492
547
ie.revision = parent_entry.revision
493
548
return self._get_delta(ie, basis_inv, path), False, None
494
549
ie.reference_revision = content_summary[3]
496
self._add_text_to_weave(ie.file_id, lines, heads, None)
550
if ie.reference_revision is None:
551
raise AssertionError("invalid content_summary for nested tree: %r"
552
% (content_summary,))
553
self._add_text_to_weave(ie.file_id, '', heads, None)
498
555
raise NotImplementedError('unknown kind')
499
556
ie.revision = self._new_revision_id
557
self._any_changes = True
500
558
return self._get_delta(ie, basis_inv, path), True, fingerprint
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]
560
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
561
_entry_factory=entry_factory):
562
"""Record a new tree via iter_changes.
564
:param tree: The tree to obtain text contents from for changed objects.
565
:param basis_revision_id: The revision id of the tree the iter_changes
566
has been generated against. Currently assumed to be the same
567
as self.parents[0] - if it is not, errors may occur.
568
:param iter_changes: An iter_changes iterator with the changes to apply
569
to basis_revision_id. The iterator must not include any items with
570
a current kind of None - missing items must be either filtered out
571
or errored-on beefore record_iter_changes sees the item.
572
:param _entry_factory: Private method to bind entry_factory locally for
574
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
577
# Create an inventory delta based on deltas between all the parents and
578
# deltas between all the parent inventories. We use inventory delta's
579
# between the inventory objects because iter_changes masks
580
# last-changed-field only changes.
582
# file_id -> change map, change is fileid, paths, changed, versioneds,
583
# parents, names, kinds, executables
585
# {file_id -> revision_id -> inventory entry, for entries in parent
586
# trees that are not parents[0]
590
revtrees = list(self.repository.revision_trees(self.parents))
591
except errors.NoSuchRevision:
592
# one or more ghosts, slow path.
594
for revision_id in self.parents:
596
revtrees.append(self.repository.revision_tree(revision_id))
597
except errors.NoSuchRevision:
599
basis_revision_id = _mod_revision.NULL_REVISION
601
revtrees.append(self.repository.revision_tree(
602
_mod_revision.NULL_REVISION))
603
# The basis inventory from a repository
605
basis_inv = revtrees[0].inventory
607
basis_inv = self.repository.revision_tree(
608
_mod_revision.NULL_REVISION).inventory
609
if len(self.parents) > 0:
610
if basis_revision_id != self.parents[0] and not ghost_basis:
612
"arbitrary basis parents not yet supported with merges")
613
for revtree in revtrees[1:]:
614
for change in revtree.inventory._make_delta(basis_inv):
615
if change[1] is None:
616
# Not present in this parent.
618
if change[2] not in merged_ids:
619
if change[0] is not None:
620
basis_entry = basis_inv[change[2]]
621
merged_ids[change[2]] = [
623
basis_entry.revision,
626
parent_entries[change[2]] = {
628
basis_entry.revision:basis_entry,
630
change[3].revision:change[3],
633
merged_ids[change[2]] = [change[3].revision]
634
parent_entries[change[2]] = {change[3].revision:change[3]}
636
merged_ids[change[2]].append(change[3].revision)
637
parent_entries[change[2]][change[3].revision] = change[3]
640
# Setup the changes from the tree:
641
# changes maps file_id -> (change, [parent revision_ids])
643
for change in iter_changes:
644
# This probably looks up in basis_inv way to much.
645
if change[1][0] is not None:
646
head_candidate = [basis_inv[change[0]].revision]
649
changes[change[0]] = change, merged_ids.get(change[0],
651
unchanged_merged = set(merged_ids) - set(changes)
652
# Extend the changes dict with synthetic changes to record merges of
654
for file_id in unchanged_merged:
655
# Record a merged version of these items that did not change vs the
656
# basis. This can be either identical parallel changes, or a revert
657
# of a specific file after a merge. The recorded content will be
658
# that of the current tree (which is the same as the basis), but
659
# the per-file graph will reflect a merge.
660
# NB:XXX: We are reconstructing path information we had, this
661
# should be preserved instead.
662
# inv delta change: (file_id, (path_in_source, path_in_target),
663
# changed_content, versioned, parent, name, kind,
666
basis_entry = basis_inv[file_id]
667
except errors.NoSuchId:
668
# a change from basis->some_parents but file_id isn't in basis
669
# so was new in the merge, which means it must have changed
670
# from basis -> current, and as it hasn't the add was reverted
671
# by the user. So we discard this change.
675
(basis_inv.id2path(file_id), tree.id2path(file_id)),
677
(basis_entry.parent_id, basis_entry.parent_id),
678
(basis_entry.name, basis_entry.name),
679
(basis_entry.kind, basis_entry.kind),
680
(basis_entry.executable, basis_entry.executable))
681
changes[file_id] = (change, merged_ids[file_id])
682
# changes contains tuples with the change and a set of inventory
683
# candidates for the file.
685
# old_path, new_path, file_id, new_inventory_entry
686
seen_root = False # Is the root in the basis delta?
687
inv_delta = self._basis_delta
688
modified_rev = self._new_revision_id
689
for change, head_candidates in changes.values():
690
if change[3][1]: # versioned in target.
691
# Several things may be happening here:
692
# We may have a fork in the per-file graph
693
# - record a change with the content from tree
694
# We may have a change against < all trees
695
# - carry over the tree that hasn't changed
696
# We may have a change against all trees
697
# - record the change with the content from tree
700
entry = _entry_factory[kind](file_id, change[5][1],
702
head_set = self._heads(change[0], set(head_candidates))
705
for head_candidate in head_candidates:
706
if head_candidate in head_set:
707
heads.append(head_candidate)
708
head_set.remove(head_candidate)
711
# Could be a carry-over situation:
712
parent_entry_revs = parent_entries.get(file_id, None)
713
if parent_entry_revs:
714
parent_entry = parent_entry_revs.get(heads[0], None)
717
if parent_entry is None:
718
# The parent iter_changes was called against is the one
719
# that is the per-file head, so any change is relevant
720
# iter_changes is valid.
721
carry_over_possible = False
723
# could be a carry over situation
724
# A change against the basis may just indicate a merge,
725
# we need to check the content against the source of the
726
# merge to determine if it was changed after the merge
728
if (parent_entry.kind != entry.kind or
729
parent_entry.parent_id != entry.parent_id or
730
parent_entry.name != entry.name):
731
# Metadata common to all entries has changed
732
# against per-file parent
733
carry_over_possible = False
735
carry_over_possible = True
736
# per-type checks for changes against the parent_entry
739
# Cannot be a carry-over situation
740
carry_over_possible = False
741
# Populate the entry in the delta
743
# XXX: There is still a small race here: If someone reverts the content of a file
744
# after iter_changes examines and decides it has changed,
745
# we will unconditionally record a new version even if some
746
# other process reverts it while commit is running (with
747
# the revert happening after iter_changes did it's
750
entry.executable = True
752
entry.executable = False
753
if (carry_over_possible and
754
parent_entry.executable == entry.executable):
755
# Check the file length, content hash after reading
757
nostore_sha = parent_entry.text_sha1
760
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
762
text = file_obj.read()
766
entry.text_sha1, entry.text_size = self._add_text_to_weave(
767
file_id, text, heads, nostore_sha)
768
yield file_id, change[1][1], (entry.text_sha1, stat_value)
769
except errors.ExistingContent:
770
# No content change against a carry_over parent
771
# Perhaps this should also yield a fs hash update?
773
entry.text_size = parent_entry.text_size
774
entry.text_sha1 = parent_entry.text_sha1
775
elif kind == 'symlink':
777
entry.symlink_target = tree.get_symlink_target(file_id)
778
if (carry_over_possible and
779
parent_entry.symlink_target == entry.symlink_target):
782
self._add_text_to_weave(change[0], '', heads, None)
783
elif kind == 'directory':
784
if carry_over_possible:
787
# Nothing to set on the entry.
788
# XXX: split into the Root and nonRoot versions.
789
if change[1][1] != '' or self.repository.supports_rich_root():
790
self._add_text_to_weave(change[0], '', heads, None)
791
elif kind == 'tree-reference':
792
if not self.repository._format.supports_tree_reference:
793
# This isn't quite sane as an error, but we shouldn't
794
# ever see this code path in practice: tree's don't
795
# permit references when the repo doesn't support tree
797
raise errors.UnsupportedOperation(tree.add_reference,
799
reference_revision = tree.get_reference_revision(change[0])
800
entry.reference_revision = reference_revision
801
if (carry_over_possible and
802
parent_entry.reference_revision == reference_revision):
805
self._add_text_to_weave(change[0], '', heads, None)
807
raise AssertionError('unknown kind %r' % kind)
809
entry.revision = modified_rev
811
entry.revision = parent_entry.revision
814
new_path = change[1][1]
815
inv_delta.append((change[1][0], new_path, change[0], entry))
818
self.new_inventory = None
820
# This should perhaps be guarded by a check that the basis we
821
# commit against is the basis for the commit and if not do a delta
823
self._any_changes = True
825
# housekeeping root entry changes do not affect no-change commits.
826
self._require_root_change(tree)
827
self.basis_delta_revision = basis_revision_id
829
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
830
parent_keys = tuple([(file_id, parent) for parent in parents])
831
return self.repository.texts._add_text(
832
(file_id, self._new_revision_id), parent_keys, new_text,
833
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
515
836
class RootCommitBuilder(CommitBuilder):
516
837
"""This commitbuilder actually records the root id"""
518
839
# the root entry gets versioned properly by this builder.
519
840
_versioned_root = True
797
1167
# The old API returned a list, should this actually be a set?
798
1168
return parent_map.keys()
1170
def _check_inventories(self, checker):
1171
"""Check the inventories found from the revision scan.
1173
This is responsible for verifying the sha1 of inventories and
1174
creating a pending_keys set that covers data referenced by inventories.
1176
bar = ui.ui_factory.nested_progress_bar()
1178
self._do_check_inventories(checker, bar)
1182
def _do_check_inventories(self, checker, bar):
1183
"""Helper for _check_inventories."""
1185
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1186
kinds = ['chk_bytes', 'texts']
1187
count = len(checker.pending_keys)
1188
bar.update("inventories", 0, 2)
1189
current_keys = checker.pending_keys
1190
checker.pending_keys = {}
1191
# Accumulate current checks.
1192
for key in current_keys:
1193
if key[0] != 'inventories' and key[0] not in kinds:
1194
checker._report_items.append('unknown key type %r' % (key,))
1195
keys[key[0]].add(key[1:])
1196
if keys['inventories']:
1197
# NB: output order *should* be roughly sorted - topo or
1198
# inverse topo depending on repository - either way decent
1199
# to just delta against. However, pre-CHK formats didn't
1200
# try to optimise inventory layout on disk. As such the
1201
# pre-CHK code path does not use inventory deltas.
1203
for record in self.inventories.check(keys=keys['inventories']):
1204
if record.storage_kind == 'absent':
1205
checker._report_items.append(
1206
'Missing inventory {%s}' % (record.key,))
1208
last_object = self._check_record('inventories', record,
1209
checker, last_object,
1210
current_keys[('inventories',) + record.key])
1211
del keys['inventories']
1214
bar.update("texts", 1)
1215
while (checker.pending_keys or keys['chk_bytes']
1217
# Something to check.
1218
current_keys = checker.pending_keys
1219
checker.pending_keys = {}
1220
# Accumulate current checks.
1221
for key in current_keys:
1222
if key[0] not in kinds:
1223
checker._report_items.append('unknown key type %r' % (key,))
1224
keys[key[0]].add(key[1:])
1225
# Check the outermost kind only - inventories || chk_bytes || texts
1229
for record in getattr(self, kind).check(keys=keys[kind]):
1230
if record.storage_kind == 'absent':
1231
checker._report_items.append(
1232
'Missing %s {%s}' % (kind, record.key,))
1234
last_object = self._check_record(kind, record,
1235
checker, last_object, current_keys[(kind,) + record.key])
1239
def _check_record(self, kind, record, checker, last_object, item_data):
1240
"""Check a single text from this repository."""
1241
if kind == 'inventories':
1242
rev_id = record.key[0]
1243
inv = self._deserialise_inventory(rev_id,
1244
record.get_bytes_as('fulltext'))
1245
if last_object is not None:
1246
delta = inv._make_delta(last_object)
1247
for old_path, path, file_id, ie in delta:
1250
ie.check(checker, rev_id, inv)
1252
for path, ie in inv.iter_entries():
1253
ie.check(checker, rev_id, inv)
1254
if self._format.fast_deltas:
1256
elif kind == 'chk_bytes':
1257
# No code written to check chk_bytes for this repo format.
1258
checker._report_items.append(
1259
'unsupported key type chk_bytes for %s' % (record.key,))
1260
elif kind == 'texts':
1261
self._check_text(record, checker, item_data)
1263
checker._report_items.append(
1264
'unknown key type %s for %s' % (kind, record.key))
1266
def _check_text(self, record, checker, item_data):
1267
"""Check a single text."""
1268
# Check it is extractable.
1269
# TODO: check length.
1270
if record.storage_kind == 'chunked':
1271
chunks = record.get_bytes_as(record.storage_kind)
1272
sha1 = osutils.sha_strings(chunks)
1273
length = sum(map(len, chunks))
1275
content = record.get_bytes_as('fulltext')
1276
sha1 = osutils.sha_string(content)
1277
length = len(content)
1278
if item_data and sha1 != item_data[1]:
1279
checker._report_items.append(
1280
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1281
(record.key, sha1, item_data[1], item_data[2]))
801
1284
def create(a_bzrdir):
802
1285
"""Construct the current default format repository in a_bzrdir."""
1076
1546
"""Commit the contents accrued within the current write group.
1078
1548
:seealso: start_write_group.
1550
:return: it may return an opaque hint that can be passed to 'pack'.
1080
1552
if self._write_group is not self.get_transaction():
1081
1553
# has an unlock or relock occured ?
1082
1554
raise errors.BzrError('mismatched lock context %r and '
1083
1555
'write group %r.' %
1084
1556
(self.get_transaction(), self._write_group))
1085
self._commit_write_group()
1557
result = self._commit_write_group()
1086
1558
self._write_group = None
1088
1561
def _commit_write_group(self):
1089
1562
"""Template method for per-repository write group cleanup.
1091
This is called before the write group is considered to be
1564
This is called before the write group is considered to be
1092
1565
finished and should ensure that all data handed to the repository
1093
for writing during the write group is safely committed (to the
1566
for writing during the write group is safely committed (to the
1094
1567
extent possible considering file system caching etc).
1097
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False):
1570
def suspend_write_group(self):
1571
raise errors.UnsuspendableWriteGroup(self)
1573
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1574
"""Return the keys of missing inventory parents for revisions added in
1577
A revision is not complete if the inventory delta for that revision
1578
cannot be calculated. Therefore if the parent inventories of a
1579
revision are not present, the revision is incomplete, and e.g. cannot
1580
be streamed by a smart server. This method finds missing inventory
1581
parents for revisions added in this write group.
1583
if not self._format.supports_external_lookups:
1584
# This is only an issue for stacked repositories
1586
if not self.is_in_write_group():
1587
raise AssertionError('not in a write group')
1589
# XXX: We assume that every added revision already has its
1590
# corresponding inventory, so we only check for parent inventories that
1591
# might be missing, rather than all inventories.
1592
parents = set(self.revisions._index.get_missing_parents())
1593
parents.discard(_mod_revision.NULL_REVISION)
1594
unstacked_inventories = self.inventories._index
1595
present_inventories = unstacked_inventories.get_parent_map(
1596
key[-1:] for key in parents)
1597
parents.difference_update(present_inventories)
1598
if len(parents) == 0:
1599
# No missing parent inventories.
1601
if not check_for_missing_texts:
1602
return set(('inventories', rev_id) for (rev_id,) in parents)
1603
# Ok, now we have a list of missing inventories. But these only matter
1604
# if the inventories that reference them are missing some texts they
1605
# appear to introduce.
1606
# XXX: Texts referenced by all added inventories need to be present,
1607
# but at the moment we're only checking for texts referenced by
1608
# inventories at the graph's edge.
1609
key_deps = self.revisions._index._key_dependencies
1610
key_deps.satisfy_refs_for_keys(present_inventories)
1611
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1612
file_ids = self.fileids_altered_by_revision_ids(referrers)
1613
missing_texts = set()
1614
for file_id, version_ids in file_ids.iteritems():
1615
missing_texts.update(
1616
(file_id, version_id) for version_id in version_ids)
1617
present_texts = self.texts.get_parent_map(missing_texts)
1618
missing_texts.difference_update(present_texts)
1619
if not missing_texts:
1620
# No texts are missing, so all revisions and their deltas are
1623
# Alternatively the text versions could be returned as the missing
1624
# keys, but this is likely to be less data.
1625
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1628
def refresh_data(self):
1629
"""Re-read any data needed to to synchronise with disk.
1631
This method is intended to be called after another repository instance
1632
(such as one used by a smart server) has inserted data into the
1633
repository. It may not be called during a write group, but may be
1634
called at any other time.
1636
if self.is_in_write_group():
1637
raise errors.InternalBzrError(
1638
"May not refresh_data while in a write group.")
1639
self._refresh_data()
1641
def resume_write_group(self, tokens):
1642
if not self.is_write_locked():
1643
raise errors.NotWriteLocked(self)
1644
if self._write_group:
1645
raise errors.BzrError('already in a write group')
1646
self._resume_write_group(tokens)
1647
# so we can detect unlock/relock - the write group is now entered.
1648
self._write_group = self.get_transaction()
1650
def _resume_write_group(self, tokens):
1651
raise errors.UnsuspendableWriteGroup(self)
1653
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1098
1655
"""Fetch the content required to construct revision_id from source.
1100
If revision_id is None all content is copied.
1657
If revision_id is None and fetch_spec is None, then all content is
1660
fetch() may not be used when the repository is in a write group -
1661
either finish the current write group before using fetch, or use
1662
fetch before starting the write group.
1101
1664
:param find_ghosts: Find and copy revisions in the source that are
1102
1665
ghosts in the target (and not reachable directly by walking out to
1103
1666
the first-present revision in target from revision_id).
1667
:param revision_id: If specified, all the content needed for this
1668
revision ID will be copied to the target. Fetch will determine for
1669
itself which content needs to be copied.
1670
:param fetch_spec: If specified, a SearchResult or
1671
PendingAncestryResult that describes which revisions to copy. This
1672
allows copying multiple heads at once. Mutually exclusive with
1675
if fetch_spec is not None and revision_id is not None:
1676
raise AssertionError(
1677
"fetch_spec and revision_id are mutually exclusive.")
1678
if self.is_in_write_group():
1679
raise errors.InternalBzrError(
1680
"May not fetch while in a write group.")
1105
1681
# fast path same-url fetch operations
1106
if self.has_same_location(source):
1682
# TODO: lift out to somewhere common with RemoteRepository
1683
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1684
if (self.has_same_location(source)
1685
and fetch_spec is None
1686
and self._has_same_fallbacks(source)):
1107
1687
# check that last_revision is in 'from' and then return a
1108
1688
# no-operation.
1109
1689
if (revision_id is not None and
1259
1852
@needs_read_lock
1260
1853
def get_revisions(self, revision_ids):
1261
"""Get many revisions at once."""
1854
"""Get many revisions at once.
1856
Repositories that need to check data on every revision read should
1857
subclass this method.
1262
1859
return self._get_revisions(revision_ids)
1264
1861
@needs_read_lock
1265
1862
def _get_revisions(self, revision_ids):
1266
1863
"""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)
1865
for revid, rev in self._iter_revisions(revision_ids):
1867
raise errors.NoSuchRevision(self, revid)
1869
return [revs[revid] for revid in revision_ids]
1871
def _iter_revisions(self, revision_ids):
1872
"""Iterate over revision objects.
1874
:param revision_ids: An iterable of revisions to examine. None may be
1875
passed to request all revisions known to the repository. Note that
1876
not all repositories can find unreferenced revisions; for those
1877
repositories only referenced ones will be returned.
1878
:return: An iterator of (revid, revision) tuples. Absent revisions (
1879
those asked for but not available) are returned as (revid, None).
1881
if revision_ids is None:
1882
revision_ids = self.all_revision_ids()
1884
for rev_id in revision_ids:
1885
if not rev_id or not isinstance(rev_id, basestring):
1886
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1270
1887
keys = [(key,) for key in revision_ids]
1271
1888
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1273
1889
for record in stream:
1890
revid = record.key[0]
1274
1891
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
1279
return [revs[revid] for revid in revision_ids]
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):
1894
text = record.get_bytes_as('fulltext')
1895
rev = self._serializer.read_revision_from_string(text)
1898
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1294
1899
"""Produce a generator of revision deltas.
1296
1901
Note that the input is a sequence of REVISIONS, not revision_ids.
1297
1902
Trees will be held in memory until the generator exits.
1298
1903
Each delta is relative to the revision's lefthand predecessor.
1905
:param specific_fileids: if not None, the result is filtered
1906
so that only those file-ids, their parents and their
1907
children are included.
1909
# Get the revision-ids of interest
1300
1910
required_trees = set()
1301
1911
for revision in revisions:
1302
1912
required_trees.add(revision.revision_id)
1303
1913
required_trees.update(revision.parent_ids[:1])
1304
trees = dict((t.get_revision_id(), t) for
1305
t in self.revision_trees(required_trees))
1915
# Get the matching filtered trees. Note that it's more
1916
# efficient to pass filtered trees to changes_from() rather
1917
# than doing the filtering afterwards. changes_from() could
1918
# arguably do the filtering itself but it's path-based, not
1919
# file-id based, so filtering before or afterwards is
1921
if specific_fileids is None:
1922
trees = dict((t.get_revision_id(), t) for
1923
t in self.revision_trees(required_trees))
1925
trees = dict((t.get_revision_id(), t) for
1926
t in self._filtered_revision_trees(required_trees,
1929
# Calculate the deltas
1306
1930
for revision in revisions:
1307
1931
if not revision.parent_ids:
1308
1932
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1662
2339
"""Get Inventory object by revision id."""
1663
2340
return self.iter_inventories([revision_id]).next()
1665
def iter_inventories(self, revision_ids):
2342
def iter_inventories(self, revision_ids, ordering=None):
1666
2343
"""Get many inventories by revision_ids.
1668
2345
This will buffer some or all of the texts used in constructing the
1669
2346
inventories in memory, but will only parse a single inventory at a
2349
:param revision_ids: The expected revision ids of the inventories.
2350
:param ordering: optional ordering, e.g. 'topological'. If not
2351
specified, the order of revision_ids will be preserved (by
2352
buffering if necessary).
1672
2353
:return: An iterator of inventories.
1674
2355
if ((None in revision_ids)
1675
2356
or (_mod_revision.NULL_REVISION in revision_ids)):
1676
2357
raise ValueError('cannot get null revision inventory')
1677
return self._iter_inventories(revision_ids)
2358
return self._iter_inventories(revision_ids, ordering)
1679
def _iter_inventories(self, revision_ids):
2360
def _iter_inventories(self, revision_ids, ordering):
1680
2361
"""single-document based inventory iteration."""
1681
for text, revision_id in self._iter_inventory_xmls(revision_ids):
1682
yield self.deserialise_inventory(revision_id, text)
2362
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2363
for text, revision_id in inv_xmls:
2364
yield self._deserialise_inventory(revision_id, text)
1684
def _iter_inventory_xmls(self, revision_ids):
2366
def _iter_inventory_xmls(self, revision_ids, ordering):
2367
if ordering is None:
2368
order_as_requested = True
2369
ordering = 'unordered'
2371
order_as_requested = False
1685
2372
keys = [(revision_id,) for revision_id in revision_ids]
1686
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2375
if order_as_requested:
2376
key_iter = iter(keys)
2377
next_key = key_iter.next()
2378
stream = self.inventories.get_record_stream(keys, ordering, True)
1687
2379
text_chunks = {}
1688
2380
for record in stream:
1689
2381
if record.storage_kind != 'absent':
1690
text_chunks[record.key] = record.get_bytes_as('chunked')
2382
chunks = record.get_bytes_as('chunked')
2383
if order_as_requested:
2384
text_chunks[record.key] = chunks
2386
yield ''.join(chunks), record.key[-1]
1692
2388
raise errors.NoSuchRevision(self, record.key)
1694
chunks = text_chunks.pop(key)
1695
yield ''.join(chunks), key[-1]
2389
if order_as_requested:
2390
# Yield as many results as we can while preserving order.
2391
while next_key in text_chunks:
2392
chunks = text_chunks.pop(next_key)
2393
yield ''.join(chunks), next_key[-1]
2395
next_key = key_iter.next()
2396
except StopIteration:
2397
# We still want to fully consume the get_record_stream,
2398
# just in case it is not actually finished at this point
1697
def deserialise_inventory(self, revision_id, xml):
1698
"""Transform the xml into an inventory object.
2402
def _deserialise_inventory(self, revision_id, xml):
2403
"""Transform the xml into an inventory object.
1700
2405
:param revision_id: The expected revision id of the inventory.
1701
2406
:param xml: A serialised inventory.
1703
2408
result = self._serializer.read_inventory_from_string(xml, revision_id,
1704
entry_cache=self._inventory_entry_cache)
2409
entry_cache=self._inventory_entry_cache,
2410
return_from_cache=self._safe_to_return_from_cache)
1705
2411
if result.revision_id != revision_id:
1706
2412
raise AssertionError('revision id mismatch %s != %s' % (
1707
2413
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)
1716
2416
def get_serializer_format(self):
1717
2417
return self._serializer.format_num
1719
2419
@needs_read_lock
1720
def get_inventory_xml(self, revision_id):
1721
"""Get inventory XML as a file object."""
1722
texts = self._iter_inventory_xmls([revision_id])
2420
def _get_inventory_xml(self, revision_id):
2421
"""Get serialized inventory as a string."""
2422
texts = self._iter_inventory_xmls([revision_id], 'unordered')
1724
2424
text, revision_id = texts.next()
1725
2425
except StopIteration:
1726
2426
raise errors.HistoryMissing(self, 'inventory', revision_id)
1730
def get_inventory_sha1(self, revision_id):
1731
"""Return the sha1 hash of the inventory entry
2429
def get_rev_id_for_revno(self, revno, known_pair):
2430
"""Return the revision id of a revno, given a later (revno, revid)
2431
pair in the same history.
2433
:return: if found (True, revid). If the available history ran out
2434
before reaching the revno, then this returns
2435
(False, (closest_revno, closest_revid)).
1733
return self.get_revision(revision_id).inventory_sha1
2437
known_revno, known_revid = known_pair
2438
partial_history = [known_revid]
2439
distance_from_known = known_revno - revno
2440
if distance_from_known < 0:
2442
'requested revno (%d) is later than given known revno (%d)'
2443
% (revno, known_revno))
2446
self, partial_history, stop_index=distance_from_known)
2447
except errors.RevisionNotPresent, err:
2448
if err.revision_id == known_revid:
2449
# The start revision (known_revid) wasn't found.
2451
# This is a stacked repository with no fallbacks, or a there's a
2452
# left-hand ghost. Either way, even though the revision named in
2453
# the error isn't in this repo, we know it's the next step in this
2454
# left-hand history.
2455
partial_history.append(err.revision_id)
2456
if len(partial_history) <= distance_from_known:
2457
# Didn't find enough history to get a revid for the revno.
2458
earliest_revno = known_revno - len(partial_history) + 1
2459
return (False, (earliest_revno, partial_history[-1]))
2460
if len(partial_history) - 1 > distance_from_known:
2461
raise AssertionError('_iter_for_revno returned too much history')
2462
return (True, partial_history[-1])
1735
2464
def iter_reverse_revision_history(self, revision_id):
1736
2465
"""Iterate backwards through revision ids in the lefthand history
2909
3671
return self.source.revision_ids_to_search_result(result_set)
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):
3674
class InterDifferingSerializer(InterRepository):
3170
3677
def _get_repo_format_to_test(self):
3174
3681
def is_compatible(source, target):
3175
3682
"""Be compatible with Knit2 source and Knit3 target"""
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)):
3683
# This is redundant with format.check_conversion_target(), however that
3684
# raises an exception, and we just want to say "False" as in we won't
3685
# support converting between these formats.
3686
if 'IDS_never' in debug.debug_flags:
3688
if source.supports_rich_root() and not target.supports_rich_root():
3690
if (source._format.supports_tree_reference
3691
and not target._format.supports_tree_reference):
3693
if target._fallback_repositories and target._format.supports_chks:
3694
# IDS doesn't know how to copy CHKs for the parent inventories it
3695
# adds to stacked repos.
3697
if 'IDS_always' in debug.debug_flags:
3699
# Only use this code path for local source and target. IDS does far
3700
# too much IO (both bandwidth and roundtrips) over a network.
3701
if not source.bzrdir.transport.base.startswith('file:///'):
3703
if not target.bzrdir.transport.base.startswith('file:///'):
3185
def _fetch_batch(self, revision_ids, basis_id, basis_tree):
3707
def _get_trees(self, revision_ids, cache):
3709
for rev_id in revision_ids:
3711
possible_trees.append((rev_id, cache[rev_id]))
3713
# Not cached, but inventory might be present anyway.
3715
tree = self.source.revision_tree(rev_id)
3716
except errors.NoSuchRevision:
3717
# Nope, parent is ghost.
3720
cache[rev_id] = tree
3721
possible_trees.append((rev_id, tree))
3722
return possible_trees
3724
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3725
"""Get the best delta and base for this revision.
3727
:return: (basis_id, delta)
3730
# Generate deltas against each tree, to find the shortest.
3731
texts_possibly_new_in_tree = set()
3732
for basis_id, basis_tree in possible_trees:
3733
delta = tree.inventory._make_delta(basis_tree.inventory)
3734
for old_path, new_path, file_id, new_entry in delta:
3735
if new_path is None:
3736
# This file_id isn't present in the new rev, so we don't
3740
# Rich roots are handled elsewhere...
3742
kind = new_entry.kind
3743
if kind != 'directory' and kind != 'file':
3744
# No text record associated with this inventory entry.
3746
# This is a directory or file that has changed somehow.
3747
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3748
deltas.append((len(delta), basis_id, delta))
3750
return deltas[0][1:]
3752
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3753
"""Find all parent revisions that are absent, but for which the
3754
inventory is present, and copy those inventories.
3756
This is necessary to preserve correctness when the source is stacked
3757
without fallbacks configured. (Note that in cases like upgrade the
3758
source may be not have _fallback_repositories even though it is
3762
for parents in parent_map.values():
3763
parent_revs.update(parents)
3764
present_parents = self.source.get_parent_map(parent_revs)
3765
absent_parents = set(parent_revs).difference(present_parents)
3766
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3767
(rev_id,) for rev_id in absent_parents)
3768
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3769
for parent_tree in self.source.revision_trees(parent_inv_ids):
3770
current_revision_id = parent_tree.get_revision_id()
3771
parents_parents_keys = parent_invs_keys_for_stacking[
3772
(current_revision_id,)]
3773
parents_parents = [key[-1] for key in parents_parents_keys]
3774
basis_id = _mod_revision.NULL_REVISION
3775
basis_tree = self.source.revision_tree(basis_id)
3776
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3777
self.target.add_inventory_by_delta(
3778
basis_id, delta, current_revision_id, parents_parents)
3779
cache[current_revision_id] = parent_tree
3781
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3186
3782
"""Fetch across a few revisions.
3188
3784
:param revision_ids: The revisions to copy
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
3785
:param basis_id: The revision_id of a tree that must be in cache, used
3786
as a basis for delta when no other base is available
3787
:param cache: A cache of RevisionTrees that we can use.
3788
:param a_graph: A Graph object to determine the heads() of the
3789
rich-root data stream.
3790
:return: The revision_id of the last converted tree. The RevisionTree
3791
for it will be in cache
3195
3793
# Walk though all revisions; get inventory deltas, copy referenced
3196
3794
# texts that delta references, insert the delta, revision and
3796
root_keys_to_create = set()
3198
3797
text_keys = set()
3199
3798
pending_deltas = []
3200
3799
pending_revisions = []
3800
parent_map = self.source.get_parent_map(revision_ids)
3801
self._fetch_parent_invs_for_stacking(parent_map, cache)
3802
self.source._safe_to_return_from_cache = True
3201
3803
for tree in self.source.revision_trees(revision_ids):
3804
# Find a inventory delta for this revision.
3805
# Find text entries that need to be copied, too.
3202
3806
current_revision_id = tree.get_revision_id()
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))
3807
parent_ids = parent_map.get(current_revision_id, ())
3808
parent_trees = self._get_trees(parent_ids, cache)
3809
possible_trees = list(parent_trees)
3810
if len(possible_trees) == 0:
3811
# There either aren't any parents, or the parents are ghosts,
3812
# so just use the last converted tree.
3813
possible_trees.append((basis_id, cache[basis_id]))
3814
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3215
3816
revision = self.source.get_revision(current_revision_id)
3216
3817
pending_deltas.append((basis_id, delta,
3217
3818
current_revision_id, revision.parent_ids))
3819
if self._converting_to_rich_root:
3820
self._revision_id_to_root_id[current_revision_id] = \
3822
# Determine which texts are in present in this revision but not in
3823
# any of the available parents.
3824
texts_possibly_new_in_tree = set()
3825
for old_path, new_path, file_id, entry in delta:
3826
if new_path is None:
3827
# This file_id isn't present in the new rev
3831
if not self.target.supports_rich_root():
3832
# The target doesn't support rich root, so we don't
3835
if self._converting_to_rich_root:
3836
# This can't be copied normally, we have to insert
3838
root_keys_to_create.add((file_id, entry.revision))
3841
texts_possibly_new_in_tree.add((file_id, entry.revision))
3842
for basis_id, basis_tree in possible_trees:
3843
basis_inv = basis_tree.inventory
3844
for file_key in list(texts_possibly_new_in_tree):
3845
file_id, file_revision = file_key
3847
entry = basis_inv[file_id]
3848
except errors.NoSuchId:
3850
if entry.revision == file_revision:
3851
texts_possibly_new_in_tree.remove(file_key)
3852
text_keys.update(texts_possibly_new_in_tree)
3218
3853
pending_revisions.append(revision)
3854
cache[current_revision_id] = tree
3219
3855
basis_id = current_revision_id
3856
self.source._safe_to_return_from_cache = False
3221
3857
# Copy file texts
3222
3858
from_texts = self.source.texts
3223
3859
to_texts = self.target.texts
3860
if root_keys_to_create:
3861
root_stream = _mod_fetch._new_root_data_stream(
3862
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3863
self.source, graph=a_graph)
3864
to_texts.insert_record_stream(root_stream)
3224
3865
to_texts.insert_record_stream(from_texts.get_record_stream(
3225
text_keys, self.target._fetch_order,
3226
not self.target._fetch_uses_deltas))
3866
text_keys, self.target._format._fetch_order,
3867
not self.target._format._fetch_uses_deltas))
3868
# insert inventory deltas
3228
3869
for delta in pending_deltas:
3229
3870
self.target.add_inventory_by_delta(*delta)
3871
if self.target._fallback_repositories:
3872
# Make sure this stacked repository has all the parent inventories
3873
# for the new revisions that we are about to insert. We do this
3874
# before adding the revisions so that no revision is added until
3875
# all the inventories it may depend on are added.
3876
# Note that this is overzealous, as we may have fetched these in an
3879
revision_ids = set()
3880
for revision in pending_revisions:
3881
revision_ids.add(revision.revision_id)
3882
parent_ids.update(revision.parent_ids)
3883
parent_ids.difference_update(revision_ids)
3884
parent_ids.discard(_mod_revision.NULL_REVISION)
3885
parent_map = self.source.get_parent_map(parent_ids)
3886
# we iterate over parent_map and not parent_ids because we don't
3887
# want to try copying any revision which is a ghost
3888
for parent_tree in self.source.revision_trees(parent_map):
3889
current_revision_id = parent_tree.get_revision_id()
3890
parents_parents = parent_map[current_revision_id]
3891
possible_trees = self._get_trees(parents_parents, cache)
3892
if len(possible_trees) == 0:
3893
# There either aren't any parents, or the parents are
3894
# ghosts, so just use the last converted tree.
3895
possible_trees.append((basis_id, cache[basis_id]))
3896
basis_id, delta = self._get_delta_for_revision(parent_tree,
3897
parents_parents, possible_trees)
3898
self.target.add_inventory_by_delta(
3899
basis_id, delta, current_revision_id, parents_parents)
3230
3900
# insert signatures and revisions
3231
3901
for revision in pending_revisions:
3309
4018
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):
3424
4021
InterRepository.register_optimiser(InterDifferingSerializer)
3425
4022
InterRepository.register_optimiser(InterSameDataRepository)
3426
4023
InterRepository.register_optimiser(InterWeaveRepo)
3427
4024
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)
3436
4027
class CopyConverter(object):
3437
4028
"""A repository conversion tool which just performs a copy of the content.
3439
4030
This is slow but quite reliable.
3587
4183
revision_graph[key] = tuple(parent for parent in parents if parent
3588
4184
in revision_graph)
3589
4185
return revision_graph
4188
class StreamSink(object):
4189
"""An object that can insert a stream into a repository.
4191
This interface handles the complexity of reserialising inventories and
4192
revisions from different formats, and allows unidirectional insertion into
4193
stacked repositories without looking for the missing basis parents
4197
def __init__(self, target_repo):
4198
self.target_repo = target_repo
4200
def insert_stream(self, stream, src_format, resume_tokens):
4201
"""Insert a stream's content into the target repository.
4203
:param src_format: a bzr repository format.
4205
:return: a list of resume tokens and an iterable of keys additional
4206
items required before the insertion can be completed.
4208
self.target_repo.lock_write()
4211
self.target_repo.resume_write_group(resume_tokens)
4214
self.target_repo.start_write_group()
4217
# locked_insert_stream performs a commit|suspend.
4218
return self._locked_insert_stream(stream, src_format, is_resume)
4220
self.target_repo.abort_write_group(suppress_errors=True)
4223
self.target_repo.unlock()
4225
def _locked_insert_stream(self, stream, src_format, is_resume):
4226
to_serializer = self.target_repo._format._serializer
4227
src_serializer = src_format._serializer
4229
if to_serializer == src_serializer:
4230
# If serializers match and the target is a pack repository, set the
4231
# write cache size on the new pack. This avoids poor performance
4232
# on transports where append is unbuffered (such as
4233
# RemoteTransport). This is safe to do because nothing should read
4234
# back from the target repository while a stream with matching
4235
# serialization is being inserted.
4236
# The exception is that a delta record from the source that should
4237
# be a fulltext may need to be expanded by the target (see
4238
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4239
# explicitly flush any buffered writes first in that rare case.
4241
new_pack = self.target_repo._pack_collection._new_pack
4242
except AttributeError:
4243
# Not a pack repository
4246
new_pack.set_write_cache_size(1024*1024)
4247
for substream_type, substream in stream:
4248
if 'stream' in debug.debug_flags:
4249
mutter('inserting substream: %s', substream_type)
4250
if substream_type == 'texts':
4251
self.target_repo.texts.insert_record_stream(substream)
4252
elif substream_type == 'inventories':
4253
if src_serializer == to_serializer:
4254
self.target_repo.inventories.insert_record_stream(
4257
self._extract_and_insert_inventories(
4258
substream, src_serializer)
4259
elif substream_type == 'inventory-deltas':
4260
ui.ui_factory.warn_cross_format_fetch(src_format,
4261
self.target_repo._format)
4262
self._extract_and_insert_inventory_deltas(
4263
substream, src_serializer)
4264
elif substream_type == 'chk_bytes':
4265
# XXX: This doesn't support conversions, as it assumes the
4266
# conversion was done in the fetch code.
4267
self.target_repo.chk_bytes.insert_record_stream(substream)
4268
elif substream_type == 'revisions':
4269
# This may fallback to extract-and-insert more often than
4270
# required if the serializers are different only in terms of
4272
if src_serializer == to_serializer:
4273
self.target_repo.revisions.insert_record_stream(
4276
self._extract_and_insert_revisions(substream,
4278
elif substream_type == 'signatures':
4279
self.target_repo.signatures.insert_record_stream(substream)
4281
raise AssertionError('kaboom! %s' % (substream_type,))
4282
# Done inserting data, and the missing_keys calculations will try to
4283
# read back from the inserted data, so flush the writes to the new pack
4284
# (if this is pack format).
4285
if new_pack is not None:
4286
new_pack._write_data('', flush=True)
4287
# Find all the new revisions (including ones from resume_tokens)
4288
missing_keys = self.target_repo.get_missing_parent_inventories(
4289
check_for_missing_texts=is_resume)
4291
for prefix, versioned_file in (
4292
('texts', self.target_repo.texts),
4293
('inventories', self.target_repo.inventories),
4294
('revisions', self.target_repo.revisions),
4295
('signatures', self.target_repo.signatures),
4296
('chk_bytes', self.target_repo.chk_bytes),
4298
if versioned_file is None:
4300
# TODO: key is often going to be a StaticTuple object
4301
# I don't believe we can define a method by which
4302
# (prefix,) + StaticTuple will work, though we could
4303
# define a StaticTuple.sq_concat that would allow you to
4304
# pass in either a tuple or a StaticTuple as the second
4305
# object, so instead we could have:
4306
# StaticTuple(prefix) + key here...
4307
missing_keys.update((prefix,) + key for key in
4308
versioned_file.get_missing_compression_parent_keys())
4309
except NotImplementedError:
4310
# cannot even attempt suspending, and missing would have failed
4311
# during stream insertion.
4312
missing_keys = set()
4315
# suspend the write group and tell the caller what we is
4316
# missing. We know we can suspend or else we would not have
4317
# entered this code path. (All repositories that can handle
4318
# missing keys can handle suspending a write group).
4319
write_group_tokens = self.target_repo.suspend_write_group()
4320
return write_group_tokens, missing_keys
4321
hint = self.target_repo.commit_write_group()
4322
if (to_serializer != src_serializer and
4323
self.target_repo._format.pack_compresses):
4324
self.target_repo.pack(hint=hint)
4327
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4328
target_rich_root = self.target_repo._format.rich_root_data
4329
target_tree_refs = self.target_repo._format.supports_tree_reference
4330
for record in substream:
4331
# Insert the delta directly
4332
inventory_delta_bytes = record.get_bytes_as('fulltext')
4333
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4335
parse_result = deserialiser.parse_text_bytes(
4336
inventory_delta_bytes)
4337
except inventory_delta.IncompatibleInventoryDelta, err:
4338
trace.mutter("Incompatible delta: %s", err.msg)
4339
raise errors.IncompatibleRevision(self.target_repo._format)
4340
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4341
revision_id = new_id
4342
parents = [key[0] for key in record.parents]
4343
self.target_repo.add_inventory_by_delta(
4344
basis_id, inv_delta, revision_id, parents)
4346
def _extract_and_insert_inventories(self, substream, serializer,
4348
"""Generate a new inventory versionedfile in target, converting data.
4350
The inventory is retrieved from the source, (deserializing it), and
4351
stored in the target (reserializing it in a different format).
4353
target_rich_root = self.target_repo._format.rich_root_data
4354
target_tree_refs = self.target_repo._format.supports_tree_reference
4355
for record in substream:
4356
# It's not a delta, so it must be a fulltext in the source
4357
# serializer's format.
4358
bytes = record.get_bytes_as('fulltext')
4359
revision_id = record.key[0]
4360
inv = serializer.read_inventory_from_string(bytes, revision_id)
4361
parents = [key[0] for key in record.parents]
4362
self.target_repo.add_inventory(revision_id, inv, parents)
4363
# No need to keep holding this full inv in memory when the rest of
4364
# the substream is likely to be all deltas.
4367
def _extract_and_insert_revisions(self, substream, serializer):
4368
for record in substream:
4369
bytes = record.get_bytes_as('fulltext')
4370
revision_id = record.key[0]
4371
rev = serializer.read_revision_from_string(bytes)
4372
if rev.revision_id != revision_id:
4373
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4374
self.target_repo.add_revision(revision_id, rev)
4377
if self.target_repo._format._fetch_reconcile:
4378
self.target_repo.reconcile()
4381
class StreamSource(object):
4382
"""A source of a stream for fetching between repositories."""
4384
def __init__(self, from_repository, to_format):
4385
"""Create a StreamSource streaming from from_repository."""
4386
self.from_repository = from_repository
4387
self.to_format = to_format
4389
def delta_on_metadata(self):
4390
"""Return True if delta's are permitted on metadata streams.
4392
That is on revisions and signatures.
4394
src_serializer = self.from_repository._format._serializer
4395
target_serializer = self.to_format._serializer
4396
return (self.to_format._fetch_uses_deltas and
4397
src_serializer == target_serializer)
4399
def _fetch_revision_texts(self, revs):
4400
# fetch signatures first and then the revision texts
4401
# may need to be a InterRevisionStore call here.
4402
from_sf = self.from_repository.signatures
4403
# A missing signature is just skipped.
4404
keys = [(rev_id,) for rev_id in revs]
4405
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4407
self.to_format._fetch_order,
4408
not self.to_format._fetch_uses_deltas))
4409
# If a revision has a delta, this is actually expanded inside the
4410
# insert_record_stream code now, which is an alternate fix for
4412
from_rf = self.from_repository.revisions
4413
revisions = from_rf.get_record_stream(
4415
self.to_format._fetch_order,
4416
not self.delta_on_metadata())
4417
return [('signatures', signatures), ('revisions', revisions)]
4419
def _generate_root_texts(self, revs):
4420
"""This will be called by get_stream between fetching weave texts and
4421
fetching the inventory weave.
4423
if self._rich_root_upgrade():
4424
return _mod_fetch.Inter1and2Helper(
4425
self.from_repository).generate_root_texts(revs)
4429
def get_stream(self, search):
4431
revs = search.get_keys()
4432
graph = self.from_repository.get_graph()
4433
revs = tsort.topo_sort(graph.get_parent_map(revs))
4434
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4436
for knit_kind, file_id, revisions in data_to_fetch:
4437
if knit_kind != phase:
4439
# Make a new progress bar for this phase
4440
if knit_kind == "file":
4441
# Accumulate file texts
4442
text_keys.extend([(file_id, revision) for revision in
4444
elif knit_kind == "inventory":
4445
# Now copy the file texts.
4446
from_texts = self.from_repository.texts
4447
yield ('texts', from_texts.get_record_stream(
4448
text_keys, self.to_format._fetch_order,
4449
not self.to_format._fetch_uses_deltas))
4450
# Cause an error if a text occurs after we have done the
4453
# Before we process the inventory we generate the root
4454
# texts (if necessary) so that the inventories references
4456
for _ in self._generate_root_texts(revs):
4458
# we fetch only the referenced inventories because we do not
4459
# know for unselected inventories whether all their required
4460
# texts are present in the other repository - it could be
4462
for info in self._get_inventory_stream(revs):
4464
elif knit_kind == "signatures":
4465
# Nothing to do here; this will be taken care of when
4466
# _fetch_revision_texts happens.
4468
elif knit_kind == "revisions":
4469
for record in self._fetch_revision_texts(revs):
4472
raise AssertionError("Unknown knit kind %r" % knit_kind)
4474
def get_stream_for_missing_keys(self, missing_keys):
4475
# missing keys can only occur when we are byte copying and not
4476
# translating (because translation means we don't send
4477
# unreconstructable deltas ever).
4479
keys['texts'] = set()
4480
keys['revisions'] = set()
4481
keys['inventories'] = set()
4482
keys['chk_bytes'] = set()
4483
keys['signatures'] = set()
4484
for key in missing_keys:
4485
keys[key[0]].add(key[1:])
4486
if len(keys['revisions']):
4487
# If we allowed copying revisions at this point, we could end up
4488
# copying a revision without copying its required texts: a
4489
# violation of the requirements for repository integrity.
4490
raise AssertionError(
4491
'cannot copy revisions to fill in missing deltas %s' % (
4492
keys['revisions'],))
4493
for substream_kind, keys in keys.iteritems():
4494
vf = getattr(self.from_repository, substream_kind)
4495
if vf is None and keys:
4496
raise AssertionError(
4497
"cannot fill in keys for a versioned file we don't"
4498
" have: %s needs %s" % (substream_kind, keys))
4500
# No need to stream something we don't have
4502
if substream_kind == 'inventories':
4503
# Some missing keys are genuinely ghosts, filter those out.
4504
present = self.from_repository.inventories.get_parent_map(keys)
4505
revs = [key[0] for key in present]
4506
# Get the inventory stream more-or-less as we do for the
4507
# original stream; there's no reason to assume that records
4508
# direct from the source will be suitable for the sink. (Think
4509
# e.g. 2a -> 1.9-rich-root).
4510
for info in self._get_inventory_stream(revs, missing=True):
4514
# Ask for full texts always so that we don't need more round trips
4515
# after this stream.
4516
# Some of the missing keys are genuinely ghosts, so filter absent
4517
# records. The Sink is responsible for doing another check to
4518
# ensure that ghosts don't introduce missing data for future
4520
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4521
self.to_format._fetch_order, True))
4522
yield substream_kind, stream
4524
def inventory_fetch_order(self):
4525
if self._rich_root_upgrade():
4526
return 'topological'
4528
return self.to_format._fetch_order
4530
def _rich_root_upgrade(self):
4531
return (not self.from_repository._format.rich_root_data and
4532
self.to_format.rich_root_data)
4534
def _get_inventory_stream(self, revision_ids, missing=False):
4535
from_format = self.from_repository._format
4536
if (from_format.supports_chks and self.to_format.supports_chks and
4537
from_format.network_name() == self.to_format.network_name()):
4538
raise AssertionError(
4539
"this case should be handled by GroupCHKStreamSource")
4540
elif 'forceinvdeltas' in debug.debug_flags:
4541
return self._get_convertable_inventory_stream(revision_ids,
4542
delta_versus_null=missing)
4543
elif from_format.network_name() == self.to_format.network_name():
4545
return self._get_simple_inventory_stream(revision_ids,
4547
elif (not from_format.supports_chks and not self.to_format.supports_chks
4548
and from_format._serializer == self.to_format._serializer):
4549
# Essentially the same format.
4550
return self._get_simple_inventory_stream(revision_ids,
4553
# Any time we switch serializations, we want to use an
4554
# inventory-delta based approach.
4555
return self._get_convertable_inventory_stream(revision_ids,
4556
delta_versus_null=missing)
4558
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4559
# NB: This currently reopens the inventory weave in source;
4560
# using a single stream interface instead would avoid this.
4561
from_weave = self.from_repository.inventories
4563
delta_closure = True
4565
delta_closure = not self.delta_on_metadata()
4566
yield ('inventories', from_weave.get_record_stream(
4567
[(rev_id,) for rev_id in revision_ids],
4568
self.inventory_fetch_order(), delta_closure))
4570
def _get_convertable_inventory_stream(self, revision_ids,
4571
delta_versus_null=False):
4572
# The two formats are sufficiently different that there is no fast
4573
# path, so we need to send just inventorydeltas, which any
4574
# sufficiently modern client can insert into any repository.
4575
# The StreamSink code expects to be able to
4576
# convert on the target, so we need to put bytes-on-the-wire that can
4577
# be converted. That means inventory deltas (if the remote is <1.19,
4578
# RemoteStreamSink will fallback to VFS to insert the deltas).
4579
yield ('inventory-deltas',
4580
self._stream_invs_as_deltas(revision_ids,
4581
delta_versus_null=delta_versus_null))
4583
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4584
"""Return a stream of inventory-deltas for the given rev ids.
4586
:param revision_ids: The list of inventories to transmit
4587
:param delta_versus_null: Don't try to find a minimal delta for this
4588
entry, instead compute the delta versus the NULL_REVISION. This
4589
effectively streams a complete inventory. Used for stuff like
4590
filling in missing parents, etc.
4592
from_repo = self.from_repository
4593
revision_keys = [(rev_id,) for rev_id in revision_ids]
4594
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4595
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4597
inventories = self.from_repository.iter_inventories(
4598
revision_ids, 'topological')
4599
format = from_repo._format
4600
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4601
inventory_cache = lru_cache.LRUCache(50)
4602
null_inventory = from_repo.revision_tree(
4603
_mod_revision.NULL_REVISION).inventory
4604
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4605
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4606
# repo back into a non-rich-root repo ought to be allowed)
4607
serializer = inventory_delta.InventoryDeltaSerializer(
4608
versioned_root=format.rich_root_data,
4609
tree_references=format.supports_tree_reference)
4610
for inv in inventories:
4611
key = (inv.revision_id,)
4612
parent_keys = parent_map.get(key, ())
4614
if not delta_versus_null and parent_keys:
4615
# The caller did not ask for complete inventories and we have
4616
# some parents that we can delta against. Make a delta against
4617
# each parent so that we can find the smallest.
4618
parent_ids = [parent_key[0] for parent_key in parent_keys]
4619
for parent_id in parent_ids:
4620
if parent_id not in invs_sent_so_far:
4621
# We don't know that the remote side has this basis, so
4624
if parent_id == _mod_revision.NULL_REVISION:
4625
parent_inv = null_inventory
4627
parent_inv = inventory_cache.get(parent_id, None)
4628
if parent_inv is None:
4629
parent_inv = from_repo.get_inventory(parent_id)
4630
candidate_delta = inv._make_delta(parent_inv)
4631
if (delta is None or
4632
len(delta) > len(candidate_delta)):
4633
delta = candidate_delta
4634
basis_id = parent_id
4636
# Either none of the parents ended up being suitable, or we
4637
# were asked to delta against NULL
4638
basis_id = _mod_revision.NULL_REVISION
4639
delta = inv._make_delta(null_inventory)
4640
invs_sent_so_far.add(inv.revision_id)
4641
inventory_cache[inv.revision_id] = inv
4642
delta_serialized = ''.join(
4643
serializer.delta_to_lines(basis_id, key[-1], delta))
4644
yield versionedfile.FulltextContentFactory(
4645
key, parent_keys, None, delta_serialized)
4648
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4649
stop_revision=None):
4650
"""Extend the partial history to include a given index
4652
If a stop_index is supplied, stop when that index has been reached.
4653
If a stop_revision is supplied, stop when that revision is
4654
encountered. Otherwise, stop when the beginning of history is
4657
:param stop_index: The index which should be present. When it is
4658
present, history extension will stop.
4659
:param stop_revision: The revision id which should be present. When
4660
it is encountered, history extension will stop.
4662
start_revision = partial_history_cache[-1]
4663
iterator = repo.iter_reverse_revision_history(start_revision)
4665
#skip the last revision in the list
4668
if (stop_index is not None and
4669
len(partial_history_cache) > stop_index):
4671
if partial_history_cache[-1] == stop_revision:
4673
revision_id = iterator.next()
4674
partial_history_cache.append(revision_id)
4675
except StopIteration:
added
removed
bzrlib/repository.py
