2346
2372
:param revision_ids: The expected revision ids of the inventories.
2373
:param ordering: optional ordering, e.g. 'topological'. If not
2374
specified, the order of revision_ids will be preserved (by
2375
buffering if necessary).
2347
2376
:return: An iterator of inventories.
2349
2378
if ((None in revision_ids)
2350
2379
or (_mod_revision.NULL_REVISION in revision_ids)):
2351
2380
raise ValueError('cannot get null revision inventory')
2352
return self._iter_inventories(revision_ids)
2381
return self._iter_inventories(revision_ids, ordering)
2354
def _iter_inventories(self, revision_ids):
2383
def _iter_inventories(self, revision_ids, ordering):
2355
2384
"""single-document based inventory iteration."""
2356
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2357
yield self.deserialise_inventory(revision_id, text)
2385
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2386
for text, revision_id in inv_xmls:
2387
yield self._deserialise_inventory(revision_id, text)
2359
def _iter_inventory_xmls(self, revision_ids):
2389
def _iter_inventory_xmls(self, revision_ids, ordering):
2390
if ordering is None:
2391
order_as_requested = True
2392
ordering = 'unordered'
2394
order_as_requested = False
2360
2395
keys = [(revision_id,) for revision_id in revision_ids]
2361
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2398
if order_as_requested:
2399
key_iter = iter(keys)
2400
next_key = key_iter.next()
2401
stream = self.inventories.get_record_stream(keys, ordering, True)
2362
2402
text_chunks = {}
2363
2403
for record in stream:
2364
2404
if record.storage_kind != 'absent':
2365
text_chunks[record.key] = record.get_bytes_as('chunked')
2405
chunks = record.get_bytes_as('chunked')
2406
if order_as_requested:
2407
text_chunks[record.key] = chunks
2409
yield ''.join(chunks), record.key[-1]
2367
2411
raise errors.NoSuchRevision(self, record.key)
2369
chunks = text_chunks.pop(key)
2370
yield ''.join(chunks), key[-1]
2412
if order_as_requested:
2413
# Yield as many results as we can while preserving order.
2414
while next_key in text_chunks:
2415
chunks = text_chunks.pop(next_key)
2416
yield ''.join(chunks), next_key[-1]
2418
next_key = key_iter.next()
2419
except StopIteration:
2420
# We still want to fully consume the get_record_stream,
2421
# just in case it is not actually finished at this point
2372
def deserialise_inventory(self, revision_id, xml):
2425
def _deserialise_inventory(self, revision_id, xml):
2373
2426
"""Transform the xml into an inventory object.
2375
2428
:param revision_id: The expected revision id of the inventory.
2376
2429
:param xml: A serialised inventory.
2378
2431
result = self._serializer.read_inventory_from_string(xml, revision_id,
2379
entry_cache=self._inventory_entry_cache)
2432
entry_cache=self._inventory_entry_cache,
2433
return_from_cache=self._safe_to_return_from_cache)
2380
2434
if result.revision_id != revision_id:
2381
2435
raise AssertionError('revision id mismatch %s != %s' % (
2382
2436
result.revision_id, revision_id))
2385
def serialise_inventory(self, inv):
2439
def _serialise_inventory(self, inv):
2386
2440
return self._serializer.write_inventory_to_string(inv)
2388
2442
def _serialise_inventory_to_lines(self, inv):
3660
3728
# This is redundant with format.check_conversion_target(), however that
3661
3729
# raises an exception, and we just want to say "False" as in we won't
3662
3730
# support converting between these formats.
3731
if 'IDS_never' in debug.debug_flags:
3663
3733
if source.supports_rich_root() and not target.supports_rich_root():
3665
3735
if (source._format.supports_tree_reference
3666
3736
and not target._format.supports_tree_reference):
3738
if target._fallback_repositories and target._format.supports_chks:
3739
# IDS doesn't know how to copy CHKs for the parent inventories it
3740
# adds to stacked repos.
3742
if 'IDS_always' in debug.debug_flags:
3744
# Only use this code path for local source and target. IDS does far
3745
# too much IO (both bandwidth and roundtrips) over a network.
3746
if not source.bzrdir.transport.base.startswith('file:///'):
3748
if not target.bzrdir.transport.base.startswith('file:///'):
3670
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3752
def _get_trees(self, revision_ids, cache):
3754
for rev_id in revision_ids:
3756
possible_trees.append((rev_id, cache[rev_id]))
3758
# Not cached, but inventory might be present anyway.
3760
tree = self.source.revision_tree(rev_id)
3761
except errors.NoSuchRevision:
3762
# Nope, parent is ghost.
3765
cache[rev_id] = tree
3766
possible_trees.append((rev_id, tree))
3767
return possible_trees
3769
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3671
3770
"""Get the best delta and base for this revision.
3673
3772
:return: (basis_id, delta)
3675
possible_trees = [(parent_id, cache[parent_id])
3676
for parent_id in parent_ids
3677
if parent_id in cache]
3678
if len(possible_trees) == 0:
3679
# There either aren't any parents, or the parents aren't in the
3680
# cache, so just use the last converted tree
3681
possible_trees.append((basis_id, cache[basis_id]))
3775
# Generate deltas against each tree, to find the shortest.
3776
texts_possibly_new_in_tree = set()
3683
3777
for basis_id, basis_tree in possible_trees:
3684
3778
delta = tree.inventory._make_delta(basis_tree.inventory)
3779
for old_path, new_path, file_id, new_entry in delta:
3780
if new_path is None:
3781
# This file_id isn't present in the new rev, so we don't
3785
# Rich roots are handled elsewhere...
3787
kind = new_entry.kind
3788
if kind != 'directory' and kind != 'file':
3789
# No text record associated with this inventory entry.
3791
# This is a directory or file that has changed somehow.
3792
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3685
3793
deltas.append((len(delta), basis_id, delta))
3687
3795
return deltas[0][1:]
3689
def _get_parent_keys(self, root_key, parent_map):
3690
"""Get the parent keys for a given root id."""
3691
root_id, rev_id = root_key
3692
# Include direct parents of the revision, but only if they used
3693
# the same root_id and are heads.
3695
for parent_id in parent_map[rev_id]:
3696
if parent_id == _mod_revision.NULL_REVISION:
3698
if parent_id not in self._revision_id_to_root_id:
3699
# We probably didn't read this revision, go spend the
3700
# extra effort to actually check
3702
tree = self.source.revision_tree(parent_id)
3703
except errors.NoSuchRevision:
3704
# Ghost, fill out _revision_id_to_root_id in case we
3705
# encounter this again.
3706
# But set parent_root_id to None since we don't really know
3707
parent_root_id = None
3709
parent_root_id = tree.get_root_id()
3710
self._revision_id_to_root_id[parent_id] = None
3712
parent_root_id = self._revision_id_to_root_id[parent_id]
3713
if root_id == parent_root_id:
3714
# With stacking we _might_ want to refer to a non-local
3715
# revision, but this code path only applies when we have the
3716
# full content available, so ghosts really are ghosts, not just
3717
# the edge of local data.
3718
parent_keys.append((parent_id,))
3720
# root_id may be in the parent anyway.
3722
tree = self.source.revision_tree(parent_id)
3723
except errors.NoSuchRevision:
3724
# ghost, can't refer to it.
3728
parent_keys.append((tree.inventory[root_id].revision,))
3729
except errors.NoSuchId:
3732
g = graph.Graph(self.source.revisions)
3733
heads = g.heads(parent_keys)
3735
for key in parent_keys:
3736
if key in heads and key not in selected_keys:
3737
selected_keys.append(key)
3738
return tuple([(root_id,)+ key for key in selected_keys])
3740
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3741
for root_key in root_keys_to_create:
3742
parent_keys = self._get_parent_keys(root_key, parent_map)
3743
yield versionedfile.FulltextContentFactory(root_key,
3744
parent_keys, None, '')
3746
def _fetch_batch(self, revision_ids, basis_id, cache):
3797
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3798
"""Find all parent revisions that are absent, but for which the
3799
inventory is present, and copy those inventories.
3801
This is necessary to preserve correctness when the source is stacked
3802
without fallbacks configured. (Note that in cases like upgrade the
3803
source may be not have _fallback_repositories even though it is
3807
for parents in parent_map.values():
3808
parent_revs.update(parents)
3809
present_parents = self.source.get_parent_map(parent_revs)
3810
absent_parents = set(parent_revs).difference(present_parents)
3811
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3812
(rev_id,) for rev_id in absent_parents)
3813
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3814
for parent_tree in self.source.revision_trees(parent_inv_ids):
3815
current_revision_id = parent_tree.get_revision_id()
3816
parents_parents_keys = parent_invs_keys_for_stacking[
3817
(current_revision_id,)]
3818
parents_parents = [key[-1] for key in parents_parents_keys]
3819
basis_id = _mod_revision.NULL_REVISION
3820
basis_tree = self.source.revision_tree(basis_id)
3821
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3822
self.target.add_inventory_by_delta(
3823
basis_id, delta, current_revision_id, parents_parents)
3824
cache[current_revision_id] = parent_tree
3826
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3747
3827
"""Fetch across a few revisions.
3749
3829
:param revision_ids: The revisions to copy
3750
3830
:param basis_id: The revision_id of a tree that must be in cache, used
3751
3831
as a basis for delta when no other base is available
3752
3832
:param cache: A cache of RevisionTrees that we can use.
3833
:param a_graph: A Graph object to determine the heads() of the
3834
rich-root data stream.
3753
3835
:return: The revision_id of the last converted tree. The RevisionTree
3754
3836
for it will be in cache
3761
3843
pending_deltas = []
3762
3844
pending_revisions = []
3763
3845
parent_map = self.source.get_parent_map(revision_ids)
3846
self._fetch_parent_invs_for_stacking(parent_map, cache)
3847
self.source._safe_to_return_from_cache = True
3764
3848
for tree in self.source.revision_trees(revision_ids):
3849
# Find a inventory delta for this revision.
3850
# Find text entries that need to be copied, too.
3765
3851
current_revision_id = tree.get_revision_id()
3766
3852
parent_ids = parent_map.get(current_revision_id, ())
3853
parent_trees = self._get_trees(parent_ids, cache)
3854
possible_trees = list(parent_trees)
3855
if len(possible_trees) == 0:
3856
# There either aren't any parents, or the parents are ghosts,
3857
# so just use the last converted tree.
3858
possible_trees.append((basis_id, cache[basis_id]))
3767
3859
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3861
revision = self.source.get_revision(current_revision_id)
3862
pending_deltas.append((basis_id, delta,
3863
current_revision_id, revision.parent_ids))
3769
3864
if self._converting_to_rich_root:
3770
3865
self._revision_id_to_root_id[current_revision_id] = \
3771
3866
tree.get_root_id()
3772
# Find text entries that need to be copied
3867
# Determine which texts are in present in this revision but not in
3868
# any of the available parents.
3869
texts_possibly_new_in_tree = set()
3773
3870
for old_path, new_path, file_id, entry in delta:
3774
if new_path is not None:
3777
if not self.target.supports_rich_root():
3778
# The target doesn't support rich root, so we don't
3781
if self._converting_to_rich_root:
3782
# This can't be copied normally, we have to insert
3784
root_keys_to_create.add((file_id, entry.revision))
3786
text_keys.add((file_id, entry.revision))
3787
revision = self.source.get_revision(current_revision_id)
3788
pending_deltas.append((basis_id, delta,
3789
current_revision_id, revision.parent_ids))
3871
if new_path is None:
3872
# This file_id isn't present in the new rev
3876
if not self.target.supports_rich_root():
3877
# The target doesn't support rich root, so we don't
3880
if self._converting_to_rich_root:
3881
# This can't be copied normally, we have to insert
3883
root_keys_to_create.add((file_id, entry.revision))
3886
texts_possibly_new_in_tree.add((file_id, entry.revision))
3887
for basis_id, basis_tree in possible_trees:
3888
basis_inv = basis_tree.inventory
3889
for file_key in list(texts_possibly_new_in_tree):
3890
file_id, file_revision = file_key
3892
entry = basis_inv[file_id]
3893
except errors.NoSuchId:
3895
if entry.revision == file_revision:
3896
texts_possibly_new_in_tree.remove(file_key)
3897
text_keys.update(texts_possibly_new_in_tree)
3790
3898
pending_revisions.append(revision)
3791
3899
cache[current_revision_id] = tree
3792
3900
basis_id = current_revision_id
3901
self.source._safe_to_return_from_cache = False
3793
3902
# Copy file texts
3794
3903
from_texts = self.source.texts
3795
3904
to_texts = self.target.texts
3796
3905
if root_keys_to_create:
3797
root_stream = self._new_root_data_stream(root_keys_to_create,
3906
root_stream = _mod_fetch._new_root_data_stream(
3907
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3908
self.source, graph=a_graph)
3799
3909
to_texts.insert_record_stream(root_stream)
3800
3910
to_texts.insert_record_stream(from_texts.get_record_stream(
3801
3911
text_keys, self.target._format._fetch_order,
4403
4576
return (not self.from_repository._format.rich_root_data and
4404
4577
self.to_format.rich_root_data)
4406
def _get_inventory_stream(self, revision_ids):
4579
def _get_inventory_stream(self, revision_ids, missing=False):
4407
4580
from_format = self.from_repository._format
4408
if (from_format.supports_chks and self.to_format.supports_chks
4409
and (from_format._serializer == self.to_format._serializer)):
4410
# Both sides support chks, and they use the same serializer, so it
4411
# is safe to transmit the chk pages and inventory pages across
4413
return self._get_chk_inventory_stream(revision_ids)
4414
elif (not from_format.supports_chks):
4415
# Source repository doesn't support chks. So we can transmit the
4416
# inventories 'as-is' and either they are just accepted on the
4417
# target, or the Sink will properly convert it.
4418
return self._get_simple_inventory_stream(revision_ids)
4581
if (from_format.supports_chks and self.to_format.supports_chks and
4582
from_format.network_name() == self.to_format.network_name()):
4583
raise AssertionError(
4584
"this case should be handled by GroupCHKStreamSource")
4585
elif 'forceinvdeltas' in debug.debug_flags:
4586
return self._get_convertable_inventory_stream(revision_ids,
4587
delta_versus_null=missing)
4588
elif from_format.network_name() == self.to_format.network_name():
4590
return self._get_simple_inventory_stream(revision_ids,
4592
elif (not from_format.supports_chks and not self.to_format.supports_chks
4593
and from_format._serializer == self.to_format._serializer):
4594
# Essentially the same format.
4595
return self._get_simple_inventory_stream(revision_ids,
4420
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4421
# inventories. Note that this should probably be done somehow
4422
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4423
# figure out how a non-chk repository could possibly handle
4424
# deserializing an inventory stream from a chk repo, as it
4425
# doesn't have a way to understand individual pages.
4426
return self._get_convertable_inventory_stream(revision_ids)
4598
# Any time we switch serializations, we want to use an
4599
# inventory-delta based approach.
4600
return self._get_convertable_inventory_stream(revision_ids,
4601
delta_versus_null=missing)
4428
def _get_simple_inventory_stream(self, revision_ids):
4603
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4604
# NB: This currently reopens the inventory weave in source;
4605
# using a single stream interface instead would avoid this.
4429
4606
from_weave = self.from_repository.inventories
4608
delta_closure = True
4610
delta_closure = not self.delta_on_metadata()
4430
4611
yield ('inventories', from_weave.get_record_stream(
4431
4612
[(rev_id,) for rev_id in revision_ids],
4432
self.inventory_fetch_order(),
4433
not self.delta_on_metadata()))
4435
def _get_chk_inventory_stream(self, revision_ids):
4436
"""Fetch the inventory texts, along with the associated chk maps."""
4437
# We want an inventory outside of the search set, so that we can filter
4438
# out uninteresting chk pages. For now we use
4439
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4440
# could use that instead.
4441
start_rev_id = self.from_repository._find_revision_outside_set(
4443
start_rev_key = (start_rev_id,)
4444
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4445
if start_rev_id != _mod_revision.NULL_REVISION:
4446
inv_keys_to_fetch.append((start_rev_id,))
4447
# Any repo that supports chk_bytes must also support out-of-order
4448
# insertion. At least, that is how we expect it to work
4449
# We use get_record_stream instead of iter_inventories because we want
4450
# to be able to insert the stream as well. We could instead fetch
4451
# allowing deltas, and then iter_inventories, but we don't know whether
4452
# source or target is more 'local' anway.
4453
inv_stream = self.from_repository.inventories.get_record_stream(
4454
inv_keys_to_fetch, 'unordered',
4455
True) # We need them as full-texts so we can find their references
4456
uninteresting_chk_roots = set()
4457
interesting_chk_roots = set()
4458
def filter_inv_stream(inv_stream):
4459
for idx, record in enumerate(inv_stream):
4460
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4461
bytes = record.get_bytes_as('fulltext')
4462
chk_inv = inventory.CHKInventory.deserialise(
4463
self.from_repository.chk_bytes, bytes, record.key)
4464
if record.key == start_rev_key:
4465
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4466
p_id_map = chk_inv.parent_id_basename_to_file_id
4467
if p_id_map is not None:
4468
uninteresting_chk_roots.add(p_id_map.key())
4471
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4472
p_id_map = chk_inv.parent_id_basename_to_file_id
4473
if p_id_map is not None:
4474
interesting_chk_roots.add(p_id_map.key())
4475
### pb.update('fetch inventory', 0, 2)
4476
yield ('inventories', filter_inv_stream(inv_stream))
4477
# Now that we have worked out all of the interesting root nodes, grab
4478
# all of the interesting pages and insert them
4479
### pb.update('fetch inventory', 1, 2)
4480
interesting = chk_map.iter_interesting_nodes(
4481
self.from_repository.chk_bytes, interesting_chk_roots,
4482
uninteresting_chk_roots)
4483
def to_stream_adapter():
4484
"""Adapt the iter_interesting_nodes result to a single stream.
4486
iter_interesting_nodes returns records as it processes them, along
4487
with keys. However, we only want to return the records themselves.
4489
for record, items in interesting:
4490
if record is not None:
4492
# XXX: We could instead call get_record_stream(records.keys())
4493
# ATM, this will always insert the records as fulltexts, and
4494
# requires that you can hang on to records once you have gone
4495
# on to the next one. Further, it causes the target to
4496
# recompress the data. Testing shows it to be faster than
4497
# requesting the records again, though.
4498
yield ('chk_bytes', to_stream_adapter())
4499
### pb.update('fetch inventory', 2, 2)
4501
def _get_convertable_inventory_stream(self, revision_ids):
4502
# XXX: One of source or target is using chks, and they don't have
4503
# compatible serializations. The StreamSink code expects to be
4504
# able to convert on the target, so we need to put
4505
# bytes-on-the-wire that can be converted
4506
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4508
def _stream_invs_as_fulltexts(self, revision_ids):
4613
self.inventory_fetch_order(), delta_closure))
4615
def _get_convertable_inventory_stream(self, revision_ids,
4616
delta_versus_null=False):
4617
# The two formats are sufficiently different that there is no fast
4618
# path, so we need to send just inventorydeltas, which any
4619
# sufficiently modern client can insert into any repository.
4620
# The StreamSink code expects to be able to
4621
# convert on the target, so we need to put bytes-on-the-wire that can
4622
# be converted. That means inventory deltas (if the remote is <1.19,
4623
# RemoteStreamSink will fallback to VFS to insert the deltas).
4624
yield ('inventory-deltas',
4625
self._stream_invs_as_deltas(revision_ids,
4626
delta_versus_null=delta_versus_null))
4628
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4629
"""Return a stream of inventory-deltas for the given rev ids.
4631
:param revision_ids: The list of inventories to transmit
4632
:param delta_versus_null: Don't try to find a minimal delta for this
4633
entry, instead compute the delta versus the NULL_REVISION. This
4634
effectively streams a complete inventory. Used for stuff like
4635
filling in missing parents, etc.
4509
4637
from_repo = self.from_repository
4510
from_serializer = from_repo._format._serializer
4511
4638
revision_keys = [(rev_id,) for rev_id in revision_ids]
4512
4639
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4513
for inv in self.from_repository.iter_inventories(revision_ids):
4514
# XXX: This is a bit hackish, but it works. Basically,
4515
# CHKSerializer 'accidentally' supports
4516
# read/write_inventory_to_string, even though that is never
4517
# the format that is stored on disk. It *does* give us a
4518
# single string representation for an inventory, so live with
4520
# This would be far better if we had a 'serialized inventory
4521
# delta' form. Then we could use 'inventory._make_delta', and
4522
# transmit that. This would both be faster to generate, and
4523
# result in fewer bytes-on-the-wire.
4524
as_bytes = from_serializer.write_inventory_to_string(inv)
4640
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4642
inventories = self.from_repository.iter_inventories(
4643
revision_ids, 'topological')
4644
format = from_repo._format
4645
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4646
inventory_cache = lru_cache.LRUCache(50)
4647
null_inventory = from_repo.revision_tree(
4648
_mod_revision.NULL_REVISION).inventory
4649
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4650
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4651
# repo back into a non-rich-root repo ought to be allowed)
4652
serializer = inventory_delta.InventoryDeltaSerializer(
4653
versioned_root=format.rich_root_data,
4654
tree_references=format.supports_tree_reference)
4655
for inv in inventories:
4525
4656
key = (inv.revision_id,)
4526
4657
parent_keys = parent_map.get(key, ())
4659
if not delta_versus_null and parent_keys:
4660
# The caller did not ask for complete inventories and we have
4661
# some parents that we can delta against. Make a delta against
4662
# each parent so that we can find the smallest.
4663
parent_ids = [parent_key[0] for parent_key in parent_keys]
4664
for parent_id in parent_ids:
4665
if parent_id not in invs_sent_so_far:
4666
# We don't know that the remote side has this basis, so
4669
if parent_id == _mod_revision.NULL_REVISION:
4670
parent_inv = null_inventory
4672
parent_inv = inventory_cache.get(parent_id, None)
4673
if parent_inv is None:
4674
parent_inv = from_repo.get_inventory(parent_id)
4675
candidate_delta = inv._make_delta(parent_inv)
4676
if (delta is None or
4677
len(delta) > len(candidate_delta)):
4678
delta = candidate_delta
4679
basis_id = parent_id
4681
# Either none of the parents ended up being suitable, or we
4682
# were asked to delta against NULL
4683
basis_id = _mod_revision.NULL_REVISION
4684
delta = inv._make_delta(null_inventory)
4685
invs_sent_so_far.add(inv.revision_id)
4686
inventory_cache[inv.revision_id] = inv
4687
delta_serialized = ''.join(
4688
serializer.delta_to_lines(basis_id, key[-1], delta))
4527
4689
yield versionedfile.FulltextContentFactory(
4528
key, parent_keys, None, as_bytes)
4690
key, parent_keys, None, delta_serialized)
4531
4693
def _iter_for_revno(repo, partial_history_cache, stop_index=None,