492
546
ie.revision = parent_entry.revision
493
547
return self._get_delta(ie, basis_inv, path), False, None
494
548
ie.reference_revision = content_summary[3]
496
self._add_text_to_weave(ie.file_id, lines, heads, None)
549
if ie.reference_revision is None:
550
raise AssertionError("invalid content_summary for nested tree: %r"
551
% (content_summary,))
552
self._add_text_to_weave(ie.file_id, '', heads, None)
498
554
raise NotImplementedError('unknown kind')
499
555
ie.revision = self._new_revision_id
556
self._any_changes = True
500
557
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]
559
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
560
_entry_factory=entry_factory):
561
"""Record a new tree via iter_changes.
563
:param tree: The tree to obtain text contents from for changed objects.
564
:param basis_revision_id: The revision id of the tree the iter_changes
565
has been generated against. Currently assumed to be the same
566
as self.parents[0] - if it is not, errors may occur.
567
:param iter_changes: An iter_changes iterator with the changes to apply
568
to basis_revision_id. The iterator must not include any items with
569
a current kind of None - missing items must be either filtered out
570
or errored-on beefore record_iter_changes sees the item.
571
:param _entry_factory: Private method to bind entry_factory locally for
573
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
576
# Create an inventory delta based on deltas between all the parents and
577
# deltas between all the parent inventories. We use inventory delta's
578
# between the inventory objects because iter_changes masks
579
# last-changed-field only changes.
581
# file_id -> change map, change is fileid, paths, changed, versioneds,
582
# parents, names, kinds, executables
584
# {file_id -> revision_id -> inventory entry, for entries in parent
585
# trees that are not parents[0]
589
revtrees = list(self.repository.revision_trees(self.parents))
590
except errors.NoSuchRevision:
591
# one or more ghosts, slow path.
593
for revision_id in self.parents:
595
revtrees.append(self.repository.revision_tree(revision_id))
596
except errors.NoSuchRevision:
598
basis_revision_id = _mod_revision.NULL_REVISION
600
revtrees.append(self.repository.revision_tree(
601
_mod_revision.NULL_REVISION))
602
# The basis inventory from a repository
604
basis_inv = revtrees[0].inventory
606
basis_inv = self.repository.revision_tree(
607
_mod_revision.NULL_REVISION).inventory
608
if len(self.parents) > 0:
609
if basis_revision_id != self.parents[0] and not ghost_basis:
611
"arbitrary basis parents not yet supported with merges")
612
for revtree in revtrees[1:]:
613
for change in revtree.inventory._make_delta(basis_inv):
614
if change[1] is None:
615
# Not present in this parent.
617
if change[2] not in merged_ids:
618
if change[0] is not None:
619
basis_entry = basis_inv[change[2]]
620
merged_ids[change[2]] = [
622
basis_entry.revision,
625
parent_entries[change[2]] = {
627
basis_entry.revision:basis_entry,
629
change[3].revision:change[3],
632
merged_ids[change[2]] = [change[3].revision]
633
parent_entries[change[2]] = {change[3].revision:change[3]}
635
merged_ids[change[2]].append(change[3].revision)
636
parent_entries[change[2]][change[3].revision] = change[3]
639
# Setup the changes from the tree:
640
# changes maps file_id -> (change, [parent revision_ids])
642
for change in iter_changes:
643
# This probably looks up in basis_inv way to much.
644
if change[1][0] is not None:
645
head_candidate = [basis_inv[change[0]].revision]
648
changes[change[0]] = change, merged_ids.get(change[0],
650
unchanged_merged = set(merged_ids) - set(changes)
651
# Extend the changes dict with synthetic changes to record merges of
653
for file_id in unchanged_merged:
654
# Record a merged version of these items that did not change vs the
655
# basis. This can be either identical parallel changes, or a revert
656
# of a specific file after a merge. The recorded content will be
657
# that of the current tree (which is the same as the basis), but
658
# the per-file graph will reflect a merge.
659
# NB:XXX: We are reconstructing path information we had, this
660
# should be preserved instead.
661
# inv delta change: (file_id, (path_in_source, path_in_target),
662
# changed_content, versioned, parent, name, kind,
665
basis_entry = basis_inv[file_id]
666
except errors.NoSuchId:
667
# a change from basis->some_parents but file_id isn't in basis
668
# so was new in the merge, which means it must have changed
669
# from basis -> current, and as it hasn't the add was reverted
670
# by the user. So we discard this change.
674
(basis_inv.id2path(file_id), tree.id2path(file_id)),
676
(basis_entry.parent_id, basis_entry.parent_id),
677
(basis_entry.name, basis_entry.name),
678
(basis_entry.kind, basis_entry.kind),
679
(basis_entry.executable, basis_entry.executable))
680
changes[file_id] = (change, merged_ids[file_id])
681
# changes contains tuples with the change and a set of inventory
682
# candidates for the file.
684
# old_path, new_path, file_id, new_inventory_entry
685
seen_root = False # Is the root in the basis delta?
686
inv_delta = self._basis_delta
687
modified_rev = self._new_revision_id
688
for change, head_candidates in changes.values():
689
if change[3][1]: # versioned in target.
690
# Several things may be happening here:
691
# We may have a fork in the per-file graph
692
# - record a change with the content from tree
693
# We may have a change against < all trees
694
# - carry over the tree that hasn't changed
695
# We may have a change against all trees
696
# - record the change with the content from tree
699
entry = _entry_factory[kind](file_id, change[5][1],
701
head_set = self._heads(change[0], set(head_candidates))
704
for head_candidate in head_candidates:
705
if head_candidate in head_set:
706
heads.append(head_candidate)
707
head_set.remove(head_candidate)
710
# Could be a carry-over situation:
711
parent_entry_revs = parent_entries.get(file_id, None)
712
if parent_entry_revs:
713
parent_entry = parent_entry_revs.get(heads[0], None)
716
if parent_entry is None:
717
# The parent iter_changes was called against is the one
718
# that is the per-file head, so any change is relevant
719
# iter_changes is valid.
720
carry_over_possible = False
722
# could be a carry over situation
723
# A change against the basis may just indicate a merge,
724
# we need to check the content against the source of the
725
# merge to determine if it was changed after the merge
727
if (parent_entry.kind != entry.kind or
728
parent_entry.parent_id != entry.parent_id or
729
parent_entry.name != entry.name):
730
# Metadata common to all entries has changed
731
# against per-file parent
732
carry_over_possible = False
734
carry_over_possible = True
735
# per-type checks for changes against the parent_entry
738
# Cannot be a carry-over situation
739
carry_over_possible = False
740
# Populate the entry in the delta
742
# XXX: There is still a small race here: If someone reverts the content of a file
743
# after iter_changes examines and decides it has changed,
744
# we will unconditionally record a new version even if some
745
# other process reverts it while commit is running (with
746
# the revert happening after iter_changes did it's
749
entry.executable = True
751
entry.executable = False
752
if (carry_over_possible and
753
parent_entry.executable == entry.executable):
754
# Check the file length, content hash after reading
756
nostore_sha = parent_entry.text_sha1
759
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
761
text = file_obj.read()
765
entry.text_sha1, entry.text_size = self._add_text_to_weave(
766
file_id, text, heads, nostore_sha)
767
yield file_id, change[1][1], (entry.text_sha1, stat_value)
768
except errors.ExistingContent:
769
# No content change against a carry_over parent
770
# Perhaps this should also yield a fs hash update?
772
entry.text_size = parent_entry.text_size
773
entry.text_sha1 = parent_entry.text_sha1
774
elif kind == 'symlink':
776
entry.symlink_target = tree.get_symlink_target(file_id)
777
if (carry_over_possible and
778
parent_entry.symlink_target == entry.symlink_target):
781
self._add_text_to_weave(change[0], '', heads, None)
782
elif kind == 'directory':
783
if carry_over_possible:
786
# Nothing to set on the entry.
787
# XXX: split into the Root and nonRoot versions.
788
if change[1][1] != '' or self.repository.supports_rich_root():
789
self._add_text_to_weave(change[0], '', heads, None)
790
elif kind == 'tree-reference':
791
if not self.repository._format.supports_tree_reference:
792
# This isn't quite sane as an error, but we shouldn't
793
# ever see this code path in practice: tree's don't
794
# permit references when the repo doesn't support tree
796
raise errors.UnsupportedOperation(tree.add_reference,
798
reference_revision = tree.get_reference_revision(change[0])
799
entry.reference_revision = reference_revision
800
if (carry_over_possible and
801
parent_entry.reference_revision == reference_revision):
804
self._add_text_to_weave(change[0], '', heads, None)
806
raise AssertionError('unknown kind %r' % kind)
808
entry.revision = modified_rev
810
entry.revision = parent_entry.revision
813
new_path = change[1][1]
814
inv_delta.append((change[1][0], new_path, change[0], entry))
817
self.new_inventory = None
819
# This should perhaps be guarded by a check that the basis we
820
# commit against is the basis for the commit and if not do a delta
822
self._any_changes = True
824
# housekeeping root entry changes do not affect no-change commits.
825
self._require_root_change(tree)
826
self.basis_delta_revision = basis_revision_id
828
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
829
parent_keys = tuple([(file_id, parent) for parent in parents])
830
return self.repository.texts._add_text(
831
(file_id, self._new_revision_id), parent_keys, new_text,
832
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
515
835
class RootCommitBuilder(CommitBuilder):
516
836
"""This commitbuilder actually records the root id"""
518
838
# the root entry gets versioned properly by this builder.
519
839
_versioned_root = True
797
1166
# The old API returned a list, should this actually be a set?
798
1167
return parent_map.keys()
1169
def _check_inventories(self, checker):
1170
"""Check the inventories found from the revision scan.
1172
This is responsible for verifying the sha1 of inventories and
1173
creating a pending_keys set that covers data referenced by inventories.
1175
bar = ui.ui_factory.nested_progress_bar()
1177
self._do_check_inventories(checker, bar)
1181
def _do_check_inventories(self, checker, bar):
1182
"""Helper for _check_inventories."""
1184
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1185
kinds = ['chk_bytes', 'texts']
1186
count = len(checker.pending_keys)
1187
bar.update("inventories", 0, 2)
1188
current_keys = checker.pending_keys
1189
checker.pending_keys = {}
1190
# Accumulate current checks.
1191
for key in current_keys:
1192
if key[0] != 'inventories' and key[0] not in kinds:
1193
checker._report_items.append('unknown key type %r' % (key,))
1194
keys[key[0]].add(key[1:])
1195
if keys['inventories']:
1196
# NB: output order *should* be roughly sorted - topo or
1197
# inverse topo depending on repository - either way decent
1198
# to just delta against. However, pre-CHK formats didn't
1199
# try to optimise inventory layout on disk. As such the
1200
# pre-CHK code path does not use inventory deltas.
1202
for record in self.inventories.check(keys=keys['inventories']):
1203
if record.storage_kind == 'absent':
1204
checker._report_items.append(
1205
'Missing inventory {%s}' % (record.key,))
1207
last_object = self._check_record('inventories', record,
1208
checker, last_object,
1209
current_keys[('inventories',) + record.key])
1210
del keys['inventories']
1213
bar.update("texts", 1)
1214
while (checker.pending_keys or keys['chk_bytes']
1216
# Something to check.
1217
current_keys = checker.pending_keys
1218
checker.pending_keys = {}
1219
# Accumulate current checks.
1220
for key in current_keys:
1221
if key[0] not in kinds:
1222
checker._report_items.append('unknown key type %r' % (key,))
1223
keys[key[0]].add(key[1:])
1224
# Check the outermost kind only - inventories || chk_bytes || texts
1228
for record in getattr(self, kind).check(keys=keys[kind]):
1229
if record.storage_kind == 'absent':
1230
checker._report_items.append(
1231
'Missing %s {%s}' % (kind, record.key,))
1233
last_object = self._check_record(kind, record,
1234
checker, last_object, current_keys[(kind,) + record.key])
1238
def _check_record(self, kind, record, checker, last_object, item_data):
1239
"""Check a single text from this repository."""
1240
if kind == 'inventories':
1241
rev_id = record.key[0]
1242
inv = self.deserialise_inventory(rev_id,
1243
record.get_bytes_as('fulltext'))
1244
if last_object is not None:
1245
delta = inv._make_delta(last_object)
1246
for old_path, path, file_id, ie in delta:
1249
ie.check(checker, rev_id, inv)
1251
for path, ie in inv.iter_entries():
1252
ie.check(checker, rev_id, inv)
1253
if self._format.fast_deltas:
1255
elif kind == 'chk_bytes':
1256
# No code written to check chk_bytes for this repo format.
1257
checker._report_items.append(
1258
'unsupported key type chk_bytes for %s' % (record.key,))
1259
elif kind == 'texts':
1260
self._check_text(record, checker, item_data)
1262
checker._report_items.append(
1263
'unknown key type %s for %s' % (kind, record.key))
1265
def _check_text(self, record, checker, item_data):
1266
"""Check a single text."""
1267
# Check it is extractable.
1268
# TODO: check length.
1269
if record.storage_kind == 'chunked':
1270
chunks = record.get_bytes_as(record.storage_kind)
1271
sha1 = osutils.sha_strings(chunks)
1272
length = sum(map(len, chunks))
1274
content = record.get_bytes_as('fulltext')
1275
sha1 = osutils.sha_string(content)
1276
length = len(content)
1277
if item_data and sha1 != item_data[1]:
1278
checker._report_items.append(
1279
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1280
(record.key, sha1, item_data[1], item_data[2]))
801
1283
def create(a_bzrdir):
802
1284
"""Construct the current default format repository in a_bzrdir."""
1076
1551
"""Commit the contents accrued within the current write group.
1078
1553
:seealso: start_write_group.
1555
:return: it may return an opaque hint that can be passed to 'pack'.
1080
1557
if self._write_group is not self.get_transaction():
1081
1558
# has an unlock or relock occured ?
1082
1559
raise errors.BzrError('mismatched lock context %r and '
1083
1560
'write group %r.' %
1084
1561
(self.get_transaction(), self._write_group))
1085
self._commit_write_group()
1562
result = self._commit_write_group()
1086
1563
self._write_group = None
1088
1566
def _commit_write_group(self):
1089
1567
"""Template method for per-repository write group cleanup.
1091
This is called before the write group is considered to be
1569
This is called before the write group is considered to be
1092
1570
finished and should ensure that all data handed to the repository
1093
for writing during the write group is safely committed (to the
1571
for writing during the write group is safely committed (to the
1094
1572
extent possible considering file system caching etc).
1097
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False):
1575
def suspend_write_group(self):
1576
raise errors.UnsuspendableWriteGroup(self)
1578
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1579
"""Return the keys of missing inventory parents for revisions added in
1582
A revision is not complete if the inventory delta for that revision
1583
cannot be calculated. Therefore if the parent inventories of a
1584
revision are not present, the revision is incomplete, and e.g. cannot
1585
be streamed by a smart server. This method finds missing inventory
1586
parents for revisions added in this write group.
1588
if not self._format.supports_external_lookups:
1589
# This is only an issue for stacked repositories
1591
if not self.is_in_write_group():
1592
raise AssertionError('not in a write group')
1594
# XXX: We assume that every added revision already has its
1595
# corresponding inventory, so we only check for parent inventories that
1596
# might be missing, rather than all inventories.
1597
parents = set(self.revisions._index.get_missing_parents())
1598
parents.discard(_mod_revision.NULL_REVISION)
1599
unstacked_inventories = self.inventories._index
1600
present_inventories = unstacked_inventories.get_parent_map(
1601
key[-1:] for key in parents)
1602
parents.difference_update(present_inventories)
1603
if len(parents) == 0:
1604
# No missing parent inventories.
1606
if not check_for_missing_texts:
1607
return set(('inventories', rev_id) for (rev_id,) in parents)
1608
# Ok, now we have a list of missing inventories. But these only matter
1609
# if the inventories that reference them are missing some texts they
1610
# appear to introduce.
1611
# XXX: Texts referenced by all added inventories need to be present,
1612
# but at the moment we're only checking for texts referenced by
1613
# inventories at the graph's edge.
1614
key_deps = self.revisions._index._key_dependencies
1615
key_deps.satisfy_refs_for_keys(present_inventories)
1616
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1617
file_ids = self.fileids_altered_by_revision_ids(referrers)
1618
missing_texts = set()
1619
for file_id, version_ids in file_ids.iteritems():
1620
missing_texts.update(
1621
(file_id, version_id) for version_id in version_ids)
1622
present_texts = self.texts.get_parent_map(missing_texts)
1623
missing_texts.difference_update(present_texts)
1624
if not missing_texts:
1625
# No texts are missing, so all revisions and their deltas are
1628
# Alternatively the text versions could be returned as the missing
1629
# keys, but this is likely to be less data.
1630
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1633
def refresh_data(self):
1634
"""Re-read any data needed to to synchronise with disk.
1636
This method is intended to be called after another repository instance
1637
(such as one used by a smart server) has inserted data into the
1638
repository. It may not be called during a write group, but may be
1639
called at any other time.
1641
if self.is_in_write_group():
1642
raise errors.InternalBzrError(
1643
"May not refresh_data while in a write group.")
1644
self._refresh_data()
1646
def resume_write_group(self, tokens):
1647
if not self.is_write_locked():
1648
raise errors.NotWriteLocked(self)
1649
if self._write_group:
1650
raise errors.BzrError('already in a write group')
1651
self._resume_write_group(tokens)
1652
# so we can detect unlock/relock - the write group is now entered.
1653
self._write_group = self.get_transaction()
1655
def _resume_write_group(self, tokens):
1656
raise errors.UnsuspendableWriteGroup(self)
1658
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1098
1660
"""Fetch the content required to construct revision_id from source.
1100
If revision_id is None all content is copied.
1662
If revision_id is None and fetch_spec is None, then all content is
1665
fetch() may not be used when the repository is in a write group -
1666
either finish the current write group before using fetch, or use
1667
fetch before starting the write group.
1101
1669
:param find_ghosts: Find and copy revisions in the source that are
1102
1670
ghosts in the target (and not reachable directly by walking out to
1103
1671
the first-present revision in target from revision_id).
1672
:param revision_id: If specified, all the content needed for this
1673
revision ID will be copied to the target. Fetch will determine for
1674
itself which content needs to be copied.
1675
:param fetch_spec: If specified, a SearchResult or
1676
PendingAncestryResult that describes which revisions to copy. This
1677
allows copying multiple heads at once. Mutually exclusive with
1680
if fetch_spec is not None and revision_id is not None:
1681
raise AssertionError(
1682
"fetch_spec and revision_id are mutually exclusive.")
1683
if self.is_in_write_group():
1684
raise errors.InternalBzrError(
1685
"May not fetch while in a write group.")
1105
1686
# fast path same-url fetch operations
1106
if self.has_same_location(source):
1687
# TODO: lift out to somewhere common with RemoteRepository
1688
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1689
if (self.has_same_location(source)
1690
and fetch_spec is None
1691
and self._has_same_fallbacks(source)):
1107
1692
# check that last_revision is in 'from' and then return a
1108
1693
# no-operation.
1109
1694
if (revision_id is not None and
2909
3723
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):
3726
class InterDifferingSerializer(InterRepository):
3170
3729
def _get_repo_format_to_test(self):
3174
3733
def is_compatible(source, target):
3175
3734
"""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)):
3735
# This is redundant with format.check_conversion_target(), however that
3736
# raises an exception, and we just want to say "False" as in we won't
3737
# support converting between these formats.
3738
if 'IDS_never' in debug.debug_flags:
3740
if source.supports_rich_root() and not target.supports_rich_root():
3742
if (source._format.supports_tree_reference
3743
and not target._format.supports_tree_reference):
3745
if target._fallback_repositories and target._format.supports_chks:
3746
# IDS doesn't know how to copy CHKs for the parent inventories it
3747
# adds to stacked repos.
3749
if 'IDS_always' in debug.debug_flags:
3751
# Only use this code path for local source and target. IDS does far
3752
# too much IO (both bandwidth and roundtrips) over a network.
3753
if not source.bzrdir.transport.base.startswith('file:///'):
3755
if not target.bzrdir.transport.base.startswith('file:///'):
3185
def _fetch_batch(self, revision_ids, basis_id, basis_tree):
3759
def _get_trees(self, revision_ids, cache):
3761
for rev_id in revision_ids:
3763
possible_trees.append((rev_id, cache[rev_id]))
3765
# Not cached, but inventory might be present anyway.
3767
tree = self.source.revision_tree(rev_id)
3768
except errors.NoSuchRevision:
3769
# Nope, parent is ghost.
3772
cache[rev_id] = tree
3773
possible_trees.append((rev_id, tree))
3774
return possible_trees
3776
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3777
"""Get the best delta and base for this revision.
3779
:return: (basis_id, delta)
3782
# Generate deltas against each tree, to find the shortest.
3783
texts_possibly_new_in_tree = set()
3784
for basis_id, basis_tree in possible_trees:
3785
delta = tree.inventory._make_delta(basis_tree.inventory)
3786
for old_path, new_path, file_id, new_entry in delta:
3787
if new_path is None:
3788
# This file_id isn't present in the new rev, so we don't
3792
# Rich roots are handled elsewhere...
3794
kind = new_entry.kind
3795
if kind != 'directory' and kind != 'file':
3796
# No text record associated with this inventory entry.
3798
# This is a directory or file that has changed somehow.
3799
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3800
deltas.append((len(delta), basis_id, delta))
3802
return deltas[0][1:]
3804
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3805
"""Find all parent revisions that are absent, but for which the
3806
inventory is present, and copy those inventories.
3808
This is necessary to preserve correctness when the source is stacked
3809
without fallbacks configured. (Note that in cases like upgrade the
3810
source may be not have _fallback_repositories even though it is
3814
for parents in parent_map.values():
3815
parent_revs.update(parents)
3816
present_parents = self.source.get_parent_map(parent_revs)
3817
absent_parents = set(parent_revs).difference(present_parents)
3818
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3819
(rev_id,) for rev_id in absent_parents)
3820
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3821
for parent_tree in self.source.revision_trees(parent_inv_ids):
3822
current_revision_id = parent_tree.get_revision_id()
3823
parents_parents_keys = parent_invs_keys_for_stacking[
3824
(current_revision_id,)]
3825
parents_parents = [key[-1] for key in parents_parents_keys]
3826
basis_id = _mod_revision.NULL_REVISION
3827
basis_tree = self.source.revision_tree(basis_id)
3828
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3829
self.target.add_inventory_by_delta(
3830
basis_id, delta, current_revision_id, parents_parents)
3831
cache[current_revision_id] = parent_tree
3833
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3186
3834
"""Fetch across a few revisions.
3188
3836
: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
3837
:param basis_id: The revision_id of a tree that must be in cache, used
3838
as a basis for delta when no other base is available
3839
:param cache: A cache of RevisionTrees that we can use.
3840
:param a_graph: A Graph object to determine the heads() of the
3841
rich-root data stream.
3842
:return: The revision_id of the last converted tree. The RevisionTree
3843
for it will be in cache
3195
3845
# Walk though all revisions; get inventory deltas, copy referenced
3196
3846
# texts that delta references, insert the delta, revision and
3848
root_keys_to_create = set()
3198
3849
text_keys = set()
3199
3850
pending_deltas = []
3200
3851
pending_revisions = []
3852
parent_map = self.source.get_parent_map(revision_ids)
3853
self._fetch_parent_invs_for_stacking(parent_map, cache)
3854
self.source._safe_to_return_from_cache = True
3201
3855
for tree in self.source.revision_trees(revision_ids):
3856
# Find a inventory delta for this revision.
3857
# Find text entries that need to be copied, too.
3202
3858
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))
3859
parent_ids = parent_map.get(current_revision_id, ())
3860
parent_trees = self._get_trees(parent_ids, cache)
3861
possible_trees = list(parent_trees)
3862
if len(possible_trees) == 0:
3863
# There either aren't any parents, or the parents are ghosts,
3864
# so just use the last converted tree.
3865
possible_trees.append((basis_id, cache[basis_id]))
3866
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3215
3868
revision = self.source.get_revision(current_revision_id)
3216
3869
pending_deltas.append((basis_id, delta,
3217
3870
current_revision_id, revision.parent_ids))
3871
if self._converting_to_rich_root:
3872
self._revision_id_to_root_id[current_revision_id] = \
3874
# Determine which texts are in present in this revision but not in
3875
# any of the available parents.
3876
texts_possibly_new_in_tree = set()
3877
for old_path, new_path, file_id, entry in delta:
3878
if new_path is None:
3879
# This file_id isn't present in the new rev
3883
if not self.target.supports_rich_root():
3884
# The target doesn't support rich root, so we don't
3887
if self._converting_to_rich_root:
3888
# This can't be copied normally, we have to insert
3890
root_keys_to_create.add((file_id, entry.revision))
3893
texts_possibly_new_in_tree.add((file_id, entry.revision))
3894
for basis_id, basis_tree in possible_trees:
3895
basis_inv = basis_tree.inventory
3896
for file_key in list(texts_possibly_new_in_tree):
3897
file_id, file_revision = file_key
3899
entry = basis_inv[file_id]
3900
except errors.NoSuchId:
3902
if entry.revision == file_revision:
3903
texts_possibly_new_in_tree.remove(file_key)
3904
text_keys.update(texts_possibly_new_in_tree)
3218
3905
pending_revisions.append(revision)
3906
cache[current_revision_id] = tree
3219
3907
basis_id = current_revision_id
3908
self.source._safe_to_return_from_cache = False
3221
3909
# Copy file texts
3222
3910
from_texts = self.source.texts
3223
3911
to_texts = self.target.texts
3912
if root_keys_to_create:
3913
root_stream = _mod_fetch._new_root_data_stream(
3914
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3915
self.source, graph=a_graph)
3916
to_texts.insert_record_stream(root_stream)
3224
3917
to_texts.insert_record_stream(from_texts.get_record_stream(
3225
text_keys, self.target._fetch_order,
3226
not self.target._fetch_uses_deltas))
3918
text_keys, self.target._format._fetch_order,
3919
not self.target._format._fetch_uses_deltas))
3920
# insert inventory deltas
3228
3921
for delta in pending_deltas:
3229
3922
self.target.add_inventory_by_delta(*delta)
3923
if self.target._fallback_repositories:
3924
# Make sure this stacked repository has all the parent inventories
3925
# for the new revisions that we are about to insert. We do this
3926
# before adding the revisions so that no revision is added until
3927
# all the inventories it may depend on are added.
3928
# Note that this is overzealous, as we may have fetched these in an
3931
revision_ids = set()
3932
for revision in pending_revisions:
3933
revision_ids.add(revision.revision_id)
3934
parent_ids.update(revision.parent_ids)
3935
parent_ids.difference_update(revision_ids)
3936
parent_ids.discard(_mod_revision.NULL_REVISION)
3937
parent_map = self.source.get_parent_map(parent_ids)
3938
# we iterate over parent_map and not parent_ids because we don't
3939
# want to try copying any revision which is a ghost
3940
for parent_tree in self.source.revision_trees(parent_map):
3941
current_revision_id = parent_tree.get_revision_id()
3942
parents_parents = parent_map[current_revision_id]
3943
possible_trees = self._get_trees(parents_parents, cache)
3944
if len(possible_trees) == 0:
3945
# There either aren't any parents, or the parents are
3946
# ghosts, so just use the last converted tree.
3947
possible_trees.append((basis_id, cache[basis_id]))
3948
basis_id, delta = self._get_delta_for_revision(parent_tree,
3949
parents_parents, possible_trees)
3950
self.target.add_inventory_by_delta(
3951
basis_id, delta, current_revision_id, parents_parents)
3230
3952
# insert signatures and revisions
3231
3953
for revision in pending_revisions:
3309
4063
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
4066
InterRepository.register_optimiser(InterDifferingSerializer)
3425
4067
InterRepository.register_optimiser(InterSameDataRepository)
3426
4068
InterRepository.register_optimiser(InterWeaveRepo)
3427
4069
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
4072
class CopyConverter(object):
3437
4073
"""A repository conversion tool which just performs a copy of the content.
3439
4075
This is slow but quite reliable.
3587
4232
revision_graph[key] = tuple(parent for parent in parents if parent
3588
4233
in revision_graph)
3589
4234
return revision_graph
4237
class StreamSink(object):
4238
"""An object that can insert a stream into a repository.
4240
This interface handles the complexity of reserialising inventories and
4241
revisions from different formats, and allows unidirectional insertion into
4242
stacked repositories without looking for the missing basis parents
4246
def __init__(self, target_repo):
4247
self.target_repo = target_repo
4249
def insert_stream(self, stream, src_format, resume_tokens):
4250
"""Insert a stream's content into the target repository.
4252
:param src_format: a bzr repository format.
4254
:return: a list of resume tokens and an iterable of keys additional
4255
items required before the insertion can be completed.
4257
self.target_repo.lock_write()
4260
self.target_repo.resume_write_group(resume_tokens)
4263
self.target_repo.start_write_group()
4266
# locked_insert_stream performs a commit|suspend.
4267
return self._locked_insert_stream(stream, src_format, is_resume)
4269
self.target_repo.abort_write_group(suppress_errors=True)
4272
self.target_repo.unlock()
4274
def _locked_insert_stream(self, stream, src_format, is_resume):
4275
to_serializer = self.target_repo._format._serializer
4276
src_serializer = src_format._serializer
4278
if to_serializer == src_serializer:
4279
# If serializers match and the target is a pack repository, set the
4280
# write cache size on the new pack. This avoids poor performance
4281
# on transports where append is unbuffered (such as
4282
# RemoteTransport). This is safe to do because nothing should read
4283
# back from the target repository while a stream with matching
4284
# serialization is being inserted.
4285
# The exception is that a delta record from the source that should
4286
# be a fulltext may need to be expanded by the target (see
4287
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4288
# explicitly flush any buffered writes first in that rare case.
4290
new_pack = self.target_repo._pack_collection._new_pack
4291
except AttributeError:
4292
# Not a pack repository
4295
new_pack.set_write_cache_size(1024*1024)
4296
for substream_type, substream in stream:
4297
if 'stream' in debug.debug_flags:
4298
mutter('inserting substream: %s', substream_type)
4299
if substream_type == 'texts':
4300
self.target_repo.texts.insert_record_stream(substream)
4301
elif substream_type == 'inventories':
4302
if src_serializer == to_serializer:
4303
self.target_repo.inventories.insert_record_stream(
4306
self._extract_and_insert_inventories(
4307
substream, src_serializer)
4308
elif substream_type == 'inventory-deltas':
4309
self._extract_and_insert_inventory_deltas(
4310
substream, src_serializer)
4311
elif substream_type == 'chk_bytes':
4312
# XXX: This doesn't support conversions, as it assumes the
4313
# conversion was done in the fetch code.
4314
self.target_repo.chk_bytes.insert_record_stream(substream)
4315
elif substream_type == 'revisions':
4316
# This may fallback to extract-and-insert more often than
4317
# required if the serializers are different only in terms of
4319
if src_serializer == to_serializer:
4320
self.target_repo.revisions.insert_record_stream(
4323
self._extract_and_insert_revisions(substream,
4325
elif substream_type == 'signatures':
4326
self.target_repo.signatures.insert_record_stream(substream)
4328
raise AssertionError('kaboom! %s' % (substream_type,))
4329
# Done inserting data, and the missing_keys calculations will try to
4330
# read back from the inserted data, so flush the writes to the new pack
4331
# (if this is pack format).
4332
if new_pack is not None:
4333
new_pack._write_data('', flush=True)
4334
# Find all the new revisions (including ones from resume_tokens)
4335
missing_keys = self.target_repo.get_missing_parent_inventories(
4336
check_for_missing_texts=is_resume)
4338
for prefix, versioned_file in (
4339
('texts', self.target_repo.texts),
4340
('inventories', self.target_repo.inventories),
4341
('revisions', self.target_repo.revisions),
4342
('signatures', self.target_repo.signatures),
4343
('chk_bytes', self.target_repo.chk_bytes),
4345
if versioned_file is None:
4347
# TODO: key is often going to be a StaticTuple object
4348
# I don't believe we can define a method by which
4349
# (prefix,) + StaticTuple will work, though we could
4350
# define a StaticTuple.sq_concat that would allow you to
4351
# pass in either a tuple or a StaticTuple as the second
4352
# object, so instead we could have:
4353
# StaticTuple(prefix) + key here...
4354
missing_keys.update((prefix,) + key for key in
4355
versioned_file.get_missing_compression_parent_keys())
4356
except NotImplementedError:
4357
# cannot even attempt suspending, and missing would have failed
4358
# during stream insertion.
4359
missing_keys = set()
4362
# suspend the write group and tell the caller what we is
4363
# missing. We know we can suspend or else we would not have
4364
# entered this code path. (All repositories that can handle
4365
# missing keys can handle suspending a write group).
4366
write_group_tokens = self.target_repo.suspend_write_group()
4367
return write_group_tokens, missing_keys
4368
hint = self.target_repo.commit_write_group()
4369
if (to_serializer != src_serializer and
4370
self.target_repo._format.pack_compresses):
4371
self.target_repo.pack(hint=hint)
4374
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4375
target_rich_root = self.target_repo._format.rich_root_data
4376
target_tree_refs = self.target_repo._format.supports_tree_reference
4377
for record in substream:
4378
# Insert the delta directly
4379
inventory_delta_bytes = record.get_bytes_as('fulltext')
4380
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4382
parse_result = deserialiser.parse_text_bytes(
4383
inventory_delta_bytes)
4384
except inventory_delta.IncompatibleInventoryDelta, err:
4385
trace.mutter("Incompatible delta: %s", err.msg)
4386
raise errors.IncompatibleRevision(self.target_repo._format)
4387
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4388
revision_id = new_id
4389
parents = [key[0] for key in record.parents]
4390
self.target_repo.add_inventory_by_delta(
4391
basis_id, inv_delta, revision_id, parents)
4393
def _extract_and_insert_inventories(self, substream, serializer,
4395
"""Generate a new inventory versionedfile in target, converting data.
4397
The inventory is retrieved from the source, (deserializing it), and
4398
stored in the target (reserializing it in a different format).
4400
target_rich_root = self.target_repo._format.rich_root_data
4401
target_tree_refs = self.target_repo._format.supports_tree_reference
4402
for record in substream:
4403
# It's not a delta, so it must be a fulltext in the source
4404
# serializer's format.
4405
bytes = record.get_bytes_as('fulltext')
4406
revision_id = record.key[0]
4407
inv = serializer.read_inventory_from_string(bytes, revision_id)
4408
parents = [key[0] for key in record.parents]
4409
self.target_repo.add_inventory(revision_id, inv, parents)
4410
# No need to keep holding this full inv in memory when the rest of
4411
# the substream is likely to be all deltas.
4414
def _extract_and_insert_revisions(self, substream, serializer):
4415
for record in substream:
4416
bytes = record.get_bytes_as('fulltext')
4417
revision_id = record.key[0]
4418
rev = serializer.read_revision_from_string(bytes)
4419
if rev.revision_id != revision_id:
4420
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4421
self.target_repo.add_revision(revision_id, rev)
4424
if self.target_repo._format._fetch_reconcile:
4425
self.target_repo.reconcile()
4428
class StreamSource(object):
4429
"""A source of a stream for fetching between repositories."""
4431
def __init__(self, from_repository, to_format):
4432
"""Create a StreamSource streaming from from_repository."""
4433
self.from_repository = from_repository
4434
self.to_format = to_format
4436
def delta_on_metadata(self):
4437
"""Return True if delta's are permitted on metadata streams.
4439
That is on revisions and signatures.
4441
src_serializer = self.from_repository._format._serializer
4442
target_serializer = self.to_format._serializer
4443
return (self.to_format._fetch_uses_deltas and
4444
src_serializer == target_serializer)
4446
def _fetch_revision_texts(self, revs):
4447
# fetch signatures first and then the revision texts
4448
# may need to be a InterRevisionStore call here.
4449
from_sf = self.from_repository.signatures
4450
# A missing signature is just skipped.
4451
keys = [(rev_id,) for rev_id in revs]
4452
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4454
self.to_format._fetch_order,
4455
not self.to_format._fetch_uses_deltas))
4456
# If a revision has a delta, this is actually expanded inside the
4457
# insert_record_stream code now, which is an alternate fix for
4459
from_rf = self.from_repository.revisions
4460
revisions = from_rf.get_record_stream(
4462
self.to_format._fetch_order,
4463
not self.delta_on_metadata())
4464
return [('signatures', signatures), ('revisions', revisions)]
4466
def _generate_root_texts(self, revs):
4467
"""This will be called by get_stream between fetching weave texts and
4468
fetching the inventory weave.
4470
if self._rich_root_upgrade():
4471
return _mod_fetch.Inter1and2Helper(
4472
self.from_repository).generate_root_texts(revs)
4476
def get_stream(self, search):
4478
revs = search.get_keys()
4479
graph = self.from_repository.get_graph()
4480
revs = tsort.topo_sort(graph.get_parent_map(revs))
4481
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4483
for knit_kind, file_id, revisions in data_to_fetch:
4484
if knit_kind != phase:
4486
# Make a new progress bar for this phase
4487
if knit_kind == "file":
4488
# Accumulate file texts
4489
text_keys.extend([(file_id, revision) for revision in
4491
elif knit_kind == "inventory":
4492
# Now copy the file texts.
4493
from_texts = self.from_repository.texts
4494
yield ('texts', from_texts.get_record_stream(
4495
text_keys, self.to_format._fetch_order,
4496
not self.to_format._fetch_uses_deltas))
4497
# Cause an error if a text occurs after we have done the
4500
# Before we process the inventory we generate the root
4501
# texts (if necessary) so that the inventories references
4503
for _ in self._generate_root_texts(revs):
4505
# we fetch only the referenced inventories because we do not
4506
# know for unselected inventories whether all their required
4507
# texts are present in the other repository - it could be
4509
for info in self._get_inventory_stream(revs):
4511
elif knit_kind == "signatures":
4512
# Nothing to do here; this will be taken care of when
4513
# _fetch_revision_texts happens.
4515
elif knit_kind == "revisions":
4516
for record in self._fetch_revision_texts(revs):
4519
raise AssertionError("Unknown knit kind %r" % knit_kind)
4521
def get_stream_for_missing_keys(self, missing_keys):
4522
# missing keys can only occur when we are byte copying and not
4523
# translating (because translation means we don't send
4524
# unreconstructable deltas ever).
4526
keys['texts'] = set()
4527
keys['revisions'] = set()
4528
keys['inventories'] = set()
4529
keys['chk_bytes'] = set()
4530
keys['signatures'] = set()
4531
for key in missing_keys:
4532
keys[key[0]].add(key[1:])
4533
if len(keys['revisions']):
4534
# If we allowed copying revisions at this point, we could end up
4535
# copying a revision without copying its required texts: a
4536
# violation of the requirements for repository integrity.
4537
raise AssertionError(
4538
'cannot copy revisions to fill in missing deltas %s' % (
4539
keys['revisions'],))
4540
for substream_kind, keys in keys.iteritems():
4541
vf = getattr(self.from_repository, substream_kind)
4542
if vf is None and keys:
4543
raise AssertionError(
4544
"cannot fill in keys for a versioned file we don't"
4545
" have: %s needs %s" % (substream_kind, keys))
4547
# No need to stream something we don't have
4549
if substream_kind == 'inventories':
4550
# Some missing keys are genuinely ghosts, filter those out.
4551
present = self.from_repository.inventories.get_parent_map(keys)
4552
revs = [key[0] for key in present]
4553
# Get the inventory stream more-or-less as we do for the
4554
# original stream; there's no reason to assume that records
4555
# direct from the source will be suitable for the sink. (Think
4556
# e.g. 2a -> 1.9-rich-root).
4557
for info in self._get_inventory_stream(revs, missing=True):
4561
# Ask for full texts always so that we don't need more round trips
4562
# after this stream.
4563
# Some of the missing keys are genuinely ghosts, so filter absent
4564
# records. The Sink is responsible for doing another check to
4565
# ensure that ghosts don't introduce missing data for future
4567
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4568
self.to_format._fetch_order, True))
4569
yield substream_kind, stream
4571
def inventory_fetch_order(self):
4572
if self._rich_root_upgrade():
4573
return 'topological'
4575
return self.to_format._fetch_order
4577
def _rich_root_upgrade(self):
4578
return (not self.from_repository._format.rich_root_data and
4579
self.to_format.rich_root_data)
4581
def _get_inventory_stream(self, revision_ids, missing=False):
4582
from_format = self.from_repository._format
4583
if (from_format.supports_chks and self.to_format.supports_chks and
4584
from_format.network_name() == self.to_format.network_name()):
4585
raise AssertionError(
4586
"this case should be handled by GroupCHKStreamSource")
4587
elif 'forceinvdeltas' in debug.debug_flags:
4588
return self._get_convertable_inventory_stream(revision_ids,
4589
delta_versus_null=missing)
4590
elif from_format.network_name() == self.to_format.network_name():
4592
return self._get_simple_inventory_stream(revision_ids,
4594
elif (not from_format.supports_chks and not self.to_format.supports_chks
4595
and from_format._serializer == self.to_format._serializer):
4596
# Essentially the same format.
4597
return self._get_simple_inventory_stream(revision_ids,
4600
# Any time we switch serializations, we want to use an
4601
# inventory-delta based approach.
4602
return self._get_convertable_inventory_stream(revision_ids,
4603
delta_versus_null=missing)
4605
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4606
# NB: This currently reopens the inventory weave in source;
4607
# using a single stream interface instead would avoid this.
4608
from_weave = self.from_repository.inventories
4610
delta_closure = True
4612
delta_closure = not self.delta_on_metadata()
4613
yield ('inventories', from_weave.get_record_stream(
4614
[(rev_id,) for rev_id in revision_ids],
4615
self.inventory_fetch_order(), delta_closure))
4617
def _get_convertable_inventory_stream(self, revision_ids,
4618
delta_versus_null=False):
4619
# The source is using CHKs, but the target either doesn't or it has a
4620
# different serializer. The StreamSink code expects to be able to
4621
# convert on the target, so we need to put bytes-on-the-wire that can
4622
# be converted. That means inventory deltas (if the remote is <1.19,
4623
# RemoteStreamSink will fallback to VFS to insert the deltas).
4624
yield ('inventory-deltas',
4625
self._stream_invs_as_deltas(revision_ids,
4626
delta_versus_null=delta_versus_null))
4628
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4629
"""Return a stream of inventory-deltas for the given rev ids.
4631
:param revision_ids: The list of inventories to transmit
4632
:param delta_versus_null: Don't try to find a minimal delta for this
4633
entry, instead compute the delta versus the NULL_REVISION. This
4634
effectively streams a complete inventory. Used for stuff like
4635
filling in missing parents, etc.
4637
from_repo = self.from_repository
4638
revision_keys = [(rev_id,) for rev_id in revision_ids]
4639
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4640
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4642
inventories = self.from_repository.iter_inventories(
4643
revision_ids, 'topological')
4644
format = from_repo._format
4645
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4646
inventory_cache = lru_cache.LRUCache(50)
4647
null_inventory = from_repo.revision_tree(
4648
_mod_revision.NULL_REVISION).inventory
4649
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4650
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4651
# repo back into a non-rich-root repo ought to be allowed)
4652
serializer = inventory_delta.InventoryDeltaSerializer(
4653
versioned_root=format.rich_root_data,
4654
tree_references=format.supports_tree_reference)
4655
for inv in inventories:
4656
key = (inv.revision_id,)
4657
parent_keys = parent_map.get(key, ())
4659
if not delta_versus_null and parent_keys:
4660
# The caller did not ask for complete inventories and we have
4661
# some parents that we can delta against. Make a delta against
4662
# each parent so that we can find the smallest.
4663
parent_ids = [parent_key[0] for parent_key in parent_keys]
4664
for parent_id in parent_ids:
4665
if parent_id not in invs_sent_so_far:
4666
# We don't know that the remote side has this basis, so
4669
if parent_id == _mod_revision.NULL_REVISION:
4670
parent_inv = null_inventory
4672
parent_inv = inventory_cache.get(parent_id, None)
4673
if parent_inv is None:
4674
parent_inv = from_repo.get_inventory(parent_id)
4675
candidate_delta = inv._make_delta(parent_inv)
4676
if (delta is None or
4677
len(delta) > len(candidate_delta)):
4678
delta = candidate_delta
4679
basis_id = parent_id
4681
# Either none of the parents ended up being suitable, or we
4682
# were asked to delta against NULL
4683
basis_id = _mod_revision.NULL_REVISION
4684
delta = inv._make_delta(null_inventory)
4685
invs_sent_so_far.add(inv.revision_id)
4686
inventory_cache[inv.revision_id] = inv
4687
delta_serialized = ''.join(
4688
serializer.delta_to_lines(basis_id, key[-1], delta))
4689
yield versionedfile.FulltextContentFactory(
4690
key, parent_keys, None, delta_serialized)
4693
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4694
stop_revision=None):
4695
"""Extend the partial history to include a given index
4697
If a stop_index is supplied, stop when that index has been reached.
4698
If a stop_revision is supplied, stop when that revision is
4699
encountered. Otherwise, stop when the beginning of history is
4702
:param stop_index: The index which should be present. When it is
4703
present, history extension will stop.
4704
:param stop_revision: The revision id which should be present. When
4705
it is encountered, history extension will stop.
4707
start_revision = partial_history_cache[-1]
4708
iterator = repo.iter_reverse_revision_history(start_revision)
4710
#skip the last revision in the list
4713
if (stop_index is not None and
4714
len(partial_history_cache) > stop_index):
4716
if partial_history_cache[-1] == stop_revision:
4718
revision_id = iterator.next()
4719
partial_history_cache.append(revision_id)
4720
except StopIteration:
added
removed
bzrlib/repository.py
