1181
1154
# The old API returned a list, should this actually be a set?
1182
1155
return parent_map.keys()
1184
def _check_inventories(self, checker):
1185
"""Check the inventories found from the revision scan.
1187
This is responsible for verifying the sha1 of inventories and
1188
creating a pending_keys set that covers data referenced by inventories.
1190
bar = ui.ui_factory.nested_progress_bar()
1192
self._do_check_inventories(checker, bar)
1196
def _do_check_inventories(self, checker, bar):
1197
"""Helper for _check_inventories."""
1199
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1200
kinds = ['chk_bytes', 'texts']
1201
count = len(checker.pending_keys)
1202
bar.update("inventories", 0, 2)
1203
current_keys = checker.pending_keys
1204
checker.pending_keys = {}
1205
# Accumulate current checks.
1206
for key in current_keys:
1207
if key[0] != 'inventories' and key[0] not in kinds:
1208
checker._report_items.append('unknown key type %r' % (key,))
1209
keys[key[0]].add(key[1:])
1210
if keys['inventories']:
1211
# NB: output order *should* be roughly sorted - topo or
1212
# inverse topo depending on repository - either way decent
1213
# to just delta against. However, pre-CHK formats didn't
1214
# try to optimise inventory layout on disk. As such the
1215
# pre-CHK code path does not use inventory deltas.
1217
for record in self.inventories.check(keys=keys['inventories']):
1218
if record.storage_kind == 'absent':
1219
checker._report_items.append(
1220
'Missing inventory {%s}' % (record.key,))
1222
last_object = self._check_record('inventories', record,
1223
checker, last_object,
1224
current_keys[('inventories',) + record.key])
1225
del keys['inventories']
1228
bar.update("texts", 1)
1229
while (checker.pending_keys or keys['chk_bytes']
1231
# Something to check.
1232
current_keys = checker.pending_keys
1233
checker.pending_keys = {}
1234
# Accumulate current checks.
1235
for key in current_keys:
1236
if key[0] not in kinds:
1237
checker._report_items.append('unknown key type %r' % (key,))
1238
keys[key[0]].add(key[1:])
1239
# Check the outermost kind only - inventories || chk_bytes || texts
1243
for record in getattr(self, kind).check(keys=keys[kind]):
1244
if record.storage_kind == 'absent':
1245
checker._report_items.append(
1246
'Missing %s {%s}' % (kind, record.key,))
1248
last_object = self._check_record(kind, record,
1249
checker, last_object, current_keys[(kind,) + record.key])
1253
def _check_record(self, kind, record, checker, last_object, item_data):
1254
"""Check a single text from this repository."""
1255
if kind == 'inventories':
1256
rev_id = record.key[0]
1257
inv = self._deserialise_inventory(rev_id,
1258
record.get_bytes_as('fulltext'))
1259
if last_object is not None:
1260
delta = inv._make_delta(last_object)
1261
for old_path, path, file_id, ie in delta:
1264
ie.check(checker, rev_id, inv)
1266
for path, ie in inv.iter_entries():
1267
ie.check(checker, rev_id, inv)
1268
if self._format.fast_deltas:
1270
elif kind == 'chk_bytes':
1271
# No code written to check chk_bytes for this repo format.
1272
checker._report_items.append(
1273
'unsupported key type chk_bytes for %s' % (record.key,))
1274
elif kind == 'texts':
1275
self._check_text(record, checker, item_data)
1277
checker._report_items.append(
1278
'unknown key type %s for %s' % (kind, record.key))
1280
def _check_text(self, record, checker, item_data):
1281
"""Check a single text."""
1282
# Check it is extractable.
1283
# TODO: check length.
1284
if record.storage_kind == 'chunked':
1285
chunks = record.get_bytes_as(record.storage_kind)
1286
sha1 = osutils.sha_strings(chunks)
1287
length = sum(map(len, chunks))
1289
content = record.get_bytes_as('fulltext')
1290
sha1 = osutils.sha_string(content)
1291
length = len(content)
1292
if item_data and sha1 != item_data[1]:
1293
checker._report_items.append(
1294
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1295
(record.key, sha1, item_data[1], item_data[2]))
1298
1158
def create(a_bzrdir):
1299
1159
"""Construct the current default format repository in a_bzrdir."""
1880
1701
@needs_read_lock
1881
1702
def get_revisions(self, revision_ids):
1882
"""Get many revisions at once.
1884
Repositories that need to check data on every revision read should
1885
subclass this method.
1703
"""Get many revisions at once."""
1887
1704
return self._get_revisions(revision_ids)
1889
1706
@needs_read_lock
1890
1707
def _get_revisions(self, revision_ids):
1891
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)
1893
for revid, rev in self._iter_revisions(revision_ids):
1895
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
1897
1721
return [revs[revid] for revid in revision_ids]
1899
def _iter_revisions(self, revision_ids):
1900
"""Iterate over revision objects.
1902
:param revision_ids: An iterable of revisions to examine. None may be
1903
passed to request all revisions known to the repository. Note that
1904
not all repositories can find unreferenced revisions; for those
1905
repositories only referenced ones will be returned.
1906
:return: An iterator of (revid, revision) tuples. Absent revisions (
1907
those asked for but not available) are returned as (revid, None).
1909
if revision_ids is None:
1910
revision_ids = self.all_revision_ids()
1912
for rev_id in revision_ids:
1913
if not rev_id or not isinstance(rev_id, basestring):
1914
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1915
keys = [(key,) for key in revision_ids]
1916
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1917
for record in stream:
1918
revid = record.key[0]
1919
if record.storage_kind == 'absent':
1922
text = record.get_bytes_as('fulltext')
1923
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()
1926
1740
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1927
1741
"""Produce a generator of revision deltas.
2377
2173
:param revision_ids: The expected revision ids of the inventories.
2378
:param ordering: optional ordering, e.g. 'topological'. If not
2379
specified, the order of revision_ids will be preserved (by
2380
buffering if necessary).
2381
2174
:return: An iterator of inventories.
2383
2176
if ((None in revision_ids)
2384
2177
or (_mod_revision.NULL_REVISION in revision_ids)):
2385
2178
raise ValueError('cannot get null revision inventory')
2386
return self._iter_inventories(revision_ids, ordering)
2179
return self._iter_inventories(revision_ids)
2388
def _iter_inventories(self, revision_ids, ordering):
2181
def _iter_inventories(self, revision_ids):
2389
2182
"""single-document based inventory iteration."""
2390
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2391
for text, revision_id in inv_xmls:
2392
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)
2394
def _iter_inventory_xmls(self, revision_ids, ordering):
2395
if ordering is None:
2396
order_as_requested = True
2397
ordering = 'unordered'
2399
order_as_requested = False
2186
def _iter_inventory_xmls(self, revision_ids):
2400
2187
keys = [(revision_id,) for revision_id in revision_ids]
2403
if order_as_requested:
2404
key_iter = iter(keys)
2405
next_key = key_iter.next()
2406
stream = self.inventories.get_record_stream(keys, ordering, True)
2188
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2407
2189
text_chunks = {}
2408
2190
for record in stream:
2409
2191
if record.storage_kind != 'absent':
2410
chunks = record.get_bytes_as('chunked')
2411
if order_as_requested:
2412
text_chunks[record.key] = chunks
2414
yield ''.join(chunks), record.key[-1]
2192
text_chunks[record.key] = record.get_bytes_as('chunked')
2416
2194
raise errors.NoSuchRevision(self, record.key)
2417
if order_as_requested:
2418
# Yield as many results as we can while preserving order.
2419
while next_key in text_chunks:
2420
chunks = text_chunks.pop(next_key)
2421
yield ''.join(chunks), next_key[-1]
2423
next_key = key_iter.next()
2424
except StopIteration:
2425
# We still want to fully consume the get_record_stream,
2426
# just in case it is not actually finished at this point
2196
chunks = text_chunks.pop(key)
2197
yield ''.join(chunks), key[-1]
2430
def _deserialise_inventory(self, revision_id, xml):
2199
def deserialise_inventory(self, revision_id, xml):
2431
2200
"""Transform the xml into an inventory object.
2433
2202
:param revision_id: The expected revision id of the inventory.
2434
2203
:param xml: A serialised inventory.
2436
2205
result = self._serializer.read_inventory_from_string(xml, revision_id,
2437
entry_cache=self._inventory_entry_cache,
2438
return_from_cache=self._safe_to_return_from_cache)
2206
entry_cache=self._inventory_entry_cache)
2439
2207
if result.revision_id != revision_id:
2440
2208
raise AssertionError('revision id mismatch %s != %s' % (
2441
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)
2444
2218
def get_serializer_format(self):
2445
2219
return self._serializer.format_num
2447
2221
@needs_read_lock
2448
def _get_inventory_xml(self, revision_id):
2449
"""Get serialized inventory as a string."""
2450
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])
2452
2226
text, revision_id = texts.next()
2453
2227
except StopIteration:
2454
2228
raise errors.HistoryMissing(self, 'inventory', revision_id)
2457
def get_rev_id_for_revno(self, revno, known_pair):
2458
"""Return the revision id of a revno, given a later (revno, revid)
2459
pair in the same history.
2461
:return: if found (True, revid). If the available history ran out
2462
before reaching the revno, then this returns
2463
(False, (closest_revno, closest_revid)).
2232
def get_inventory_sha1(self, revision_id):
2233
"""Return the sha1 hash of the inventory entry
2465
known_revno, known_revid = known_pair
2466
partial_history = [known_revid]
2467
distance_from_known = known_revno - revno
2468
if distance_from_known < 0:
2470
'requested revno (%d) is later than given known revno (%d)'
2471
% (revno, known_revno))
2474
self, partial_history, stop_index=distance_from_known)
2475
except errors.RevisionNotPresent, err:
2476
if err.revision_id == known_revid:
2477
# The start revision (known_revid) wasn't found.
2479
# This is a stacked repository with no fallbacks, or a there's a
2480
# left-hand ghost. Either way, even though the revision named in
2481
# the error isn't in this repo, we know it's the next step in this
2482
# left-hand history.
2483
partial_history.append(err.revision_id)
2484
if len(partial_history) <= distance_from_known:
2485
# Didn't find enough history to get a revid for the revno.
2486
earliest_revno = known_revno - len(partial_history) + 1
2487
return (False, (earliest_revno, partial_history[-1]))
2488
if len(partial_history) - 1 > distance_from_known:
2489
raise AssertionError('_iter_for_revno returned too much history')
2490
return (True, partial_history[-1])
2235
return self.get_revision(revision_id).inventory_sha1
2492
2237
def iter_reverse_revision_history(self, revision_id):
2493
2238
"""Iterate backwards through revision ids in the lefthand history
3726
3415
return self.source.revision_ids_to_search_result(result_set)
3418
class InterPackRepo(InterSameDataRepository):
3419
"""Optimised code paths between Pack based repositories."""
3422
def _get_repo_format_to_test(self):
3423
from bzrlib.repofmt import pack_repo
3424
return pack_repo.RepositoryFormatKnitPack6RichRoot()
3427
def is_compatible(source, target):
3428
"""Be compatible with known Pack formats.
3430
We don't test for the stores being of specific types because that
3431
could lead to confusing results, and there is no need to be
3434
InterPackRepo does not support CHK based repositories.
3436
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3437
from bzrlib.repofmt.groupcompress_repo import RepositoryFormatCHK1
3439
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3440
isinstance(target._format, RepositoryFormatPack))
3441
not_packs = (isinstance(source._format, RepositoryFormatCHK1) or
3442
isinstance(target._format, RepositoryFormatCHK1))
3443
except AttributeError:
3445
if not_packs or not are_packs:
3447
return InterRepository._same_model(source, target)
3450
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3452
"""See InterRepository.fetch()."""
3453
if (len(self.source._fallback_repositories) > 0 or
3454
len(self.target._fallback_repositories) > 0):
3455
# The pack layer is not aware of fallback repositories, so when
3456
# fetching from a stacked repository or into a stacked repository
3457
# we use the generic fetch logic which uses the VersionedFiles
3458
# attributes on repository.
3459
from bzrlib.fetch import RepoFetcher
3460
fetcher = RepoFetcher(self.target, self.source, revision_id,
3461
pb, find_ghosts, fetch_spec=fetch_spec)
3462
if fetch_spec is not None:
3463
if len(list(fetch_spec.heads)) != 1:
3464
raise AssertionError(
3465
"InterPackRepo.fetch doesn't support "
3466
"fetching multiple heads yet.")
3467
revision_id = list(fetch_spec.heads)[0]
3469
if revision_id is None:
3471
# everything to do - use pack logic
3472
# to fetch from all packs to one without
3473
# inventory parsing etc, IFF nothing to be copied is in the target.
3475
source_revision_ids = frozenset(self.source.all_revision_ids())
3476
revision_ids = source_revision_ids - \
3477
frozenset(self.target.get_parent_map(source_revision_ids))
3478
revision_keys = [(revid,) for revid in revision_ids]
3479
index = self.target._pack_collection.revision_index.combined_index
3480
present_revision_ids = set(item[1][0] for item in
3481
index.iter_entries(revision_keys))
3482
revision_ids = set(revision_ids) - present_revision_ids
3483
# implementing the TODO will involve:
3484
# - detecting when all of a pack is selected
3485
# - avoiding as much as possible pre-selection, so the
3486
# more-core routines such as create_pack_from_packs can filter in
3487
# a just-in-time fashion. (though having a HEADS list on a
3488
# repository might make this a lot easier, because we could
3489
# sensibly detect 'new revisions' without doing a full index scan.
3490
elif _mod_revision.is_null(revision_id):
3494
revision_ids = self.search_missing_revision_ids(revision_id,
3495
find_ghosts=find_ghosts).get_keys()
3496
if len(revision_ids) == 0:
3498
return self._pack(self.source, self.target, revision_ids)
3500
def _pack(self, source, target, revision_ids):
3501
from bzrlib.repofmt.pack_repo import Packer
3502
packs = source._pack_collection.all_packs()
3503
pack = Packer(self.target._pack_collection, packs, '.fetch',
3504
revision_ids).pack()
3505
if pack is not None:
3506
self.target._pack_collection._save_pack_names()
3507
copied_revs = pack.get_revision_count()
3508
# Trigger an autopack. This may duplicate effort as we've just done
3509
# a pack creation, but for now it is simpler to think about as
3510
# 'upload data, then repack if needed'.
3511
self.target._pack_collection.autopack()
3512
return (copied_revs, [])
3517
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3518
"""See InterRepository.missing_revision_ids().
3520
:param find_ghosts: Find ghosts throughout the ancestry of
3523
if not find_ghosts and revision_id is not None:
3524
return self._walk_to_common_revisions([revision_id])
3525
elif revision_id is not None:
3526
# Find ghosts: search for revisions pointing from one repository to
3527
# the other, and vice versa, anywhere in the history of revision_id.
3528
graph = self.target.get_graph(other_repository=self.source)
3529
searcher = graph._make_breadth_first_searcher([revision_id])
3533
next_revs, ghosts = searcher.next_with_ghosts()
3534
except StopIteration:
3536
if revision_id in ghosts:
3537
raise errors.NoSuchRevision(self.source, revision_id)
3538
found_ids.update(next_revs)
3539
found_ids.update(ghosts)
3540
found_ids = frozenset(found_ids)
3541
# Double query here: should be able to avoid this by changing the
3542
# graph api further.
3543
result_set = found_ids - frozenset(
3544
self.target.get_parent_map(found_ids))
3546
source_ids = self.source.all_revision_ids()
3547
# source_ids is the worst possible case we may need to pull.
3548
# now we want to filter source_ids against what we actually
3549
# have in target, but don't try to check for existence where we know
3550
# we do not have a revision as that would be pointless.
3551
target_ids = set(self.target.all_revision_ids())
3552
result_set = set(source_ids).difference(target_ids)
3553
return self.source.revision_ids_to_search_result(result_set)
3729
3556
class InterDifferingSerializer(InterRepository):
3738
3565
# This is redundant with format.check_conversion_target(), however that
3739
3566
# raises an exception, and we just want to say "False" as in we won't
3740
3567
# support converting between these formats.
3741
if 'IDS_never' in debug.debug_flags:
3743
3568
if source.supports_rich_root() and not target.supports_rich_root():
3745
3570
if (source._format.supports_tree_reference
3746
3571
and not target._format.supports_tree_reference):
3748
if target._fallback_repositories and target._format.supports_chks:
3749
# IDS doesn't know how to copy CHKs for the parent inventories it
3750
# adds to stacked repos.
3752
if 'IDS_always' in debug.debug_flags:
3754
# Only use this code path for local source and target. IDS does far
3755
# too much IO (both bandwidth and roundtrips) over a network.
3756
if not source.bzrdir.transport.base.startswith('file:///'):
3758
if not target.bzrdir.transport.base.startswith('file:///'):
3762
def _get_trees(self, revision_ids, cache):
3764
for rev_id in revision_ids:
3766
possible_trees.append((rev_id, cache[rev_id]))
3768
# Not cached, but inventory might be present anyway.
3770
tree = self.source.revision_tree(rev_id)
3771
except errors.NoSuchRevision:
3772
# Nope, parent is ghost.
3775
cache[rev_id] = tree
3776
possible_trees.append((rev_id, tree))
3777
return possible_trees
3779
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3575
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3780
3576
"""Get the best delta and base for this revision.
3782
3578
:return: (basis_id, delta)
3580
possible_trees = [(parent_id, cache[parent_id])
3581
for parent_id in parent_ids
3582
if parent_id in cache]
3583
if len(possible_trees) == 0:
3584
# There either aren't any parents, or the parents aren't in the
3585
# cache, so just use the last converted tree
3586
possible_trees.append((basis_id, cache[basis_id]))
3785
# Generate deltas against each tree, to find the shortest.
3786
texts_possibly_new_in_tree = set()
3787
3588
for basis_id, basis_tree in possible_trees:
3788
3589
delta = tree.inventory._make_delta(basis_tree.inventory)
3789
for old_path, new_path, file_id, new_entry in delta:
3790
if new_path is None:
3791
# This file_id isn't present in the new rev, so we don't
3795
# Rich roots are handled elsewhere...
3797
kind = new_entry.kind
3798
if kind != 'directory' and kind != 'file':
3799
# No text record associated with this inventory entry.
3801
# This is a directory or file that has changed somehow.
3802
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3803
3590
deltas.append((len(delta), basis_id, delta))
3805
3592
return deltas[0][1:]
3807
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3808
"""Find all parent revisions that are absent, but for which the
3809
inventory is present, and copy those inventories.
3811
This is necessary to preserve correctness when the source is stacked
3812
without fallbacks configured. (Note that in cases like upgrade the
3813
source may be not have _fallback_repositories even though it is
3817
for parents in parent_map.values():
3818
parent_revs.update(parents)
3819
present_parents = self.source.get_parent_map(parent_revs)
3820
absent_parents = set(parent_revs).difference(present_parents)
3821
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3822
(rev_id,) for rev_id in absent_parents)
3823
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3824
for parent_tree in self.source.revision_trees(parent_inv_ids):
3825
current_revision_id = parent_tree.get_revision_id()
3826
parents_parents_keys = parent_invs_keys_for_stacking[
3827
(current_revision_id,)]
3828
parents_parents = [key[-1] for key in parents_parents_keys]
3829
basis_id = _mod_revision.NULL_REVISION
3830
basis_tree = self.source.revision_tree(basis_id)
3831
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3832
self.target.add_inventory_by_delta(
3833
basis_id, delta, current_revision_id, parents_parents)
3834
cache[current_revision_id] = parent_tree
3836
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3594
def _get_parent_keys(self, root_key, parent_map):
3595
"""Get the parent keys for a given root id."""
3596
root_id, rev_id = root_key
3597
# Include direct parents of the revision, but only if they used
3598
# the same root_id and are heads.
3600
for parent_id in parent_map[rev_id]:
3601
if parent_id == _mod_revision.NULL_REVISION:
3603
if parent_id not in self._revision_id_to_root_id:
3604
# We probably didn't read this revision, go spend the
3605
# extra effort to actually check
3607
tree = self.source.revision_tree(parent_id)
3608
except errors.NoSuchRevision:
3609
# Ghost, fill out _revision_id_to_root_id in case we
3610
# encounter this again.
3611
# But set parent_root_id to None since we don't really know
3612
parent_root_id = None
3614
parent_root_id = tree.get_root_id()
3615
self._revision_id_to_root_id[parent_id] = None
3617
parent_root_id = self._revision_id_to_root_id[parent_id]
3618
if root_id == parent_root_id:
3619
# With stacking we _might_ want to refer to a non-local
3620
# revision, but this code path only applies when we have the
3621
# full content available, so ghosts really are ghosts, not just
3622
# the edge of local data.
3623
parent_keys.append((parent_id,))
3625
# root_id may be in the parent anyway.
3627
tree = self.source.revision_tree(parent_id)
3628
except errors.NoSuchRevision:
3629
# ghost, can't refer to it.
3633
parent_keys.append((tree.inventory[root_id].revision,))
3634
except errors.NoSuchId:
3637
g = graph.Graph(self.source.revisions)
3638
heads = g.heads(parent_keys)
3640
for key in parent_keys:
3641
if key in heads and key not in selected_keys:
3642
selected_keys.append(key)
3643
return tuple([(root_id,)+ key for key in selected_keys])
3645
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3646
for root_key in root_keys_to_create:
3647
parent_keys = self._get_parent_keys(root_key, parent_map)
3648
yield versionedfile.FulltextContentFactory(root_key,
3649
parent_keys, None, '')
3651
def _fetch_batch(self, revision_ids, basis_id, cache):
3837
3652
"""Fetch across a few revisions.
3839
3654
:param revision_ids: The revisions to copy
3840
3655
:param basis_id: The revision_id of a tree that must be in cache, used
3841
3656
as a basis for delta when no other base is available
3842
3657
:param cache: A cache of RevisionTrees that we can use.
3843
:param a_graph: A Graph object to determine the heads() of the
3844
rich-root data stream.
3845
3658
:return: The revision_id of the last converted tree. The RevisionTree
3846
3659
for it will be in cache
3853
3666
pending_deltas = []
3854
3667
pending_revisions = []
3855
3668
parent_map = self.source.get_parent_map(revision_ids)
3856
self._fetch_parent_invs_for_stacking(parent_map, cache)
3857
self.source._safe_to_return_from_cache = True
3858
3669
for tree in self.source.revision_trees(revision_ids):
3859
# Find a inventory delta for this revision.
3860
# Find text entries that need to be copied, too.
3861
3670
current_revision_id = tree.get_revision_id()
3862
3671
parent_ids = parent_map.get(current_revision_id, ())
3863
parent_trees = self._get_trees(parent_ids, cache)
3864
possible_trees = list(parent_trees)
3865
if len(possible_trees) == 0:
3866
# There either aren't any parents, or the parents are ghosts,
3867
# so just use the last converted tree.
3868
possible_trees.append((basis_id, cache[basis_id]))
3869
3672
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3871
revision = self.source.get_revision(current_revision_id)
3872
pending_deltas.append((basis_id, delta,
3873
current_revision_id, revision.parent_ids))
3874
3674
if self._converting_to_rich_root:
3875
3675
self._revision_id_to_root_id[current_revision_id] = \
3876
3676
tree.get_root_id()
3877
# Determine which texts are in present in this revision but not in
3878
# any of the available parents.
3879
texts_possibly_new_in_tree = set()
3677
# Find text entries that need to be copied
3880
3678
for old_path, new_path, file_id, entry in delta:
3881
if new_path is None:
3882
# This file_id isn't present in the new rev
3886
if not self.target.supports_rich_root():
3887
# The target doesn't support rich root, so we don't
3890
if self._converting_to_rich_root:
3891
# This can't be copied normally, we have to insert
3893
root_keys_to_create.add((file_id, entry.revision))
3896
texts_possibly_new_in_tree.add((file_id, entry.revision))
3897
for basis_id, basis_tree in possible_trees:
3898
basis_inv = basis_tree.inventory
3899
for file_key in list(texts_possibly_new_in_tree):
3900
file_id, file_revision = file_key
3902
entry = basis_inv[file_id]
3903
except errors.NoSuchId:
3905
if entry.revision == file_revision:
3906
texts_possibly_new_in_tree.remove(file_key)
3907
text_keys.update(texts_possibly_new_in_tree)
3679
if new_path is not None:
3682
if not self.target.supports_rich_root():
3683
# The target doesn't support rich root, so we don't
3686
if self._converting_to_rich_root:
3687
# This can't be copied normally, we have to insert
3689
root_keys_to_create.add((file_id, entry.revision))
3691
text_keys.add((file_id, entry.revision))
3692
revision = self.source.get_revision(current_revision_id)
3693
pending_deltas.append((basis_id, delta,
3694
current_revision_id, revision.parent_ids))
3908
3695
pending_revisions.append(revision)
3909
3696
cache[current_revision_id] = tree
3910
3697
basis_id = current_revision_id
3911
self.source._safe_to_return_from_cache = False
3912
3698
# Copy file texts
3913
3699
from_texts = self.source.texts
3914
3700
to_texts = self.target.texts
3915
3701
if root_keys_to_create:
3916
root_stream = _mod_fetch._new_root_data_stream(
3917
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3918
self.source, graph=a_graph)
3702
root_stream = self._new_root_data_stream(root_keys_to_create,
3919
3704
to_texts.insert_record_stream(root_stream)
3920
3705
to_texts.insert_record_stream(from_texts.get_record_stream(
3921
3706
text_keys, self.target._format._fetch_order,
4370
4112
# missing keys can handle suspending a write group).
4371
4113
write_group_tokens = self.target_repo.suspend_write_group()
4372
4114
return write_group_tokens, missing_keys
4373
hint = self.target_repo.commit_write_group()
4374
if (to_serializer != src_serializer and
4375
self.target_repo._format.pack_compresses):
4376
self.target_repo.pack(hint=hint)
4115
self.target_repo.commit_write_group()
4377
4116
return [], set()
4379
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4380
target_rich_root = self.target_repo._format.rich_root_data
4381
target_tree_refs = self.target_repo._format.supports_tree_reference
4382
for record in substream:
4383
# Insert the delta directly
4384
inventory_delta_bytes = record.get_bytes_as('fulltext')
4385
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4387
parse_result = deserialiser.parse_text_bytes(
4388
inventory_delta_bytes)
4389
except inventory_delta.IncompatibleInventoryDelta, err:
4390
trace.mutter("Incompatible delta: %s", err.msg)
4391
raise errors.IncompatibleRevision(self.target_repo._format)
4392
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4393
revision_id = new_id
4394
parents = [key[0] for key in record.parents]
4395
self.target_repo.add_inventory_by_delta(
4396
basis_id, inv_delta, revision_id, parents)
4398
def _extract_and_insert_inventories(self, substream, serializer,
4118
def _extract_and_insert_inventories(self, substream, serializer):
4400
4119
"""Generate a new inventory versionedfile in target, converting data.
4402
4121
The inventory is retrieved from the source, (deserializing it), and
4403
4122
stored in the target (reserializing it in a different format).
4405
target_rich_root = self.target_repo._format.rich_root_data
4406
target_tree_refs = self.target_repo._format.supports_tree_reference
4407
4124
for record in substream:
4408
# It's not a delta, so it must be a fulltext in the source
4409
# serializer's format.
4410
4125
bytes = record.get_bytes_as('fulltext')
4411
4126
revision_id = record.key[0]
4412
4127
inv = serializer.read_inventory_from_string(bytes, revision_id)
4413
4128
parents = [key[0] for key in record.parents]
4414
4129
self.target_repo.add_inventory(revision_id, inv, parents)
4415
# No need to keep holding this full inv in memory when the rest of
4416
# the substream is likely to be all deltas.
4419
4131
def _extract_and_insert_revisions(self, substream, serializer):
4420
4132
for record in substream:
4583
4290
return (not self.from_repository._format.rich_root_data and
4584
4291
self.to_format.rich_root_data)
4586
def _get_inventory_stream(self, revision_ids, missing=False):
4293
def _get_inventory_stream(self, revision_ids):
4587
4294
from_format = self.from_repository._format
4588
if (from_format.supports_chks and self.to_format.supports_chks and
4589
from_format.network_name() == self.to_format.network_name()):
4590
raise AssertionError(
4591
"this case should be handled by GroupCHKStreamSource")
4592
elif 'forceinvdeltas' in debug.debug_flags:
4593
return self._get_convertable_inventory_stream(revision_ids,
4594
delta_versus_null=missing)
4595
elif from_format.network_name() == self.to_format.network_name():
4597
return self._get_simple_inventory_stream(revision_ids,
4599
elif (not from_format.supports_chks and not self.to_format.supports_chks
4600
and from_format._serializer == self.to_format._serializer):
4601
# Essentially the same format.
4602
return self._get_simple_inventory_stream(revision_ids,
4295
if (from_format.supports_chks and self.to_format.supports_chks
4296
and (from_format._serializer == self.to_format._serializer)):
4297
# Both sides support chks, and they use the same serializer, so it
4298
# is safe to transmit the chk pages and inventory pages across
4300
return self._get_chk_inventory_stream(revision_ids)
4301
elif (not from_format.supports_chks):
4302
# Source repository doesn't support chks. So we can transmit the
4303
# inventories 'as-is' and either they are just accepted on the
4304
# target, or the Sink will properly convert it.
4305
return self._get_simple_inventory_stream(revision_ids)
4605
# Any time we switch serializations, we want to use an
4606
# inventory-delta based approach.
4607
return self._get_convertable_inventory_stream(revision_ids,
4608
delta_versus_null=missing)
4307
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4308
# inventories. Note that this should probably be done somehow
4309
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4310
# figure out how a non-chk repository could possibly handle
4311
# deserializing an inventory stream from a chk repo, as it
4312
# doesn't have a way to understand individual pages.
4313
return self._get_convertable_inventory_stream(revision_ids)
4610
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4611
# NB: This currently reopens the inventory weave in source;
4612
# using a single stream interface instead would avoid this.
4315
def _get_simple_inventory_stream(self, revision_ids):
4613
4316
from_weave = self.from_repository.inventories
4615
delta_closure = True
4617
delta_closure = not self.delta_on_metadata()
4618
4317
yield ('inventories', from_weave.get_record_stream(
4619
4318
[(rev_id,) for rev_id in revision_ids],
4620
self.inventory_fetch_order(), delta_closure))
4622
def _get_convertable_inventory_stream(self, revision_ids,
4623
delta_versus_null=False):
4624
# The two formats are sufficiently different that there is no fast
4625
# path, so we need to send just inventorydeltas, which any
4626
# sufficiently modern client can insert into any repository.
4627
# The StreamSink code expects to be able to
4628
# convert on the target, so we need to put bytes-on-the-wire that can
4629
# be converted. That means inventory deltas (if the remote is <1.19,
4630
# RemoteStreamSink will fallback to VFS to insert the deltas).
4631
yield ('inventory-deltas',
4632
self._stream_invs_as_deltas(revision_ids,
4633
delta_versus_null=delta_versus_null))
4635
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4636
"""Return a stream of inventory-deltas for the given rev ids.
4638
:param revision_ids: The list of inventories to transmit
4639
:param delta_versus_null: Don't try to find a minimal delta for this
4640
entry, instead compute the delta versus the NULL_REVISION. This
4641
effectively streams a complete inventory. Used for stuff like
4642
filling in missing parents, etc.
4319
self.inventory_fetch_order(),
4320
not self.delta_on_metadata()))
4322
def _get_chk_inventory_stream(self, revision_ids):
4323
"""Fetch the inventory texts, along with the associated chk maps."""
4324
# We want an inventory outside of the search set, so that we can filter
4325
# out uninteresting chk pages. For now we use
4326
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4327
# could use that instead.
4328
start_rev_id = self.from_repository._find_revision_outside_set(
4330
start_rev_key = (start_rev_id,)
4331
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4332
if start_rev_id != _mod_revision.NULL_REVISION:
4333
inv_keys_to_fetch.append((start_rev_id,))
4334
# Any repo that supports chk_bytes must also support out-of-order
4335
# insertion. At least, that is how we expect it to work
4336
# We use get_record_stream instead of iter_inventories because we want
4337
# to be able to insert the stream as well. We could instead fetch
4338
# allowing deltas, and then iter_inventories, but we don't know whether
4339
# source or target is more 'local' anway.
4340
inv_stream = self.from_repository.inventories.get_record_stream(
4341
inv_keys_to_fetch, 'unordered',
4342
True) # We need them as full-texts so we can find their references
4343
uninteresting_chk_roots = set()
4344
interesting_chk_roots = set()
4345
def filter_inv_stream(inv_stream):
4346
for idx, record in enumerate(inv_stream):
4347
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4348
bytes = record.get_bytes_as('fulltext')
4349
chk_inv = inventory.CHKInventory.deserialise(
4350
self.from_repository.chk_bytes, bytes, record.key)
4351
if record.key == start_rev_key:
4352
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4353
p_id_map = chk_inv.parent_id_basename_to_file_id
4354
if p_id_map is not None:
4355
uninteresting_chk_roots.add(p_id_map.key())
4358
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4359
p_id_map = chk_inv.parent_id_basename_to_file_id
4360
if p_id_map is not None:
4361
interesting_chk_roots.add(p_id_map.key())
4362
### pb.update('fetch inventory', 0, 2)
4363
yield ('inventories', filter_inv_stream(inv_stream))
4364
# Now that we have worked out all of the interesting root nodes, grab
4365
# all of the interesting pages and insert them
4366
### pb.update('fetch inventory', 1, 2)
4367
interesting = chk_map.iter_interesting_nodes(
4368
self.from_repository.chk_bytes, interesting_chk_roots,
4369
uninteresting_chk_roots)
4370
def to_stream_adapter():
4371
"""Adapt the iter_interesting_nodes result to a single stream.
4373
iter_interesting_nodes returns records as it processes them, along
4374
with keys. However, we only want to return the records themselves.
4376
for record, items in interesting:
4377
if record is not None:
4379
# XXX: We could instead call get_record_stream(records.keys())
4380
# ATM, this will always insert the records as fulltexts, and
4381
# requires that you can hang on to records once you have gone
4382
# on to the next one. Further, it causes the target to
4383
# recompress the data. Testing shows it to be faster than
4384
# requesting the records again, though.
4385
yield ('chk_bytes', to_stream_adapter())
4386
### pb.update('fetch inventory', 2, 2)
4388
def _get_convertable_inventory_stream(self, revision_ids):
4389
# XXX: One of source or target is using chks, and they don't have
4390
# compatible serializations. The StreamSink code expects to be
4391
# able to convert on the target, so we need to put
4392
# bytes-on-the-wire that can be converted
4393
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4395
def _stream_invs_as_fulltexts(self, revision_ids):
4644
4396
from_repo = self.from_repository
4397
from_serializer = from_repo._format._serializer
4645
4398
revision_keys = [(rev_id,) for rev_id in revision_ids]
4646
4399
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4647
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4649
inventories = self.from_repository.iter_inventories(
4650
revision_ids, 'topological')
4651
format = from_repo._format
4652
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4653
inventory_cache = lru_cache.LRUCache(50)
4654
null_inventory = from_repo.revision_tree(
4655
_mod_revision.NULL_REVISION).inventory
4656
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4657
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4658
# repo back into a non-rich-root repo ought to be allowed)
4659
serializer = inventory_delta.InventoryDeltaSerializer(
4660
versioned_root=format.rich_root_data,
4661
tree_references=format.supports_tree_reference)
4662
for inv in inventories:
4400
for inv in self.from_repository.iter_inventories(revision_ids):
4401
# XXX: This is a bit hackish, but it works. Basically,
4402
# CHKSerializer 'accidentally' supports
4403
# read/write_inventory_to_string, even though that is never
4404
# the format that is stored on disk. It *does* give us a
4405
# single string representation for an inventory, so live with
4407
# This would be far better if we had a 'serialized inventory
4408
# delta' form. Then we could use 'inventory._make_delta', and
4409
# transmit that. This would both be faster to generate, and
4410
# result in fewer bytes-on-the-wire.
4411
as_bytes = from_serializer.write_inventory_to_string(inv)
4663
4412
key = (inv.revision_id,)
4664
4413
parent_keys = parent_map.get(key, ())
4666
if not delta_versus_null and parent_keys:
4667
# The caller did not ask for complete inventories and we have
4668
# some parents that we can delta against. Make a delta against
4669
# each parent so that we can find the smallest.
4670
parent_ids = [parent_key[0] for parent_key in parent_keys]
4671
for parent_id in parent_ids:
4672
if parent_id not in invs_sent_so_far:
4673
# We don't know that the remote side has this basis, so
4676
if parent_id == _mod_revision.NULL_REVISION:
4677
parent_inv = null_inventory
4679
parent_inv = inventory_cache.get(parent_id, None)
4680
if parent_inv is None:
4681
parent_inv = from_repo.get_inventory(parent_id)
4682
candidate_delta = inv._make_delta(parent_inv)
4683
if (delta is None or
4684
len(delta) > len(candidate_delta)):
4685
delta = candidate_delta
4686
basis_id = parent_id
4688
# Either none of the parents ended up being suitable, or we
4689
# were asked to delta against NULL
4690
basis_id = _mod_revision.NULL_REVISION
4691
delta = inv._make_delta(null_inventory)
4692
invs_sent_so_far.add(inv.revision_id)
4693
inventory_cache[inv.revision_id] = inv
4694
delta_serialized = ''.join(
4695
serializer.delta_to_lines(basis_id, key[-1], delta))
4696
4414
yield versionedfile.FulltextContentFactory(
4697
key, parent_keys, None, delta_serialized)
4700
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4701
stop_revision=None):
4702
"""Extend the partial history to include a given index
4704
If a stop_index is supplied, stop when that index has been reached.
4705
If a stop_revision is supplied, stop when that revision is
4706
encountered. Otherwise, stop when the beginning of history is
4709
:param stop_index: The index which should be present. When it is
4710
present, history extension will stop.
4711
:param stop_revision: The revision id which should be present. When
4712
it is encountered, history extension will stop.
4714
start_revision = partial_history_cache[-1]
4715
iterator = repo.iter_reverse_revision_history(start_revision)
4717
#skip the last revision in the list
4720
if (stop_index is not None and
4721
len(partial_history_cache) > stop_index):
4723
if partial_history_cache[-1] == stop_revision:
4725
revision_id = iterator.next()
4726
partial_history_cache.append(revision_id)
4727
except StopIteration:
4415
key, parent_keys, None, as_bytes)
added
removed
bzrlib/repository.py
