548
492
ie.revision = parent_entry.revision
549
493
return self._get_delta(ie, basis_inv, path), False, None
550
494
ie.reference_revision = content_summary[3]
551
if ie.reference_revision is None:
552
raise AssertionError("invalid content_summary for nested tree: %r"
553
% (content_summary,))
554
self._add_text_to_weave(ie.file_id, '', heads, None)
496
self._add_text_to_weave(ie.file_id, lines, heads, None)
556
498
raise NotImplementedError('unknown kind')
557
499
ie.revision = self._new_revision_id
558
self._any_changes = True
559
500
return self._get_delta(ie, basis_inv, path), True, fingerprint
561
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
562
_entry_factory=entry_factory):
563
"""Record a new tree via iter_changes.
565
:param tree: The tree to obtain text contents from for changed objects.
566
:param basis_revision_id: The revision id of the tree the iter_changes
567
has been generated against. Currently assumed to be the same
568
as self.parents[0] - if it is not, errors may occur.
569
:param iter_changes: An iter_changes iterator with the changes to apply
570
to basis_revision_id. The iterator must not include any items with
571
a current kind of None - missing items must be either filtered out
572
or errored-on beefore record_iter_changes sees the item.
573
:param _entry_factory: Private method to bind entry_factory locally for
575
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
578
# Create an inventory delta based on deltas between all the parents and
579
# deltas between all the parent inventories. We use inventory delta's
580
# between the inventory objects because iter_changes masks
581
# last-changed-field only changes.
583
# file_id -> change map, change is fileid, paths, changed, versioneds,
584
# parents, names, kinds, executables
586
# {file_id -> revision_id -> inventory entry, for entries in parent
587
# trees that are not parents[0]
591
revtrees = list(self.repository.revision_trees(self.parents))
592
except errors.NoSuchRevision:
593
# one or more ghosts, slow path.
595
for revision_id in self.parents:
597
revtrees.append(self.repository.revision_tree(revision_id))
598
except errors.NoSuchRevision:
600
basis_revision_id = _mod_revision.NULL_REVISION
602
revtrees.append(self.repository.revision_tree(
603
_mod_revision.NULL_REVISION))
604
# The basis inventory from a repository
606
basis_inv = revtrees[0].inventory
608
basis_inv = self.repository.revision_tree(
609
_mod_revision.NULL_REVISION).inventory
610
if len(self.parents) > 0:
611
if basis_revision_id != self.parents[0] and not ghost_basis:
613
"arbitrary basis parents not yet supported with merges")
614
for revtree in revtrees[1:]:
615
for change in revtree.inventory._make_delta(basis_inv):
616
if change[1] is None:
617
# Not present in this parent.
619
if change[2] not in merged_ids:
620
if change[0] is not None:
621
basis_entry = basis_inv[change[2]]
622
merged_ids[change[2]] = [
624
basis_entry.revision,
627
parent_entries[change[2]] = {
629
basis_entry.revision:basis_entry,
631
change[3].revision:change[3],
634
merged_ids[change[2]] = [change[3].revision]
635
parent_entries[change[2]] = {change[3].revision:change[3]}
637
merged_ids[change[2]].append(change[3].revision)
638
parent_entries[change[2]][change[3].revision] = change[3]
641
# Setup the changes from the tree:
642
# changes maps file_id -> (change, [parent revision_ids])
644
for change in iter_changes:
645
# This probably looks up in basis_inv way to much.
646
if change[1][0] is not None:
647
head_candidate = [basis_inv[change[0]].revision]
650
changes[change[0]] = change, merged_ids.get(change[0],
652
unchanged_merged = set(merged_ids) - set(changes)
653
# Extend the changes dict with synthetic changes to record merges of
655
for file_id in unchanged_merged:
656
# Record a merged version of these items that did not change vs the
657
# basis. This can be either identical parallel changes, or a revert
658
# of a specific file after a merge. The recorded content will be
659
# that of the current tree (which is the same as the basis), but
660
# the per-file graph will reflect a merge.
661
# NB:XXX: We are reconstructing path information we had, this
662
# should be preserved instead.
663
# inv delta change: (file_id, (path_in_source, path_in_target),
664
# changed_content, versioned, parent, name, kind,
667
basis_entry = basis_inv[file_id]
668
except errors.NoSuchId:
669
# a change from basis->some_parents but file_id isn't in basis
670
# so was new in the merge, which means it must have changed
671
# from basis -> current, and as it hasn't the add was reverted
672
# by the user. So we discard this change.
676
(basis_inv.id2path(file_id), tree.id2path(file_id)),
678
(basis_entry.parent_id, basis_entry.parent_id),
679
(basis_entry.name, basis_entry.name),
680
(basis_entry.kind, basis_entry.kind),
681
(basis_entry.executable, basis_entry.executable))
682
changes[file_id] = (change, merged_ids[file_id])
683
# changes contains tuples with the change and a set of inventory
684
# candidates for the file.
686
# old_path, new_path, file_id, new_inventory_entry
687
seen_root = False # Is the root in the basis delta?
688
inv_delta = self._basis_delta
689
modified_rev = self._new_revision_id
690
for change, head_candidates in changes.values():
691
if change[3][1]: # versioned in target.
692
# Several things may be happening here:
693
# We may have a fork in the per-file graph
694
# - record a change with the content from tree
695
# We may have a change against < all trees
696
# - carry over the tree that hasn't changed
697
# We may have a change against all trees
698
# - record the change with the content from tree
701
entry = _entry_factory[kind](file_id, change[5][1],
703
head_set = self._heads(change[0], set(head_candidates))
706
for head_candidate in head_candidates:
707
if head_candidate in head_set:
708
heads.append(head_candidate)
709
head_set.remove(head_candidate)
712
# Could be a carry-over situation:
713
parent_entry_revs = parent_entries.get(file_id, None)
714
if parent_entry_revs:
715
parent_entry = parent_entry_revs.get(heads[0], None)
718
if parent_entry is None:
719
# The parent iter_changes was called against is the one
720
# that is the per-file head, so any change is relevant
721
# iter_changes is valid.
722
carry_over_possible = False
724
# could be a carry over situation
725
# A change against the basis may just indicate a merge,
726
# we need to check the content against the source of the
727
# merge to determine if it was changed after the merge
729
if (parent_entry.kind != entry.kind or
730
parent_entry.parent_id != entry.parent_id or
731
parent_entry.name != entry.name):
732
# Metadata common to all entries has changed
733
# against per-file parent
734
carry_over_possible = False
736
carry_over_possible = True
737
# per-type checks for changes against the parent_entry
740
# Cannot be a carry-over situation
741
carry_over_possible = False
742
# Populate the entry in the delta
744
# XXX: There is still a small race here: If someone reverts the content of a file
745
# after iter_changes examines and decides it has changed,
746
# we will unconditionally record a new version even if some
747
# other process reverts it while commit is running (with
748
# the revert happening after iter_changes did it's
751
entry.executable = True
753
entry.executable = False
754
if (carry_over_possible and
755
parent_entry.executable == entry.executable):
756
# Check the file length, content hash after reading
758
nostore_sha = parent_entry.text_sha1
761
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
763
text = file_obj.read()
767
entry.text_sha1, entry.text_size = self._add_text_to_weave(
768
file_id, text, heads, nostore_sha)
769
yield file_id, change[1][1], (entry.text_sha1, stat_value)
770
except errors.ExistingContent:
771
# No content change against a carry_over parent
772
# Perhaps this should also yield a fs hash update?
774
entry.text_size = parent_entry.text_size
775
entry.text_sha1 = parent_entry.text_sha1
776
elif kind == 'symlink':
778
entry.symlink_target = tree.get_symlink_target(file_id)
779
if (carry_over_possible and
780
parent_entry.symlink_target == entry.symlink_target):
783
self._add_text_to_weave(change[0], '', heads, None)
784
elif kind == 'directory':
785
if carry_over_possible:
788
# Nothing to set on the entry.
789
# XXX: split into the Root and nonRoot versions.
790
if change[1][1] != '' or self.repository.supports_rich_root():
791
self._add_text_to_weave(change[0], '', heads, None)
792
elif kind == 'tree-reference':
793
if not self.repository._format.supports_tree_reference:
794
# This isn't quite sane as an error, but we shouldn't
795
# ever see this code path in practice: tree's don't
796
# permit references when the repo doesn't support tree
798
raise errors.UnsupportedOperation(tree.add_reference,
800
reference_revision = tree.get_reference_revision(change[0])
801
entry.reference_revision = reference_revision
802
if (carry_over_possible and
803
parent_entry.reference_revision == reference_revision):
806
self._add_text_to_weave(change[0], '', heads, None)
808
raise AssertionError('unknown kind %r' % kind)
810
entry.revision = modified_rev
812
entry.revision = parent_entry.revision
815
new_path = change[1][1]
816
inv_delta.append((change[1][0], new_path, change[0], entry))
819
self.new_inventory = None
821
# This should perhaps be guarded by a check that the basis we
822
# commit against is the basis for the commit and if not do a delta
824
self._any_changes = True
826
# housekeeping root entry changes do not affect no-change commits.
827
self._require_root_change(tree)
828
self.basis_delta_revision = basis_revision_id
830
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
831
parent_keys = tuple([(file_id, parent) for parent in parents])
832
return self.repository.texts._add_text(
833
(file_id, self._new_revision_id), parent_keys, new_text,
834
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
502
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
503
# Note: as we read the content directly from the tree, we know its not
504
# been turned into unicode or badly split - but a broken tree
505
# implementation could give us bad output from readlines() so this is
506
# not a guarantee of safety. What would be better is always checking
507
# the content during test suite execution. RBC 20070912
508
parent_keys = tuple((file_id, parent) for parent in parents)
509
return self.repository.texts.add_lines(
510
(file_id, self._new_revision_id), parent_keys, new_lines,
511
nostore_sha=nostore_sha, random_id=self.random_revid,
512
check_content=False)[0:2]
837
515
class RootCommitBuilder(CommitBuilder):
838
516
"""This commitbuilder actually records the root id"""
840
518
# the root entry gets versioned properly by this builder.
841
519
_versioned_root = True
1181
797
# The old API returned a list, should this actually be a set?
1182
798
return parent_map.keys()
1184
def _check_inventories(self, checker):
1185
"""Check the inventories found from the revision scan.
1187
This is responsible for verifying the sha1 of inventories and
1188
creating a pending_keys set that covers data referenced by inventories.
1190
bar = ui.ui_factory.nested_progress_bar()
1192
self._do_check_inventories(checker, bar)
1196
def _do_check_inventories(self, checker, bar):
1197
"""Helper for _check_inventories."""
1199
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1200
kinds = ['chk_bytes', 'texts']
1201
count = len(checker.pending_keys)
1202
bar.update("inventories", 0, 2)
1203
current_keys = checker.pending_keys
1204
checker.pending_keys = {}
1205
# Accumulate current checks.
1206
for key in current_keys:
1207
if key[0] != 'inventories' and key[0] not in kinds:
1208
checker._report_items.append('unknown key type %r' % (key,))
1209
keys[key[0]].add(key[1:])
1210
if keys['inventories']:
1211
# NB: output order *should* be roughly sorted - topo or
1212
# inverse topo depending on repository - either way decent
1213
# to just delta against. However, pre-CHK formats didn't
1214
# try to optimise inventory layout on disk. As such the
1215
# pre-CHK code path does not use inventory deltas.
1217
for record in self.inventories.check(keys=keys['inventories']):
1218
if record.storage_kind == 'absent':
1219
checker._report_items.append(
1220
'Missing inventory {%s}' % (record.key,))
1222
last_object = self._check_record('inventories', record,
1223
checker, last_object,
1224
current_keys[('inventories',) + record.key])
1225
del keys['inventories']
1228
bar.update("texts", 1)
1229
while (checker.pending_keys or keys['chk_bytes']
1231
# Something to check.
1232
current_keys = checker.pending_keys
1233
checker.pending_keys = {}
1234
# Accumulate current checks.
1235
for key in current_keys:
1236
if key[0] not in kinds:
1237
checker._report_items.append('unknown key type %r' % (key,))
1238
keys[key[0]].add(key[1:])
1239
# Check the outermost kind only - inventories || chk_bytes || texts
1243
for record in getattr(self, kind).check(keys=keys[kind]):
1244
if record.storage_kind == 'absent':
1245
checker._report_items.append(
1246
'Missing %s {%s}' % (kind, record.key,))
1248
last_object = self._check_record(kind, record,
1249
checker, last_object, current_keys[(kind,) + record.key])
1253
def _check_record(self, kind, record, checker, last_object, item_data):
1254
"""Check a single text from this repository."""
1255
if kind == 'inventories':
1256
rev_id = record.key[0]
1257
inv = self._deserialise_inventory(rev_id,
1258
record.get_bytes_as('fulltext'))
1259
if last_object is not None:
1260
delta = inv._make_delta(last_object)
1261
for old_path, path, file_id, ie in delta:
1264
ie.check(checker, rev_id, inv)
1266
for path, ie in inv.iter_entries():
1267
ie.check(checker, rev_id, inv)
1268
if self._format.fast_deltas:
1270
elif kind == 'chk_bytes':
1271
# No code written to check chk_bytes for this repo format.
1272
checker._report_items.append(
1273
'unsupported key type chk_bytes for %s' % (record.key,))
1274
elif kind == 'texts':
1275
self._check_text(record, checker, item_data)
1277
checker._report_items.append(
1278
'unknown key type %s for %s' % (kind, record.key))
1280
def _check_text(self, record, checker, item_data):
1281
"""Check a single text."""
1282
# Check it is extractable.
1283
# TODO: check length.
1284
if record.storage_kind == 'chunked':
1285
chunks = record.get_bytes_as(record.storage_kind)
1286
sha1 = osutils.sha_strings(chunks)
1287
length = sum(map(len, chunks))
1289
content = record.get_bytes_as('fulltext')
1290
sha1 = osutils.sha_string(content)
1291
length = len(content)
1292
if item_data and sha1 != item_data[1]:
1293
checker._report_items.append(
1294
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1295
(record.key, sha1, item_data[1], item_data[2]))
1298
801
def create(a_bzrdir):
1299
802
"""Construct the current default format repository in a_bzrdir."""
1574
1076
"""Commit the contents accrued within the current write group.
1576
1078
:seealso: start_write_group.
1578
:return: it may return an opaque hint that can be passed to 'pack'.
1580
1080
if self._write_group is not self.get_transaction():
1581
1081
# has an unlock or relock occured ?
1582
1082
raise errors.BzrError('mismatched lock context %r and '
1583
1083
'write group %r.' %
1584
1084
(self.get_transaction(), self._write_group))
1585
result = self._commit_write_group()
1085
self._commit_write_group()
1586
1086
self._write_group = None
1589
1088
def _commit_write_group(self):
1590
1089
"""Template method for per-repository write group cleanup.
1592
This is called before the write group is considered to be
1091
This is called before the write group is considered to be
1593
1092
finished and should ensure that all data handed to the repository
1594
for writing during the write group is safely committed (to the
1093
for writing during the write group is safely committed (to the
1595
1094
extent possible considering file system caching etc).
1598
def suspend_write_group(self):
1599
raise errors.UnsuspendableWriteGroup(self)
1601
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1602
"""Return the keys of missing inventory parents for revisions added in
1605
A revision is not complete if the inventory delta for that revision
1606
cannot be calculated. Therefore if the parent inventories of a
1607
revision are not present, the revision is incomplete, and e.g. cannot
1608
be streamed by a smart server. This method finds missing inventory
1609
parents for revisions added in this write group.
1611
if not self._format.supports_external_lookups:
1612
# This is only an issue for stacked repositories
1614
if not self.is_in_write_group():
1615
raise AssertionError('not in a write group')
1617
# XXX: We assume that every added revision already has its
1618
# corresponding inventory, so we only check for parent inventories that
1619
# might be missing, rather than all inventories.
1620
parents = set(self.revisions._index.get_missing_parents())
1621
parents.discard(_mod_revision.NULL_REVISION)
1622
unstacked_inventories = self.inventories._index
1623
present_inventories = unstacked_inventories.get_parent_map(
1624
key[-1:] for key in parents)
1625
parents.difference_update(present_inventories)
1626
if len(parents) == 0:
1627
# No missing parent inventories.
1629
if not check_for_missing_texts:
1630
return set(('inventories', rev_id) for (rev_id,) in parents)
1631
# Ok, now we have a list of missing inventories. But these only matter
1632
# if the inventories that reference them are missing some texts they
1633
# appear to introduce.
1634
# XXX: Texts referenced by all added inventories need to be present,
1635
# but at the moment we're only checking for texts referenced by
1636
# inventories at the graph's edge.
1637
key_deps = self.revisions._index._key_dependencies
1638
key_deps.satisfy_refs_for_keys(present_inventories)
1639
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1640
file_ids = self.fileids_altered_by_revision_ids(referrers)
1641
missing_texts = set()
1642
for file_id, version_ids in file_ids.iteritems():
1643
missing_texts.update(
1644
(file_id, version_id) for version_id in version_ids)
1645
present_texts = self.texts.get_parent_map(missing_texts)
1646
missing_texts.difference_update(present_texts)
1647
if not missing_texts:
1648
# No texts are missing, so all revisions and their deltas are
1651
# Alternatively the text versions could be returned as the missing
1652
# keys, but this is likely to be less data.
1653
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1656
def refresh_data(self):
1657
"""Re-read any data needed to to synchronise with disk.
1659
This method is intended to be called after another repository instance
1660
(such as one used by a smart server) has inserted data into the
1661
repository. It may not be called during a write group, but may be
1662
called at any other time.
1664
if self.is_in_write_group():
1665
raise errors.InternalBzrError(
1666
"May not refresh_data while in a write group.")
1667
self._refresh_data()
1669
def resume_write_group(self, tokens):
1670
if not self.is_write_locked():
1671
raise errors.NotWriteLocked(self)
1672
if self._write_group:
1673
raise errors.BzrError('already in a write group')
1674
self._resume_write_group(tokens)
1675
# so we can detect unlock/relock - the write group is now entered.
1676
self._write_group = self.get_transaction()
1678
def _resume_write_group(self, tokens):
1679
raise errors.UnsuspendableWriteGroup(self)
1681
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1097
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False):
1683
1098
"""Fetch the content required to construct revision_id from source.
1685
If revision_id is None and fetch_spec is None, then all content is
1688
fetch() may not be used when the repository is in a write group -
1689
either finish the current write group before using fetch, or use
1690
fetch before starting the write group.
1100
If revision_id is None all content is copied.
1692
1101
:param find_ghosts: Find and copy revisions in the source that are
1693
1102
ghosts in the target (and not reachable directly by walking out to
1694
1103
the first-present revision in target from revision_id).
1695
:param revision_id: If specified, all the content needed for this
1696
revision ID will be copied to the target. Fetch will determine for
1697
itself which content needs to be copied.
1698
:param fetch_spec: If specified, a SearchResult or
1699
PendingAncestryResult that describes which revisions to copy. This
1700
allows copying multiple heads at once. Mutually exclusive with
1703
if fetch_spec is not None and revision_id is not None:
1704
raise AssertionError(
1705
"fetch_spec and revision_id are mutually exclusive.")
1706
if self.is_in_write_group():
1707
raise errors.InternalBzrError(
1708
"May not fetch while in a write group.")
1709
1105
# fast path same-url fetch operations
1710
# TODO: lift out to somewhere common with RemoteRepository
1711
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1712
if (self.has_same_location(source)
1713
and fetch_spec is None
1714
and self._has_same_fallbacks(source)):
1106
if self.has_same_location(source):
1715
1107
# check that last_revision is in 'from' and then return a
1716
1108
# no-operation.
1717
1109
if (revision_id is not None and
1880
1259
@needs_read_lock
1881
1260
def get_revisions(self, revision_ids):
1882
"""Get many revisions at once.
1884
Repositories that need to check data on every revision read should
1885
subclass this method.
1261
"""Get many revisions at once."""
1887
1262
return self._get_revisions(revision_ids)
1889
1264
@needs_read_lock
1890
1265
def _get_revisions(self, revision_ids):
1891
1266
"""Core work logic to get many revisions without sanity checks."""
1267
for rev_id in revision_ids:
1268
if not rev_id or not isinstance(rev_id, basestring):
1269
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1270
keys = [(key,) for key in revision_ids]
1271
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1893
for revid, rev in self._iter_revisions(revision_ids):
1895
raise errors.NoSuchRevision(self, revid)
1273
for record in stream:
1274
if record.storage_kind == 'absent':
1275
raise errors.NoSuchRevision(self, record.key[0])
1276
text = record.get_bytes_as('fulltext')
1277
rev = self._serializer.read_revision_from_string(text)
1278
revs[record.key[0]] = rev
1897
1279
return [revs[revid] for revid in revision_ids]
1899
def _iter_revisions(self, revision_ids):
1900
"""Iterate over revision objects.
1902
:param revision_ids: An iterable of revisions to examine. None may be
1903
passed to request all revisions known to the repository. Note that
1904
not all repositories can find unreferenced revisions; for those
1905
repositories only referenced ones will be returned.
1906
:return: An iterator of (revid, revision) tuples. Absent revisions (
1907
those asked for but not available) are returned as (revid, None).
1909
if revision_ids is None:
1910
revision_ids = self.all_revision_ids()
1912
for rev_id in revision_ids:
1913
if not rev_id or not isinstance(rev_id, basestring):
1914
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1915
keys = [(key,) for key in revision_ids]
1916
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1917
for record in stream:
1918
revid = record.key[0]
1919
if record.storage_kind == 'absent':
1922
text = record.get_bytes_as('fulltext')
1923
rev = self._serializer.read_revision_from_string(text)
1926
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1282
def get_revision_xml(self, revision_id):
1283
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1284
# would have already do it.
1285
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1286
rev = self.get_revision(revision_id)
1287
rev_tmp = cStringIO.StringIO()
1288
# the current serializer..
1289
self._serializer.write_revision(rev, rev_tmp)
1291
return rev_tmp.getvalue()
1293
def get_deltas_for_revisions(self, revisions):
1927
1294
"""Produce a generator of revision deltas.
1929
1296
Note that the input is a sequence of REVISIONS, not revision_ids.
1930
1297
Trees will be held in memory until the generator exits.
1931
1298
Each delta is relative to the revision's lefthand predecessor.
1933
:param specific_fileids: if not None, the result is filtered
1934
so that only those file-ids, their parents and their
1935
children are included.
1937
# Get the revision-ids of interest
1938
1300
required_trees = set()
1939
1301
for revision in revisions:
1940
1302
required_trees.add(revision.revision_id)
1941
1303
required_trees.update(revision.parent_ids[:1])
1943
# Get the matching filtered trees. Note that it's more
1944
# efficient to pass filtered trees to changes_from() rather
1945
# than doing the filtering afterwards. changes_from() could
1946
# arguably do the filtering itself but it's path-based, not
1947
# file-id based, so filtering before or afterwards is
1949
if specific_fileids is None:
1950
trees = dict((t.get_revision_id(), t) for
1951
t in self.revision_trees(required_trees))
1953
trees = dict((t.get_revision_id(), t) for
1954
t in self._filtered_revision_trees(required_trees,
1957
# Calculate the deltas
1304
trees = dict((t.get_revision_id(), t) for
1305
t in self.revision_trees(required_trees))
1958
1306
for revision in revisions:
1959
1307
if not revision.parent_ids:
1960
1308
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
2367
1662
"""Get Inventory object by revision id."""
2368
1663
return self.iter_inventories([revision_id]).next()
2370
def iter_inventories(self, revision_ids, ordering=None):
1665
def iter_inventories(self, revision_ids):
2371
1666
"""Get many inventories by revision_ids.
2373
1668
This will buffer some or all of the texts used in constructing the
2374
1669
inventories in memory, but will only parse a single inventory at a
2377
:param revision_ids: The expected revision ids of the inventories.
2378
:param ordering: optional ordering, e.g. 'topological'. If not
2379
specified, the order of revision_ids will be preserved (by
2380
buffering if necessary).
2381
1672
:return: An iterator of inventories.
2383
1674
if ((None in revision_ids)
2384
1675
or (_mod_revision.NULL_REVISION in revision_ids)):
2385
1676
raise ValueError('cannot get null revision inventory')
2386
return self._iter_inventories(revision_ids, ordering)
1677
return self._iter_inventories(revision_ids)
2388
def _iter_inventories(self, revision_ids, ordering):
1679
def _iter_inventories(self, revision_ids):
2389
1680
"""single-document based inventory iteration."""
2390
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2391
for text, revision_id in inv_xmls:
2392
yield self._deserialise_inventory(revision_id, text)
1681
for text, revision_id in self._iter_inventory_xmls(revision_ids):
1682
yield self.deserialise_inventory(revision_id, text)
2394
def _iter_inventory_xmls(self, revision_ids, ordering):
2395
if ordering is None:
2396
order_as_requested = True
2397
ordering = 'unordered'
2399
order_as_requested = False
1684
def _iter_inventory_xmls(self, revision_ids):
2400
1685
keys = [(revision_id,) for revision_id in revision_ids]
2403
if order_as_requested:
2404
key_iter = iter(keys)
2405
next_key = key_iter.next()
2406
stream = self.inventories.get_record_stream(keys, ordering, True)
1686
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2407
1687
text_chunks = {}
2408
1688
for record in stream:
2409
1689
if record.storage_kind != 'absent':
2410
chunks = record.get_bytes_as('chunked')
2411
if order_as_requested:
2412
text_chunks[record.key] = chunks
2414
yield ''.join(chunks), record.key[-1]
1690
text_chunks[record.key] = record.get_bytes_as('chunked')
2416
1692
raise errors.NoSuchRevision(self, record.key)
2417
if order_as_requested:
2418
# Yield as many results as we can while preserving order.
2419
while next_key in text_chunks:
2420
chunks = text_chunks.pop(next_key)
2421
yield ''.join(chunks), next_key[-1]
2423
next_key = key_iter.next()
2424
except StopIteration:
2425
# We still want to fully consume the get_record_stream,
2426
# just in case it is not actually finished at this point
1694
chunks = text_chunks.pop(key)
1695
yield ''.join(chunks), key[-1]
2430
def _deserialise_inventory(self, revision_id, xml):
2431
"""Transform the xml into an inventory object.
1697
def deserialise_inventory(self, revision_id, xml):
1698
"""Transform the xml into an inventory object.
2433
1700
:param revision_id: The expected revision id of the inventory.
2434
1701
:param xml: A serialised inventory.
2436
1703
result = self._serializer.read_inventory_from_string(xml, revision_id,
2437
entry_cache=self._inventory_entry_cache,
2438
return_from_cache=self._safe_to_return_from_cache)
1704
entry_cache=self._inventory_entry_cache)
2439
1705
if result.revision_id != revision_id:
2440
1706
raise AssertionError('revision id mismatch %s != %s' % (
2441
1707
result.revision_id, revision_id))
1710
def serialise_inventory(self, inv):
1711
return self._serializer.write_inventory_to_string(inv)
1713
def _serialise_inventory_to_lines(self, inv):
1714
return self._serializer.write_inventory_to_lines(inv)
2444
1716
def get_serializer_format(self):
2445
1717
return self._serializer.format_num
2447
1719
@needs_read_lock
2448
def _get_inventory_xml(self, revision_id):
2449
"""Get serialized inventory as a string."""
2450
texts = self._iter_inventory_xmls([revision_id], 'unordered')
1720
def get_inventory_xml(self, revision_id):
1721
"""Get inventory XML as a file object."""
1722
texts = self._iter_inventory_xmls([revision_id])
2452
1724
text, revision_id = texts.next()
2453
1725
except StopIteration:
2454
1726
raise errors.HistoryMissing(self, 'inventory', revision_id)
2457
def get_rev_id_for_revno(self, revno, known_pair):
2458
"""Return the revision id of a revno, given a later (revno, revid)
2459
pair in the same history.
2461
:return: if found (True, revid). If the available history ran out
2462
before reaching the revno, then this returns
2463
(False, (closest_revno, closest_revid)).
1730
def get_inventory_sha1(self, revision_id):
1731
"""Return the sha1 hash of the inventory entry
2465
known_revno, known_revid = known_pair
2466
partial_history = [known_revid]
2467
distance_from_known = known_revno - revno
2468
if distance_from_known < 0:
2470
'requested revno (%d) is later than given known revno (%d)'
2471
% (revno, known_revno))
2474
self, partial_history, stop_index=distance_from_known)
2475
except errors.RevisionNotPresent, err:
2476
if err.revision_id == known_revid:
2477
# The start revision (known_revid) wasn't found.
2479
# This is a stacked repository with no fallbacks, or a there's a
2480
# left-hand ghost. Either way, even though the revision named in
2481
# the error isn't in this repo, we know it's the next step in this
2482
# left-hand history.
2483
partial_history.append(err.revision_id)
2484
if len(partial_history) <= distance_from_known:
2485
# Didn't find enough history to get a revid for the revno.
2486
earliest_revno = known_revno - len(partial_history) + 1
2487
return (False, (earliest_revno, partial_history[-1]))
2488
if len(partial_history) - 1 > distance_from_known:
2489
raise AssertionError('_iter_for_revno returned too much history')
2490
return (True, partial_history[-1])
1733
return self.get_revision(revision_id).inventory_sha1
2492
1735
def iter_reverse_revision_history(self, revision_id):
2493
1736
"""Iterate backwards through revision ids in the lefthand history
3726
2909
return self.source.revision_ids_to_search_result(result_set)
3729
class InterDifferingSerializer(InterRepository):
2912
class InterPackRepo(InterSameDataRepository):
2913
"""Optimised code paths between Pack based repositories."""
2916
def _get_repo_format_to_test(self):
2917
from bzrlib.repofmt import pack_repo
2918
return pack_repo.RepositoryFormatKnitPack1()
2921
def is_compatible(source, target):
2922
"""Be compatible with known Pack formats.
2924
We don't test for the stores being of specific types because that
2925
could lead to confusing results, and there is no need to be
2928
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
2930
are_packs = (isinstance(source._format, RepositoryFormatPack) and
2931
isinstance(target._format, RepositoryFormatPack))
2932
except AttributeError:
2934
return are_packs and InterRepository._same_model(source, target)
2937
def fetch(self, revision_id=None, pb=None, find_ghosts=False):
2938
"""See InterRepository.fetch()."""
2939
if (len(self.source._fallback_repositories) > 0 or
2940
len(self.target._fallback_repositories) > 0):
2941
# The pack layer is not aware of fallback repositories, so when
2942
# fetching from a stacked repository or into a stacked repository
2943
# we use the generic fetch logic which uses the VersionedFiles
2944
# attributes on repository.
2945
from bzrlib.fetch import RepoFetcher
2946
fetcher = RepoFetcher(self.target, self.source, revision_id,
2948
return fetcher.count_copied, fetcher.failed_revisions
2949
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
2950
self.source, self.source._format, self.target, self.target._format)
2951
self.count_copied = 0
2952
if revision_id is None:
2954
# everything to do - use pack logic
2955
# to fetch from all packs to one without
2956
# inventory parsing etc, IFF nothing to be copied is in the target.
2958
source_revision_ids = frozenset(self.source.all_revision_ids())
2959
revision_ids = source_revision_ids - \
2960
frozenset(self.target_get_parent_map(source_revision_ids))
2961
revision_keys = [(revid,) for revid in revision_ids]
2962
target_pack_collection = self._get_target_pack_collection()
2963
index = target_pack_collection.revision_index.combined_index
2964
present_revision_ids = set(item[1][0] for item in
2965
index.iter_entries(revision_keys))
2966
revision_ids = set(revision_ids) - present_revision_ids
2967
# implementing the TODO will involve:
2968
# - detecting when all of a pack is selected
2969
# - avoiding as much as possible pre-selection, so the
2970
# more-core routines such as create_pack_from_packs can filter in
2971
# a just-in-time fashion. (though having a HEADS list on a
2972
# repository might make this a lot easier, because we could
2973
# sensibly detect 'new revisions' without doing a full index scan.
2974
elif _mod_revision.is_null(revision_id):
2979
revision_ids = self.search_missing_revision_ids(revision_id,
2980
find_ghosts=find_ghosts).get_keys()
2981
except errors.NoSuchRevision:
2982
raise errors.InstallFailed([revision_id])
2983
if len(revision_ids) == 0:
2985
return self._pack(self.source, self.target, revision_ids)
2987
def _pack(self, source, target, revision_ids):
2988
from bzrlib.repofmt.pack_repo import Packer
2989
target_pack_collection = self._get_target_pack_collection()
2990
packs = source._pack_collection.all_packs()
2991
pack = Packer(target_pack_collection, packs, '.fetch',
2992
revision_ids).pack()
2993
if pack is not None:
2994
target_pack_collection._save_pack_names()
2995
copied_revs = pack.get_revision_count()
2996
# Trigger an autopack. This may duplicate effort as we've just done
2997
# a pack creation, but for now it is simpler to think about as
2998
# 'upload data, then repack if needed'.
3000
return (copied_revs, [])
3004
def _autopack(self):
3005
self.target._pack_collection.autopack()
3007
def _get_target_pack_collection(self):
3008
return self.target._pack_collection
3011
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3012
"""See InterRepository.missing_revision_ids().
3014
:param find_ghosts: Find ghosts throughout the ancestry of
3017
if not find_ghosts and revision_id is not None:
3018
return self._walk_to_common_revisions([revision_id])
3019
elif revision_id is not None:
3020
# Find ghosts: search for revisions pointing from one repository to
3021
# the other, and vice versa, anywhere in the history of revision_id.
3022
graph = self.target_get_graph(other_repository=self.source)
3023
searcher = graph._make_breadth_first_searcher([revision_id])
3027
next_revs, ghosts = searcher.next_with_ghosts()
3028
except StopIteration:
3030
if revision_id in ghosts:
3031
raise errors.NoSuchRevision(self.source, revision_id)
3032
found_ids.update(next_revs)
3033
found_ids.update(ghosts)
3034
found_ids = frozenset(found_ids)
3035
# Double query here: should be able to avoid this by changing the
3036
# graph api further.
3037
result_set = found_ids - frozenset(
3038
self.target_get_parent_map(found_ids))
3040
source_ids = self.source.all_revision_ids()
3041
# source_ids is the worst possible case we may need to pull.
3042
# now we want to filter source_ids against what we actually
3043
# have in target, but don't try to check for existence where we know
3044
# we do not have a revision as that would be pointless.
3045
target_ids = set(self.target.all_revision_ids())
3046
result_set = set(source_ids).difference(target_ids)
3047
return self.source.revision_ids_to_search_result(result_set)
3050
class InterModel1and2(InterRepository):
3053
def _get_repo_format_to_test(self):
3057
def is_compatible(source, target):
3058
if not source.supports_rich_root() and target.supports_rich_root():
3064
def fetch(self, revision_id=None, pb=None, find_ghosts=False):
3065
"""See InterRepository.fetch()."""
3066
from bzrlib.fetch import Model1toKnit2Fetcher
3067
f = Model1toKnit2Fetcher(to_repository=self.target,
3068
from_repository=self.source,
3069
last_revision=revision_id,
3070
pb=pb, find_ghosts=find_ghosts)
3071
return f.count_copied, f.failed_revisions
3074
def copy_content(self, revision_id=None):
3075
"""Make a complete copy of the content in self into destination.
3077
This is a destructive operation! Do not use it on existing
3080
:param revision_id: Only copy the content needed to construct
3081
revision_id and its parents.
3084
self.target.set_make_working_trees(self.source.make_working_trees())
3085
except NotImplementedError:
3087
# but don't bother fetching if we have the needed data now.
3088
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
3089
self.target.has_revision(revision_id)):
3091
self.target.fetch(self.source, revision_id=revision_id)
3094
class InterKnit1and2(InterKnitRepo):
3097
def _get_repo_format_to_test(self):
3101
def is_compatible(source, target):
3102
"""Be compatible with Knit1 source and Knit3 target"""
3104
from bzrlib.repofmt.knitrepo import (
3105
RepositoryFormatKnit1,
3106
RepositoryFormatKnit3,
3108
from bzrlib.repofmt.pack_repo import (
3109
RepositoryFormatKnitPack1,
3110
RepositoryFormatKnitPack3,
3111
RepositoryFormatKnitPack4,
3112
RepositoryFormatKnitPack5,
3113
RepositoryFormatKnitPack5RichRoot,
3114
RepositoryFormatKnitPack6,
3115
RepositoryFormatKnitPack6RichRoot,
3116
RepositoryFormatPackDevelopment2,
3117
RepositoryFormatPackDevelopment2Subtree,
3120
RepositoryFormatKnit1, # no rr, no subtree
3121
RepositoryFormatKnitPack1, # no rr, no subtree
3122
RepositoryFormatPackDevelopment2, # no rr, no subtree
3123
RepositoryFormatKnitPack5, # no rr, no subtree
3124
RepositoryFormatKnitPack6, # no rr, no subtree
3127
RepositoryFormatKnit3, # rr, subtree
3128
RepositoryFormatKnitPack3, # rr, subtree
3129
RepositoryFormatKnitPack4, # rr, no subtree
3130
RepositoryFormatKnitPack5RichRoot,# rr, no subtree
3131
RepositoryFormatKnitPack6RichRoot,# rr, no subtree
3132
RepositoryFormatPackDevelopment2Subtree, # rr, subtree
3134
for format in norichroot:
3135
if format.rich_root_data:
3136
raise AssertionError('Format %s is a rich-root format'
3137
' but is included in the non-rich-root list'
3139
for format in richroot:
3140
if not format.rich_root_data:
3141
raise AssertionError('Format %s is not a rich-root format'
3142
' but is included in the rich-root list'
3144
# TODO: One alternative is to just check format.rich_root_data,
3145
# instead of keeping membership lists. However, the formats
3146
# *also* have to use the same 'Knit' style of storage
3147
# (line-deltas, fulltexts, etc.)
3148
return (isinstance(source._format, norichroot) and
3149
isinstance(target._format, richroot))
3150
except AttributeError:
3154
def fetch(self, revision_id=None, pb=None, find_ghosts=False):
3155
"""See InterRepository.fetch()."""
3156
from bzrlib.fetch import Knit1to2Fetcher
3157
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
3158
self.source, self.source._format, self.target,
3159
self.target._format)
3160
f = Knit1to2Fetcher(to_repository=self.target,
3161
from_repository=self.source,
3162
last_revision=revision_id,
3163
pb=pb, find_ghosts=find_ghosts)
3164
return f.count_copied, f.failed_revisions
3167
class InterDifferingSerializer(InterKnitRepo):
3732
3170
def _get_repo_format_to_test(self):
3736
3174
def is_compatible(source, target):
3737
3175
"""Be compatible with Knit2 source and Knit3 target"""
3738
# This is redundant with format.check_conversion_target(), however that
3739
# raises an exception, and we just want to say "False" as in we won't
3740
# support converting between these formats.
3741
if 'IDS_never' in debug.debug_flags:
3743
if source.supports_rich_root() and not target.supports_rich_root():
3745
if (source._format.supports_tree_reference
3746
and not target._format.supports_tree_reference):
3748
if target._fallback_repositories and target._format.supports_chks:
3749
# IDS doesn't know how to copy CHKs for the parent inventories it
3750
# adds to stacked repos.
3752
if 'IDS_always' in debug.debug_flags:
3754
# Only use this code path for local source and target. IDS does far
3755
# too much IO (both bandwidth and roundtrips) over a network.
3756
if not source.bzrdir.transport.base.startswith('file:///'):
3758
if not target.bzrdir.transport.base.startswith('file:///'):
3176
if source.supports_rich_root() != target.supports_rich_root():
3178
# Ideally, we'd support fetching if the source had no tree references
3179
# even if it supported them...
3180
if (getattr(source, '_format.supports_tree_reference', False) and
3181
not getattr(target, '_format.supports_tree_reference', False)):
3762
def _get_trees(self, revision_ids, cache):
3764
for rev_id in revision_ids:
3766
possible_trees.append((rev_id, cache[rev_id]))
3768
# Not cached, but inventory might be present anyway.
3770
tree = self.source.revision_tree(rev_id)
3771
except errors.NoSuchRevision:
3772
# Nope, parent is ghost.
3775
cache[rev_id] = tree
3776
possible_trees.append((rev_id, tree))
3777
return possible_trees
3779
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3780
"""Get the best delta and base for this revision.
3782
:return: (basis_id, delta)
3785
# Generate deltas against each tree, to find the shortest.
3786
texts_possibly_new_in_tree = set()
3787
for basis_id, basis_tree in possible_trees:
3788
delta = tree.inventory._make_delta(basis_tree.inventory)
3789
for old_path, new_path, file_id, new_entry in delta:
3790
if new_path is None:
3791
# This file_id isn't present in the new rev, so we don't
3795
# Rich roots are handled elsewhere...
3797
kind = new_entry.kind
3798
if kind != 'directory' and kind != 'file':
3799
# No text record associated with this inventory entry.
3801
# This is a directory or file that has changed somehow.
3802
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3803
deltas.append((len(delta), basis_id, delta))
3805
return deltas[0][1:]
3807
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3808
"""Find all parent revisions that are absent, but for which the
3809
inventory is present, and copy those inventories.
3811
This is necessary to preserve correctness when the source is stacked
3812
without fallbacks configured. (Note that in cases like upgrade the
3813
source may be not have _fallback_repositories even though it is
3817
for parents in parent_map.values():
3818
parent_revs.update(parents)
3819
present_parents = self.source.get_parent_map(parent_revs)
3820
absent_parents = set(parent_revs).difference(present_parents)
3821
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3822
(rev_id,) for rev_id in absent_parents)
3823
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3824
for parent_tree in self.source.revision_trees(parent_inv_ids):
3825
current_revision_id = parent_tree.get_revision_id()
3826
parents_parents_keys = parent_invs_keys_for_stacking[
3827
(current_revision_id,)]
3828
parents_parents = [key[-1] for key in parents_parents_keys]
3829
basis_id = _mod_revision.NULL_REVISION
3830
basis_tree = self.source.revision_tree(basis_id)
3831
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3832
self.target.add_inventory_by_delta(
3833
basis_id, delta, current_revision_id, parents_parents)
3834
cache[current_revision_id] = parent_tree
3836
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3185
def _fetch_batch(self, revision_ids, basis_id, basis_tree):
3837
3186
"""Fetch across a few revisions.
3839
3188
:param revision_ids: The revisions to copy
3840
:param basis_id: The revision_id of a tree that must be in cache, used
3841
as a basis for delta when no other base is available
3842
:param cache: A cache of RevisionTrees that we can use.
3843
:param a_graph: A Graph object to determine the heads() of the
3844
rich-root data stream.
3845
:return: The revision_id of the last converted tree. The RevisionTree
3846
for it will be in cache
3189
:param basis_id: The revision_id of basis_tree
3190
:param basis_tree: A tree that is not in revision_ids which should
3191
already exist in the target.
3192
:return: (basis_id, basis_tree) A new basis to use now that these trees
3848
3195
# Walk though all revisions; get inventory deltas, copy referenced
3849
3196
# texts that delta references, insert the delta, revision and
3851
root_keys_to_create = set()
3852
3198
text_keys = set()
3853
3199
pending_deltas = []
3854
3200
pending_revisions = []
3855
parent_map = self.source.get_parent_map(revision_ids)
3856
self._fetch_parent_invs_for_stacking(parent_map, cache)
3857
self.source._safe_to_return_from_cache = True
3858
3201
for tree in self.source.revision_trees(revision_ids):
3859
# Find a inventory delta for this revision.
3860
# Find text entries that need to be copied, too.
3861
3202
current_revision_id = tree.get_revision_id()
3862
parent_ids = parent_map.get(current_revision_id, ())
3863
parent_trees = self._get_trees(parent_ids, cache)
3864
possible_trees = list(parent_trees)
3865
if len(possible_trees) == 0:
3866
# There either aren't any parents, or the parents are ghosts,
3867
# so just use the last converted tree.
3868
possible_trees.append((basis_id, cache[basis_id]))
3869
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3203
delta = tree.inventory._make_delta(basis_tree.inventory)
3204
for old_path, new_path, file_id, entry in delta:
3205
if new_path is not None:
3206
if not (new_path or self.target.supports_rich_root()):
3207
# We leave the inventory delta in, because that
3208
# will have the deserialised inventory root
3212
# "if entry.revision == current_revision_id" ?
3213
if entry.revision == current_revision_id:
3214
text_keys.add((file_id, entry.revision))
3871
3215
revision = self.source.get_revision(current_revision_id)
3872
3216
pending_deltas.append((basis_id, delta,
3873
3217
current_revision_id, revision.parent_ids))
3874
if self._converting_to_rich_root:
3875
self._revision_id_to_root_id[current_revision_id] = \
3877
# Determine which texts are in present in this revision but not in
3878
# any of the available parents.
3879
texts_possibly_new_in_tree = set()
3880
for old_path, new_path, file_id, entry in delta:
3881
if new_path is None:
3882
# This file_id isn't present in the new rev
3886
if not self.target.supports_rich_root():
3887
# The target doesn't support rich root, so we don't
3890
if self._converting_to_rich_root:
3891
# This can't be copied normally, we have to insert
3893
root_keys_to_create.add((file_id, entry.revision))
3896
texts_possibly_new_in_tree.add((file_id, entry.revision))
3897
for basis_id, basis_tree in possible_trees:
3898
basis_inv = basis_tree.inventory
3899
for file_key in list(texts_possibly_new_in_tree):
3900
file_id, file_revision = file_key
3902
entry = basis_inv[file_id]
3903
except errors.NoSuchId:
3905
if entry.revision == file_revision:
3906
texts_possibly_new_in_tree.remove(file_key)
3907
text_keys.update(texts_possibly_new_in_tree)
3908
3218
pending_revisions.append(revision)
3909
cache[current_revision_id] = tree
3910
3219
basis_id = current_revision_id
3911
self.source._safe_to_return_from_cache = False
3912
3221
# Copy file texts
3913
3222
from_texts = self.source.texts
3914
3223
to_texts = self.target.texts
3915
if root_keys_to_create:
3916
root_stream = _mod_fetch._new_root_data_stream(
3917
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3918
self.source, graph=a_graph)
3919
to_texts.insert_record_stream(root_stream)
3920
3224
to_texts.insert_record_stream(from_texts.get_record_stream(
3921
text_keys, self.target._format._fetch_order,
3922
not self.target._format._fetch_uses_deltas))
3923
# insert inventory deltas
3225
text_keys, self.target._fetch_order,
3226
not self.target._fetch_uses_deltas))
3924
3228
for delta in pending_deltas:
3925
3229
self.target.add_inventory_by_delta(*delta)
3926
if self.target._fallback_repositories:
3927
# Make sure this stacked repository has all the parent inventories
3928
# for the new revisions that we are about to insert. We do this
3929
# before adding the revisions so that no revision is added until
3930
# all the inventories it may depend on are added.
3931
# Note that this is overzealous, as we may have fetched these in an
3934
revision_ids = set()
3935
for revision in pending_revisions:
3936
revision_ids.add(revision.revision_id)
3937
parent_ids.update(revision.parent_ids)
3938
parent_ids.difference_update(revision_ids)
3939
parent_ids.discard(_mod_revision.NULL_REVISION)
3940
parent_map = self.source.get_parent_map(parent_ids)
3941
# we iterate over parent_map and not parent_ids because we don't
3942
# want to try copying any revision which is a ghost
3943
for parent_tree in self.source.revision_trees(parent_map):
3944
current_revision_id = parent_tree.get_revision_id()
3945
parents_parents = parent_map[current_revision_id]
3946
possible_trees = self._get_trees(parents_parents, cache)
3947
if len(possible_trees) == 0:
3948
# There either aren't any parents, or the parents are
3949
# ghosts, so just use the last converted tree.
3950
possible_trees.append((basis_id, cache[basis_id]))
3951
basis_id, delta = self._get_delta_for_revision(parent_tree,
3952
parents_parents, possible_trees)
3953
self.target.add_inventory_by_delta(
3954
basis_id, delta, current_revision_id, parents_parents)
3955
3230
# insert signatures and revisions
3956
3231
for revision in pending_revisions:
4072
3309
return basis_id, basis_tree
3312
class InterOtherToRemote(InterRepository):
3313
"""An InterRepository that simply delegates to the 'real' InterRepository
3314
calculated for (source, target._real_repository).
3317
_walk_to_common_revisions_batch_size = 50
3319
def __init__(self, source, target):
3320
InterRepository.__init__(self, source, target)
3321
self._real_inter = None
3324
def is_compatible(source, target):
3325
if isinstance(target, remote.RemoteRepository):
3329
def _ensure_real_inter(self):
3330
if self._real_inter is None:
3331
self.target._ensure_real()
3332
real_target = self.target._real_repository
3333
self._real_inter = InterRepository.get(self.source, real_target)
3334
# Make _real_inter use the RemoteRepository for get_parent_map
3335
self._real_inter.target_get_graph = self.target.get_graph
3336
self._real_inter.target_get_parent_map = self.target.get_parent_map
3338
def copy_content(self, revision_id=None):
3339
self._ensure_real_inter()
3340
self._real_inter.copy_content(revision_id=revision_id)
3342
def fetch(self, revision_id=None, pb=None, find_ghosts=False):
3343
self._ensure_real_inter()
3344
return self._real_inter.fetch(revision_id=revision_id, pb=pb,
3345
find_ghosts=find_ghosts)
3348
def _get_repo_format_to_test(self):
3352
class InterRemoteToOther(InterRepository):
3354
def __init__(self, source, target):
3355
InterRepository.__init__(self, source, target)
3356
self._real_inter = None
3359
def is_compatible(source, target):
3360
if not isinstance(source, remote.RemoteRepository):
3362
# Is source's model compatible with target's model?
3363
source._ensure_real()
3364
real_source = source._real_repository
3365
if isinstance(real_source, remote.RemoteRepository):
3366
raise NotImplementedError(
3367
"We don't support remote repos backed by remote repos yet.")
3368
return InterRepository._same_model(real_source, target)
3370
def _ensure_real_inter(self):
3371
if self._real_inter is None:
3372
self.source._ensure_real()
3373
real_source = self.source._real_repository
3374
self._real_inter = InterRepository.get(real_source, self.target)
3376
def fetch(self, revision_id=None, pb=None, find_ghosts=False):
3377
self._ensure_real_inter()
3378
return self._real_inter.fetch(revision_id=revision_id, pb=pb,
3379
find_ghosts=find_ghosts)
3381
def copy_content(self, revision_id=None):
3382
self._ensure_real_inter()
3383
self._real_inter.copy_content(revision_id=revision_id)
3386
def _get_repo_format_to_test(self):
3391
class InterPackToRemotePack(InterPackRepo):
3392
"""A specialisation of InterPackRepo for a target that is a
3395
This will use the get_parent_map RPC rather than plain readvs, and also
3396
uses an RPC for autopacking.
3399
_walk_to_common_revisions_batch_size = 50
3402
def is_compatible(source, target):
3403
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3404
if isinstance(source._format, RepositoryFormatPack):
3405
if isinstance(target, remote.RemoteRepository):
3406
target._ensure_real()
3407
if isinstance(target._real_repository._format,
3408
RepositoryFormatPack):
3409
if InterRepository._same_model(source, target):
3413
def _autopack(self):
3414
self.target.autopack()
3416
def _get_target_pack_collection(self):
3417
return self.target._real_repository._pack_collection
3420
def _get_repo_format_to_test(self):
4075
3424
InterRepository.register_optimiser(InterDifferingSerializer)
4076
3425
InterRepository.register_optimiser(InterSameDataRepository)
4077
3426
InterRepository.register_optimiser(InterWeaveRepo)
4078
3427
InterRepository.register_optimiser(InterKnitRepo)
3428
InterRepository.register_optimiser(InterModel1and2)
3429
InterRepository.register_optimiser(InterKnit1and2)
3430
InterRepository.register_optimiser(InterPackRepo)
3431
InterRepository.register_optimiser(InterOtherToRemote)
3432
InterRepository.register_optimiser(InterRemoteToOther)
3433
InterRepository.register_optimiser(InterPackToRemotePack)
4081
3436
class CopyConverter(object):
4082
3437
"""A repository conversion tool which just performs a copy of the content.
4084
3439
This is slow but quite reliable.
4237
3587
revision_graph[key] = tuple(parent for parent in parents if parent
4238
3588
in revision_graph)
4239
3589
return revision_graph
4242
class StreamSink(object):
4243
"""An object that can insert a stream into a repository.
4245
This interface handles the complexity of reserialising inventories and
4246
revisions from different formats, and allows unidirectional insertion into
4247
stacked repositories without looking for the missing basis parents
4251
def __init__(self, target_repo):
4252
self.target_repo = target_repo
4254
def insert_stream(self, stream, src_format, resume_tokens):
4255
"""Insert a stream's content into the target repository.
4257
:param src_format: a bzr repository format.
4259
:return: a list of resume tokens and an iterable of keys additional
4260
items required before the insertion can be completed.
4262
self.target_repo.lock_write()
4265
self.target_repo.resume_write_group(resume_tokens)
4268
self.target_repo.start_write_group()
4271
# locked_insert_stream performs a commit|suspend.
4272
return self._locked_insert_stream(stream, src_format, is_resume)
4274
self.target_repo.abort_write_group(suppress_errors=True)
4277
self.target_repo.unlock()
4279
def _locked_insert_stream(self, stream, src_format, is_resume):
4280
to_serializer = self.target_repo._format._serializer
4281
src_serializer = src_format._serializer
4283
if to_serializer == src_serializer:
4284
# If serializers match and the target is a pack repository, set the
4285
# write cache size on the new pack. This avoids poor performance
4286
# on transports where append is unbuffered (such as
4287
# RemoteTransport). This is safe to do because nothing should read
4288
# back from the target repository while a stream with matching
4289
# serialization is being inserted.
4290
# The exception is that a delta record from the source that should
4291
# be a fulltext may need to be expanded by the target (see
4292
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4293
# explicitly flush any buffered writes first in that rare case.
4295
new_pack = self.target_repo._pack_collection._new_pack
4296
except AttributeError:
4297
# Not a pack repository
4300
new_pack.set_write_cache_size(1024*1024)
4301
for substream_type, substream in stream:
4302
if 'stream' in debug.debug_flags:
4303
mutter('inserting substream: %s', substream_type)
4304
if substream_type == 'texts':
4305
self.target_repo.texts.insert_record_stream(substream)
4306
elif substream_type == 'inventories':
4307
if src_serializer == to_serializer:
4308
self.target_repo.inventories.insert_record_stream(
4311
self._extract_and_insert_inventories(
4312
substream, src_serializer)
4313
elif substream_type == 'inventory-deltas':
4314
self._extract_and_insert_inventory_deltas(
4315
substream, src_serializer)
4316
elif substream_type == 'chk_bytes':
4317
# XXX: This doesn't support conversions, as it assumes the
4318
# conversion was done in the fetch code.
4319
self.target_repo.chk_bytes.insert_record_stream(substream)
4320
elif substream_type == 'revisions':
4321
# This may fallback to extract-and-insert more often than
4322
# required if the serializers are different only in terms of
4324
if src_serializer == to_serializer:
4325
self.target_repo.revisions.insert_record_stream(
4328
self._extract_and_insert_revisions(substream,
4330
elif substream_type == 'signatures':
4331
self.target_repo.signatures.insert_record_stream(substream)
4333
raise AssertionError('kaboom! %s' % (substream_type,))
4334
# Done inserting data, and the missing_keys calculations will try to
4335
# read back from the inserted data, so flush the writes to the new pack
4336
# (if this is pack format).
4337
if new_pack is not None:
4338
new_pack._write_data('', flush=True)
4339
# Find all the new revisions (including ones from resume_tokens)
4340
missing_keys = self.target_repo.get_missing_parent_inventories(
4341
check_for_missing_texts=is_resume)
4343
for prefix, versioned_file in (
4344
('texts', self.target_repo.texts),
4345
('inventories', self.target_repo.inventories),
4346
('revisions', self.target_repo.revisions),
4347
('signatures', self.target_repo.signatures),
4348
('chk_bytes', self.target_repo.chk_bytes),
4350
if versioned_file is None:
4352
# TODO: key is often going to be a StaticTuple object
4353
# I don't believe we can define a method by which
4354
# (prefix,) + StaticTuple will work, though we could
4355
# define a StaticTuple.sq_concat that would allow you to
4356
# pass in either a tuple or a StaticTuple as the second
4357
# object, so instead we could have:
4358
# StaticTuple(prefix) + key here...
4359
missing_keys.update((prefix,) + key for key in
4360
versioned_file.get_missing_compression_parent_keys())
4361
except NotImplementedError:
4362
# cannot even attempt suspending, and missing would have failed
4363
# during stream insertion.
4364
missing_keys = set()
4367
# suspend the write group and tell the caller what we is
4368
# missing. We know we can suspend or else we would not have
4369
# entered this code path. (All repositories that can handle
4370
# missing keys can handle suspending a write group).
4371
write_group_tokens = self.target_repo.suspend_write_group()
4372
return write_group_tokens, missing_keys
4373
hint = self.target_repo.commit_write_group()
4374
if (to_serializer != src_serializer and
4375
self.target_repo._format.pack_compresses):
4376
self.target_repo.pack(hint=hint)
4379
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4380
target_rich_root = self.target_repo._format.rich_root_data
4381
target_tree_refs = self.target_repo._format.supports_tree_reference
4382
for record in substream:
4383
# Insert the delta directly
4384
inventory_delta_bytes = record.get_bytes_as('fulltext')
4385
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4387
parse_result = deserialiser.parse_text_bytes(
4388
inventory_delta_bytes)
4389
except inventory_delta.IncompatibleInventoryDelta, err:
4390
trace.mutter("Incompatible delta: %s", err.msg)
4391
raise errors.IncompatibleRevision(self.target_repo._format)
4392
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4393
revision_id = new_id
4394
parents = [key[0] for key in record.parents]
4395
self.target_repo.add_inventory_by_delta(
4396
basis_id, inv_delta, revision_id, parents)
4398
def _extract_and_insert_inventories(self, substream, serializer,
4400
"""Generate a new inventory versionedfile in target, converting data.
4402
The inventory is retrieved from the source, (deserializing it), and
4403
stored in the target (reserializing it in a different format).
4405
target_rich_root = self.target_repo._format.rich_root_data
4406
target_tree_refs = self.target_repo._format.supports_tree_reference
4407
for record in substream:
4408
# It's not a delta, so it must be a fulltext in the source
4409
# serializer's format.
4410
bytes = record.get_bytes_as('fulltext')
4411
revision_id = record.key[0]
4412
inv = serializer.read_inventory_from_string(bytes, revision_id)
4413
parents = [key[0] for key in record.parents]
4414
self.target_repo.add_inventory(revision_id, inv, parents)
4415
# No need to keep holding this full inv in memory when the rest of
4416
# the substream is likely to be all deltas.
4419
def _extract_and_insert_revisions(self, substream, serializer):
4420
for record in substream:
4421
bytes = record.get_bytes_as('fulltext')
4422
revision_id = record.key[0]
4423
rev = serializer.read_revision_from_string(bytes)
4424
if rev.revision_id != revision_id:
4425
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4426
self.target_repo.add_revision(revision_id, rev)
4429
if self.target_repo._format._fetch_reconcile:
4430
self.target_repo.reconcile()
4433
class StreamSource(object):
4434
"""A source of a stream for fetching between repositories."""
4436
def __init__(self, from_repository, to_format):
4437
"""Create a StreamSource streaming from from_repository."""
4438
self.from_repository = from_repository
4439
self.to_format = to_format
4441
def delta_on_metadata(self):
4442
"""Return True if delta's are permitted on metadata streams.
4444
That is on revisions and signatures.
4446
src_serializer = self.from_repository._format._serializer
4447
target_serializer = self.to_format._serializer
4448
return (self.to_format._fetch_uses_deltas and
4449
src_serializer == target_serializer)
4451
def _fetch_revision_texts(self, revs):
4452
# fetch signatures first and then the revision texts
4453
# may need to be a InterRevisionStore call here.
4454
from_sf = self.from_repository.signatures
4455
# A missing signature is just skipped.
4456
keys = [(rev_id,) for rev_id in revs]
4457
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4459
self.to_format._fetch_order,
4460
not self.to_format._fetch_uses_deltas))
4461
# If a revision has a delta, this is actually expanded inside the
4462
# insert_record_stream code now, which is an alternate fix for
4464
from_rf = self.from_repository.revisions
4465
revisions = from_rf.get_record_stream(
4467
self.to_format._fetch_order,
4468
not self.delta_on_metadata())
4469
return [('signatures', signatures), ('revisions', revisions)]
4471
def _generate_root_texts(self, revs):
4472
"""This will be called by get_stream between fetching weave texts and
4473
fetching the inventory weave.
4475
if self._rich_root_upgrade():
4476
return _mod_fetch.Inter1and2Helper(
4477
self.from_repository).generate_root_texts(revs)
4481
def get_stream(self, search):
4483
revs = search.get_keys()
4484
graph = self.from_repository.get_graph()
4485
revs = tsort.topo_sort(graph.get_parent_map(revs))
4486
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4488
for knit_kind, file_id, revisions in data_to_fetch:
4489
if knit_kind != phase:
4491
# Make a new progress bar for this phase
4492
if knit_kind == "file":
4493
# Accumulate file texts
4494
text_keys.extend([(file_id, revision) for revision in
4496
elif knit_kind == "inventory":
4497
# Now copy the file texts.
4498
from_texts = self.from_repository.texts
4499
yield ('texts', from_texts.get_record_stream(
4500
text_keys, self.to_format._fetch_order,
4501
not self.to_format._fetch_uses_deltas))
4502
# Cause an error if a text occurs after we have done the
4505
# Before we process the inventory we generate the root
4506
# texts (if necessary) so that the inventories references
4508
for _ in self._generate_root_texts(revs):
4510
# we fetch only the referenced inventories because we do not
4511
# know for unselected inventories whether all their required
4512
# texts are present in the other repository - it could be
4514
for info in self._get_inventory_stream(revs):
4516
elif knit_kind == "signatures":
4517
# Nothing to do here; this will be taken care of when
4518
# _fetch_revision_texts happens.
4520
elif knit_kind == "revisions":
4521
for record in self._fetch_revision_texts(revs):
4524
raise AssertionError("Unknown knit kind %r" % knit_kind)
4526
def get_stream_for_missing_keys(self, missing_keys):
4527
# missing keys can only occur when we are byte copying and not
4528
# translating (because translation means we don't send
4529
# unreconstructable deltas ever).
4531
keys['texts'] = set()
4532
keys['revisions'] = set()
4533
keys['inventories'] = set()
4534
keys['chk_bytes'] = set()
4535
keys['signatures'] = set()
4536
for key in missing_keys:
4537
keys[key[0]].add(key[1:])
4538
if len(keys['revisions']):
4539
# If we allowed copying revisions at this point, we could end up
4540
# copying a revision without copying its required texts: a
4541
# violation of the requirements for repository integrity.
4542
raise AssertionError(
4543
'cannot copy revisions to fill in missing deltas %s' % (
4544
keys['revisions'],))
4545
for substream_kind, keys in keys.iteritems():
4546
vf = getattr(self.from_repository, substream_kind)
4547
if vf is None and keys:
4548
raise AssertionError(
4549
"cannot fill in keys for a versioned file we don't"
4550
" have: %s needs %s" % (substream_kind, keys))
4552
# No need to stream something we don't have
4554
if substream_kind == 'inventories':
4555
# Some missing keys are genuinely ghosts, filter those out.
4556
present = self.from_repository.inventories.get_parent_map(keys)
4557
revs = [key[0] for key in present]
4558
# Get the inventory stream more-or-less as we do for the
4559
# original stream; there's no reason to assume that records
4560
# direct from the source will be suitable for the sink. (Think
4561
# e.g. 2a -> 1.9-rich-root).
4562
for info in self._get_inventory_stream(revs, missing=True):
4566
# Ask for full texts always so that we don't need more round trips
4567
# after this stream.
4568
# Some of the missing keys are genuinely ghosts, so filter absent
4569
# records. The Sink is responsible for doing another check to
4570
# ensure that ghosts don't introduce missing data for future
4572
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4573
self.to_format._fetch_order, True))
4574
yield substream_kind, stream
4576
def inventory_fetch_order(self):
4577
if self._rich_root_upgrade():
4578
return 'topological'
4580
return self.to_format._fetch_order
4582
def _rich_root_upgrade(self):
4583
return (not self.from_repository._format.rich_root_data and
4584
self.to_format.rich_root_data)
4586
def _get_inventory_stream(self, revision_ids, missing=False):
4587
from_format = self.from_repository._format
4588
if (from_format.supports_chks and self.to_format.supports_chks and
4589
from_format.network_name() == self.to_format.network_name()):
4590
raise AssertionError(
4591
"this case should be handled by GroupCHKStreamSource")
4592
elif 'forceinvdeltas' in debug.debug_flags:
4593
return self._get_convertable_inventory_stream(revision_ids,
4594
delta_versus_null=missing)
4595
elif from_format.network_name() == self.to_format.network_name():
4597
return self._get_simple_inventory_stream(revision_ids,
4599
elif (not from_format.supports_chks and not self.to_format.supports_chks
4600
and from_format._serializer == self.to_format._serializer):
4601
# Essentially the same format.
4602
return self._get_simple_inventory_stream(revision_ids,
4605
# Any time we switch serializations, we want to use an
4606
# inventory-delta based approach.
4607
return self._get_convertable_inventory_stream(revision_ids,
4608
delta_versus_null=missing)
4610
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4611
# NB: This currently reopens the inventory weave in source;
4612
# using a single stream interface instead would avoid this.
4613
from_weave = self.from_repository.inventories
4615
delta_closure = True
4617
delta_closure = not self.delta_on_metadata()
4618
yield ('inventories', from_weave.get_record_stream(
4619
[(rev_id,) for rev_id in revision_ids],
4620
self.inventory_fetch_order(), delta_closure))
4622
def _get_convertable_inventory_stream(self, revision_ids,
4623
delta_versus_null=False):
4624
# The two formats are sufficiently different that there is no fast
4625
# path, so we need to send just inventorydeltas, which any
4626
# sufficiently modern client can insert into any repository.
4627
# The StreamSink code expects to be able to
4628
# convert on the target, so we need to put bytes-on-the-wire that can
4629
# be converted. That means inventory deltas (if the remote is <1.19,
4630
# RemoteStreamSink will fallback to VFS to insert the deltas).
4631
yield ('inventory-deltas',
4632
self._stream_invs_as_deltas(revision_ids,
4633
delta_versus_null=delta_versus_null))
4635
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4636
"""Return a stream of inventory-deltas for the given rev ids.
4638
:param revision_ids: The list of inventories to transmit
4639
:param delta_versus_null: Don't try to find a minimal delta for this
4640
entry, instead compute the delta versus the NULL_REVISION. This
4641
effectively streams a complete inventory. Used for stuff like
4642
filling in missing parents, etc.
4644
from_repo = self.from_repository
4645
revision_keys = [(rev_id,) for rev_id in revision_ids]
4646
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4647
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4649
inventories = self.from_repository.iter_inventories(
4650
revision_ids, 'topological')
4651
format = from_repo._format
4652
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4653
inventory_cache = lru_cache.LRUCache(50)
4654
null_inventory = from_repo.revision_tree(
4655
_mod_revision.NULL_REVISION).inventory
4656
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4657
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4658
# repo back into a non-rich-root repo ought to be allowed)
4659
serializer = inventory_delta.InventoryDeltaSerializer(
4660
versioned_root=format.rich_root_data,
4661
tree_references=format.supports_tree_reference)
4662
for inv in inventories:
4663
key = (inv.revision_id,)
4664
parent_keys = parent_map.get(key, ())
4666
if not delta_versus_null and parent_keys:
4667
# The caller did not ask for complete inventories and we have
4668
# some parents that we can delta against. Make a delta against
4669
# each parent so that we can find the smallest.
4670
parent_ids = [parent_key[0] for parent_key in parent_keys]
4671
for parent_id in parent_ids:
4672
if parent_id not in invs_sent_so_far:
4673
# We don't know that the remote side has this basis, so
4676
if parent_id == _mod_revision.NULL_REVISION:
4677
parent_inv = null_inventory
4679
parent_inv = inventory_cache.get(parent_id, None)
4680
if parent_inv is None:
4681
parent_inv = from_repo.get_inventory(parent_id)
4682
candidate_delta = inv._make_delta(parent_inv)
4683
if (delta is None or
4684
len(delta) > len(candidate_delta)):
4685
delta = candidate_delta
4686
basis_id = parent_id
4688
# Either none of the parents ended up being suitable, or we
4689
# were asked to delta against NULL
4690
basis_id = _mod_revision.NULL_REVISION
4691
delta = inv._make_delta(null_inventory)
4692
invs_sent_so_far.add(inv.revision_id)
4693
inventory_cache[inv.revision_id] = inv
4694
delta_serialized = ''.join(
4695
serializer.delta_to_lines(basis_id, key[-1], delta))
4696
yield versionedfile.FulltextContentFactory(
4697
key, parent_keys, None, delta_serialized)
4700
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4701
stop_revision=None):
4702
"""Extend the partial history to include a given index
4704
If a stop_index is supplied, stop when that index has been reached.
4705
If a stop_revision is supplied, stop when that revision is
4706
encountered. Otherwise, stop when the beginning of history is
4709
:param stop_index: The index which should be present. When it is
4710
present, history extension will stop.
4711
:param stop_revision: The revision id which should be present. When
4712
it is encountered, history extension will stop.
4714
start_revision = partial_history_cache[-1]
4715
iterator = repo.iter_reverse_revision_history(start_revision)
4717
#skip the last revision in the list
4720
if (stop_index is not None and
4721
len(partial_history_cache) > stop_index):
4723
if partial_history_cache[-1] == stop_revision:
4725
revision_id = iterator.next()
4726
partial_history_cache.append(revision_id)
4727
except StopIteration:
added
removed
bzrlib/repository.py
