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
992
return InterRepository._assert_same_model(self, repository)
994
def add_inventory(self, revision_id, inv, parents):
995
"""Add the inventory inv to the repository as revision_id.
997
:param parents: The revision ids of the parents that revision_id
998
is known to have and are in the repository already.
1000
:returns: The validator(which is a sha1 digest, though what is sha'd is
1001
repository format specific) of the serialized inventory.
1003
if not self.is_in_write_group():
1004
raise AssertionError("%r not in write group" % (self,))
1005
_mod_revision.check_not_reserved_id(revision_id)
1006
if not (inv.revision_id is None or inv.revision_id == revision_id):
1007
raise AssertionError(
1008
"Mismatch between inventory revision"
1009
" id and insertion revid (%r, %r)"
1010
% (inv.revision_id, revision_id))
1011
if inv.root is None:
1012
raise AssertionError()
1013
return self._add_inventory_checked(revision_id, inv, parents)
1015
def _add_inventory_checked(self, revision_id, inv, parents):
1016
"""Add inv to the repository after checking the inputs.
1018
This function can be overridden to allow different inventory styles.
1020
:seealso: add_inventory, for the contract.
1022
inv_lines = self._serialise_inventory_to_lines(inv)
1023
return self._inventory_add_lines(revision_id, parents,
1024
inv_lines, check_content=False)
1026
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1027
parents, basis_inv=None, propagate_caches=False):
1028
"""Add a new inventory expressed as a delta against another revision.
1030
:param basis_revision_id: The inventory id the delta was created
1031
against. (This does not have to be a direct parent.)
1032
:param delta: The inventory delta (see Inventory.apply_delta for
1034
:param new_revision_id: The revision id that the inventory is being
1036
:param parents: The revision ids of the parents that revision_id is
1037
known to have and are in the repository already. These are supplied
1038
for repositories that depend on the inventory graph for revision
1039
graph access, as well as for those that pun ancestry with delta
1041
:param basis_inv: The basis inventory if it is already known,
1043
:param propagate_caches: If True, the caches for this inventory are
1044
copied to and updated for the result if possible.
1046
:returns: (validator, new_inv)
1047
The validator(which is a sha1 digest, though what is sha'd is
1048
repository format specific) of the serialized inventory, and the
1049
resulting inventory.
1051
if not self.is_in_write_group():
1052
raise AssertionError("%r not in write group" % (self,))
1053
_mod_revision.check_not_reserved_id(new_revision_id)
1054
basis_tree = self.revision_tree(basis_revision_id)
1055
basis_tree.lock_read()
1057
# Note that this mutates the inventory of basis_tree, which not all
1058
# inventory implementations may support: A better idiom would be to
1059
# return a new inventory, but as there is no revision tree cache in
1060
# repository this is safe for now - RBC 20081013
1061
if basis_inv is None:
1062
basis_inv = basis_tree.inventory
1063
basis_inv.apply_delta(delta)
1064
basis_inv.revision_id = new_revision_id
1065
return (self.add_inventory(new_revision_id, basis_inv, parents),
1070
def _inventory_add_lines(self, revision_id, parents, lines,
1071
check_content=True):
1072
"""Store lines in inv_vf and return the sha1 of the inventory."""
1073
parents = [(parent,) for parent in parents]
1074
return self.inventories.add_lines((revision_id,), parents, lines,
1075
check_content=check_content)[0]
1077
def add_revision(self, revision_id, rev, inv=None, config=None):
1078
"""Add rev to the revision store as revision_id.
1080
:param revision_id: the revision id to use.
1081
:param rev: The revision object.
1082
:param inv: The inventory for the revision. if None, it will be looked
1083
up in the inventory storer
1084
:param config: If None no digital signature will be created.
1085
If supplied its signature_needed method will be used
1086
to determine if a signature should be made.
1088
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1090
_mod_revision.check_not_reserved_id(revision_id)
1091
if config is not None and config.signature_needed():
1093
inv = self.get_inventory(revision_id)
1094
plaintext = Testament(rev, inv).as_short_text()
1095
self.store_revision_signature(
1096
gpg.GPGStrategy(config), plaintext, revision_id)
1097
# check inventory present
1098
if not self.inventories.get_parent_map([(revision_id,)]):
1100
raise errors.WeaveRevisionNotPresent(revision_id,
1103
# yes, this is not suitable for adding with ghosts.
1104
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1107
key = (revision_id,)
1108
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1109
self._add_revision(rev)
1111
def _add_revision(self, revision):
1112
text = self._serializer.write_revision_to_string(revision)
1113
key = (revision.revision_id,)
1114
parents = tuple((parent,) for parent in revision.parent_ids)
1115
self.revisions.add_lines(key, parents, osutils.split_lines(text))
315
1117
def all_revision_ids(self):
316
1118
"""Returns a list of all the revision ids in the repository.
909
1796
signature = gpg_strategy.sign(plaintext)
910
1797
self.add_signature_text(revision_id, signature)
912
1800
def add_signature_text(self, revision_id, signature):
913
"""Store a signature text for a revision.
915
:param revision_id: Revision id of the revision
916
:param signature: Signature text.
918
raise NotImplementedError(self.add_signature_text)
920
def _find_parent_ids_of_revisions(self, revision_ids):
921
"""Find all parent ids that are mentioned in the revision graph.
923
:return: set of revisions that are parents of revision_ids which are
924
not part of revision_ids themselves
926
parent_map = self.get_parent_map(revision_ids)
928
map(parent_ids.update, parent_map.itervalues())
929
parent_ids.difference_update(revision_ids)
930
parent_ids.discard(_mod_revision.NULL_REVISION)
933
def fileids_altered_by_revision_ids(self, revision_ids):
1801
self.signatures.add_lines((revision_id,), (),
1802
osutils.split_lines(signature))
1804
def find_text_key_references(self):
1805
"""Find the text key references within the repository.
1807
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1808
to whether they were referred to by the inventory of the
1809
revision_id that they contain. The inventory texts from all present
1810
revision ids are assessed to generate this report.
1812
revision_keys = self.revisions.keys()
1813
w = self.inventories
1814
pb = ui.ui_factory.nested_progress_bar()
1816
return self._find_text_key_references_from_xml_inventory_lines(
1817
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1821
def _find_text_key_references_from_xml_inventory_lines(self,
1823
"""Core routine for extracting references to texts from inventories.
1825
This performs the translation of xml lines to revision ids.
1827
:param line_iterator: An iterator of lines, origin_version_id
1828
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1829
to whether they were referred to by the inventory of the
1830
revision_id that they contain. Note that if that revision_id was
1831
not part of the line_iterator's output then False will be given -
1832
even though it may actually refer to that key.
1834
if not self._serializer.support_altered_by_hack:
1835
raise AssertionError(
1836
"_find_text_key_references_from_xml_inventory_lines only "
1837
"supported for branches which store inventory as unnested xml"
1838
", not on %r" % self)
1841
# this code needs to read every new line in every inventory for the
1842
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1843
# not present in one of those inventories is unnecessary but not
1844
# harmful because we are filtering by the revision id marker in the
1845
# inventory lines : we only select file ids altered in one of those
1846
# revisions. We don't need to see all lines in the inventory because
1847
# only those added in an inventory in rev X can contain a revision=X
1849
unescape_revid_cache = {}
1850
unescape_fileid_cache = {}
1852
# jam 20061218 In a big fetch, this handles hundreds of thousands
1853
# of lines, so it has had a lot of inlining and optimizing done.
1854
# Sorry that it is a little bit messy.
1855
# Move several functions to be local variables, since this is a long
1857
search = self._file_ids_altered_regex.search
1858
unescape = _unescape_xml
1859
setdefault = result.setdefault
1860
for line, line_key in line_iterator:
1861
match = search(line)
1864
# One call to match.group() returning multiple items is quite a
1865
# bit faster than 2 calls to match.group() each returning 1
1866
file_id, revision_id = match.group('file_id', 'revision_id')
1868
# Inlining the cache lookups helps a lot when you make 170,000
1869
# lines and 350k ids, versus 8.4 unique ids.
1870
# Using a cache helps in 2 ways:
1871
# 1) Avoids unnecessary decoding calls
1872
# 2) Re-uses cached strings, which helps in future set and
1874
# (2) is enough that removing encoding entirely along with
1875
# the cache (so we are using plain strings) results in no
1876
# performance improvement.
1878
revision_id = unescape_revid_cache[revision_id]
1880
unescaped = unescape(revision_id)
1881
unescape_revid_cache[revision_id] = unescaped
1882
revision_id = unescaped
1884
# Note that unconditionally unescaping means that we deserialise
1885
# every fileid, which for general 'pull' is not great, but we don't
1886
# really want to have some many fulltexts that this matters anyway.
1889
file_id = unescape_fileid_cache[file_id]
1891
unescaped = unescape(file_id)
1892
unescape_fileid_cache[file_id] = unescaped
1895
key = (file_id, revision_id)
1896
setdefault(key, False)
1897
if revision_id == line_key[-1]:
1901
def _inventory_xml_lines_for_keys(self, keys):
1902
"""Get a line iterator of the sort needed for findind references.
1904
Not relevant for non-xml inventory repositories.
1906
Ghosts in revision_keys are ignored.
1908
:param revision_keys: The revision keys for the inventories to inspect.
1909
:return: An iterator over (inventory line, revid) for the fulltexts of
1910
all of the xml inventories specified by revision_keys.
1912
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1913
for record in stream:
1914
if record.storage_kind != 'absent':
1915
chunks = record.get_bytes_as('chunked')
1916
revid = record.key[-1]
1917
lines = osutils.chunks_to_lines(chunks)
1921
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1923
"""Helper routine for fileids_altered_by_revision_ids.
1925
This performs the translation of xml lines to revision ids.
1927
:param line_iterator: An iterator of lines, origin_version_id
1928
:param revision_ids: The revision ids to filter for. This should be a
1929
set or other type which supports efficient __contains__ lookups, as
1930
the revision id from each parsed line will be looked up in the
1931
revision_ids filter.
1932
:return: a dictionary mapping altered file-ids to an iterable of
1933
revision_ids. Each altered file-ids has the exact revision_ids that
1934
altered it listed explicitly.
1936
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1937
line_iterator).iterkeys())
1938
# Note that revision_ids are revision keys.
1939
parent_maps = self.revisions.get_parent_map(revision_ids)
1941
map(parents.update, parent_maps.itervalues())
1942
parents.difference_update(revision_ids)
1943
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1944
self._inventory_xml_lines_for_keys(parents)))
1945
new_keys = seen - parent_seen
1947
setdefault = result.setdefault
1948
for key in new_keys:
1949
setdefault(key[0], set()).add(key[-1])
1952
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
934
1953
"""Find the file ids and versions affected by revisions.
936
1955
:param revisions: an iterable containing revision ids.
1956
:param _inv_weave: The inventory weave from this repository or None.
1957
If None, the inventory weave will be opened automatically.
937
1958
:return: a dictionary mapping altered file-ids to an iterable of
938
revision_ids. Each altered file-ids has the exact revision_ids
939
that altered it listed explicitly.
1959
revision_ids. Each altered file-ids has the exact revision_ids that
1960
altered it listed explicitly.
941
raise NotImplementedError(self.fileids_altered_by_revision_ids)
1962
selected_keys = set((revid,) for revid in revision_ids)
1963
w = _inv_weave or self.inventories
1964
pb = ui.ui_factory.nested_progress_bar()
1966
return self._find_file_ids_from_xml_inventory_lines(
1967
w.iter_lines_added_or_present_in_keys(
1968
selected_keys, pb=pb),
943
1973
def iter_files_bytes(self, desired_files):
944
1974
"""Iterate through file versions.
951
1981
uniquely identify the file version in the caller's context. (Examples:
952
1982
an index number or a TreeTransform trans_id.)
1984
bytes_iterator is an iterable of bytestrings for the file. The
1985
kind of iterable and length of the bytestrings are unspecified, but for
1986
this implementation, it is a list of bytes produced by
1987
VersionedFile.get_record_stream().
954
1989
:param desired_files: a list of (file_id, revision_id, identifier)
957
raise NotImplementedError(self.iter_files_bytes)
959
def get_rev_id_for_revno(self, revno, known_pair):
960
"""Return the revision id of a revno, given a later (revno, revid)
961
pair in the same history.
963
:return: if found (True, revid). If the available history ran out
964
before reaching the revno, then this returns
965
(False, (closest_revno, closest_revid)).
967
known_revno, known_revid = known_pair
968
partial_history = [known_revid]
969
distance_from_known = known_revno - revno
970
if distance_from_known < 0:
972
'requested revno (%d) is later than given known revno (%d)'
973
% (revno, known_revno))
976
self, partial_history, stop_index=distance_from_known)
977
except errors.RevisionNotPresent, err:
978
if err.revision_id == known_revid:
979
# The start revision (known_revid) wasn't found.
981
# This is a stacked repository with no fallbacks, or a there's a
982
# left-hand ghost. Either way, even though the revision named in
983
# the error isn't in this repo, we know it's the next step in this
985
partial_history.append(err.revision_id)
986
if len(partial_history) <= distance_from_known:
987
# Didn't find enough history to get a revid for the revno.
988
earliest_revno = known_revno - len(partial_history) + 1
989
return (False, (earliest_revno, partial_history[-1]))
990
if len(partial_history) - 1 > distance_from_known:
991
raise AssertionError('_iter_for_revno returned too much history')
992
return (True, partial_history[-1])
994
@symbol_versioning.deprecated_method(symbol_versioning.deprecated_in((2, 4, 0)))
1993
for file_id, revision_id, callable_data in desired_files:
1994
text_keys[(file_id, revision_id)] = callable_data
1995
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1996
if record.storage_kind == 'absent':
1997
raise errors.RevisionNotPresent(record.key, self)
1998
yield text_keys[record.key], record.get_bytes_as('chunked')
2000
def _generate_text_key_index(self, text_key_references=None,
2002
"""Generate a new text key index for the repository.
2004
This is an expensive function that will take considerable time to run.
2006
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2007
list of parents, also text keys. When a given key has no parents,
2008
the parents list will be [NULL_REVISION].
2010
# All revisions, to find inventory parents.
2011
if ancestors is None:
2012
graph = self.get_graph()
2013
ancestors = graph.get_parent_map(self.all_revision_ids())
2014
if text_key_references is None:
2015
text_key_references = self.find_text_key_references()
2016
pb = ui.ui_factory.nested_progress_bar()
2018
return self._do_generate_text_key_index(ancestors,
2019
text_key_references, pb)
2023
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2024
"""Helper for _generate_text_key_index to avoid deep nesting."""
2025
revision_order = tsort.topo_sort(ancestors)
2026
invalid_keys = set()
2028
for revision_id in revision_order:
2029
revision_keys[revision_id] = set()
2030
text_count = len(text_key_references)
2031
# a cache of the text keys to allow reuse; costs a dict of all the
2032
# keys, but saves a 2-tuple for every child of a given key.
2034
for text_key, valid in text_key_references.iteritems():
2036
invalid_keys.add(text_key)
2038
revision_keys[text_key[1]].add(text_key)
2039
text_key_cache[text_key] = text_key
2040
del text_key_references
2042
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2043
NULL_REVISION = _mod_revision.NULL_REVISION
2044
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2045
# too small for large or very branchy trees. However, for 55K path
2046
# trees, it would be easy to use too much memory trivially. Ideally we
2047
# could gauge this by looking at available real memory etc, but this is
2048
# always a tricky proposition.
2049
inventory_cache = lru_cache.LRUCache(10)
2050
batch_size = 10 # should be ~150MB on a 55K path tree
2051
batch_count = len(revision_order) / batch_size + 1
2053
pb.update("Calculating text parents", processed_texts, text_count)
2054
for offset in xrange(batch_count):
2055
to_query = revision_order[offset * batch_size:(offset + 1) *
2059
for rev_tree in self.revision_trees(to_query):
2060
revision_id = rev_tree.get_revision_id()
2061
parent_ids = ancestors[revision_id]
2062
for text_key in revision_keys[revision_id]:
2063
pb.update("Calculating text parents", processed_texts)
2064
processed_texts += 1
2065
candidate_parents = []
2066
for parent_id in parent_ids:
2067
parent_text_key = (text_key[0], parent_id)
2069
check_parent = parent_text_key not in \
2070
revision_keys[parent_id]
2072
# the parent parent_id is a ghost:
2073
check_parent = False
2074
# truncate the derived graph against this ghost.
2075
parent_text_key = None
2077
# look at the parent commit details inventories to
2078
# determine possible candidates in the per file graph.
2081
inv = inventory_cache[parent_id]
2083
inv = self.revision_tree(parent_id).inventory
2084
inventory_cache[parent_id] = inv
2086
parent_entry = inv[text_key[0]]
2087
except (KeyError, errors.NoSuchId):
2089
if parent_entry is not None:
2091
text_key[0], parent_entry.revision)
2093
parent_text_key = None
2094
if parent_text_key is not None:
2095
candidate_parents.append(
2096
text_key_cache[parent_text_key])
2097
parent_heads = text_graph.heads(candidate_parents)
2098
new_parents = list(parent_heads)
2099
new_parents.sort(key=lambda x:candidate_parents.index(x))
2100
if new_parents == []:
2101
new_parents = [NULL_REVISION]
2102
text_index[text_key] = new_parents
2104
for text_key in invalid_keys:
2105
text_index[text_key] = [NULL_REVISION]
2108
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2109
"""Get an iterable listing the keys of all the data introduced by a set
2112
The keys will be ordered so that the corresponding items can be safely
2113
fetched and inserted in that order.
2115
:returns: An iterable producing tuples of (knit-kind, file-id,
2116
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2117
'revisions'. file-id is None unless knit-kind is 'file'.
2119
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2122
for result in self._find_non_file_keys_to_fetch(revision_ids):
2125
def _find_file_keys_to_fetch(self, revision_ids, pb):
2126
# XXX: it's a bit weird to control the inventory weave caching in this
2127
# generator. Ideally the caching would be done in fetch.py I think. Or
2128
# maybe this generator should explicitly have the contract that it
2129
# should not be iterated until the previously yielded item has been
2131
inv_w = self.inventories
2133
# file ids that changed
2134
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2136
num_file_ids = len(file_ids)
2137
for file_id, altered_versions in file_ids.iteritems():
2139
pb.update("fetch texts", count, num_file_ids)
2141
yield ("file", file_id, altered_versions)
2143
def _find_non_file_keys_to_fetch(self, revision_ids):
2145
yield ("inventory", None, revision_ids)
2148
# XXX: Note ATM no callers actually pay attention to this return
2149
# instead they just use the list of revision ids and ignore
2150
# missing sigs. Consider removing this work entirely
2151
revisions_with_signatures = set(self.signatures.get_parent_map(
2152
[(r,) for r in revision_ids]))
2153
revisions_with_signatures = set(
2154
[r for (r,) in revisions_with_signatures])
2155
revisions_with_signatures.intersection_update(revision_ids)
2156
yield ("signatures", None, revisions_with_signatures)
2159
yield ("revisions", None, revision_ids)
2162
def get_inventory(self, revision_id):
2163
"""Get Inventory object by revision id."""
2164
return self.iter_inventories([revision_id]).next()
2166
def iter_inventories(self, revision_ids):
2167
"""Get many inventories by revision_ids.
2169
This will buffer some or all of the texts used in constructing the
2170
inventories in memory, but will only parse a single inventory at a
2173
:param revision_ids: The expected revision ids of the inventories.
2174
:return: An iterator of inventories.
2176
if ((None in revision_ids)
2177
or (_mod_revision.NULL_REVISION in revision_ids)):
2178
raise ValueError('cannot get null revision inventory')
2179
return self._iter_inventories(revision_ids)
2181
def _iter_inventories(self, revision_ids):
2182
"""single-document based inventory iteration."""
2183
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2184
yield self.deserialise_inventory(revision_id, text)
2186
def _iter_inventory_xmls(self, revision_ids):
2187
keys = [(revision_id,) for revision_id in revision_ids]
2188
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2190
for record in stream:
2191
if record.storage_kind != 'absent':
2192
text_chunks[record.key] = record.get_bytes_as('chunked')
2194
raise errors.NoSuchRevision(self, record.key)
2196
chunks = text_chunks.pop(key)
2197
yield ''.join(chunks), key[-1]
2199
def deserialise_inventory(self, revision_id, xml):
2200
"""Transform the xml into an inventory object.
2202
:param revision_id: The expected revision id of the inventory.
2203
:param xml: A serialised inventory.
2205
result = self._serializer.read_inventory_from_string(xml, revision_id,
2206
entry_cache=self._inventory_entry_cache)
2207
if result.revision_id != revision_id:
2208
raise AssertionError('revision id mismatch %s != %s' % (
2209
result.revision_id, revision_id))
2212
def serialise_inventory(self, inv):
2213
return self._serializer.write_inventory_to_string(inv)
2215
def _serialise_inventory_to_lines(self, inv):
2216
return self._serializer.write_inventory_to_lines(inv)
2218
def get_serializer_format(self):
2219
return self._serializer.format_num
2222
def get_inventory_xml(self, revision_id):
2223
"""Get inventory XML as a file object."""
2224
texts = self._iter_inventory_xmls([revision_id])
2226
text, revision_id = texts.next()
2227
except StopIteration:
2228
raise errors.HistoryMissing(self, 'inventory', revision_id)
2232
def get_inventory_sha1(self, revision_id):
2233
"""Return the sha1 hash of the inventory entry
2235
return self.get_revision(revision_id).inventory_sha1
995
2237
def iter_reverse_revision_history(self, revision_id):
996
2238
"""Iterate backwards through revision ids in the lefthand history
1277
2525
except UnicodeDecodeError:
1278
2526
raise errors.NonAsciiRevisionId(method, self)
2528
def revision_graph_can_have_wrong_parents(self):
2529
"""Is it possible for this repository to have a revision graph with
2532
If True, then this repository must also implement
2533
_find_inconsistent_revision_parents so that check and reconcile can
2534
check for inconsistencies before proceeding with other checks that may
2535
depend on the revision index being consistent.
2537
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2540
# remove these delegates a while after bzr 0.15
2541
def __make_delegated(name, from_module):
2542
def _deprecated_repository_forwarder():
2543
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2544
% (name, from_module),
2547
m = __import__(from_module, globals(), locals(), [name])
2549
return getattr(m, name)
2550
except AttributeError:
2551
raise AttributeError('module %s has no name %s'
2553
globals()[name] = _deprecated_repository_forwarder
2556
'AllInOneRepository',
2557
'WeaveMetaDirRepository',
2558
'PreSplitOutRepositoryFormat',
2559
'RepositoryFormat4',
2560
'RepositoryFormat5',
2561
'RepositoryFormat6',
2562
'RepositoryFormat7',
2564
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2568
'RepositoryFormatKnit',
2569
'RepositoryFormatKnit1',
2571
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2574
def install_revision(repository, rev, revision_tree):
2575
"""Install all revision data into a repository."""
2576
install_revisions(repository, [(rev, revision_tree, None)])
2579
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2580
"""Install all revision data into a repository.
2582
Accepts an iterable of revision, tree, signature tuples. The signature
2585
repository.start_write_group()
2587
inventory_cache = lru_cache.LRUCache(10)
2588
for n, (revision, revision_tree, signature) in enumerate(iterable):
2589
_install_revision(repository, revision, revision_tree, signature,
2592
pb.update('Transferring revisions', n + 1, num_revisions)
2594
repository.abort_write_group()
2597
repository.commit_write_group()
2600
def _install_revision(repository, rev, revision_tree, signature,
2602
"""Install all revision data into a repository."""
2603
present_parents = []
2605
for p_id in rev.parent_ids:
2606
if repository.has_revision(p_id):
2607
present_parents.append(p_id)
2608
parent_trees[p_id] = repository.revision_tree(p_id)
2610
parent_trees[p_id] = repository.revision_tree(
2611
_mod_revision.NULL_REVISION)
2613
inv = revision_tree.inventory
2614
entries = inv.iter_entries()
2615
# backwards compatibility hack: skip the root id.
2616
if not repository.supports_rich_root():
2617
path, root = entries.next()
2618
if root.revision != rev.revision_id:
2619
raise errors.IncompatibleRevision(repr(repository))
2621
for path, ie in entries:
2622
text_keys[(ie.file_id, ie.revision)] = ie
2623
text_parent_map = repository.texts.get_parent_map(text_keys)
2624
missing_texts = set(text_keys) - set(text_parent_map)
2625
# Add the texts that are not already present
2626
for text_key in missing_texts:
2627
ie = text_keys[text_key]
2629
# FIXME: TODO: The following loop overlaps/duplicates that done by
2630
# commit to determine parents. There is a latent/real bug here where
2631
# the parents inserted are not those commit would do - in particular
2632
# they are not filtered by heads(). RBC, AB
2633
for revision, tree in parent_trees.iteritems():
2634
if ie.file_id not in tree:
2636
parent_id = tree.inventory[ie.file_id].revision
2637
if parent_id in text_parents:
2639
text_parents.append((ie.file_id, parent_id))
2640
lines = revision_tree.get_file(ie.file_id).readlines()
2641
repository.texts.add_lines(text_key, text_parents, lines)
2643
# install the inventory
2644
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2645
# Cache this inventory
2646
inventory_cache[rev.revision_id] = inv
2648
basis_inv = inventory_cache[rev.parent_ids[0]]
2650
repository.add_inventory(rev.revision_id, inv, present_parents)
2652
delta = inv._make_delta(basis_inv)
2653
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2654
rev.revision_id, present_parents)
2656
repository.add_inventory(rev.revision_id, inv, present_parents)
2657
except errors.RevisionAlreadyPresent:
2659
if signature is not None:
2660
repository.add_signature_text(rev.revision_id, signature)
2661
repository.add_revision(rev.revision_id, rev, inv)
1281
2664
class MetaDirRepository(Repository):
1282
2665
"""Repositories in the new meta-dir layout.
1773
3222
"different serializers")
3225
class InterSameDataRepository(InterRepository):
3226
"""Code for converting between repositories that represent the same data.
3228
Data format and model must match for this to work.
3232
def _get_repo_format_to_test(self):
3233
"""Repository format for testing with.
3235
InterSameData can pull from subtree to subtree and from non-subtree to
3236
non-subtree, so we test this with the richest repository format.
3238
from bzrlib.repofmt import knitrepo
3239
return knitrepo.RepositoryFormatKnit3()
3242
def is_compatible(source, target):
3243
return InterRepository._same_model(source, target)
3246
class InterWeaveRepo(InterSameDataRepository):
3247
"""Optimised code paths between Weave based repositories.
3249
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3250
implemented lazy inter-object optimisation.
3254
def _get_repo_format_to_test(self):
3255
from bzrlib.repofmt import weaverepo
3256
return weaverepo.RepositoryFormat7()
3259
def is_compatible(source, target):
3260
"""Be compatible with known Weave formats.
3262
We don't test for the stores being of specific types because that
3263
could lead to confusing results, and there is no need to be
3266
from bzrlib.repofmt.weaverepo import (
3272
return (isinstance(source._format, (RepositoryFormat5,
3274
RepositoryFormat7)) and
3275
isinstance(target._format, (RepositoryFormat5,
3277
RepositoryFormat7)))
3278
except AttributeError:
3282
def copy_content(self, revision_id=None):
3283
"""See InterRepository.copy_content()."""
3284
# weave specific optimised path:
3286
self.target.set_make_working_trees(self.source.make_working_trees())
3287
except (errors.RepositoryUpgradeRequired, NotImplemented):
3289
# FIXME do not peek!
3290
if self.source._transport.listable():
3291
pb = ui.ui_factory.nested_progress_bar()
3293
self.target.texts.insert_record_stream(
3294
self.source.texts.get_record_stream(
3295
self.source.texts.keys(), 'topological', False))
3296
pb.update('copying inventory', 0, 1)
3297
self.target.inventories.insert_record_stream(
3298
self.source.inventories.get_record_stream(
3299
self.source.inventories.keys(), 'topological', False))
3300
self.target.signatures.insert_record_stream(
3301
self.source.signatures.get_record_stream(
3302
self.source.signatures.keys(),
3304
self.target.revisions.insert_record_stream(
3305
self.source.revisions.get_record_stream(
3306
self.source.revisions.keys(),
3307
'topological', True))
3311
self.target.fetch(self.source, revision_id=revision_id)
3314
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3315
"""See InterRepository.missing_revision_ids()."""
3316
# we want all revisions to satisfy revision_id in source.
3317
# but we don't want to stat every file here and there.
3318
# we want then, all revisions other needs to satisfy revision_id
3319
# checked, but not those that we have locally.
3320
# so the first thing is to get a subset of the revisions to
3321
# satisfy revision_id in source, and then eliminate those that
3322
# we do already have.
3323
# this is slow on high latency connection to self, but as this
3324
# disk format scales terribly for push anyway due to rewriting
3325
# inventory.weave, this is considered acceptable.
3327
if revision_id is not None:
3328
source_ids = self.source.get_ancestry(revision_id)
3329
if source_ids[0] is not None:
3330
raise AssertionError()
3333
source_ids = self.source._all_possible_ids()
3334
source_ids_set = set(source_ids)
3335
# source_ids is the worst possible case we may need to pull.
3336
# now we want to filter source_ids against what we actually
3337
# have in target, but don't try to check for existence where we know
3338
# we do not have a revision as that would be pointless.
3339
target_ids = set(self.target._all_possible_ids())
3340
possibly_present_revisions = target_ids.intersection(source_ids_set)
3341
actually_present_revisions = set(
3342
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3343
required_revisions = source_ids_set.difference(actually_present_revisions)
3344
if revision_id is not None:
3345
# we used get_ancestry to determine source_ids then we are assured all
3346
# revisions referenced are present as they are installed in topological order.
3347
# and the tip revision was validated by get_ancestry.
3348
result_set = required_revisions
3350
# if we just grabbed the possibly available ids, then
3351
# we only have an estimate of whats available and need to validate
3352
# that against the revision records.
3354
self.source._eliminate_revisions_not_present(required_revisions))
3355
return self.source.revision_ids_to_search_result(result_set)
3358
class InterKnitRepo(InterSameDataRepository):
3359
"""Optimised code paths between Knit based repositories."""
3362
def _get_repo_format_to_test(self):
3363
from bzrlib.repofmt import knitrepo
3364
return knitrepo.RepositoryFormatKnit1()
3367
def is_compatible(source, target):
3368
"""Be compatible with known Knit formats.
3370
We don't test for the stores being of specific types because that
3371
could lead to confusing results, and there is no need to be
3374
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3376
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3377
isinstance(target._format, RepositoryFormatKnit))
3378
except AttributeError:
3380
return are_knits and InterRepository._same_model(source, target)
3383
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3384
"""See InterRepository.missing_revision_ids()."""
3385
if revision_id is not None:
3386
source_ids = self.source.get_ancestry(revision_id)
3387
if source_ids[0] is not None:
3388
raise AssertionError()
3391
source_ids = self.source.all_revision_ids()
3392
source_ids_set = set(source_ids)
3393
# source_ids is the worst possible case we may need to pull.
3394
# now we want to filter source_ids against what we actually
3395
# have in target, but don't try to check for existence where we know
3396
# we do not have a revision as that would be pointless.
3397
target_ids = set(self.target.all_revision_ids())
3398
possibly_present_revisions = target_ids.intersection(source_ids_set)
3399
actually_present_revisions = set(
3400
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3401
required_revisions = source_ids_set.difference(actually_present_revisions)
3402
if revision_id is not None:
3403
# we used get_ancestry to determine source_ids then we are assured all
3404
# revisions referenced are present as they are installed in topological order.
3405
# and the tip revision was validated by get_ancestry.
3406
result_set = required_revisions
3408
# if we just grabbed the possibly available ids, then
3409
# we only have an estimate of whats available and need to validate
3410
# that against the revision records.
3412
self.source._eliminate_revisions_not_present(required_revisions))
3413
return self.source.revision_ids_to_search_result(result_set)
3416
class InterPackRepo(InterSameDataRepository):
3417
"""Optimised code paths between Pack based repositories."""
3420
def _get_repo_format_to_test(self):
3421
from bzrlib.repofmt import pack_repo
3422
return pack_repo.RepositoryFormatKnitPack6RichRoot()
3425
def is_compatible(source, target):
3426
"""Be compatible with known Pack formats.
3428
We don't test for the stores being of specific types because that
3429
could lead to confusing results, and there is no need to be
3432
InterPackRepo does not support CHK based repositories.
3434
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3435
from bzrlib.repofmt.groupcompress_repo import RepositoryFormatCHK1
3437
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3438
isinstance(target._format, RepositoryFormatPack))
3439
not_packs = (isinstance(source._format, RepositoryFormatCHK1) or
3440
isinstance(target._format, RepositoryFormatCHK1))
3441
except AttributeError:
3443
if not_packs or not are_packs:
3445
return InterRepository._same_model(source, target)
3448
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3450
"""See InterRepository.fetch()."""
3451
if (len(self.source._fallback_repositories) > 0 or
3452
len(self.target._fallback_repositories) > 0):
3453
# The pack layer is not aware of fallback repositories, so when
3454
# fetching from a stacked repository or into a stacked repository
3455
# we use the generic fetch logic which uses the VersionedFiles
3456
# attributes on repository.
3457
from bzrlib.fetch import RepoFetcher
3458
fetcher = RepoFetcher(self.target, self.source, revision_id,
3459
pb, find_ghosts, fetch_spec=fetch_spec)
3460
if fetch_spec is not None:
3461
if len(list(fetch_spec.heads)) != 1:
3462
raise AssertionError(
3463
"InterPackRepo.fetch doesn't support "
3464
"fetching multiple heads yet.")
3465
revision_id = list(fetch_spec.heads)[0]
3467
if revision_id is None:
3469
# everything to do - use pack logic
3470
# to fetch from all packs to one without
3471
# inventory parsing etc, IFF nothing to be copied is in the target.
3473
source_revision_ids = frozenset(self.source.all_revision_ids())
3474
revision_ids = source_revision_ids - \
3475
frozenset(self.target.get_parent_map(source_revision_ids))
3476
revision_keys = [(revid,) for revid in revision_ids]
3477
index = self.target._pack_collection.revision_index.combined_index
3478
present_revision_ids = set(item[1][0] for item in
3479
index.iter_entries(revision_keys))
3480
revision_ids = set(revision_ids) - present_revision_ids
3481
# implementing the TODO will involve:
3482
# - detecting when all of a pack is selected
3483
# - avoiding as much as possible pre-selection, so the
3484
# more-core routines such as create_pack_from_packs can filter in
3485
# a just-in-time fashion. (though having a HEADS list on a
3486
# repository might make this a lot easier, because we could
3487
# sensibly detect 'new revisions' without doing a full index scan.
3488
elif _mod_revision.is_null(revision_id):
3492
revision_ids = self.search_missing_revision_ids(revision_id,
3493
find_ghosts=find_ghosts).get_keys()
3494
if len(revision_ids) == 0:
3496
return self._pack(self.source, self.target, revision_ids)
3498
def _pack(self, source, target, revision_ids):
3499
from bzrlib.repofmt.pack_repo import Packer
3500
packs = source._pack_collection.all_packs()
3501
pack = Packer(self.target._pack_collection, packs, '.fetch',
3502
revision_ids).pack()
3503
if pack is not None:
3504
self.target._pack_collection._save_pack_names()
3505
copied_revs = pack.get_revision_count()
3506
# Trigger an autopack. This may duplicate effort as we've just done
3507
# a pack creation, but for now it is simpler to think about as
3508
# 'upload data, then repack if needed'.
3509
self.target._pack_collection.autopack()
3510
return (copied_revs, [])
3515
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3516
"""See InterRepository.missing_revision_ids().
3518
:param find_ghosts: Find ghosts throughout the ancestry of
3521
if not find_ghosts and revision_id is not None:
3522
return self._walk_to_common_revisions([revision_id])
3523
elif revision_id is not None:
3524
# Find ghosts: search for revisions pointing from one repository to
3525
# the other, and vice versa, anywhere in the history of revision_id.
3526
graph = self.target.get_graph(other_repository=self.source)
3527
searcher = graph._make_breadth_first_searcher([revision_id])
3531
next_revs, ghosts = searcher.next_with_ghosts()
3532
except StopIteration:
3534
if revision_id in ghosts:
3535
raise errors.NoSuchRevision(self.source, revision_id)
3536
found_ids.update(next_revs)
3537
found_ids.update(ghosts)
3538
found_ids = frozenset(found_ids)
3539
# Double query here: should be able to avoid this by changing the
3540
# graph api further.
3541
result_set = found_ids - frozenset(
3542
self.target.get_parent_map(found_ids))
3544
source_ids = self.source.all_revision_ids()
3545
# source_ids is the worst possible case we may need to pull.
3546
# now we want to filter source_ids against what we actually
3547
# have in target, but don't try to check for existence where we know
3548
# we do not have a revision as that would be pointless.
3549
target_ids = set(self.target.all_revision_ids())
3550
result_set = set(source_ids).difference(target_ids)
3551
return self.source.revision_ids_to_search_result(result_set)
3554
class InterDifferingSerializer(InterRepository):
3557
def _get_repo_format_to_test(self):
3561
def is_compatible(source, target):
3562
"""Be compatible with Knit2 source and Knit3 target"""
3563
# This is redundant with format.check_conversion_target(), however that
3564
# raises an exception, and we just want to say "False" as in we won't
3565
# support converting between these formats.
3566
if source.supports_rich_root() and not target.supports_rich_root():
3568
if (source._format.supports_tree_reference
3569
and not target._format.supports_tree_reference):
3573
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3574
"""Get the best delta and base for this revision.
3576
:return: (basis_id, delta)
3578
possible_trees = [(parent_id, cache[parent_id])
3579
for parent_id in parent_ids
3580
if parent_id in cache]
3581
if len(possible_trees) == 0:
3582
# There either aren't any parents, or the parents aren't in the
3583
# cache, so just use the last converted tree
3584
possible_trees.append((basis_id, cache[basis_id]))
3586
for basis_id, basis_tree in possible_trees:
3587
delta = tree.inventory._make_delta(basis_tree.inventory)
3588
deltas.append((len(delta), basis_id, delta))
3590
return deltas[0][1:]
3592
def _get_parent_keys(self, root_key, parent_map):
3593
"""Get the parent keys for a given root id."""
3594
root_id, rev_id = root_key
3595
# Include direct parents of the revision, but only if they used
3596
# the same root_id and are heads.
3598
for parent_id in parent_map[rev_id]:
3599
if parent_id == _mod_revision.NULL_REVISION:
3601
if parent_id not in self._revision_id_to_root_id:
3602
# We probably didn't read this revision, go spend the
3603
# extra effort to actually check
3605
tree = self.source.revision_tree(parent_id)
3606
except errors.NoSuchRevision:
3607
# Ghost, fill out _revision_id_to_root_id in case we
3608
# encounter this again.
3609
# But set parent_root_id to None since we don't really know
3610
parent_root_id = None
3612
parent_root_id = tree.get_root_id()
3613
self._revision_id_to_root_id[parent_id] = None
3615
parent_root_id = self._revision_id_to_root_id[parent_id]
3616
if root_id == parent_root_id:
3617
# With stacking we _might_ want to refer to a non-local
3618
# revision, but this code path only applies when we have the
3619
# full content available, so ghosts really are ghosts, not just
3620
# the edge of local data.
3621
parent_keys.append((parent_id,))
3623
# root_id may be in the parent anyway.
3625
tree = self.source.revision_tree(parent_id)
3626
except errors.NoSuchRevision:
3627
# ghost, can't refer to it.
3631
parent_keys.append((tree.inventory[root_id].revision,))
3632
except errors.NoSuchId:
3635
g = graph.Graph(self.source.revisions)
3636
heads = g.heads(parent_keys)
3638
for key in parent_keys:
3639
if key in heads and key not in selected_keys:
3640
selected_keys.append(key)
3641
return tuple([(root_id,)+ key for key in selected_keys])
3643
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3644
for root_key in root_keys_to_create:
3645
parent_keys = self._get_parent_keys(root_key, parent_map)
3646
yield versionedfile.FulltextContentFactory(root_key,
3647
parent_keys, None, '')
3649
def _fetch_batch(self, revision_ids, basis_id, cache):
3650
"""Fetch across a few revisions.
3652
:param revision_ids: The revisions to copy
3653
:param basis_id: The revision_id of a tree that must be in cache, used
3654
as a basis for delta when no other base is available
3655
:param cache: A cache of RevisionTrees that we can use.
3656
:return: The revision_id of the last converted tree. The RevisionTree
3657
for it will be in cache
3659
# Walk though all revisions; get inventory deltas, copy referenced
3660
# texts that delta references, insert the delta, revision and
3662
root_keys_to_create = set()
3665
pending_revisions = []
3666
parent_map = self.source.get_parent_map(revision_ids)
3667
for tree in self.source.revision_trees(revision_ids):
3668
current_revision_id = tree.get_revision_id()
3669
parent_ids = parent_map.get(current_revision_id, ())
3670
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3672
if self._converting_to_rich_root:
3673
self._revision_id_to_root_id[current_revision_id] = \
3675
# Find text entries that need to be copied
3676
for old_path, new_path, file_id, entry in delta:
3677
if new_path is not None:
3680
if not self.target.supports_rich_root():
3681
# The target doesn't support rich root, so we don't
3684
if self._converting_to_rich_root:
3685
# This can't be copied normally, we have to insert
3687
root_keys_to_create.add((file_id, entry.revision))
3689
text_keys.add((file_id, entry.revision))
3690
revision = self.source.get_revision(current_revision_id)
3691
pending_deltas.append((basis_id, delta,
3692
current_revision_id, revision.parent_ids))
3693
pending_revisions.append(revision)
3694
cache[current_revision_id] = tree
3695
basis_id = current_revision_id
3697
from_texts = self.source.texts
3698
to_texts = self.target.texts
3699
if root_keys_to_create:
3700
root_stream = self._new_root_data_stream(root_keys_to_create,
3702
to_texts.insert_record_stream(root_stream)
3703
to_texts.insert_record_stream(from_texts.get_record_stream(
3704
text_keys, self.target._format._fetch_order,
3705
not self.target._format._fetch_uses_deltas))
3706
# insert inventory deltas
3707
for delta in pending_deltas:
3708
self.target.add_inventory_by_delta(*delta)
3709
if self.target._fallback_repositories:
3710
# Make sure this stacked repository has all the parent inventories
3711
# for the new revisions that we are about to insert. We do this
3712
# before adding the revisions so that no revision is added until
3713
# all the inventories it may depend on are added.
3715
revision_ids = set()
3716
for revision in pending_revisions:
3717
revision_ids.add(revision.revision_id)
3718
parent_ids.update(revision.parent_ids)
3719
parent_ids.difference_update(revision_ids)
3720
parent_ids.discard(_mod_revision.NULL_REVISION)
3721
parent_map = self.source.get_parent_map(parent_ids)
3722
for parent_tree in self.source.revision_trees(parent_ids):
3723
basis_id, delta = self._get_delta_for_revision(tree, parent_ids, basis_id, cache)
3724
current_revision_id = parent_tree.get_revision_id()
3725
parents_parents = parent_map[current_revision_id]
3726
self.target.add_inventory_by_delta(
3727
basis_id, delta, current_revision_id, parents_parents)
3728
# insert signatures and revisions
3729
for revision in pending_revisions:
3731
signature = self.source.get_signature_text(
3732
revision.revision_id)
3733
self.target.add_signature_text(revision.revision_id,
3735
except errors.NoSuchRevision:
3737
self.target.add_revision(revision.revision_id, revision)
3740
def _fetch_all_revisions(self, revision_ids, pb):
3741
"""Fetch everything for the list of revisions.
3743
:param revision_ids: The list of revisions to fetch. Must be in
3745
:param pb: A ProgressBar
3748
basis_id, basis_tree = self._get_basis(revision_ids[0])
3750
cache = lru_cache.LRUCache(100)
3751
cache[basis_id] = basis_tree
3752
del basis_tree # We don't want to hang on to it here
3753
for offset in range(0, len(revision_ids), batch_size):
3754
self.target.start_write_group()
3756
pb.update('Transferring revisions', offset,
3758
batch = revision_ids[offset:offset+batch_size]
3759
basis_id = self._fetch_batch(batch, basis_id, cache)
3761
self.target.abort_write_group()
3764
self.target.commit_write_group()
3765
pb.update('Transferring revisions', len(revision_ids),
3769
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3771
"""See InterRepository.fetch()."""
3772
if fetch_spec is not None:
3773
raise AssertionError("Not implemented yet...")
3774
if (not self.source.supports_rich_root()
3775
and self.target.supports_rich_root()):
3776
self._converting_to_rich_root = True
3777
self._revision_id_to_root_id = {}
3779
self._converting_to_rich_root = False
3780
revision_ids = self.target.search_missing_revision_ids(self.source,
3781
revision_id, find_ghosts=find_ghosts).get_keys()
3782
if not revision_ids:
3784
revision_ids = tsort.topo_sort(
3785
self.source.get_graph().get_parent_map(revision_ids))
3786
if not revision_ids:
3788
# Walk though all revisions; get inventory deltas, copy referenced
3789
# texts that delta references, insert the delta, revision and
3791
first_rev = self.source.get_revision(revision_ids[0])
3793
my_pb = ui.ui_factory.nested_progress_bar()
3796
symbol_versioning.warn(
3797
symbol_versioning.deprecated_in((1, 14, 0))
3798
% "pb parameter to fetch()")
3801
self._fetch_all_revisions(revision_ids, pb)
3803
if my_pb is not None:
3805
return len(revision_ids), 0
3807
def _get_basis(self, first_revision_id):
3808
"""Get a revision and tree which exists in the target.
3810
This assumes that first_revision_id is selected for transmission
3811
because all other ancestors are already present. If we can't find an
3812
ancestor we fall back to NULL_REVISION since we know that is safe.
3814
:return: (basis_id, basis_tree)
3816
first_rev = self.source.get_revision(first_revision_id)
3818
basis_id = first_rev.parent_ids[0]
3819
# only valid as a basis if the target has it
3820
self.target.get_revision(basis_id)
3821
# Try to get a basis tree - if its a ghost it will hit the
3822
# NoSuchRevision case.
3823
basis_tree = self.source.revision_tree(basis_id)
3824
except (IndexError, errors.NoSuchRevision):
3825
basis_id = _mod_revision.NULL_REVISION
3826
basis_tree = self.source.revision_tree(basis_id)
3827
return basis_id, basis_tree
3830
InterRepository.register_optimiser(InterDifferingSerializer)
3831
InterRepository.register_optimiser(InterSameDataRepository)
3832
InterRepository.register_optimiser(InterWeaveRepo)
3833
InterRepository.register_optimiser(InterKnitRepo)
3834
InterRepository.register_optimiser(InterPackRepo)
1776
3837
class CopyConverter(object):
1777
3838
"""A repository conversion tool which just performs a copy of the content.
1792
3853
:param to_convert: The disk object to convert.
1793
3854
:param pb: a progress bar to use for progress information.
1795
pb = ui.ui_factory.nested_progress_bar()
1798
3859
# this is only useful with metadir layouts - separated repo content.
1799
3860
# trigger an assertion if not such
1800
3861
repo._format.get_format_string()
1801
3862
self.repo_dir = repo.bzrdir
1802
pb.update('Moving repository to repository.backup')
3863
self.step('Moving repository to repository.backup')
1803
3864
self.repo_dir.transport.move('repository', 'repository.backup')
1804
3865
backup_transport = self.repo_dir.transport.clone('repository.backup')
1805
3866
repo._format.check_conversion_target(self.target_format)
1806
3867
self.source_repo = repo._format.open(self.repo_dir,
1808
3869
_override_transport=backup_transport)
1809
pb.update('Creating new repository')
3870
self.step('Creating new repository')
1810
3871
converted = self.target_format.initialize(self.repo_dir,
1811
3872
self.source_repo.is_shared())
1812
3873
converted.lock_write()
1814
pb.update('Copying content')
3875
self.step('Copying content into repository.')
1815
3876
self.source_repo.copy_content_into(converted)
1817
3878
converted.unlock()
1818
pb.update('Deleting old repository content')
3879
self.step('Deleting old repository content.')
1819
3880
self.repo_dir.transport.delete_tree('repository.backup')
1820
ui.ui_factory.note('repository converted')
3881
self.pb.note('repository converted')
3883
def step(self, message):
3884
"""Update the pb by a step."""
3886
self.pb.update(message, self.count, self.total)
3898
def _unescaper(match, _map=_unescape_map):
3899
code = match.group(1)
3903
if not code.startswith('#'):
3905
return unichr(int(code[1:])).encode('utf8')
3911
def _unescape_xml(data):
3912
"""Unescape predefined XML entities in a string of data."""
3914
if _unescape_re is None:
3915
_unescape_re = re.compile('\&([^;]*);')
3916
return _unescape_re.sub(_unescaper, data)
3919
class _VersionedFileChecker(object):
3921
def __init__(self, repository, text_key_references=None):
3922
self.repository = repository
3923
self.text_index = self.repository._generate_text_key_index(
3924
text_key_references=text_key_references)
3926
def calculate_file_version_parents(self, text_key):
3927
"""Calculate the correct parents for a file version according to
3930
parent_keys = self.text_index[text_key]
3931
if parent_keys == [_mod_revision.NULL_REVISION]:
3933
return tuple(parent_keys)
3935
def check_file_version_parents(self, texts, progress_bar=None):
3936
"""Check the parents stored in a versioned file are correct.
3938
It also detects file versions that are not referenced by their
3939
corresponding revision's inventory.
3941
:returns: A tuple of (wrong_parents, dangling_file_versions).
3942
wrong_parents is a dict mapping {revision_id: (stored_parents,
3943
correct_parents)} for each revision_id where the stored parents
3944
are not correct. dangling_file_versions is a set of (file_id,
3945
revision_id) tuples for versions that are present in this versioned
3946
file, but not used by the corresponding inventory.
3949
self.file_ids = set([file_id for file_id, _ in
3950
self.text_index.iterkeys()])
3951
# text keys is now grouped by file_id
3952
n_weaves = len(self.file_ids)
3953
files_in_revisions = {}
3954
revisions_of_files = {}
3955
n_versions = len(self.text_index)
3956
progress_bar.update('loading text store', 0, n_versions)
3957
parent_map = self.repository.texts.get_parent_map(self.text_index)
3958
# On unlistable transports this could well be empty/error...
3959
text_keys = self.repository.texts.keys()
3960
unused_keys = frozenset(text_keys) - set(self.text_index)
3961
for num, key in enumerate(self.text_index.iterkeys()):
3962
if progress_bar is not None:
3963
progress_bar.update('checking text graph', num, n_versions)
3964
correct_parents = self.calculate_file_version_parents(key)
3966
knit_parents = parent_map[key]
3967
except errors.RevisionNotPresent:
3970
if correct_parents != knit_parents:
3971
wrong_parents[key] = (knit_parents, correct_parents)
3972
return wrong_parents, unused_keys
3975
def _old_get_graph(repository, revision_id):
3976
"""DO NOT USE. That is all. I'm serious."""
3977
graph = repository.get_graph()
3978
revision_graph = dict(((key, value) for key, value in
3979
graph.iter_ancestry([revision_id]) if value is not None))
3980
return _strip_NULL_ghosts(revision_graph)
1824
3983
def _strip_NULL_ghosts(revision_graph):
1832
3991
return revision_graph
1835
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
1836
stop_revision=None):
1837
"""Extend the partial history to include a given index
1839
If a stop_index is supplied, stop when that index has been reached.
1840
If a stop_revision is supplied, stop when that revision is
1841
encountered. Otherwise, stop when the beginning of history is
1844
:param stop_index: The index which should be present. When it is
1845
present, history extension will stop.
1846
:param stop_revision: The revision id which should be present. When
1847
it is encountered, history extension will stop.
1849
start_revision = partial_history_cache[-1]
1850
graph = repo.get_graph()
1851
iterator = graph.iter_lefthand_ancestry(start_revision,
1852
(_mod_revision.NULL_REVISION,))
1854
# skip the last revision in the list
1857
if (stop_index is not None and
1858
len(partial_history_cache) > stop_index):
1860
if partial_history_cache[-1] == stop_revision:
1862
revision_id = iterator.next()
1863
partial_history_cache.append(revision_id)
1864
except StopIteration:
1869
class _LazyListJoin(object):
1870
"""An iterable yielding the contents of many lists as one list.
1872
Each iterator made from this will reflect the current contents of the lists
1873
at the time the iterator is made.
1875
This is used by Repository's _make_parents_provider implementation so that
1878
pp = repo._make_parents_provider() # uses a list of fallback repos
1879
pp.add_fallback_repository(other_repo) # appends to that list
1880
result = pp.get_parent_map(...)
1881
# The result will include revs from other_repo
1884
def __init__(self, *list_parts):
1885
self.list_parts = list_parts
1889
for list_part in self.list_parts:
1890
full_list.extend(list_part)
1891
return iter(full_list)
3994
class StreamSink(object):
3995
"""An object that can insert a stream into a repository.
3997
This interface handles the complexity of reserialising inventories and
3998
revisions from different formats, and allows unidirectional insertion into
3999
stacked repositories without looking for the missing basis parents
4003
def __init__(self, target_repo):
4004
self.target_repo = target_repo
4006
def insert_stream(self, stream, src_format, resume_tokens):
4007
"""Insert a stream's content into the target repository.
4009
:param src_format: a bzr repository format.
4011
:return: a list of resume tokens and an iterable of keys additional
4012
items required before the insertion can be completed.
4014
self.target_repo.lock_write()
4017
self.target_repo.resume_write_group(resume_tokens)
4020
self.target_repo.start_write_group()
4023
# locked_insert_stream performs a commit|suspend.
4024
return self._locked_insert_stream(stream, src_format, is_resume)
4026
self.target_repo.abort_write_group(suppress_errors=True)
4029
self.target_repo.unlock()
4031
def _locked_insert_stream(self, stream, src_format, is_resume):
4032
to_serializer = self.target_repo._format._serializer
4033
src_serializer = src_format._serializer
4035
if to_serializer == src_serializer:
4036
# If serializers match and the target is a pack repository, set the
4037
# write cache size on the new pack. This avoids poor performance
4038
# on transports where append is unbuffered (such as
4039
# RemoteTransport). This is safe to do because nothing should read
4040
# back from the target repository while a stream with matching
4041
# serialization is being inserted.
4042
# The exception is that a delta record from the source that should
4043
# be a fulltext may need to be expanded by the target (see
4044
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4045
# explicitly flush any buffered writes first in that rare case.
4047
new_pack = self.target_repo._pack_collection._new_pack
4048
except AttributeError:
4049
# Not a pack repository
4052
new_pack.set_write_cache_size(1024*1024)
4053
for substream_type, substream in stream:
4054
if substream_type == 'texts':
4055
self.target_repo.texts.insert_record_stream(substream)
4056
elif substream_type == 'inventories':
4057
if src_serializer == to_serializer:
4058
self.target_repo.inventories.insert_record_stream(
4061
self._extract_and_insert_inventories(
4062
substream, src_serializer)
4063
elif substream_type == 'chk_bytes':
4064
# XXX: This doesn't support conversions, as it assumes the
4065
# conversion was done in the fetch code.
4066
self.target_repo.chk_bytes.insert_record_stream(substream)
4067
elif substream_type == 'revisions':
4068
# This may fallback to extract-and-insert more often than
4069
# required if the serializers are different only in terms of
4071
if src_serializer == to_serializer:
4072
self.target_repo.revisions.insert_record_stream(
4075
self._extract_and_insert_revisions(substream,
4077
elif substream_type == 'signatures':
4078
self.target_repo.signatures.insert_record_stream(substream)
4080
raise AssertionError('kaboom! %s' % (substream_type,))
4081
# Done inserting data, and the missing_keys calculations will try to
4082
# read back from the inserted data, so flush the writes to the new pack
4083
# (if this is pack format).
4084
if new_pack is not None:
4085
new_pack._write_data('', flush=True)
4086
# Find all the new revisions (including ones from resume_tokens)
4087
missing_keys = self.target_repo.get_missing_parent_inventories(
4088
check_for_missing_texts=is_resume)
4090
for prefix, versioned_file in (
4091
('texts', self.target_repo.texts),
4092
('inventories', self.target_repo.inventories),
4093
('revisions', self.target_repo.revisions),
4094
('signatures', self.target_repo.signatures),
4095
('chk_bytes', self.target_repo.chk_bytes),
4097
if versioned_file is None:
4099
missing_keys.update((prefix,) + key for key in
4100
versioned_file.get_missing_compression_parent_keys())
4101
except NotImplementedError:
4102
# cannot even attempt suspending, and missing would have failed
4103
# during stream insertion.
4104
missing_keys = set()
4107
# suspend the write group and tell the caller what we is
4108
# missing. We know we can suspend or else we would not have
4109
# entered this code path. (All repositories that can handle
4110
# missing keys can handle suspending a write group).
4111
write_group_tokens = self.target_repo.suspend_write_group()
4112
return write_group_tokens, missing_keys
4113
self.target_repo.commit_write_group()
4116
def _extract_and_insert_inventories(self, substream, serializer):
4117
"""Generate a new inventory versionedfile in target, converting data.
4119
The inventory is retrieved from the source, (deserializing it), and
4120
stored in the target (reserializing it in a different format).
4122
for record in substream:
4123
bytes = record.get_bytes_as('fulltext')
4124
revision_id = record.key[0]
4125
inv = serializer.read_inventory_from_string(bytes, revision_id)
4126
parents = [key[0] for key in record.parents]
4127
self.target_repo.add_inventory(revision_id, inv, parents)
4129
def _extract_and_insert_revisions(self, substream, serializer):
4130
for record in substream:
4131
bytes = record.get_bytes_as('fulltext')
4132
revision_id = record.key[0]
4133
rev = serializer.read_revision_from_string(bytes)
4134
if rev.revision_id != revision_id:
4135
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4136
self.target_repo.add_revision(revision_id, rev)
4139
if self.target_repo._format._fetch_reconcile:
4140
self.target_repo.reconcile()
4143
class StreamSource(object):
4144
"""A source of a stream for fetching between repositories."""
4146
def __init__(self, from_repository, to_format):
4147
"""Create a StreamSource streaming from from_repository."""
4148
self.from_repository = from_repository
4149
self.to_format = to_format
4151
def delta_on_metadata(self):
4152
"""Return True if delta's are permitted on metadata streams.
4154
That is on revisions and signatures.
4156
src_serializer = self.from_repository._format._serializer
4157
target_serializer = self.to_format._serializer
4158
return (self.to_format._fetch_uses_deltas and
4159
src_serializer == target_serializer)
4161
def _fetch_revision_texts(self, revs):
4162
# fetch signatures first and then the revision texts
4163
# may need to be a InterRevisionStore call here.
4164
from_sf = self.from_repository.signatures
4165
# A missing signature is just skipped.
4166
keys = [(rev_id,) for rev_id in revs]
4167
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4169
self.to_format._fetch_order,
4170
not self.to_format._fetch_uses_deltas))
4171
# If a revision has a delta, this is actually expanded inside the
4172
# insert_record_stream code now, which is an alternate fix for
4174
from_rf = self.from_repository.revisions
4175
revisions = from_rf.get_record_stream(
4177
self.to_format._fetch_order,
4178
not self.delta_on_metadata())
4179
return [('signatures', signatures), ('revisions', revisions)]
4181
def _generate_root_texts(self, revs):
4182
"""This will be called by __fetch between fetching weave texts and
4183
fetching the inventory weave.
4185
Subclasses should override this if they need to generate root texts
4186
after fetching weave texts.
4188
if self._rich_root_upgrade():
4190
return bzrlib.fetch.Inter1and2Helper(
4191
self.from_repository).generate_root_texts(revs)
4195
def get_stream(self, search):
4197
revs = search.get_keys()
4198
graph = self.from_repository.get_graph()
4199
revs = list(graph.iter_topo_order(revs))
4200
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4202
for knit_kind, file_id, revisions in data_to_fetch:
4203
if knit_kind != phase:
4205
# Make a new progress bar for this phase
4206
if knit_kind == "file":
4207
# Accumulate file texts
4208
text_keys.extend([(file_id, revision) for revision in
4210
elif knit_kind == "inventory":
4211
# Now copy the file texts.
4212
from_texts = self.from_repository.texts
4213
yield ('texts', from_texts.get_record_stream(
4214
text_keys, self.to_format._fetch_order,
4215
not self.to_format._fetch_uses_deltas))
4216
# Cause an error if a text occurs after we have done the
4219
# Before we process the inventory we generate the root
4220
# texts (if necessary) so that the inventories references
4222
for _ in self._generate_root_texts(revs):
4224
# NB: This currently reopens the inventory weave in source;
4225
# using a single stream interface instead would avoid this.
4226
from_weave = self.from_repository.inventories
4227
# we fetch only the referenced inventories because we do not
4228
# know for unselected inventories whether all their required
4229
# texts are present in the other repository - it could be
4231
for info in self._get_inventory_stream(revs):
4233
elif knit_kind == "signatures":
4234
# Nothing to do here; this will be taken care of when
4235
# _fetch_revision_texts happens.
4237
elif knit_kind == "revisions":
4238
for record in self._fetch_revision_texts(revs):
4241
raise AssertionError("Unknown knit kind %r" % knit_kind)
4243
def get_stream_for_missing_keys(self, missing_keys):
4244
# missing keys can only occur when we are byte copying and not
4245
# translating (because translation means we don't send
4246
# unreconstructable deltas ever).
4248
keys['texts'] = set()
4249
keys['revisions'] = set()
4250
keys['inventories'] = set()
4251
keys['chk_bytes'] = set()
4252
keys['signatures'] = set()
4253
for key in missing_keys:
4254
keys[key[0]].add(key[1:])
4255
if len(keys['revisions']):
4256
# If we allowed copying revisions at this point, we could end up
4257
# copying a revision without copying its required texts: a
4258
# violation of the requirements for repository integrity.
4259
raise AssertionError(
4260
'cannot copy revisions to fill in missing deltas %s' % (
4261
keys['revisions'],))
4262
for substream_kind, keys in keys.iteritems():
4263
vf = getattr(self.from_repository, substream_kind)
4264
if vf is None and keys:
4265
raise AssertionError(
4266
"cannot fill in keys for a versioned file we don't"
4267
" have: %s needs %s" % (substream_kind, keys))
4269
# No need to stream something we don't have
4271
# Ask for full texts always so that we don't need more round trips
4272
# after this stream.
4273
# Some of the missing keys are genuinely ghosts, so filter absent
4274
# records. The Sink is responsible for doing another check to
4275
# ensure that ghosts don't introduce missing data for future
4277
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4278
self.to_format._fetch_order, True))
4279
yield substream_kind, stream
4281
def inventory_fetch_order(self):
4282
if self._rich_root_upgrade():
4283
return 'topological'
4285
return self.to_format._fetch_order
4287
def _rich_root_upgrade(self):
4288
return (not self.from_repository._format.rich_root_data and
4289
self.to_format.rich_root_data)
4291
def _get_inventory_stream(self, revision_ids):
4292
from_format = self.from_repository._format
4293
if (from_format.supports_chks and self.to_format.supports_chks
4294
and (from_format._serializer == self.to_format._serializer)):
4295
# Both sides support chks, and they use the same serializer, so it
4296
# is safe to transmit the chk pages and inventory pages across
4298
return self._get_chk_inventory_stream(revision_ids)
4299
elif (not from_format.supports_chks):
4300
# Source repository doesn't support chks. So we can transmit the
4301
# inventories 'as-is' and either they are just accepted on the
4302
# target, or the Sink will properly convert it.
4303
return self._get_simple_inventory_stream(revision_ids)
4305
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4306
# inventories. Note that this should probably be done somehow
4307
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4308
# figure out how a non-chk repository could possibly handle
4309
# deserializing an inventory stream from a chk repo, as it
4310
# doesn't have a way to understand individual pages.
4311
return self._get_convertable_inventory_stream(revision_ids)
4313
def _get_simple_inventory_stream(self, revision_ids):
4314
from_weave = self.from_repository.inventories
4315
yield ('inventories', from_weave.get_record_stream(
4316
[(rev_id,) for rev_id in revision_ids],
4317
self.inventory_fetch_order(),
4318
not self.delta_on_metadata()))
4320
def _get_chk_inventory_stream(self, revision_ids):
4321
"""Fetch the inventory texts, along with the associated chk maps."""
4322
# We want an inventory outside of the search set, so that we can filter
4323
# out uninteresting chk pages. For now we use
4324
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4325
# could use that instead.
4326
start_rev_id = self.from_repository._find_revision_outside_set(
4328
start_rev_key = (start_rev_id,)
4329
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4330
if start_rev_id != _mod_revision.NULL_REVISION:
4331
inv_keys_to_fetch.append((start_rev_id,))
4332
# Any repo that supports chk_bytes must also support out-of-order
4333
# insertion. At least, that is how we expect it to work
4334
# We use get_record_stream instead of iter_inventories because we want
4335
# to be able to insert the stream as well. We could instead fetch
4336
# allowing deltas, and then iter_inventories, but we don't know whether
4337
# source or target is more 'local' anway.
4338
inv_stream = self.from_repository.inventories.get_record_stream(
4339
inv_keys_to_fetch, 'unordered',
4340
True) # We need them as full-texts so we can find their references
4341
uninteresting_chk_roots = set()
4342
interesting_chk_roots = set()
4343
def filter_inv_stream(inv_stream):
4344
for idx, record in enumerate(inv_stream):
4345
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4346
bytes = record.get_bytes_as('fulltext')
4347
chk_inv = inventory.CHKInventory.deserialise(
4348
self.from_repository.chk_bytes, bytes, record.key)
4349
if record.key == start_rev_key:
4350
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4351
p_id_map = chk_inv.parent_id_basename_to_file_id
4352
if p_id_map is not None:
4353
uninteresting_chk_roots.add(p_id_map.key())
4356
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4357
p_id_map = chk_inv.parent_id_basename_to_file_id
4358
if p_id_map is not None:
4359
interesting_chk_roots.add(p_id_map.key())
4360
### pb.update('fetch inventory', 0, 2)
4361
yield ('inventories', filter_inv_stream(inv_stream))
4362
# Now that we have worked out all of the interesting root nodes, grab
4363
# all of the interesting pages and insert them
4364
### pb.update('fetch inventory', 1, 2)
4365
interesting = chk_map.iter_interesting_nodes(
4366
self.from_repository.chk_bytes, interesting_chk_roots,
4367
uninteresting_chk_roots)
4368
def to_stream_adapter():
4369
"""Adapt the iter_interesting_nodes result to a single stream.
4371
iter_interesting_nodes returns records as it processes them, along
4372
with keys. However, we only want to return the records themselves.
4374
for record, items in interesting:
4375
if record is not None:
4377
# XXX: We could instead call get_record_stream(records.keys())
4378
# ATM, this will always insert the records as fulltexts, and
4379
# requires that you can hang on to records once you have gone
4380
# on to the next one. Further, it causes the target to
4381
# recompress the data. Testing shows it to be faster than
4382
# requesting the records again, though.
4383
yield ('chk_bytes', to_stream_adapter())
4384
### pb.update('fetch inventory', 2, 2)
4386
def _get_convertable_inventory_stream(self, revision_ids):
4387
# XXX: One of source or target is using chks, and they don't have
4388
# compatible serializations. The StreamSink code expects to be
4389
# able to convert on the target, so we need to put
4390
# bytes-on-the-wire that can be converted
4391
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4393
def _stream_invs_as_fulltexts(self, revision_ids):
4394
from_repo = self.from_repository
4395
from_serializer = from_repo._format._serializer
4396
revision_keys = [(rev_id,) for rev_id in revision_ids]
4397
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4398
for inv in self.from_repository.iter_inventories(revision_ids):
4399
# XXX: This is a bit hackish, but it works. Basically,
4400
# CHKSerializer 'accidentally' supports
4401
# read/write_inventory_to_string, even though that is never
4402
# the format that is stored on disk. It *does* give us a
4403
# single string representation for an inventory, so live with
4405
# This would be far better if we had a 'serialized inventory
4406
# delta' form. Then we could use 'inventory._make_delta', and
4407
# transmit that. This would both be faster to generate, and
4408
# result in fewer bytes-on-the-wire.
4409
as_bytes = from_serializer.write_inventory_to_string(inv)
4410
key = (inv.revision_id,)
4411
parent_keys = parent_map.get(key, ())
4412
yield versionedfile.FulltextContentFactory(
4413
key, parent_keys, None, as_bytes)
added
removed
bzrlib/repository.py
