1168
1146
# The old API returned a list, should this actually be a set?
1169
1147
return parent_map.keys()
1171
def _check_inventories(self, checker):
1172
"""Check the inventories found from the revision scan.
1174
This is responsible for verifying the sha1 of inventories and
1175
creating a pending_keys set that covers data referenced by inventories.
1177
bar = ui.ui_factory.nested_progress_bar()
1179
self._do_check_inventories(checker, bar)
1183
def _do_check_inventories(self, checker, bar):
1184
"""Helper for _check_inventories."""
1186
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1187
kinds = ['chk_bytes', 'texts']
1188
count = len(checker.pending_keys)
1189
bar.update("inventories", 0, 2)
1190
current_keys = checker.pending_keys
1191
checker.pending_keys = {}
1192
# Accumulate current checks.
1193
for key in current_keys:
1194
if key[0] != 'inventories' and key[0] not in kinds:
1195
checker._report_items.append('unknown key type %r' % (key,))
1196
keys[key[0]].add(key[1:])
1197
if keys['inventories']:
1198
# NB: output order *should* be roughly sorted - topo or
1199
# inverse topo depending on repository - either way decent
1200
# to just delta against. However, pre-CHK formats didn't
1201
# try to optimise inventory layout on disk. As such the
1202
# pre-CHK code path does not use inventory deltas.
1204
for record in self.inventories.check(keys=keys['inventories']):
1205
if record.storage_kind == 'absent':
1206
checker._report_items.append(
1207
'Missing inventory {%s}' % (record.key,))
1209
last_object = self._check_record('inventories', record,
1210
checker, last_object,
1211
current_keys[('inventories',) + record.key])
1212
del keys['inventories']
1215
bar.update("texts", 1)
1216
while (checker.pending_keys or keys['chk_bytes']
1218
# Something to check.
1219
current_keys = checker.pending_keys
1220
checker.pending_keys = {}
1221
# Accumulate current checks.
1222
for key in current_keys:
1223
if key[0] not in kinds:
1224
checker._report_items.append('unknown key type %r' % (key,))
1225
keys[key[0]].add(key[1:])
1226
# Check the outermost kind only - inventories || chk_bytes || texts
1230
for record in getattr(self, kind).check(keys=keys[kind]):
1231
if record.storage_kind == 'absent':
1232
checker._report_items.append(
1233
'Missing %s {%s}' % (kind, record.key,))
1235
last_object = self._check_record(kind, record,
1236
checker, last_object, current_keys[(kind,) + record.key])
1240
def _check_record(self, kind, record, checker, last_object, item_data):
1241
"""Check a single text from this repository."""
1242
if kind == 'inventories':
1243
rev_id = record.key[0]
1244
inv = self._deserialise_inventory(rev_id,
1245
record.get_bytes_as('fulltext'))
1246
if last_object is not None:
1247
delta = inv._make_delta(last_object)
1248
for old_path, path, file_id, ie in delta:
1251
ie.check(checker, rev_id, inv)
1253
for path, ie in inv.iter_entries():
1254
ie.check(checker, rev_id, inv)
1255
if self._format.fast_deltas:
1257
elif kind == 'chk_bytes':
1258
# No code written to check chk_bytes for this repo format.
1259
checker._report_items.append(
1260
'unsupported key type chk_bytes for %s' % (record.key,))
1261
elif kind == 'texts':
1262
self._check_text(record, checker, item_data)
1264
checker._report_items.append(
1265
'unknown key type %s for %s' % (kind, record.key))
1267
def _check_text(self, record, checker, item_data):
1268
"""Check a single text."""
1269
# Check it is extractable.
1270
# TODO: check length.
1271
if record.storage_kind == 'chunked':
1272
chunks = record.get_bytes_as(record.storage_kind)
1273
sha1 = osutils.sha_strings(chunks)
1274
length = sum(map(len, chunks))
1276
content = record.get_bytes_as('fulltext')
1277
sha1 = osutils.sha_string(content)
1278
length = len(content)
1279
if item_data and sha1 != item_data[1]:
1280
checker._report_items.append(
1281
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1282
(record.key, sha1, item_data[1], item_data[2]))
1285
1150
def create(a_bzrdir):
1286
1151
"""Construct the current default format repository in a_bzrdir."""
1860
1715
@needs_read_lock
1861
1716
def get_revisions(self, revision_ids):
1862
"""Get many revisions at once.
1864
Repositories that need to check data on every revision read should
1865
subclass this method.
1717
"""Get many revisions at once."""
1867
1718
return self._get_revisions(revision_ids)
1869
1720
@needs_read_lock
1870
1721
def _get_revisions(self, revision_ids):
1871
1722
"""Core work logic to get many revisions without sanity checks."""
1723
for rev_id in revision_ids:
1724
if not rev_id or not isinstance(rev_id, basestring):
1725
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1726
keys = [(key,) for key in revision_ids]
1727
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1873
for revid, rev in self._iter_revisions(revision_ids):
1875
raise errors.NoSuchRevision(self, revid)
1729
for record in stream:
1730
if record.storage_kind == 'absent':
1731
raise errors.NoSuchRevision(self, record.key[0])
1732
text = record.get_bytes_as('fulltext')
1733
rev = self._serializer.read_revision_from_string(text)
1734
revs[record.key[0]] = rev
1877
1735
return [revs[revid] for revid in revision_ids]
1879
def _iter_revisions(self, revision_ids):
1880
"""Iterate over revision objects.
1882
:param revision_ids: An iterable of revisions to examine. None may be
1883
passed to request all revisions known to the repository. Note that
1884
not all repositories can find unreferenced revisions; for those
1885
repositories only referenced ones will be returned.
1886
:return: An iterator of (revid, revision) tuples. Absent revisions (
1887
those asked for but not available) are returned as (revid, None).
1889
if revision_ids is None:
1890
revision_ids = self.all_revision_ids()
1892
for rev_id in revision_ids:
1893
if not rev_id or not isinstance(rev_id, basestring):
1894
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1895
keys = [(key,) for key in revision_ids]
1896
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1897
for record in stream:
1898
revid = record.key[0]
1899
if record.storage_kind == 'absent':
1902
text = record.get_bytes_as('fulltext')
1903
rev = self._serializer.read_revision_from_string(text)
1738
def get_revision_xml(self, revision_id):
1739
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1740
# would have already do it.
1741
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1742
# TODO: this can't just be replaced by:
1743
# return self._serializer.write_revision_to_string(
1744
# self.get_revision(revision_id))
1745
# as cStringIO preservers the encoding unlike write_revision_to_string
1746
# or some other call down the path.
1747
rev = self.get_revision(revision_id)
1748
rev_tmp = cStringIO.StringIO()
1749
# the current serializer..
1750
self._serializer.write_revision(rev, rev_tmp)
1752
return rev_tmp.getvalue()
1906
1754
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1907
1755
"""Produce a generator of revision deltas.
2357
2210
:param revision_ids: The expected revision ids of the inventories.
2358
:param ordering: optional ordering, e.g. 'topological'. If not
2359
specified, the order of revision_ids will be preserved (by
2360
buffering if necessary).
2361
2211
:return: An iterator of inventories.
2363
2213
if ((None in revision_ids)
2364
2214
or (_mod_revision.NULL_REVISION in revision_ids)):
2365
2215
raise ValueError('cannot get null revision inventory')
2366
return self._iter_inventories(revision_ids, ordering)
2216
return self._iter_inventories(revision_ids)
2368
def _iter_inventories(self, revision_ids, ordering):
2218
def _iter_inventories(self, revision_ids):
2369
2219
"""single-document based inventory iteration."""
2370
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2371
for text, revision_id in inv_xmls:
2372
yield self._deserialise_inventory(revision_id, text)
2220
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2221
yield self.deserialise_inventory(revision_id, text)
2374
def _iter_inventory_xmls(self, revision_ids, ordering):
2375
if ordering is None:
2376
order_as_requested = True
2377
ordering = 'unordered'
2379
order_as_requested = False
2223
def _iter_inventory_xmls(self, revision_ids):
2380
2224
keys = [(revision_id,) for revision_id in revision_ids]
2383
if order_as_requested:
2384
key_iter = iter(keys)
2385
next_key = key_iter.next()
2386
stream = self.inventories.get_record_stream(keys, ordering, True)
2225
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2387
2226
text_chunks = {}
2388
2227
for record in stream:
2389
2228
if record.storage_kind != 'absent':
2390
chunks = record.get_bytes_as('chunked')
2391
if order_as_requested:
2392
text_chunks[record.key] = chunks
2394
yield ''.join(chunks), record.key[-1]
2229
text_chunks[record.key] = record.get_bytes_as('chunked')
2396
2231
raise errors.NoSuchRevision(self, record.key)
2397
if order_as_requested:
2398
# Yield as many results as we can while preserving order.
2399
while next_key in text_chunks:
2400
chunks = text_chunks.pop(next_key)
2401
yield ''.join(chunks), next_key[-1]
2403
next_key = key_iter.next()
2404
except StopIteration:
2405
# We still want to fully consume the get_record_stream,
2406
# just in case it is not actually finished at this point
2233
chunks = text_chunks.pop(key)
2234
yield ''.join(chunks), key[-1]
2410
def _deserialise_inventory(self, revision_id, xml):
2236
def deserialise_inventory(self, revision_id, xml):
2411
2237
"""Transform the xml into an inventory object.
2413
2239
:param revision_id: The expected revision id of the inventory.
2414
2240
:param xml: A serialised inventory.
2416
2242
result = self._serializer.read_inventory_from_string(xml, revision_id,
2417
entry_cache=self._inventory_entry_cache,
2418
return_from_cache=self._safe_to_return_from_cache)
2243
entry_cache=self._inventory_entry_cache)
2419
2244
if result.revision_id != revision_id:
2420
2245
raise AssertionError('revision id mismatch %s != %s' % (
2421
2246
result.revision_id, revision_id))
2249
def serialise_inventory(self, inv):
2250
return self._serializer.write_inventory_to_string(inv)
2252
def _serialise_inventory_to_lines(self, inv):
2253
return self._serializer.write_inventory_to_lines(inv)
2424
2255
def get_serializer_format(self):
2425
2256
return self._serializer.format_num
2427
2258
@needs_read_lock
2428
def _get_inventory_xml(self, revision_id):
2429
"""Get serialized inventory as a string."""
2430
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2259
def get_inventory_xml(self, revision_id):
2260
"""Get inventory XML as a file object."""
2261
texts = self._iter_inventory_xmls([revision_id])
2432
2263
text, revision_id = texts.next()
2433
2264
except StopIteration:
2434
2265
raise errors.HistoryMissing(self, 'inventory', revision_id)
2269
def get_inventory_sha1(self, revision_id):
2270
"""Return the sha1 hash of the inventory entry
2272
return self.get_revision(revision_id).inventory_sha1
2437
2274
def get_rev_id_for_revno(self, revno, known_pair):
2438
2275
"""Return the revision id of a revno, given a later (revno, revid)
2439
2276
pair in the same history.
3718
3514
# This is redundant with format.check_conversion_target(), however that
3719
3515
# raises an exception, and we just want to say "False" as in we won't
3720
3516
# support converting between these formats.
3721
if 'IDS_never' in debug.debug_flags:
3723
3517
if source.supports_rich_root() and not target.supports_rich_root():
3725
3519
if (source._format.supports_tree_reference
3726
3520
and not target._format.supports_tree_reference):
3728
if target._fallback_repositories and target._format.supports_chks:
3729
# IDS doesn't know how to copy CHKs for the parent inventories it
3730
# adds to stacked repos.
3732
if 'IDS_always' in debug.debug_flags:
3734
# Only use this code path for local source and target. IDS does far
3735
# too much IO (both bandwidth and roundtrips) over a network.
3736
if not source.bzrdir.transport.base.startswith('file:///'):
3738
if not target.bzrdir.transport.base.startswith('file:///'):
3742
def _get_trees(self, revision_ids, cache):
3744
for rev_id in revision_ids:
3746
possible_trees.append((rev_id, cache[rev_id]))
3748
# Not cached, but inventory might be present anyway.
3750
tree = self.source.revision_tree(rev_id)
3751
except errors.NoSuchRevision:
3752
# Nope, parent is ghost.
3755
cache[rev_id] = tree
3756
possible_trees.append((rev_id, tree))
3757
return possible_trees
3759
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3524
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3760
3525
"""Get the best delta and base for this revision.
3762
3527
:return: (basis_id, delta)
3529
possible_trees = [(parent_id, cache[parent_id])
3530
for parent_id in parent_ids
3531
if parent_id in cache]
3532
if len(possible_trees) == 0:
3533
# There either aren't any parents, or the parents aren't in the
3534
# cache, so just use the last converted tree
3535
possible_trees.append((basis_id, cache[basis_id]))
3765
# Generate deltas against each tree, to find the shortest.
3766
texts_possibly_new_in_tree = set()
3767
3537
for basis_id, basis_tree in possible_trees:
3768
3538
delta = tree.inventory._make_delta(basis_tree.inventory)
3769
for old_path, new_path, file_id, new_entry in delta:
3770
if new_path is None:
3771
# This file_id isn't present in the new rev, so we don't
3775
# Rich roots are handled elsewhere...
3777
kind = new_entry.kind
3778
if kind != 'directory' and kind != 'file':
3779
# No text record associated with this inventory entry.
3781
# This is a directory or file that has changed somehow.
3782
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3783
3539
deltas.append((len(delta), basis_id, delta))
3785
3541
return deltas[0][1:]
3787
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3788
"""Find all parent revisions that are absent, but for which the
3789
inventory is present, and copy those inventories.
3791
This is necessary to preserve correctness when the source is stacked
3792
without fallbacks configured. (Note that in cases like upgrade the
3793
source may be not have _fallback_repositories even though it is
3797
for parents in parent_map.values():
3798
parent_revs.update(parents)
3799
present_parents = self.source.get_parent_map(parent_revs)
3800
absent_parents = set(parent_revs).difference(present_parents)
3801
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3802
(rev_id,) for rev_id in absent_parents)
3803
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3804
for parent_tree in self.source.revision_trees(parent_inv_ids):
3805
current_revision_id = parent_tree.get_revision_id()
3806
parents_parents_keys = parent_invs_keys_for_stacking[
3807
(current_revision_id,)]
3808
parents_parents = [key[-1] for key in parents_parents_keys]
3809
basis_id = _mod_revision.NULL_REVISION
3810
basis_tree = self.source.revision_tree(basis_id)
3811
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3812
self.target.add_inventory_by_delta(
3813
basis_id, delta, current_revision_id, parents_parents)
3814
cache[current_revision_id] = parent_tree
3816
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3543
def _get_parent_keys(self, root_key, parent_map):
3544
"""Get the parent keys for a given root id."""
3545
root_id, rev_id = root_key
3546
# Include direct parents of the revision, but only if they used
3547
# the same root_id and are heads.
3549
for parent_id in parent_map[rev_id]:
3550
if parent_id == _mod_revision.NULL_REVISION:
3552
if parent_id not in self._revision_id_to_root_id:
3553
# We probably didn't read this revision, go spend the
3554
# extra effort to actually check
3556
tree = self.source.revision_tree(parent_id)
3557
except errors.NoSuchRevision:
3558
# Ghost, fill out _revision_id_to_root_id in case we
3559
# encounter this again.
3560
# But set parent_root_id to None since we don't really know
3561
parent_root_id = None
3563
parent_root_id = tree.get_root_id()
3564
self._revision_id_to_root_id[parent_id] = None
3566
parent_root_id = self._revision_id_to_root_id[parent_id]
3567
if root_id == parent_root_id:
3568
# With stacking we _might_ want to refer to a non-local
3569
# revision, but this code path only applies when we have the
3570
# full content available, so ghosts really are ghosts, not just
3571
# the edge of local data.
3572
parent_keys.append((parent_id,))
3574
# root_id may be in the parent anyway.
3576
tree = self.source.revision_tree(parent_id)
3577
except errors.NoSuchRevision:
3578
# ghost, can't refer to it.
3582
parent_keys.append((tree.inventory[root_id].revision,))
3583
except errors.NoSuchId:
3586
g = graph.Graph(self.source.revisions)
3587
heads = g.heads(parent_keys)
3589
for key in parent_keys:
3590
if key in heads and key not in selected_keys:
3591
selected_keys.append(key)
3592
return tuple([(root_id,)+ key for key in selected_keys])
3594
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3595
for root_key in root_keys_to_create:
3596
parent_keys = self._get_parent_keys(root_key, parent_map)
3597
yield versionedfile.FulltextContentFactory(root_key,
3598
parent_keys, None, '')
3600
def _fetch_batch(self, revision_ids, basis_id, cache):
3817
3601
"""Fetch across a few revisions.
3819
3603
:param revision_ids: The revisions to copy
3820
3604
:param basis_id: The revision_id of a tree that must be in cache, used
3821
3605
as a basis for delta when no other base is available
3822
3606
:param cache: A cache of RevisionTrees that we can use.
3823
:param a_graph: A Graph object to determine the heads() of the
3824
rich-root data stream.
3825
3607
:return: The revision_id of the last converted tree. The RevisionTree
3826
3608
for it will be in cache
3833
3615
pending_deltas = []
3834
3616
pending_revisions = []
3835
3617
parent_map = self.source.get_parent_map(revision_ids)
3836
self._fetch_parent_invs_for_stacking(parent_map, cache)
3837
self.source._safe_to_return_from_cache = True
3838
3618
for tree in self.source.revision_trees(revision_ids):
3839
# Find a inventory delta for this revision.
3840
# Find text entries that need to be copied, too.
3841
3619
current_revision_id = tree.get_revision_id()
3842
3620
parent_ids = parent_map.get(current_revision_id, ())
3843
parent_trees = self._get_trees(parent_ids, cache)
3844
possible_trees = list(parent_trees)
3845
if len(possible_trees) == 0:
3846
# There either aren't any parents, or the parents are ghosts,
3847
# so just use the last converted tree.
3848
possible_trees.append((basis_id, cache[basis_id]))
3849
3621
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3851
revision = self.source.get_revision(current_revision_id)
3852
pending_deltas.append((basis_id, delta,
3853
current_revision_id, revision.parent_ids))
3854
3623
if self._converting_to_rich_root:
3855
3624
self._revision_id_to_root_id[current_revision_id] = \
3856
3625
tree.get_root_id()
3857
# Determine which texts are in present in this revision but not in
3858
# any of the available parents.
3859
texts_possibly_new_in_tree = set()
3626
# Find text entries that need to be copied
3860
3627
for old_path, new_path, file_id, entry in delta:
3861
if new_path is None:
3862
# This file_id isn't present in the new rev
3866
if not self.target.supports_rich_root():
3867
# The target doesn't support rich root, so we don't
3870
if self._converting_to_rich_root:
3871
# This can't be copied normally, we have to insert
3873
root_keys_to_create.add((file_id, entry.revision))
3876
texts_possibly_new_in_tree.add((file_id, entry.revision))
3877
for basis_id, basis_tree in possible_trees:
3878
basis_inv = basis_tree.inventory
3879
for file_key in list(texts_possibly_new_in_tree):
3880
file_id, file_revision = file_key
3882
entry = basis_inv[file_id]
3883
except errors.NoSuchId:
3885
if entry.revision == file_revision:
3886
texts_possibly_new_in_tree.remove(file_key)
3887
text_keys.update(texts_possibly_new_in_tree)
3628
if new_path is not None:
3631
if not self.target.supports_rich_root():
3632
# The target doesn't support rich root, so we don't
3635
if self._converting_to_rich_root:
3636
# This can't be copied normally, we have to insert
3638
root_keys_to_create.add((file_id, entry.revision))
3640
text_keys.add((file_id, entry.revision))
3641
revision = self.source.get_revision(current_revision_id)
3642
pending_deltas.append((basis_id, delta,
3643
current_revision_id, revision.parent_ids))
3888
3644
pending_revisions.append(revision)
3889
3645
cache[current_revision_id] = tree
3890
3646
basis_id = current_revision_id
3891
self.source._safe_to_return_from_cache = False
3892
3647
# Copy file texts
3893
3648
from_texts = self.source.texts
3894
3649
to_texts = self.target.texts
3895
3650
if root_keys_to_create:
3896
root_stream = _mod_fetch._new_root_data_stream(
3897
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3898
self.source, graph=a_graph)
3651
root_stream = self._new_root_data_stream(root_keys_to_create,
3899
3653
to_texts.insert_record_stream(root_stream)
3900
3654
to_texts.insert_record_stream(from_texts.get_record_stream(
3901
3655
text_keys, self.target._format._fetch_order,
4563
4246
return (not self.from_repository._format.rich_root_data and
4564
4247
self.to_format.rich_root_data)
4566
def _get_inventory_stream(self, revision_ids, missing=False):
4249
def _get_inventory_stream(self, revision_ids):
4567
4250
from_format = self.from_repository._format
4568
if (from_format.supports_chks and self.to_format.supports_chks and
4569
from_format.network_name() == self.to_format.network_name()):
4570
raise AssertionError(
4571
"this case should be handled by GroupCHKStreamSource")
4572
elif 'forceinvdeltas' in debug.debug_flags:
4573
return self._get_convertable_inventory_stream(revision_ids,
4574
delta_versus_null=missing)
4575
elif from_format.network_name() == self.to_format.network_name():
4577
return self._get_simple_inventory_stream(revision_ids,
4579
elif (not from_format.supports_chks and not self.to_format.supports_chks
4580
and from_format._serializer == self.to_format._serializer):
4581
# Essentially the same format.
4582
return self._get_simple_inventory_stream(revision_ids,
4251
if (from_format.supports_chks and self.to_format.supports_chks
4252
and (from_format._serializer == self.to_format._serializer)):
4253
# Both sides support chks, and they use the same serializer, so it
4254
# is safe to transmit the chk pages and inventory pages across
4256
return self._get_chk_inventory_stream(revision_ids)
4257
elif (not from_format.supports_chks):
4258
# Source repository doesn't support chks. So we can transmit the
4259
# inventories 'as-is' and either they are just accepted on the
4260
# target, or the Sink will properly convert it.
4261
return self._get_simple_inventory_stream(revision_ids)
4585
# Any time we switch serializations, we want to use an
4586
# inventory-delta based approach.
4587
return self._get_convertable_inventory_stream(revision_ids,
4588
delta_versus_null=missing)
4263
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4264
# inventories. Note that this should probably be done somehow
4265
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4266
# figure out how a non-chk repository could possibly handle
4267
# deserializing an inventory stream from a chk repo, as it
4268
# doesn't have a way to understand individual pages.
4269
return self._get_convertable_inventory_stream(revision_ids)
4590
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4591
# NB: This currently reopens the inventory weave in source;
4592
# using a single stream interface instead would avoid this.
4271
def _get_simple_inventory_stream(self, revision_ids):
4593
4272
from_weave = self.from_repository.inventories
4595
delta_closure = True
4597
delta_closure = not self.delta_on_metadata()
4598
4273
yield ('inventories', from_weave.get_record_stream(
4599
4274
[(rev_id,) for rev_id in revision_ids],
4600
self.inventory_fetch_order(), delta_closure))
4602
def _get_convertable_inventory_stream(self, revision_ids,
4603
delta_versus_null=False):
4604
# The two formats are sufficiently different that there is no fast
4605
# path, so we need to send just inventorydeltas, which any
4606
# sufficiently modern client can insert into any repository.
4607
# The StreamSink code expects to be able to
4608
# convert on the target, so we need to put bytes-on-the-wire that can
4609
# be converted. That means inventory deltas (if the remote is <1.19,
4610
# RemoteStreamSink will fallback to VFS to insert the deltas).
4611
yield ('inventory-deltas',
4612
self._stream_invs_as_deltas(revision_ids,
4613
delta_versus_null=delta_versus_null))
4615
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4616
"""Return a stream of inventory-deltas for the given rev ids.
4618
:param revision_ids: The list of inventories to transmit
4619
:param delta_versus_null: Don't try to find a minimal delta for this
4620
entry, instead compute the delta versus the NULL_REVISION. This
4621
effectively streams a complete inventory. Used for stuff like
4622
filling in missing parents, etc.
4275
self.inventory_fetch_order(),
4276
not self.delta_on_metadata()))
4278
def _get_chk_inventory_stream(self, revision_ids):
4279
"""Fetch the inventory texts, along with the associated chk maps."""
4280
# We want an inventory outside of the search set, so that we can filter
4281
# out uninteresting chk pages. For now we use
4282
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4283
# could use that instead.
4284
start_rev_id = self.from_repository._find_revision_outside_set(
4286
start_rev_key = (start_rev_id,)
4287
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4288
if start_rev_id != _mod_revision.NULL_REVISION:
4289
inv_keys_to_fetch.append((start_rev_id,))
4290
# Any repo that supports chk_bytes must also support out-of-order
4291
# insertion. At least, that is how we expect it to work
4292
# We use get_record_stream instead of iter_inventories because we want
4293
# to be able to insert the stream as well. We could instead fetch
4294
# allowing deltas, and then iter_inventories, but we don't know whether
4295
# source or target is more 'local' anway.
4296
inv_stream = self.from_repository.inventories.get_record_stream(
4297
inv_keys_to_fetch, 'unordered',
4298
True) # We need them as full-texts so we can find their references
4299
uninteresting_chk_roots = set()
4300
interesting_chk_roots = set()
4301
def filter_inv_stream(inv_stream):
4302
for idx, record in enumerate(inv_stream):
4303
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4304
bytes = record.get_bytes_as('fulltext')
4305
chk_inv = inventory.CHKInventory.deserialise(
4306
self.from_repository.chk_bytes, bytes, record.key)
4307
if record.key == start_rev_key:
4308
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4309
p_id_map = chk_inv.parent_id_basename_to_file_id
4310
if p_id_map is not None:
4311
uninteresting_chk_roots.add(p_id_map.key())
4314
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4315
p_id_map = chk_inv.parent_id_basename_to_file_id
4316
if p_id_map is not None:
4317
interesting_chk_roots.add(p_id_map.key())
4318
### pb.update('fetch inventory', 0, 2)
4319
yield ('inventories', filter_inv_stream(inv_stream))
4320
# Now that we have worked out all of the interesting root nodes, grab
4321
# all of the interesting pages and insert them
4322
### pb.update('fetch inventory', 1, 2)
4323
interesting = chk_map.iter_interesting_nodes(
4324
self.from_repository.chk_bytes, interesting_chk_roots,
4325
uninteresting_chk_roots)
4326
def to_stream_adapter():
4327
"""Adapt the iter_interesting_nodes result to a single stream.
4329
iter_interesting_nodes returns records as it processes them, along
4330
with keys. However, we only want to return the records themselves.
4332
for record, items in interesting:
4333
if record is not None:
4335
# XXX: We could instead call get_record_stream(records.keys())
4336
# ATM, this will always insert the records as fulltexts, and
4337
# requires that you can hang on to records once you have gone
4338
# on to the next one. Further, it causes the target to
4339
# recompress the data. Testing shows it to be faster than
4340
# requesting the records again, though.
4341
yield ('chk_bytes', to_stream_adapter())
4342
### pb.update('fetch inventory', 2, 2)
4344
def _get_convertable_inventory_stream(self, revision_ids):
4345
# XXX: One of source or target is using chks, and they don't have
4346
# compatible serializations. The StreamSink code expects to be
4347
# able to convert on the target, so we need to put
4348
# bytes-on-the-wire that can be converted
4349
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4351
def _stream_invs_as_fulltexts(self, revision_ids):
4624
4352
from_repo = self.from_repository
4353
from_serializer = from_repo._format._serializer
4625
4354
revision_keys = [(rev_id,) for rev_id in revision_ids]
4626
4355
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4627
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4629
inventories = self.from_repository.iter_inventories(
4630
revision_ids, 'topological')
4631
format = from_repo._format
4632
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4633
inventory_cache = lru_cache.LRUCache(50)
4634
null_inventory = from_repo.revision_tree(
4635
_mod_revision.NULL_REVISION).inventory
4636
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4637
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4638
# repo back into a non-rich-root repo ought to be allowed)
4639
serializer = inventory_delta.InventoryDeltaSerializer(
4640
versioned_root=format.rich_root_data,
4641
tree_references=format.supports_tree_reference)
4642
for inv in inventories:
4356
for inv in self.from_repository.iter_inventories(revision_ids):
4357
# XXX: This is a bit hackish, but it works. Basically,
4358
# CHKSerializer 'accidentally' supports
4359
# read/write_inventory_to_string, even though that is never
4360
# the format that is stored on disk. It *does* give us a
4361
# single string representation for an inventory, so live with
4363
# This would be far better if we had a 'serialized inventory
4364
# delta' form. Then we could use 'inventory._make_delta', and
4365
# transmit that. This would both be faster to generate, and
4366
# result in fewer bytes-on-the-wire.
4367
as_bytes = from_serializer.write_inventory_to_string(inv)
4643
4368
key = (inv.revision_id,)
4644
4369
parent_keys = parent_map.get(key, ())
4646
if not delta_versus_null and parent_keys:
4647
# The caller did not ask for complete inventories and we have
4648
# some parents that we can delta against. Make a delta against
4649
# each parent so that we can find the smallest.
4650
parent_ids = [parent_key[0] for parent_key in parent_keys]
4651
for parent_id in parent_ids:
4652
if parent_id not in invs_sent_so_far:
4653
# We don't know that the remote side has this basis, so
4656
if parent_id == _mod_revision.NULL_REVISION:
4657
parent_inv = null_inventory
4659
parent_inv = inventory_cache.get(parent_id, None)
4660
if parent_inv is None:
4661
parent_inv = from_repo.get_inventory(parent_id)
4662
candidate_delta = inv._make_delta(parent_inv)
4663
if (delta is None or
4664
len(delta) > len(candidate_delta)):
4665
delta = candidate_delta
4666
basis_id = parent_id
4668
# Either none of the parents ended up being suitable, or we
4669
# were asked to delta against NULL
4670
basis_id = _mod_revision.NULL_REVISION
4671
delta = inv._make_delta(null_inventory)
4672
invs_sent_so_far.add(inv.revision_id)
4673
inventory_cache[inv.revision_id] = inv
4674
delta_serialized = ''.join(
4675
serializer.delta_to_lines(basis_id, key[-1], delta))
4676
4370
yield versionedfile.FulltextContentFactory(
4677
key, parent_keys, None, delta_serialized)
4371
key, parent_keys, None, as_bytes)
4680
4374
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
added
removed
bzrlib/repository.py
