547
417
ie.revision = parent_entry.revision
548
418
return self._get_delta(ie, basis_inv, path), False, None
549
419
ie.reference_revision = content_summary[3]
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)
421
self._add_text_to_weave(ie.file_id, lines, heads, None)
555
423
raise NotImplementedError('unknown kind')
556
424
ie.revision = self._new_revision_id
557
self._any_changes = True
558
425
return self._get_delta(ie, basis_inv, path), True, fingerprint
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]
427
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
428
# Note: as we read the content directly from the tree, we know its not
429
# been turned into unicode or badly split - but a broken tree
430
# implementation could give us bad output from readlines() so this is
431
# not a guarantee of safety. What would be better is always checking
432
# the content during test suite execution. RBC 20070912
433
parent_keys = tuple((file_id, parent) for parent in parents)
434
return self.repository.texts.add_lines(
435
(file_id, self._new_revision_id), parent_keys, new_lines,
436
nostore_sha=nostore_sha, random_id=self.random_revid,
437
check_content=False)[0:2]
836
440
class RootCommitBuilder(CommitBuilder):
837
441
"""This commitbuilder actually records the root id"""
839
443
# the root entry gets versioned properly by this builder.
840
444
_versioned_root = True
1167
670
# The old API returned a list, should this actually be a set?
1168
671
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]))
1284
674
def create(a_bzrdir):
1285
675
"""Construct the current default format repository in a_bzrdir."""
1552
947
"""Commit the contents accrued within the current write group.
1554
949
:seealso: start_write_group.
1556
:return: it may return an opaque hint that can be passed to 'pack'.
1558
951
if self._write_group is not self.get_transaction():
1559
952
# has an unlock or relock occured ?
1560
953
raise errors.BzrError('mismatched lock context %r and '
1561
954
'write group %r.' %
1562
955
(self.get_transaction(), self._write_group))
1563
result = self._commit_write_group()
956
self._commit_write_group()
1564
957
self._write_group = None
1567
959
def _commit_write_group(self):
1568
960
"""Template method for per-repository write group cleanup.
1570
This is called before the write group is considered to be
962
This is called before the write group is considered to be
1571
963
finished and should ensure that all data handed to the repository
1572
for writing during the write group is safely committed (to the
964
for writing during the write group is safely committed (to the
1573
965
extent possible considering file system caching etc).
1576
def suspend_write_group(self):
1577
raise errors.UnsuspendableWriteGroup(self)
1579
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1580
"""Return the keys of missing inventory parents for revisions added in
1583
A revision is not complete if the inventory delta for that revision
1584
cannot be calculated. Therefore if the parent inventories of a
1585
revision are not present, the revision is incomplete, and e.g. cannot
1586
be streamed by a smart server. This method finds missing inventory
1587
parents for revisions added in this write group.
1589
if not self._format.supports_external_lookups:
1590
# This is only an issue for stacked repositories
1592
if not self.is_in_write_group():
1593
raise AssertionError('not in a write group')
1595
# XXX: We assume that every added revision already has its
1596
# corresponding inventory, so we only check for parent inventories that
1597
# might be missing, rather than all inventories.
1598
parents = set(self.revisions._index.get_missing_parents())
1599
parents.discard(_mod_revision.NULL_REVISION)
1600
unstacked_inventories = self.inventories._index
1601
present_inventories = unstacked_inventories.get_parent_map(
1602
key[-1:] for key in parents)
1603
parents.difference_update(present_inventories)
1604
if len(parents) == 0:
1605
# No missing parent inventories.
1607
if not check_for_missing_texts:
1608
return set(('inventories', rev_id) for (rev_id,) in parents)
1609
# Ok, now we have a list of missing inventories. But these only matter
1610
# if the inventories that reference them are missing some texts they
1611
# appear to introduce.
1612
# XXX: Texts referenced by all added inventories need to be present,
1613
# but at the moment we're only checking for texts referenced by
1614
# inventories at the graph's edge.
1615
key_deps = self.revisions._index._key_dependencies
1616
key_deps.satisfy_refs_for_keys(present_inventories)
1617
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1618
file_ids = self.fileids_altered_by_revision_ids(referrers)
1619
missing_texts = set()
1620
for file_id, version_ids in file_ids.iteritems():
1621
missing_texts.update(
1622
(file_id, version_id) for version_id in version_ids)
1623
present_texts = self.texts.get_parent_map(missing_texts)
1624
missing_texts.difference_update(present_texts)
1625
if not missing_texts:
1626
# No texts are missing, so all revisions and their deltas are
1629
# Alternatively the text versions could be returned as the missing
1630
# keys, but this is likely to be less data.
1631
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1634
def refresh_data(self):
1635
"""Re-read any data needed to to synchronise with disk.
1637
This method is intended to be called after another repository instance
1638
(such as one used by a smart server) has inserted data into the
1639
repository. It may not be called during a write group, but may be
1640
called at any other time.
1642
if self.is_in_write_group():
1643
raise errors.InternalBzrError(
1644
"May not refresh_data while in a write group.")
1645
self._refresh_data()
1647
def resume_write_group(self, tokens):
1648
if not self.is_write_locked():
1649
raise errors.NotWriteLocked(self)
1650
if self._write_group:
1651
raise errors.BzrError('already in a write group')
1652
self._resume_write_group(tokens)
1653
# so we can detect unlock/relock - the write group is now entered.
1654
self._write_group = self.get_transaction()
1656
def _resume_write_group(self, tokens):
1657
raise errors.UnsuspendableWriteGroup(self)
1659
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
968
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False):
1661
969
"""Fetch the content required to construct revision_id from source.
1663
If revision_id is None and fetch_spec is None, then all content is
1666
fetch() may not be used when the repository is in a write group -
1667
either finish the current write group before using fetch, or use
1668
fetch before starting the write group.
971
If revision_id is None all content is copied.
1670
972
:param find_ghosts: Find and copy revisions in the source that are
1671
973
ghosts in the target (and not reachable directly by walking out to
1672
974
the first-present revision in target from revision_id).
1673
:param revision_id: If specified, all the content needed for this
1674
revision ID will be copied to the target. Fetch will determine for
1675
itself which content needs to be copied.
1676
:param fetch_spec: If specified, a SearchResult or
1677
PendingAncestryResult that describes which revisions to copy. This
1678
allows copying multiple heads at once. Mutually exclusive with
1681
if fetch_spec is not None and revision_id is not None:
1682
raise AssertionError(
1683
"fetch_spec and revision_id are mutually exclusive.")
1684
if self.is_in_write_group():
1685
raise errors.InternalBzrError(
1686
"May not fetch while in a write group.")
1687
976
# fast path same-url fetch operations
1688
# TODO: lift out to somewhere common with RemoteRepository
1689
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1690
if (self.has_same_location(source)
1691
and fetch_spec is None
1692
and self._has_same_fallbacks(source)):
977
if self.has_same_location(source):
1693
978
# check that last_revision is in 'from' and then return a
1695
980
if (revision_id is not None and
2366
1533
"""Get Inventory object by revision id."""
2367
1534
return self.iter_inventories([revision_id]).next()
2369
def iter_inventories(self, revision_ids, ordering=None):
1536
def iter_inventories(self, revision_ids):
2370
1537
"""Get many inventories by revision_ids.
2372
1539
This will buffer some or all of the texts used in constructing the
2373
1540
inventories in memory, but will only parse a single inventory at a
2376
:param revision_ids: The expected revision ids of the inventories.
2377
:param ordering: optional ordering, e.g. 'topological'. If not
2378
specified, the order of revision_ids will be preserved (by
2379
buffering if necessary).
2380
1543
:return: An iterator of inventories.
2382
1545
if ((None in revision_ids)
2383
1546
or (_mod_revision.NULL_REVISION in revision_ids)):
2384
1547
raise ValueError('cannot get null revision inventory')
2385
return self._iter_inventories(revision_ids, ordering)
1548
return self._iter_inventories(revision_ids)
2387
def _iter_inventories(self, revision_ids, ordering):
1550
def _iter_inventories(self, revision_ids):
2388
1551
"""single-document based inventory iteration."""
2389
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2390
for text, revision_id in inv_xmls:
2391
yield self._deserialise_inventory(revision_id, text)
1552
for text, revision_id in self._iter_inventory_xmls(revision_ids):
1553
yield self.deserialise_inventory(revision_id, text)
2393
def _iter_inventory_xmls(self, revision_ids, ordering):
2394
if ordering is None:
2395
order_as_requested = True
2396
ordering = 'unordered'
2398
order_as_requested = False
1555
def _iter_inventory_xmls(self, revision_ids):
2399
1556
keys = [(revision_id,) for revision_id in revision_ids]
2402
if order_as_requested:
2403
key_iter = iter(keys)
2404
next_key = key_iter.next()
2405
stream = self.inventories.get_record_stream(keys, ordering, True)
1557
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2407
1559
for record in stream:
2408
1560
if record.storage_kind != 'absent':
2409
chunks = record.get_bytes_as('chunked')
2410
if order_as_requested:
2411
text_chunks[record.key] = chunks
2413
yield ''.join(chunks), record.key[-1]
1561
texts[record.key] = record.get_bytes_as('fulltext')
2415
1563
raise errors.NoSuchRevision(self, record.key)
2416
if order_as_requested:
2417
# Yield as many results as we can while preserving order.
2418
while next_key in text_chunks:
2419
chunks = text_chunks.pop(next_key)
2420
yield ''.join(chunks), next_key[-1]
2422
next_key = key_iter.next()
2423
except StopIteration:
2424
# We still want to fully consume the get_record_stream,
2425
# just in case it is not actually finished at this point
1565
yield texts[key], key[-1]
2429
def _deserialise_inventory(self, revision_id, xml):
2430
"""Transform the xml into an inventory object.
1567
def deserialise_inventory(self, revision_id, xml):
1568
"""Transform the xml into an inventory object.
2432
1570
:param revision_id: The expected revision id of the inventory.
2433
1571
:param xml: A serialised inventory.
2435
result = self._serializer.read_inventory_from_string(xml, revision_id,
2436
entry_cache=self._inventory_entry_cache,
2437
return_from_cache=self._safe_to_return_from_cache)
1573
result = self._serializer.read_inventory_from_string(xml, revision_id)
2438
1574
if result.revision_id != revision_id:
2439
1575
raise AssertionError('revision id mismatch %s != %s' % (
2440
1576
result.revision_id, revision_id))
2443
def _serialise_inventory(self, inv):
1579
def serialise_inventory(self, inv):
2444
1580
return self._serializer.write_inventory_to_string(inv)
2446
1582
def _serialise_inventory_to_lines(self, inv):
3718
2732
return self.source.revision_ids_to_search_result(result_set)
3721
class InterDifferingSerializer(InterRepository):
2735
class InterPackRepo(InterSameDataRepository):
2736
"""Optimised code paths between Pack based repositories."""
2739
def _get_repo_format_to_test(self):
2740
from bzrlib.repofmt import pack_repo
2741
return pack_repo.RepositoryFormatKnitPack1()
2744
def is_compatible(source, target):
2745
"""Be compatible with known Pack formats.
2747
We don't test for the stores being of specific types because that
2748
could lead to confusing results, and there is no need to be
2751
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
2753
are_packs = (isinstance(source._format, RepositoryFormatPack) and
2754
isinstance(target._format, RepositoryFormatPack))
2755
except AttributeError:
2757
return are_packs and InterRepository._same_model(source, target)
2760
def fetch(self, revision_id=None, pb=None, find_ghosts=False):
2761
"""See InterRepository.fetch()."""
2762
if (len(self.source._fallback_repositories) > 0 or
2763
len(self.target._fallback_repositories) > 0):
2764
# The pack layer is not aware of fallback repositories, so when
2765
# fetching from a stacked repository or into a stacked repository
2766
# we use the generic fetch logic which uses the VersionedFiles
2767
# attributes on repository.
2768
from bzrlib.fetch import RepoFetcher
2769
fetcher = RepoFetcher(self.target, self.source, revision_id,
2771
return fetcher.count_copied, fetcher.failed_revisions
2772
from bzrlib.repofmt.pack_repo import Packer
2773
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
2774
self.source, self.source._format, self.target, self.target._format)
2775
self.count_copied = 0
2776
if revision_id is None:
2778
# everything to do - use pack logic
2779
# to fetch from all packs to one without
2780
# inventory parsing etc, IFF nothing to be copied is in the target.
2782
source_revision_ids = frozenset(self.source.all_revision_ids())
2783
revision_ids = source_revision_ids - \
2784
frozenset(self.target.get_parent_map(source_revision_ids))
2785
revision_keys = [(revid,) for revid in revision_ids]
2786
index = self.target._pack_collection.revision_index.combined_index
2787
present_revision_ids = set(item[1][0] for item in
2788
index.iter_entries(revision_keys))
2789
revision_ids = set(revision_ids) - present_revision_ids
2790
# implementing the TODO will involve:
2791
# - detecting when all of a pack is selected
2792
# - avoiding as much as possible pre-selection, so the
2793
# more-core routines such as create_pack_from_packs can filter in
2794
# a just-in-time fashion. (though having a HEADS list on a
2795
# repository might make this a lot easier, because we could
2796
# sensibly detect 'new revisions' without doing a full index scan.
2797
elif _mod_revision.is_null(revision_id):
2802
revision_ids = self.search_missing_revision_ids(revision_id,
2803
find_ghosts=find_ghosts).get_keys()
2804
except errors.NoSuchRevision:
2805
raise errors.InstallFailed([revision_id])
2806
if len(revision_ids) == 0:
2808
packs = self.source._pack_collection.all_packs()
2809
pack = Packer(self.target._pack_collection, packs, '.fetch',
2810
revision_ids).pack()
2811
if pack is not None:
2812
self.target._pack_collection._save_pack_names()
2813
# Trigger an autopack. This may duplicate effort as we've just done
2814
# a pack creation, but for now it is simpler to think about as
2815
# 'upload data, then repack if needed'.
2816
self.target._pack_collection.autopack()
2817
return (pack.get_revision_count(), [])
2822
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
2823
"""See InterRepository.missing_revision_ids().
2825
:param find_ghosts: Find ghosts throughout the ancestry of
2828
if not find_ghosts and revision_id is not None:
2829
return self._walk_to_common_revisions([revision_id])
2830
elif revision_id is not None:
2831
# Find ghosts: search for revisions pointing from one repository to
2832
# the other, and vice versa, anywhere in the history of revision_id.
2833
graph = self.target.get_graph(other_repository=self.source)
2834
searcher = graph._make_breadth_first_searcher([revision_id])
2838
next_revs, ghosts = searcher.next_with_ghosts()
2839
except StopIteration:
2841
if revision_id in ghosts:
2842
raise errors.NoSuchRevision(self.source, revision_id)
2843
found_ids.update(next_revs)
2844
found_ids.update(ghosts)
2845
found_ids = frozenset(found_ids)
2846
# Double query here: should be able to avoid this by changing the
2847
# graph api further.
2848
result_set = found_ids - frozenset(
2849
self.target.get_parent_map(found_ids))
2851
source_ids = self.source.all_revision_ids()
2852
# source_ids is the worst possible case we may need to pull.
2853
# now we want to filter source_ids against what we actually
2854
# have in target, but don't try to check for existence where we know
2855
# we do not have a revision as that would be pointless.
2856
target_ids = set(self.target.all_revision_ids())
2857
result_set = set(source_ids).difference(target_ids)
2858
return self.source.revision_ids_to_search_result(result_set)
2861
class InterModel1and2(InterRepository):
2864
def _get_repo_format_to_test(self):
2868
def is_compatible(source, target):
2869
if not source.supports_rich_root() and target.supports_rich_root():
2875
def fetch(self, revision_id=None, pb=None, find_ghosts=False):
2876
"""See InterRepository.fetch()."""
2877
from bzrlib.fetch import Model1toKnit2Fetcher
2878
f = Model1toKnit2Fetcher(to_repository=self.target,
2879
from_repository=self.source,
2880
last_revision=revision_id,
2881
pb=pb, find_ghosts=find_ghosts)
2882
return f.count_copied, f.failed_revisions
2885
def copy_content(self, revision_id=None):
2886
"""Make a complete copy of the content in self into destination.
2888
This is a destructive operation! Do not use it on existing
2891
:param revision_id: Only copy the content needed to construct
2892
revision_id and its parents.
2895
self.target.set_make_working_trees(self.source.make_working_trees())
2896
except NotImplementedError:
2898
# but don't bother fetching if we have the needed data now.
2899
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
2900
self.target.has_revision(revision_id)):
2902
self.target.fetch(self.source, revision_id=revision_id)
2905
class InterKnit1and2(InterKnitRepo):
2908
def _get_repo_format_to_test(self):
2912
def is_compatible(source, target):
2913
"""Be compatible with Knit1 source and Knit3 target"""
2915
from bzrlib.repofmt.knitrepo import (
2916
RepositoryFormatKnit1,
2917
RepositoryFormatKnit3,
2919
from bzrlib.repofmt.pack_repo import (
2920
RepositoryFormatKnitPack1,
2921
RepositoryFormatKnitPack3,
2922
RepositoryFormatKnitPack4,
2923
RepositoryFormatKnitPack5,
2924
RepositoryFormatKnitPack5RichRoot,
2925
RepositoryFormatPackDevelopment2,
2926
RepositoryFormatPackDevelopment2Subtree,
2929
RepositoryFormatKnit1, # no rr, no subtree
2930
RepositoryFormatKnitPack1, # no rr, no subtree
2931
RepositoryFormatPackDevelopment2, # no rr, no subtree
2932
RepositoryFormatKnitPack5, # no rr, no subtree
2935
RepositoryFormatKnit3, # rr, subtree
2936
RepositoryFormatKnitPack3, # rr, subtree
2937
RepositoryFormatKnitPack4, # rr, no subtree
2938
RepositoryFormatKnitPack5RichRoot,# rr, no subtree
2939
RepositoryFormatPackDevelopment2Subtree, # rr, subtree
2941
for format in norichroot:
2942
if format.rich_root_data:
2943
raise AssertionError('Format %s is a rich-root format'
2944
' but is included in the non-rich-root list'
2946
for format in richroot:
2947
if not format.rich_root_data:
2948
raise AssertionError('Format %s is not a rich-root format'
2949
' but is included in the rich-root list'
2951
# TODO: One alternative is to just check format.rich_root_data,
2952
# instead of keeping membership lists. However, the formats
2953
# *also* have to use the same 'Knit' style of storage
2954
# (line-deltas, fulltexts, etc.)
2955
return (isinstance(source._format, norichroot) and
2956
isinstance(target._format, richroot))
2957
except AttributeError:
2961
def fetch(self, revision_id=None, pb=None, find_ghosts=False):
2962
"""See InterRepository.fetch()."""
2963
from bzrlib.fetch import Knit1to2Fetcher
2964
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
2965
self.source, self.source._format, self.target,
2966
self.target._format)
2967
f = Knit1to2Fetcher(to_repository=self.target,
2968
from_repository=self.source,
2969
last_revision=revision_id,
2970
pb=pb, find_ghosts=find_ghosts)
2971
return f.count_copied, f.failed_revisions
2974
class InterDifferingSerializer(InterKnitRepo):
3724
2977
def _get_repo_format_to_test(self):
3728
2981
def is_compatible(source, target):
3729
2982
"""Be compatible with Knit2 source and Knit3 target"""
3730
# This is redundant with format.check_conversion_target(), however that
3731
# raises an exception, and we just want to say "False" as in we won't
3732
# support converting between these formats.
3733
if 'IDS_never' in debug.debug_flags:
3735
if source.supports_rich_root() and not target.supports_rich_root():
3737
if (source._format.supports_tree_reference
3738
and not target._format.supports_tree_reference):
3740
if target._fallback_repositories and target._format.supports_chks:
3741
# IDS doesn't know how to copy CHKs for the parent inventories it
3742
# adds to stacked repos.
3744
if 'IDS_always' in debug.debug_flags:
3746
# Only use this code path for local source and target. IDS does far
3747
# too much IO (both bandwidth and roundtrips) over a network.
3748
if not source.bzrdir.transport.base.startswith('file:///'):
3750
if not target.bzrdir.transport.base.startswith('file:///'):
2983
if source.supports_rich_root() != target.supports_rich_root():
2985
# Ideally, we'd support fetching if the source had no tree references
2986
# even if it supported them...
2987
if (getattr(source, '_format.supports_tree_reference', False) and
2988
not getattr(target, '_format.supports_tree_reference', False)):
3754
def _get_trees(self, revision_ids, cache):
3756
for rev_id in revision_ids:
3758
possible_trees.append((rev_id, cache[rev_id]))
3760
# Not cached, but inventory might be present anyway.
3762
tree = self.source.revision_tree(rev_id)
3763
except errors.NoSuchRevision:
3764
# Nope, parent is ghost.
3767
cache[rev_id] = tree
3768
possible_trees.append((rev_id, tree))
3769
return possible_trees
3771
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3772
"""Get the best delta and base for this revision.
3774
:return: (basis_id, delta)
3777
# Generate deltas against each tree, to find the shortest.
3778
texts_possibly_new_in_tree = set()
3779
for basis_id, basis_tree in possible_trees:
3780
delta = tree.inventory._make_delta(basis_tree.inventory)
3781
for old_path, new_path, file_id, new_entry in delta:
3782
if new_path is None:
3783
# This file_id isn't present in the new rev, so we don't
3787
# Rich roots are handled elsewhere...
3789
kind = new_entry.kind
3790
if kind != 'directory' and kind != 'file':
3791
# No text record associated with this inventory entry.
3793
# This is a directory or file that has changed somehow.
3794
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3795
deltas.append((len(delta), basis_id, delta))
3797
return deltas[0][1:]
3799
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3800
"""Find all parent revisions that are absent, but for which the
3801
inventory is present, and copy those inventories.
3803
This is necessary to preserve correctness when the source is stacked
3804
without fallbacks configured. (Note that in cases like upgrade the
3805
source may be not have _fallback_repositories even though it is
3809
for parents in parent_map.values():
3810
parent_revs.update(parents)
3811
present_parents = self.source.get_parent_map(parent_revs)
3812
absent_parents = set(parent_revs).difference(present_parents)
3813
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3814
(rev_id,) for rev_id in absent_parents)
3815
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3816
for parent_tree in self.source.revision_trees(parent_inv_ids):
3817
current_revision_id = parent_tree.get_revision_id()
3818
parents_parents_keys = parent_invs_keys_for_stacking[
3819
(current_revision_id,)]
3820
parents_parents = [key[-1] for key in parents_parents_keys]
3821
basis_id = _mod_revision.NULL_REVISION
3822
basis_tree = self.source.revision_tree(basis_id)
3823
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3824
self.target.add_inventory_by_delta(
3825
basis_id, delta, current_revision_id, parents_parents)
3826
cache[current_revision_id] = parent_tree
3828
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3829
"""Fetch across a few revisions.
3831
:param revision_ids: The revisions to copy
3832
:param basis_id: The revision_id of a tree that must be in cache, used
3833
as a basis for delta when no other base is available
3834
:param cache: A cache of RevisionTrees that we can use.
3835
:param a_graph: A Graph object to determine the heads() of the
3836
rich-root data stream.
3837
:return: The revision_id of the last converted tree. The RevisionTree
3838
for it will be in cache
3840
# Walk though all revisions; get inventory deltas, copy referenced
3841
# texts that delta references, insert the delta, revision and
3843
root_keys_to_create = set()
3846
pending_revisions = []
3847
parent_map = self.source.get_parent_map(revision_ids)
3848
self._fetch_parent_invs_for_stacking(parent_map, cache)
3849
self.source._safe_to_return_from_cache = True
3850
for tree in self.source.revision_trees(revision_ids):
3851
# Find a inventory delta for this revision.
3852
# Find text entries that need to be copied, too.
3853
current_revision_id = tree.get_revision_id()
3854
parent_ids = parent_map.get(current_revision_id, ())
3855
parent_trees = self._get_trees(parent_ids, cache)
3856
possible_trees = list(parent_trees)
3857
if len(possible_trees) == 0:
3858
# There either aren't any parents, or the parents are ghosts,
3859
# so just use the last converted tree.
3860
possible_trees.append((basis_id, cache[basis_id]))
3861
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3863
revision = self.source.get_revision(current_revision_id)
3864
pending_deltas.append((basis_id, delta,
3865
current_revision_id, revision.parent_ids))
3866
if self._converting_to_rich_root:
3867
self._revision_id_to_root_id[current_revision_id] = \
3869
# Determine which texts are in present in this revision but not in
3870
# any of the available parents.
3871
texts_possibly_new_in_tree = set()
3872
for old_path, new_path, file_id, entry in delta:
3873
if new_path is None:
3874
# This file_id isn't present in the new rev
3878
if not self.target.supports_rich_root():
3879
# The target doesn't support rich root, so we don't
3882
if self._converting_to_rich_root:
3883
# This can't be copied normally, we have to insert
3885
root_keys_to_create.add((file_id, entry.revision))
3888
texts_possibly_new_in_tree.add((file_id, entry.revision))
3889
for basis_id, basis_tree in possible_trees:
3890
basis_inv = basis_tree.inventory
3891
for file_key in list(texts_possibly_new_in_tree):
3892
file_id, file_revision = file_key
3894
entry = basis_inv[file_id]
3895
except errors.NoSuchId:
3897
if entry.revision == file_revision:
3898
texts_possibly_new_in_tree.remove(file_key)
3899
text_keys.update(texts_possibly_new_in_tree)
3900
pending_revisions.append(revision)
3901
cache[current_revision_id] = tree
3902
basis_id = current_revision_id
3903
self.source._safe_to_return_from_cache = False
3905
from_texts = self.source.texts
3906
to_texts = self.target.texts
3907
if root_keys_to_create:
3908
root_stream = _mod_fetch._new_root_data_stream(
3909
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3910
self.source, graph=a_graph)
3911
to_texts.insert_record_stream(root_stream)
3912
to_texts.insert_record_stream(from_texts.get_record_stream(
3913
text_keys, self.target._format._fetch_order,
3914
not self.target._format._fetch_uses_deltas))
3915
# insert inventory deltas
3916
for delta in pending_deltas:
3917
self.target.add_inventory_by_delta(*delta)
3918
if self.target._fallback_repositories:
3919
# Make sure this stacked repository has all the parent inventories
3920
# for the new revisions that we are about to insert. We do this
3921
# before adding the revisions so that no revision is added until
3922
# all the inventories it may depend on are added.
3923
# Note that this is overzealous, as we may have fetched these in an
3926
revision_ids = set()
3927
for revision in pending_revisions:
3928
revision_ids.add(revision.revision_id)
3929
parent_ids.update(revision.parent_ids)
3930
parent_ids.difference_update(revision_ids)
3931
parent_ids.discard(_mod_revision.NULL_REVISION)
3932
parent_map = self.source.get_parent_map(parent_ids)
3933
# we iterate over parent_map and not parent_ids because we don't
3934
# want to try copying any revision which is a ghost
3935
for parent_tree in self.source.revision_trees(parent_map):
3936
current_revision_id = parent_tree.get_revision_id()
3937
parents_parents = parent_map[current_revision_id]
3938
possible_trees = self._get_trees(parents_parents, cache)
3939
if len(possible_trees) == 0:
3940
# There either aren't any parents, or the parents are
3941
# ghosts, so just use the last converted tree.
3942
possible_trees.append((basis_id, cache[basis_id]))
3943
basis_id, delta = self._get_delta_for_revision(parent_tree,
3944
parents_parents, possible_trees)
3945
self.target.add_inventory_by_delta(
3946
basis_id, delta, current_revision_id, parents_parents)
3947
# insert signatures and revisions
3948
for revision in pending_revisions:
3950
signature = self.source.get_signature_text(
3951
revision.revision_id)
3952
self.target.add_signature_text(revision.revision_id,
3954
except errors.NoSuchRevision:
3956
self.target.add_revision(revision.revision_id, revision)
3959
def _fetch_all_revisions(self, revision_ids, pb):
3960
"""Fetch everything for the list of revisions.
3962
:param revision_ids: The list of revisions to fetch. Must be in
3964
:param pb: A ProgressTask
3967
basis_id, basis_tree = self._get_basis(revision_ids[0])
3969
cache = lru_cache.LRUCache(100)
3970
cache[basis_id] = basis_tree
3971
del basis_tree # We don't want to hang on to it here
3973
if self._converting_to_rich_root and len(revision_ids) > 100:
3974
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3979
for offset in range(0, len(revision_ids), batch_size):
3980
self.target.start_write_group()
3982
pb.update('Transferring revisions', offset,
3984
batch = revision_ids[offset:offset+batch_size]
3985
basis_id = self._fetch_batch(batch, basis_id, cache,
3988
self.source._safe_to_return_from_cache = False
3989
self.target.abort_write_group()
3992
hint = self.target.commit_write_group()
3995
if hints and self.target._format.pack_compresses:
3996
self.target.pack(hint=hints)
3997
pb.update('Transferring revisions', len(revision_ids),
4000
2992
@needs_write_lock
4001
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
2993
def fetch(self, revision_id=None, pb=None, find_ghosts=False):
4003
2994
"""See InterRepository.fetch()."""
4004
if fetch_spec is not None:
4005
raise AssertionError("Not implemented yet...")
4006
# See <https://launchpad.net/bugs/456077> asking for a warning here
4008
# nb this is only active for local-local fetches; other things using
4010
ui.ui_factory.warn_cross_format_fetch(self.source._format,
4011
self.target._format)
4012
if (not self.source.supports_rich_root()
4013
and self.target.supports_rich_root()):
4014
self._converting_to_rich_root = True
4015
self._revision_id_to_root_id = {}
4017
self._converting_to_rich_root = False
4018
2995
revision_ids = self.target.search_missing_revision_ids(self.source,
4019
2996
revision_id, find_ghosts=find_ghosts).get_keys()
4020
if not revision_ids:
4022
2997
revision_ids = tsort.topo_sort(
4023
2998
self.source.get_graph().get_parent_map(revision_ids))
4024
if not revision_ids:
4026
# Walk though all revisions; get inventory deltas, copy referenced
4027
# texts that delta references, insert the delta, revision and
2999
def revisions_iterator():
3000
for current_revision_id in revision_ids:
3001
revision = self.source.get_revision(current_revision_id)
3002
tree = self.source.revision_tree(current_revision_id)
3004
signature = self.source.get_signature_text(
3005
current_revision_id)
3006
except errors.NoSuchRevision:
3008
yield revision, tree, signature
4030
3010
my_pb = ui.ui_factory.nested_progress_bar()
4033
symbol_versioning.warn(
4034
symbol_versioning.deprecated_in((1, 14, 0))
4035
% "pb parameter to fetch()")
4038
self._fetch_all_revisions(revision_ids, pb)
3015
install_revisions(self.target, revisions_iterator(),
3016
len(revision_ids), pb)
4040
3018
if my_pb is not None:
4041
3019
my_pb.finished()
4042
3020
return len(revision_ids), 0
4044
def _get_basis(self, first_revision_id):
4045
"""Get a revision and tree which exists in the target.
4047
This assumes that first_revision_id is selected for transmission
4048
because all other ancestors are already present. If we can't find an
4049
ancestor we fall back to NULL_REVISION since we know that is safe.
4051
:return: (basis_id, basis_tree)
4053
first_rev = self.source.get_revision(first_revision_id)
4055
basis_id = first_rev.parent_ids[0]
4056
# only valid as a basis if the target has it
4057
self.target.get_revision(basis_id)
4058
# Try to get a basis tree - if its a ghost it will hit the
4059
# NoSuchRevision case.
4060
basis_tree = self.source.revision_tree(basis_id)
4061
except (IndexError, errors.NoSuchRevision):
4062
basis_id = _mod_revision.NULL_REVISION
4063
basis_tree = self.source.revision_tree(basis_id)
4064
return basis_id, basis_tree
3023
class InterOtherToRemote(InterRepository):
3025
def __init__(self, source, target):
3026
InterRepository.__init__(self, source, target)
3027
self._real_inter = None
3030
def is_compatible(source, target):
3031
if isinstance(target, remote.RemoteRepository):
3035
def _ensure_real_inter(self):
3036
if self._real_inter is None:
3037
self.target._ensure_real()
3038
real_target = self.target._real_repository
3039
self._real_inter = InterRepository.get(self.source, real_target)
3041
def copy_content(self, revision_id=None):
3042
self._ensure_real_inter()
3043
self._real_inter.copy_content(revision_id=revision_id)
3045
def fetch(self, revision_id=None, pb=None, find_ghosts=False):
3046
self._ensure_real_inter()
3047
return self._real_inter.fetch(revision_id=revision_id, pb=pb,
3048
find_ghosts=find_ghosts)
3051
def _get_repo_format_to_test(self):
3055
class InterRemoteToOther(InterRepository):
3057
def __init__(self, source, target):
3058
InterRepository.__init__(self, source, target)
3059
self._real_inter = None
3062
def is_compatible(source, target):
3063
if not isinstance(source, remote.RemoteRepository):
3065
# Is source's model compatible with target's model?
3066
source._ensure_real()
3067
real_source = source._real_repository
3068
if isinstance(real_source, remote.RemoteRepository):
3069
raise NotImplementedError(
3070
"We don't support remote repos backed by remote repos yet.")
3071
return InterRepository._same_model(real_source, target)
3073
def _ensure_real_inter(self):
3074
if self._real_inter is None:
3075
self.source._ensure_real()
3076
real_source = self.source._real_repository
3077
self._real_inter = InterRepository.get(real_source, self.target)
3079
def fetch(self, revision_id=None, pb=None, find_ghosts=False):
3080
self._ensure_real_inter()
3081
return self._real_inter.fetch(revision_id=revision_id, pb=pb,
3082
find_ghosts=find_ghosts)
3084
def copy_content(self, revision_id=None):
3085
self._ensure_real_inter()
3086
self._real_inter.copy_content(revision_id=revision_id)
3089
def _get_repo_format_to_test(self):
4067
3094
InterRepository.register_optimiser(InterDifferingSerializer)
4068
3095
InterRepository.register_optimiser(InterSameDataRepository)
4069
3096
InterRepository.register_optimiser(InterWeaveRepo)
4070
3097
InterRepository.register_optimiser(InterKnitRepo)
3098
InterRepository.register_optimiser(InterModel1and2)
3099
InterRepository.register_optimiser(InterKnit1and2)
3100
InterRepository.register_optimiser(InterPackRepo)
3101
InterRepository.register_optimiser(InterOtherToRemote)
3102
InterRepository.register_optimiser(InterRemoteToOther)
4073
3105
class CopyConverter(object):
4074
3106
"""A repository conversion tool which just performs a copy of the content.
4076
3108
This is slow but quite reliable.
4233
3256
revision_graph[key] = tuple(parent for parent in parents if parent
4234
3257
in revision_graph)
4235
3258
return revision_graph
4238
class StreamSink(object):
4239
"""An object that can insert a stream into a repository.
4241
This interface handles the complexity of reserialising inventories and
4242
revisions from different formats, and allows unidirectional insertion into
4243
stacked repositories without looking for the missing basis parents
4247
def __init__(self, target_repo):
4248
self.target_repo = target_repo
4250
def insert_stream(self, stream, src_format, resume_tokens):
4251
"""Insert a stream's content into the target repository.
4253
:param src_format: a bzr repository format.
4255
:return: a list of resume tokens and an iterable of keys additional
4256
items required before the insertion can be completed.
4258
self.target_repo.lock_write()
4261
self.target_repo.resume_write_group(resume_tokens)
4264
self.target_repo.start_write_group()
4267
# locked_insert_stream performs a commit|suspend.
4268
return self._locked_insert_stream(stream, src_format, is_resume)
4270
self.target_repo.abort_write_group(suppress_errors=True)
4273
self.target_repo.unlock()
4275
def _locked_insert_stream(self, stream, src_format, is_resume):
4276
to_serializer = self.target_repo._format._serializer
4277
src_serializer = src_format._serializer
4279
if to_serializer == src_serializer:
4280
# If serializers match and the target is a pack repository, set the
4281
# write cache size on the new pack. This avoids poor performance
4282
# on transports where append is unbuffered (such as
4283
# RemoteTransport). This is safe to do because nothing should read
4284
# back from the target repository while a stream with matching
4285
# serialization is being inserted.
4286
# The exception is that a delta record from the source that should
4287
# be a fulltext may need to be expanded by the target (see
4288
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4289
# explicitly flush any buffered writes first in that rare case.
4291
new_pack = self.target_repo._pack_collection._new_pack
4292
except AttributeError:
4293
# Not a pack repository
4296
new_pack.set_write_cache_size(1024*1024)
4297
for substream_type, substream in stream:
4298
if 'stream' in debug.debug_flags:
4299
mutter('inserting substream: %s', substream_type)
4300
if substream_type == 'texts':
4301
self.target_repo.texts.insert_record_stream(substream)
4302
elif substream_type == 'inventories':
4303
if src_serializer == to_serializer:
4304
self.target_repo.inventories.insert_record_stream(
4307
self._extract_and_insert_inventories(
4308
substream, src_serializer)
4309
elif substream_type == 'inventory-deltas':
4310
ui.ui_factory.warn_cross_format_fetch(src_format,
4311
self.target_repo._format)
4312
self._extract_and_insert_inventory_deltas(
4313
substream, src_serializer)
4314
elif substream_type == 'chk_bytes':
4315
# XXX: This doesn't support conversions, as it assumes the
4316
# conversion was done in the fetch code.
4317
self.target_repo.chk_bytes.insert_record_stream(substream)
4318
elif substream_type == 'revisions':
4319
# This may fallback to extract-and-insert more often than
4320
# required if the serializers are different only in terms of
4322
if src_serializer == to_serializer:
4323
self.target_repo.revisions.insert_record_stream(
4326
self._extract_and_insert_revisions(substream,
4328
elif substream_type == 'signatures':
4329
self.target_repo.signatures.insert_record_stream(substream)
4331
raise AssertionError('kaboom! %s' % (substream_type,))
4332
# Done inserting data, and the missing_keys calculations will try to
4333
# read back from the inserted data, so flush the writes to the new pack
4334
# (if this is pack format).
4335
if new_pack is not None:
4336
new_pack._write_data('', flush=True)
4337
# Find all the new revisions (including ones from resume_tokens)
4338
missing_keys = self.target_repo.get_missing_parent_inventories(
4339
check_for_missing_texts=is_resume)
4341
for prefix, versioned_file in (
4342
('texts', self.target_repo.texts),
4343
('inventories', self.target_repo.inventories),
4344
('revisions', self.target_repo.revisions),
4345
('signatures', self.target_repo.signatures),
4346
('chk_bytes', self.target_repo.chk_bytes),
4348
if versioned_file is None:
4350
# TODO: key is often going to be a StaticTuple object
4351
# I don't believe we can define a method by which
4352
# (prefix,) + StaticTuple will work, though we could
4353
# define a StaticTuple.sq_concat that would allow you to
4354
# pass in either a tuple or a StaticTuple as the second
4355
# object, so instead we could have:
4356
# StaticTuple(prefix) + key here...
4357
missing_keys.update((prefix,) + key for key in
4358
versioned_file.get_missing_compression_parent_keys())
4359
except NotImplementedError:
4360
# cannot even attempt suspending, and missing would have failed
4361
# during stream insertion.
4362
missing_keys = set()
4365
# suspend the write group and tell the caller what we is
4366
# missing. We know we can suspend or else we would not have
4367
# entered this code path. (All repositories that can handle
4368
# missing keys can handle suspending a write group).
4369
write_group_tokens = self.target_repo.suspend_write_group()
4370
return write_group_tokens, missing_keys
4371
hint = self.target_repo.commit_write_group()
4372
if (to_serializer != src_serializer and
4373
self.target_repo._format.pack_compresses):
4374
self.target_repo.pack(hint=hint)
4377
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4378
target_rich_root = self.target_repo._format.rich_root_data
4379
target_tree_refs = self.target_repo._format.supports_tree_reference
4380
for record in substream:
4381
# Insert the delta directly
4382
inventory_delta_bytes = record.get_bytes_as('fulltext')
4383
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4385
parse_result = deserialiser.parse_text_bytes(
4386
inventory_delta_bytes)
4387
except inventory_delta.IncompatibleInventoryDelta, err:
4388
trace.mutter("Incompatible delta: %s", err.msg)
4389
raise errors.IncompatibleRevision(self.target_repo._format)
4390
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4391
revision_id = new_id
4392
parents = [key[0] for key in record.parents]
4393
self.target_repo.add_inventory_by_delta(
4394
basis_id, inv_delta, revision_id, parents)
4396
def _extract_and_insert_inventories(self, substream, serializer,
4398
"""Generate a new inventory versionedfile in target, converting data.
4400
The inventory is retrieved from the source, (deserializing it), and
4401
stored in the target (reserializing it in a different format).
4403
target_rich_root = self.target_repo._format.rich_root_data
4404
target_tree_refs = self.target_repo._format.supports_tree_reference
4405
for record in substream:
4406
# It's not a delta, so it must be a fulltext in the source
4407
# serializer's format.
4408
bytes = record.get_bytes_as('fulltext')
4409
revision_id = record.key[0]
4410
inv = serializer.read_inventory_from_string(bytes, revision_id)
4411
parents = [key[0] for key in record.parents]
4412
self.target_repo.add_inventory(revision_id, inv, parents)
4413
# No need to keep holding this full inv in memory when the rest of
4414
# the substream is likely to be all deltas.
4417
def _extract_and_insert_revisions(self, substream, serializer):
4418
for record in substream:
4419
bytes = record.get_bytes_as('fulltext')
4420
revision_id = record.key[0]
4421
rev = serializer.read_revision_from_string(bytes)
4422
if rev.revision_id != revision_id:
4423
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4424
self.target_repo.add_revision(revision_id, rev)
4427
if self.target_repo._format._fetch_reconcile:
4428
self.target_repo.reconcile()
4431
class StreamSource(object):
4432
"""A source of a stream for fetching between repositories."""
4434
def __init__(self, from_repository, to_format):
4435
"""Create a StreamSource streaming from from_repository."""
4436
self.from_repository = from_repository
4437
self.to_format = to_format
4439
def delta_on_metadata(self):
4440
"""Return True if delta's are permitted on metadata streams.
4442
That is on revisions and signatures.
4444
src_serializer = self.from_repository._format._serializer
4445
target_serializer = self.to_format._serializer
4446
return (self.to_format._fetch_uses_deltas and
4447
src_serializer == target_serializer)
4449
def _fetch_revision_texts(self, revs):
4450
# fetch signatures first and then the revision texts
4451
# may need to be a InterRevisionStore call here.
4452
from_sf = self.from_repository.signatures
4453
# A missing signature is just skipped.
4454
keys = [(rev_id,) for rev_id in revs]
4455
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4457
self.to_format._fetch_order,
4458
not self.to_format._fetch_uses_deltas))
4459
# If a revision has a delta, this is actually expanded inside the
4460
# insert_record_stream code now, which is an alternate fix for
4462
from_rf = self.from_repository.revisions
4463
revisions = from_rf.get_record_stream(
4465
self.to_format._fetch_order,
4466
not self.delta_on_metadata())
4467
return [('signatures', signatures), ('revisions', revisions)]
4469
def _generate_root_texts(self, revs):
4470
"""This will be called by get_stream between fetching weave texts and
4471
fetching the inventory weave.
4473
if self._rich_root_upgrade():
4474
return _mod_fetch.Inter1and2Helper(
4475
self.from_repository).generate_root_texts(revs)
4479
def get_stream(self, search):
4481
revs = search.get_keys()
4482
graph = self.from_repository.get_graph()
4483
revs = tsort.topo_sort(graph.get_parent_map(revs))
4484
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4486
for knit_kind, file_id, revisions in data_to_fetch:
4487
if knit_kind != phase:
4489
# Make a new progress bar for this phase
4490
if knit_kind == "file":
4491
# Accumulate file texts
4492
text_keys.extend([(file_id, revision) for revision in
4494
elif knit_kind == "inventory":
4495
# Now copy the file texts.
4496
from_texts = self.from_repository.texts
4497
yield ('texts', from_texts.get_record_stream(
4498
text_keys, self.to_format._fetch_order,
4499
not self.to_format._fetch_uses_deltas))
4500
# Cause an error if a text occurs after we have done the
4503
# Before we process the inventory we generate the root
4504
# texts (if necessary) so that the inventories references
4506
for _ in self._generate_root_texts(revs):
4508
# we fetch only the referenced inventories because we do not
4509
# know for unselected inventories whether all their required
4510
# texts are present in the other repository - it could be
4512
for info in self._get_inventory_stream(revs):
4514
elif knit_kind == "signatures":
4515
# Nothing to do here; this will be taken care of when
4516
# _fetch_revision_texts happens.
4518
elif knit_kind == "revisions":
4519
for record in self._fetch_revision_texts(revs):
4522
raise AssertionError("Unknown knit kind %r" % knit_kind)
4524
def get_stream_for_missing_keys(self, missing_keys):
4525
# missing keys can only occur when we are byte copying and not
4526
# translating (because translation means we don't send
4527
# unreconstructable deltas ever).
4529
keys['texts'] = set()
4530
keys['revisions'] = set()
4531
keys['inventories'] = set()
4532
keys['chk_bytes'] = set()
4533
keys['signatures'] = set()
4534
for key in missing_keys:
4535
keys[key[0]].add(key[1:])
4536
if len(keys['revisions']):
4537
# If we allowed copying revisions at this point, we could end up
4538
# copying a revision without copying its required texts: a
4539
# violation of the requirements for repository integrity.
4540
raise AssertionError(
4541
'cannot copy revisions to fill in missing deltas %s' % (
4542
keys['revisions'],))
4543
for substream_kind, keys in keys.iteritems():
4544
vf = getattr(self.from_repository, substream_kind)
4545
if vf is None and keys:
4546
raise AssertionError(
4547
"cannot fill in keys for a versioned file we don't"
4548
" have: %s needs %s" % (substream_kind, keys))
4550
# No need to stream something we don't have
4552
if substream_kind == 'inventories':
4553
# Some missing keys are genuinely ghosts, filter those out.
4554
present = self.from_repository.inventories.get_parent_map(keys)
4555
revs = [key[0] for key in present]
4556
# Get the inventory stream more-or-less as we do for the
4557
# original stream; there's no reason to assume that records
4558
# direct from the source will be suitable for the sink. (Think
4559
# e.g. 2a -> 1.9-rich-root).
4560
for info in self._get_inventory_stream(revs, missing=True):
4564
# Ask for full texts always so that we don't need more round trips
4565
# after this stream.
4566
# Some of the missing keys are genuinely ghosts, so filter absent
4567
# records. The Sink is responsible for doing another check to
4568
# ensure that ghosts don't introduce missing data for future
4570
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4571
self.to_format._fetch_order, True))
4572
yield substream_kind, stream
4574
def inventory_fetch_order(self):
4575
if self._rich_root_upgrade():
4576
return 'topological'
4578
return self.to_format._fetch_order
4580
def _rich_root_upgrade(self):
4581
return (not self.from_repository._format.rich_root_data and
4582
self.to_format.rich_root_data)
4584
def _get_inventory_stream(self, revision_ids, missing=False):
4585
from_format = self.from_repository._format
4586
if (from_format.supports_chks and self.to_format.supports_chks and
4587
from_format.network_name() == self.to_format.network_name()):
4588
raise AssertionError(
4589
"this case should be handled by GroupCHKStreamSource")
4590
elif 'forceinvdeltas' in debug.debug_flags:
4591
return self._get_convertable_inventory_stream(revision_ids,
4592
delta_versus_null=missing)
4593
elif from_format.network_name() == self.to_format.network_name():
4595
return self._get_simple_inventory_stream(revision_ids,
4597
elif (not from_format.supports_chks and not self.to_format.supports_chks
4598
and from_format._serializer == self.to_format._serializer):
4599
# Essentially the same format.
4600
return self._get_simple_inventory_stream(revision_ids,
4603
# Any time we switch serializations, we want to use an
4604
# inventory-delta based approach.
4605
return self._get_convertable_inventory_stream(revision_ids,
4606
delta_versus_null=missing)
4608
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4609
# NB: This currently reopens the inventory weave in source;
4610
# using a single stream interface instead would avoid this.
4611
from_weave = self.from_repository.inventories
4613
delta_closure = True
4615
delta_closure = not self.delta_on_metadata()
4616
yield ('inventories', from_weave.get_record_stream(
4617
[(rev_id,) for rev_id in revision_ids],
4618
self.inventory_fetch_order(), delta_closure))
4620
def _get_convertable_inventory_stream(self, revision_ids,
4621
delta_versus_null=False):
4622
# The two formats are sufficiently different that there is no fast
4623
# path, so we need to send just inventorydeltas, which any
4624
# sufficiently modern client can insert into any repository.
4625
# The StreamSink code expects to be able to
4626
# convert on the target, so we need to put bytes-on-the-wire that can
4627
# be converted. That means inventory deltas (if the remote is <1.19,
4628
# RemoteStreamSink will fallback to VFS to insert the deltas).
4629
yield ('inventory-deltas',
4630
self._stream_invs_as_deltas(revision_ids,
4631
delta_versus_null=delta_versus_null))
4633
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4634
"""Return a stream of inventory-deltas for the given rev ids.
4636
:param revision_ids: The list of inventories to transmit
4637
:param delta_versus_null: Don't try to find a minimal delta for this
4638
entry, instead compute the delta versus the NULL_REVISION. This
4639
effectively streams a complete inventory. Used for stuff like
4640
filling in missing parents, etc.
4642
from_repo = self.from_repository
4643
revision_keys = [(rev_id,) for rev_id in revision_ids]
4644
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4645
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4647
inventories = self.from_repository.iter_inventories(
4648
revision_ids, 'topological')
4649
format = from_repo._format
4650
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4651
inventory_cache = lru_cache.LRUCache(50)
4652
null_inventory = from_repo.revision_tree(
4653
_mod_revision.NULL_REVISION).inventory
4654
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4655
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4656
# repo back into a non-rich-root repo ought to be allowed)
4657
serializer = inventory_delta.InventoryDeltaSerializer(
4658
versioned_root=format.rich_root_data,
4659
tree_references=format.supports_tree_reference)
4660
for inv in inventories:
4661
key = (inv.revision_id,)
4662
parent_keys = parent_map.get(key, ())
4664
if not delta_versus_null and parent_keys:
4665
# The caller did not ask for complete inventories and we have
4666
# some parents that we can delta against. Make a delta against
4667
# each parent so that we can find the smallest.
4668
parent_ids = [parent_key[0] for parent_key in parent_keys]
4669
for parent_id in parent_ids:
4670
if parent_id not in invs_sent_so_far:
4671
# We don't know that the remote side has this basis, so
4674
if parent_id == _mod_revision.NULL_REVISION:
4675
parent_inv = null_inventory
4677
parent_inv = inventory_cache.get(parent_id, None)
4678
if parent_inv is None:
4679
parent_inv = from_repo.get_inventory(parent_id)
4680
candidate_delta = inv._make_delta(parent_inv)
4681
if (delta is None or
4682
len(delta) > len(candidate_delta)):
4683
delta = candidate_delta
4684
basis_id = parent_id
4686
# Either none of the parents ended up being suitable, or we
4687
# were asked to delta against NULL
4688
basis_id = _mod_revision.NULL_REVISION
4689
delta = inv._make_delta(null_inventory)
4690
invs_sent_so_far.add(inv.revision_id)
4691
inventory_cache[inv.revision_id] = inv
4692
delta_serialized = ''.join(
4693
serializer.delta_to_lines(basis_id, key[-1], delta))
4694
yield versionedfile.FulltextContentFactory(
4695
key, parent_keys, None, delta_serialized)
4698
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4699
stop_revision=None):
4700
"""Extend the partial history to include a given index
4702
If a stop_index is supplied, stop when that index has been reached.
4703
If a stop_revision is supplied, stop when that revision is
4704
encountered. Otherwise, stop when the beginning of history is
4707
:param stop_index: The index which should be present. When it is
4708
present, history extension will stop.
4709
:param stop_revision: The revision id which should be present. When
4710
it is encountered, history extension will stop.
4712
start_revision = partial_history_cache[-1]
4713
iterator = repo.iter_reverse_revision_history(start_revision)
4715
#skip the last revision in the list
4718
if (stop_index is not None and
4719
len(partial_history_cache) > stop_index):
4721
if partial_history_cache[-1] == stop_revision:
4723
revision_id = iterator.next()
4724
partial_history_cache.append(revision_id)
4725
except StopIteration:
added
removed
bzrlib/repository.py
