis_id)
3750
return basis_id, basis_tree
3753
InterRepository.register_optimiser(InterDifferingSerializer)
3754
InterRepository.register_optimiser(InterSameDataRepository)
3755
InterRepository.register_optimiser(InterWeaveRepo)
3756
InterRepository.register_optimiser(InterKnitRepo)
3757
InterRepository.register_optimiser(InterPackRepo)
1788
3760
class CopyConverter(object):
1789
3761
"""A repository conversion tool which just performs a copy of the content.
1804
3776
:param to_convert: The disk object to convert.
1805
3777
:param pb: a progress bar to use for progress information.
1807
pb = ui.ui_factory.nested_progress_bar()
1810
3782
# this is only useful with metadir layouts - separated repo content.
1811
3783
# trigger an assertion if not such
1812
3784
repo._format.get_format_string()
1813
3785
self.repo_dir = repo.bzrdir
1814
pb.update('Moving repository to repository.backup')
3786
self.step('Moving repository to repository.backup')
1815
3787
self.repo_dir.transport.move('repository', 'repository.backup')
1816
3788
backup_transport = self.repo_dir.transport.clone('repository.backup')
1817
3789
repo._format.check_conversion_target(self.target_format)
1818
3790
self.source_repo = repo._format.open(self.repo_dir,
1820
3792
_override_transport=backup_transport)
1821
pb.update('Creating new repository')
3793
self.step('Creating new repository')
1822
3794
converted = self.target_format.initialize(self.repo_dir,
1823
3795
self.source_repo.is_shared())
1824
3796
converted.lock_write()
1826
pb.update('Copying content')
3798
self.step('Copying content into repository.')
1827
3799
self.source_repo.copy_content_into(converted)
1829
3801
converted.unlock()
1830
pb.update('Deleting old repository content')
3802
self.step('Deleting old repository content.')
1831
3803
self.repo_dir.transport.delete_tree('repository.backup')
1832
ui.ui_factory.note('repository converted')
3804
self.pb.note('repository converted')
3806
def step(self, message):
3807
"""Update the pb by a step."""
3809
self.pb.update(message, self.count, self.total)
3821
def _unescaper(match, _map=_unescape_map):
3822
code = match.group(1)
3826
if not code.startswith('#'):
3828
return unichr(int(code[1:])).encode('utf8')
3834
def _unescape_xml(data):
3835
"""Unescape predefined XML entities in a string of data."""
3837
if _unescape_re is None:
3838
_unescape_re = re.compile('\&([^;]*);')
3839
return _unescape_re.sub(_unescaper, data)
3842
class _VersionedFileChecker(object):
3844
def __init__(self, repository, text_key_references=None):
3845
self.repository = repository
3846
self.text_index = self.repository._generate_text_key_index(
3847
text_key_references=text_key_references)
3849
def calculate_file_version_parents(self, text_key):
3850
"""Calculate the correct parents for a file version according to
3853
parent_keys = self.text_index[text_key]
3854
if parent_keys == [_mod_revision.NULL_REVISION]:
3856
return tuple(parent_keys)
3858
def check_file_version_parents(self, texts, progress_bar=None):
3859
"""Check the parents stored in a versioned file are correct.
3861
It also detects file versions that are not referenced by their
3862
corresponding revision's inventory.
3864
:returns: A tuple of (wrong_parents, dangling_file_versions).
3865
wrong_parents is a dict mapping {revision_id: (stored_parents,
3866
correct_parents)} for each revision_id where the stored parents
3867
are not correct. dangling_file_versions is a set of (file_id,
3868
revision_id) tuples for versions that are present in this versioned
3869
file, but not used by the corresponding inventory.
3872
self.file_ids = set([file_id for file_id, _ in
3873
self.text_index.iterkeys()])
3874
# text keys is now grouped by file_id
3875
n_weaves = len(self.file_ids)
3876
files_in_revisions = {}
3877
revisions_of_files = {}
3878
n_versions = len(self.text_index)
3879
progress_bar.update('loading text store', 0, n_versions)
3880
parent_map = self.repository.texts.get_parent_map(self.text_index)
3881
# On unlistable transports this could well be empty/error...
3882
text_keys = self.repository.texts.keys()
3883
unused_keys = frozenset(text_keys) - set(self.text_index)
3884
for num, key in enumerate(self.text_index.iterkeys()):
3885
if progress_bar is not None:
3886
progress_bar.update('checking text graph', num, n_versions)
3887
correct_parents = self.calculate_file_version_parents(key)
3889
knit_parents = parent_map[key]
3890
except errors.RevisionNotPresent:
3893
if correct_parents != knit_parents:
3894
wrong_parents[key] = (knit_parents, correct_parents)
3895
return wrong_parents, unused_keys
3898
def _old_get_graph(repository, revision_id):
3899
"""DO NOT USE. That is all. I'm serious."""
3900
graph = repository.get_graph()
3901
revision_graph = dict(((key, value) for key, value in
3902
graph.iter_ancestry([revision_id]) if value is not None))
3903
return _strip_NULL_ghosts(revision_graph)
1836
3906
def _strip_NULL_ghosts(revision_graph):
1844
3914
return revision_graph
1847
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
1848
stop_revision=None):
1849
"""Extend the partial history to include a given index
1851
If a stop_index is supplied, stop when that index has been reached.
1852
If a stop_revision is supplied, stop when that revision is
1853
encountered. Otherwise, stop when the beginning of history is
1856
:param stop_index: The index which should be present. When it is
1857
present, history extension will stop.
1858
:param stop_revision: The revision id which should be present. When
1859
it is encountered, history extension will stop.
1861
start_revision = partial_history_cache[-1]
1862
graph = repo.get_graph()
1863
iterator = graph.iter_lefthand_ancestry(start_revision,
1864
(_mod_revision.NULL_REVISION,))
1866
# skip the last revision in the list
1869
if (stop_index is not None and
1870
len(partial_history_cache) > stop_index):
1872
if partial_history_cache[-1] == stop_revision:
1874
revision_id = iterator.next()
1875
partial_history_cache.append(revision_id)
1876
except StopIteration:
1881
class _LazyListJoin(object):
1882
"""An iterable yielding the contents of many lists as one list.
1884
Each iterator made from this will reflect the current contents of the lists
1885
at the time the iterator is made.
1887
This is used by Repository's _make_parents_provider implementation so that
1890
pp = repo._make_parents_provider() # uses a list of fallback repos
1891
pp.add_fallback_repository(other_repo) # appends to that list
1892
result = pp.get_parent_map(...)
1893
# The result will include revs from other_repo
1896
def __init__(self, *list_parts):
1897
self.list_parts = list_parts
1901
for list_part in self.list_parts:
1902
full_list.extend(list_part)
1903
return iter(full_list)
3917
class StreamSink(object):
3918
"""An object that can insert a stream into a repository.
3920
This interface handles the complexity of reserialising inventories and
3921
revisions from different formats, and allows unidirectional insertion into
3922
stacked repositories without looking for the missing basis parents
3926
def __init__(self, target_repo):
3927
self.target_repo = target_repo
3929
def insert_stream(self, stream, src_format, resume_tokens):
3930
"""Insert a stream's content into the target repository.
3932
:param src_format: a bzr repository format.
3934
:return: a list of resume tokens and an iterable of keys additional
3935
items required before the insertion can be completed.
3937
self.target_repo.lock_write()
3940
self.target_repo.resume_write_group(resume_tokens)
3942
self.target_repo.start_write_group()
3944
# locked_insert_stream performs a commit|suspend.
3945
return self._locked_insert_stream(stream, src_format)
3947
self.target_repo.abort_write_group(suppress_errors=True)
3950
self.target_repo.unlock()
3952
def _locked_insert_stream(self, stream, src_format):
3953
to_serializer = self.target_repo._format._serializer
3954
src_serializer = src_format._serializer
3955
if to_serializer == src_serializer:
3956
# If serializers match and the target is a pack repository, set the
3957
# write cache size on the new pack. This avoids poor performance
3958
# on transports where append is unbuffered (such as
3959
# RemoteTransport). This is safe to do because nothing should read
3960
# back from the target repository while a stream with matching
3961
# serialization is being inserted.
3962
# The exception is that a delta record from the source that should
3963
# be a fulltext may need to be expanded by the target (see
3964
# test_fetch_revisions_with_deltas_into_pack); but we take care to
3965
# explicitly flush any buffered writes first in that rare case.
3967
new_pack = self.target_repo._pack_collection._new_pack
3968
except AttributeError:
3969
# Not a pack repository
3972
new_pack.set_write_cache_size(1024*1024)
3973
for substream_type, substream in stream:
3974
if substream_type == 'texts':
3975
self.target_repo.texts.insert_record_stream(substream)
3976
elif substream_type == 'inventories':
3977
if src_serializer == to_serializer:
3978
self.target_repo.inventories.insert_record_stream(
3981
self._extract_and_insert_inventories(
3982
substream, src_serializer)
3983
elif substream_type == 'chk_bytes':
3984
# XXX: This doesn't support conversions, as it assumes the
3985
# conversion was done in the fetch code.
3986
self.target_repo.chk_bytes.insert_record_stream(substream)
3987
elif substream_type == 'revisions':
3988
# This may fallback to extract-and-insert more often than
3989
# required if the serializers are different only in terms of
3991
if src_serializer == to_serializer:
3992
self.target_repo.revisions.insert_record_stream(
3995
self._extract_and_insert_revisions(substream,
3997
elif substream_type == 'signatures':
3998
self.target_repo.signatures.insert_record_stream(substream)
4000
raise AssertionError('kaboom! %s' % (substream_type,))
4002
missing_keys = set()
4003
for prefix, versioned_file in (
4004
('texts', self.target_repo.texts),
4005
('inventories', self.target_repo.inventories),
4006
('revisions', self.target_repo.revisions),
4007
('signatures', self.target_repo.signatures),
4009
missing_keys.update((prefix,) + key for key in
4010
versioned_file.get_missing_compression_parent_keys())
4011
except NotImplementedError:
4012
# cannot even attempt suspending, and missing would have failed
4013
# during stream insertion.
4014
missing_keys = set()
4017
# suspend the write group and tell the caller what we is
4018
# missing. We know we can suspend or else we would not have
4019
# entered this code path. (All repositories that can handle
4020
# missing keys can handle suspending a write group).
4021
write_group_tokens = self.target_repo.suspend_write_group()
4022
return write_group_tokens, missing_keys
4023
self.target_repo.commit_write_group()
4026
def _extract_and_insert_inventories(self, substream, serializer):
4027
"""Generate a new inventory versionedfile in target, converting data.
4029
The inventory is retrieved from the source, (deserializing it), and
4030
stored in the target (reserializing it in a different format).
4032
for record in substream:
4033
bytes = record.get_bytes_as('fulltext')
4034
revision_id = record.key[0]
4035
inv = serializer.read_inventory_from_string(bytes, revision_id)
4036
parents = [key[0] for key in record.parents]
4037
self.target_repo.add_inventory(revision_id, inv, parents)
4039
def _extract_and_insert_revisions(self, substream, serializer):
4040
for record in substream:
4041
bytes = record.get_bytes_as('fulltext')
4042
revision_id = record.key[0]
4043
rev = serializer.read_revision_from_string(bytes)
4044
if rev.revision_id != revision_id:
4045
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4046
self.target_repo.add_revision(revision_id, rev)
4049
if self.target_repo._format._fetch_reconcile:
4050
self.target_repo.reconcile()
4053
class StreamSource(object):
4054
"""A source of a stream for fetching between repositories."""
4056
def __init__(self, from_repository, to_format):
4057
"""Create a StreamSource streaming from from_repository."""
4058
self.from_repository = from_repository
4059
self.to_format = to_format
4061
def delta_on_metadata(self):
4062
"""Return True if delta's are permitted on metadata streams.
4064
That is on revisions and signatures.
4066
src_serializer = self.from_repository._format._serializer
4067
target_serializer = self.to_format._serializer
4068
return (self.to_format._fetch_uses_deltas and
4069
src_serializer == target_serializer)
4071
def _fetch_revision_texts(self, revs):
4072
# fetch signatures first and then the revision texts
4073
# may need to be a InterRevisionStore call here.
4074
from_sf = self.from_repository.signatures
4075
# A missing signature is just skipped.
4076
keys = [(rev_id,) for rev_id in revs]
4077
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4079
self.to_format._fetch_order,
4080
not self.to_format._fetch_uses_deltas))
4081
# If a revision has a delta, this is actually expanded inside the
4082
# insert_record_stream code now, which is an alternate fix for
4084
from_rf = self.from_repository.revisions
4085
revisions = from_rf.get_record_stream(
4087
self.to_format._fetch_order,
4088
not self.delta_on_metadata())
4089
return [('signatures', signatures), ('revisions', revisions)]
4091
def _generate_root_texts(self, revs):
4092
"""This will be called by __fetch between fetching weave texts and
4093
fetching the inventory weave.
4095
Subclasses should override this if they need to generate root texts
4096
after fetching weave texts.
4098
if self._rich_root_upgrade():
4100
return bzrlib.fetch.Inter1and2Helper(
4101
self.from_repository).generate_root_texts(revs)
4105
def get_stream(self, search):
4107
revs = search.get_keys()
4108
graph = self.from_repository.get_graph()
4109
revs = list(graph.iter_topo_order(revs))
4110
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4112
for knit_kind, file_id, revisions in data_to_fetch:
4113
if knit_kind != phase:
4115
# Make a new progress bar for this phase
4116
if knit_kind == "file":
4117
# Accumulate file texts
4118
text_keys.extend([(file_id, revision) for revision in
4120
elif knit_kind == "inventory":
4121
# Now copy the file texts.
4122
from_texts = self.from_repository.texts
4123
yield ('texts', from_texts.get_record_stream(
4124
text_keys, self.to_format._fetch_order,
4125
not self.to_format._fetch_uses_deltas))
4126
# Cause an error if a text occurs after we have done the
4129
# Before we process the inventory we generate the root
4130
# texts (if necessary) so that the inventories references
4132
for _ in self._generate_root_texts(revs):
4134
# NB: This currently reopens the inventory weave in source;
4135
# using a single stream interface instead would avoid this.
4136
from_weave = self.from_repository.inventories
4137
# we fetch only the referenced inventories because we do not
4138
# know for unselected inventories whether all their required
4139
# texts are present in the other repository - it could be
4141
for info in self._get_inventory_stream(revs):
4143
elif knit_kind == "signatures":
4144
# Nothing to do here; this will be taken care of when
4145
# _fetch_revision_texts happens.
4147
elif knit_kind == "revisions":
4148
for record in self._fetch_revision_texts(revs):
4151
raise AssertionError("Unknown knit kind %r" % knit_kind)
4153
def get_stream_for_missing_keys(self, missing_keys):
4154
# missing keys can only occur when we are byte copying and not
4155
# translating (because translation means we don't send
4156
# unreconstructable deltas ever).
4158
keys['texts'] = set()
4159
keys['revisions'] = set()
4160
keys['inventories'] = set()
4161
keys['signatures'] = set()
4162
for key in missing_keys:
4163
keys[key[0]].add(key[1:])
4164
if len(keys['revisions']):
4165
# If we allowed copying revisions at this point, we could end up
4166
# copying a revision without copying its required texts: a
4167
# violation of the requirements for repository integrity.
4168
raise AssertionError(
4169
'cannot copy revisions to fill in missing deltas %s' % (
4170
keys['revisions'],))
4171
for substream_kind, keys in keys.iteritems():
4172
vf = getattr(self.from_repository, substream_kind)
4173
# Ask for full texts always so that we don't need more round trips
4174
# after this stream.
4175
stream = vf.get_record_stream(keys,
4176
self.to_format._fetch_order, True)
4177
yield substream_kind, stream
4179
def inventory_fetch_order(self):
4180
if self._rich_root_upgrade():
4181
return 'topological'
4183
return self.to_format._fetch_order
4185
def _rich_root_upgrade(self):
4186
return (not self.from_repository._format.rich_root_data and
4187
self.to_format.rich_root_data)
4189
def _get_inventory_stream(self, revision_ids):
4190
from_format = self.from_repository._format
4191
if (from_format.supports_chks and self.to_format.supports_chks
4192
and (from_format._serializer == self.to_format._serializer)):
4193
# Both sides support chks, and they use the same serializer, so it
4194
# is safe to transmit the chk pages and inventory pages across
4196
return self._get_chk_inventory_stream(revision_ids)
4197
elif (not from_format.supports_chks):
4198
# Source repository doesn't support chks. So we can transmit the
4199
# inventories 'as-is' and either they are just accepted on the
4200
# target, or the Sink will properly convert it.
4201
return self._get_simple_inventory_stream(revision_ids)
4203
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4204
# inventories. Note that this should probably be done somehow
4205
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4206
# figure out how a non-chk repository could possibly handle
4207
# deserializing an inventory stream from a chk repo, as it
4208
# doesn't have a way to understand individual pages.
4209
return self._get_convertable_inventory_stream(revision_ids)
4211
def _get_simple_inventory_stream(self, revision_ids):
4212
from_weave = self.from_repository.inventories
4213
yield ('inventories', from_weave.get_record_stream(
4214
[(rev_id,) for rev_id in revision_ids],
4215
self.inventory_fetch_order(),
4216
not self.delta_on_metadata()))
4218
def _get_chk_inventory_stream(self, revision_ids):
4219
"""Fetch the inventory texts, along with the associated chk maps."""
4220
# We want an inventory outside of the search set, so that we can filter
4221
# out uninteresting chk pages. For now we use
4222
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4223
# could use that instead.
4224
start_rev_id = self.from_repository._find_revision_outside_set(
4226
start_rev_key = (start_rev_id,)
4227
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4228
if start_rev_id != _mod_revision.NULL_REVISION:
4229
inv_keys_to_fetch.append((start_rev_id,))
4230
# Any repo that supports chk_bytes must also support out-of-order
4231
# insertion. At least, that is how we expect it to work
4232
# We use get_record_stream instead of iter_inventories because we want
4233
# to be able to insert the stream as well. We could instead fetch
4234
# allowing deltas, and then iter_inventories, but we don't know whether
4235
# source or target is more 'local' anway.
4236
inv_stream = self.from_repository.inventories.get_record_stream(
4237
inv_keys_to_fetch, 'unordered',
4238
True) # We need them as full-texts so we can find their references
4239
uninteresting_chk_roots = set()
4240
interesting_chk_roots = set()
4241
def filter_inv_stream(inv_stream):
4242
for idx, record in enumerate(inv_stream):
4243
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4244
bytes = record.get_bytes_as('fulltext')
4245
chk_inv = inventory.CHKInventory.deserialise(
4246
self.from_repository.chk_bytes, bytes, record.key)
4247
if record.key == start_rev_key:
4248
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4249
p_id_map = chk_inv.parent_id_basename_to_file_id
4250
if p_id_map is not None:
4251
uninteresting_chk_roots.add(p_id_map.key())
4254
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4255
p_id_map = chk_inv.parent_id_basename_to_file_id
4256
if p_id_map is not None:
4257
interesting_chk_roots.add(p_id_map.key())
4258
### pb.update('fetch inventory', 0, 2)
4259
yield ('inventories', filter_inv_stream(inv_stream))
4260
# Now that we have worked out all of the interesting root nodes, grab
4261
# all of the interesting pages and insert them
4262
### pb.update('fetch inventory', 1, 2)
4263
interesting = chk_map.iter_interesting_nodes(
4264
self.from_repository.chk_bytes, interesting_chk_roots,
4265
uninteresting_chk_roots)
4266
def to_stream_adapter():
4267
"""Adapt the iter_interesting_nodes result to a single stream.
4269
iter_interesting_nodes returns records as it processes them, along
4270
with keys. However, we only want to return the records themselves.
4272
for record, items in interesting:
4273
if record is not None:
4275
# XXX: We could instead call get_record_stream(records.keys())
4276
# ATM, this will always insert the records as fulltexts, and
4277
# requires that you can hang on to records once you have gone
4278
# on to the next one. Further, it causes the target to
4279
# recompress the data. Testing shows it to be faster than
4280
# requesting the records again, though.
4281
yield ('chk_bytes', to_stream_adapter())
4282
### pb.update('fetch inventory', 2, 2)
4284
def _get_convertable_inventory_stream(self, revision_ids):
4285
# XXX: One of source or target is using chks, and they don't have
4286
# compatible serializations. The StreamSink code expects to be
4287
# able to convert on the target, so we need to put
4288
# bytes-on-the-wire that can be converted
4289
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4291
def _stream_invs_as_fulltexts(self, revision_ids):
4292
from_repo = self.from_repository
4293
from_serializer = from_repo._format._serializer
4294
revision_keys = [(rev_id,) for rev_id in revision_ids]
4295
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4296
for inv in self.from_repository.iter_inventories(revision_ids):
4297
# XXX: This is a bit hackish, but it works. Basically,
4298
# CHKSerializer 'accidentally' supports
4299
# read/write_inventory_to_string, even though that is never
4300
# the format that is stored on disk. It *does* give us a
4301
# single string representation for an inventory, so live with
4303
# This would be far better if we had a 'serialized inventory
4304
# delta' form. Then we could use 'inventory._make_delta', and
4305
# transmit that. This would both be faster to generate, and
4306
# result in fewer bytes-on-the-wire.
4307
as_bytes = from_serializer.write_inventory_to_string(inv)
4308
key = (inv.revision_id,)
4309
parent_keys = parent_map.get(key, ())
4310
yield versionedfile.FulltextContentFactory(
4311
key, parent_keys, None, as_bytes)