1185
1154
# The old API returned a list, should this actually be a set?
1186
1155
return parent_map.keys()
1188
def _check_inventories(self, checker):
1189
"""Check the inventories found from the revision scan.
1191
This is responsible for verifying the sha1 of inventories and
1192
creating a pending_keys set that covers data referenced by inventories.
1194
bar = ui.ui_factory.nested_progress_bar()
1196
self._do_check_inventories(checker, bar)
1200
def _do_check_inventories(self, checker, bar):
1201
"""Helper for _check_inventories."""
1203
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1204
kinds = ['chk_bytes', 'texts']
1205
count = len(checker.pending_keys)
1206
bar.update("inventories", 0, 2)
1207
current_keys = checker.pending_keys
1208
checker.pending_keys = {}
1209
# Accumulate current checks.
1210
for key in current_keys:
1211
if key[0] != 'inventories' and key[0] not in kinds:
1212
checker._report_items.append('unknown key type %r' % (key,))
1213
keys[key[0]].add(key[1:])
1214
if keys['inventories']:
1215
# NB: output order *should* be roughly sorted - topo or
1216
# inverse topo depending on repository - either way decent
1217
# to just delta against. However, pre-CHK formats didn't
1218
# try to optimise inventory layout on disk. As such the
1219
# pre-CHK code path does not use inventory deltas.
1221
for record in self.inventories.check(keys=keys['inventories']):
1222
if record.storage_kind == 'absent':
1223
checker._report_items.append(
1224
'Missing inventory {%s}' % (record.key,))
1226
last_object = self._check_record('inventories', record,
1227
checker, last_object,
1228
current_keys[('inventories',) + record.key])
1229
del keys['inventories']
1232
bar.update("texts", 1)
1233
while (checker.pending_keys or keys['chk_bytes']
1235
# Something to check.
1236
current_keys = checker.pending_keys
1237
checker.pending_keys = {}
1238
# Accumulate current checks.
1239
for key in current_keys:
1240
if key[0] not in kinds:
1241
checker._report_items.append('unknown key type %r' % (key,))
1242
keys[key[0]].add(key[1:])
1243
# Check the outermost kind only - inventories || chk_bytes || texts
1247
for record in getattr(self, kind).check(keys=keys[kind]):
1248
if record.storage_kind == 'absent':
1249
checker._report_items.append(
1250
'Missing %s {%s}' % (kind, record.key,))
1252
last_object = self._check_record(kind, record,
1253
checker, last_object, current_keys[(kind,) + record.key])
1257
def _check_record(self, kind, record, checker, last_object, item_data):
1258
"""Check a single text from this repository."""
1259
if kind == 'inventories':
1260
rev_id = record.key[0]
1261
inv = self._deserialise_inventory(rev_id,
1262
record.get_bytes_as('fulltext'))
1263
if last_object is not None:
1264
delta = inv._make_delta(last_object)
1265
for old_path, path, file_id, ie in delta:
1268
ie.check(checker, rev_id, inv)
1270
for path, ie in inv.iter_entries():
1271
ie.check(checker, rev_id, inv)
1272
if self._format.fast_deltas:
1274
elif kind == 'chk_bytes':
1275
# No code written to check chk_bytes for this repo format.
1276
checker._report_items.append(
1277
'unsupported key type chk_bytes for %s' % (record.key,))
1278
elif kind == 'texts':
1279
self._check_text(record, checker, item_data)
1281
checker._report_items.append(
1282
'unknown key type %s for %s' % (kind, record.key))
1284
def _check_text(self, record, checker, item_data):
1285
"""Check a single text."""
1286
# Check it is extractable.
1287
# TODO: check length.
1288
if record.storage_kind == 'chunked':
1289
chunks = record.get_bytes_as(record.storage_kind)
1290
sha1 = osutils.sha_strings(chunks)
1291
length = sum(map(len, chunks))
1293
content = record.get_bytes_as('fulltext')
1294
sha1 = osutils.sha_string(content)
1295
length = len(content)
1296
if item_data and sha1 != item_data[1]:
1297
checker._report_items.append(
1298
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1299
(record.key, sha1, item_data[1], item_data[2]))
1302
1158
def create(a_bzrdir):
1303
1159
"""Construct the current default format repository in a_bzrdir."""
1884
1701
@needs_read_lock
1885
1702
def get_revisions(self, revision_ids):
1886
"""Get many revisions at once.
1888
Repositories that need to check data on every revision read should
1889
subclass this method.
1703
"""Get many revisions at once."""
1891
1704
return self._get_revisions(revision_ids)
1893
1706
@needs_read_lock
1894
1707
def _get_revisions(self, revision_ids):
1895
1708
"""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)
1712
keys = [(key,) for key in revision_ids]
1713
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1897
for revid, rev in self._iter_revisions(revision_ids):
1899
raise errors.NoSuchRevision(self, revid)
1715
for record in stream:
1716
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
1901
1721
return [revs[revid] for revid in revision_ids]
1903
def _iter_revisions(self, revision_ids):
1904
"""Iterate over revision objects.
1906
:param revision_ids: An iterable of revisions to examine. None may be
1907
passed to request all revisions known to the repository. Note that
1908
not all repositories can find unreferenced revisions; for those
1909
repositories only referenced ones will be returned.
1910
:return: An iterator of (revid, revision) tuples. Absent revisions (
1911
those asked for but not available) are returned as (revid, None).
1913
if revision_ids is None:
1914
revision_ids = self.all_revision_ids()
1916
for rev_id in revision_ids:
1917
if not rev_id or not isinstance(rev_id, basestring):
1918
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1919
keys = [(key,) for key in revision_ids]
1920
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1921
for record in stream:
1922
revid = record.key[0]
1923
if record.storage_kind == 'absent':
1926
text = record.get_bytes_as('fulltext')
1927
rev = self._serializer.read_revision_from_string(text)
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()
1930
1740
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1931
1741
"""Produce a generator of revision deltas.
2381
2173
:param revision_ids: The expected revision ids of the inventories.
2382
:param ordering: optional ordering, e.g. 'topological'. If not
2383
specified, the order of revision_ids will be preserved (by
2384
buffering if necessary).
2385
2174
:return: An iterator of inventories.
2387
2176
if ((None in revision_ids)
2388
2177
or (_mod_revision.NULL_REVISION in revision_ids)):
2389
2178
raise ValueError('cannot get null revision inventory')
2390
return self._iter_inventories(revision_ids, ordering)
2179
return self._iter_inventories(revision_ids)
2392
def _iter_inventories(self, revision_ids, ordering):
2181
def _iter_inventories(self, revision_ids):
2393
2182
"""single-document based inventory iteration."""
2394
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2395
for text, revision_id in inv_xmls:
2396
yield self._deserialise_inventory(revision_id, text)
2183
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2184
yield self.deserialise_inventory(revision_id, text)
2398
def _iter_inventory_xmls(self, revision_ids, ordering):
2399
if ordering is None:
2400
order_as_requested = True
2401
ordering = 'unordered'
2403
order_as_requested = False
2186
def _iter_inventory_xmls(self, revision_ids):
2404
2187
keys = [(revision_id,) for revision_id in revision_ids]
2407
if order_as_requested:
2408
key_iter = iter(keys)
2409
next_key = key_iter.next()
2410
stream = self.inventories.get_record_stream(keys, ordering, True)
2188
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2411
2189
text_chunks = {}
2412
2190
for record in stream:
2413
2191
if record.storage_kind != 'absent':
2414
chunks = record.get_bytes_as('chunked')
2415
if order_as_requested:
2416
text_chunks[record.key] = chunks
2418
yield ''.join(chunks), record.key[-1]
2192
text_chunks[record.key] = record.get_bytes_as('chunked')
2420
2194
raise errors.NoSuchRevision(self, record.key)
2421
if order_as_requested:
2422
# Yield as many results as we can while preserving order.
2423
while next_key in text_chunks:
2424
chunks = text_chunks.pop(next_key)
2425
yield ''.join(chunks), next_key[-1]
2427
next_key = key_iter.next()
2428
except StopIteration:
2429
# We still want to fully consume the get_record_stream,
2430
# just in case it is not actually finished at this point
2196
chunks = text_chunks.pop(key)
2197
yield ''.join(chunks), key[-1]
2434
def _deserialise_inventory(self, revision_id, xml):
2199
def deserialise_inventory(self, revision_id, xml):
2435
2200
"""Transform the xml into an inventory object.
2437
2202
:param revision_id: The expected revision id of the inventory.
2438
2203
:param xml: A serialised inventory.
2440
2205
result = self._serializer.read_inventory_from_string(xml, revision_id,
2441
entry_cache=self._inventory_entry_cache,
2442
return_from_cache=self._safe_to_return_from_cache)
2206
entry_cache=self._inventory_entry_cache)
2443
2207
if result.revision_id != revision_id:
2444
2208
raise AssertionError('revision id mismatch %s != %s' % (
2445
2209
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)
2448
2218
def get_serializer_format(self):
2449
2219
return self._serializer.format_num
2451
2221
@needs_read_lock
2452
def _get_inventory_xml(self, revision_id):
2453
"""Get serialized inventory as a string."""
2454
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2222
def get_inventory_xml(self, revision_id):
2223
"""Get inventory XML as a file object."""
2224
texts = self._iter_inventory_xmls([revision_id])
2456
2226
text, revision_id = texts.next()
2457
2227
except StopIteration:
2458
2228
raise errors.HistoryMissing(self, 'inventory', revision_id)
2461
def get_rev_id_for_revno(self, revno, known_pair):
2462
"""Return the revision id of a revno, given a later (revno, revid)
2463
pair in the same history.
2465
:return: if found (True, revid). If the available history ran out
2466
before reaching the revno, then this returns
2467
(False, (closest_revno, closest_revid)).
2232
def get_inventory_sha1(self, revision_id):
2233
"""Return the sha1 hash of the inventory entry
2469
known_revno, known_revid = known_pair
2470
partial_history = [known_revid]
2471
distance_from_known = known_revno - revno
2472
if distance_from_known < 0:
2474
'requested revno (%d) is later than given known revno (%d)'
2475
% (revno, known_revno))
2478
self, partial_history, stop_index=distance_from_known)
2479
except errors.RevisionNotPresent, err:
2480
if err.revision_id == known_revid:
2481
# The start revision (known_revid) wasn't found.
2483
# This is a stacked repository with no fallbacks, or a there's a
2484
# left-hand ghost. Either way, even though the revision named in
2485
# the error isn't in this repo, we know it's the next step in this
2486
# left-hand history.
2487
partial_history.append(err.revision_id)
2488
if len(partial_history) <= distance_from_known:
2489
# Didn't find enough history to get a revid for the revno.
2490
earliest_revno = known_revno - len(partial_history) + 1
2491
return (False, (earliest_revno, partial_history[-1]))
2492
if len(partial_history) - 1 > distance_from_known:
2493
raise AssertionError('_iter_for_revno returned too much history')
2494
return (True, partial_history[-1])
2235
return self.get_revision(revision_id).inventory_sha1
2496
2237
def iter_reverse_revision_history(self, revision_id):
2497
2238
"""Iterate backwards through revision ids in the lefthand history
3730
3408
return self.source.revision_ids_to_search_result(result_set)
3411
class InterPackRepo(InterSameDataRepository):
3412
"""Optimised code paths between Pack based repositories."""
3415
def _get_repo_format_to_test(self):
3416
from bzrlib.repofmt import pack_repo
3417
return pack_repo.RepositoryFormatKnitPack6RichRoot()
3420
def is_compatible(source, target):
3421
"""Be compatible with known Pack formats.
3423
We don't test for the stores being of specific types because that
3424
could lead to confusing results, and there is no need to be
3427
InterPackRepo does not support CHK based repositories.
3429
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3430
from bzrlib.repofmt.groupcompress_repo import RepositoryFormatCHK1
3432
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3433
isinstance(target._format, RepositoryFormatPack))
3434
not_packs = (isinstance(source._format, RepositoryFormatCHK1) or
3435
isinstance(target._format, RepositoryFormatCHK1))
3436
except AttributeError:
3438
if not_packs or not are_packs:
3440
return InterRepository._same_model(source, target)
3443
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3445
"""See InterRepository.fetch()."""
3446
if (len(self.source._fallback_repositories) > 0 or
3447
len(self.target._fallback_repositories) > 0):
3448
# The pack layer is not aware of fallback repositories, so when
3449
# fetching from a stacked repository or into a stacked repository
3450
# we use the generic fetch logic which uses the VersionedFiles
3451
# attributes on repository.
3452
from bzrlib.fetch import RepoFetcher
3453
fetcher = RepoFetcher(self.target, self.source, revision_id,
3454
pb, find_ghosts, fetch_spec=fetch_spec)
3455
if fetch_spec is not None:
3456
if len(list(fetch_spec.heads)) != 1:
3457
raise AssertionError(
3458
"InterPackRepo.fetch doesn't support "
3459
"fetching multiple heads yet.")
3460
revision_id = list(fetch_spec.heads)[0]
3462
if revision_id is None:
3464
# everything to do - use pack logic
3465
# to fetch from all packs to one without
3466
# inventory parsing etc, IFF nothing to be copied is in the target.
3468
source_revision_ids = frozenset(self.source.all_revision_ids())
3469
revision_ids = source_revision_ids - \
3470
frozenset(self.target.get_parent_map(source_revision_ids))
3471
revision_keys = [(revid,) for revid in revision_ids]
3472
index = self.target._pack_collection.revision_index.combined_index
3473
present_revision_ids = set(item[1][0] for item in
3474
index.iter_entries(revision_keys))
3475
revision_ids = set(revision_ids) - present_revision_ids
3476
# implementing the TODO will involve:
3477
# - detecting when all of a pack is selected
3478
# - avoiding as much as possible pre-selection, so the
3479
# more-core routines such as create_pack_from_packs can filter in
3480
# a just-in-time fashion. (though having a HEADS list on a
3481
# repository might make this a lot easier, because we could
3482
# sensibly detect 'new revisions' without doing a full index scan.
3483
elif _mod_revision.is_null(revision_id):
3487
revision_ids = self.search_missing_revision_ids(revision_id,
3488
find_ghosts=find_ghosts).get_keys()
3489
if len(revision_ids) == 0:
3491
return self._pack(self.source, self.target, revision_ids)
3493
def _pack(self, source, target, revision_ids):
3494
from bzrlib.repofmt.pack_repo import Packer
3495
packs = source._pack_collection.all_packs()
3496
pack = Packer(self.target._pack_collection, packs, '.fetch',
3497
revision_ids).pack()
3498
if pack is not None:
3499
self.target._pack_collection._save_pack_names()
3500
copied_revs = pack.get_revision_count()
3501
# Trigger an autopack. This may duplicate effort as we've just done
3502
# a pack creation, but for now it is simpler to think about as
3503
# 'upload data, then repack if needed'.
3504
self.target._pack_collection.autopack()
3505
return (copied_revs, [])
3510
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3511
"""See InterRepository.missing_revision_ids().
3513
:param find_ghosts: Find ghosts throughout the ancestry of
3516
if not find_ghosts and revision_id is not None:
3517
return self._walk_to_common_revisions([revision_id])
3518
elif revision_id is not None:
3519
# Find ghosts: search for revisions pointing from one repository to
3520
# the other, and vice versa, anywhere in the history of revision_id.
3521
graph = self.target.get_graph(other_repository=self.source)
3522
searcher = graph._make_breadth_first_searcher([revision_id])
3526
next_revs, ghosts = searcher.next_with_ghosts()
3527
except StopIteration:
3529
if revision_id in ghosts:
3530
raise errors.NoSuchRevision(self.source, revision_id)
3531
found_ids.update(next_revs)
3532
found_ids.update(ghosts)
3533
found_ids = frozenset(found_ids)
3534
# Double query here: should be able to avoid this by changing the
3535
# graph api further.
3536
result_set = found_ids - frozenset(
3537
self.target.get_parent_map(found_ids))
3539
source_ids = self.source.all_revision_ids()
3540
# source_ids is the worst possible case we may need to pull.
3541
# now we want to filter source_ids against what we actually
3542
# have in target, but don't try to check for existence where we know
3543
# we do not have a revision as that would be pointless.
3544
target_ids = set(self.target.all_revision_ids())
3545
result_set = set(source_ids).difference(target_ids)
3546
return self.source.revision_ids_to_search_result(result_set)
3733
3549
class InterDifferingSerializer(InterRepository):
3742
3558
# This is redundant with format.check_conversion_target(), however that
3743
3559
# raises an exception, and we just want to say "False" as in we won't
3744
3560
# support converting between these formats.
3745
if 'IDS_never' in debug.debug_flags:
3747
3561
if source.supports_rich_root() and not target.supports_rich_root():
3749
3563
if (source._format.supports_tree_reference
3750
3564
and not target._format.supports_tree_reference):
3752
if target._fallback_repositories and target._format.supports_chks:
3753
# IDS doesn't know how to copy CHKs for the parent inventories it
3754
# adds to stacked repos.
3756
if 'IDS_always' in debug.debug_flags:
3758
# Only use this code path for local source and target. IDS does far
3759
# too much IO (both bandwidth and roundtrips) over a network.
3760
if not source.bzrdir.transport.base.startswith('file:///'):
3762
if not target.bzrdir.transport.base.startswith('file:///'):
3766
def _get_trees(self, revision_ids, cache):
3768
for rev_id in revision_ids:
3770
possible_trees.append((rev_id, cache[rev_id]))
3772
# Not cached, but inventory might be present anyway.
3774
tree = self.source.revision_tree(rev_id)
3775
except errors.NoSuchRevision:
3776
# Nope, parent is ghost.
3779
cache[rev_id] = tree
3780
possible_trees.append((rev_id, tree))
3781
return possible_trees
3783
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3568
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3784
3569
"""Get the best delta and base for this revision.
3786
3571
:return: (basis_id, delta)
3573
possible_trees = [(parent_id, cache[parent_id])
3574
for parent_id in parent_ids
3575
if parent_id in cache]
3576
if len(possible_trees) == 0:
3577
# There either aren't any parents, or the parents aren't in the
3578
# cache, so just use the last converted tree
3579
possible_trees.append((basis_id, cache[basis_id]))
3789
# Generate deltas against each tree, to find the shortest.
3790
texts_possibly_new_in_tree = set()
3791
3581
for basis_id, basis_tree in possible_trees:
3792
3582
delta = tree.inventory._make_delta(basis_tree.inventory)
3793
for old_path, new_path, file_id, new_entry in delta:
3794
if new_path is None:
3795
# This file_id isn't present in the new rev, so we don't
3799
# Rich roots are handled elsewhere...
3801
kind = new_entry.kind
3802
if kind != 'directory' and kind != 'file':
3803
# No text record associated with this inventory entry.
3805
# This is a directory or file that has changed somehow.
3806
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3807
3583
deltas.append((len(delta), basis_id, delta))
3809
3585
return deltas[0][1:]
3811
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3812
"""Find all parent revisions that are absent, but for which the
3813
inventory is present, and copy those inventories.
3815
This is necessary to preserve correctness when the source is stacked
3816
without fallbacks configured. (Note that in cases like upgrade the
3817
source may be not have _fallback_repositories even though it is
3821
for parents in parent_map.values():
3822
parent_revs.update(parents)
3823
present_parents = self.source.get_parent_map(parent_revs)
3824
absent_parents = set(parent_revs).difference(present_parents)
3825
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3826
(rev_id,) for rev_id in absent_parents)
3827
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3828
for parent_tree in self.source.revision_trees(parent_inv_ids):
3829
current_revision_id = parent_tree.get_revision_id()
3830
parents_parents_keys = parent_invs_keys_for_stacking[
3831
(current_revision_id,)]
3832
parents_parents = [key[-1] for key in parents_parents_keys]
3833
basis_id = _mod_revision.NULL_REVISION
3834
basis_tree = self.source.revision_tree(basis_id)
3835
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3836
self.target.add_inventory_by_delta(
3837
basis_id, delta, current_revision_id, parents_parents)
3838
cache[current_revision_id] = parent_tree
3840
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3587
def _get_parent_keys(self, root_key, parent_map):
3588
"""Get the parent keys for a given root id."""
3589
root_id, rev_id = root_key
3590
# Include direct parents of the revision, but only if they used
3591
# the same root_id and are heads.
3593
for parent_id in parent_map[rev_id]:
3594
if parent_id == _mod_revision.NULL_REVISION:
3596
if parent_id not in self._revision_id_to_root_id:
3597
# We probably didn't read this revision, go spend the
3598
# extra effort to actually check
3600
tree = self.source.revision_tree(parent_id)
3601
except errors.NoSuchRevision:
3602
# Ghost, fill out _revision_id_to_root_id in case we
3603
# encounter this again.
3604
# But set parent_root_id to None since we don't really know
3605
parent_root_id = None
3607
parent_root_id = tree.get_root_id()
3608
self._revision_id_to_root_id[parent_id] = None
3610
parent_root_id = self._revision_id_to_root_id[parent_id]
3611
if root_id == parent_root_id:
3612
# With stacking we _might_ want to refer to a non-local
3613
# revision, but this code path only applies when we have the
3614
# full content available, so ghosts really are ghosts, not just
3615
# the edge of local data.
3616
parent_keys.append((parent_id,))
3618
# root_id may be in the parent anyway.
3620
tree = self.source.revision_tree(parent_id)
3621
except errors.NoSuchRevision:
3622
# ghost, can't refer to it.
3626
parent_keys.append((tree.inventory[root_id].revision,))
3627
except errors.NoSuchId:
3630
g = graph.Graph(self.source.revisions)
3631
heads = g.heads(parent_keys)
3633
for key in parent_keys:
3634
if key in heads and key not in selected_keys:
3635
selected_keys.append(key)
3636
return tuple([(root_id,)+ key for key in selected_keys])
3638
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3639
for root_key in root_keys_to_create:
3640
parent_keys = self._get_parent_keys(root_key, parent_map)
3641
yield versionedfile.FulltextContentFactory(root_key,
3642
parent_keys, None, '')
3644
def _fetch_batch(self, revision_ids, basis_id, cache):
3841
3645
"""Fetch across a few revisions.
3843
3647
:param revision_ids: The revisions to copy
3844
3648
:param basis_id: The revision_id of a tree that must be in cache, used
3845
3649
as a basis for delta when no other base is available
3846
3650
:param cache: A cache of RevisionTrees that we can use.
3847
:param a_graph: A Graph object to determine the heads() of the
3848
rich-root data stream.
3849
3651
:return: The revision_id of the last converted tree. The RevisionTree
3850
3652
for it will be in cache
3857
3659
pending_deltas = []
3858
3660
pending_revisions = []
3859
3661
parent_map = self.source.get_parent_map(revision_ids)
3860
self._fetch_parent_invs_for_stacking(parent_map, cache)
3861
self.source._safe_to_return_from_cache = True
3862
3662
for tree in self.source.revision_trees(revision_ids):
3863
# Find a inventory delta for this revision.
3864
# Find text entries that need to be copied, too.
3865
3663
current_revision_id = tree.get_revision_id()
3866
3664
parent_ids = parent_map.get(current_revision_id, ())
3867
parent_trees = self._get_trees(parent_ids, cache)
3868
possible_trees = list(parent_trees)
3869
if len(possible_trees) == 0:
3870
# There either aren't any parents, or the parents are ghosts,
3871
# so just use the last converted tree.
3872
possible_trees.append((basis_id, cache[basis_id]))
3873
3665
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3875
revision = self.source.get_revision(current_revision_id)
3876
pending_deltas.append((basis_id, delta,
3877
current_revision_id, revision.parent_ids))
3878
3667
if self._converting_to_rich_root:
3879
3668
self._revision_id_to_root_id[current_revision_id] = \
3880
3669
tree.get_root_id()
3881
# Determine which texts are in present in this revision but not in
3882
# any of the available parents.
3883
texts_possibly_new_in_tree = set()
3670
# Find text entries that need to be copied
3884
3671
for old_path, new_path, file_id, entry in delta:
3885
if new_path is None:
3886
# This file_id isn't present in the new rev
3890
if not self.target.supports_rich_root():
3891
# The target doesn't support rich root, so we don't
3894
if self._converting_to_rich_root:
3895
# This can't be copied normally, we have to insert
3897
root_keys_to_create.add((file_id, entry.revision))
3900
texts_possibly_new_in_tree.add((file_id, entry.revision))
3901
for basis_id, basis_tree in possible_trees:
3902
basis_inv = basis_tree.inventory
3903
for file_key in list(texts_possibly_new_in_tree):
3904
file_id, file_revision = file_key
3906
entry = basis_inv[file_id]
3907
except errors.NoSuchId:
3909
if entry.revision == file_revision:
3910
texts_possibly_new_in_tree.remove(file_key)
3911
text_keys.update(texts_possibly_new_in_tree)
3672
if new_path is not None:
3675
if not self.target.supports_rich_root():
3676
# The target doesn't support rich root, so we don't
3679
if self._converting_to_rich_root:
3680
# This can't be copied normally, we have to insert
3682
root_keys_to_create.add((file_id, entry.revision))
3684
text_keys.add((file_id, entry.revision))
3685
revision = self.source.get_revision(current_revision_id)
3686
pending_deltas.append((basis_id, delta,
3687
current_revision_id, revision.parent_ids))
3912
3688
pending_revisions.append(revision)
3913
3689
cache[current_revision_id] = tree
3914
3690
basis_id = current_revision_id
3915
self.source._safe_to_return_from_cache = False
3916
3691
# Copy file texts
3917
3692
from_texts = self.source.texts
3918
3693
to_texts = self.target.texts
3919
3694
if root_keys_to_create:
3920
root_stream = _mod_fetch._new_root_data_stream(
3921
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3922
self.source, graph=a_graph)
3695
root_stream = self._new_root_data_stream(root_keys_to_create,
3923
3697
to_texts.insert_record_stream(root_stream)
3924
3698
to_texts.insert_record_stream(from_texts.get_record_stream(
3925
3699
text_keys, self.target._format._fetch_order,
4374
4105
# missing keys can handle suspending a write group).
4375
4106
write_group_tokens = self.target_repo.suspend_write_group()
4376
4107
return write_group_tokens, missing_keys
4377
hint = self.target_repo.commit_write_group()
4378
if (to_serializer != src_serializer and
4379
self.target_repo._format.pack_compresses):
4380
self.target_repo.pack(hint=hint)
4108
self.target_repo.commit_write_group()
4381
4109
return [], set()
4383
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4384
target_rich_root = self.target_repo._format.rich_root_data
4385
target_tree_refs = self.target_repo._format.supports_tree_reference
4386
for record in substream:
4387
# Insert the delta directly
4388
inventory_delta_bytes = record.get_bytes_as('fulltext')
4389
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4391
parse_result = deserialiser.parse_text_bytes(
4392
inventory_delta_bytes)
4393
except inventory_delta.IncompatibleInventoryDelta, err:
4394
trace.mutter("Incompatible delta: %s", err.msg)
4395
raise errors.IncompatibleRevision(self.target_repo._format)
4396
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4397
revision_id = new_id
4398
parents = [key[0] for key in record.parents]
4399
self.target_repo.add_inventory_by_delta(
4400
basis_id, inv_delta, revision_id, parents)
4402
def _extract_and_insert_inventories(self, substream, serializer,
4111
def _extract_and_insert_inventories(self, substream, serializer):
4404
4112
"""Generate a new inventory versionedfile in target, converting data.
4406
4114
The inventory is retrieved from the source, (deserializing it), and
4407
4115
stored in the target (reserializing it in a different format).
4409
target_rich_root = self.target_repo._format.rich_root_data
4410
target_tree_refs = self.target_repo._format.supports_tree_reference
4411
4117
for record in substream:
4412
# It's not a delta, so it must be a fulltext in the source
4413
# serializer's format.
4414
4118
bytes = record.get_bytes_as('fulltext')
4415
4119
revision_id = record.key[0]
4416
4120
inv = serializer.read_inventory_from_string(bytes, revision_id)
4417
4121
parents = [key[0] for key in record.parents]
4418
4122
self.target_repo.add_inventory(revision_id, inv, parents)
4419
# No need to keep holding this full inv in memory when the rest of
4420
# the substream is likely to be all deltas.
4423
4124
def _extract_and_insert_revisions(self, substream, serializer):
4424
4125
for record in substream:
4587
4279
return (not self.from_repository._format.rich_root_data and
4588
4280
self.to_format.rich_root_data)
4590
def _get_inventory_stream(self, revision_ids, missing=False):
4282
def _get_inventory_stream(self, revision_ids):
4591
4283
from_format = self.from_repository._format
4592
if (from_format.supports_chks and self.to_format.supports_chks and
4593
from_format.network_name() == self.to_format.network_name()):
4594
raise AssertionError(
4595
"this case should be handled by GroupCHKStreamSource")
4596
elif 'forceinvdeltas' in debug.debug_flags:
4597
return self._get_convertable_inventory_stream(revision_ids,
4598
delta_versus_null=missing)
4599
elif from_format.network_name() == self.to_format.network_name():
4601
return self._get_simple_inventory_stream(revision_ids,
4603
elif (not from_format.supports_chks and not self.to_format.supports_chks
4604
and from_format._serializer == self.to_format._serializer):
4605
# Essentially the same format.
4606
return self._get_simple_inventory_stream(revision_ids,
4284
if (from_format.supports_chks and self.to_format.supports_chks
4285
and (from_format._serializer == self.to_format._serializer)):
4286
# Both sides support chks, and they use the same serializer, so it
4287
# is safe to transmit the chk pages and inventory pages across
4289
return self._get_chk_inventory_stream(revision_ids)
4290
elif (not from_format.supports_chks):
4291
# Source repository doesn't support chks. So we can transmit the
4292
# inventories 'as-is' and either they are just accepted on the
4293
# target, or the Sink will properly convert it.
4294
return self._get_simple_inventory_stream(revision_ids)
4609
# Any time we switch serializations, we want to use an
4610
# inventory-delta based approach.
4611
return self._get_convertable_inventory_stream(revision_ids,
4612
delta_versus_null=missing)
4296
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4297
# inventories. Note that this should probably be done somehow
4298
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4299
# figure out how a non-chk repository could possibly handle
4300
# deserializing an inventory stream from a chk repo, as it
4301
# doesn't have a way to understand individual pages.
4302
return self._get_convertable_inventory_stream(revision_ids)
4614
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4615
# NB: This currently reopens the inventory weave in source;
4616
# using a single stream interface instead would avoid this.
4304
def _get_simple_inventory_stream(self, revision_ids):
4617
4305
from_weave = self.from_repository.inventories
4619
delta_closure = True
4621
delta_closure = not self.delta_on_metadata()
4622
4306
yield ('inventories', from_weave.get_record_stream(
4623
4307
[(rev_id,) for rev_id in revision_ids],
4624
self.inventory_fetch_order(), delta_closure))
4626
def _get_convertable_inventory_stream(self, revision_ids,
4627
delta_versus_null=False):
4628
# The two formats are sufficiently different that there is no fast
4629
# path, so we need to send just inventorydeltas, which any
4630
# sufficiently modern client can insert into any repository.
4631
# The StreamSink code expects to be able to
4632
# convert on the target, so we need to put bytes-on-the-wire that can
4633
# be converted. That means inventory deltas (if the remote is <1.19,
4634
# RemoteStreamSink will fallback to VFS to insert the deltas).
4635
yield ('inventory-deltas',
4636
self._stream_invs_as_deltas(revision_ids,
4637
delta_versus_null=delta_versus_null))
4639
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4640
"""Return a stream of inventory-deltas for the given rev ids.
4642
:param revision_ids: The list of inventories to transmit
4643
:param delta_versus_null: Don't try to find a minimal delta for this
4644
entry, instead compute the delta versus the NULL_REVISION. This
4645
effectively streams a complete inventory. Used for stuff like
4646
filling in missing parents, etc.
4308
self.inventory_fetch_order(),
4309
not self.delta_on_metadata()))
4311
def _get_chk_inventory_stream(self, revision_ids):
4312
"""Fetch the inventory texts, along with the associated chk maps."""
4313
# We want an inventory outside of the search set, so that we can filter
4314
# out uninteresting chk pages. For now we use
4315
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4316
# could use that instead.
4317
start_rev_id = self.from_repository._find_revision_outside_set(
4319
start_rev_key = (start_rev_id,)
4320
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4321
if start_rev_id != _mod_revision.NULL_REVISION:
4322
inv_keys_to_fetch.append((start_rev_id,))
4323
# Any repo that supports chk_bytes must also support out-of-order
4324
# insertion. At least, that is how we expect it to work
4325
# We use get_record_stream instead of iter_inventories because we want
4326
# to be able to insert the stream as well. We could instead fetch
4327
# allowing deltas, and then iter_inventories, but we don't know whether
4328
# source or target is more 'local' anway.
4329
inv_stream = self.from_repository.inventories.get_record_stream(
4330
inv_keys_to_fetch, 'unordered',
4331
True) # We need them as full-texts so we can find their references
4332
uninteresting_chk_roots = set()
4333
interesting_chk_roots = set()
4334
def filter_inv_stream(inv_stream):
4335
for idx, record in enumerate(inv_stream):
4336
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4337
bytes = record.get_bytes_as('fulltext')
4338
chk_inv = inventory.CHKInventory.deserialise(
4339
self.from_repository.chk_bytes, bytes, record.key)
4340
if record.key == start_rev_key:
4341
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4342
p_id_map = chk_inv.parent_id_basename_to_file_id
4343
if p_id_map is not None:
4344
uninteresting_chk_roots.add(p_id_map.key())
4347
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4348
p_id_map = chk_inv.parent_id_basename_to_file_id
4349
if p_id_map is not None:
4350
interesting_chk_roots.add(p_id_map.key())
4351
### pb.update('fetch inventory', 0, 2)
4352
yield ('inventories', filter_inv_stream(inv_stream))
4353
# Now that we have worked out all of the interesting root nodes, grab
4354
# all of the interesting pages and insert them
4355
### pb.update('fetch inventory', 1, 2)
4356
interesting = chk_map.iter_interesting_nodes(
4357
self.from_repository.chk_bytes, interesting_chk_roots,
4358
uninteresting_chk_roots)
4359
def to_stream_adapter():
4360
"""Adapt the iter_interesting_nodes result to a single stream.
4362
iter_interesting_nodes returns records as it processes them, along
4363
with keys. However, we only want to return the records themselves.
4365
for record, items in interesting:
4366
if record is not None:
4368
# XXX: We could instead call get_record_stream(records.keys())
4369
# ATM, this will always insert the records as fulltexts, and
4370
# requires that you can hang on to records once you have gone
4371
# on to the next one. Further, it causes the target to
4372
# recompress the data. Testing shows it to be faster than
4373
# requesting the records again, though.
4374
yield ('chk_bytes', to_stream_adapter())
4375
### pb.update('fetch inventory', 2, 2)
4377
def _get_convertable_inventory_stream(self, revision_ids):
4378
# XXX: One of source or target is using chks, and they don't have
4379
# compatible serializations. The StreamSink code expects to be
4380
# able to convert on the target, so we need to put
4381
# bytes-on-the-wire that can be converted
4382
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4384
def _stream_invs_as_fulltexts(self, revision_ids):
4648
4385
from_repo = self.from_repository
4386
from_serializer = from_repo._format._serializer
4649
4387
revision_keys = [(rev_id,) for rev_id in revision_ids]
4650
4388
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4651
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4653
inventories = self.from_repository.iter_inventories(
4654
revision_ids, 'topological')
4655
format = from_repo._format
4656
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4657
inventory_cache = lru_cache.LRUCache(50)
4658
null_inventory = from_repo.revision_tree(
4659
_mod_revision.NULL_REVISION).inventory
4660
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4661
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4662
# repo back into a non-rich-root repo ought to be allowed)
4663
serializer = inventory_delta.InventoryDeltaSerializer(
4664
versioned_root=format.rich_root_data,
4665
tree_references=format.supports_tree_reference)
4666
for inv in inventories:
4389
for inv in self.from_repository.iter_inventories(revision_ids):
4390
# XXX: This is a bit hackish, but it works. Basically,
4391
# CHKSerializer 'accidentally' supports
4392
# read/write_inventory_to_string, even though that is never
4393
# the format that is stored on disk. It *does* give us a
4394
# single string representation for an inventory, so live with
4396
# This would be far better if we had a 'serialized inventory
4397
# delta' form. Then we could use 'inventory._make_delta', and
4398
# transmit that. This would both be faster to generate, and
4399
# result in fewer bytes-on-the-wire.
4400
as_bytes = from_serializer.write_inventory_to_string(inv)
4667
4401
key = (inv.revision_id,)
4668
4402
parent_keys = parent_map.get(key, ())
4670
if not delta_versus_null and parent_keys:
4671
# The caller did not ask for complete inventories and we have
4672
# some parents that we can delta against. Make a delta against
4673
# each parent so that we can find the smallest.
4674
parent_ids = [parent_key[0] for parent_key in parent_keys]
4675
for parent_id in parent_ids:
4676
if parent_id not in invs_sent_so_far:
4677
# We don't know that the remote side has this basis, so
4680
if parent_id == _mod_revision.NULL_REVISION:
4681
parent_inv = null_inventory
4683
parent_inv = inventory_cache.get(parent_id, None)
4684
if parent_inv is None:
4685
parent_inv = from_repo.get_inventory(parent_id)
4686
candidate_delta = inv._make_delta(parent_inv)
4687
if (delta is None or
4688
len(delta) > len(candidate_delta)):
4689
delta = candidate_delta
4690
basis_id = parent_id
4692
# Either none of the parents ended up being suitable, or we
4693
# were asked to delta against NULL
4694
basis_id = _mod_revision.NULL_REVISION
4695
delta = inv._make_delta(null_inventory)
4696
invs_sent_so_far.add(inv.revision_id)
4697
inventory_cache[inv.revision_id] = inv
4698
delta_serialized = ''.join(
4699
serializer.delta_to_lines(basis_id, key[-1], delta))
4700
4403
yield versionedfile.FulltextContentFactory(
4701
key, parent_keys, None, delta_serialized)
4704
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4705
stop_revision=None):
4706
"""Extend the partial history to include a given index
4708
If a stop_index is supplied, stop when that index has been reached.
4709
If a stop_revision is supplied, stop when that revision is
4710
encountered. Otherwise, stop when the beginning of history is
4713
:param stop_index: The index which should be present. When it is
4714
present, history extension will stop.
4715
:param stop_revision: The revision id which should be present. When
4716
it is encountered, history extension will stop.
4718
start_revision = partial_history_cache[-1]
4719
iterator = repo.iter_reverse_revision_history(start_revision)
4721
#skip the last revision in the list
4724
if (stop_index is not None and
4725
len(partial_history_cache) > stop_index):
4727
if partial_history_cache[-1] == stop_revision:
4729
revision_id = iterator.next()
4730
partial_history_cache.append(revision_id)
4731
except StopIteration:
4404
key, parent_keys, None, as_bytes)
added
removed
bzrlib/repository.py
