556
396
# unchanged, carry over.
557
397
ie.reference_revision = parent_entry.reference_revision
558
398
ie.revision = parent_entry.revision
559
return self._get_delta(ie, basis_inv, path), False, None
399
return self._get_delta(ie, basis_inv, path), False
560
400
ie.reference_revision = content_summary[3]
561
if ie.reference_revision is None:
562
raise AssertionError("invalid content_summary for nested tree: %r"
563
% (content_summary,))
564
self._add_text_to_weave(ie.file_id, '', heads, None)
402
self._add_text_to_weave(ie.file_id, lines, heads, None)
566
404
raise NotImplementedError('unknown kind')
567
405
ie.revision = self._new_revision_id
568
self._any_changes = True
569
return self._get_delta(ie, basis_inv, path), True, fingerprint
571
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
572
_entry_factory=entry_factory):
573
"""Record a new tree via iter_changes.
575
:param tree: The tree to obtain text contents from for changed objects.
576
:param basis_revision_id: The revision id of the tree the iter_changes
577
has been generated against. Currently assumed to be the same
578
as self.parents[0] - if it is not, errors may occur.
579
:param iter_changes: An iter_changes iterator with the changes to apply
580
to basis_revision_id. The iterator must not include any items with
581
a current kind of None - missing items must be either filtered out
582
or errored-on beefore record_iter_changes sees the item.
583
:param _entry_factory: Private method to bind entry_factory locally for
585
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
588
# Create an inventory delta based on deltas between all the parents and
589
# deltas between all the parent inventories. We use inventory delta's
590
# between the inventory objects because iter_changes masks
591
# last-changed-field only changes.
593
# file_id -> change map, change is fileid, paths, changed, versioneds,
594
# parents, names, kinds, executables
596
# {file_id -> revision_id -> inventory entry, for entries in parent
597
# trees that are not parents[0]
406
return self._get_delta(ie, basis_inv, path), True
408
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
409
versionedfile = self.repository.weave_store.get_weave_or_empty(
410
file_id, self.repository.get_transaction())
411
# Don't change this to add_lines - add_lines_with_ghosts is cheaper
412
# than add_lines, and allows committing when a parent is ghosted for
414
# Note: as we read the content directly from the tree, we know its not
415
# been turned into unicode or badly split - but a broken tree
416
# implementation could give us bad output from readlines() so this is
417
# not a guarantee of safety. What would be better is always checking
418
# the content during test suite execution. RBC 20070912
601
revtrees = list(self.repository.revision_trees(self.parents))
602
except errors.NoSuchRevision:
603
# one or more ghosts, slow path.
605
for revision_id in self.parents:
607
revtrees.append(self.repository.revision_tree(revision_id))
608
except errors.NoSuchRevision:
610
basis_revision_id = _mod_revision.NULL_REVISION
612
revtrees.append(self.repository.revision_tree(
613
_mod_revision.NULL_REVISION))
614
# The basis inventory from a repository
616
basis_inv = revtrees[0].inventory
618
basis_inv = self.repository.revision_tree(
619
_mod_revision.NULL_REVISION).inventory
620
if len(self.parents) > 0:
621
if basis_revision_id != self.parents[0] and not ghost_basis:
623
"arbitrary basis parents not yet supported with merges")
624
for revtree in revtrees[1:]:
625
for change in revtree.inventory._make_delta(basis_inv):
626
if change[1] is None:
627
# Not present in this parent.
629
if change[2] not in merged_ids:
630
if change[0] is not None:
631
basis_entry = basis_inv[change[2]]
632
merged_ids[change[2]] = [
634
basis_entry.revision,
637
parent_entries[change[2]] = {
639
basis_entry.revision:basis_entry,
641
change[3].revision:change[3],
644
merged_ids[change[2]] = [change[3].revision]
645
parent_entries[change[2]] = {change[3].revision:change[3]}
647
merged_ids[change[2]].append(change[3].revision)
648
parent_entries[change[2]][change[3].revision] = change[3]
651
# Setup the changes from the tree:
652
# changes maps file_id -> (change, [parent revision_ids])
654
for change in iter_changes:
655
# This probably looks up in basis_inv way to much.
656
if change[1][0] is not None:
657
head_candidate = [basis_inv[change[0]].revision]
660
changes[change[0]] = change, merged_ids.get(change[0],
662
unchanged_merged = set(merged_ids) - set(changes)
663
# Extend the changes dict with synthetic changes to record merges of
665
for file_id in unchanged_merged:
666
# Record a merged version of these items that did not change vs the
667
# basis. This can be either identical parallel changes, or a revert
668
# of a specific file after a merge. The recorded content will be
669
# that of the current tree (which is the same as the basis), but
670
# the per-file graph will reflect a merge.
671
# NB:XXX: We are reconstructing path information we had, this
672
# should be preserved instead.
673
# inv delta change: (file_id, (path_in_source, path_in_target),
674
# changed_content, versioned, parent, name, kind,
677
basis_entry = basis_inv[file_id]
678
except errors.NoSuchId:
679
# a change from basis->some_parents but file_id isn't in basis
680
# so was new in the merge, which means it must have changed
681
# from basis -> current, and as it hasn't the add was reverted
682
# by the user. So we discard this change.
686
(basis_inv.id2path(file_id), tree.id2path(file_id)),
688
(basis_entry.parent_id, basis_entry.parent_id),
689
(basis_entry.name, basis_entry.name),
690
(basis_entry.kind, basis_entry.kind),
691
(basis_entry.executable, basis_entry.executable))
692
changes[file_id] = (change, merged_ids[file_id])
693
# changes contains tuples with the change and a set of inventory
694
# candidates for the file.
696
# old_path, new_path, file_id, new_inventory_entry
697
seen_root = False # Is the root in the basis delta?
698
inv_delta = self._basis_delta
699
modified_rev = self._new_revision_id
700
for change, head_candidates in changes.values():
701
if change[3][1]: # versioned in target.
702
# Several things may be happening here:
703
# We may have a fork in the per-file graph
704
# - record a change with the content from tree
705
# We may have a change against < all trees
706
# - carry over the tree that hasn't changed
707
# We may have a change against all trees
708
# - record the change with the content from tree
711
entry = _entry_factory[kind](file_id, change[5][1],
713
head_set = self._heads(change[0], set(head_candidates))
716
for head_candidate in head_candidates:
717
if head_candidate in head_set:
718
heads.append(head_candidate)
719
head_set.remove(head_candidate)
722
# Could be a carry-over situation:
723
parent_entry_revs = parent_entries.get(file_id, None)
724
if parent_entry_revs:
725
parent_entry = parent_entry_revs.get(heads[0], None)
728
if parent_entry is None:
729
# The parent iter_changes was called against is the one
730
# that is the per-file head, so any change is relevant
731
# iter_changes is valid.
732
carry_over_possible = False
734
# could be a carry over situation
735
# A change against the basis may just indicate a merge,
736
# we need to check the content against the source of the
737
# merge to determine if it was changed after the merge
739
if (parent_entry.kind != entry.kind or
740
parent_entry.parent_id != entry.parent_id or
741
parent_entry.name != entry.name):
742
# Metadata common to all entries has changed
743
# against per-file parent
744
carry_over_possible = False
746
carry_over_possible = True
747
# per-type checks for changes against the parent_entry
750
# Cannot be a carry-over situation
751
carry_over_possible = False
752
# Populate the entry in the delta
754
# XXX: There is still a small race here: If someone reverts the content of a file
755
# after iter_changes examines and decides it has changed,
756
# we will unconditionally record a new version even if some
757
# other process reverts it while commit is running (with
758
# the revert happening after iter_changes did it's
761
entry.executable = True
763
entry.executable = False
764
if (carry_over_possible and
765
parent_entry.executable == entry.executable):
766
# Check the file length, content hash after reading
768
nostore_sha = parent_entry.text_sha1
771
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
773
text = file_obj.read()
777
entry.text_sha1, entry.text_size = self._add_text_to_weave(
778
file_id, text, heads, nostore_sha)
779
yield file_id, change[1][1], (entry.text_sha1, stat_value)
780
except errors.ExistingContent:
781
# No content change against a carry_over parent
782
# Perhaps this should also yield a fs hash update?
784
entry.text_size = parent_entry.text_size
785
entry.text_sha1 = parent_entry.text_sha1
786
elif kind == 'symlink':
788
entry.symlink_target = tree.get_symlink_target(file_id)
789
if (carry_over_possible and
790
parent_entry.symlink_target == entry.symlink_target):
793
self._add_text_to_weave(change[0], '', heads, None)
794
elif kind == 'directory':
795
if carry_over_possible:
798
# Nothing to set on the entry.
799
# XXX: split into the Root and nonRoot versions.
800
if change[1][1] != '' or self.repository.supports_rich_root():
801
self._add_text_to_weave(change[0], '', heads, None)
802
elif kind == 'tree-reference':
803
if not self.repository._format.supports_tree_reference:
804
# This isn't quite sane as an error, but we shouldn't
805
# ever see this code path in practice: tree's don't
806
# permit references when the repo doesn't support tree
808
raise errors.UnsupportedOperation(tree.add_reference,
810
reference_revision = tree.get_reference_revision(change[0])
811
entry.reference_revision = reference_revision
812
if (carry_over_possible and
813
parent_entry.reference_revision == reference_revision):
816
self._add_text_to_weave(change[0], '', heads, None)
818
raise AssertionError('unknown kind %r' % kind)
820
entry.revision = modified_rev
822
entry.revision = parent_entry.revision
825
new_path = change[1][1]
826
inv_delta.append((change[1][0], new_path, change[0], entry))
829
self.new_inventory = None
831
# This should perhaps be guarded by a check that the basis we
832
# commit against is the basis for the commit and if not do a delta
834
self._any_changes = True
836
# housekeeping root entry changes do not affect no-change commits.
837
self._require_root_change(tree)
838
self.basis_delta_revision = basis_revision_id
840
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
841
parent_keys = tuple([(file_id, parent) for parent in parents])
842
return self.repository.texts._add_text(
843
(file_id, self._new_revision_id), parent_keys, new_text,
844
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
420
return versionedfile.add_lines_with_ghosts(
421
self._new_revision_id, parents, new_lines,
422
nostore_sha=nostore_sha, random_id=self.random_revid,
423
check_content=False)[0:2]
425
versionedfile.clear_cache()
847
428
class RootCommitBuilder(CommitBuilder):
848
429
"""This commitbuilder actually records the root id"""
850
431
# the root entry gets versioned properly by this builder.
851
432
_versioned_root = True
1001
def add_fallback_repository(self, repository):
1002
"""Add a repository to use for looking up data not held locally.
1004
:param repository: A repository.
1006
if not self._format.supports_external_lookups:
1007
raise errors.UnstackableRepositoryFormat(self._format, self.base)
1008
if self.is_locked():
1009
# This repository will call fallback.unlock() when we transition to
1010
# the unlocked state, so we make sure to increment the lock count
1011
repository.lock_read()
1012
self._check_fallback_repository(repository)
1013
self._fallback_repositories.append(repository)
1014
self.texts.add_fallback_versioned_files(repository.texts)
1015
self.inventories.add_fallback_versioned_files(repository.inventories)
1016
self.revisions.add_fallback_versioned_files(repository.revisions)
1017
self.signatures.add_fallback_versioned_files(repository.signatures)
1018
if self.chk_bytes is not None:
1019
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
1021
def _check_fallback_repository(self, repository):
1022
"""Check that this repository can fallback to repository safely.
1024
Raise an error if not.
1026
:param repository: A repository to fallback to.
1028
return InterRepository._assert_same_model(self, repository)
1030
495
def add_inventory(self, revision_id, inv, parents):
1031
496
"""Add the inventory inv to the repository as revision_id.
1033
498
:param parents: The revision ids of the parents that revision_id
1034
499
is known to have and are in the repository already.
1036
:returns: The validator(which is a sha1 digest, though what is sha'd is
1037
repository format specific) of the serialized inventory.
501
returns the sha1 of the serialized inventory.
1039
if not self.is_in_write_group():
1040
raise AssertionError("%r not in write group" % (self,))
1041
503
_mod_revision.check_not_reserved_id(revision_id)
1042
if not (inv.revision_id is None or inv.revision_id == revision_id):
1043
raise AssertionError(
1044
"Mismatch between inventory revision"
1045
" id and insertion revid (%r, %r)"
1046
% (inv.revision_id, revision_id))
1047
if inv.root is None:
1048
raise AssertionError()
1049
return self._add_inventory_checked(revision_id, inv, parents)
1051
def _add_inventory_checked(self, revision_id, inv, parents):
1052
"""Add inv to the repository after checking the inputs.
1054
This function can be overridden to allow different inventory styles.
1056
:seealso: add_inventory, for the contract.
1058
inv_lines = self._serializer.write_inventory_to_lines(inv)
1059
return self._inventory_add_lines(revision_id, parents,
504
assert inv.revision_id is None or inv.revision_id == revision_id, \
505
"Mismatch between inventory revision" \
506
" id and insertion revid (%r, %r)" % (inv.revision_id, revision_id)
507
assert inv.root is not None
508
inv_lines = self._serialise_inventory_to_lines(inv)
509
inv_vf = self.get_inventory_weave()
510
return self._inventory_add_lines(inv_vf, revision_id, parents,
1060
511
inv_lines, check_content=False)
1062
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1063
parents, basis_inv=None, propagate_caches=False):
1064
"""Add a new inventory expressed as a delta against another revision.
1066
See the inventory developers documentation for the theory behind
1069
:param basis_revision_id: The inventory id the delta was created
1070
against. (This does not have to be a direct parent.)
1071
:param delta: The inventory delta (see Inventory.apply_delta for
1073
:param new_revision_id: The revision id that the inventory is being
1075
:param parents: The revision ids of the parents that revision_id is
1076
known to have and are in the repository already. These are supplied
1077
for repositories that depend on the inventory graph for revision
1078
graph access, as well as for those that pun ancestry with delta
1080
:param basis_inv: The basis inventory if it is already known,
1082
:param propagate_caches: If True, the caches for this inventory are
1083
copied to and updated for the result if possible.
1085
:returns: (validator, new_inv)
1086
The validator(which is a sha1 digest, though what is sha'd is
1087
repository format specific) of the serialized inventory, and the
1088
resulting inventory.
1090
if not self.is_in_write_group():
1091
raise AssertionError("%r not in write group" % (self,))
1092
_mod_revision.check_not_reserved_id(new_revision_id)
1093
basis_tree = self.revision_tree(basis_revision_id)
1094
basis_tree.lock_read()
1096
# Note that this mutates the inventory of basis_tree, which not all
1097
# inventory implementations may support: A better idiom would be to
1098
# return a new inventory, but as there is no revision tree cache in
1099
# repository this is safe for now - RBC 20081013
1100
if basis_inv is None:
1101
basis_inv = basis_tree.inventory
1102
basis_inv.apply_delta(delta)
1103
basis_inv.revision_id = new_revision_id
1104
return (self.add_inventory(new_revision_id, basis_inv, parents),
1109
def _inventory_add_lines(self, revision_id, parents, lines,
513
def _inventory_add_lines(self, inv_vf, revision_id, parents, lines,
1110
514
check_content=True):
1111
515
"""Store lines in inv_vf and return the sha1 of the inventory."""
1112
parents = [(parent,) for parent in parents]
1113
result = self.inventories.add_lines((revision_id,), parents, lines,
517
for parent in parents:
519
final_parents.append(parent)
520
return inv_vf.add_lines(revision_id, final_parents, lines,
1114
521
check_content=check_content)[0]
1115
self.inventories._access.flush()
1118
524
def add_revision(self, revision_id, rev, inv=None, config=None):
1119
525
"""Add rev to the revision store as revision_id.
1190
590
Returns a set of the present revisions.
1193
graph = self.get_graph()
1194
parent_map = graph.get_parent_map(revision_ids)
1195
# The old API returned a list, should this actually be a set?
1196
return parent_map.keys()
1198
def _check_inventories(self, checker):
1199
"""Check the inventories found from the revision scan.
1201
This is responsible for verifying the sha1 of inventories and
1202
creating a pending_keys set that covers data referenced by inventories.
1204
bar = ui.ui_factory.nested_progress_bar()
1206
self._do_check_inventories(checker, bar)
1210
def _do_check_inventories(self, checker, bar):
1211
"""Helper for _check_inventories."""
1213
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1214
kinds = ['chk_bytes', 'texts']
1215
count = len(checker.pending_keys)
1216
bar.update("inventories", 0, 2)
1217
current_keys = checker.pending_keys
1218
checker.pending_keys = {}
1219
# Accumulate current checks.
1220
for key in current_keys:
1221
if key[0] != 'inventories' and key[0] not in kinds:
1222
checker._report_items.append('unknown key type %r' % (key,))
1223
keys[key[0]].add(key[1:])
1224
if keys['inventories']:
1225
# NB: output order *should* be roughly sorted - topo or
1226
# inverse topo depending on repository - either way decent
1227
# to just delta against. However, pre-CHK formats didn't
1228
# try to optimise inventory layout on disk. As such the
1229
# pre-CHK code path does not use inventory deltas.
1231
for record in self.inventories.check(keys=keys['inventories']):
1232
if record.storage_kind == 'absent':
1233
checker._report_items.append(
1234
'Missing inventory {%s}' % (record.key,))
1236
last_object = self._check_record('inventories', record,
1237
checker, last_object,
1238
current_keys[('inventories',) + record.key])
1239
del keys['inventories']
1242
bar.update("texts", 1)
1243
while (checker.pending_keys or keys['chk_bytes']
1245
# Something to check.
1246
current_keys = checker.pending_keys
1247
checker.pending_keys = {}
1248
# Accumulate current checks.
1249
for key in current_keys:
1250
if key[0] not in kinds:
1251
checker._report_items.append('unknown key type %r' % (key,))
1252
keys[key[0]].add(key[1:])
1253
# Check the outermost kind only - inventories || chk_bytes || texts
1257
for record in getattr(self, kind).check(keys=keys[kind]):
1258
if record.storage_kind == 'absent':
1259
checker._report_items.append(
1260
'Missing %s {%s}' % (kind, record.key,))
1262
last_object = self._check_record(kind, record,
1263
checker, last_object, current_keys[(kind,) + record.key])
1267
def _check_record(self, kind, record, checker, last_object, item_data):
1268
"""Check a single text from this repository."""
1269
if kind == 'inventories':
1270
rev_id = record.key[0]
1271
inv = self._deserialise_inventory(rev_id,
1272
record.get_bytes_as('fulltext'))
1273
if last_object is not None:
1274
delta = inv._make_delta(last_object)
1275
for old_path, path, file_id, ie in delta:
1278
ie.check(checker, rev_id, inv)
1280
for path, ie in inv.iter_entries():
1281
ie.check(checker, rev_id, inv)
1282
if self._format.fast_deltas:
1284
elif kind == 'chk_bytes':
1285
# No code written to check chk_bytes for this repo format.
1286
checker._report_items.append(
1287
'unsupported key type chk_bytes for %s' % (record.key,))
1288
elif kind == 'texts':
1289
self._check_text(record, checker, item_data)
1291
checker._report_items.append(
1292
'unknown key type %s for %s' % (kind, record.key))
1294
def _check_text(self, record, checker, item_data):
1295
"""Check a single text."""
1296
# Check it is extractable.
1297
# TODO: check length.
1298
if record.storage_kind == 'chunked':
1299
chunks = record.get_bytes_as(record.storage_kind)
1300
sha1 = osutils.sha_strings(chunks)
1301
length = sum(map(len, chunks))
1303
content = record.get_bytes_as('fulltext')
1304
sha1 = osutils.sha_string(content)
1305
length = len(content)
1306
if item_data and sha1 != item_data[1]:
1307
checker._report_items.append(
1308
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1309
(record.key, sha1, item_data[1], item_data[2]))
593
for id in revision_ids:
594
if self.has_revision(id):
1312
599
def create(a_bzrdir):
1313
600
"""Construct the current default format repository in a_bzrdir."""
1314
601
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1316
def __init__(self, _format, a_bzrdir, control_files):
603
def __init__(self, _format, a_bzrdir, control_files, _revision_store, control_store, text_store):
1317
604
"""instantiate a Repository.
1319
606
:param _format: The format of the repository on disk.
1320
607
:param a_bzrdir: The BzrDir of the repository.
609
In the future we will have a single api for all stores for
610
getting file texts, inventories and revisions, then
611
this construct will accept instances of those things.
1322
# In the future we will have a single api for all stores for
1323
# getting file texts, inventories and revisions, then
1324
# this construct will accept instances of those things.
1325
613
super(Repository, self).__init__()
1326
614
self._format = _format
1327
615
# the following are part of the public API for Repository:
1328
616
self.bzrdir = a_bzrdir
1329
617
self.control_files = control_files
1330
self._transport = control_files._transport
1331
self.base = self._transport.base
618
self._revision_store = _revision_store
619
# backwards compatibility
620
self.weave_store = text_store
1333
622
self._reconcile_does_inventory_gc = True
1334
self._reconcile_fixes_text_parents = False
1335
self._reconcile_backsup_inventory = True
623
# not right yet - should be more semantically clear ?
625
self.control_store = control_store
626
self.control_weaves = control_store
627
# TODO: make sure to construct the right store classes, etc, depending
628
# on whether escaping is required.
629
self._warn_if_deprecated()
1336
630
self._write_group = None
1337
# Additional places to query for data.
1338
self._fallback_repositories = []
1339
# An InventoryEntry cache, used during deserialization
1340
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1341
# Is it safe to return inventory entries directly from the entry cache,
1342
# rather copying them?
1343
self._safe_to_return_from_cache = False
1346
def user_transport(self):
1347
return self.bzrdir.user_transport
1350
def control_transport(self):
1351
return self._transport
631
self.base = control_files._transport.base
1353
633
def __repr__(self):
1354
if self._fallback_repositories:
1355
return '%s(%r, fallback_repositories=%r)' % (
1356
self.__class__.__name__,
1358
self._fallback_repositories)
1360
return '%s(%r)' % (self.__class__.__name__,
1363
def _has_same_fallbacks(self, other_repo):
1364
"""Returns true if the repositories have the same fallbacks."""
1365
my_fb = self._fallback_repositories
1366
other_fb = other_repo._fallback_repositories
1367
if len(my_fb) != len(other_fb):
1369
for f, g in zip(my_fb, other_fb):
1370
if not f.has_same_location(g):
634
return '%s(%r)' % (self.__class__.__name__,
1374
637
def has_same_location(self, other):
1375
638
"""Returns a boolean indicating if this repository is at the same
1588
787
"""Commit the contents accrued within the current write group.
1590
789
:seealso: start_write_group.
1592
:return: it may return an opaque hint that can be passed to 'pack'.
1594
791
if self._write_group is not self.get_transaction():
1595
792
# has an unlock or relock occured ?
1596
793
raise errors.BzrError('mismatched lock context %r and '
1597
794
'write group %r.' %
1598
795
(self.get_transaction(), self._write_group))
1599
result = self._commit_write_group()
796
self._commit_write_group()
1600
797
self._write_group = None
1603
799
def _commit_write_group(self):
1604
800
"""Template method for per-repository write group cleanup.
1606
This is called before the write group is considered to be
802
This is called before the write group is considered to be
1607
803
finished and should ensure that all data handed to the repository
1608
for writing during the write group is safely committed (to the
804
for writing during the write group is safely committed (to the
1609
805
extent possible considering file system caching etc).
1612
def suspend_write_group(self):
1613
raise errors.UnsuspendableWriteGroup(self)
1615
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1616
"""Return the keys of missing inventory parents for revisions added in
1619
A revision is not complete if the inventory delta for that revision
1620
cannot be calculated. Therefore if the parent inventories of a
1621
revision are not present, the revision is incomplete, and e.g. cannot
1622
be streamed by a smart server. This method finds missing inventory
1623
parents for revisions added in this write group.
1625
if not self._format.supports_external_lookups:
1626
# This is only an issue for stacked repositories
1628
if not self.is_in_write_group():
1629
raise AssertionError('not in a write group')
1631
# XXX: We assume that every added revision already has its
1632
# corresponding inventory, so we only check for parent inventories that
1633
# might be missing, rather than all inventories.
1634
parents = set(self.revisions._index.get_missing_parents())
1635
parents.discard(_mod_revision.NULL_REVISION)
1636
unstacked_inventories = self.inventories._index
1637
present_inventories = unstacked_inventories.get_parent_map(
1638
key[-1:] for key in parents)
1639
parents.difference_update(present_inventories)
1640
if len(parents) == 0:
1641
# No missing parent inventories.
1643
if not check_for_missing_texts:
1644
return set(('inventories', rev_id) for (rev_id,) in parents)
1645
# Ok, now we have a list of missing inventories. But these only matter
1646
# if the inventories that reference them are missing some texts they
1647
# appear to introduce.
1648
# XXX: Texts referenced by all added inventories need to be present,
1649
# but at the moment we're only checking for texts referenced by
1650
# inventories at the graph's edge.
1651
key_deps = self.revisions._index._key_dependencies
1652
key_deps.satisfy_refs_for_keys(present_inventories)
1653
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1654
file_ids = self.fileids_altered_by_revision_ids(referrers)
1655
missing_texts = set()
1656
for file_id, version_ids in file_ids.iteritems():
1657
missing_texts.update(
1658
(file_id, version_id) for version_id in version_ids)
1659
present_texts = self.texts.get_parent_map(missing_texts)
1660
missing_texts.difference_update(present_texts)
1661
if not missing_texts:
1662
# No texts are missing, so all revisions and their deltas are
1665
# Alternatively the text versions could be returned as the missing
1666
# keys, but this is likely to be less data.
1667
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1670
def refresh_data(self):
1671
"""Re-read any data needed to synchronise with disk.
1673
This method is intended to be called after another repository instance
1674
(such as one used by a smart server) has inserted data into the
1675
repository. On all repositories this will work outside of write groups.
1676
Some repository formats (pack and newer for bzrlib native formats)
1677
support refresh_data inside write groups. If called inside a write
1678
group on a repository that does not support refreshing in a write group
1679
IsInWriteGroupError will be raised.
1681
self._refresh_data()
1683
def resume_write_group(self, tokens):
1684
if not self.is_write_locked():
1685
raise errors.NotWriteLocked(self)
1686
if self._write_group:
1687
raise errors.BzrError('already in a write group')
1688
self._resume_write_group(tokens)
1689
# so we can detect unlock/relock - the write group is now entered.
1690
self._write_group = self.get_transaction()
1692
def _resume_write_group(self, tokens):
1693
raise errors.UnsuspendableWriteGroup(self)
1695
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
808
def fetch(self, source, revision_id=None, pb=None):
1697
809
"""Fetch the content required to construct revision_id from source.
1699
If revision_id is None and fetch_spec is None, then all content is
1702
fetch() may not be used when the repository is in a write group -
1703
either finish the current write group before using fetch, or use
1704
fetch before starting the write group.
1706
:param find_ghosts: Find and copy revisions in the source that are
1707
ghosts in the target (and not reachable directly by walking out to
1708
the first-present revision in target from revision_id).
1709
:param revision_id: If specified, all the content needed for this
1710
revision ID will be copied to the target. Fetch will determine for
1711
itself which content needs to be copied.
1712
:param fetch_spec: If specified, a SearchResult or
1713
PendingAncestryResult that describes which revisions to copy. This
1714
allows copying multiple heads at once. Mutually exclusive with
811
If revision_id is None all content is copied.
1717
if fetch_spec is not None and revision_id is not None:
1718
raise AssertionError(
1719
"fetch_spec and revision_id are mutually exclusive.")
1720
if self.is_in_write_group():
1721
raise errors.InternalBzrError(
1722
"May not fetch while in a write group.")
1723
813
# fast path same-url fetch operations
1724
# TODO: lift out to somewhere common with RemoteRepository
1725
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1726
if (self.has_same_location(source)
1727
and fetch_spec is None
1728
and self._has_same_fallbacks(source)):
814
if self.has_same_location(source):
1729
815
# check that last_revision is in 'from' and then return a
1731
817
if (revision_id is not None and
1732
818
not _mod_revision.is_null(revision_id)):
1733
819
self.get_revision(revision_id)
1735
# if there is no specific appropriate InterRepository, this will get
1736
# the InterRepository base class, which raises an
1737
# IncompatibleRepositories when asked to fetch.
1738
821
inter = InterRepository.get(source, self)
1739
return inter.fetch(revision_id=revision_id, pb=pb,
1740
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
823
return inter.fetch(revision_id=revision_id, pb=pb)
824
except NotImplementedError:
825
raise errors.IncompatibleRepositories(source, self)
1742
827
def create_bundle(self, target, base, fileobj, format=None):
1743
828
return serializer.write_bundle(self, target, base, fileobj, format)
1894
948
@needs_read_lock
1895
949
def get_revisions(self, revision_ids):
1896
"""Get many revisions at once.
1898
Repositories that need to check data on every revision read should
1899
subclass this method.
950
"""Get many revisions at once."""
1901
951
return self._get_revisions(revision_ids)
1903
953
@needs_read_lock
1904
954
def _get_revisions(self, revision_ids):
1905
955
"""Core work logic to get many revisions without sanity checks."""
1907
for revid, rev in self._iter_revisions(revision_ids):
1909
raise errors.NoSuchRevision(self, revid)
1911
return [revs[revid] for revid in revision_ids]
1913
def _iter_revisions(self, revision_ids):
1914
"""Iterate over revision objects.
1916
:param revision_ids: An iterable of revisions to examine. None may be
1917
passed to request all revisions known to the repository. Note that
1918
not all repositories can find unreferenced revisions; for those
1919
repositories only referenced ones will be returned.
1920
:return: An iterator of (revid, revision) tuples. Absent revisions (
1921
those asked for but not available) are returned as (revid, None).
1923
if revision_ids is None:
1924
revision_ids = self.all_revision_ids()
1926
for rev_id in revision_ids:
1927
if not rev_id or not isinstance(rev_id, basestring):
1928
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1929
keys = [(key,) for key in revision_ids]
1930
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1931
for record in stream:
1932
revid = record.key[0]
1933
if record.storage_kind == 'absent':
1936
text = record.get_bytes_as('fulltext')
1937
rev = self._serializer.read_revision_from_string(text)
1940
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
956
for rev_id in revision_ids:
957
if not rev_id or not isinstance(rev_id, basestring):
958
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
959
revs = self._revision_store.get_revisions(revision_ids,
960
self.get_transaction())
962
assert not isinstance(rev.revision_id, unicode)
963
for parent_id in rev.parent_ids:
964
assert not isinstance(parent_id, unicode)
968
def get_revision_xml(self, revision_id):
969
# TODO: jam 20070210 This shouldn't be necessary since get_revision
970
# would have already do it.
971
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
972
rev = self.get_revision(revision_id)
974
# the current serializer..
975
self._revision_store._serializer.write_revision(rev, rev_tmp)
977
return rev_tmp.getvalue()
980
def get_deltas_for_revisions(self, revisions):
1941
981
"""Produce a generator of revision deltas.
1943
983
Note that the input is a sequence of REVISIONS, not revision_ids.
1944
984
Trees will be held in memory until the generator exits.
1945
985
Each delta is relative to the revision's lefthand predecessor.
1947
:param specific_fileids: if not None, the result is filtered
1948
so that only those file-ids, their parents and their
1949
children are included.
1951
# Get the revision-ids of interest
1952
987
required_trees = set()
1953
988
for revision in revisions:
1954
989
required_trees.add(revision.revision_id)
1955
990
required_trees.update(revision.parent_ids[:1])
1957
# Get the matching filtered trees. Note that it's more
1958
# efficient to pass filtered trees to changes_from() rather
1959
# than doing the filtering afterwards. changes_from() could
1960
# arguably do the filtering itself but it's path-based, not
1961
# file-id based, so filtering before or afterwards is
1963
if specific_fileids is None:
1964
trees = dict((t.get_revision_id(), t) for
1965
t in self.revision_trees(required_trees))
1967
trees = dict((t.get_revision_id(), t) for
1968
t in self._filtered_revision_trees(required_trees,
1971
# Calculate the deltas
991
trees = dict((t.get_revision_id(), t) for
992
t in self.revision_trees(required_trees))
1972
993
for revision in revisions:
1973
994
if not revision.parent_ids:
1974
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
995
old_tree = self.revision_tree(None)
1976
997
old_tree = trees[revision.parent_ids[0]]
1977
998
yield trees[revision.revision_id].changes_from(old_tree)
1979
1000
@needs_read_lock
1980
def get_revision_delta(self, revision_id, specific_fileids=None):
1001
def get_revision_delta(self, revision_id):
1981
1002
"""Return the delta for one revision.
1983
1004
The delta is relative to the left-hand predecessor of the
1986
:param specific_fileids: if not None, the result is filtered
1987
so that only those file-ids, their parents and their
1988
children are included.
1990
1007
r = self.get_revision(revision_id)
1991
return list(self.get_deltas_for_revisions([r],
1992
specific_fileids=specific_fileids))[0]
1008
return list(self.get_deltas_for_revisions([r]))[0]
1994
1010
@needs_write_lock
1995
1011
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1996
1012
signature = gpg_strategy.sign(plaintext)
1997
self.add_signature_text(revision_id, signature)
2000
def add_signature_text(self, revision_id, signature):
2001
self.signatures.add_lines((revision_id,), (),
2002
osutils.split_lines(signature))
2004
def find_text_key_references(self):
2005
"""Find the text key references within the repository.
2007
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2008
to whether they were referred to by the inventory of the
2009
revision_id that they contain. The inventory texts from all present
2010
revision ids are assessed to generate this report.
2012
revision_keys = self.revisions.keys()
2013
w = self.inventories
2014
pb = ui.ui_factory.nested_progress_bar()
2016
return self._find_text_key_references_from_xml_inventory_lines(
2017
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2021
def _find_text_key_references_from_xml_inventory_lines(self,
2023
"""Core routine for extracting references to texts from inventories.
2025
This performs the translation of xml lines to revision ids.
2027
:param line_iterator: An iterator of lines, origin_version_id
2028
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2029
to whether they were referred to by the inventory of the
2030
revision_id that they contain. Note that if that revision_id was
2031
not part of the line_iterator's output then False will be given -
2032
even though it may actually refer to that key.
2034
if not self._serializer.support_altered_by_hack:
2035
raise AssertionError(
2036
"_find_text_key_references_from_xml_inventory_lines only "
2037
"supported for branches which store inventory as unnested xml"
2038
", not on %r" % self)
1013
self._revision_store.add_revision_signature_text(revision_id,
1015
self.get_transaction())
1017
def fileids_altered_by_revision_ids(self, revision_ids):
1018
"""Find the file ids and versions affected by revisions.
1020
:param revisions: an iterable containing revision ids.
1021
:return: a dictionary mapping altered file-ids to an iterable of
1022
revision_ids. Each altered file-ids has the exact revision_ids that
1023
altered it listed explicitly.
1025
assert self._serializer.support_altered_by_hack, \
1026
("fileids_altered_by_revision_ids only supported for branches "
1027
"which store inventory as unnested xml, not on %r" % self)
1028
selected_revision_ids = set(revision_ids)
1029
w = self.get_inventory_weave()
2041
1032
# this code needs to read every new line in every inventory for the
2042
1033
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2043
# not present in one of those inventories is unnecessary but not
1034
# not present in one of those inventories is unnecessary but not
2044
1035
# harmful because we are filtering by the revision id marker in the
2045
# inventory lines : we only select file ids altered in one of those
1036
# inventory lines : we only select file ids altered in one of those
2046
1037
# revisions. We don't need to see all lines in the inventory because
2047
1038
# only those added in an inventory in rev X can contain a revision=X
2057
1048
search = self._file_ids_altered_regex.search
2058
1049
unescape = _unescape_xml
2059
1050
setdefault = result.setdefault
2060
for line, line_key in line_iterator:
2061
match = search(line)
2064
# One call to match.group() returning multiple items is quite a
2065
# bit faster than 2 calls to match.group() each returning 1
2066
file_id, revision_id = match.group('file_id', 'revision_id')
2068
# Inlining the cache lookups helps a lot when you make 170,000
2069
# lines and 350k ids, versus 8.4 unique ids.
2070
# Using a cache helps in 2 ways:
2071
# 1) Avoids unnecessary decoding calls
2072
# 2) Re-uses cached strings, which helps in future set and
2074
# (2) is enough that removing encoding entirely along with
2075
# the cache (so we are using plain strings) results in no
2076
# performance improvement.
2078
revision_id = unescape_revid_cache[revision_id]
2080
unescaped = unescape(revision_id)
2081
unescape_revid_cache[revision_id] = unescaped
2082
revision_id = unescaped
2084
# Note that unconditionally unescaping means that we deserialise
2085
# every fileid, which for general 'pull' is not great, but we don't
2086
# really want to have some many fulltexts that this matters anyway.
2089
file_id = unescape_fileid_cache[file_id]
2091
unescaped = unescape(file_id)
2092
unescape_fileid_cache[file_id] = unescaped
2095
key = (file_id, revision_id)
2096
setdefault(key, False)
2097
if revision_id == line_key[-1]:
2101
def _inventory_xml_lines_for_keys(self, keys):
2102
"""Get a line iterator of the sort needed for findind references.
2104
Not relevant for non-xml inventory repositories.
2106
Ghosts in revision_keys are ignored.
2108
:param revision_keys: The revision keys for the inventories to inspect.
2109
:return: An iterator over (inventory line, revid) for the fulltexts of
2110
all of the xml inventories specified by revision_keys.
2112
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2113
for record in stream:
2114
if record.storage_kind != 'absent':
2115
chunks = record.get_bytes_as('chunked')
2116
revid = record.key[-1]
2117
lines = osutils.chunks_to_lines(chunks)
2121
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2123
"""Helper routine for fileids_altered_by_revision_ids.
2125
This performs the translation of xml lines to revision ids.
2127
:param line_iterator: An iterator of lines, origin_version_id
2128
:param revision_keys: The revision ids to filter for. This should be a
2129
set or other type which supports efficient __contains__ lookups, as
2130
the revision key from each parsed line will be looked up in the
2131
revision_keys filter.
2132
:return: a dictionary mapping altered file-ids to an iterable of
2133
revision_ids. Each altered file-ids has the exact revision_ids that
2134
altered it listed explicitly.
2136
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2137
line_iterator).iterkeys())
2138
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2139
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2140
self._inventory_xml_lines_for_keys(parent_keys)))
2141
new_keys = seen - parent_seen
2143
setdefault = result.setdefault
2144
for key in new_keys:
2145
setdefault(key[0], set()).add(key[-1])
2148
def _find_parent_ids_of_revisions(self, revision_ids):
2149
"""Find all parent ids that are mentioned in the revision graph.
2151
:return: set of revisions that are parents of revision_ids which are
2152
not part of revision_ids themselves
2154
parent_map = self.get_parent_map(revision_ids)
2156
map(parent_ids.update, parent_map.itervalues())
2157
parent_ids.difference_update(revision_ids)
2158
parent_ids.discard(_mod_revision.NULL_REVISION)
2161
def _find_parent_keys_of_revisions(self, revision_keys):
2162
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2164
:param revision_keys: An iterable of revision_keys.
2165
:return: The parents of all revision_keys that are not already in
2168
parent_map = self.revisions.get_parent_map(revision_keys)
2170
map(parent_keys.update, parent_map.itervalues())
2171
parent_keys.difference_update(revision_keys)
2172
parent_keys.discard(_mod_revision.NULL_REVISION)
2175
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2176
"""Find the file ids and versions affected by revisions.
2178
:param revisions: an iterable containing revision ids.
2179
:param _inv_weave: The inventory weave from this repository or None.
2180
If None, the inventory weave will be opened automatically.
2181
:return: a dictionary mapping altered file-ids to an iterable of
2182
revision_ids. Each altered file-ids has the exact revision_ids that
2183
altered it listed explicitly.
2185
selected_keys = set((revid,) for revid in revision_ids)
2186
w = _inv_weave or self.inventories
2187
return self._find_file_ids_from_xml_inventory_lines(
2188
w.iter_lines_added_or_present_in_keys(
2189
selected_keys, pb=None),
1051
pb = ui.ui_factory.nested_progress_bar()
1053
for line in w.iter_lines_added_or_present_in_versions(
1054
selected_revision_ids, pb=pb):
1055
match = search(line)
1058
# One call to match.group() returning multiple items is quite a
1059
# bit faster than 2 calls to match.group() each returning 1
1060
file_id, revision_id = match.group('file_id', 'revision_id')
1062
# Inlining the cache lookups helps a lot when you make 170,000
1063
# lines and 350k ids, versus 8.4 unique ids.
1064
# Using a cache helps in 2 ways:
1065
# 1) Avoids unnecessary decoding calls
1066
# 2) Re-uses cached strings, which helps in future set and
1068
# (2) is enough that removing encoding entirely along with
1069
# the cache (so we are using plain strings) results in no
1070
# performance improvement.
1072
revision_id = unescape_revid_cache[revision_id]
1074
unescaped = unescape(revision_id)
1075
unescape_revid_cache[revision_id] = unescaped
1076
revision_id = unescaped
1078
if revision_id in selected_revision_ids:
1080
file_id = unescape_fileid_cache[file_id]
1082
unescaped = unescape(file_id)
1083
unescape_fileid_cache[file_id] = unescaped
1085
setdefault(file_id, set()).add(revision_id)
2192
1090
def iter_files_bytes(self, desired_files):
2193
1091
"""Iterate through file versions.
2203
1101
bytes_iterator is an iterable of bytestrings for the file. The
2204
1102
kind of iterable and length of the bytestrings are unspecified, but for
2205
this implementation, it is a list of bytes produced by
2206
VersionedFile.get_record_stream().
1103
this implementation, it is a list of lines produced by
1104
VersionedFile.get_lines().
2208
1106
:param desired_files: a list of (file_id, revision_id, identifier)
1109
transaction = self.get_transaction()
2212
1110
for file_id, revision_id, callable_data in desired_files:
2213
text_keys[(file_id, revision_id)] = callable_data
2214
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2215
if record.storage_kind == 'absent':
2216
raise errors.RevisionNotPresent(record.key, self)
2217
yield text_keys[record.key], record.get_bytes_as('chunked')
2219
def _generate_text_key_index(self, text_key_references=None,
2221
"""Generate a new text key index for the repository.
2223
This is an expensive function that will take considerable time to run.
2225
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2226
list of parents, also text keys. When a given key has no parents,
2227
the parents list will be [NULL_REVISION].
2229
# All revisions, to find inventory parents.
2230
if ancestors is None:
2231
graph = self.get_graph()
2232
ancestors = graph.get_parent_map(self.all_revision_ids())
2233
if text_key_references is None:
2234
text_key_references = self.find_text_key_references()
2235
pb = ui.ui_factory.nested_progress_bar()
2237
return self._do_generate_text_key_index(ancestors,
2238
text_key_references, pb)
2242
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2243
"""Helper for _generate_text_key_index to avoid deep nesting."""
2244
revision_order = tsort.topo_sort(ancestors)
2245
invalid_keys = set()
2247
for revision_id in revision_order:
2248
revision_keys[revision_id] = set()
2249
text_count = len(text_key_references)
2250
# a cache of the text keys to allow reuse; costs a dict of all the
2251
# keys, but saves a 2-tuple for every child of a given key.
2253
for text_key, valid in text_key_references.iteritems():
2255
invalid_keys.add(text_key)
2257
revision_keys[text_key[1]].add(text_key)
2258
text_key_cache[text_key] = text_key
2259
del text_key_references
2261
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2262
NULL_REVISION = _mod_revision.NULL_REVISION
2263
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2264
# too small for large or very branchy trees. However, for 55K path
2265
# trees, it would be easy to use too much memory trivially. Ideally we
2266
# could gauge this by looking at available real memory etc, but this is
2267
# always a tricky proposition.
2268
inventory_cache = lru_cache.LRUCache(10)
2269
batch_size = 10 # should be ~150MB on a 55K path tree
2270
batch_count = len(revision_order) / batch_size + 1
2272
pb.update("Calculating text parents", processed_texts, text_count)
2273
for offset in xrange(batch_count):
2274
to_query = revision_order[offset * batch_size:(offset + 1) *
2278
for revision_id in to_query:
2279
parent_ids = ancestors[revision_id]
2280
for text_key in revision_keys[revision_id]:
2281
pb.update("Calculating text parents", processed_texts)
2282
processed_texts += 1
2283
candidate_parents = []
2284
for parent_id in parent_ids:
2285
parent_text_key = (text_key[0], parent_id)
2287
check_parent = parent_text_key not in \
2288
revision_keys[parent_id]
2290
# the parent parent_id is a ghost:
2291
check_parent = False
2292
# truncate the derived graph against this ghost.
2293
parent_text_key = None
2295
# look at the parent commit details inventories to
2296
# determine possible candidates in the per file graph.
2299
inv = inventory_cache[parent_id]
2301
inv = self.revision_tree(parent_id).inventory
2302
inventory_cache[parent_id] = inv
2304
parent_entry = inv[text_key[0]]
2305
except (KeyError, errors.NoSuchId):
2307
if parent_entry is not None:
2309
text_key[0], parent_entry.revision)
2311
parent_text_key = None
2312
if parent_text_key is not None:
2313
candidate_parents.append(
2314
text_key_cache[parent_text_key])
2315
parent_heads = text_graph.heads(candidate_parents)
2316
new_parents = list(parent_heads)
2317
new_parents.sort(key=lambda x:candidate_parents.index(x))
2318
if new_parents == []:
2319
new_parents = [NULL_REVISION]
2320
text_index[text_key] = new_parents
2322
for text_key in invalid_keys:
2323
text_index[text_key] = [NULL_REVISION]
1112
weave = self.weave_store.get_weave(file_id, transaction)
1113
except errors.NoSuchFile:
1114
raise errors.NoSuchIdInRepository(self, file_id)
1115
yield callable_data, weave.get_lines(revision_id)
2326
1117
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2327
1118
"""Get an iterable listing the keys of all the data introduced by a set
2334
1125
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2335
1126
'revisions'. file-id is None unless knit-kind is 'file'.
2337
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2340
for result in self._find_non_file_keys_to_fetch(revision_ids):
2343
def _find_file_keys_to_fetch(self, revision_ids, pb):
2344
1128
# XXX: it's a bit weird to control the inventory weave caching in this
2345
1129
# generator. Ideally the caching would be done in fetch.py I think. Or
2346
1130
# maybe this generator should explicitly have the contract that it
2347
1131
# should not be iterated until the previously yielded item has been
2349
inv_w = self.inventories
1133
inv_w = self.get_inventory_weave()
1134
inv_w.enable_cache()
2351
1136
# file ids that changed
2352
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
1137
file_ids = self.fileids_altered_by_revision_ids(revision_ids)
2354
1139
num_file_ids = len(file_ids)
2355
1140
for file_id, altered_versions in file_ids.iteritems():
2357
pb.update("Fetch texts", count, num_file_ids)
1141
if _files_pb is not None:
1142
_files_pb.update("fetch texts", count, num_file_ids)
2359
1144
yield ("file", file_id, altered_versions)
1145
# We're done with the files_pb. Note that it finished by the caller,
1146
# just as it was created by the caller.
2361
def _find_non_file_keys_to_fetch(self, revision_ids):
2363
1150
yield ("inventory", None, revision_ids)
2366
# XXX: Note ATM no callers actually pay attention to this return
2367
# instead they just use the list of revision ids and ignore
2368
# missing sigs. Consider removing this work entirely
2369
revisions_with_signatures = set(self.signatures.get_parent_map(
2370
[(r,) for r in revision_ids]))
2371
revisions_with_signatures = set(
2372
[r for (r,) in revisions_with_signatures])
2373
revisions_with_signatures.intersection_update(revision_ids)
1154
revisions_with_signatures = set()
1155
for rev_id in revision_ids:
1157
self.get_signature_text(rev_id)
1158
except errors.NoSuchRevision:
1162
revisions_with_signatures.add(rev_id)
2374
1163
yield ("signatures", None, revisions_with_signatures)
2377
1166
yield ("revisions", None, revision_ids)
2379
1168
@needs_read_lock
1169
def get_inventory_weave(self):
1170
return self.control_weaves.get_weave('inventory',
1171
self.get_transaction())
2380
1174
def get_inventory(self, revision_id):
2381
"""Get Inventory object by revision id."""
2382
return self.iter_inventories([revision_id]).next()
2384
def iter_inventories(self, revision_ids, ordering=None):
2385
"""Get many inventories by revision_ids.
2387
This will buffer some or all of the texts used in constructing the
2388
inventories in memory, but will only parse a single inventory at a
2391
:param revision_ids: The expected revision ids of the inventories.
2392
:param ordering: optional ordering, e.g. 'topological'. If not
2393
specified, the order of revision_ids will be preserved (by
2394
buffering if necessary).
2395
:return: An iterator of inventories.
2397
if ((None in revision_ids)
2398
or (_mod_revision.NULL_REVISION in revision_ids)):
2399
raise ValueError('cannot get null revision inventory')
2400
return self._iter_inventories(revision_ids, ordering)
2402
def _iter_inventories(self, revision_ids, ordering):
2403
"""single-document based inventory iteration."""
2404
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2405
for text, revision_id in inv_xmls:
2406
yield self._deserialise_inventory(revision_id, text)
2408
def _iter_inventory_xmls(self, revision_ids, ordering):
2409
if ordering is None:
2410
order_as_requested = True
2411
ordering = 'unordered'
2413
order_as_requested = False
2414
keys = [(revision_id,) for revision_id in revision_ids]
2417
if order_as_requested:
2418
key_iter = iter(keys)
2419
next_key = key_iter.next()
2420
stream = self.inventories.get_record_stream(keys, ordering, True)
2422
for record in stream:
2423
if record.storage_kind != 'absent':
2424
chunks = record.get_bytes_as('chunked')
2425
if order_as_requested:
2426
text_chunks[record.key] = chunks
2428
yield ''.join(chunks), record.key[-1]
2430
raise errors.NoSuchRevision(self, record.key)
2431
if order_as_requested:
2432
# Yield as many results as we can while preserving order.
2433
while next_key in text_chunks:
2434
chunks = text_chunks.pop(next_key)
2435
yield ''.join(chunks), next_key[-1]
2437
next_key = key_iter.next()
2438
except StopIteration:
2439
# We still want to fully consume the get_record_stream,
2440
# just in case it is not actually finished at this point
2444
def _deserialise_inventory(self, revision_id, xml):
2445
"""Transform the xml into an inventory object.
1175
"""Get Inventory object by hash."""
1176
return self.deserialise_inventory(
1177
revision_id, self.get_inventory_xml(revision_id))
1179
def deserialise_inventory(self, revision_id, xml):
1180
"""Transform the xml into an inventory object.
2447
1182
:param revision_id: The expected revision id of the inventory.
2448
1183
:param xml: A serialised inventory.
2450
result = self._serializer.read_inventory_from_string(xml, revision_id,
2451
entry_cache=self._inventory_entry_cache,
2452
return_from_cache=self._safe_to_return_from_cache)
2453
if result.revision_id != revision_id:
2454
raise AssertionError('revision id mismatch %s != %s' % (
2455
result.revision_id, revision_id))
1185
return self._serializer.read_inventory_from_string(xml, revision_id)
1187
def serialise_inventory(self, inv):
1188
return self._serializer.write_inventory_to_string(inv)
1190
def _serialise_inventory_to_lines(self, inv):
1191
return self._serializer.write_inventory_to_lines(inv)
2458
1193
def get_serializer_format(self):
2459
1194
return self._serializer.format_num
2461
1196
@needs_read_lock
2462
def _get_inventory_xml(self, revision_id):
2463
"""Get serialized inventory as a string."""
2464
texts = self._iter_inventory_xmls([revision_id], 'unordered')
1197
def get_inventory_xml(self, revision_id):
1198
"""Get inventory XML as a file object."""
2466
text, revision_id = texts.next()
2467
except StopIteration:
1200
assert isinstance(revision_id, str), type(revision_id)
1201
iw = self.get_inventory_weave()
1202
return iw.get_text(revision_id)
2468
1204
raise errors.HistoryMissing(self, 'inventory', revision_id)
2471
def get_rev_id_for_revno(self, revno, known_pair):
2472
"""Return the revision id of a revno, given a later (revno, revid)
2473
pair in the same history.
2475
:return: if found (True, revid). If the available history ran out
2476
before reaching the revno, then this returns
2477
(False, (closest_revno, closest_revid)).
2479
known_revno, known_revid = known_pair
2480
partial_history = [known_revid]
2481
distance_from_known = known_revno - revno
2482
if distance_from_known < 0:
2484
'requested revno (%d) is later than given known revno (%d)'
2485
% (revno, known_revno))
2488
self, partial_history, stop_index=distance_from_known)
2489
except errors.RevisionNotPresent, err:
2490
if err.revision_id == known_revid:
2491
# The start revision (known_revid) wasn't found.
2493
# This is a stacked repository with no fallbacks, or a there's a
2494
# left-hand ghost. Either way, even though the revision named in
2495
# the error isn't in this repo, we know it's the next step in this
2496
# left-hand history.
2497
partial_history.append(err.revision_id)
2498
if len(partial_history) <= distance_from_known:
2499
# Didn't find enough history to get a revid for the revno.
2500
earliest_revno = known_revno - len(partial_history) + 1
2501
return (False, (earliest_revno, partial_history[-1]))
2502
if len(partial_history) - 1 > distance_from_known:
2503
raise AssertionError('_iter_for_revno returned too much history')
2504
return (True, partial_history[-1])
1207
def get_inventory_sha1(self, revision_id):
1208
"""Return the sha1 hash of the inventory entry
1210
return self.get_revision(revision_id).inventory_sha1
1213
def get_revision_graph(self, revision_id=None):
1214
"""Return a dictionary containing the revision graph.
1216
NB: This method should not be used as it accesses the entire graph all
1217
at once, which is much more data than most operations should require.
1219
:param revision_id: The revision_id to get a graph from. If None, then
1220
the entire revision graph is returned. This is a deprecated mode of
1221
operation and will be removed in the future.
1222
:return: a dictionary of revision_id->revision_parents_list.
1224
raise NotImplementedError(self.get_revision_graph)
1227
def get_revision_graph_with_ghosts(self, revision_ids=None):
1228
"""Return a graph of the revisions with ghosts marked as applicable.
1230
:param revision_ids: an iterable of revisions to graph or None for all.
1231
:return: a Graph object with the graph reachable from revision_ids.
1233
if 'evil' in debug.debug_flags:
1235
"get_revision_graph_with_ghosts scales with size of history.")
1236
result = deprecated_graph.Graph()
1237
if not revision_ids:
1238
pending = set(self.all_revision_ids())
1241
pending = set(revision_ids)
1242
# special case NULL_REVISION
1243
if _mod_revision.NULL_REVISION in pending:
1244
pending.remove(_mod_revision.NULL_REVISION)
1245
required = set(pending)
1248
revision_id = pending.pop()
1250
rev = self.get_revision(revision_id)
1251
except errors.NoSuchRevision:
1252
if revision_id in required:
1255
result.add_ghost(revision_id)
1257
for parent_id in rev.parent_ids:
1258
# is this queued or done ?
1259
if (parent_id not in pending and
1260
parent_id not in done):
1262
pending.add(parent_id)
1263
result.add_node(revision_id, rev.parent_ids)
1264
done.add(revision_id)
1267
def _get_history_vf(self):
1268
"""Get a versionedfile whose history graph reflects all revisions.
1270
For weave repositories, this is the inventory weave.
1272
return self.get_inventory_weave()
2506
1274
def iter_reverse_revision_history(self, revision_id):
2507
1275
"""Iterate backwards through revision ids in the lefthand history
2543
1325
for repositories to maintain loaded indices across multiple locks
2544
1326
by checking inside their implementation of this method to see
2545
1327
whether their indices are still valid. This depends of course on
2546
the disk format being validatable in this manner. This method is
2547
also called by the refresh_data() public interface to cause a refresh
2548
to occur while in a write lock so that data inserted by a smart server
2549
push operation is visible on the client's instance of the physical
1328
the disk format being validatable in this manner.
2553
1331
@needs_read_lock
2554
1332
def revision_tree(self, revision_id):
2555
1333
"""Return Tree for a revision on this branch.
2557
`revision_id` may be NULL_REVISION for the empty tree revision.
1335
`revision_id` may be None for the empty tree revision.
2559
revision_id = _mod_revision.ensure_null(revision_id)
2560
1337
# TODO: refactor this to use an existing revision object
2561
1338
# so we don't need to read it in twice.
2562
if revision_id == _mod_revision.NULL_REVISION:
2563
return RevisionTree(self, Inventory(root_id=None),
1339
if revision_id is None or revision_id == _mod_revision.NULL_REVISION:
1340
return RevisionTree(self, Inventory(root_id=None),
2564
1341
_mod_revision.NULL_REVISION)
2566
inv = self.get_inventory(revision_id)
1343
inv = self.get_revision_inventory(revision_id)
2567
1344
return RevisionTree(self, inv, revision_id)
2569
1347
def revision_trees(self, revision_ids):
2570
"""Return Trees for revisions in this repository.
2572
:param revision_ids: a sequence of revision-ids;
2573
a revision-id may not be None or 'null:'
2575
inventories = self.iter_inventories(revision_ids)
2576
for inv in inventories:
2577
yield RevisionTree(self, inv, inv.revision_id)
2579
def _filtered_revision_trees(self, revision_ids, file_ids):
2580
"""Return Tree for a revision on this branch with only some files.
2582
:param revision_ids: a sequence of revision-ids;
2583
a revision-id may not be None or 'null:'
2584
:param file_ids: if not None, the result is filtered
2585
so that only those file-ids, their parents and their
2586
children are included.
2588
inventories = self.iter_inventories(revision_ids)
2589
for inv in inventories:
2590
# Should we introduce a FilteredRevisionTree class rather
2591
# than pre-filter the inventory here?
2592
filtered_inv = inv.filter(file_ids)
2593
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
1348
"""Return Tree for a revision on this branch.
1350
`revision_id` may not be None or 'null:'"""
1351
assert None not in revision_ids
1352
assert _mod_revision.NULL_REVISION not in revision_ids
1353
texts = self.get_inventory_weave().get_texts(revision_ids)
1354
for text, revision_id in zip(texts, revision_ids):
1355
inv = self.deserialise_inventory(revision_id, text)
1356
yield RevisionTree(self, inv, revision_id)
2595
1358
@needs_read_lock
2596
1359
def get_ancestry(self, revision_id, topo_sorted=True):
2597
1360
"""Return a list of revision-ids integrated by a revision.
2599
The first element of the list is always None, indicating the origin
2600
revision. This might change when we have history horizons, or
1362
The first element of the list is always None, indicating the origin
1363
revision. This might change when we have history horizons, or
2601
1364
perhaps we should have a new API.
2603
1366
This is topologically sorted.
2605
1368
if _mod_revision.is_null(revision_id):
2607
1370
if not self.has_revision(revision_id):
2608
1371
raise errors.NoSuchRevision(self, revision_id)
2609
graph = self.get_graph()
2611
search = graph._make_breadth_first_searcher([revision_id])
2614
found, ghosts = search.next_with_ghosts()
2615
except StopIteration:
2618
if _mod_revision.NULL_REVISION in keys:
2619
keys.remove(_mod_revision.NULL_REVISION)
2621
parent_map = graph.get_parent_map(keys)
2622
keys = tsort.topo_sort(parent_map)
2623
return [None] + list(keys)
1372
w = self.get_inventory_weave()
1373
candidates = w.get_ancestry(revision_id, topo_sorted)
1374
return [None] + candidates # self._eliminate_revisions_not_present(candidates)
2625
def pack(self, hint=None, clean_obsolete_packs=False):
2626
1377
"""Compress the data within the repository.
2628
1379
This operation only makes sense for some repository types. For other
2629
1380
types it should be a no-op that just returns.
2631
1382
This stub method does not require a lock, but subclasses should use
2632
@needs_write_lock as this is a long running call its reasonable to
1383
@needs_write_lock as this is a long running call its reasonable to
2633
1384
implicitly lock for the user.
2635
:param hint: If not supplied, the whole repository is packed.
2636
If supplied, the repository may use the hint parameter as a
2637
hint for the parts of the repository to pack. A hint can be
2638
obtained from the result of commit_write_group(). Out of
2639
date hints are simply ignored, because concurrent operations
2640
can obsolete them rapidly.
2642
:param clean_obsolete_packs: Clean obsolete packs immediately after
1388
def print_file(self, file, revision_id):
1389
"""Print `file` to stdout.
1391
FIXME RBC 20060125 as John Meinel points out this is a bad api
1392
- it writes to stdout, it assumes that that is valid etc. Fix
1393
by creating a new more flexible convenience function.
1395
tree = self.revision_tree(revision_id)
1396
# use inventory as it was in that revision
1397
file_id = tree.inventory.path2id(file)
1399
# TODO: jam 20060427 Write a test for this code path
1400
# it had a bug in it, and was raising the wrong
1402
raise errors.BzrError("%r is not present in revision %s" % (file, revision_id))
1403
tree.print_file(file_id)
2646
1405
def get_transaction(self):
2647
1406
return self.control_files.get_transaction()
2649
def get_parent_map(self, revision_ids):
2650
"""See graph.StackedParentsProvider.get_parent_map"""
2651
# revisions index works in keys; this just works in revisions
2652
# therefore wrap and unwrap
1408
def revision_parents(self, revision_id):
1409
return self.get_inventory_weave().parent_names(revision_id)
1411
def get_parents(self, revision_ids):
1412
"""See StackedParentsProvider.get_parents"""
2655
1414
for revision_id in revision_ids:
2656
1415
if revision_id == _mod_revision.NULL_REVISION:
2657
result[revision_id] = ()
2658
elif revision_id is None:
2659
raise ValueError('get_parent_map(None) is not valid')
2661
query_keys.append((revision_id ,))
2662
for ((revision_id,), parent_keys) in \
2663
self.revisions.get_parent_map(query_keys).iteritems():
2665
result[revision_id] = tuple([parent_revid
2666
for (parent_revid,) in parent_keys])
2668
result[revision_id] = (_mod_revision.NULL_REVISION,)
1419
parents = self.get_revision(revision_id).parent_ids
1420
except errors.NoSuchRevision:
1423
if len(parents) == 0:
1424
parents = [_mod_revision.NULL_REVISION]
1425
parents_list.append(parents)
2671
1428
def _make_parents_provider(self):
2675
def get_known_graph_ancestry(self, revision_ids):
2676
"""Return the known graph for a set of revision ids and their ancestors.
2678
st = static_tuple.StaticTuple
2679
revision_keys = [st(r_id).intern() for r_id in revision_ids]
2680
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
2681
return graph.GraphThunkIdsToKeys(known_graph)
2683
1431
def get_graph(self, other_repository=None):
2684
1432
"""Return the graph walker for this repository format"""
2685
1433
parents_provider = self._make_parents_provider()
2686
1434
if (other_repository is not None and
2687
not self.has_same_location(other_repository)):
2688
parents_provider = graph.StackedParentsProvider(
1435
other_repository.bzrdir.transport.base !=
1436
self.bzrdir.transport.base):
1437
parents_provider = graph._StackedParentsProvider(
2689
1438
[parents_provider, other_repository._make_parents_provider()])
2690
1439
return graph.Graph(parents_provider)
2692
def _get_versioned_file_checker(self, text_key_references=None,
2694
"""Return an object suitable for checking versioned files.
2696
:param text_key_references: if non-None, an already built
2697
dictionary mapping text keys ((fileid, revision_id) tuples)
2698
to whether they were referred to by the inventory of the
2699
revision_id that they contain. If None, this will be
2701
:param ancestors: Optional result from
2702
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2705
return _VersionedFileChecker(self,
2706
text_key_references=text_key_references, ancestors=ancestors)
2708
def revision_ids_to_search_result(self, result_set):
2709
"""Convert a set of revision ids to a graph SearchResult."""
2710
result_parents = set()
2711
for parents in self.get_graph().get_parent_map(
2712
result_set).itervalues():
2713
result_parents.update(parents)
2714
included_keys = result_set.intersection(result_parents)
2715
start_keys = result_set.difference(included_keys)
2716
exclude_keys = result_parents.difference(result_set)
2717
result = graph.SearchResult(start_keys, exclude_keys,
2718
len(result_set), result_set)
2721
1441
@needs_write_lock
2722
1442
def set_make_working_trees(self, new_value):
2723
1443
"""Set the policy flag for making working trees when creating branches.
3402
1907
InterRepository.get(other).method_name(parameters).
3405
_walk_to_common_revisions_batch_size = 50
3406
1910
_optimisers = []
3407
1911
"""The available optimised InterRepository types."""
3410
1913
def copy_content(self, revision_id=None):
3411
"""Make a complete copy of the content in self into destination.
3413
This is a destructive operation! Do not use it on existing
3416
:param revision_id: Only copy the content needed to construct
3417
revision_id and its parents.
3420
self.target.set_make_working_trees(self.source.make_working_trees())
3421
except NotImplementedError:
3423
self.target.fetch(self.source, revision_id=revision_id)
3426
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
1914
raise NotImplementedError(self.copy_content)
1916
def fetch(self, revision_id=None, pb=None):
3428
1917
"""Fetch the content required to construct revision_id.
3430
1919
The content is copied from self.source to self.target.
3432
1921
:param revision_id: if None all content is copied, if NULL_REVISION no
3433
1922
content is copied.
3437
ui.ui_factory.warn_experimental_format_fetch(self)
3438
from bzrlib.fetch import RepoFetcher
3439
# See <https://launchpad.net/bugs/456077> asking for a warning here
3440
if self.source._format.network_name() != self.target._format.network_name():
3441
ui.ui_factory.show_user_warning('cross_format_fetch',
3442
from_format=self.source._format,
3443
to_format=self.target._format)
3444
f = RepoFetcher(to_repository=self.target,
3445
from_repository=self.source,
3446
last_revision=revision_id,
3447
fetch_spec=fetch_spec,
3448
find_ghosts=find_ghosts)
3450
def _walk_to_common_revisions(self, revision_ids):
3451
"""Walk out from revision_ids in source to revisions target has.
3453
:param revision_ids: The start point for the search.
3454
:return: A set of revision ids.
3456
target_graph = self.target.get_graph()
3457
revision_ids = frozenset(revision_ids)
3458
missing_revs = set()
3459
source_graph = self.source.get_graph()
3460
# ensure we don't pay silly lookup costs.
3461
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3462
null_set = frozenset([_mod_revision.NULL_REVISION])
3463
searcher_exhausted = False
3467
# Iterate the searcher until we have enough next_revs
3468
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3470
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3471
next_revs.update(next_revs_part)
3472
ghosts.update(ghosts_part)
3473
except StopIteration:
3474
searcher_exhausted = True
3476
# If there are ghosts in the source graph, and the caller asked for
3477
# them, make sure that they are present in the target.
3478
# We don't care about other ghosts as we can't fetch them and
3479
# haven't been asked to.
3480
ghosts_to_check = set(revision_ids.intersection(ghosts))
3481
revs_to_get = set(next_revs).union(ghosts_to_check)
3483
have_revs = set(target_graph.get_parent_map(revs_to_get))
3484
# we always have NULL_REVISION present.
3485
have_revs = have_revs.union(null_set)
3486
# Check if the target is missing any ghosts we need.
3487
ghosts_to_check.difference_update(have_revs)
3489
# One of the caller's revision_ids is a ghost in both the
3490
# source and the target.
3491
raise errors.NoSuchRevision(
3492
self.source, ghosts_to_check.pop())
3493
missing_revs.update(next_revs - have_revs)
3494
# Because we may have walked past the original stop point, make
3495
# sure everything is stopped
3496
stop_revs = searcher.find_seen_ancestors(have_revs)
3497
searcher.stop_searching_any(stop_revs)
3498
if searcher_exhausted:
3500
return searcher.get_result()
1923
:param pb: optional progress bar to use for progress reports. If not
1924
provided a default one will be created.
1926
Returns the copied revision count and the failed revisions in a tuple:
1929
raise NotImplementedError(self.fetch)
3502
1931
@needs_read_lock
3503
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1932
def missing_revision_ids(self, revision_id=None):
3504
1933
"""Return the revision ids that source has that target does not.
1935
These are returned in topological order.
3506
1937
:param revision_id: only return revision ids included by this
3508
:param find_ghosts: If True find missing revisions in deep history
3509
rather than just finding the surface difference.
3510
:return: A bzrlib.graph.SearchResult.
3512
# stop searching at found target revisions.
3513
if not find_ghosts and revision_id is not None:
3514
return self._walk_to_common_revisions([revision_id])
3515
1940
# generic, possibly worst case, slow code path.
3516
1941
target_ids = set(self.target.all_revision_ids())
3517
1942
if revision_id is not None:
3518
1943
source_ids = self.source.get_ancestry(revision_id)
3519
if source_ids[0] is not None:
3520
raise AssertionError()
1944
assert source_ids[0] is None
3521
1945
source_ids.pop(0)
3523
1947
source_ids = self.source.all_revision_ids()
3524
1948
result_set = set(source_ids).difference(target_ids)
3525
return self.source.revision_ids_to_search_result(result_set)
1949
# this may look like a no-op: its not. It preserves the ordering
1950
# other_ids had while only returning the members from other_ids
1951
# that we've decided we need.
1952
return [rev_id for rev_id in source_ids if rev_id in result_set]
3528
1955
def _same_model(source, target):
3529
"""True if source and target have the same data representation.
3531
Note: this is always called on the base class; overriding it in a
3532
subclass will have no effect.
3535
InterRepository._assert_same_model(source, target)
3537
except errors.IncompatibleRepositories, e:
3541
def _assert_same_model(source, target):
3542
"""Raise an exception if two repositories do not use the same model.
1956
"""True if source and target have the same data representation."""
3544
1957
if source.supports_rich_root() != target.supports_rich_root():
3545
raise errors.IncompatibleRepositories(source, target,
3546
"different rich-root support")
3547
1959
if source._serializer != target._serializer:
3548
raise errors.IncompatibleRepositories(source, target,
3549
"different serializers")
3552
1964
class InterSameDataRepository(InterRepository):
3553
1965
"""Code for converting between repositories that represent the same data.
3555
1967
Data format and model must match for this to work.
3559
1971
def _get_repo_format_to_test(self):
3560
1972
"""Repository format for testing with.
3562
1974
InterSameData can pull from subtree to subtree and from non-subtree to
3563
1975
non-subtree, so we test this with the richest repository format.
3723
2189
# we do not have a revision as that would be pointless.
3724
2190
target_ids = set(self.target.all_revision_ids())
3725
2191
possibly_present_revisions = target_ids.intersection(source_ids_set)
3726
actually_present_revisions = set(
3727
self.target._eliminate_revisions_not_present(possibly_present_revisions))
2192
actually_present_revisions = set(self.target._eliminate_revisions_not_present(possibly_present_revisions))
3728
2193
required_revisions = source_ids_set.difference(actually_present_revisions)
2194
required_topo_revisions = [rev_id for rev_id in source_ids if rev_id in required_revisions]
3729
2195
if revision_id is not None:
3730
2196
# we used get_ancestry to determine source_ids then we are assured all
3731
2197
# revisions referenced are present as they are installed in topological order.
3732
2198
# and the tip revision was validated by get_ancestry.
3733
result_set = required_revisions
2199
return required_topo_revisions
3735
# if we just grabbed the possibly available ids, then
2201
# if we just grabbed the possibly available ids, then
3736
2202
# we only have an estimate of whats available and need to validate
3737
2203
# that against the revision records.
3739
self.source._eliminate_revisions_not_present(required_revisions))
3740
return self.source.revision_ids_to_search_result(result_set)
3743
class InterDifferingSerializer(InterRepository):
3746
def _get_repo_format_to_test(self):
3750
def is_compatible(source, target):
3751
"""Be compatible with Knit2 source and Knit3 target"""
3752
# This is redundant with format.check_conversion_target(), however that
3753
# raises an exception, and we just want to say "False" as in we won't
3754
# support converting between these formats.
3755
if 'IDS_never' in debug.debug_flags:
3757
if source.supports_rich_root() and not target.supports_rich_root():
3759
if (source._format.supports_tree_reference
3760
and not target._format.supports_tree_reference):
3762
if target._fallback_repositories and target._format.supports_chks:
3763
# IDS doesn't know how to copy CHKs for the parent inventories it
3764
# adds to stacked repos.
3766
if 'IDS_always' in debug.debug_flags:
3768
# Only use this code path for local source and target. IDS does far
3769
# too much IO (both bandwidth and roundtrips) over a network.
3770
if not source.bzrdir.transport.base.startswith('file:///'):
3772
if not target.bzrdir.transport.base.startswith('file:///'):
3776
def _get_trees(self, revision_ids, cache):
3778
for rev_id in revision_ids:
3780
possible_trees.append((rev_id, cache[rev_id]))
3782
# Not cached, but inventory might be present anyway.
3784
tree = self.source.revision_tree(rev_id)
3785
except errors.NoSuchRevision:
3786
# Nope, parent is ghost.
3789
cache[rev_id] = tree
3790
possible_trees.append((rev_id, tree))
3791
return possible_trees
3793
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3794
"""Get the best delta and base for this revision.
3796
:return: (basis_id, delta)
3799
# Generate deltas against each tree, to find the shortest.
3800
texts_possibly_new_in_tree = set()
3801
for basis_id, basis_tree in possible_trees:
3802
delta = tree.inventory._make_delta(basis_tree.inventory)
3803
for old_path, new_path, file_id, new_entry in delta:
3804
if new_path is None:
3805
# This file_id isn't present in the new rev, so we don't
3809
# Rich roots are handled elsewhere...
3811
kind = new_entry.kind
3812
if kind != 'directory' and kind != 'file':
3813
# No text record associated with this inventory entry.
3815
# This is a directory or file that has changed somehow.
3816
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3817
deltas.append((len(delta), basis_id, delta))
3819
return deltas[0][1:]
3821
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3822
"""Find all parent revisions that are absent, but for which the
3823
inventory is present, and copy those inventories.
3825
This is necessary to preserve correctness when the source is stacked
3826
without fallbacks configured. (Note that in cases like upgrade the
3827
source may be not have _fallback_repositories even though it is
3831
for parents in parent_map.values():
3832
parent_revs.update(parents)
3833
present_parents = self.source.get_parent_map(parent_revs)
3834
absent_parents = set(parent_revs).difference(present_parents)
3835
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3836
(rev_id,) for rev_id in absent_parents)
3837
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3838
for parent_tree in self.source.revision_trees(parent_inv_ids):
3839
current_revision_id = parent_tree.get_revision_id()
3840
parents_parents_keys = parent_invs_keys_for_stacking[
3841
(current_revision_id,)]
3842
parents_parents = [key[-1] for key in parents_parents_keys]
3843
basis_id = _mod_revision.NULL_REVISION
3844
basis_tree = self.source.revision_tree(basis_id)
3845
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3846
self.target.add_inventory_by_delta(
3847
basis_id, delta, current_revision_id, parents_parents)
3848
cache[current_revision_id] = parent_tree
3850
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3851
"""Fetch across a few revisions.
3853
:param revision_ids: The revisions to copy
3854
:param basis_id: The revision_id of a tree that must be in cache, used
3855
as a basis for delta when no other base is available
3856
:param cache: A cache of RevisionTrees that we can use.
3857
:param a_graph: A Graph object to determine the heads() of the
3858
rich-root data stream.
3859
:return: The revision_id of the last converted tree. The RevisionTree
3860
for it will be in cache
3862
# Walk though all revisions; get inventory deltas, copy referenced
3863
# texts that delta references, insert the delta, revision and
3865
root_keys_to_create = set()
3868
pending_revisions = []
3869
parent_map = self.source.get_parent_map(revision_ids)
3870
self._fetch_parent_invs_for_stacking(parent_map, cache)
3871
self.source._safe_to_return_from_cache = True
3872
for tree in self.source.revision_trees(revision_ids):
3873
# Find a inventory delta for this revision.
3874
# Find text entries that need to be copied, too.
3875
current_revision_id = tree.get_revision_id()
3876
parent_ids = parent_map.get(current_revision_id, ())
3877
parent_trees = self._get_trees(parent_ids, cache)
3878
possible_trees = list(parent_trees)
3879
if len(possible_trees) == 0:
3880
# There either aren't any parents, or the parents are ghosts,
3881
# so just use the last converted tree.
3882
possible_trees.append((basis_id, cache[basis_id]))
3883
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3885
revision = self.source.get_revision(current_revision_id)
3886
pending_deltas.append((basis_id, delta,
3887
current_revision_id, revision.parent_ids))
3888
if self._converting_to_rich_root:
3889
self._revision_id_to_root_id[current_revision_id] = \
3891
# Determine which texts are in present in this revision but not in
3892
# any of the available parents.
3893
texts_possibly_new_in_tree = set()
3894
for old_path, new_path, file_id, entry in delta:
3895
if new_path is None:
3896
# This file_id isn't present in the new rev
3900
if not self.target.supports_rich_root():
3901
# The target doesn't support rich root, so we don't
3904
if self._converting_to_rich_root:
3905
# This can't be copied normally, we have to insert
3907
root_keys_to_create.add((file_id, entry.revision))
3910
texts_possibly_new_in_tree.add((file_id, entry.revision))
3911
for basis_id, basis_tree in possible_trees:
3912
basis_inv = basis_tree.inventory
3913
for file_key in list(texts_possibly_new_in_tree):
3914
file_id, file_revision = file_key
3916
entry = basis_inv[file_id]
3917
except errors.NoSuchId:
3919
if entry.revision == file_revision:
3920
texts_possibly_new_in_tree.remove(file_key)
3921
text_keys.update(texts_possibly_new_in_tree)
3922
pending_revisions.append(revision)
3923
cache[current_revision_id] = tree
3924
basis_id = current_revision_id
3925
self.source._safe_to_return_from_cache = False
3927
from_texts = self.source.texts
3928
to_texts = self.target.texts
3929
if root_keys_to_create:
3930
root_stream = _mod_fetch._new_root_data_stream(
3931
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3932
self.source, graph=a_graph)
3933
to_texts.insert_record_stream(root_stream)
3934
to_texts.insert_record_stream(from_texts.get_record_stream(
3935
text_keys, self.target._format._fetch_order,
3936
not self.target._format._fetch_uses_deltas))
3937
# insert inventory deltas
3938
for delta in pending_deltas:
3939
self.target.add_inventory_by_delta(*delta)
3940
if self.target._fallback_repositories:
3941
# Make sure this stacked repository has all the parent inventories
3942
# for the new revisions that we are about to insert. We do this
3943
# before adding the revisions so that no revision is added until
3944
# all the inventories it may depend on are added.
3945
# Note that this is overzealous, as we may have fetched these in an
3948
revision_ids = set()
3949
for revision in pending_revisions:
3950
revision_ids.add(revision.revision_id)
3951
parent_ids.update(revision.parent_ids)
3952
parent_ids.difference_update(revision_ids)
3953
parent_ids.discard(_mod_revision.NULL_REVISION)
3954
parent_map = self.source.get_parent_map(parent_ids)
3955
# we iterate over parent_map and not parent_ids because we don't
3956
# want to try copying any revision which is a ghost
3957
for parent_tree in self.source.revision_trees(parent_map):
3958
current_revision_id = parent_tree.get_revision_id()
3959
parents_parents = parent_map[current_revision_id]
3960
possible_trees = self._get_trees(parents_parents, cache)
3961
if len(possible_trees) == 0:
3962
# There either aren't any parents, or the parents are
3963
# ghosts, so just use the last converted tree.
3964
possible_trees.append((basis_id, cache[basis_id]))
3965
basis_id, delta = self._get_delta_for_revision(parent_tree,
3966
parents_parents, possible_trees)
3967
self.target.add_inventory_by_delta(
3968
basis_id, delta, current_revision_id, parents_parents)
3969
# insert signatures and revisions
3970
for revision in pending_revisions:
3972
signature = self.source.get_signature_text(
3973
revision.revision_id)
3974
self.target.add_signature_text(revision.revision_id,
3976
except errors.NoSuchRevision:
3978
self.target.add_revision(revision.revision_id, revision)
3981
def _fetch_all_revisions(self, revision_ids, pb):
3982
"""Fetch everything for the list of revisions.
3984
:param revision_ids: The list of revisions to fetch. Must be in
3986
:param pb: A ProgressTask
3989
basis_id, basis_tree = self._get_basis(revision_ids[0])
3991
cache = lru_cache.LRUCache(100)
3992
cache[basis_id] = basis_tree
3993
del basis_tree # We don't want to hang on to it here
3995
if self._converting_to_rich_root and len(revision_ids) > 100:
3996
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
4001
for offset in range(0, len(revision_ids), batch_size):
4002
self.target.start_write_group()
4004
pb.update('Transferring revisions', offset,
4006
batch = revision_ids[offset:offset+batch_size]
4007
basis_id = self._fetch_batch(batch, basis_id, cache,
4010
self.source._safe_to_return_from_cache = False
4011
self.target.abort_write_group()
4014
hint = self.target.commit_write_group()
4017
if hints and self.target._format.pack_compresses:
4018
self.target.pack(hint=hints)
4019
pb.update('Transferring revisions', len(revision_ids),
4023
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
4025
"""See InterRepository.fetch()."""
4026
if fetch_spec is not None:
4027
raise AssertionError("Not implemented yet...")
4028
ui.ui_factory.warn_experimental_format_fetch(self)
4029
if (not self.source.supports_rich_root()
4030
and self.target.supports_rich_root()):
4031
self._converting_to_rich_root = True
4032
self._revision_id_to_root_id = {}
4034
self._converting_to_rich_root = False
4035
# See <https://launchpad.net/bugs/456077> asking for a warning here
4036
if self.source._format.network_name() != self.target._format.network_name():
4037
ui.ui_factory.show_user_warning('cross_format_fetch',
4038
from_format=self.source._format,
4039
to_format=self.target._format)
4040
revision_ids = self.target.search_missing_revision_ids(self.source,
4041
revision_id, find_ghosts=find_ghosts).get_keys()
4042
if not revision_ids:
4044
revision_ids = tsort.topo_sort(
4045
self.source.get_graph().get_parent_map(revision_ids))
4046
if not revision_ids:
4048
# Walk though all revisions; get inventory deltas, copy referenced
4049
# texts that delta references, insert the delta, revision and
4052
my_pb = ui.ui_factory.nested_progress_bar()
4055
symbol_versioning.warn(
4056
symbol_versioning.deprecated_in((1, 14, 0))
4057
% "pb parameter to fetch()")
4060
self._fetch_all_revisions(revision_ids, pb)
4062
if my_pb is not None:
4064
return len(revision_ids), 0
4066
def _get_basis(self, first_revision_id):
4067
"""Get a revision and tree which exists in the target.
4069
This assumes that first_revision_id is selected for transmission
4070
because all other ancestors are already present. If we can't find an
4071
ancestor we fall back to NULL_REVISION since we know that is safe.
4073
:return: (basis_id, basis_tree)
4075
first_rev = self.source.get_revision(first_revision_id)
4077
basis_id = first_rev.parent_ids[0]
4078
# only valid as a basis if the target has it
4079
self.target.get_revision(basis_id)
4080
# Try to get a basis tree - if its a ghost it will hit the
4081
# NoSuchRevision case.
4082
basis_tree = self.source.revision_tree(basis_id)
4083
except (IndexError, errors.NoSuchRevision):
4084
basis_id = _mod_revision.NULL_REVISION
4085
basis_tree = self.source.revision_tree(basis_id)
4086
return basis_id, basis_tree
4089
InterRepository.register_optimiser(InterDifferingSerializer)
2204
return self.source._eliminate_revisions_not_present(required_topo_revisions)
2207
class InterModel1and2(InterRepository):
2210
def _get_repo_format_to_test(self):
2214
def is_compatible(source, target):
2215
if not source.supports_rich_root() and target.supports_rich_root():
2221
def fetch(self, revision_id=None, pb=None):
2222
"""See InterRepository.fetch()."""
2223
from bzrlib.fetch import Model1toKnit2Fetcher
2224
f = Model1toKnit2Fetcher(to_repository=self.target,
2225
from_repository=self.source,
2226
last_revision=revision_id,
2228
return f.count_copied, f.failed_revisions
2231
def copy_content(self, revision_id=None):
2232
"""Make a complete copy of the content in self into destination.
2234
This is a destructive operation! Do not use it on existing
2237
:param revision_id: Only copy the content needed to construct
2238
revision_id and its parents.
2241
self.target.set_make_working_trees(self.source.make_working_trees())
2242
except NotImplementedError:
2244
# but don't bother fetching if we have the needed data now.
2245
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
2246
self.target.has_revision(revision_id)):
2248
self.target.fetch(self.source, revision_id=revision_id)
2251
class InterKnit1and2(InterKnitRepo):
2254
def _get_repo_format_to_test(self):
2258
def is_compatible(source, target):
2259
"""Be compatible with Knit1 source and Knit3 target"""
2260
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit3
2262
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1, \
2263
RepositoryFormatKnit3
2264
return (isinstance(source._format, (RepositoryFormatKnit1)) and
2265
isinstance(target._format, (RepositoryFormatKnit3)))
2266
except AttributeError:
2270
def fetch(self, revision_id=None, pb=None):
2271
"""See InterRepository.fetch()."""
2272
from bzrlib.fetch import Knit1to2Fetcher
2273
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
2274
self.source, self.source._format, self.target,
2275
self.target._format)
2276
f = Knit1to2Fetcher(to_repository=self.target,
2277
from_repository=self.source,
2278
last_revision=revision_id,
2280
return f.count_copied, f.failed_revisions
2283
class InterRemoteRepository(InterRepository):
2284
"""Code for converting between RemoteRepository objects.
2286
This just gets an non-remote repository from the RemoteRepository, and calls
2287
InterRepository.get again.
2290
def __init__(self, source, target):
2291
if isinstance(source, remote.RemoteRepository):
2292
source._ensure_real()
2293
real_source = source._real_repository
2295
real_source = source
2296
if isinstance(target, remote.RemoteRepository):
2297
target._ensure_real()
2298
real_target = target._real_repository
2300
real_target = target
2301
self.real_inter = InterRepository.get(real_source, real_target)
2304
def is_compatible(source, target):
2305
if isinstance(source, remote.RemoteRepository):
2307
if isinstance(target, remote.RemoteRepository):
2311
def copy_content(self, revision_id=None):
2312
self.real_inter.copy_content(revision_id=revision_id)
2314
def fetch(self, revision_id=None, pb=None):
2315
self.real_inter.fetch(revision_id=revision_id, pb=pb)
2318
def _get_repo_format_to_test(self):
4090
2322
InterRepository.register_optimiser(InterSameDataRepository)
4091
2323
InterRepository.register_optimiser(InterWeaveRepo)
4092
2324
InterRepository.register_optimiser(InterKnitRepo)
2325
InterRepository.register_optimiser(InterModel1and2)
2326
InterRepository.register_optimiser(InterKnit1and2)
2327
InterRepository.register_optimiser(InterRemoteRepository)
4095
2330
class CopyConverter(object):
4096
2331
"""A repository conversion tool which just performs a copy of the content.
4098
2333
This is slow but quite reliable.
4168
2407
if _unescape_re is None:
4169
2408
_unescape_re = re.compile('\&([^;]*);')
4170
2409
return _unescape_re.sub(_unescaper, data)
4173
class _VersionedFileChecker(object):
4175
def __init__(self, repository, text_key_references=None, ancestors=None):
4176
self.repository = repository
4177
self.text_index = self.repository._generate_text_key_index(
4178
text_key_references=text_key_references, ancestors=ancestors)
4180
def calculate_file_version_parents(self, text_key):
4181
"""Calculate the correct parents for a file version according to
4184
parent_keys = self.text_index[text_key]
4185
if parent_keys == [_mod_revision.NULL_REVISION]:
4187
return tuple(parent_keys)
4189
def check_file_version_parents(self, texts, progress_bar=None):
4190
"""Check the parents stored in a versioned file are correct.
4192
It also detects file versions that are not referenced by their
4193
corresponding revision's inventory.
4195
:returns: A tuple of (wrong_parents, dangling_file_versions).
4196
wrong_parents is a dict mapping {revision_id: (stored_parents,
4197
correct_parents)} for each revision_id where the stored parents
4198
are not correct. dangling_file_versions is a set of (file_id,
4199
revision_id) tuples for versions that are present in this versioned
4200
file, but not used by the corresponding inventory.
4202
local_progress = None
4203
if progress_bar is None:
4204
local_progress = ui.ui_factory.nested_progress_bar()
4205
progress_bar = local_progress
4207
return self._check_file_version_parents(texts, progress_bar)
4210
local_progress.finished()
4212
def _check_file_version_parents(self, texts, progress_bar):
4213
"""See check_file_version_parents."""
4215
self.file_ids = set([file_id for file_id, _ in
4216
self.text_index.iterkeys()])
4217
# text keys is now grouped by file_id
4218
n_versions = len(self.text_index)
4219
progress_bar.update('loading text store', 0, n_versions)
4220
parent_map = self.repository.texts.get_parent_map(self.text_index)
4221
# On unlistable transports this could well be empty/error...
4222
text_keys = self.repository.texts.keys()
4223
unused_keys = frozenset(text_keys) - set(self.text_index)
4224
for num, key in enumerate(self.text_index.iterkeys()):
4225
progress_bar.update('checking text graph', num, n_versions)
4226
correct_parents = self.calculate_file_version_parents(key)
4228
knit_parents = parent_map[key]
4229
except errors.RevisionNotPresent:
4232
if correct_parents != knit_parents:
4233
wrong_parents[key] = (knit_parents, correct_parents)
4234
return wrong_parents, unused_keys
4237
def _old_get_graph(repository, revision_id):
4238
"""DO NOT USE. That is all. I'm serious."""
4239
graph = repository.get_graph()
4240
revision_graph = dict(((key, value) for key, value in
4241
graph.iter_ancestry([revision_id]) if value is not None))
4242
return _strip_NULL_ghosts(revision_graph)
4245
def _strip_NULL_ghosts(revision_graph):
4246
"""Also don't use this. more compatibility code for unmigrated clients."""
4247
# Filter ghosts, and null:
4248
if _mod_revision.NULL_REVISION in revision_graph:
4249
del revision_graph[_mod_revision.NULL_REVISION]
4250
for key, parents in revision_graph.items():
4251
revision_graph[key] = tuple(parent for parent in parents if parent
4253
return revision_graph
4256
class StreamSink(object):
4257
"""An object that can insert a stream into a repository.
4259
This interface handles the complexity of reserialising inventories and
4260
revisions from different formats, and allows unidirectional insertion into
4261
stacked repositories without looking for the missing basis parents
4265
def __init__(self, target_repo):
4266
self.target_repo = target_repo
4268
def insert_stream(self, stream, src_format, resume_tokens):
4269
"""Insert a stream's content into the target repository.
4271
:param src_format: a bzr repository format.
4273
:return: a list of resume tokens and an iterable of keys additional
4274
items required before the insertion can be completed.
4276
self.target_repo.lock_write()
4279
self.target_repo.resume_write_group(resume_tokens)
4282
self.target_repo.start_write_group()
4285
# locked_insert_stream performs a commit|suspend.
4286
return self._locked_insert_stream(stream, src_format, is_resume)
4288
self.target_repo.abort_write_group(suppress_errors=True)
4291
self.target_repo.unlock()
4293
def _locked_insert_stream(self, stream, src_format, is_resume):
4294
to_serializer = self.target_repo._format._serializer
4295
src_serializer = src_format._serializer
4297
if to_serializer == src_serializer:
4298
# If serializers match and the target is a pack repository, set the
4299
# write cache size on the new pack. This avoids poor performance
4300
# on transports where append is unbuffered (such as
4301
# RemoteTransport). This is safe to do because nothing should read
4302
# back from the target repository while a stream with matching
4303
# serialization is being inserted.
4304
# The exception is that a delta record from the source that should
4305
# be a fulltext may need to be expanded by the target (see
4306
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4307
# explicitly flush any buffered writes first in that rare case.
4309
new_pack = self.target_repo._pack_collection._new_pack
4310
except AttributeError:
4311
# Not a pack repository
4314
new_pack.set_write_cache_size(1024*1024)
4315
for substream_type, substream in stream:
4316
if 'stream' in debug.debug_flags:
4317
mutter('inserting substream: %s', substream_type)
4318
if substream_type == 'texts':
4319
self.target_repo.texts.insert_record_stream(substream)
4320
elif substream_type == 'inventories':
4321
if src_serializer == to_serializer:
4322
self.target_repo.inventories.insert_record_stream(
4325
self._extract_and_insert_inventories(
4326
substream, src_serializer)
4327
elif substream_type == 'inventory-deltas':
4328
self._extract_and_insert_inventory_deltas(
4329
substream, src_serializer)
4330
elif substream_type == 'chk_bytes':
4331
# XXX: This doesn't support conversions, as it assumes the
4332
# conversion was done in the fetch code.
4333
self.target_repo.chk_bytes.insert_record_stream(substream)
4334
elif substream_type == 'revisions':
4335
# This may fallback to extract-and-insert more often than
4336
# required if the serializers are different only in terms of
4338
if src_serializer == to_serializer:
4339
self.target_repo.revisions.insert_record_stream(
4342
self._extract_and_insert_revisions(substream,
4344
elif substream_type == 'signatures':
4345
self.target_repo.signatures.insert_record_stream(substream)
4347
raise AssertionError('kaboom! %s' % (substream_type,))
4348
# Done inserting data, and the missing_keys calculations will try to
4349
# read back from the inserted data, so flush the writes to the new pack
4350
# (if this is pack format).
4351
if new_pack is not None:
4352
new_pack._write_data('', flush=True)
4353
# Find all the new revisions (including ones from resume_tokens)
4354
missing_keys = self.target_repo.get_missing_parent_inventories(
4355
check_for_missing_texts=is_resume)
4357
for prefix, versioned_file in (
4358
('texts', self.target_repo.texts),
4359
('inventories', self.target_repo.inventories),
4360
('revisions', self.target_repo.revisions),
4361
('signatures', self.target_repo.signatures),
4362
('chk_bytes', self.target_repo.chk_bytes),
4364
if versioned_file is None:
4366
# TODO: key is often going to be a StaticTuple object
4367
# I don't believe we can define a method by which
4368
# (prefix,) + StaticTuple will work, though we could
4369
# define a StaticTuple.sq_concat that would allow you to
4370
# pass in either a tuple or a StaticTuple as the second
4371
# object, so instead we could have:
4372
# StaticTuple(prefix) + key here...
4373
missing_keys.update((prefix,) + key for key in
4374
versioned_file.get_missing_compression_parent_keys())
4375
except NotImplementedError:
4376
# cannot even attempt suspending, and missing would have failed
4377
# during stream insertion.
4378
missing_keys = set()
4381
# suspend the write group and tell the caller what we is
4382
# missing. We know we can suspend or else we would not have
4383
# entered this code path. (All repositories that can handle
4384
# missing keys can handle suspending a write group).
4385
write_group_tokens = self.target_repo.suspend_write_group()
4386
return write_group_tokens, missing_keys
4387
hint = self.target_repo.commit_write_group()
4388
if (to_serializer != src_serializer and
4389
self.target_repo._format.pack_compresses):
4390
self.target_repo.pack(hint=hint)
4393
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4394
target_rich_root = self.target_repo._format.rich_root_data
4395
target_tree_refs = self.target_repo._format.supports_tree_reference
4396
for record in substream:
4397
# Insert the delta directly
4398
inventory_delta_bytes = record.get_bytes_as('fulltext')
4399
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4401
parse_result = deserialiser.parse_text_bytes(
4402
inventory_delta_bytes)
4403
except inventory_delta.IncompatibleInventoryDelta, err:
4404
trace.mutter("Incompatible delta: %s", err.msg)
4405
raise errors.IncompatibleRevision(self.target_repo._format)
4406
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4407
revision_id = new_id
4408
parents = [key[0] for key in record.parents]
4409
self.target_repo.add_inventory_by_delta(
4410
basis_id, inv_delta, revision_id, parents)
4412
def _extract_and_insert_inventories(self, substream, serializer,
4414
"""Generate a new inventory versionedfile in target, converting data.
4416
The inventory is retrieved from the source, (deserializing it), and
4417
stored in the target (reserializing it in a different format).
4419
target_rich_root = self.target_repo._format.rich_root_data
4420
target_tree_refs = self.target_repo._format.supports_tree_reference
4421
for record in substream:
4422
# It's not a delta, so it must be a fulltext in the source
4423
# serializer's format.
4424
bytes = record.get_bytes_as('fulltext')
4425
revision_id = record.key[0]
4426
inv = serializer.read_inventory_from_string(bytes, revision_id)
4427
parents = [key[0] for key in record.parents]
4428
self.target_repo.add_inventory(revision_id, inv, parents)
4429
# No need to keep holding this full inv in memory when the rest of
4430
# the substream is likely to be all deltas.
4433
def _extract_and_insert_revisions(self, substream, serializer):
4434
for record in substream:
4435
bytes = record.get_bytes_as('fulltext')
4436
revision_id = record.key[0]
4437
rev = serializer.read_revision_from_string(bytes)
4438
if rev.revision_id != revision_id:
4439
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4440
self.target_repo.add_revision(revision_id, rev)
4443
if self.target_repo._format._fetch_reconcile:
4444
self.target_repo.reconcile()
4447
class StreamSource(object):
4448
"""A source of a stream for fetching between repositories."""
4450
def __init__(self, from_repository, to_format):
4451
"""Create a StreamSource streaming from from_repository."""
4452
self.from_repository = from_repository
4453
self.to_format = to_format
4455
def delta_on_metadata(self):
4456
"""Return True if delta's are permitted on metadata streams.
4458
That is on revisions and signatures.
4460
src_serializer = self.from_repository._format._serializer
4461
target_serializer = self.to_format._serializer
4462
return (self.to_format._fetch_uses_deltas and
4463
src_serializer == target_serializer)
4465
def _fetch_revision_texts(self, revs):
4466
# fetch signatures first and then the revision texts
4467
# may need to be a InterRevisionStore call here.
4468
from_sf = self.from_repository.signatures
4469
# A missing signature is just skipped.
4470
keys = [(rev_id,) for rev_id in revs]
4471
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4473
self.to_format._fetch_order,
4474
not self.to_format._fetch_uses_deltas))
4475
# If a revision has a delta, this is actually expanded inside the
4476
# insert_record_stream code now, which is an alternate fix for
4478
from_rf = self.from_repository.revisions
4479
revisions = from_rf.get_record_stream(
4481
self.to_format._fetch_order,
4482
not self.delta_on_metadata())
4483
return [('signatures', signatures), ('revisions', revisions)]
4485
def _generate_root_texts(self, revs):
4486
"""This will be called by get_stream between fetching weave texts and
4487
fetching the inventory weave.
4489
if self._rich_root_upgrade():
4490
return _mod_fetch.Inter1and2Helper(
4491
self.from_repository).generate_root_texts(revs)
4495
def get_stream(self, search):
4497
revs = search.get_keys()
4498
graph = self.from_repository.get_graph()
4499
revs = tsort.topo_sort(graph.get_parent_map(revs))
4500
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4502
for knit_kind, file_id, revisions in data_to_fetch:
4503
if knit_kind != phase:
4505
# Make a new progress bar for this phase
4506
if knit_kind == "file":
4507
# Accumulate file texts
4508
text_keys.extend([(file_id, revision) for revision in
4510
elif knit_kind == "inventory":
4511
# Now copy the file texts.
4512
from_texts = self.from_repository.texts
4513
yield ('texts', from_texts.get_record_stream(
4514
text_keys, self.to_format._fetch_order,
4515
not self.to_format._fetch_uses_deltas))
4516
# Cause an error if a text occurs after we have done the
4519
# Before we process the inventory we generate the root
4520
# texts (if necessary) so that the inventories references
4522
for _ in self._generate_root_texts(revs):
4524
# we fetch only the referenced inventories because we do not
4525
# know for unselected inventories whether all their required
4526
# texts are present in the other repository - it could be
4528
for info in self._get_inventory_stream(revs):
4530
elif knit_kind == "signatures":
4531
# Nothing to do here; this will be taken care of when
4532
# _fetch_revision_texts happens.
4534
elif knit_kind == "revisions":
4535
for record in self._fetch_revision_texts(revs):
4538
raise AssertionError("Unknown knit kind %r" % knit_kind)
4540
def get_stream_for_missing_keys(self, missing_keys):
4541
# missing keys can only occur when we are byte copying and not
4542
# translating (because translation means we don't send
4543
# unreconstructable deltas ever).
4545
keys['texts'] = set()
4546
keys['revisions'] = set()
4547
keys['inventories'] = set()
4548
keys['chk_bytes'] = set()
4549
keys['signatures'] = set()
4550
for key in missing_keys:
4551
keys[key[0]].add(key[1:])
4552
if len(keys['revisions']):
4553
# If we allowed copying revisions at this point, we could end up
4554
# copying a revision without copying its required texts: a
4555
# violation of the requirements for repository integrity.
4556
raise AssertionError(
4557
'cannot copy revisions to fill in missing deltas %s' % (
4558
keys['revisions'],))
4559
for substream_kind, keys in keys.iteritems():
4560
vf = getattr(self.from_repository, substream_kind)
4561
if vf is None and keys:
4562
raise AssertionError(
4563
"cannot fill in keys for a versioned file we don't"
4564
" have: %s needs %s" % (substream_kind, keys))
4566
# No need to stream something we don't have
4568
if substream_kind == 'inventories':
4569
# Some missing keys are genuinely ghosts, filter those out.
4570
present = self.from_repository.inventories.get_parent_map(keys)
4571
revs = [key[0] for key in present]
4572
# Get the inventory stream more-or-less as we do for the
4573
# original stream; there's no reason to assume that records
4574
# direct from the source will be suitable for the sink. (Think
4575
# e.g. 2a -> 1.9-rich-root).
4576
for info in self._get_inventory_stream(revs, missing=True):
4580
# Ask for full texts always so that we don't need more round trips
4581
# after this stream.
4582
# Some of the missing keys are genuinely ghosts, so filter absent
4583
# records. The Sink is responsible for doing another check to
4584
# ensure that ghosts don't introduce missing data for future
4586
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4587
self.to_format._fetch_order, True))
4588
yield substream_kind, stream
4590
def inventory_fetch_order(self):
4591
if self._rich_root_upgrade():
4592
return 'topological'
4594
return self.to_format._fetch_order
4596
def _rich_root_upgrade(self):
4597
return (not self.from_repository._format.rich_root_data and
4598
self.to_format.rich_root_data)
4600
def _get_inventory_stream(self, revision_ids, missing=False):
4601
from_format = self.from_repository._format
4602
if (from_format.supports_chks and self.to_format.supports_chks and
4603
from_format.network_name() == self.to_format.network_name()):
4604
raise AssertionError(
4605
"this case should be handled by GroupCHKStreamSource")
4606
elif 'forceinvdeltas' in debug.debug_flags:
4607
return self._get_convertable_inventory_stream(revision_ids,
4608
delta_versus_null=missing)
4609
elif from_format.network_name() == self.to_format.network_name():
4611
return self._get_simple_inventory_stream(revision_ids,
4613
elif (not from_format.supports_chks and not self.to_format.supports_chks
4614
and from_format._serializer == self.to_format._serializer):
4615
# Essentially the same format.
4616
return self._get_simple_inventory_stream(revision_ids,
4619
# Any time we switch serializations, we want to use an
4620
# inventory-delta based approach.
4621
return self._get_convertable_inventory_stream(revision_ids,
4622
delta_versus_null=missing)
4624
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4625
# NB: This currently reopens the inventory weave in source;
4626
# using a single stream interface instead would avoid this.
4627
from_weave = self.from_repository.inventories
4629
delta_closure = True
4631
delta_closure = not self.delta_on_metadata()
4632
yield ('inventories', from_weave.get_record_stream(
4633
[(rev_id,) for rev_id in revision_ids],
4634
self.inventory_fetch_order(), delta_closure))
4636
def _get_convertable_inventory_stream(self, revision_ids,
4637
delta_versus_null=False):
4638
# The two formats are sufficiently different that there is no fast
4639
# path, so we need to send just inventorydeltas, which any
4640
# sufficiently modern client can insert into any repository.
4641
# The StreamSink code expects to be able to
4642
# convert on the target, so we need to put bytes-on-the-wire that can
4643
# be converted. That means inventory deltas (if the remote is <1.19,
4644
# RemoteStreamSink will fallback to VFS to insert the deltas).
4645
yield ('inventory-deltas',
4646
self._stream_invs_as_deltas(revision_ids,
4647
delta_versus_null=delta_versus_null))
4649
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4650
"""Return a stream of inventory-deltas for the given rev ids.
4652
:param revision_ids: The list of inventories to transmit
4653
:param delta_versus_null: Don't try to find a minimal delta for this
4654
entry, instead compute the delta versus the NULL_REVISION. This
4655
effectively streams a complete inventory. Used for stuff like
4656
filling in missing parents, etc.
4658
from_repo = self.from_repository
4659
revision_keys = [(rev_id,) for rev_id in revision_ids]
4660
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4661
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4663
inventories = self.from_repository.iter_inventories(
4664
revision_ids, 'topological')
4665
format = from_repo._format
4666
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4667
inventory_cache = lru_cache.LRUCache(50)
4668
null_inventory = from_repo.revision_tree(
4669
_mod_revision.NULL_REVISION).inventory
4670
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4671
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4672
# repo back into a non-rich-root repo ought to be allowed)
4673
serializer = inventory_delta.InventoryDeltaSerializer(
4674
versioned_root=format.rich_root_data,
4675
tree_references=format.supports_tree_reference)
4676
for inv in inventories:
4677
key = (inv.revision_id,)
4678
parent_keys = parent_map.get(key, ())
4680
if not delta_versus_null and parent_keys:
4681
# The caller did not ask for complete inventories and we have
4682
# some parents that we can delta against. Make a delta against
4683
# each parent so that we can find the smallest.
4684
parent_ids = [parent_key[0] for parent_key in parent_keys]
4685
for parent_id in parent_ids:
4686
if parent_id not in invs_sent_so_far:
4687
# We don't know that the remote side has this basis, so
4690
if parent_id == _mod_revision.NULL_REVISION:
4691
parent_inv = null_inventory
4693
parent_inv = inventory_cache.get(parent_id, None)
4694
if parent_inv is None:
4695
parent_inv = from_repo.get_inventory(parent_id)
4696
candidate_delta = inv._make_delta(parent_inv)
4697
if (delta is None or
4698
len(delta) > len(candidate_delta)):
4699
delta = candidate_delta
4700
basis_id = parent_id
4702
# Either none of the parents ended up being suitable, or we
4703
# were asked to delta against NULL
4704
basis_id = _mod_revision.NULL_REVISION
4705
delta = inv._make_delta(null_inventory)
4706
invs_sent_so_far.add(inv.revision_id)
4707
inventory_cache[inv.revision_id] = inv
4708
delta_serialized = ''.join(
4709
serializer.delta_to_lines(basis_id, key[-1], delta))
4710
yield versionedfile.FulltextContentFactory(
4711
key, parent_keys, None, delta_serialized)
4714
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4715
stop_revision=None):
4716
"""Extend the partial history to include a given index
4718
If a stop_index is supplied, stop when that index has been reached.
4719
If a stop_revision is supplied, stop when that revision is
4720
encountered. Otherwise, stop when the beginning of history is
4723
:param stop_index: The index which should be present. When it is
4724
present, history extension will stop.
4725
:param stop_revision: The revision id which should be present. When
4726
it is encountered, history extension will stop.
4728
start_revision = partial_history_cache[-1]
4729
iterator = repo.iter_reverse_revision_history(start_revision)
4731
#skip the last revision in the list
4734
if (stop_index is not None and
4735
len(partial_history_cache) > stop_index):
4737
if partial_history_cache[-1] == stop_revision:
4739
revision_id = iterator.next()
4740
partial_history_cache.append(revision_id)
4741
except StopIteration:
added
removed
bzrlib/repository.py
