1167
1123
# The old API returned a list, should this actually be a set?
1168
1124
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
1127
def create(a_bzrdir):
1285
1128
"""Construct the current default format repository in a_bzrdir."""
1570
1397
def suspend_write_group(self):
1571
1398
raise errors.UnsuspendableWriteGroup(self)
1573
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1574
"""Return the keys of missing inventory parents for revisions added in
1577
A revision is not complete if the inventory delta for that revision
1578
cannot be calculated. Therefore if the parent inventories of a
1579
revision are not present, the revision is incomplete, and e.g. cannot
1580
be streamed by a smart server. This method finds missing inventory
1581
parents for revisions added in this write group.
1583
if not self._format.supports_external_lookups:
1584
# This is only an issue for stacked repositories
1586
if not self.is_in_write_group():
1587
raise AssertionError('not in a write group')
1589
# XXX: We assume that every added revision already has its
1590
# corresponding inventory, so we only check for parent inventories that
1591
# might be missing, rather than all inventories.
1592
parents = set(self.revisions._index.get_missing_parents())
1593
parents.discard(_mod_revision.NULL_REVISION)
1594
unstacked_inventories = self.inventories._index
1595
present_inventories = unstacked_inventories.get_parent_map(
1596
key[-1:] for key in parents)
1597
parents.difference_update(present_inventories)
1598
if len(parents) == 0:
1599
# No missing parent inventories.
1601
if not check_for_missing_texts:
1602
return set(('inventories', rev_id) for (rev_id,) in parents)
1603
# Ok, now we have a list of missing inventories. But these only matter
1604
# if the inventories that reference them are missing some texts they
1605
# appear to introduce.
1606
# XXX: Texts referenced by all added inventories need to be present,
1607
# but at the moment we're only checking for texts referenced by
1608
# inventories at the graph's edge.
1609
key_deps = self.revisions._index._key_dependencies
1610
key_deps.satisfy_refs_for_keys(present_inventories)
1611
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1612
file_ids = self.fileids_altered_by_revision_ids(referrers)
1613
missing_texts = set()
1614
for file_id, version_ids in file_ids.iteritems():
1615
missing_texts.update(
1616
(file_id, version_id) for version_id in version_ids)
1617
present_texts = self.texts.get_parent_map(missing_texts)
1618
missing_texts.difference_update(present_texts)
1619
if not missing_texts:
1620
# No texts are missing, so all revisions and their deltas are
1623
# Alternatively the text versions could be returned as the missing
1624
# keys, but this is likely to be less data.
1625
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1628
1400
def refresh_data(self):
1629
1401
"""Re-read any data needed to to synchronise with disk.
1852
1615
@needs_read_lock
1853
1616
def get_revisions(self, revision_ids):
1854
"""Get many revisions at once.
1856
Repositories that need to check data on every revision read should
1857
subclass this method.
1617
"""Get many revisions at once."""
1859
1618
return self._get_revisions(revision_ids)
1861
1620
@needs_read_lock
1862
1621
def _get_revisions(self, revision_ids):
1863
1622
"""Core work logic to get many revisions without sanity checks."""
1623
for rev_id in revision_ids:
1624
if not rev_id or not isinstance(rev_id, basestring):
1625
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1626
keys = [(key,) for key in revision_ids]
1627
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1865
for revid, rev in self._iter_revisions(revision_ids):
1867
raise errors.NoSuchRevision(self, revid)
1629
for record in stream:
1630
if record.storage_kind == 'absent':
1631
raise errors.NoSuchRevision(self, record.key[0])
1632
text = record.get_bytes_as('fulltext')
1633
rev = self._serializer.read_revision_from_string(text)
1634
revs[record.key[0]] = rev
1869
1635
return [revs[revid] for revid in revision_ids]
1871
def _iter_revisions(self, revision_ids):
1872
"""Iterate over revision objects.
1874
:param revision_ids: An iterable of revisions to examine. None may be
1875
passed to request all revisions known to the repository. Note that
1876
not all repositories can find unreferenced revisions; for those
1877
repositories only referenced ones will be returned.
1878
:return: An iterator of (revid, revision) tuples. Absent revisions (
1879
those asked for but not available) are returned as (revid, None).
1881
if revision_ids is None:
1882
revision_ids = self.all_revision_ids()
1884
for rev_id in revision_ids:
1885
if not rev_id or not isinstance(rev_id, basestring):
1886
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1887
keys = [(key,) for key in revision_ids]
1888
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1889
for record in stream:
1890
revid = record.key[0]
1891
if record.storage_kind == 'absent':
1894
text = record.get_bytes_as('fulltext')
1895
rev = self._serializer.read_revision_from_string(text)
1638
def get_revision_xml(self, revision_id):
1639
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1640
# would have already do it.
1641
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1642
# TODO: this can't just be replaced by:
1643
# return self._serializer.write_revision_to_string(
1644
# self.get_revision(revision_id))
1645
# as cStringIO preservers the encoding unlike write_revision_to_string
1646
# or some other call down the path.
1647
rev = self.get_revision(revision_id)
1648
rev_tmp = cStringIO.StringIO()
1649
# the current serializer..
1650
self._serializer.write_revision(rev, rev_tmp)
1652
return rev_tmp.getvalue()
1898
1654
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1899
1655
"""Produce a generator of revision deltas.
2349
2079
:param revision_ids: The expected revision ids of the inventories.
2350
:param ordering: optional ordering, e.g. 'topological'. If not
2351
specified, the order of revision_ids will be preserved (by
2352
buffering if necessary).
2353
2080
:return: An iterator of inventories.
2355
2082
if ((None in revision_ids)
2356
2083
or (_mod_revision.NULL_REVISION in revision_ids)):
2357
2084
raise ValueError('cannot get null revision inventory')
2358
return self._iter_inventories(revision_ids, ordering)
2085
return self._iter_inventories(revision_ids)
2360
def _iter_inventories(self, revision_ids, ordering):
2087
def _iter_inventories(self, revision_ids):
2361
2088
"""single-document based inventory iteration."""
2362
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2363
for text, revision_id in inv_xmls:
2364
yield self._deserialise_inventory(revision_id, text)
2089
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2090
yield self.deserialise_inventory(revision_id, text)
2366
def _iter_inventory_xmls(self, revision_ids, ordering):
2367
if ordering is None:
2368
order_as_requested = True
2369
ordering = 'unordered'
2371
order_as_requested = False
2092
def _iter_inventory_xmls(self, revision_ids):
2372
2093
keys = [(revision_id,) for revision_id in revision_ids]
2375
if order_as_requested:
2376
key_iter = iter(keys)
2377
next_key = key_iter.next()
2378
stream = self.inventories.get_record_stream(keys, ordering, True)
2094
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2379
2095
text_chunks = {}
2380
2096
for record in stream:
2381
2097
if record.storage_kind != 'absent':
2382
chunks = record.get_bytes_as('chunked')
2383
if order_as_requested:
2384
text_chunks[record.key] = chunks
2386
yield ''.join(chunks), record.key[-1]
2098
text_chunks[record.key] = record.get_bytes_as('chunked')
2388
2100
raise errors.NoSuchRevision(self, record.key)
2389
if order_as_requested:
2390
# Yield as many results as we can while preserving order.
2391
while next_key in text_chunks:
2392
chunks = text_chunks.pop(next_key)
2393
yield ''.join(chunks), next_key[-1]
2395
next_key = key_iter.next()
2396
except StopIteration:
2397
# We still want to fully consume the get_record_stream,
2398
# just in case it is not actually finished at this point
2102
chunks = text_chunks.pop(key)
2103
yield ''.join(chunks), key[-1]
2402
def _deserialise_inventory(self, revision_id, xml):
2105
def deserialise_inventory(self, revision_id, xml):
2403
2106
"""Transform the xml into an inventory object.
2405
2108
:param revision_id: The expected revision id of the inventory.
2406
2109
:param xml: A serialised inventory.
2408
2111
result = self._serializer.read_inventory_from_string(xml, revision_id,
2409
entry_cache=self._inventory_entry_cache,
2410
return_from_cache=self._safe_to_return_from_cache)
2112
entry_cache=self._inventory_entry_cache)
2411
2113
if result.revision_id != revision_id:
2412
2114
raise AssertionError('revision id mismatch %s != %s' % (
2413
2115
result.revision_id, revision_id))
2118
def serialise_inventory(self, inv):
2119
return self._serializer.write_inventory_to_string(inv)
2121
def _serialise_inventory_to_lines(self, inv):
2122
return self._serializer.write_inventory_to_lines(inv)
2416
2124
def get_serializer_format(self):
2417
2125
return self._serializer.format_num
2419
2127
@needs_read_lock
2420
def _get_inventory_xml(self, revision_id):
2421
"""Get serialized inventory as a string."""
2422
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2128
def get_inventory_xml(self, revision_id):
2129
"""Get inventory XML as a file object."""
2130
texts = self._iter_inventory_xmls([revision_id])
2424
2132
text, revision_id = texts.next()
2425
2133
except StopIteration:
2426
2134
raise errors.HistoryMissing(self, 'inventory', revision_id)
2429
def get_rev_id_for_revno(self, revno, known_pair):
2430
"""Return the revision id of a revno, given a later (revno, revid)
2431
pair in the same history.
2433
:return: if found (True, revid). If the available history ran out
2434
before reaching the revno, then this returns
2435
(False, (closest_revno, closest_revid)).
2138
def get_inventory_sha1(self, revision_id):
2139
"""Return the sha1 hash of the inventory entry
2437
known_revno, known_revid = known_pair
2438
partial_history = [known_revid]
2439
distance_from_known = known_revno - revno
2440
if distance_from_known < 0:
2442
'requested revno (%d) is later than given known revno (%d)'
2443
% (revno, known_revno))
2446
self, partial_history, stop_index=distance_from_known)
2447
except errors.RevisionNotPresent, err:
2448
if err.revision_id == known_revid:
2449
# The start revision (known_revid) wasn't found.
2451
# This is a stacked repository with no fallbacks, or a there's a
2452
# left-hand ghost. Either way, even though the revision named in
2453
# the error isn't in this repo, we know it's the next step in this
2454
# left-hand history.
2455
partial_history.append(err.revision_id)
2456
if len(partial_history) <= distance_from_known:
2457
# Didn't find enough history to get a revid for the revno.
2458
earliest_revno = known_revno - len(partial_history) + 1
2459
return (False, (earliest_revno, partial_history[-1]))
2460
if len(partial_history) - 1 > distance_from_known:
2461
raise AssertionError('_iter_for_revno returned too much history')
2462
return (True, partial_history[-1])
2141
return self.get_revision(revision_id).inventory_sha1
2464
2143
def iter_reverse_revision_history(self, revision_id):
2465
2144
"""Iterate backwards through revision ids in the lefthand history
2705
2391
return record.get_bytes_as('fulltext')
2707
2393
@needs_read_lock
2708
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2394
def check(self, revision_ids=None):
2709
2395
"""Check consistency of all history of given revision_ids.
2711
2397
Different repository implementations should override _check().
2713
2399
:param revision_ids: A non-empty list of revision_ids whose ancestry
2714
2400
will be checked. Typically the last revision_id of a branch.
2715
:param callback_refs: A dict of check-refs to resolve and callback
2716
the check/_check method on the items listed as wanting the ref.
2718
:param check_repo: If False do not check the repository contents, just
2719
calculate the data callback_refs requires and call them back.
2721
return self._check(revision_ids, callback_refs=callback_refs,
2722
check_repo=check_repo)
2402
return self._check(revision_ids)
2724
def _check(self, revision_ids, callback_refs, check_repo):
2725
result = check.Check(self, check_repo=check_repo)
2726
result.check(callback_refs)
2404
def _check(self, revision_ids):
2405
result = check.Check(self)
2729
def _warn_if_deprecated(self, branch=None):
2409
def _warn_if_deprecated(self):
2730
2410
global _deprecation_warning_done
2731
2411
if _deprecation_warning_done:
2735
conf = config.GlobalConfig()
2737
conf = branch.get_config()
2738
if conf.suppress_warning('format_deprecation'):
2740
warning("Format %s for %s is deprecated -"
2741
" please use 'bzr upgrade' to get better performance"
2742
% (self._format, self.bzrdir.transport.base))
2744
_deprecation_warning_done = True
2413
_deprecation_warning_done = True
2414
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2415
% (self._format, self.bzrdir.transport.base))
2746
2417
def supports_rich_root(self):
2747
2418
return self._format.rich_root_data
3671
3298
return self.source.revision_ids_to_search_result(result_set)
3674
class InterDifferingSerializer(InterRepository):
3301
class InterPackRepo(InterSameDataRepository):
3302
"""Optimised code paths between Pack based repositories."""
3305
def _get_repo_format_to_test(self):
3306
from bzrlib.repofmt import pack_repo
3307
return pack_repo.RepositoryFormatKnitPack1()
3310
def is_compatible(source, target):
3311
"""Be compatible with known Pack formats.
3313
We don't test for the stores being of specific types because that
3314
could lead to confusing results, and there is no need to be
3317
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3319
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3320
isinstance(target._format, RepositoryFormatPack))
3321
except AttributeError:
3323
return are_packs and InterRepository._same_model(source, target)
3326
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3328
"""See InterRepository.fetch()."""
3329
if (len(self.source._fallback_repositories) > 0 or
3330
len(self.target._fallback_repositories) > 0):
3331
# The pack layer is not aware of fallback repositories, so when
3332
# fetching from a stacked repository or into a stacked repository
3333
# we use the generic fetch logic which uses the VersionedFiles
3334
# attributes on repository.
3335
from bzrlib.fetch import RepoFetcher
3336
fetcher = RepoFetcher(self.target, self.source, revision_id,
3337
pb, find_ghosts, fetch_spec=fetch_spec)
3338
if fetch_spec is not None:
3339
if len(list(fetch_spec.heads)) != 1:
3340
raise AssertionError(
3341
"InterPackRepo.fetch doesn't support "
3342
"fetching multiple heads yet.")
3343
revision_id = list(fetch_spec.heads)[0]
3345
if revision_id is None:
3347
# everything to do - use pack logic
3348
# to fetch from all packs to one without
3349
# inventory parsing etc, IFF nothing to be copied is in the target.
3351
source_revision_ids = frozenset(self.source.all_revision_ids())
3352
revision_ids = source_revision_ids - \
3353
frozenset(self.target.get_parent_map(source_revision_ids))
3354
revision_keys = [(revid,) for revid in revision_ids]
3355
index = self.target._pack_collection.revision_index.combined_index
3356
present_revision_ids = set(item[1][0] for item in
3357
index.iter_entries(revision_keys))
3358
revision_ids = set(revision_ids) - present_revision_ids
3359
# implementing the TODO will involve:
3360
# - detecting when all of a pack is selected
3361
# - avoiding as much as possible pre-selection, so the
3362
# more-core routines such as create_pack_from_packs can filter in
3363
# a just-in-time fashion. (though having a HEADS list on a
3364
# repository might make this a lot easier, because we could
3365
# sensibly detect 'new revisions' without doing a full index scan.
3366
elif _mod_revision.is_null(revision_id):
3371
revision_ids = self.search_missing_revision_ids(revision_id,
3372
find_ghosts=find_ghosts).get_keys()
3373
except errors.NoSuchRevision:
3374
raise errors.InstallFailed([revision_id])
3375
if len(revision_ids) == 0:
3377
return self._pack(self.source, self.target, revision_ids)
3379
def _pack(self, source, target, revision_ids):
3380
from bzrlib.repofmt.pack_repo import Packer
3381
packs = source._pack_collection.all_packs()
3382
pack = Packer(self.target._pack_collection, packs, '.fetch',
3383
revision_ids).pack()
3384
if pack is not None:
3385
self.target._pack_collection._save_pack_names()
3386
copied_revs = pack.get_revision_count()
3387
# Trigger an autopack. This may duplicate effort as we've just done
3388
# a pack creation, but for now it is simpler to think about as
3389
# 'upload data, then repack if needed'.
3390
self.target._pack_collection.autopack()
3391
return (copied_revs, [])
3396
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3397
"""See InterRepository.missing_revision_ids().
3399
:param find_ghosts: Find ghosts throughout the ancestry of
3402
if not find_ghosts and revision_id is not None:
3403
return self._walk_to_common_revisions([revision_id])
3404
elif revision_id is not None:
3405
# Find ghosts: search for revisions pointing from one repository to
3406
# the other, and vice versa, anywhere in the history of revision_id.
3407
graph = self.target.get_graph(other_repository=self.source)
3408
searcher = graph._make_breadth_first_searcher([revision_id])
3412
next_revs, ghosts = searcher.next_with_ghosts()
3413
except StopIteration:
3415
if revision_id in ghosts:
3416
raise errors.NoSuchRevision(self.source, revision_id)
3417
found_ids.update(next_revs)
3418
found_ids.update(ghosts)
3419
found_ids = frozenset(found_ids)
3420
# Double query here: should be able to avoid this by changing the
3421
# graph api further.
3422
result_set = found_ids - frozenset(
3423
self.target.get_parent_map(found_ids))
3425
source_ids = self.source.all_revision_ids()
3426
# source_ids is the worst possible case we may need to pull.
3427
# now we want to filter source_ids against what we actually
3428
# have in target, but don't try to check for existence where we know
3429
# we do not have a revision as that would be pointless.
3430
target_ids = set(self.target.all_revision_ids())
3431
result_set = set(source_ids).difference(target_ids)
3432
return self.source.revision_ids_to_search_result(result_set)
3435
class InterDifferingSerializer(InterKnitRepo):
3677
3438
def _get_repo_format_to_test(self):
3681
3442
def is_compatible(source, target):
3682
3443
"""Be compatible with Knit2 source and Knit3 target"""
3683
# This is redundant with format.check_conversion_target(), however that
3684
# raises an exception, and we just want to say "False" as in we won't
3685
# support converting between these formats.
3686
if 'IDS_never' in debug.debug_flags:
3688
if source.supports_rich_root() and not target.supports_rich_root():
3690
if (source._format.supports_tree_reference
3691
and not target._format.supports_tree_reference):
3693
if target._fallback_repositories and target._format.supports_chks:
3694
# IDS doesn't know how to copy CHKs for the parent inventories it
3695
# adds to stacked repos.
3697
if 'IDS_always' in debug.debug_flags:
3699
# Only use this code path for local source and target. IDS does far
3700
# too much IO (both bandwidth and roundtrips) over a network.
3701
if not source.bzrdir.transport.base.startswith('file:///'):
3703
if not target.bzrdir.transport.base.startswith('file:///'):
3444
if source.supports_rich_root() != target.supports_rich_root():
3446
# Ideally, we'd support fetching if the source had no tree references
3447
# even if it supported them...
3448
if (getattr(source, '_format.supports_tree_reference', False) and
3449
not getattr(target, '_format.supports_tree_reference', False)):
3707
def _get_trees(self, revision_ids, cache):
3709
for rev_id in revision_ids:
3711
possible_trees.append((rev_id, cache[rev_id]))
3713
# Not cached, but inventory might be present anyway.
3715
tree = self.source.revision_tree(rev_id)
3716
except errors.NoSuchRevision:
3717
# Nope, parent is ghost.
3720
cache[rev_id] = tree
3721
possible_trees.append((rev_id, tree))
3722
return possible_trees
3724
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3453
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3725
3454
"""Get the best delta and base for this revision.
3727
3456
:return: (basis_id, delta)
3458
possible_trees = [(parent_id, cache[parent_id])
3459
for parent_id in parent_ids
3460
if parent_id in cache]
3461
if len(possible_trees) == 0:
3462
# There either aren't any parents, or the parents aren't in the
3463
# cache, so just use the last converted tree
3464
possible_trees.append((basis_id, cache[basis_id]))
3730
# Generate deltas against each tree, to find the shortest.
3731
texts_possibly_new_in_tree = set()
3732
3466
for basis_id, basis_tree in possible_trees:
3733
3467
delta = tree.inventory._make_delta(basis_tree.inventory)
3734
for old_path, new_path, file_id, new_entry in delta:
3735
if new_path is None:
3736
# This file_id isn't present in the new rev, so we don't
3740
# Rich roots are handled elsewhere...
3742
kind = new_entry.kind
3743
if kind != 'directory' and kind != 'file':
3744
# No text record associated with this inventory entry.
3746
# This is a directory or file that has changed somehow.
3747
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3748
3468
deltas.append((len(delta), basis_id, delta))
3750
3470
return deltas[0][1:]
3752
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3753
"""Find all parent revisions that are absent, but for which the
3754
inventory is present, and copy those inventories.
3756
This is necessary to preserve correctness when the source is stacked
3757
without fallbacks configured. (Note that in cases like upgrade the
3758
source may be not have _fallback_repositories even though it is
3762
for parents in parent_map.values():
3763
parent_revs.update(parents)
3764
present_parents = self.source.get_parent_map(parent_revs)
3765
absent_parents = set(parent_revs).difference(present_parents)
3766
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3767
(rev_id,) for rev_id in absent_parents)
3768
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3769
for parent_tree in self.source.revision_trees(parent_inv_ids):
3770
current_revision_id = parent_tree.get_revision_id()
3771
parents_parents_keys = parent_invs_keys_for_stacking[
3772
(current_revision_id,)]
3773
parents_parents = [key[-1] for key in parents_parents_keys]
3774
basis_id = _mod_revision.NULL_REVISION
3775
basis_tree = self.source.revision_tree(basis_id)
3776
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3777
self.target.add_inventory_by_delta(
3778
basis_id, delta, current_revision_id, parents_parents)
3779
cache[current_revision_id] = parent_tree
3781
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3472
def _fetch_batch(self, revision_ids, basis_id, cache):
3782
3473
"""Fetch across a few revisions.
3784
3475
:param revision_ids: The revisions to copy
3785
3476
:param basis_id: The revision_id of a tree that must be in cache, used
3786
3477
as a basis for delta when no other base is available
3787
3478
:param cache: A cache of RevisionTrees that we can use.
3788
:param a_graph: A Graph object to determine the heads() of the
3789
rich-root data stream.
3790
3479
:return: The revision_id of the last converted tree. The RevisionTree
3791
3480
for it will be in cache
3793
3482
# Walk though all revisions; get inventory deltas, copy referenced
3794
3483
# texts that delta references, insert the delta, revision and
3796
root_keys_to_create = set()
3797
3485
text_keys = set()
3798
3486
pending_deltas = []
3799
3487
pending_revisions = []
3800
3488
parent_map = self.source.get_parent_map(revision_ids)
3801
self._fetch_parent_invs_for_stacking(parent_map, cache)
3802
self.source._safe_to_return_from_cache = True
3803
3489
for tree in self.source.revision_trees(revision_ids):
3804
# Find a inventory delta for this revision.
3805
# Find text entries that need to be copied, too.
3806
3490
current_revision_id = tree.get_revision_id()
3807
3491
parent_ids = parent_map.get(current_revision_id, ())
3808
parent_trees = self._get_trees(parent_ids, cache)
3809
possible_trees = list(parent_trees)
3810
if len(possible_trees) == 0:
3811
# There either aren't any parents, or the parents are ghosts,
3812
# so just use the last converted tree.
3813
possible_trees.append((basis_id, cache[basis_id]))
3814
3492
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3494
# Find text entries that need to be copied
3495
for old_path, new_path, file_id, entry in delta:
3496
if new_path is not None:
3497
if not (new_path or self.target.supports_rich_root()):
3498
# We don't copy the text for the root node unless the
3499
# target supports_rich_root.
3501
text_keys.add((file_id, entry.revision))
3816
3502
revision = self.source.get_revision(current_revision_id)
3817
3503
pending_deltas.append((basis_id, delta,
3818
3504
current_revision_id, revision.parent_ids))
3819
if self._converting_to_rich_root:
3820
self._revision_id_to_root_id[current_revision_id] = \
3822
# Determine which texts are in present in this revision but not in
3823
# any of the available parents.
3824
texts_possibly_new_in_tree = set()
3825
for old_path, new_path, file_id, entry in delta:
3826
if new_path is None:
3827
# This file_id isn't present in the new rev
3831
if not self.target.supports_rich_root():
3832
# The target doesn't support rich root, so we don't
3835
if self._converting_to_rich_root:
3836
# This can't be copied normally, we have to insert
3838
root_keys_to_create.add((file_id, entry.revision))
3841
texts_possibly_new_in_tree.add((file_id, entry.revision))
3842
for basis_id, basis_tree in possible_trees:
3843
basis_inv = basis_tree.inventory
3844
for file_key in list(texts_possibly_new_in_tree):
3845
file_id, file_revision = file_key
3847
entry = basis_inv[file_id]
3848
except errors.NoSuchId:
3850
if entry.revision == file_revision:
3851
texts_possibly_new_in_tree.remove(file_key)
3852
text_keys.update(texts_possibly_new_in_tree)
3853
3505
pending_revisions.append(revision)
3854
3506
cache[current_revision_id] = tree
3855
3507
basis_id = current_revision_id
3856
self.source._safe_to_return_from_cache = False
3857
3508
# Copy file texts
3858
3509
from_texts = self.source.texts
3859
3510
to_texts = self.target.texts
3860
if root_keys_to_create:
3861
root_stream = _mod_fetch._new_root_data_stream(
3862
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3863
self.source, graph=a_graph)
3864
to_texts.insert_record_stream(root_stream)
3865
3511
to_texts.insert_record_stream(from_texts.get_record_stream(
3866
3512
text_keys, self.target._format._fetch_order,
3867
3513
not self.target._format._fetch_uses_deltas))
3868
# insert inventory deltas
3869
3515
for delta in pending_deltas:
3870
3516
self.target.add_inventory_by_delta(*delta)
3871
if self.target._fallback_repositories:
3872
# Make sure this stacked repository has all the parent inventories
3873
# for the new revisions that we are about to insert. We do this
3874
# before adding the revisions so that no revision is added until
3875
# all the inventories it may depend on are added.
3876
# Note that this is overzealous, as we may have fetched these in an
3879
revision_ids = set()
3880
for revision in pending_revisions:
3881
revision_ids.add(revision.revision_id)
3882
parent_ids.update(revision.parent_ids)
3883
parent_ids.difference_update(revision_ids)
3884
parent_ids.discard(_mod_revision.NULL_REVISION)
3885
parent_map = self.source.get_parent_map(parent_ids)
3886
# we iterate over parent_map and not parent_ids because we don't
3887
# want to try copying any revision which is a ghost
3888
for parent_tree in self.source.revision_trees(parent_map):
3889
current_revision_id = parent_tree.get_revision_id()
3890
parents_parents = parent_map[current_revision_id]
3891
possible_trees = self._get_trees(parents_parents, cache)
3892
if len(possible_trees) == 0:
3893
# There either aren't any parents, or the parents are
3894
# ghosts, so just use the last converted tree.
3895
possible_trees.append((basis_id, cache[basis_id]))
3896
basis_id, delta = self._get_delta_for_revision(parent_tree,
3897
parents_parents, possible_trees)
3898
self.target.add_inventory_by_delta(
3899
basis_id, delta, current_revision_id, parents_parents)
3900
3517
# insert signatures and revisions
3901
3518
for revision in pending_revisions:
4318
3870
# missing keys can handle suspending a write group).
4319
3871
write_group_tokens = self.target_repo.suspend_write_group()
4320
3872
return write_group_tokens, missing_keys
4321
hint = self.target_repo.commit_write_group()
4322
if (to_serializer != src_serializer and
4323
self.target_repo._format.pack_compresses):
4324
self.target_repo.pack(hint=hint)
3873
self.target_repo.commit_write_group()
4325
3874
return [], set()
4327
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4328
target_rich_root = self.target_repo._format.rich_root_data
4329
target_tree_refs = self.target_repo._format.supports_tree_reference
4330
for record in substream:
4331
# Insert the delta directly
4332
inventory_delta_bytes = record.get_bytes_as('fulltext')
4333
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4335
parse_result = deserialiser.parse_text_bytes(
4336
inventory_delta_bytes)
4337
except inventory_delta.IncompatibleInventoryDelta, err:
4338
trace.mutter("Incompatible delta: %s", err.msg)
4339
raise errors.IncompatibleRevision(self.target_repo._format)
4340
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4341
revision_id = new_id
4342
parents = [key[0] for key in record.parents]
4343
self.target_repo.add_inventory_by_delta(
4344
basis_id, inv_delta, revision_id, parents)
4346
def _extract_and_insert_inventories(self, substream, serializer,
3876
def _extract_and_insert_inventories(self, substream, serializer):
4348
3877
"""Generate a new inventory versionedfile in target, converting data.
4350
3879
The inventory is retrieved from the source, (deserializing it), and
4351
3880
stored in the target (reserializing it in a different format).
4353
target_rich_root = self.target_repo._format.rich_root_data
4354
target_tree_refs = self.target_repo._format.supports_tree_reference
4355
3882
for record in substream:
4356
# It's not a delta, so it must be a fulltext in the source
4357
# serializer's format.
4358
3883
bytes = record.get_bytes_as('fulltext')
4359
3884
revision_id = record.key[0]
4360
3885
inv = serializer.read_inventory_from_string(bytes, revision_id)
4361
3886
parents = [key[0] for key in record.parents]
4362
3887
self.target_repo.add_inventory(revision_id, inv, parents)
4363
# No need to keep holding this full inv in memory when the rest of
4364
# the substream is likely to be all deltas.
4367
3889
def _extract_and_insert_revisions(self, substream, serializer):
4368
3890
for record in substream:
4531
4038
return (not self.from_repository._format.rich_root_data and
4532
4039
self.to_format.rich_root_data)
4534
def _get_inventory_stream(self, revision_ids, missing=False):
4535
from_format = self.from_repository._format
4536
if (from_format.supports_chks and self.to_format.supports_chks and
4537
from_format.network_name() == self.to_format.network_name()):
4538
raise AssertionError(
4539
"this case should be handled by GroupCHKStreamSource")
4540
elif 'forceinvdeltas' in debug.debug_flags:
4541
return self._get_convertable_inventory_stream(revision_ids,
4542
delta_versus_null=missing)
4543
elif from_format.network_name() == self.to_format.network_name():
4545
return self._get_simple_inventory_stream(revision_ids,
4547
elif (not from_format.supports_chks and not self.to_format.supports_chks
4548
and from_format._serializer == self.to_format._serializer):
4549
# Essentially the same format.
4550
return self._get_simple_inventory_stream(revision_ids,
4553
# Any time we switch serializations, we want to use an
4554
# inventory-delta based approach.
4555
return self._get_convertable_inventory_stream(revision_ids,
4556
delta_versus_null=missing)
4558
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4559
# NB: This currently reopens the inventory weave in source;
4560
# using a single stream interface instead would avoid this.
4561
from_weave = self.from_repository.inventories
4563
delta_closure = True
4565
delta_closure = not self.delta_on_metadata()
4566
yield ('inventories', from_weave.get_record_stream(
4567
[(rev_id,) for rev_id in revision_ids],
4568
self.inventory_fetch_order(), delta_closure))
4570
def _get_convertable_inventory_stream(self, revision_ids,
4571
delta_versus_null=False):
4572
# The two formats are sufficiently different that there is no fast
4573
# path, so we need to send just inventorydeltas, which any
4574
# sufficiently modern client can insert into any repository.
4575
# The StreamSink code expects to be able to
4576
# convert on the target, so we need to put bytes-on-the-wire that can
4577
# be converted. That means inventory deltas (if the remote is <1.19,
4578
# RemoteStreamSink will fallback to VFS to insert the deltas).
4579
yield ('inventory-deltas',
4580
self._stream_invs_as_deltas(revision_ids,
4581
delta_versus_null=delta_versus_null))
4583
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4584
"""Return a stream of inventory-deltas for the given rev ids.
4586
:param revision_ids: The list of inventories to transmit
4587
:param delta_versus_null: Don't try to find a minimal delta for this
4588
entry, instead compute the delta versus the NULL_REVISION. This
4589
effectively streams a complete inventory. Used for stuff like
4590
filling in missing parents, etc.
4592
from_repo = self.from_repository
4593
revision_keys = [(rev_id,) for rev_id in revision_ids]
4594
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4595
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4597
inventories = self.from_repository.iter_inventories(
4598
revision_ids, 'topological')
4599
format = from_repo._format
4600
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4601
inventory_cache = lru_cache.LRUCache(50)
4602
null_inventory = from_repo.revision_tree(
4603
_mod_revision.NULL_REVISION).inventory
4604
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4605
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4606
# repo back into a non-rich-root repo ought to be allowed)
4607
serializer = inventory_delta.InventoryDeltaSerializer(
4608
versioned_root=format.rich_root_data,
4609
tree_references=format.supports_tree_reference)
4610
for inv in inventories:
4611
key = (inv.revision_id,)
4612
parent_keys = parent_map.get(key, ())
4614
if not delta_versus_null and parent_keys:
4615
# The caller did not ask for complete inventories and we have
4616
# some parents that we can delta against. Make a delta against
4617
# each parent so that we can find the smallest.
4618
parent_ids = [parent_key[0] for parent_key in parent_keys]
4619
for parent_id in parent_ids:
4620
if parent_id not in invs_sent_so_far:
4621
# We don't know that the remote side has this basis, so
4624
if parent_id == _mod_revision.NULL_REVISION:
4625
parent_inv = null_inventory
4627
parent_inv = inventory_cache.get(parent_id, None)
4628
if parent_inv is None:
4629
parent_inv = from_repo.get_inventory(parent_id)
4630
candidate_delta = inv._make_delta(parent_inv)
4631
if (delta is None or
4632
len(delta) > len(candidate_delta)):
4633
delta = candidate_delta
4634
basis_id = parent_id
4636
# Either none of the parents ended up being suitable, or we
4637
# were asked to delta against NULL
4638
basis_id = _mod_revision.NULL_REVISION
4639
delta = inv._make_delta(null_inventory)
4640
invs_sent_so_far.add(inv.revision_id)
4641
inventory_cache[inv.revision_id] = inv
4642
delta_serialized = ''.join(
4643
serializer.delta_to_lines(basis_id, key[-1], delta))
4644
yield versionedfile.FulltextContentFactory(
4645
key, parent_keys, None, delta_serialized)
4648
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4649
stop_revision=None):
4650
"""Extend the partial history to include a given index
4652
If a stop_index is supplied, stop when that index has been reached.
4653
If a stop_revision is supplied, stop when that revision is
4654
encountered. Otherwise, stop when the beginning of history is
4657
:param stop_index: The index which should be present. When it is
4658
present, history extension will stop.
4659
:param stop_revision: The revision id which should be present. When
4660
it is encountered, history extension will stop.
4662
start_revision = partial_history_cache[-1]
4663
iterator = repo.iter_reverse_revision_history(start_revision)
4665
#skip the last revision in the list
4668
if (stop_index is not None and
4669
len(partial_history_cache) > stop_index):
4671
if partial_history_cache[-1] == stop_revision:
4673
revision_id = iterator.next()
4674
partial_history_cache.append(revision_id)
4675
except StopIteration:
added
removed
bzrlib/repository.py
