1185
1142
# The old API returned a list, should this actually be a set?
1186
1143
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
1146
def create(a_bzrdir):
1303
1147
"""Construct the current default format repository in a_bzrdir."""
1884
1690
@needs_read_lock
1885
1691
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.
1692
"""Get many revisions at once."""
1891
1693
return self._get_revisions(revision_ids)
1893
1695
@needs_read_lock
1894
1696
def _get_revisions(self, revision_ids):
1895
1697
"""Core work logic to get many revisions without sanity checks."""
1698
for rev_id in revision_ids:
1699
if not rev_id or not isinstance(rev_id, basestring):
1700
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1701
keys = [(key,) for key in revision_ids]
1702
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)
1704
for record in stream:
1705
if record.storage_kind == 'absent':
1706
raise errors.NoSuchRevision(self, record.key[0])
1707
text = record.get_bytes_as('fulltext')
1708
rev = self._serializer.read_revision_from_string(text)
1709
revs[record.key[0]] = rev
1901
1710
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)
1713
def get_revision_xml(self, revision_id):
1714
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1715
# would have already do it.
1716
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1717
# TODO: this can't just be replaced by:
1718
# return self._serializer.write_revision_to_string(
1719
# self.get_revision(revision_id))
1720
# as cStringIO preservers the encoding unlike write_revision_to_string
1721
# or some other call down the path.
1722
rev = self.get_revision(revision_id)
1723
rev_tmp = cStringIO.StringIO()
1724
# the current serializer..
1725
self._serializer.write_revision(rev, rev_tmp)
1727
return rev_tmp.getvalue()
1930
1729
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1931
1730
"""Produce a generator of revision deltas.
2381
2185
: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
2186
:return: An iterator of inventories.
2387
2188
if ((None in revision_ids)
2388
2189
or (_mod_revision.NULL_REVISION in revision_ids)):
2389
2190
raise ValueError('cannot get null revision inventory')
2390
return self._iter_inventories(revision_ids, ordering)
2191
return self._iter_inventories(revision_ids)
2392
def _iter_inventories(self, revision_ids, ordering):
2193
def _iter_inventories(self, revision_ids):
2393
2194
"""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)
2195
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2196
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
2198
def _iter_inventory_xmls(self, revision_ids):
2404
2199
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)
2200
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2411
2201
text_chunks = {}
2412
2202
for record in stream:
2413
2203
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]
2204
text_chunks[record.key] = record.get_bytes_as('chunked')
2420
2206
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
2208
chunks = text_chunks.pop(key)
2209
yield ''.join(chunks), key[-1]
2434
def _deserialise_inventory(self, revision_id, xml):
2211
def deserialise_inventory(self, revision_id, xml):
2435
2212
"""Transform the xml into an inventory object.
2437
2214
:param revision_id: The expected revision id of the inventory.
2438
2215
:param xml: A serialised inventory.
2440
2217
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)
2218
entry_cache=self._inventory_entry_cache)
2443
2219
if result.revision_id != revision_id:
2444
2220
raise AssertionError('revision id mismatch %s != %s' % (
2445
2221
result.revision_id, revision_id))
2224
def serialise_inventory(self, inv):
2225
return self._serializer.write_inventory_to_string(inv)
2227
def _serialise_inventory_to_lines(self, inv):
2228
return self._serializer.write_inventory_to_lines(inv)
2448
2230
def get_serializer_format(self):
2449
2231
return self._serializer.format_num
2451
2233
@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')
2234
def get_inventory_xml(self, revision_id):
2235
"""Get inventory XML as a file object."""
2236
texts = self._iter_inventory_xmls([revision_id])
2456
2238
text, revision_id = texts.next()
2457
2239
except StopIteration:
2458
2240
raise errors.HistoryMissing(self, 'inventory', revision_id)
2244
def get_inventory_sha1(self, revision_id):
2245
"""Return the sha1 hash of the inventory entry
2247
return self.get_revision(revision_id).inventory_sha1
2461
2249
def get_rev_id_for_revno(self, revno, known_pair):
2462
2250
"""Return the revision id of a revno, given a later (revno, revid)
2463
2251
pair in the same history.
3742
3489
# This is redundant with format.check_conversion_target(), however that
3743
3490
# raises an exception, and we just want to say "False" as in we won't
3744
3491
# support converting between these formats.
3745
if 'IDS_never' in debug.debug_flags:
3747
3492
if source.supports_rich_root() and not target.supports_rich_root():
3749
3494
if (source._format.supports_tree_reference
3750
3495
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):
3499
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3784
3500
"""Get the best delta and base for this revision.
3786
3502
:return: (basis_id, delta)
3504
possible_trees = [(parent_id, cache[parent_id])
3505
for parent_id in parent_ids
3506
if parent_id in cache]
3507
if len(possible_trees) == 0:
3508
# There either aren't any parents, or the parents aren't in the
3509
# cache, so just use the last converted tree
3510
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
3512
for basis_id, basis_tree in possible_trees:
3792
3513
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
3514
deltas.append((len(delta), basis_id, delta))
3809
3516
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):
3518
def _get_parent_keys(self, root_key, parent_map):
3519
"""Get the parent keys for a given root id."""
3520
root_id, rev_id = root_key
3521
# Include direct parents of the revision, but only if they used
3522
# the same root_id and are heads.
3524
for parent_id in parent_map[rev_id]:
3525
if parent_id == _mod_revision.NULL_REVISION:
3527
if parent_id not in self._revision_id_to_root_id:
3528
# We probably didn't read this revision, go spend the
3529
# extra effort to actually check
3531
tree = self.source.revision_tree(parent_id)
3532
except errors.NoSuchRevision:
3533
# Ghost, fill out _revision_id_to_root_id in case we
3534
# encounter this again.
3535
# But set parent_root_id to None since we don't really know
3536
parent_root_id = None
3538
parent_root_id = tree.get_root_id()
3539
self._revision_id_to_root_id[parent_id] = None
3541
parent_root_id = self._revision_id_to_root_id[parent_id]
3542
if root_id == parent_root_id:
3543
# With stacking we _might_ want to refer to a non-local
3544
# revision, but this code path only applies when we have the
3545
# full content available, so ghosts really are ghosts, not just
3546
# the edge of local data.
3547
parent_keys.append((parent_id,))
3549
# root_id may be in the parent anyway.
3551
tree = self.source.revision_tree(parent_id)
3552
except errors.NoSuchRevision:
3553
# ghost, can't refer to it.
3557
parent_keys.append((tree.inventory[root_id].revision,))
3558
except errors.NoSuchId:
3561
g = graph.Graph(self.source.revisions)
3562
heads = g.heads(parent_keys)
3564
for key in parent_keys:
3565
if key in heads and key not in selected_keys:
3566
selected_keys.append(key)
3567
return tuple([(root_id,)+ key for key in selected_keys])
3569
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3570
for root_key in root_keys_to_create:
3571
parent_keys = self._get_parent_keys(root_key, parent_map)
3572
yield versionedfile.FulltextContentFactory(root_key,
3573
parent_keys, None, '')
3575
def _fetch_batch(self, revision_ids, basis_id, cache):
3841
3576
"""Fetch across a few revisions.
3843
3578
:param revision_ids: The revisions to copy
3844
3579
:param basis_id: The revision_id of a tree that must be in cache, used
3845
3580
as a basis for delta when no other base is available
3846
3581
: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
3582
:return: The revision_id of the last converted tree. The RevisionTree
3850
3583
for it will be in cache
3857
3590
pending_deltas = []
3858
3591
pending_revisions = []
3859
3592
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
3593
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
3594
current_revision_id = tree.get_revision_id()
3866
3595
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
3596
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
3598
if self._converting_to_rich_root:
3879
3599
self._revision_id_to_root_id[current_revision_id] = \
3880
3600
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()
3601
# Find text entries that need to be copied
3884
3602
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)
3603
if new_path is not None:
3606
if not self.target.supports_rich_root():
3607
# The target doesn't support rich root, so we don't
3610
if self._converting_to_rich_root:
3611
# This can't be copied normally, we have to insert
3613
root_keys_to_create.add((file_id, entry.revision))
3615
text_keys.add((file_id, entry.revision))
3616
revision = self.source.get_revision(current_revision_id)
3617
pending_deltas.append((basis_id, delta,
3618
current_revision_id, revision.parent_ids))
3912
3619
pending_revisions.append(revision)
3913
3620
cache[current_revision_id] = tree
3914
3621
basis_id = current_revision_id
3915
self.source._safe_to_return_from_cache = False
3916
3622
# Copy file texts
3917
3623
from_texts = self.source.texts
3918
3624
to_texts = self.target.texts
3919
3625
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)
3626
root_stream = self._new_root_data_stream(root_keys_to_create,
3923
3628
to_texts.insert_record_stream(root_stream)
3924
3629
to_texts.insert_record_stream(from_texts.get_record_stream(
3925
3630
text_keys, self.target._format._fetch_order,
4587
4221
return (not self.from_repository._format.rich_root_data and
4588
4222
self.to_format.rich_root_data)
4590
def _get_inventory_stream(self, revision_ids, missing=False):
4224
def _get_inventory_stream(self, revision_ids):
4591
4225
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,
4226
if (from_format.supports_chks and self.to_format.supports_chks
4227
and (from_format._serializer == self.to_format._serializer)):
4228
# Both sides support chks, and they use the same serializer, so it
4229
# is safe to transmit the chk pages and inventory pages across
4231
return self._get_chk_inventory_stream(revision_ids)
4232
elif (not from_format.supports_chks):
4233
# Source repository doesn't support chks. So we can transmit the
4234
# inventories 'as-is' and either they are just accepted on the
4235
# target, or the Sink will properly convert it.
4236
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)
4238
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4239
# inventories. Note that this should probably be done somehow
4240
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4241
# figure out how a non-chk repository could possibly handle
4242
# deserializing an inventory stream from a chk repo, as it
4243
# doesn't have a way to understand individual pages.
4244
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.
4246
def _get_simple_inventory_stream(self, revision_ids):
4617
4247
from_weave = self.from_repository.inventories
4619
delta_closure = True
4621
delta_closure = not self.delta_on_metadata()
4622
4248
yield ('inventories', from_weave.get_record_stream(
4623
4249
[(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.
4250
self.inventory_fetch_order(),
4251
not self.delta_on_metadata()))
4253
def _get_chk_inventory_stream(self, revision_ids):
4254
"""Fetch the inventory texts, along with the associated chk maps."""
4255
# We want an inventory outside of the search set, so that we can filter
4256
# out uninteresting chk pages. For now we use
4257
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4258
# could use that instead.
4259
start_rev_id = self.from_repository._find_revision_outside_set(
4261
start_rev_key = (start_rev_id,)
4262
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4263
if start_rev_id != _mod_revision.NULL_REVISION:
4264
inv_keys_to_fetch.append((start_rev_id,))
4265
# Any repo that supports chk_bytes must also support out-of-order
4266
# insertion. At least, that is how we expect it to work
4267
# We use get_record_stream instead of iter_inventories because we want
4268
# to be able to insert the stream as well. We could instead fetch
4269
# allowing deltas, and then iter_inventories, but we don't know whether
4270
# source or target is more 'local' anway.
4271
inv_stream = self.from_repository.inventories.get_record_stream(
4272
inv_keys_to_fetch, 'unordered',
4273
True) # We need them as full-texts so we can find their references
4274
uninteresting_chk_roots = set()
4275
interesting_chk_roots = set()
4276
def filter_inv_stream(inv_stream):
4277
for idx, record in enumerate(inv_stream):
4278
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4279
bytes = record.get_bytes_as('fulltext')
4280
chk_inv = inventory.CHKInventory.deserialise(
4281
self.from_repository.chk_bytes, bytes, record.key)
4282
if record.key == start_rev_key:
4283
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4284
p_id_map = chk_inv.parent_id_basename_to_file_id
4285
if p_id_map is not None:
4286
uninteresting_chk_roots.add(p_id_map.key())
4289
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4290
p_id_map = chk_inv.parent_id_basename_to_file_id
4291
if p_id_map is not None:
4292
interesting_chk_roots.add(p_id_map.key())
4293
### pb.update('fetch inventory', 0, 2)
4294
yield ('inventories', filter_inv_stream(inv_stream))
4295
# Now that we have worked out all of the interesting root nodes, grab
4296
# all of the interesting pages and insert them
4297
### pb.update('fetch inventory', 1, 2)
4298
interesting = chk_map.iter_interesting_nodes(
4299
self.from_repository.chk_bytes, interesting_chk_roots,
4300
uninteresting_chk_roots)
4301
def to_stream_adapter():
4302
"""Adapt the iter_interesting_nodes result to a single stream.
4304
iter_interesting_nodes returns records as it processes them, along
4305
with keys. However, we only want to return the records themselves.
4307
for record, items in interesting:
4308
if record is not None:
4310
# XXX: We could instead call get_record_stream(records.keys())
4311
# ATM, this will always insert the records as fulltexts, and
4312
# requires that you can hang on to records once you have gone
4313
# on to the next one. Further, it causes the target to
4314
# recompress the data. Testing shows it to be faster than
4315
# requesting the records again, though.
4316
yield ('chk_bytes', to_stream_adapter())
4317
### pb.update('fetch inventory', 2, 2)
4319
def _get_convertable_inventory_stream(self, revision_ids):
4320
# XXX: One of source or target is using chks, and they don't have
4321
# compatible serializations. The StreamSink code expects to be
4322
# able to convert on the target, so we need to put
4323
# bytes-on-the-wire that can be converted
4324
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4326
def _stream_invs_as_fulltexts(self, revision_ids):
4648
4327
from_repo = self.from_repository
4328
from_serializer = from_repo._format._serializer
4649
4329
revision_keys = [(rev_id,) for rev_id in revision_ids]
4650
4330
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:
4331
for inv in self.from_repository.iter_inventories(revision_ids):
4332
# XXX: This is a bit hackish, but it works. Basically,
4333
# CHKSerializer 'accidentally' supports
4334
# read/write_inventory_to_string, even though that is never
4335
# the format that is stored on disk. It *does* give us a
4336
# single string representation for an inventory, so live with
4338
# This would be far better if we had a 'serialized inventory
4339
# delta' form. Then we could use 'inventory._make_delta', and
4340
# transmit that. This would both be faster to generate, and
4341
# result in fewer bytes-on-the-wire.
4342
as_bytes = from_serializer.write_inventory_to_string(inv)
4667
4343
key = (inv.revision_id,)
4668
4344
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
4345
yield versionedfile.FulltextContentFactory(
4701
key, parent_keys, None, delta_serialized)
4346
key, parent_keys, None, as_bytes)
4704
4349
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
added
removed
bzrlib/repository.py
