1154
1168
# The old API returned a list, should this actually be a set?
1155
1169
return parent_map.keys()
1171
def _check_inventories(self, checker):
1172
"""Check the inventories found from the revision scan.
1174
This is responsible for verifying the sha1 of inventories and
1175
creating a pending_keys set that covers data referenced by inventories.
1177
bar = ui.ui_factory.nested_progress_bar()
1179
self._do_check_inventories(checker, bar)
1183
def _do_check_inventories(self, checker, bar):
1184
"""Helper for _check_inventories."""
1186
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1187
kinds = ['chk_bytes', 'texts']
1188
count = len(checker.pending_keys)
1189
bar.update("inventories", 0, 2)
1190
current_keys = checker.pending_keys
1191
checker.pending_keys = {}
1192
# Accumulate current checks.
1193
for key in current_keys:
1194
if key[0] != 'inventories' and key[0] not in kinds:
1195
checker._report_items.append('unknown key type %r' % (key,))
1196
keys[key[0]].add(key[1:])
1197
if keys['inventories']:
1198
# NB: output order *should* be roughly sorted - topo or
1199
# inverse topo depending on repository - either way decent
1200
# to just delta against. However, pre-CHK formats didn't
1201
# try to optimise inventory layout on disk. As such the
1202
# pre-CHK code path does not use inventory deltas.
1204
for record in self.inventories.check(keys=keys['inventories']):
1205
if record.storage_kind == 'absent':
1206
checker._report_items.append(
1207
'Missing inventory {%s}' % (record.key,))
1209
last_object = self._check_record('inventories', record,
1210
checker, last_object,
1211
current_keys[('inventories',) + record.key])
1212
del keys['inventories']
1215
bar.update("texts", 1)
1216
while (checker.pending_keys or keys['chk_bytes']
1218
# Something to check.
1219
current_keys = checker.pending_keys
1220
checker.pending_keys = {}
1221
# Accumulate current checks.
1222
for key in current_keys:
1223
if key[0] not in kinds:
1224
checker._report_items.append('unknown key type %r' % (key,))
1225
keys[key[0]].add(key[1:])
1226
# Check the outermost kind only - inventories || chk_bytes || texts
1230
for record in getattr(self, kind).check(keys=keys[kind]):
1231
if record.storage_kind == 'absent':
1232
checker._report_items.append(
1233
'Missing %s {%s}' % (kind, record.key,))
1235
last_object = self._check_record(kind, record,
1236
checker, last_object, current_keys[(kind,) + record.key])
1240
def _check_record(self, kind, record, checker, last_object, item_data):
1241
"""Check a single text from this repository."""
1242
if kind == 'inventories':
1243
rev_id = record.key[0]
1244
inv = self._deserialise_inventory(rev_id,
1245
record.get_bytes_as('fulltext'))
1246
if last_object is not None:
1247
delta = inv._make_delta(last_object)
1248
for old_path, path, file_id, ie in delta:
1251
ie.check(checker, rev_id, inv)
1253
for path, ie in inv.iter_entries():
1254
ie.check(checker, rev_id, inv)
1255
if self._format.fast_deltas:
1257
elif kind == 'chk_bytes':
1258
# No code written to check chk_bytes for this repo format.
1259
checker._report_items.append(
1260
'unsupported key type chk_bytes for %s' % (record.key,))
1261
elif kind == 'texts':
1262
self._check_text(record, checker, item_data)
1264
checker._report_items.append(
1265
'unknown key type %s for %s' % (kind, record.key))
1267
def _check_text(self, record, checker, item_data):
1268
"""Check a single text."""
1269
# Check it is extractable.
1270
# TODO: check length.
1271
if record.storage_kind == 'chunked':
1272
chunks = record.get_bytes_as(record.storage_kind)
1273
sha1 = osutils.sha_strings(chunks)
1274
length = sum(map(len, chunks))
1276
content = record.get_bytes_as('fulltext')
1277
sha1 = osutils.sha_string(content)
1278
length = len(content)
1279
if item_data and sha1 != item_data[1]:
1280
checker._report_items.append(
1281
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1282
(record.key, sha1, item_data[1], item_data[2]))
1158
1285
def create(a_bzrdir):
1159
1286
"""Construct the current default format repository in a_bzrdir."""
1701
1860
@needs_read_lock
1702
1861
def get_revisions(self, revision_ids):
1703
"""Get many revisions at once."""
1862
"""Get many revisions at once.
1864
Repositories that need to check data on every revision read should
1865
subclass this method.
1704
1867
return self._get_revisions(revision_ids)
1706
1869
@needs_read_lock
1707
1870
def _get_revisions(self, revision_ids):
1708
1871
"""Core work logic to get many revisions without sanity checks."""
1709
for rev_id in revision_ids:
1710
if not rev_id or not isinstance(rev_id, basestring):
1711
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1873
for revid, rev in self._iter_revisions(revision_ids):
1875
raise errors.NoSuchRevision(self, revid)
1877
return [revs[revid] for revid in revision_ids]
1879
def _iter_revisions(self, revision_ids):
1880
"""Iterate over revision objects.
1882
:param revision_ids: An iterable of revisions to examine. None may be
1883
passed to request all revisions known to the repository. Note that
1884
not all repositories can find unreferenced revisions; for those
1885
repositories only referenced ones will be returned.
1886
:return: An iterator of (revid, revision) tuples. Absent revisions (
1887
those asked for but not available) are returned as (revid, None).
1889
if revision_ids is None:
1890
revision_ids = self.all_revision_ids()
1892
for rev_id in revision_ids:
1893
if not rev_id or not isinstance(rev_id, basestring):
1894
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1712
1895
keys = [(key,) for key in revision_ids]
1713
1896
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1715
1897
for record in stream:
1898
revid = record.key[0]
1716
1899
if record.storage_kind == 'absent':
1717
raise errors.NoSuchRevision(self, record.key[0])
1718
text = record.get_bytes_as('fulltext')
1719
rev = self._serializer.read_revision_from_string(text)
1720
revs[record.key[0]] = rev
1721
return [revs[revid] for revid in revision_ids]
1724
def get_revision_xml(self, revision_id):
1725
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1726
# would have already do it.
1727
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1728
# TODO: this can't just be replaced by:
1729
# return self._serializer.write_revision_to_string(
1730
# self.get_revision(revision_id))
1731
# as cStringIO preservers the encoding unlike write_revision_to_string
1732
# or some other call down the path.
1733
rev = self.get_revision(revision_id)
1734
rev_tmp = cStringIO.StringIO()
1735
# the current serializer..
1736
self._serializer.write_revision(rev, rev_tmp)
1738
return rev_tmp.getvalue()
1902
text = record.get_bytes_as('fulltext')
1903
rev = self._serializer.read_revision_from_string(text)
1740
1906
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1741
1907
"""Produce a generator of revision deltas.
2173
2357
:param revision_ids: The expected revision ids of the inventories.
2358
:param ordering: optional ordering, e.g. 'topological'. If not
2359
specified, the order of revision_ids will be preserved (by
2360
buffering if necessary).
2174
2361
:return: An iterator of inventories.
2176
2363
if ((None in revision_ids)
2177
2364
or (_mod_revision.NULL_REVISION in revision_ids)):
2178
2365
raise ValueError('cannot get null revision inventory')
2179
return self._iter_inventories(revision_ids)
2366
return self._iter_inventories(revision_ids, ordering)
2181
def _iter_inventories(self, revision_ids):
2368
def _iter_inventories(self, revision_ids, ordering):
2182
2369
"""single-document based inventory iteration."""
2183
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2184
yield self.deserialise_inventory(revision_id, text)
2370
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2371
for text, revision_id in inv_xmls:
2372
yield self._deserialise_inventory(revision_id, text)
2186
def _iter_inventory_xmls(self, revision_ids):
2374
def _iter_inventory_xmls(self, revision_ids, ordering):
2375
if ordering is None:
2376
order_as_requested = True
2377
ordering = 'unordered'
2379
order_as_requested = False
2187
2380
keys = [(revision_id,) for revision_id in revision_ids]
2188
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2383
if order_as_requested:
2384
key_iter = iter(keys)
2385
next_key = key_iter.next()
2386
stream = self.inventories.get_record_stream(keys, ordering, True)
2189
2387
text_chunks = {}
2190
2388
for record in stream:
2191
2389
if record.storage_kind != 'absent':
2192
text_chunks[record.key] = record.get_bytes_as('chunked')
2390
chunks = record.get_bytes_as('chunked')
2391
if order_as_requested:
2392
text_chunks[record.key] = chunks
2394
yield ''.join(chunks), record.key[-1]
2194
2396
raise errors.NoSuchRevision(self, record.key)
2196
chunks = text_chunks.pop(key)
2197
yield ''.join(chunks), key[-1]
2397
if order_as_requested:
2398
# Yield as many results as we can while preserving order.
2399
while next_key in text_chunks:
2400
chunks = text_chunks.pop(next_key)
2401
yield ''.join(chunks), next_key[-1]
2403
next_key = key_iter.next()
2404
except StopIteration:
2405
# We still want to fully consume the get_record_stream,
2406
# just in case it is not actually finished at this point
2199
def deserialise_inventory(self, revision_id, xml):
2410
def _deserialise_inventory(self, revision_id, xml):
2200
2411
"""Transform the xml into an inventory object.
2202
2413
:param revision_id: The expected revision id of the inventory.
2203
2414
:param xml: A serialised inventory.
2205
2416
result = self._serializer.read_inventory_from_string(xml, revision_id,
2206
entry_cache=self._inventory_entry_cache)
2417
entry_cache=self._inventory_entry_cache,
2418
return_from_cache=self._safe_to_return_from_cache)
2207
2419
if result.revision_id != revision_id:
2208
2420
raise AssertionError('revision id mismatch %s != %s' % (
2209
2421
result.revision_id, revision_id))
2212
def serialise_inventory(self, inv):
2213
return self._serializer.write_inventory_to_string(inv)
2215
def _serialise_inventory_to_lines(self, inv):
2216
return self._serializer.write_inventory_to_lines(inv)
2218
2424
def get_serializer_format(self):
2219
2425
return self._serializer.format_num
2221
2427
@needs_read_lock
2222
def get_inventory_xml(self, revision_id):
2223
"""Get inventory XML as a file object."""
2224
texts = self._iter_inventory_xmls([revision_id])
2428
def _get_inventory_xml(self, revision_id):
2429
"""Get serialized inventory as a string."""
2430
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2226
2432
text, revision_id = texts.next()
2227
2433
except StopIteration:
2228
2434
raise errors.HistoryMissing(self, 'inventory', revision_id)
2232
def get_inventory_sha1(self, revision_id):
2233
"""Return the sha1 hash of the inventory entry
2235
return self.get_revision(revision_id).inventory_sha1
2237
2437
def get_rev_id_for_revno(self, revno, known_pair):
2238
2438
"""Return the revision id of a revno, given a later (revno, revid)
2239
2439
pair in the same history.
3453
3706
return self.source.revision_ids_to_search_result(result_set)
3456
class InterPackRepo(InterSameDataRepository):
3457
"""Optimised code paths between Pack based repositories."""
3460
def _get_repo_format_to_test(self):
3461
from bzrlib.repofmt import pack_repo
3462
return pack_repo.RepositoryFormatKnitPack6RichRoot()
3465
def is_compatible(source, target):
3466
"""Be compatible with known Pack formats.
3468
We don't test for the stores being of specific types because that
3469
could lead to confusing results, and there is no need to be
3472
InterPackRepo does not support CHK based repositories.
3474
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3475
from bzrlib.repofmt.groupcompress_repo import RepositoryFormatCHK1
3477
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3478
isinstance(target._format, RepositoryFormatPack))
3479
not_packs = (isinstance(source._format, RepositoryFormatCHK1) or
3480
isinstance(target._format, RepositoryFormatCHK1))
3481
except AttributeError:
3483
if not_packs or not are_packs:
3485
return InterRepository._same_model(source, target)
3488
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3490
"""See InterRepository.fetch()."""
3491
if (len(self.source._fallback_repositories) > 0 or
3492
len(self.target._fallback_repositories) > 0):
3493
# The pack layer is not aware of fallback repositories, so when
3494
# fetching from a stacked repository or into a stacked repository
3495
# we use the generic fetch logic which uses the VersionedFiles
3496
# attributes on repository.
3497
from bzrlib.fetch import RepoFetcher
3498
fetcher = RepoFetcher(self.target, self.source, revision_id,
3499
pb, find_ghosts, fetch_spec=fetch_spec)
3500
if fetch_spec is not None:
3501
if len(list(fetch_spec.heads)) != 1:
3502
raise AssertionError(
3503
"InterPackRepo.fetch doesn't support "
3504
"fetching multiple heads yet.")
3505
revision_id = list(fetch_spec.heads)[0]
3507
if revision_id is None:
3509
# everything to do - use pack logic
3510
# to fetch from all packs to one without
3511
# inventory parsing etc, IFF nothing to be copied is in the target.
3513
source_revision_ids = frozenset(self.source.all_revision_ids())
3514
revision_ids = source_revision_ids - \
3515
frozenset(self.target.get_parent_map(source_revision_ids))
3516
revision_keys = [(revid,) for revid in revision_ids]
3517
index = self.target._pack_collection.revision_index.combined_index
3518
present_revision_ids = set(item[1][0] for item in
3519
index.iter_entries(revision_keys))
3520
revision_ids = set(revision_ids) - present_revision_ids
3521
# implementing the TODO will involve:
3522
# - detecting when all of a pack is selected
3523
# - avoiding as much as possible pre-selection, so the
3524
# more-core routines such as create_pack_from_packs can filter in
3525
# a just-in-time fashion. (though having a HEADS list on a
3526
# repository might make this a lot easier, because we could
3527
# sensibly detect 'new revisions' without doing a full index scan.
3528
elif _mod_revision.is_null(revision_id):
3532
revision_ids = self.search_missing_revision_ids(revision_id,
3533
find_ghosts=find_ghosts).get_keys()
3534
if len(revision_ids) == 0:
3536
return self._pack(self.source, self.target, revision_ids)
3538
def _pack(self, source, target, revision_ids):
3539
from bzrlib.repofmt.pack_repo import Packer
3540
packs = source._pack_collection.all_packs()
3541
pack = Packer(self.target._pack_collection, packs, '.fetch',
3542
revision_ids).pack()
3543
if pack is not None:
3544
self.target._pack_collection._save_pack_names()
3545
copied_revs = pack.get_revision_count()
3546
# Trigger an autopack. This may duplicate effort as we've just done
3547
# a pack creation, but for now it is simpler to think about as
3548
# 'upload data, then repack if needed'.
3549
self.target._pack_collection.autopack()
3550
return (copied_revs, [])
3555
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3556
"""See InterRepository.missing_revision_ids().
3558
:param find_ghosts: Find ghosts throughout the ancestry of
3561
if not find_ghosts and revision_id is not None:
3562
return self._walk_to_common_revisions([revision_id])
3563
elif revision_id is not None:
3564
# Find ghosts: search for revisions pointing from one repository to
3565
# the other, and vice versa, anywhere in the history of revision_id.
3566
graph = self.target.get_graph(other_repository=self.source)
3567
searcher = graph._make_breadth_first_searcher([revision_id])
3571
next_revs, ghosts = searcher.next_with_ghosts()
3572
except StopIteration:
3574
if revision_id in ghosts:
3575
raise errors.NoSuchRevision(self.source, revision_id)
3576
found_ids.update(next_revs)
3577
found_ids.update(ghosts)
3578
found_ids = frozenset(found_ids)
3579
# Double query here: should be able to avoid this by changing the
3580
# graph api further.
3581
result_set = found_ids - frozenset(
3582
self.target.get_parent_map(found_ids))
3584
source_ids = self.source.all_revision_ids()
3585
# source_ids is the worst possible case we may need to pull.
3586
# now we want to filter source_ids against what we actually
3587
# have in target, but don't try to check for existence where we know
3588
# we do not have a revision as that would be pointless.
3589
target_ids = set(self.target.all_revision_ids())
3590
result_set = set(source_ids).difference(target_ids)
3591
return self.source.revision_ids_to_search_result(result_set)
3594
3709
class InterDifferingSerializer(InterRepository):
3603
3718
# This is redundant with format.check_conversion_target(), however that
3604
3719
# raises an exception, and we just want to say "False" as in we won't
3605
3720
# support converting between these formats.
3721
if 'IDS_never' in debug.debug_flags:
3606
3723
if source.supports_rich_root() and not target.supports_rich_root():
3608
3725
if (source._format.supports_tree_reference
3609
3726
and not target._format.supports_tree_reference):
3728
if target._fallback_repositories and target._format.supports_chks:
3729
# IDS doesn't know how to copy CHKs for the parent inventories it
3730
# adds to stacked repos.
3732
if 'IDS_always' in debug.debug_flags:
3734
# Only use this code path for local source and target. IDS does far
3735
# too much IO (both bandwidth and roundtrips) over a network.
3736
if not source.bzrdir.transport.base.startswith('file:///'):
3738
if not target.bzrdir.transport.base.startswith('file:///'):
3613
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3742
def _get_trees(self, revision_ids, cache):
3744
for rev_id in revision_ids:
3746
possible_trees.append((rev_id, cache[rev_id]))
3748
# Not cached, but inventory might be present anyway.
3750
tree = self.source.revision_tree(rev_id)
3751
except errors.NoSuchRevision:
3752
# Nope, parent is ghost.
3755
cache[rev_id] = tree
3756
possible_trees.append((rev_id, tree))
3757
return possible_trees
3759
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3614
3760
"""Get the best delta and base for this revision.
3616
3762
:return: (basis_id, delta)
3618
possible_trees = [(parent_id, cache[parent_id])
3619
for parent_id in parent_ids
3620
if parent_id in cache]
3621
if len(possible_trees) == 0:
3622
# There either aren't any parents, or the parents aren't in the
3623
# cache, so just use the last converted tree
3624
possible_trees.append((basis_id, cache[basis_id]))
3765
# Generate deltas against each tree, to find the shortest.
3766
texts_possibly_new_in_tree = set()
3626
3767
for basis_id, basis_tree in possible_trees:
3627
3768
delta = tree.inventory._make_delta(basis_tree.inventory)
3769
for old_path, new_path, file_id, new_entry in delta:
3770
if new_path is None:
3771
# This file_id isn't present in the new rev, so we don't
3775
# Rich roots are handled elsewhere...
3777
kind = new_entry.kind
3778
if kind != 'directory' and kind != 'file':
3779
# No text record associated with this inventory entry.
3781
# This is a directory or file that has changed somehow.
3782
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3628
3783
deltas.append((len(delta), basis_id, delta))
3630
3785
return deltas[0][1:]
3632
def _get_parent_keys(self, root_key, parent_map):
3633
"""Get the parent keys for a given root id."""
3634
root_id, rev_id = root_key
3635
# Include direct parents of the revision, but only if they used
3636
# the same root_id and are heads.
3638
for parent_id in parent_map[rev_id]:
3639
if parent_id == _mod_revision.NULL_REVISION:
3641
if parent_id not in self._revision_id_to_root_id:
3642
# We probably didn't read this revision, go spend the
3643
# extra effort to actually check
3645
tree = self.source.revision_tree(parent_id)
3646
except errors.NoSuchRevision:
3647
# Ghost, fill out _revision_id_to_root_id in case we
3648
# encounter this again.
3649
# But set parent_root_id to None since we don't really know
3650
parent_root_id = None
3652
parent_root_id = tree.get_root_id()
3653
self._revision_id_to_root_id[parent_id] = None
3655
parent_root_id = self._revision_id_to_root_id[parent_id]
3656
if root_id == parent_root_id:
3657
# With stacking we _might_ want to refer to a non-local
3658
# revision, but this code path only applies when we have the
3659
# full content available, so ghosts really are ghosts, not just
3660
# the edge of local data.
3661
parent_keys.append((parent_id,))
3663
# root_id may be in the parent anyway.
3665
tree = self.source.revision_tree(parent_id)
3666
except errors.NoSuchRevision:
3667
# ghost, can't refer to it.
3671
parent_keys.append((tree.inventory[root_id].revision,))
3672
except errors.NoSuchId:
3675
g = graph.Graph(self.source.revisions)
3676
heads = g.heads(parent_keys)
3678
for key in parent_keys:
3679
if key in heads and key not in selected_keys:
3680
selected_keys.append(key)
3681
return tuple([(root_id,)+ key for key in selected_keys])
3683
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3684
for root_key in root_keys_to_create:
3685
parent_keys = self._get_parent_keys(root_key, parent_map)
3686
yield versionedfile.FulltextContentFactory(root_key,
3687
parent_keys, None, '')
3689
def _fetch_batch(self, revision_ids, basis_id, cache):
3787
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3788
"""Find all parent revisions that are absent, but for which the
3789
inventory is present, and copy those inventories.
3791
This is necessary to preserve correctness when the source is stacked
3792
without fallbacks configured. (Note that in cases like upgrade the
3793
source may be not have _fallback_repositories even though it is
3797
for parents in parent_map.values():
3798
parent_revs.update(parents)
3799
present_parents = self.source.get_parent_map(parent_revs)
3800
absent_parents = set(parent_revs).difference(present_parents)
3801
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3802
(rev_id,) for rev_id in absent_parents)
3803
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3804
for parent_tree in self.source.revision_trees(parent_inv_ids):
3805
current_revision_id = parent_tree.get_revision_id()
3806
parents_parents_keys = parent_invs_keys_for_stacking[
3807
(current_revision_id,)]
3808
parents_parents = [key[-1] for key in parents_parents_keys]
3809
basis_id = _mod_revision.NULL_REVISION
3810
basis_tree = self.source.revision_tree(basis_id)
3811
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3812
self.target.add_inventory_by_delta(
3813
basis_id, delta, current_revision_id, parents_parents)
3814
cache[current_revision_id] = parent_tree
3816
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3690
3817
"""Fetch across a few revisions.
3692
3819
:param revision_ids: The revisions to copy
3693
3820
:param basis_id: The revision_id of a tree that must be in cache, used
3694
3821
as a basis for delta when no other base is available
3695
3822
:param cache: A cache of RevisionTrees that we can use.
3823
:param a_graph: A Graph object to determine the heads() of the
3824
rich-root data stream.
3696
3825
:return: The revision_id of the last converted tree. The RevisionTree
3697
3826
for it will be in cache
3704
3833
pending_deltas = []
3705
3834
pending_revisions = []
3706
3835
parent_map = self.source.get_parent_map(revision_ids)
3836
self._fetch_parent_invs_for_stacking(parent_map, cache)
3837
self.source._safe_to_return_from_cache = True
3707
3838
for tree in self.source.revision_trees(revision_ids):
3839
# Find a inventory delta for this revision.
3840
# Find text entries that need to be copied, too.
3708
3841
current_revision_id = tree.get_revision_id()
3709
3842
parent_ids = parent_map.get(current_revision_id, ())
3843
parent_trees = self._get_trees(parent_ids, cache)
3844
possible_trees = list(parent_trees)
3845
if len(possible_trees) == 0:
3846
# There either aren't any parents, or the parents are ghosts,
3847
# so just use the last converted tree.
3848
possible_trees.append((basis_id, cache[basis_id]))
3710
3849
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3851
revision = self.source.get_revision(current_revision_id)
3852
pending_deltas.append((basis_id, delta,
3853
current_revision_id, revision.parent_ids))
3712
3854
if self._converting_to_rich_root:
3713
3855
self._revision_id_to_root_id[current_revision_id] = \
3714
3856
tree.get_root_id()
3715
# Find text entries that need to be copied
3857
# Determine which texts are in present in this revision but not in
3858
# any of the available parents.
3859
texts_possibly_new_in_tree = set()
3716
3860
for old_path, new_path, file_id, entry in delta:
3717
if new_path is not None:
3720
if not self.target.supports_rich_root():
3721
# The target doesn't support rich root, so we don't
3724
if self._converting_to_rich_root:
3725
# This can't be copied normally, we have to insert
3727
root_keys_to_create.add((file_id, entry.revision))
3729
text_keys.add((file_id, entry.revision))
3730
revision = self.source.get_revision(current_revision_id)
3731
pending_deltas.append((basis_id, delta,
3732
current_revision_id, revision.parent_ids))
3861
if new_path is None:
3862
# This file_id isn't present in the new rev
3866
if not self.target.supports_rich_root():
3867
# The target doesn't support rich root, so we don't
3870
if self._converting_to_rich_root:
3871
# This can't be copied normally, we have to insert
3873
root_keys_to_create.add((file_id, entry.revision))
3876
texts_possibly_new_in_tree.add((file_id, entry.revision))
3877
for basis_id, basis_tree in possible_trees:
3878
basis_inv = basis_tree.inventory
3879
for file_key in list(texts_possibly_new_in_tree):
3880
file_id, file_revision = file_key
3882
entry = basis_inv[file_id]
3883
except errors.NoSuchId:
3885
if entry.revision == file_revision:
3886
texts_possibly_new_in_tree.remove(file_key)
3887
text_keys.update(texts_possibly_new_in_tree)
3733
3888
pending_revisions.append(revision)
3734
3889
cache[current_revision_id] = tree
3735
3890
basis_id = current_revision_id
3891
self.source._safe_to_return_from_cache = False
3736
3892
# Copy file texts
3737
3893
from_texts = self.source.texts
3738
3894
to_texts = self.target.texts
3739
3895
if root_keys_to_create:
3740
root_stream = self._new_root_data_stream(root_keys_to_create,
3896
root_stream = _mod_fetch._new_root_data_stream(
3897
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3898
self.source, graph=a_graph)
3742
3899
to_texts.insert_record_stream(root_stream)
3743
3900
to_texts.insert_record_stream(from_texts.get_record_stream(
3744
3901
text_keys, self.target._format._fetch_order,
4150
4350
# missing keys can handle suspending a write group).
4151
4351
write_group_tokens = self.target_repo.suspend_write_group()
4152
4352
return write_group_tokens, missing_keys
4153
self.target_repo.commit_write_group()
4353
hint = self.target_repo.commit_write_group()
4354
if (to_serializer != src_serializer and
4355
self.target_repo._format.pack_compresses):
4356
self.target_repo.pack(hint=hint)
4154
4357
return [], set()
4156
def _extract_and_insert_inventories(self, substream, serializer):
4359
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4360
target_rich_root = self.target_repo._format.rich_root_data
4361
target_tree_refs = self.target_repo._format.supports_tree_reference
4362
for record in substream:
4363
# Insert the delta directly
4364
inventory_delta_bytes = record.get_bytes_as('fulltext')
4365
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4367
parse_result = deserialiser.parse_text_bytes(
4368
inventory_delta_bytes)
4369
except inventory_delta.IncompatibleInventoryDelta, err:
4370
trace.mutter("Incompatible delta: %s", err.msg)
4371
raise errors.IncompatibleRevision(self.target_repo._format)
4372
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4373
revision_id = new_id
4374
parents = [key[0] for key in record.parents]
4375
self.target_repo.add_inventory_by_delta(
4376
basis_id, inv_delta, revision_id, parents)
4378
def _extract_and_insert_inventories(self, substream, serializer,
4157
4380
"""Generate a new inventory versionedfile in target, converting data.
4159
4382
The inventory is retrieved from the source, (deserializing it), and
4160
4383
stored in the target (reserializing it in a different format).
4385
target_rich_root = self.target_repo._format.rich_root_data
4386
target_tree_refs = self.target_repo._format.supports_tree_reference
4162
4387
for record in substream:
4388
# It's not a delta, so it must be a fulltext in the source
4389
# serializer's format.
4163
4390
bytes = record.get_bytes_as('fulltext')
4164
4391
revision_id = record.key[0]
4165
4392
inv = serializer.read_inventory_from_string(bytes, revision_id)
4166
4393
parents = [key[0] for key in record.parents]
4167
4394
self.target_repo.add_inventory(revision_id, inv, parents)
4395
# No need to keep holding this full inv in memory when the rest of
4396
# the substream is likely to be all deltas.
4169
4399
def _extract_and_insert_revisions(self, substream, serializer):
4170
4400
for record in substream:
4328
4563
return (not self.from_repository._format.rich_root_data and
4329
4564
self.to_format.rich_root_data)
4331
def _get_inventory_stream(self, revision_ids):
4566
def _get_inventory_stream(self, revision_ids, missing=False):
4332
4567
from_format = self.from_repository._format
4333
if (from_format.supports_chks and self.to_format.supports_chks
4334
and (from_format._serializer == self.to_format._serializer)):
4335
# Both sides support chks, and they use the same serializer, so it
4336
# is safe to transmit the chk pages and inventory pages across
4338
return self._get_chk_inventory_stream(revision_ids)
4339
elif (not from_format.supports_chks):
4340
# Source repository doesn't support chks. So we can transmit the
4341
# inventories 'as-is' and either they are just accepted on the
4342
# target, or the Sink will properly convert it.
4343
return self._get_simple_inventory_stream(revision_ids)
4568
if (from_format.supports_chks and self.to_format.supports_chks and
4569
from_format.network_name() == self.to_format.network_name()):
4570
raise AssertionError(
4571
"this case should be handled by GroupCHKStreamSource")
4572
elif 'forceinvdeltas' in debug.debug_flags:
4573
return self._get_convertable_inventory_stream(revision_ids,
4574
delta_versus_null=missing)
4575
elif from_format.network_name() == self.to_format.network_name():
4577
return self._get_simple_inventory_stream(revision_ids,
4579
elif (not from_format.supports_chks and not self.to_format.supports_chks
4580
and from_format._serializer == self.to_format._serializer):
4581
# Essentially the same format.
4582
return self._get_simple_inventory_stream(revision_ids,
4345
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4346
# inventories. Note that this should probably be done somehow
4347
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4348
# figure out how a non-chk repository could possibly handle
4349
# deserializing an inventory stream from a chk repo, as it
4350
# doesn't have a way to understand individual pages.
4351
return self._get_convertable_inventory_stream(revision_ids)
4585
# Any time we switch serializations, we want to use an
4586
# inventory-delta based approach.
4587
return self._get_convertable_inventory_stream(revision_ids,
4588
delta_versus_null=missing)
4353
def _get_simple_inventory_stream(self, revision_ids):
4590
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4591
# NB: This currently reopens the inventory weave in source;
4592
# using a single stream interface instead would avoid this.
4354
4593
from_weave = self.from_repository.inventories
4595
delta_closure = True
4597
delta_closure = not self.delta_on_metadata()
4355
4598
yield ('inventories', from_weave.get_record_stream(
4356
4599
[(rev_id,) for rev_id in revision_ids],
4357
self.inventory_fetch_order(),
4358
not self.delta_on_metadata()))
4360
def _get_chk_inventory_stream(self, revision_ids):
4361
"""Fetch the inventory texts, along with the associated chk maps."""
4362
# We want an inventory outside of the search set, so that we can filter
4363
# out uninteresting chk pages. For now we use
4364
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4365
# could use that instead.
4366
start_rev_id = self.from_repository._find_revision_outside_set(
4368
start_rev_key = (start_rev_id,)
4369
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4370
if start_rev_id != _mod_revision.NULL_REVISION:
4371
inv_keys_to_fetch.append((start_rev_id,))
4372
# Any repo that supports chk_bytes must also support out-of-order
4373
# insertion. At least, that is how we expect it to work
4374
# We use get_record_stream instead of iter_inventories because we want
4375
# to be able to insert the stream as well. We could instead fetch
4376
# allowing deltas, and then iter_inventories, but we don't know whether
4377
# source or target is more 'local' anway.
4378
inv_stream = self.from_repository.inventories.get_record_stream(
4379
inv_keys_to_fetch, 'unordered',
4380
True) # We need them as full-texts so we can find their references
4381
uninteresting_chk_roots = set()
4382
interesting_chk_roots = set()
4383
def filter_inv_stream(inv_stream):
4384
for idx, record in enumerate(inv_stream):
4385
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4386
bytes = record.get_bytes_as('fulltext')
4387
chk_inv = inventory.CHKInventory.deserialise(
4388
self.from_repository.chk_bytes, bytes, record.key)
4389
if record.key == start_rev_key:
4390
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4391
p_id_map = chk_inv.parent_id_basename_to_file_id
4392
if p_id_map is not None:
4393
uninteresting_chk_roots.add(p_id_map.key())
4396
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4397
p_id_map = chk_inv.parent_id_basename_to_file_id
4398
if p_id_map is not None:
4399
interesting_chk_roots.add(p_id_map.key())
4400
### pb.update('fetch inventory', 0, 2)
4401
yield ('inventories', filter_inv_stream(inv_stream))
4402
# Now that we have worked out all of the interesting root nodes, grab
4403
# all of the interesting pages and insert them
4404
### pb.update('fetch inventory', 1, 2)
4405
interesting = chk_map.iter_interesting_nodes(
4406
self.from_repository.chk_bytes, interesting_chk_roots,
4407
uninteresting_chk_roots)
4408
def to_stream_adapter():
4409
"""Adapt the iter_interesting_nodes result to a single stream.
4411
iter_interesting_nodes returns records as it processes them, along
4412
with keys. However, we only want to return the records themselves.
4414
for record, items in interesting:
4415
if record is not None:
4417
# XXX: We could instead call get_record_stream(records.keys())
4418
# ATM, this will always insert the records as fulltexts, and
4419
# requires that you can hang on to records once you have gone
4420
# on to the next one. Further, it causes the target to
4421
# recompress the data. Testing shows it to be faster than
4422
# requesting the records again, though.
4423
yield ('chk_bytes', to_stream_adapter())
4424
### pb.update('fetch inventory', 2, 2)
4426
def _get_convertable_inventory_stream(self, revision_ids):
4427
# XXX: One of source or target is using chks, and they don't have
4428
# compatible serializations. The StreamSink code expects to be
4429
# able to convert on the target, so we need to put
4430
# bytes-on-the-wire that can be converted
4431
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4433
def _stream_invs_as_fulltexts(self, revision_ids):
4600
self.inventory_fetch_order(), delta_closure))
4602
def _get_convertable_inventory_stream(self, revision_ids,
4603
delta_versus_null=False):
4604
# The two formats are sufficiently different that there is no fast
4605
# path, so we need to send just inventorydeltas, which any
4606
# sufficiently modern client can insert into any repository.
4607
# The StreamSink code expects to be able to
4608
# convert on the target, so we need to put bytes-on-the-wire that can
4609
# be converted. That means inventory deltas (if the remote is <1.19,
4610
# RemoteStreamSink will fallback to VFS to insert the deltas).
4611
yield ('inventory-deltas',
4612
self._stream_invs_as_deltas(revision_ids,
4613
delta_versus_null=delta_versus_null))
4615
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4616
"""Return a stream of inventory-deltas for the given rev ids.
4618
:param revision_ids: The list of inventories to transmit
4619
:param delta_versus_null: Don't try to find a minimal delta for this
4620
entry, instead compute the delta versus the NULL_REVISION. This
4621
effectively streams a complete inventory. Used for stuff like
4622
filling in missing parents, etc.
4434
4624
from_repo = self.from_repository
4435
from_serializer = from_repo._format._serializer
4436
4625
revision_keys = [(rev_id,) for rev_id in revision_ids]
4437
4626
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4438
for inv in self.from_repository.iter_inventories(revision_ids):
4439
# XXX: This is a bit hackish, but it works. Basically,
4440
# CHKSerializer 'accidentally' supports
4441
# read/write_inventory_to_string, even though that is never
4442
# the format that is stored on disk. It *does* give us a
4443
# single string representation for an inventory, so live with
4445
# This would be far better if we had a 'serialized inventory
4446
# delta' form. Then we could use 'inventory._make_delta', and
4447
# transmit that. This would both be faster to generate, and
4448
# result in fewer bytes-on-the-wire.
4449
as_bytes = from_serializer.write_inventory_to_string(inv)
4627
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4629
inventories = self.from_repository.iter_inventories(
4630
revision_ids, 'topological')
4631
format = from_repo._format
4632
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4633
inventory_cache = lru_cache.LRUCache(50)
4634
null_inventory = from_repo.revision_tree(
4635
_mod_revision.NULL_REVISION).inventory
4636
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4637
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4638
# repo back into a non-rich-root repo ought to be allowed)
4639
serializer = inventory_delta.InventoryDeltaSerializer(
4640
versioned_root=format.rich_root_data,
4641
tree_references=format.supports_tree_reference)
4642
for inv in inventories:
4450
4643
key = (inv.revision_id,)
4451
4644
parent_keys = parent_map.get(key, ())
4646
if not delta_versus_null and parent_keys:
4647
# The caller did not ask for complete inventories and we have
4648
# some parents that we can delta against. Make a delta against
4649
# each parent so that we can find the smallest.
4650
parent_ids = [parent_key[0] for parent_key in parent_keys]
4651
for parent_id in parent_ids:
4652
if parent_id not in invs_sent_so_far:
4653
# We don't know that the remote side has this basis, so
4656
if parent_id == _mod_revision.NULL_REVISION:
4657
parent_inv = null_inventory
4659
parent_inv = inventory_cache.get(parent_id, None)
4660
if parent_inv is None:
4661
parent_inv = from_repo.get_inventory(parent_id)
4662
candidate_delta = inv._make_delta(parent_inv)
4663
if (delta is None or
4664
len(delta) > len(candidate_delta)):
4665
delta = candidate_delta
4666
basis_id = parent_id
4668
# Either none of the parents ended up being suitable, or we
4669
# were asked to delta against NULL
4670
basis_id = _mod_revision.NULL_REVISION
4671
delta = inv._make_delta(null_inventory)
4672
invs_sent_so_far.add(inv.revision_id)
4673
inventory_cache[inv.revision_id] = inv
4674
delta_serialized = ''.join(
4675
serializer.delta_to_lines(basis_id, key[-1], delta))
4452
4676
yield versionedfile.FulltextContentFactory(
4453
key, parent_keys, None, as_bytes)
4677
key, parent_keys, None, delta_serialized)
4456
4680
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
added
removed
bzrlib/repository.py
