558
492
ie.revision = parent_entry.revision
559
493
return self._get_delta(ie, basis_inv, path), False, None
560
494
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)
496
self._add_text_to_weave(ie.file_id, lines, heads, None)
566
498
raise NotImplementedError('unknown kind')
567
499
ie.revision = self._new_revision_id
568
self._any_changes = True
569
500
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]
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]
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]
847
515
class RootCommitBuilder(CommitBuilder):
848
516
"""This commitbuilder actually records the root id"""
850
518
# the root entry gets versioned properly by this builder.
851
519
_versioned_root = True
1195
797
# The old API returned a list, should this actually be a set?
1196
798
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]))
1312
801
def create(a_bzrdir):
1313
802
"""Construct the current default format repository in a_bzrdir."""
1588
1076
"""Commit the contents accrued within the current write group.
1590
1078
:seealso: start_write_group.
1592
:return: it may return an opaque hint that can be passed to 'pack'.
1594
1080
if self._write_group is not self.get_transaction():
1595
1081
# has an unlock or relock occured ?
1596
1082
raise errors.BzrError('mismatched lock context %r and '
1597
1083
'write group %r.' %
1598
1084
(self.get_transaction(), self._write_group))
1599
result = self._commit_write_group()
1085
self._commit_write_group()
1600
1086
self._write_group = None
1603
1088
def _commit_write_group(self):
1604
1089
"""Template method for per-repository write group cleanup.
1606
This is called before the write group is considered to be
1091
This is called before the write group is considered to be
1607
1092
finished and should ensure that all data handed to the repository
1608
for writing during the write group is safely committed (to the
1093
for writing during the write group is safely committed (to the
1609
1094
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,
1097
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False):
1697
1098
"""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.
1100
If revision_id is None all content is copied.
1706
1101
:param find_ghosts: Find and copy revisions in the source that are
1707
1102
ghosts in the target (and not reachable directly by walking out to
1708
1103
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
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
1105
# fast path same-url fetch operations
1724
# TODO: lift out to somewhere common with RemoteRepository
1725
# <https://bugs.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)):
1106
if self.has_same_location(source):
1729
1107
# check that last_revision is in 'from' and then return a
1730
1108
# no-operation.
1731
1109
if (revision_id is not None and
1894
1259
@needs_read_lock
1895
1260
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.
1261
"""Get many revisions at once."""
1901
1262
return self._get_revisions(revision_ids)
1903
1264
@needs_read_lock
1904
1265
def _get_revisions(self, revision_ids):
1905
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)
1907
for revid, rev in self._iter_revisions(revision_ids):
1909
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
1911
1279
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):
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):
1941
1294
"""Produce a generator of revision deltas.
1943
1296
Note that the input is a sequence of REVISIONS, not revision_ids.
1944
1297
Trees will be held in memory until the generator exits.
1945
1298
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
1300
required_trees = set()
1953
1301
for revision in revisions:
1954
1302
required_trees.add(revision.revision_id)
1955
1303
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
1304
trees = dict((t.get_revision_id(), t) for
1305
t in self.revision_trees(required_trees))
1972
1306
for revision in revisions:
1973
1307
if not revision.parent_ids:
1974
1308
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
2381
1662
"""Get Inventory object by revision id."""
2382
1663
return self.iter_inventories([revision_id]).next()
2384
def iter_inventories(self, revision_ids, ordering=None):
1665
def iter_inventories(self, revision_ids):
2385
1666
"""Get many inventories by revision_ids.
2387
1668
This will buffer some or all of the texts used in constructing the
2388
1669
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
1672
:return: An iterator of inventories.
2397
1674
if ((None in revision_ids)
2398
1675
or (_mod_revision.NULL_REVISION in revision_ids)):
2399
1676
raise ValueError('cannot get null revision inventory')
2400
return self._iter_inventories(revision_ids, ordering)
1677
return self._iter_inventories(revision_ids)
2402
def _iter_inventories(self, revision_ids, ordering):
1679
def _iter_inventories(self, revision_ids):
2403
1680
"""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)
1681
for text, revision_id in self._iter_inventory_xmls(revision_ids):
1682
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
1684
def _iter_inventory_xmls(self, revision_ids):
2414
1685
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)
1686
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2421
1687
text_chunks = {}
2422
1688
for record in stream:
2423
1689
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]
1690
text_chunks[record.key] = record.get_bytes_as('chunked')
2430
1692
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
1694
chunks = text_chunks.pop(key)
1695
yield ''.join(chunks), key[-1]
2444
def _deserialise_inventory(self, revision_id, xml):
2445
"""Transform the xml into an inventory object.
1697
def deserialise_inventory(self, revision_id, xml):
1698
"""Transform the xml into an inventory object.
2447
1700
:param revision_id: The expected revision id of the inventory.
2448
1701
:param xml: A serialised inventory.
2450
1703
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)
1704
entry_cache=self._inventory_entry_cache)
2453
1705
if result.revision_id != revision_id:
2454
1706
raise AssertionError('revision id mismatch %s != %s' % (
2455
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)
2458
1716
def get_serializer_format(self):
2459
1717
return self._serializer.format_num
2461
1719
@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')
1720
def get_inventory_xml(self, revision_id):
1721
"""Get inventory XML as a file object."""
1722
texts = self._iter_inventory_xmls([revision_id])
2466
1724
text, revision_id = texts.next()
2467
1725
except StopIteration:
2468
1726
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)).
1730
def get_inventory_sha1(self, revision_id):
1731
"""Return the sha1 hash of the inventory entry
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])
1733
return self.get_revision(revision_id).inventory_sha1
2506
1735
def iter_reverse_revision_history(self, revision_id):
2507
1736
"""Iterate backwards through revision ids in the lefthand history
3740
2909
return self.source.revision_ids_to_search_result(result_set)
3743
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):
3746
3170
def _get_repo_format_to_test(self):
3750
3174
def is_compatible(source, target):
3751
3175
"""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:///'):
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)):
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):
3185
def _fetch_batch(self, revision_ids, basis_id, basis_tree):
3851
3186
"""Fetch across a few revisions.
3853
3188
: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
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
3862
3195
# Walk though all revisions; get inventory deltas, copy referenced
3863
3196
# texts that delta references, insert the delta, revision and
3865
root_keys_to_create = set()
3866
3198
text_keys = set()
3867
3199
pending_deltas = []
3868
3200
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
3201
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
3202
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,
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))
3885
3215
revision = self.source.get_revision(current_revision_id)
3886
3216
pending_deltas.append((basis_id, delta,
3887
3217
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
3218
pending_revisions.append(revision)
3923
cache[current_revision_id] = tree
3924
3219
basis_id = current_revision_id
3925
self.source._safe_to_return_from_cache = False
3926
3221
# Copy file texts
3927
3222
from_texts = self.source.texts
3928
3223
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
3224
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
3225
text_keys, self.target._fetch_order,
3226
not self.target._fetch_uses_deltas))
3938
3228
for delta in pending_deltas:
3939
3229
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
3230
# insert signatures and revisions
3970
3231
for revision in pending_revisions:
4086
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):
4089
3424
InterRepository.register_optimiser(InterDifferingSerializer)
4090
3425
InterRepository.register_optimiser(InterSameDataRepository)
4091
3426
InterRepository.register_optimiser(InterWeaveRepo)
4092
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)
4095
3436
class CopyConverter(object):
4096
3437
"""A repository conversion tool which just performs a copy of the content.
4098
3439
This is slow but quite reliable.
4251
3587
revision_graph[key] = tuple(parent for parent in parents if parent
4252
3588
in revision_graph)
4253
3589
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,
4289
self.target_repo.abort_write_group(suppress_errors=True)
4292
self.target_repo.unlock()
4294
def _locked_insert_stream(self, stream, src_format, is_resume):
4295
to_serializer = self.target_repo._format._serializer
4296
src_serializer = src_format._serializer
4298
if to_serializer == src_serializer:
4299
# If serializers match and the target is a pack repository, set the
4300
# write cache size on the new pack. This avoids poor performance
4301
# on transports where append is unbuffered (such as
4302
# RemoteTransport). This is safe to do because nothing should read
4303
# back from the target repository while a stream with matching
4304
# serialization is being inserted.
4305
# The exception is that a delta record from the source that should
4306
# be a fulltext may need to be expanded by the target (see
4307
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4308
# explicitly flush any buffered writes first in that rare case.
4310
new_pack = self.target_repo._pack_collection._new_pack
4311
except AttributeError:
4312
# Not a pack repository
4315
new_pack.set_write_cache_size(1024*1024)
4316
for substream_type, substream in stream:
4317
if 'stream' in debug.debug_flags:
4318
mutter('inserting substream: %s', substream_type)
4319
if substream_type == 'texts':
4320
self.target_repo.texts.insert_record_stream(substream)
4321
elif substream_type == 'inventories':
4322
if src_serializer == to_serializer:
4323
self.target_repo.inventories.insert_record_stream(
4326
self._extract_and_insert_inventories(
4327
substream, src_serializer)
4328
elif substream_type == 'inventory-deltas':
4329
self._extract_and_insert_inventory_deltas(
4330
substream, src_serializer)
4331
elif substream_type == 'chk_bytes':
4332
# XXX: This doesn't support conversions, as it assumes the
4333
# conversion was done in the fetch code.
4334
self.target_repo.chk_bytes.insert_record_stream(substream)
4335
elif substream_type == 'revisions':
4336
# This may fallback to extract-and-insert more often than
4337
# required if the serializers are different only in terms of
4339
if src_serializer == to_serializer:
4340
self.target_repo.revisions.insert_record_stream(substream)
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
4454
self._record_counter = RecordCounter()
4456
def delta_on_metadata(self):
4457
"""Return True if delta's are permitted on metadata streams.
4459
That is on revisions and signatures.
4461
src_serializer = self.from_repository._format._serializer
4462
target_serializer = self.to_format._serializer
4463
return (self.to_format._fetch_uses_deltas and
4464
src_serializer == target_serializer)
4466
def _fetch_revision_texts(self, revs):
4467
# fetch signatures first and then the revision texts
4468
# may need to be a InterRevisionStore call here.
4469
from_sf = self.from_repository.signatures
4470
# A missing signature is just skipped.
4471
keys = [(rev_id,) for rev_id in revs]
4472
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4474
self.to_format._fetch_order,
4475
not self.to_format._fetch_uses_deltas))
4476
# If a revision has a delta, this is actually expanded inside the
4477
# insert_record_stream code now, which is an alternate fix for
4479
from_rf = self.from_repository.revisions
4480
revisions = from_rf.get_record_stream(
4482
self.to_format._fetch_order,
4483
not self.delta_on_metadata())
4484
return [('signatures', signatures), ('revisions', revisions)]
4486
def _generate_root_texts(self, revs):
4487
"""This will be called by get_stream between fetching weave texts and
4488
fetching the inventory weave.
4490
if self._rich_root_upgrade():
4491
return _mod_fetch.Inter1and2Helper(
4492
self.from_repository).generate_root_texts(revs)
4496
def get_stream(self, search):
4498
revs = search.get_keys()
4499
graph = self.from_repository.get_graph()
4500
revs = tsort.topo_sort(graph.get_parent_map(revs))
4501
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4503
for knit_kind, file_id, revisions in data_to_fetch:
4504
if knit_kind != phase:
4506
# Make a new progress bar for this phase
4507
if knit_kind == "file":
4508
# Accumulate file texts
4509
text_keys.extend([(file_id, revision) for revision in
4511
elif knit_kind == "inventory":
4512
# Now copy the file texts.
4513
from_texts = self.from_repository.texts
4514
yield ('texts', from_texts.get_record_stream(
4515
text_keys, self.to_format._fetch_order,
4516
not self.to_format._fetch_uses_deltas))
4517
# Cause an error if a text occurs after we have done the
4520
# Before we process the inventory we generate the root
4521
# texts (if necessary) so that the inventories references
4523
for _ in self._generate_root_texts(revs):
4525
# we fetch only the referenced inventories because we do not
4526
# know for unselected inventories whether all their required
4527
# texts are present in the other repository - it could be
4529
for info in self._get_inventory_stream(revs):
4531
elif knit_kind == "signatures":
4532
# Nothing to do here; this will be taken care of when
4533
# _fetch_revision_texts happens.
4535
elif knit_kind == "revisions":
4536
for record in self._fetch_revision_texts(revs):
4539
raise AssertionError("Unknown knit kind %r" % knit_kind)
4541
def get_stream_for_missing_keys(self, missing_keys):
4542
# missing keys can only occur when we are byte copying and not
4543
# translating (because translation means we don't send
4544
# unreconstructable deltas ever).
4546
keys['texts'] = set()
4547
keys['revisions'] = set()
4548
keys['inventories'] = set()
4549
keys['chk_bytes'] = set()
4550
keys['signatures'] = set()
4551
for key in missing_keys:
4552
keys[key[0]].add(key[1:])
4553
if len(keys['revisions']):
4554
# If we allowed copying revisions at this point, we could end up
4555
# copying a revision without copying its required texts: a
4556
# violation of the requirements for repository integrity.
4557
raise AssertionError(
4558
'cannot copy revisions to fill in missing deltas %s' % (
4559
keys['revisions'],))
4560
for substream_kind, keys in keys.iteritems():
4561
vf = getattr(self.from_repository, substream_kind)
4562
if vf is None and keys:
4563
raise AssertionError(
4564
"cannot fill in keys for a versioned file we don't"
4565
" have: %s needs %s" % (substream_kind, keys))
4567
# No need to stream something we don't have
4569
if substream_kind == 'inventories':
4570
# Some missing keys are genuinely ghosts, filter those out.
4571
present = self.from_repository.inventories.get_parent_map(keys)
4572
revs = [key[0] for key in present]
4573
# Get the inventory stream more-or-less as we do for the
4574
# original stream; there's no reason to assume that records
4575
# direct from the source will be suitable for the sink. (Think
4576
# e.g. 2a -> 1.9-rich-root).
4577
for info in self._get_inventory_stream(revs, missing=True):
4581
# Ask for full texts always so that we don't need more round trips
4582
# after this stream.
4583
# Some of the missing keys are genuinely ghosts, so filter absent
4584
# records. The Sink is responsible for doing another check to
4585
# ensure that ghosts don't introduce missing data for future
4587
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4588
self.to_format._fetch_order, True))
4589
yield substream_kind, stream
4591
def inventory_fetch_order(self):
4592
if self._rich_root_upgrade():
4593
return 'topological'
4595
return self.to_format._fetch_order
4597
def _rich_root_upgrade(self):
4598
return (not self.from_repository._format.rich_root_data and
4599
self.to_format.rich_root_data)
4601
def _get_inventory_stream(self, revision_ids, missing=False):
4602
from_format = self.from_repository._format
4603
if (from_format.supports_chks and self.to_format.supports_chks and
4604
from_format.network_name() == self.to_format.network_name()):
4605
raise AssertionError(
4606
"this case should be handled by GroupCHKStreamSource")
4607
elif 'forceinvdeltas' in debug.debug_flags:
4608
return self._get_convertable_inventory_stream(revision_ids,
4609
delta_versus_null=missing)
4610
elif from_format.network_name() == self.to_format.network_name():
4612
return self._get_simple_inventory_stream(revision_ids,
4614
elif (not from_format.supports_chks and not self.to_format.supports_chks
4615
and from_format._serializer == self.to_format._serializer):
4616
# Essentially the same format.
4617
return self._get_simple_inventory_stream(revision_ids,
4620
# Any time we switch serializations, we want to use an
4621
# inventory-delta based approach.
4622
return self._get_convertable_inventory_stream(revision_ids,
4623
delta_versus_null=missing)
4625
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4626
# NB: This currently reopens the inventory weave in source;
4627
# using a single stream interface instead would avoid this.
4628
from_weave = self.from_repository.inventories
4630
delta_closure = True
4632
delta_closure = not self.delta_on_metadata()
4633
yield ('inventories', from_weave.get_record_stream(
4634
[(rev_id,) for rev_id in revision_ids],
4635
self.inventory_fetch_order(), delta_closure))
4637
def _get_convertable_inventory_stream(self, revision_ids,
4638
delta_versus_null=False):
4639
# The two formats are sufficiently different that there is no fast
4640
# path, so we need to send just inventorydeltas, which any
4641
# sufficiently modern client can insert into any repository.
4642
# The StreamSink code expects to be able to
4643
# convert on the target, so we need to put bytes-on-the-wire that can
4644
# be converted. That means inventory deltas (if the remote is <1.19,
4645
# RemoteStreamSink will fallback to VFS to insert the deltas).
4646
yield ('inventory-deltas',
4647
self._stream_invs_as_deltas(revision_ids,
4648
delta_versus_null=delta_versus_null))
4650
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4651
"""Return a stream of inventory-deltas for the given rev ids.
4653
:param revision_ids: The list of inventories to transmit
4654
:param delta_versus_null: Don't try to find a minimal delta for this
4655
entry, instead compute the delta versus the NULL_REVISION. This
4656
effectively streams a complete inventory. Used for stuff like
4657
filling in missing parents, etc.
4659
from_repo = self.from_repository
4660
revision_keys = [(rev_id,) for rev_id in revision_ids]
4661
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4662
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4664
inventories = self.from_repository.iter_inventories(
4665
revision_ids, 'topological')
4666
format = from_repo._format
4667
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4668
inventory_cache = lru_cache.LRUCache(50)
4669
null_inventory = from_repo.revision_tree(
4670
_mod_revision.NULL_REVISION).inventory
4671
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4672
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4673
# repo back into a non-rich-root repo ought to be allowed)
4674
serializer = inventory_delta.InventoryDeltaSerializer(
4675
versioned_root=format.rich_root_data,
4676
tree_references=format.supports_tree_reference)
4677
for inv in inventories:
4678
key = (inv.revision_id,)
4679
parent_keys = parent_map.get(key, ())
4681
if not delta_versus_null and parent_keys:
4682
# The caller did not ask for complete inventories and we have
4683
# some parents that we can delta against. Make a delta against
4684
# each parent so that we can find the smallest.
4685
parent_ids = [parent_key[0] for parent_key in parent_keys]
4686
for parent_id in parent_ids:
4687
if parent_id not in invs_sent_so_far:
4688
# We don't know that the remote side has this basis, so
4691
if parent_id == _mod_revision.NULL_REVISION:
4692
parent_inv = null_inventory
4694
parent_inv = inventory_cache.get(parent_id, None)
4695
if parent_inv is None:
4696
parent_inv = from_repo.get_inventory(parent_id)
4697
candidate_delta = inv._make_delta(parent_inv)
4698
if (delta is None or
4699
len(delta) > len(candidate_delta)):
4700
delta = candidate_delta
4701
basis_id = parent_id
4703
# Either none of the parents ended up being suitable, or we
4704
# were asked to delta against NULL
4705
basis_id = _mod_revision.NULL_REVISION
4706
delta = inv._make_delta(null_inventory)
4707
invs_sent_so_far.add(inv.revision_id)
4708
inventory_cache[inv.revision_id] = inv
4709
delta_serialized = ''.join(
4710
serializer.delta_to_lines(basis_id, key[-1], delta))
4711
yield versionedfile.FulltextContentFactory(
4712
key, parent_keys, None, delta_serialized)
4715
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4716
stop_revision=None):
4717
"""Extend the partial history to include a given index
4719
If a stop_index is supplied, stop when that index has been reached.
4720
If a stop_revision is supplied, stop when that revision is
4721
encountered. Otherwise, stop when the beginning of history is
4724
:param stop_index: The index which should be present. When it is
4725
present, history extension will stop.
4726
:param stop_revision: The revision id which should be present. When
4727
it is encountered, history extension will stop.
4729
start_revision = partial_history_cache[-1]
4730
iterator = repo.iter_reverse_revision_history(start_revision)
4732
#skip the last revision in the list
4735
if (stop_index is not None and
4736
len(partial_history_cache) > stop_index):
4738
if partial_history_cache[-1] == stop_revision:
4740
revision_id = iterator.next()
4741
partial_history_cache.append(revision_id)
4742
except StopIteration:
added
removed
bzrlib/repository.py
