« back to all changes in this revision
Viewing changes to bzrlib/knit.py
- browse files at revision 3350.6.4
- compare with another revision
- compare with revision 1185.65.19
- download diff
- download tarball
- view history from revision 3350.6.4
- Committer: Robert Collins
- Date: 2008-06-11 04:20:16 UTC
- mto: (3221.12.10 Development1) (3517.4.1 annotate)
- mto: This revision was merged to the branch mainline in revision 3510.
- Revision ID: robertc@robertcollins.net-20080611042016-661dds0v1rjz542g
First cut at pluralised VersionedFiles. Some rather massive API incompatabilities, primarily because of the difficulty of coherence among competing stores.
- files removed:
- bzrlib/store/revision/__init__.py
- bzrlib/tests/interversionedfile_implementations
- bzrlib/tests/revisionstore_implementations
- files modified:
- bzrlib/annotate.py
added
removed
bzrlib/knit.py
65
65
from itertools import izip, chain
66
66
import operator
67
67
import os
68
import urllib
68
69
import sys
69
70
import warnings
70
71
from zlib import Z_DEFAULT_COMPRESSION
75
76
from bzrlib import (
76
77
annotate,
77
78
graph as _mod_graph,
79
index as _mod_index,
78
80
lru_cache,
79
81
pack,
80
82
trace,
115
117
from bzrlib.versionedfile import (
116
118
AbsentContentFactory,
117
119
adapter_registry,
120
ConstantMapper,
118
121
ContentFactory,
119
InterVersionedFile,
122
FulltextContentFactory,
120
123
VersionedFile,
124
VersionedFiles,
121
125
)
122
126
import bzrlib.weave
123
127
148
152
:param basis_vf: A versioned file to access basis texts of deltas from.
149
153
May be None for adapters that do not need to access basis texts.
150
154
"""
151
self._data = _KnitData(None)
155
self._data = KnitVersionedFiles(None, None)
152
156
self._annotate_factory = KnitAnnotateFactory()
153
157
self._plain_factory = KnitPlainFactory()
154
158
self._basis_vf = basis_vf
161
165
rec, contents = \
162
166
self._data._parse_record_unchecked(annotated_compressed_bytes)
163
167
content = self._annotate_factory.parse_fulltext(contents, rec[1])
164
size, bytes = self._data._record_to_data(rec[1], rec[3], content.text())
168
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
165
169
return bytes
166
170
167
171
174
178
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
175
179
plain=True)
176
180
contents = self._plain_factory.lower_line_delta(delta)
177
size, bytes = self._data._record_to_data(rec[1], rec[3], contents)
181
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
178
182
return bytes
179
183
180
184
184
188
def get_bytes(self, factory, annotated_compressed_bytes):
185
189
rec, contents = \
186
190
self._data._parse_record_unchecked(annotated_compressed_bytes)
187
content, delta = self._annotate_factory.parse_record(factory.key[0],
191
content, delta = self._annotate_factory.parse_record(factory.key[-1],
188
192
contents, factory._build_details, None)
189
193
return ''.join(content.text())
190
194
197
201
self._data._parse_record_unchecked(annotated_compressed_bytes)
198
202
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
199
203
plain=True)
200
compression_parent = factory.parents[0][0]
201
basis_lines = self._basis_vf.get_lines(compression_parent)
204
compression_parent = factory.parents[0]
205
basis_entry = self._basis_vf.get_record_stream(
206
[compression_parent], 'unordered', True).next()
207
if basis_entry.storage_kind == 'absent':
208
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
209
basis_lines = split_lines(basis_entry.get_bytes_as('fulltext'))
202
210
# Manually apply the delta because we have one annotated content and
203
211
# one plain.
204
212
basis_content = PlainKnitContent(basis_lines, compression_parent)
213
221
def get_bytes(self, factory, compressed_bytes):
214
222
rec, contents = \
215
223
self._data._parse_record_unchecked(compressed_bytes)
216
content, delta = self._plain_factory.parse_record(factory.key[0],
224
content, delta = self._plain_factory.parse_record(factory.key[-1],
217
225
contents, factory._build_details, None)
218
226
return ''.join(content.text())
219
227
225
233
rec, contents = \
226
234
self._data._parse_record_unchecked(compressed_bytes)
227
235
delta = self._plain_factory.parse_line_delta(contents, rec[1])
228
compression_parent = factory.parents[0][0]
229
basis_lines = self._basis_vf.get_lines(compression_parent)
236
compression_parent = factory.parents[0]
237
# XXX: string splitting overhead.
238
basis_entry = self._basis_vf.get_record_stream(
239
[compression_parent], 'unordered', True).next()
240
if basis_entry.storage_kind == 'absent':
241
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
242
basis_lines = split_lines(basis_entry.get_bytes_as('fulltext'))
230
243
basis_content = PlainKnitContent(basis_lines, compression_parent)
231
244
# Manually apply the delta because we have one annotated content and
232
245
# one plain.
241
254
:seealso ContentFactory:
242
255
"""
243
256
244
def __init__(self, version, parents, build_details, sha1, raw_record,
257
def __init__(self, key, parents, build_details, sha1, raw_record,
245
258
annotated, knit=None):
246
"""Create a KnitContentFactory for version.
259
"""Create a KnitContentFactory for key.
247
260
248
:param version: The version.
261
:param key: The key.
249
262
:param parents: The parents.
250
263
:param build_details: The build details as returned from
251
264
get_build_details.
255
268
"""
256
269
ContentFactory.__init__(self)
257
270
self.sha1 = sha1
258
self.key = (version,)
259
self.parents = tuple((parent,) for parent in parents)
271
self.key = key
272
self.parents = parents
260
273
if build_details[0] == 'line-delta':
261
274
kind = 'delta'
262
275
else:
562
575
for origin, text in lines)
563
576
return out
564
577
565
def annotate(self, knit, version_id):
566
content = knit._get_content(version_id)
567
return content.annotate()
578
def annotate(self, knit, key):
579
content = knit._get_content(key)
580
# adjust for the fact that serialised annotations are only key suffixes
581
# for this factory.
582
if type(key) == tuple:
583
prefix = key[:-1]
584
origins = content.annotate()
585
result = []
586
for origin, line in origins:
587
result.append((prefix + (origin,), line))
588
return result
589
else:
590
return content.annotate()
568
591
569
592
570
593
class KnitPlainFactory(_KnitFactory):
624
647
out.extend(lines)
625
648
return out
626
649
627
def annotate(self, knit, version_id):
650
def annotate(self, knit, key):
628
651
annotator = _KnitAnnotator(knit)
629
return annotator.annotate(version_id)
630
631
632
def make_empty_knit(transport, relpath):
633
"""Construct a empty knit at the specified location."""
634
k = make_file_knit(transport, relpath, 'w', KnitPlainFactory)
635
636
637
def make_file_knit(name, transport, file_mode=None, access_mode='w',
638
factory=None, delta=True, create=False, create_parent_dir=False,
639
delay_create=False, dir_mode=None, get_scope=None):
640
"""Factory to create a KnitVersionedFile for a .knit/.kndx file pair."""
641
if factory is None:
642
factory = KnitAnnotateFactory()
643
if get_scope is None:
644
get_scope = lambda:None
645
index = _KnitIndex(transport, name + INDEX_SUFFIX,
646
access_mode, create=create, file_mode=file_mode,
647
create_parent_dir=create_parent_dir, delay_create=delay_create,
648
dir_mode=dir_mode, get_scope=get_scope)
649
access = _KnitAccess(transport, name + DATA_SUFFIX, file_mode,
650
dir_mode, ((create and not len(index)) and delay_create),
651
create_parent_dir)
652
return KnitVersionedFile(name, transport, factory=factory,
653
create=create, delay_create=delay_create, index=index,
654
access_method=access)
655
656
657
def get_suffixes():
658
"""Return the suffixes used by file based knits."""
659
return [DATA_SUFFIX, INDEX_SUFFIX]
660
make_file_knit.get_suffixes = get_suffixes
661
662
663
class KnitVersionedFile(VersionedFile):
664
"""Weave-like structure with faster random access.
665
666
A knit stores a number of texts and a summary of the relationships
667
between them. Texts are identified by a string version-id. Texts
668
are normally stored and retrieved as a series of lines, but can
669
also be passed as single strings.
670
671
Lines are stored with the trailing newline (if any) included, to
672
avoid special cases for files with no final newline. Lines are
673
composed of 8-bit characters, not unicode. The combination of
674
these approaches should mean any 'binary' file can be safely
675
stored and retrieved.
676
"""
677
678
def __init__(self, relpath, transport, file_mode=None,
679
factory=None, delta=True, create=False, create_parent_dir=False,
680
delay_create=False, dir_mode=None, index=None, access_method=None):
681
"""Construct a knit at location specified by relpath.
682
683
:param create: If not True, only open an existing knit.
684
:param create_parent_dir: If True, create the parent directory if
685
creating the file fails. (This is used for stores with
686
hash-prefixes that may not exist yet)
687
:param delay_create: The calling code is aware that the knit won't
688
actually be created until the first data is stored.
689
:param index: An index to use for the knit.
652
return annotator.annotate(key)
653
654
655
656
def make_file_factory(annotated, mapper):
657
"""Create a factory for creating a file based KnitVersionedFiles.
658
659
:param annotated: knit annotations are wanted.
660
:param mapper: The mapper from keys to paths.
661
"""
662
def factory(transport):
663
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
664
access = _KnitKeyAccess(transport, mapper)
665
return KnitVersionedFiles(index, access, annotated=annotated)
666
return factory
667
668
669
def make_pack_factory(graph, delta, keylength):
670
"""Create a factory for creating a pack based VersionedFiles.
671
672
This is only functional enough to run interface tests, it doesn't try to
673
provide a full pack environment.
674
675
:param graph: Store a graph.
676
:param delta: Delta compress contents.
677
:param keylength: How long should keys be.
678
"""
679
def factory(transport):
680
parents = graph or delta
681
ref_length = 0
682
if graph:
683
ref_length += 1
684
if delta:
685
ref_length += 1
686
max_delta_chain = 200
687
else:
688
max_delta_chain = 0
689
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
690
key_elements=keylength)
691
stream = transport.open_write_stream('newpack')
692
writer = pack.ContainerWriter(stream.write)
693
writer.begin()
694
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
695
deltas=delta, add_callback=graph_index.add_nodes)
696
access = _DirectPackAccess({})
697
access.set_writer(writer, graph_index, (transport, 'newpack'))
698
result = KnitVersionedFiles(index, access,
699
max_delta_chain=max_delta_chain)
700
result.stream = stream
701
result.writer = writer
702
return result
703
return factory
704
705
706
def cleanup_pack_knit(versioned_files):
707
versioned_files.stream.close()
708
versioned_files.writer.end()
709
710
711
class KnitVersionedFiles(VersionedFiles):
712
"""Storage for many versioned files using knit compression.
713
714
Backend storage is managed by indices and data objects.
715
"""
716
717
def __init__(self, index, data_access, max_delta_chain=200,
718
annotated=False):
719
"""Create a KnitVersionedFiles with index and data_access.
720
721
:param index: The index for the knit data.
722
:param data_access: The access object to store and retrieve knit
723
records.
724
:param max_delta_chain: The maximum number of deltas to permit during
725
insertion. Set to 0 to prohibit the use of deltas.
726
:param annotated: Set to True to cause annotations to be calculated and
727
stored during insertion.
690
728
"""
691
super(KnitVersionedFile, self).__init__()
692
self.transport = transport
693
self.filename = relpath
694
self.factory = factory or KnitAnnotateFactory()
695
self.delta = delta
696
697
self._max_delta_chain = 200
698
699
if None in (access_method, index):
700
raise ValueError("No default access_method or index any more")
701
729
self._index = index
702
_access = access_method
703
if create and not len(self) and not delay_create:
704
_access.create()
705
self._data = _KnitData(_access)
706
707
def __repr__(self):
708
return '%s(%s)' % (self.__class__.__name__,
709
self.transport.abspath(self.filename))
710
711
def _check_should_delta(self, first_parents):
730
self._access = data_access
731
self._max_delta_chain = max_delta_chain
732
if annotated:
733
self._factory = KnitAnnotateFactory()
734
else:
735
self._factory = KnitPlainFactory()
736
737
def add_lines(self, key, parents, lines, parent_texts=None,
738
left_matching_blocks=None, nostore_sha=None, random_id=False,
739
check_content=True):
740
"""See VersionedFiles.add_lines()."""
741
self._index._check_write_ok()
742
self._check_add(key, lines, random_id, check_content)
743
if parents is None:
744
# For no-graph knits, have the public interface use None for
745
# parents.
746
parents = ()
747
return self._add(key, lines, parents,
748
parent_texts, left_matching_blocks, nostore_sha, random_id)
749
750
def _add(self, key, lines, parents, parent_texts,
751
left_matching_blocks, nostore_sha, random_id):
752
"""Add a set of lines on top of version specified by parents.
753
754
Any versions not present will be converted into ghosts.
755
"""
756
# first thing, if the content is something we don't need to store, find
757
# that out.
758
line_bytes = ''.join(lines)
759
digest = sha_string(line_bytes)
760
if nostore_sha == digest:
761
raise errors.ExistingContent
762
763
present_parents = []
764
if parent_texts is None:
765
parent_texts = {}
766
# Do a single query to ascertain parent presence.
767
present_parent_map = self.get_parent_map(parents)
768
for parent in parents:
769
if parent in present_parent_map:
770
present_parents.append(parent)
771
772
# Currently we can only compress against the left most present parent.
773
if (len(present_parents) == 0 or
774
present_parents[0] != parents[0]):
775
delta = False
776
else:
777
# To speed the extract of texts the delta chain is limited
778
# to a fixed number of deltas. This should minimize both
779
# I/O and the time spend applying deltas.
780
delta = self._check_should_delta(present_parents[0])
781
782
text_length = len(line_bytes)
783
options = []
784
if lines:
785
if lines[-1][-1] != '\n':
786
# copy the contents of lines.
787
lines = lines[:]
788
options.append('no-eol')
789
lines[-1] = lines[-1] + '\n'
790
line_bytes += '\n'
791
792
for element in key:
793
if type(element) != str:
794
raise TypeError("key contains non-strings: %r" % (key,))
795
# Knit hunks are still last-element only
796
version_id = key[-1]
797
content = self._factory.make(lines, version_id)
798
if 'no-eol' in options:
799
# Hint to the content object that its text() call should strip the
800
# EOL.
801
content._should_strip_eol = True
802
if delta or (self._factory.annotated and len(present_parents) > 0):
803
# Merge annotations from parent texts if needed.
804
delta_hunks = self._merge_annotations(content, present_parents,
805
parent_texts, delta, self._factory.annotated,
806
left_matching_blocks)
807
808
if delta:
809
options.append('line-delta')
810
store_lines = self._factory.lower_line_delta(delta_hunks)
811
size, bytes = self._record_to_data(key, digest,
812
store_lines)
813
else:
814
options.append('fulltext')
815
# isinstance is slower and we have no hierarchy.
816
if self._factory.__class__ == KnitPlainFactory:
817
# Use the already joined bytes saving iteration time in
818
# _record_to_data.
819
size, bytes = self._record_to_data(key, digest,
820
lines, [line_bytes])
821
else:
822
# get mixed annotation + content and feed it into the
823
# serialiser.
824
store_lines = self._factory.lower_fulltext(content)
825
size, bytes = self._record_to_data(key, digest,
826
store_lines)
827
828
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
829
self._index.add_records(
830
((key, options, access_memo, parents),),
831
random_id=random_id)
832
return digest, text_length, content
833
834
def annotate(self, key):
835
"""See VersionedFiles.annotate."""
836
return self._factory.annotate(self, key)
837
838
def check(self, progress_bar=None):
839
"""See VersionedFiles.check()."""
840
# This doesn't actually test extraction of everything, but that will
841
# impact 'bzr check' substantially, and needs to be integrated with
842
# care. However, it does check for the obvious problem of a delta with
843
# no basis.
844
keys = self.keys()
845
parent_map = self.get_parent_map(keys)
846
for key in keys:
847
if self._index.get_method(key) != 'fulltext':
848
compression_parent = parent_map[key][0]
849
if compression_parent not in parent_map:
850
raise errors.KnitCorrupt(self,
851
"Missing basis parent %s for %s" % (
852
compression_parent, key))
853
854
def _check_add(self, key, lines, random_id, check_content):
855
"""check that version_id and lines are safe to add."""
856
if contains_whitespace(key[-1]):
857
raise InvalidRevisionId(key[-1], self.filename)
858
self.check_not_reserved_id(key[-1])
859
# Technically this could be avoided if we are happy to allow duplicate
860
# id insertion when other things than bzr core insert texts, but it
861
# seems useful for folk using the knit api directly to have some safety
862
# blanket that we can disable.
863
##if not random_id and self.has_version(key):
864
## raise RevisionAlreadyPresent(key, self)
865
if check_content:
866
self._check_lines_not_unicode(lines)
867
self._check_lines_are_lines(lines)
868
869
def _check_header(self, key, line):
870
rec = self._split_header(line)
871
self._check_header_version(rec, key[-1])
872
return rec
873
874
def _check_header_version(self, rec, version_id):
875
"""Checks the header version on original format knit records.
876
877
These have the last component of the key embedded in the record.
878
"""
879
if rec[1] != version_id:
880
raise KnitCorrupt(self,
881
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
882
883
def _check_should_delta(self, parent):
712
884
"""Iterate back through the parent listing, looking for a fulltext.
713
885
714
886
This is used when we want to decide whether to add a delta or a new
721
893
"""
722
894
delta_size = 0
723
895
fulltext_size = None
724
delta_parents = first_parents
725
896
for count in xrange(self._max_delta_chain):
726
parent = delta_parents[0]
897
# XXX: Collapse these two queries:
727
898
method = self._index.get_method(parent)
728
899
index, pos, size = self._index.get_position(parent)
729
900
if method == 'fulltext':
730
901
fulltext_size = size
731
902
break
732
903
delta_size += size
733
delta_parents = self._index.get_parent_map([parent])[parent]
904
# No exception here because we stop at first fulltext anyway, an
905
# absent parent indicates a corrupt knit anyway.
906
# TODO: This should be asking for compression parent, not graph
907
# parent.
908
parent = self._index.get_parent_map([parent])[parent][0]
734
909
else:
735
910
# We couldn't find a fulltext, so we must create a new one
736
911
return False
737
912
# Simple heuristic - if the total I/O wold be greater as a delta than
913
# the originally installed fulltext, we create a new fulltext.
738
914
return fulltext_size > delta_size
739
915
740
def _check_write_ok(self):
741
return self._index._check_write_ok()
742
743
def _add_raw_records(self, records, data):
744
"""Add all the records 'records' with data pre-joined in 'data'.
745
746
:param records: A list of tuples(version_id, options, parents, size).
747
:param data: The data for the records. When it is written, the records
748
are adjusted to have pos pointing into data by the sum of
749
the preceding records sizes.
750
"""
751
# write all the data
752
raw_record_sizes = [record[3] for record in records]
753
positions = self._data.add_raw_records(raw_record_sizes, data)
754
index_entries = []
755
for (version_id, options, parents, _), access_memo in zip(
756
records, positions):
757
index_entries.append((version_id, options, access_memo, parents))
758
self._index.add_versions(index_entries)
759
760
def copy_to(self, name, transport):
761
"""See VersionedFile.copy_to()."""
762
# copy the current index to a temp index to avoid racing with local
763
# writes
764
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
765
self.transport.get(self._index._filename))
766
# copy the data file
767
f = self._data._open_file()
768
try:
769
transport.put_file(name + DATA_SUFFIX, f)
770
finally:
771
f.close()
772
# move the copied index into place
773
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
774
775
def get_data_stream(self, required_versions):
776
"""Get a data stream for the specified versions.
777
778
Versions may be returned in any order, not necessarily the order
779
specified. They are returned in a partial order by compression
780
parent, so that the deltas can be applied as the data stream is
781
inserted; however note that compression parents will not be sent
782
unless they were specifically requested, as the client may already
783
have them.
784
785
:param required_versions: The exact set of versions to be extracted.
786
Unlike some other knit methods, this is not used to generate a
787
transitive closure, rather it is used precisely as given.
788
789
:returns: format_signature, list of (version, options, length, parents),
790
reader_callable.
791
"""
792
required_version_set = frozenset(required_versions)
793
version_index = {}
794
# list of revisions that can just be sent without waiting for their
795
# compression parent
796
ready_to_send = []
797
# map from revision to the children based on it
798
deferred = {}
799
# first, read all relevant index data, enough to sort into the right
800
# order to return
801
for version_id in required_versions:
802
options = self._index.get_options(version_id)
803
parents = self._index.get_parents_with_ghosts(version_id)
804
index_memo = self._index.get_position(version_id)
805
version_index[version_id] = (index_memo, options, parents)
806
if ('line-delta' in options
807
and parents[0] in required_version_set):
808
# must wait until the parent has been sent
809
deferred.setdefault(parents[0], []). \
810
append(version_id)
811
else:
812
# either a fulltext, or a delta whose parent the client did
813
# not ask for and presumably already has
814
ready_to_send.append(version_id)
815
# build a list of results to return, plus instructions for data to
816
# read from the file
817
copy_queue_records = []
818
temp_version_list = []
819
while ready_to_send:
820
# XXX: pushing and popping lists may be a bit inefficient
821
version_id = ready_to_send.pop(0)
822
(index_memo, options, parents) = version_index[version_id]
823
copy_queue_records.append((version_id, index_memo))
824
none, data_pos, data_size = index_memo
825
temp_version_list.append((version_id, options, data_size,
826
parents))
827
if version_id in deferred:
828
# now we can send all the children of this revision - we could
829
# put them in anywhere, but we hope that sending them soon
830
# after the fulltext will give good locality in the receiver
831
ready_to_send[:0] = deferred.pop(version_id)
832
if not (len(deferred) == 0):
833
raise AssertionError("Still have compressed child versions waiting to be sent")
834
# XXX: The stream format is such that we cannot stream it - we have to
835
# know the length of all the data a-priori.
836
raw_datum = []
837
result_version_list = []
838
for (version_id, raw_data, _), \
839
(version_id2, options, _, parents) in \
840
izip(self._data.read_records_iter_raw(copy_queue_records),
841
temp_version_list):
842
if not (version_id == version_id2):
843
raise AssertionError('logic error, inconsistent results')
844
raw_datum.append(raw_data)
845
result_version_list.append(
846
(version_id, options, len(raw_data), parents))
847
# provide a callback to get data incrementally.
848
pseudo_file = StringIO(''.join(raw_datum))
849
def read(length):
850
if length is None:
851
return pseudo_file.read()
852
else:
853
return pseudo_file.read(length)
854
return (self.get_format_signature(), result_version_list, read)
855
856
def get_record_stream(self, versions, ordering, include_delta_closure):
857
"""Get a stream of records for versions.
858
859
:param versions: The versions to include. Each version is a tuple
860
(version,).
916
def _build_details_to_components(self, build_details):
917
"""Convert a build_details tuple to a position tuple."""
918
# record_details, access_memo, compression_parent
919
return build_details[3], build_details[0], build_details[1]
920
921
def _get_components_positions(self, keys, noraise=False):
922
"""Produce a map of position data for the components of keys.
923
924
This data is intended to be used for retrieving the knit records.
925
926
A dict of key to (record_details, index_memo, next, parents) is
927
returned.
928
method is the way referenced data should be applied.
929
index_memo is the handle to pass to the data access to actually get the
930
data
931
next is the build-parent of the version, or None for fulltexts.
932
parents is the version_ids of the parents of this version
933
934
:param noraise: If True do not raise an error on a missing component,
935
just ignore it.
936
"""
937
component_data = {}
938
pending_components = keys
939
while pending_components:
940
build_details = self._index.get_build_details(pending_components)
941
current_components = set(pending_components)
942
pending_components = set()
943
for key, details in build_details.iteritems():
944
(index_memo, compression_parent, parents,
945
record_details) = details
946
method = record_details[0]
947
if compression_parent is not None:
948
pending_components.add(compression_parent)
949
component_data[key] = self._build_details_to_components(details)
950
missing = current_components.difference(build_details)
951
if missing and not noraise:
952
raise errors.RevisionNotPresent(missing.pop(), self)
953
return component_data
954
955
def _get_content(self, key, parent_texts={}):
956
"""Returns a content object that makes up the specified
957
version."""
958
cached_version = parent_texts.get(key, None)
959
if cached_version is not None:
960
# Ensure the cache dict is valid.
961
if not self.get_parent_map([key]):
962
raise RevisionNotPresent(key, self)
963
return cached_version
964
text_map, contents_map = self._get_content_maps([key])
965
return contents_map[key]
966
967
def _get_content_maps(self, keys):
968
"""Produce maps of text and KnitContents
969
970
:return: (text_map, content_map) where text_map contains the texts for
971
the requested versions and content_map contains the KnitContents.
972
"""
973
# FUTURE: This function could be improved for the 'extract many' case
974
# by tracking each component and only doing the copy when the number of
975
# children than need to apply delta's to it is > 1 or it is part of the
976
# final output.
977
keys = list(keys)
978
multiple_versions = len(keys) != 1
979
record_map = self._get_record_map(keys)
980
981
text_map = {}
982
content_map = {}
983
final_content = {}
984
for key in keys:
985
components = []
986
cursor = key
987
while cursor is not None:
988
record, record_details, digest, next = record_map[cursor]
989
components.append((cursor, record, record_details, digest))
990
if cursor in content_map:
991
break
992
cursor = next
993
994
content = None
995
for (component_id, record, record_details,
996
digest) in reversed(components):
997
if component_id in content_map:
998
content = content_map[component_id]
999
else:
1000
content, delta = self._factory.parse_record(key[-1],
1001
record, record_details, content,
1002
copy_base_content=multiple_versions)
1003
if multiple_versions:
1004
content_map[component_id] = content
1005
1006
content.cleanup_eol(copy_on_mutate=multiple_versions)
1007
final_content[key] = content
1008
1009
# digest here is the digest from the last applied component.
1010
text = content.text()
1011
actual_sha = sha_strings(text)
1012
if actual_sha != digest:
1013
raise KnitCorrupt(self,
1014
'\n sha-1 %s'
1015
'\n of reconstructed text does not match'
1016
'\n expected %s'
1017
'\n for version %s' %
1018
(actual_sha, digest, key))
1019
text_map[key] = text
1020
return text_map, final_content
1021
1022
def get_parent_map(self, keys):
1023
"""Get a map of the parents of keys.
1024
1025
:param keys: The keys to look up parents for.
1026
:return: A mapping from keys to parents. Absent keys are absent from
1027
the mapping.
1028
"""
1029
return self._index.get_parent_map(keys)
1030
1031
def _get_record_map(self, keys):
1032
"""Produce a dictionary of knit records.
1033
1034
:return: {key:(record, record_details, digest, next)}
1035
record
1036
data returned from read_records
1037
record_details
1038
opaque information to pass to parse_record
1039
digest
1040
SHA1 digest of the full text after all steps are done
1041
next
1042
build-parent of the version, i.e. the leftmost ancestor.
1043
Will be None if the record is not a delta.
1044
"""
1045
position_map = self._get_components_positions(keys)
1046
# key = component_id, r = record_details, i_m = index_memo, n = next
1047
records = [(key, i_m) for key, (r, i_m, n)
1048
in position_map.iteritems()]
1049
record_map = {}
1050
for key, record, digest in \
1051
self._read_records_iter(records):
1052
(record_details, index_memo, next) = position_map[key]
1053
record_map[key] = record, record_details, digest, next
1054
return record_map
1055
1056
def get_record_stream(self, keys, ordering, include_delta_closure):
1057
"""Get a stream of records for keys.
1058
1059
:param keys: The keys to include.
861
1060
:param ordering: Either 'unordered' or 'topological'. A topologically
862
1061
sorted stream has compression parents strictly before their
863
1062
children.
866
1065
:return: An iterator of ContentFactory objects, each of which is only
867
1066
valid until the iterator is advanced.
868
1067
"""
1068
# keys might be a generator
1069
keys = set(keys)
1070
if not self._index.has_graph:
1071
# Cannot topological order when no graph has been stored.
1072
ordering = 'unordered'
869
1073
if include_delta_closure:
870
# Nb: what we should do is plan the data to stream to allow
871
# reconstruction of all the texts without excessive buffering,
872
# including re-sending common bases as needed. This makes the most
873
# sense when we start serialising these streams though, so for now
874
# we just fallback to individual text construction behind the
875
# abstraction barrier.
876
knit = self
1074
positions = self._get_components_positions(keys, noraise=True)
877
1075
else:
878
knit = None
879
# We end up doing multiple index lookups here for parents details and
880
# disk layout details - we need a unified api ?
881
parent_map = self.get_parent_map(versions)
882
absent_versions = set(versions) - set(parent_map)
1076
build_details = self._index.get_build_details(keys)
1077
positions = dict((key, self._build_details_to_components(details))
1078
for key, details in build_details.iteritems())
1079
absent_keys = keys.difference(set(positions))
1080
# There may be more absent keys : if we're missing the basis component
1081
# and are trying to include the delta closure.
1082
if include_delta_closure:
1083
# key:True means key can be reconstructed
1084
checked_keys = {}
1085
for key in keys:
1086
# the delta chain
1087
try:
1088
chain = [key, positions[key][2]]
1089
except KeyError:
1090
absent_keys.add(key)
1091
continue
1092
result = True
1093
while chain[-1] is not None:
1094
if chain[-1] in checked_keys:
1095
result = checked_keys[chain[-1]]
1096
break
1097
else:
1098
try:
1099
chain.append(positions[chain[-1]][2])
1100
except KeyError:
1101
# missing basis component
1102
result = False
1103
break
1104
for chain_key in chain[:-1]:
1105
checked_keys[chain_key] = result
1106
if not result:
1107
absent_keys.add(key)
1108
for key in absent_keys:
1109
yield AbsentContentFactory(key)
1110
# restrict our view to the keys we can answer.
1111
keys = keys - absent_keys
1112
# Double index lookups here : need a unified api ?
1113
parent_map = self.get_parent_map(keys)
883
1114
if ordering == 'topological':
884
present_versions = topo_sort(parent_map)
885
else:
886
# List comprehension to keep the requested order (as that seems
887
# marginally useful, at least until we start doing IO optimising
888
# here.
889
present_versions = [version for version in versions if version in
890
parent_map]
891
position_map = self._get_components_positions(present_versions)
892
records = [(version, position_map[version][1]) for version in
893
present_versions]
894
record_map = {}
895
for version in absent_versions:
896
yield AbsentContentFactory((version,))
897
for version, raw_data, sha1 in \
898
self._data.read_records_iter_raw(records):
899
(record_details, index_memo, _) = position_map[version]
900
yield KnitContentFactory(version, parent_map[version],
901
record_details, sha1, raw_data, self.factory.annotated, knit)
902
903
def _extract_blocks(self, version_id, source, target):
904
if self._index.get_method(version_id) != 'line-delta':
905
return None
906
parent, sha1, noeol, delta = self.get_delta(version_id)
907
return KnitContent.get_line_delta_blocks(delta, source, target)
908
909
def get_delta(self, version_id):
910
"""Get a delta for constructing version from some other version."""
911
self.check_not_reserved_id(version_id)
912
parents = self.get_parent_map([version_id])[version_id]
913
if len(parents):
914
parent = parents[0]
915
else:
916
parent = None
917
index_memo = self._index.get_position(version_id)
918
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
919
noeol = 'no-eol' in self._index.get_options(version_id)
920
if 'fulltext' == self._index.get_method(version_id):
921
new_content = self.factory.parse_fulltext(data, version_id)
922
if parent is not None:
923
reference_content = self._get_content(parent)
924
old_texts = reference_content.text()
925
else:
926
old_texts = []
927
new_texts = new_content.text()
928
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
929
new_texts)
930
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
931
else:
932
delta = self.factory.parse_line_delta(data, version_id)
933
return parent, sha1, noeol, delta
934
935
def get_format_signature(self):
936
"""See VersionedFile.get_format_signature()."""
937
if self.factory.annotated:
938
annotated_part = "annotated"
939
else:
940
annotated_part = "plain"
941
return "knit-%s" % (annotated_part,)
942
943
def get_sha1s(self, version_ids):
944
"""See VersionedFile.get_sha1s()."""
945
record_map = self._get_record_map(version_ids)
1115
present_keys = topo_sort(parent_map)
1116
else:
1117
present_keys = keys
1118
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1119
# XXX: At that point we need to consider double reads by utilising
1120
# components multiple times.
1121
if include_delta_closure:
1122
# XXX: get_content_maps performs its own index queries; allow state
1123
# to be passed in.
1124
text_map, _ = self._get_content_maps(present_keys)
1125
for key in present_keys:
1126
yield FulltextContentFactory(key, parent_map[key], None,
1127
''.join(text_map[key]))
1128
else:
1129
records = [(key, positions[key][1]) for key in present_keys]
1130
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1131
(record_details, index_memo, _) = positions[key]
1132
yield KnitContentFactory(key, parent_map[key],
1133
record_details, sha1, raw_data, self._factory.annotated, None)
1134
1135
def get_sha1s(self, keys):
1136
"""See VersionedFiles.get_sha1s()."""
1137
record_map = self._get_record_map(keys)
946
1138
# record entry 2 is the 'digest'.
947
return [record_map[v][2] for v in version_ids]
948
949
def insert_data_stream(self, (format, data_list, reader_callable)):
950
"""Insert knit records from a data stream into this knit.
951
952
If a version in the stream is already present in this knit, it will not
953
be inserted a second time. It will be checked for consistency with the
954
stored version however, and may cause a KnitCorrupt error to be raised
955
if the data in the stream disagrees with the already stored data.
956
957
:seealso: get_data_stream
958
"""
959
if format != self.get_format_signature():
960
if 'knit' in debug.debug_flags:
961
trace.mutter(
962
'incompatible format signature inserting to %r', self)
963
source = self._knit_from_datastream(
964
(format, data_list, reader_callable))
965
stream = source.get_record_stream(source.versions(), 'unordered', False)
966
self.insert_record_stream(stream)
967
return
968
969
for version_id, options, length, parents in data_list:
970
if self.has_version(version_id):
971
# First check: the list of parents.
972
my_parents = self.get_parents_with_ghosts(version_id)
973
if tuple(my_parents) != tuple(parents):
974
# XXX: KnitCorrupt is not quite the right exception here.
975
raise KnitCorrupt(
976
self.filename,
977
'parents list %r from data stream does not match '
978
'already recorded parents %r for %s'
979
% (parents, my_parents, version_id))
980
981
# Also check the SHA-1 of the fulltext this content will
982
# produce.
983
raw_data = reader_callable(length)
984
my_fulltext_sha1 = self.get_sha1s([version_id])[0]
985
df, rec = self._data._parse_record_header(version_id, raw_data)
986
stream_fulltext_sha1 = rec[3]
987
if my_fulltext_sha1 != stream_fulltext_sha1:
988
# Actually, we don't know if it's this knit that's corrupt,
989
# or the data stream we're trying to insert.
990
raise KnitCorrupt(
991
self.filename, 'sha-1 does not match %s' % version_id)
992
else:
993
if 'line-delta' in options:
994
# Make sure that this knit record is actually useful: a
995
# line-delta is no use unless we have its parent.
996
# Fetching from a broken repository with this problem
997
# shouldn't break the target repository.
998
#
999
# See https://bugs.launchpad.net/bzr/+bug/164443
1000
if not self._index.has_version(parents[0]):
1001
raise KnitCorrupt(
1002
self.filename,
1003
'line-delta from stream '
1004
'for version %s '
1005
'references '
1006
'missing parent %s\n'
1007
'Try running "bzr check" '
1008
'on the source repository, and "bzr reconcile" '
1009
'if necessary.' %
1010
(version_id, parents[0]))
1011
if not self.delta:
1012
# We received a line-delta record for a non-delta knit.
1013
# Convert it to a fulltext.
1014
gzip_bytes = reader_callable(length)
1015
self._convert_line_delta_to_fulltext(
1016
gzip_bytes, version_id, parents)
1017
continue
1018
1019
self._add_raw_records(
1020
[(version_id, options, parents, length)],
1021
reader_callable(length))
1022
1023
def _convert_line_delta_to_fulltext(self, gzip_bytes, version_id, parents):
1024
lines, sha1 = self._data._parse_record(version_id, gzip_bytes)
1025
delta = self.factory.parse_line_delta(lines, version_id)
1026
content = self.factory.make(self.get_lines(parents[0]), parents[0])
1027
content.apply_delta(delta, version_id)
1028
digest, len, content = self.add_lines(
1029
version_id, parents, content.text())
1030
if digest != sha1:
1031
raise errors.VersionedFileInvalidChecksum(version_id)
1032
1033
def _knit_from_datastream(self, (format, data_list, reader_callable)):
1034
"""Create a knit object from a data stream.
1035
1036
This method exists to allow conversion of data streams that do not
1037
match the signature of this knit. Generally it will be slower and use
1038
more memory to use this method to insert data, but it will work.
1039
1040
:seealso: get_data_stream for details on datastreams.
1041
:return: A knit versioned file which can be used to join the datastream
1042
into self.
1043
"""
1044
if format == "knit-plain":
1045
factory = KnitPlainFactory()
1046
elif format == "knit-annotated":
1047
factory = KnitAnnotateFactory()
1048
else:
1049
raise errors.KnitDataStreamUnknown(format)
1050
index = _StreamIndex(data_list, self._index)
1051
access = _StreamAccess(reader_callable, index, self, factory)
1052
return KnitVersionedFile(self.filename, self.transport,
1053
factory=factory, index=index, access_method=access)
1139
return [record_map[key][2] for key in keys]
1054
1140
1055
1141
def insert_record_stream(self, stream):
1056
"""Insert a record stream into this versioned file.
1142
"""Insert a record stream into this container.
1057
1143
1058
1144
:param stream: A stream of records to insert.
1059
1145
:return: None
1060
:seealso VersionedFile.get_record_stream:
1146
:seealso VersionedFiles.get_record_stream:
1061
1147
"""
1062
1148
def get_adapter(adapter_key):
1063
1149
try:
1067
1153
adapter = adapter_factory(self)
1068
1154
adapters[adapter_key] = adapter
1069
1155
return adapter
1070
if self.factory.annotated:
1156
if self._factory.annotated:
1071
1157
# self is annotated, we need annotated knits to use directly.
1072
1158
annotated = "annotated-"
1073
1159
convertibles = []
1074
1160
else:
1075
1161
# self is not annotated, but we can strip annotations cheaply.
1076
1162
annotated = ""
1077
convertibles = set(["knit-annotated-delta-gz",
1078
"knit-annotated-ft-gz"])
1163
convertibles = set(["knit-annotated-ft-gz"])
1164
if self._max_delta_chain:
1165
convertibles.add("knit-annotated-delta-gz")
1079
1166
# The set of types we can cheaply adapt without needing basis texts.
1080
1167
native_types = set()
1081
native_types.add("knit-%sdelta-gz" % annotated)
1168
if self._max_delta_chain:
1169
native_types.add("knit-%sdelta-gz" % annotated)
1082
1170
native_types.add("knit-%sft-gz" % annotated)
1083
1171
knit_types = native_types.union(convertibles)
1084
1172
adapters = {}
1091
1179
# key = basis_parent, value = index entry to add
1092
1180
buffered_index_entries = {}
1093
1181
for record in stream:
1182
parents = record.parents
1094
1183
# Raise an error when a record is missing.
1095
1184
if record.storage_kind == 'absent':
1096
raise RevisionNotPresent([record.key[0]], self)
1097
# adapt to non-tuple interface
1098
parents = [parent[0] for parent in record.parents]
1185
raise RevisionNotPresent([record.key], self)
1099
1186
if record.storage_kind in knit_types:
1100
1187
if record.storage_kind not in native_types:
1101
1188
try:
1119
1206
# deprecated format this is tolerable. It can be fixed if
1120
1207
# needed by in the kndx index support raising on a duplicate
1121
1208
# add with identical parents and options.
1122
access_memo = self._data.add_raw_records([len(bytes)], bytes)[0]
1123
index_entry = (record.key[0], options, access_memo, parents)
1209
access_memo = self._access.add_raw_records(
1210
[(record.key, len(bytes))], bytes)[0]
1211
index_entry = (record.key, options, access_memo, parents)
1124
1212
buffered = False
1125
1213
if 'fulltext' not in options:
1126
1214
basis_parent = parents[0]
1127
if not self.has_version(basis_parent):
1215
# Note that pack backed knits don't need to buffer here
1216
# because they buffer all writes to the transaction level,
1217
# but we don't expose that differnet at the index level. If
1218
# the query here has sufficient cost to show up in
1219
# profiling we should do that.
1220
if basis_parent not in self.get_parent_map([basis_parent]):
1128
1221
pending = buffered_index_entries.setdefault(
1129
1222
basis_parent, [])
1130
1223
pending.append(index_entry)
1131
1224
buffered = True
1132
1225
if not buffered:
1133
self._index.add_versions([index_entry])
1226
self._index.add_records([index_entry])
1134
1227
elif record.storage_kind == 'fulltext':
1135
self.add_lines(record.key[0], parents,
1228
self.add_lines(record.key, parents,
1136
1229
split_lines(record.get_bytes_as('fulltext')))
1137
1230
else:
1138
1231
adapter_key = record.storage_kind, 'fulltext'
1140
1233
lines = split_lines(adapter.get_bytes(
1141
1234
record, record.get_bytes_as(record.storage_kind)))
1142
1235
try:
1143
self.add_lines(record.key[0], parents, lines)
1236
self.add_lines(record.key, parents, lines)
1144
1237
except errors.RevisionAlreadyPresent:
1145
1238
pass
1146
1239
# Add any records whose basis parent is now available.
1147
added_keys = [record.key[0]]
1240
added_keys = [record.key]
1148
1241
while added_keys:
1149
1242
key = added_keys.pop(0)
1150
1243
if key in buffered_index_entries:
1151
1244
index_entries = buffered_index_entries[key]
1152
self._index.add_versions(index_entries)
1245
self._index.add_records(index_entries)
1153
1246
added_keys.extend(
1154
1247
[index_entry[0] for index_entry in index_entries])
1155
1248
del buffered_index_entries[key]
1158
1251
raise errors.RevisionNotPresent(buffered_index_entries.keys()[0],
1159
1252
self)
1160
1253
1161
def versions(self):
1162
"""See VersionedFile.versions."""
1163
if 'evil' in debug.debug_flags:
1164
trace.mutter_callsite(2, "versions scales with size of history")
1165
return self._index.get_versions()
1166
1167
def has_version(self, version_id):
1168
"""See VersionedFile.has_version."""
1169
if 'evil' in debug.debug_flags:
1170
trace.mutter_callsite(2, "has_version is a LBYL scenario")
1171
return self._index.has_version(version_id)
1172
1173
__contains__ = has_version
1254
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1255
"""Iterate over the lines in the versioned files from keys.
1256
1257
This may return lines from other keys. Each item the returned
1258
iterator yields is a tuple of a line and a text version that that line
1259
is present in (not introduced in).
1260
1261
Ordering of results is in whatever order is most suitable for the
1262
underlying storage format.
1263
1264
If a progress bar is supplied, it may be used to indicate progress.
1265
The caller is responsible for cleaning up progress bars (because this
1266
is an iterator).
1267
1268
NOTES:
1269
* Lines are normalised by the underlying store: they will all have \n
1270
terminators.
1271
* Lines are returned in arbitrary order.
1272
1273
:return: An iterator over (line, key).
1274
"""
1275
if pb is None:
1276
pb = progress.DummyProgress()
1277
keys = set(keys)
1278
# filter for available keys
1279
parent_map = self.get_parent_map(keys)
1280
if len(parent_map) != len(keys):
1281
missing = set(parent_map) - requested_keys
1282
raise RevisionNotPresent(key, self.filename)
1283
# we don't care about inclusions, the caller cares.
1284
# but we need to setup a list of records to visit.
1285
# we need key, position, length
1286
key_records = []
1287
build_details = self._index.get_build_details(keys)
1288
for key in keys:
1289
key_records.append((key, build_details[key][0]))
1290
total = len(key_records)
1291
for key_idx, (key, data, sha_value) in \
1292
enumerate(self._read_records_iter(key_records)):
1293
pb.update('Walking content.', key_idx, total)
1294
compression_parent = build_details[key][1]
1295
if compression_parent is None:
1296
# fulltext
1297
line_iterator = self._factory.get_fulltext_content(data)
1298
else:
1299
# Delta
1300
line_iterator = self._factory.get_linedelta_content(data)
1301
# XXX: It might be more efficient to yield (key,
1302
# line_iterator) in the future. However for now, this is a simpler
1303
# change to integrate into the rest of the codebase. RBC 20071110
1304
for line in line_iterator:
1305
yield line, key
1306
pb.update('Walking content.', total, total)
1307
1308
def _make_line_delta(self, delta_seq, new_content):
1309
"""Generate a line delta from delta_seq and new_content."""
1310
diff_hunks = []
1311
for op in delta_seq.get_opcodes():
1312
if op[0] == 'equal':
1313
continue
1314
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1315
return diff_hunks
1174
1316
1175
1317
def _merge_annotations(self, content, parents, parent_texts={},
1176
1318
delta=None, annotated=None,
1177
1319
left_matching_blocks=None):
1178
"""Merge annotations for content. This is done by comparing
1179
the annotations based on changed to the text.
1320
"""Merge annotations for content and generate deltas.
1321
1322
This is done by comparing the annotations based on changes to the text
1323
and generating a delta on the resulting full texts. If annotations are
1324
not being created then a simple delta is created.
1180
1325
"""
1181
1326
if left_matching_blocks is not None:
1182
1327
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1183
1328
else:
1184
1329
delta_seq = None
1185
1330
if annotated:
1186
for parent_id in parents:
1187
merge_content = self._get_content(parent_id, parent_texts)
1188
if (parent_id == parents[0] and delta_seq is not None):
1331
for parent_key in parents:
1332
merge_content = self._get_content(parent_key, parent_texts)
1333
if (parent_key == parents[0] and delta_seq is not None):
1189
1334
seq = delta_seq
1190
1335
else:
1191
1336
seq = patiencediff.PatienceSequenceMatcher(
1197
1342
# new content for any line that matches the last-checked
1198
1343
# parent.
1199
1344
content._lines[j:j+n] = merge_content._lines[i:i+n]
1345
if content._lines and content._lines[-1][1][-1] != '\n':
1346
# The copied annotation was from a line without a trailing EOL,
1347
# reinstate one for the content object, to ensure correct
1348
# serialization.
1349
line = content._lines[-1][1] + '\n'
1350
content._lines[-1] = (content._lines[-1][0], line)
1200
1351
if delta:
1201
1352
if delta_seq is None:
1202
1353
reference_content = self._get_content(parents[0], parent_texts)
1206
1357
None, old_texts, new_texts)
1207
1358
return self._make_line_delta(delta_seq, content)
1208
1359
1209
def _make_line_delta(self, delta_seq, new_content):
1210
"""Generate a line delta from delta_seq and new_content."""
1211
diff_hunks = []
1212
for op in delta_seq.get_opcodes():
1213
if op[0] == 'equal':
1214
continue
1215
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1216
return diff_hunks
1217
1218
def _get_components_positions(self, version_ids):
1219
"""Produce a map of position data for the components of versions.
1220
1221
This data is intended to be used for retrieving the knit records.
1222
1223
A dict of version_id to (record_details, index_memo, next, parents) is
1224
returned.
1225
method is the way referenced data should be applied.
1226
index_memo is the handle to pass to the data access to actually get the
1227
data
1228
next is the build-parent of the version, or None for fulltexts.
1229
parents is the version_ids of the parents of this version
1230
"""
1231
component_data = {}
1232
pending_components = version_ids
1233
while pending_components:
1234
build_details = self._index.get_build_details(pending_components)
1235
current_components = set(pending_components)
1236
pending_components = set()
1237
for version_id, details in build_details.iteritems():
1238
(index_memo, compression_parent, parents,
1239
record_details) = details
1240
method = record_details[0]
1241
if compression_parent is not None:
1242
pending_components.add(compression_parent)
1243
component_data[version_id] = (record_details, index_memo,
1244
compression_parent)
1245
missing = current_components.difference(build_details)
1246
if missing:
1247
raise errors.RevisionNotPresent(missing.pop(), self.filename)
1248
return component_data
1249
1250
def _get_content(self, version_id, parent_texts={}):
1251
"""Returns a content object that makes up the specified
1252
version."""
1253
cached_version = parent_texts.get(version_id, None)
1254
if cached_version is not None:
1255
if not self.has_version(version_id):
1256
raise RevisionNotPresent(version_id, self.filename)
1257
return cached_version
1258
1259
text_map, contents_map = self._get_content_maps([version_id])
1260
return contents_map[version_id]
1261
1262
def _check_versions_present(self, version_ids):
1263
"""Check that all specified versions are present."""
1264
self._index.check_versions_present(version_ids)
1265
1266
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
1267
nostore_sha, random_id, check_content, left_matching_blocks):
1268
"""See VersionedFile.add_lines_with_ghosts()."""
1269
self._check_add(version_id, lines, random_id, check_content)
1270
return self._add(version_id, lines, parents, self.delta,
1271
parent_texts, left_matching_blocks, nostore_sha, random_id)
1272
1273
def _add_lines(self, version_id, parents, lines, parent_texts,
1274
left_matching_blocks, nostore_sha, random_id, check_content):
1275
"""See VersionedFile.add_lines."""
1276
self._check_add(version_id, lines, random_id, check_content)
1277
self._check_versions_present(parents)
1278
return self._add(version_id, lines[:], parents, self.delta,
1279
parent_texts, left_matching_blocks, nostore_sha, random_id)
1280
1281
def _check_add(self, version_id, lines, random_id, check_content):
1282
"""check that version_id and lines are safe to add."""
1283
if contains_whitespace(version_id):
1284
raise InvalidRevisionId(version_id, self.filename)
1285
self.check_not_reserved_id(version_id)
1286
# Technically this could be avoided if we are happy to allow duplicate
1287
# id insertion when other things than bzr core insert texts, but it
1288
# seems useful for folk using the knit api directly to have some safety
1289
# blanket that we can disable.
1290
if not random_id and self.has_version(version_id):
1291
raise RevisionAlreadyPresent(version_id, self.filename)
1292
if check_content:
1293
self._check_lines_not_unicode(lines)
1294
self._check_lines_are_lines(lines)
1295
1296
def _add(self, version_id, lines, parents, delta, parent_texts,
1297
left_matching_blocks, nostore_sha, random_id):
1298
"""Add a set of lines on top of version specified by parents.
1299
1300
If delta is true, compress the text as a line-delta against
1301
the first parent.
1302
1303
Any versions not present will be converted into ghosts.
1304
"""
1305
# first thing, if the content is something we don't need to store, find
1306
# that out.
1307
line_bytes = ''.join(lines)
1308
digest = sha_string(line_bytes)
1309
if nostore_sha == digest:
1310
raise errors.ExistingContent
1311
1312
present_parents = []
1313
if parent_texts is None:
1314
parent_texts = {}
1315
for parent in parents:
1316
if self.has_version(parent):
1317
present_parents.append(parent)
1318
1319
# can only compress against the left most present parent.
1320
if (delta and
1321
(len(present_parents) == 0 or
1322
present_parents[0] != parents[0])):
1323
delta = False
1324
1325
text_length = len(line_bytes)
1326
options = []
1327
if lines:
1328
if lines[-1][-1] != '\n':
1329
# copy the contents of lines.
1330
lines = lines[:]
1331
options.append('no-eol')
1332
lines[-1] = lines[-1] + '\n'
1333
line_bytes += '\n'
1334
1335
if delta:
1336
# To speed the extract of texts the delta chain is limited
1337
# to a fixed number of deltas. This should minimize both
1338
# I/O and the time spend applying deltas.
1339
delta = self._check_should_delta(present_parents)
1340
1341
content = self.factory.make(lines, version_id)
1342
if delta or (self.factory.annotated and len(present_parents) > 0):
1343
# Merge annotations from parent texts if needed.
1344
delta_hunks = self._merge_annotations(content, present_parents,
1345
parent_texts, delta, self.factory.annotated,
1346
left_matching_blocks)
1347
1348
if delta:
1349
options.append('line-delta')
1350
store_lines = self.factory.lower_line_delta(delta_hunks)
1351
size, bytes = self._data._record_to_data(version_id, digest,
1352
store_lines)
1353
else:
1354
options.append('fulltext')
1355
# isinstance is slower and we have no hierarchy.
1356
if self.factory.__class__ == KnitPlainFactory:
1357
# Use the already joined bytes saving iteration time in
1358
# _record_to_data.
1359
size, bytes = self._data._record_to_data(version_id, digest,
1360
lines, [line_bytes])
1361
else:
1362
# get mixed annotation + content and feed it into the
1363
# serialiser.
1364
store_lines = self.factory.lower_fulltext(content)
1365
size, bytes = self._data._record_to_data(version_id, digest,
1366
store_lines)
1367
1368
access_memo = self._data.add_raw_records([size], bytes)[0]
1369
self._index.add_versions(
1370
((version_id, options, access_memo, parents),),
1371
random_id=random_id)
1372
return digest, text_length, content
1373
1374
def check(self, progress_bar=None):
1375
"""See VersionedFile.check()."""
1376
# This doesn't actually test extraction of everything, but that will
1377
# impact 'bzr check' substantially, and needs to be integrated with
1378
# care. However, it does check for the obvious problem of a delta with
1379
# no basis.
1380
versions = self.versions()
1381
parent_map = self.get_parent_map(versions)
1382
for version in versions:
1383
if self._index.get_method(version) != 'fulltext':
1384
compression_parent = parent_map[version][0]
1385
if compression_parent not in parent_map:
1386
raise errors.KnitCorrupt(self,
1387
"Missing basis parent %s for %s" % (
1388
compression_parent, version))
1389
1390
def get_lines(self, version_id):
1391
"""See VersionedFile.get_lines()."""
1392
return self.get_line_list([version_id])[0]
1393
1394
def _get_record_map(self, version_ids):
1395
"""Produce a dictionary of knit records.
1396
1397
:return: {version_id:(record, record_details, digest, next)}
1398
record
1399
data returned from read_records
1400
record_details
1401
opaque information to pass to parse_record
1402
digest
1403
SHA1 digest of the full text after all steps are done
1404
next
1405
build-parent of the version, i.e. the leftmost ancestor.
1406
Will be None if the record is not a delta.
1407
"""
1408
position_map = self._get_components_positions(version_ids)
1409
# c = component_id, r = record_details, i_m = index_memo, n = next
1410
records = [(c, i_m) for c, (r, i_m, n)
1411
in position_map.iteritems()]
1412
record_map = {}
1413
for component_id, record, digest in \
1414
self._data.read_records_iter(records):
1415
(record_details, index_memo, next) = position_map[component_id]
1416
record_map[component_id] = record, record_details, digest, next
1417
1418
return record_map
1419
1420
def get_text(self, version_id):
1421
"""See VersionedFile.get_text"""
1422
return self.get_texts([version_id])[0]
1423
1424
def get_texts(self, version_ids):
1425
return [''.join(l) for l in self.get_line_list(version_ids)]
1426
1427
def get_line_list(self, version_ids):
1428
"""Return the texts of listed versions as a list of strings."""
1429
for version_id in version_ids:
1430
self.check_not_reserved_id(version_id)
1431
text_map, content_map = self._get_content_maps(version_ids)
1432
return [text_map[v] for v in version_ids]
1433
1434
_get_lf_split_line_list = get_line_list
1435
1436
def _get_content_maps(self, version_ids):
1437
"""Produce maps of text and KnitContents
1438
1439
:return: (text_map, content_map) where text_map contains the texts for
1440
the requested versions and content_map contains the KnitContents.
1441
Both dicts take version_ids as their keys.
1442
"""
1443
# FUTURE: This function could be improved for the 'extract many' case
1444
# by tracking each component and only doing the copy when the number of
1445
# children than need to apply delta's to it is > 1 or it is part of the
1446
# final output.
1447
version_ids = list(version_ids)
1448
multiple_versions = len(version_ids) != 1
1449
record_map = self._get_record_map(version_ids)
1450
1451
text_map = {}
1452
content_map = {}
1453
final_content = {}
1454
for version_id in version_ids:
1455
components = []
1456
cursor = version_id
1457
while cursor is not None:
1458
record, record_details, digest, next = record_map[cursor]
1459
components.append((cursor, record, record_details, digest))
1460
if cursor in content_map:
1461
break
1462
cursor = next
1463
1464
content = None
1465
for (component_id, record, record_details,
1466
digest) in reversed(components):
1467
if component_id in content_map:
1468
content = content_map[component_id]
1469
else:
1470
content, delta = self.factory.parse_record(version_id,
1471
record, record_details, content,
1472
copy_base_content=multiple_versions)
1473
if multiple_versions:
1474
content_map[component_id] = content
1475
1476
content.cleanup_eol(copy_on_mutate=multiple_versions)
1477
final_content[version_id] = content
1478
1479
# digest here is the digest from the last applied component.
1480
text = content.text()
1481
actual_sha = sha_strings(text)
1482
if actual_sha != digest:
1483
raise KnitCorrupt(self.filename,
1484
'\n sha-1 %s'
1485
'\n of reconstructed text does not match'
1486
'\n expected %s'
1487
'\n for version %s' %
1488
(actual_sha, digest, version_id))
1489
text_map[version_id] = text
1490
return text_map, final_content
1491
1492
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1493
pb=None):
1494
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1495
if version_ids is None:
1496
version_ids = self.versions()
1497
if pb is None:
1498
pb = progress.DummyProgress()
1499
# we don't care about inclusions, the caller cares.
1500
# but we need to setup a list of records to visit.
1501
# we need version_id, position, length
1502
version_id_records = []
1503
requested_versions = set(version_ids)
1504
# filter for available versions
1505
for version_id in requested_versions:
1506
if not self.has_version(version_id):
1507
raise RevisionNotPresent(version_id, self.filename)
1508
# get a in-component-order queue:
1509
for version_id in self.versions():
1510
if version_id in requested_versions:
1511
index_memo = self._index.get_position(version_id)
1512
version_id_records.append((version_id, index_memo))
1513
1514
total = len(version_id_records)
1515
for version_idx, (version_id, data, sha_value) in \
1516
enumerate(self._data.read_records_iter(version_id_records)):
1517
pb.update('Walking content.', version_idx, total)
1518
method = self._index.get_method(version_id)
1519
if method == 'fulltext':
1520
line_iterator = self.factory.get_fulltext_content(data)
1521
elif method == 'line-delta':
1522
line_iterator = self.factory.get_linedelta_content(data)
1523
else:
1524
raise ValueError('invalid method %r' % (method,))
1525
# XXX: It might be more efficient to yield (version_id,
1526
# line_iterator) in the future. However for now, this is a simpler
1527
# change to integrate into the rest of the codebase. RBC 20071110
1528
for line in line_iterator:
1529
yield line, version_id
1530
1531
pb.update('Walking content.', total, total)
1532
1533
def num_versions(self):
1534
"""See VersionedFile.num_versions()."""
1535
return self._index.num_versions()
1536
1537
__len__ = num_versions
1538
1539
def annotate(self, version_id):
1540
"""See VersionedFile.annotate."""
1541
return self.factory.annotate(self, version_id)
1542
1543
def get_parent_map(self, version_ids):
1544
"""See VersionedFile.get_parent_map."""
1545
return self._index.get_parent_map(version_ids)
1546
1547
def get_ancestry(self, versions, topo_sorted=True):
1548
"""See VersionedFile.get_ancestry."""
1549
if isinstance(versions, basestring):
1550
versions = [versions]
1551
if not versions:
1552
return []
1553
return self._index.get_ancestry(versions, topo_sorted)
1554
1555
def get_ancestry_with_ghosts(self, versions):
1556
"""See VersionedFile.get_ancestry_with_ghosts."""
1557
if isinstance(versions, basestring):
1558
versions = [versions]
1559
if not versions:
1560
return []
1561
return self._index.get_ancestry_with_ghosts(versions)
1562
1563
def plan_merge(self, ver_a, ver_b):
1564
"""See VersionedFile.plan_merge."""
1565
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1566
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1567
annotated_a = self.annotate(ver_a)
1568
annotated_b = self.annotate(ver_b)
1569
return merge._plan_annotate_merge(annotated_a, annotated_b,
1570
ancestors_a, ancestors_b)
1571
1572
1573
class _KnitComponentFile(object):
1574
"""One of the files used to implement a knit database"""
1575
1576
def __init__(self, transport, filename, mode, file_mode=None,
1577
create_parent_dir=False, dir_mode=None):
1578
self._transport = transport
1579
self._filename = filename
1580
self._mode = mode
1581
self._file_mode = file_mode
1582
self._dir_mode = dir_mode
1583
self._create_parent_dir = create_parent_dir
1584
self._need_to_create = False
1585
1586
def _full_path(self):
1587
"""Return the full path to this file."""
1588
return self._transport.base + self._filename
1589
1590
def check_header(self, fp):
1591
line = fp.readline()
1592
if line == '':
1593
# An empty file can actually be treated as though the file doesn't
1594
# exist yet.
1595
raise errors.NoSuchFile(self._full_path())
1596
if line != self.HEADER:
1597
raise KnitHeaderError(badline=line,
1598
filename=self._transport.abspath(self._filename))
1599
1600
def __repr__(self):
1601
return '%s(%s)' % (self.__class__.__name__, self._filename)
1602
1603
1604
class _KnitIndex(_KnitComponentFile):
1605
"""Manages knit index file.
1606
1607
The index is already kept in memory and read on startup, to enable
1360
def _parse_record(self, version_id, data):
1361
"""Parse an original format knit record.
1362
1363
These have the last element of the key only present in the stored data.
1364
"""
1365
rec, record_contents = self._parse_record_unchecked(data)
1366
self._check_header_version(rec, version_id)
1367
return record_contents, rec[3]
1368
1369
def _parse_record_header(self, key, raw_data):
1370
"""Parse a record header for consistency.
1371
1372
:return: the header and the decompressor stream.
1373
as (stream, header_record)
1374
"""
1375
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1376
try:
1377
# Current serialise
1378
rec = self._check_header(key, df.readline())
1379
except Exception, e:
1380
raise KnitCorrupt(self,
1381
"While reading {%s} got %s(%s)"
1382
% (key, e.__class__.__name__, str(e)))
1383
return df, rec
1384
1385
def _parse_record_unchecked(self, data):
1386
# profiling notes:
1387
# 4168 calls in 2880 217 internal
1388
# 4168 calls to _parse_record_header in 2121
1389
# 4168 calls to readlines in 330
1390
df = GzipFile(mode='rb', fileobj=StringIO(data))
1391
try:
1392
record_contents = df.readlines()
1393
except Exception, e:
1394
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1395
(data, e.__class__.__name__, str(e)))
1396
header = record_contents.pop(0)
1397
rec = self._split_header(header)
1398
last_line = record_contents.pop()
1399
if len(record_contents) != int(rec[2]):
1400
raise KnitCorrupt(self,
1401
'incorrect number of lines %s != %s'
1402
' for version {%s} %s'
1403
% (len(record_contents), int(rec[2]),
1404
rec[1], record_contents))
1405
if last_line != 'end %s\n' % rec[1]:
1406
raise KnitCorrupt(self,
1407
'unexpected version end line %r, wanted %r'
1408
% (last_line, rec[1]))
1409
df.close()
1410
return rec, record_contents
1411
1412
def _read_records_iter(self, records):
1413
"""Read text records from data file and yield result.
1414
1415
The result will be returned in whatever is the fastest to read.
1416
Not by the order requested. Also, multiple requests for the same
1417
record will only yield 1 response.
1418
:param records: A list of (key, access_memo) entries
1419
:return: Yields (key, contents, digest) in the order
1420
read, not the order requested
1421
"""
1422
if not records:
1423
return
1424
1425
# XXX: This smells wrong, IO may not be getting ordered right.
1426
needed_records = sorted(set(records), key=operator.itemgetter(1))
1427
if not needed_records:
1428
return
1429
1430
# The transport optimizes the fetching as well
1431
# (ie, reads continuous ranges.)
1432
raw_data = self._access.get_raw_records(
1433
[index_memo for key, index_memo in needed_records])
1434
1435
for (key, index_memo), data in \
1436
izip(iter(needed_records), raw_data):
1437
content, digest = self._parse_record(key[-1], data)
1438
yield key, content, digest
1439
1440
def _read_records_iter_raw(self, records):
1441
"""Read text records from data file and yield raw data.
1442
1443
This unpacks enough of the text record to validate the id is
1444
as expected but thats all.
1445
1446
Each item the iterator yields is (key, bytes, sha1_of_full_text).
1447
"""
1448
# setup an iterator of the external records:
1449
# uses readv so nice and fast we hope.
1450
if len(records):
1451
# grab the disk data needed.
1452
needed_offsets = [index_memo for key, index_memo
1453
in records]
1454
raw_records = self._access.get_raw_records(needed_offsets)
1455
1456
for key, index_memo in records:
1457
data = raw_records.next()
1458
# validate the header (note that we can only use the suffix in
1459
# current knit records).
1460
df, rec = self._parse_record_header(key, data)
1461
df.close()
1462
yield key, data, rec[3]
1463
1464
def _record_to_data(self, key, digest, lines, dense_lines=None):
1465
"""Convert key, digest, lines into a raw data block.
1466
1467
:param key: The key of the record. Currently keys are always serialised
1468
using just the trailing component.
1469
:param dense_lines: The bytes of lines but in a denser form. For
1470
instance, if lines is a list of 1000 bytestrings each ending in \n,
1471
dense_lines may be a list with one line in it, containing all the
1472
1000's lines and their \n's. Using dense_lines if it is already
1473
known is a win because the string join to create bytes in this
1474
function spends less time resizing the final string.
1475
:return: (len, a StringIO instance with the raw data ready to read.)
1476
"""
1477
# Note: using a string copy here increases memory pressure with e.g.
1478
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1479
# when doing the initial commit of a mozilla tree. RBC 20070921
1480
bytes = ''.join(chain(
1481
["version %s %d %s\n" % (key[-1],
1482
len(lines),
1483
digest)],
1484
dense_lines or lines,
1485
["end %s\n" % key[-1]]))
1486
if type(bytes) != str:
1487
raise AssertionError(
1488
'data must be plain bytes was %s' % type(bytes))
1489
if lines and lines[-1][-1] != '\n':
1490
raise ValueError('corrupt lines value %r' % lines)
1491
compressed_bytes = bytes_to_gzip(bytes)
1492
return len(compressed_bytes), compressed_bytes
1493
1494
def _split_header(self, line):
1495
rec = line.split()
1496
if len(rec) != 4:
1497
raise KnitCorrupt(self,
1498
'unexpected number of elements in record header')
1499
return rec
1500
1501
def keys(self):
1502
"""See VersionedFiles.keys."""
1503
if 'evil' in debug.debug_flags:
1504
trace.mutter_callsite(2, "keys scales with size of history")
1505
return self._index.keys()
1506
1507
1508
class _KndxIndex(object):
1509
"""Manages knit index files
1510
1511
The index is kept in memorya already kept in memory and read on startup, to enable
1608
1512
fast lookups of revision information. The cursor of the index
1609
1513
file is always pointing to the end, making it easy to append
1610
1514
entries.
1656
1560
1657
1561
HEADER = "# bzr knit index 8\n"
1658
1562
1659
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1660
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1661
1662
def _cache_version(self, version_id, options, pos, size, parents):
1563
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
1564
"""Create a _KndxIndex on transport using mapper."""
1565
self._transport = transport
1566
self._mapper = mapper
1567
self._get_scope = get_scope
1568
self._allow_writes = allow_writes
1569
self._is_locked = is_locked
1570
self._reset_cache()
1571
self.has_graph = True
1572
1573
def add_records(self, records, random_id=False):
1574
"""Add multiple records to the index.
1575
1576
:param records: a list of tuples:
1577
(key, options, access_memo, parents).
1578
:param random_id: If True the ids being added were randomly generated
1579
and no check for existence will be performed.
1580
"""
1581
paths = {}
1582
for record in records:
1583
key = record[0]
1584
prefix = key[:-1]
1585
path = self._mapper.map(key) + '.kndx'
1586
path_keys = paths.setdefault(path, (prefix, []))
1587
path_keys[1].append(record)
1588
for path in sorted(paths):
1589
prefix, path_keys = paths[path]
1590
self._load_prefixes([prefix])
1591
lines = []
1592
orig_history = self._kndx_cache[prefix][1][:]
1593
orig_cache = self._kndx_cache[prefix][0].copy()
1594
1595
try:
1596
for key, options, (_, pos, size), parents in path_keys:
1597
if parents is None:
1598
# kndx indices cannot be parentless.
1599
parents = ()
1600
line = "\n%s %s %s %s %s :" % (
1601
key[-1], ','.join(options), pos, size,
1602
self._dictionary_compress(parents))
1603
if type(line) != str:
1604
raise AssertionError(
1605
'data must be utf8 was %s' % type(line))
1606
lines.append(line)
1607
self._cache_key(key, options, pos, size, parents)
1608
if len(orig_history):
1609
self._transport.append_bytes(path, ''.join(lines))
1610
else:
1611
self._init_index(path, lines)
1612
except:
1613
# If any problems happen, restore the original values and re-raise
1614
self._kndx_cache[prefix] = (orig_cache, orig_history)
1615
raise
1616
1617
def _cache_key(self, key, options, pos, size, parent_keys):
1663
1618
"""Cache a version record in the history array and index cache.
1664
1619
1665
1620
This is inlined into _load_data for performance. KEEP IN SYNC.
1666
1621
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1667
1622
indexes).
1668
1623
"""
1624
prefix = key[:-1]
1625
version_id = key[-1]
1626
# last-element only for compatibilty with the C load_data.
1627
parents = tuple(parent[-1] for parent in parent_keys)
1628
for parent in parent_keys:
1629
if parent[:-1] != prefix:
1630
raise ValueError("mismatched prefixes for %r, %r" % (
1631
key, parent_keys))
1632
cache, history = self._kndx_cache[prefix]
1669
1633
# only want the _history index to reference the 1st index entry
1670
1634
# for version_id
1671
if version_id not in self._cache:
1672
index = len(self._history)
1673
self._history.append(version_id)
1635
if version_id not in cache:
1636
index = len(history)
1637
history.append(version_id)
1674
1638
else:
1675
index = self._cache[version_id][5]
1676
self._cache[version_id] = (version_id,
1639
index = cache[version_id][5]
1640
cache[version_id] = (version_id,
1677
1641
options,
1678
1642
pos,
1679
1643
size,
1680
1644
parents,
1681
1645
index)
1682
1646
1647
def check_header(self, fp):
1648
line = fp.readline()
1649
if line == '':
1650
# An empty file can actually be treated as though the file doesn't
1651
# exist yet.
1652
raise errors.NoSuchFile(self)
1653
if line != self.HEADER:
1654
raise KnitHeaderError(badline=line, filename=self)
1655
1656
def _check_read(self):
1657
if not self._is_locked():
1658
raise errors.ObjectNotLocked(self)
1659
if self._get_scope() != self._scope:
1660
self._reset_cache()
1661
1683
1662
def _check_write_ok(self):
1663
"""Assert if not writes are permitted."""
1664
if not self._is_locked():
1665
raise errors.ObjectNotLocked(self)
1684
1666
if self._get_scope() != self._scope:
1685
raise errors.OutSideTransaction()
1667
self._reset_cache()
1686
1668
if self._mode != 'w':
1687
1669
raise errors.ReadOnlyObjectDirtiedError(self)
1688
1670
1689
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1690
create_parent_dir=False, delay_create=False, dir_mode=None,
1691
get_scope=None):
1692
_KnitComponentFile.__init__(self, transport, filename, mode,
1693
file_mode=file_mode,
1694
create_parent_dir=create_parent_dir,
1695
dir_mode=dir_mode)
1696
self._cache = {}
1697
# position in _history is the 'official' index for a revision
1698
# but the values may have come from a newer entry.
1699
# so - wc -l of a knit index is != the number of unique names
1700
# in the knit.
1701
self._history = []
1702
try:
1703
fp = self._transport.get(self._filename)
1704
try:
1705
# _load_data may raise NoSuchFile if the target knit is
1706
# completely empty.
1707
_load_data(self, fp)
1708
finally:
1709
fp.close()
1710
except NoSuchFile:
1711
if mode != 'w' or not create:
1712
raise
1713
elif delay_create:
1714
self._need_to_create = True
1715
else:
1716
self._transport.put_bytes_non_atomic(
1717
self._filename, self.HEADER, mode=self._file_mode)
1718
self._scope = get_scope()
1719
self._get_scope = get_scope
1720
1721
def get_ancestry(self, versions, topo_sorted=True):
1722
"""See VersionedFile.get_ancestry."""
1723
# get a graph of all the mentioned versions:
1724
graph = {}
1725
pending = set(versions)
1726
cache = self._cache
1727
while pending:
1728
version = pending.pop()
1729
# trim ghosts
1730
try:
1731
parents = [p for p in cache[version][4] if p in cache]
1732
except KeyError:
1733
raise RevisionNotPresent(version, self._filename)
1734
# if not completed and not a ghost
1735
pending.update([p for p in parents if p not in graph])
1736
graph[version] = parents
1737
if not topo_sorted:
1738
return graph.keys()
1739
return topo_sort(graph.items())
1740
1741
def get_ancestry_with_ghosts(self, versions):
1742
"""See VersionedFile.get_ancestry_with_ghosts."""
1743
# get a graph of all the mentioned versions:
1744
self.check_versions_present(versions)
1745
cache = self._cache
1746
graph = {}
1747
pending = set(versions)
1748
while pending:
1749
version = pending.pop()
1750
try:
1751
parents = cache[version][4]
1752
except KeyError:
1753
# ghost, fake it
1754
graph[version] = []
1755
else:
1756
# if not completed
1757
pending.update([p for p in parents if p not in graph])
1758
graph[version] = parents
1759
return topo_sort(graph.items())
1760
1761
def get_build_details(self, version_ids):
1762
"""Get the method, index_memo and compression parent for version_ids.
1671
def get_build_details(self, keys):
1672
"""Get the method, index_memo and compression parent for keys.
1763
1673
1764
1674
Ghosts are omitted from the result.
1765
1675
1766
:param version_ids: An iterable of version_ids.
1767
:return: A dict of version_id:(index_memo, compression_parent,
1768
parents, record_details).
1676
:param keys: An iterable of keys.
1677
:return: A dict of key:(access_memo, compression_parent, parents,
1678
record_details).
1769
1679
index_memo
1770
1680
opaque structure to pass to read_records to extract the raw
1771
1681
data
1777
1687
extra information about the content which needs to be passed to
1778
1688
Factory.parse_record
1779
1689
"""
1690
prefixes = self._partition_keys(keys)
1691
parent_map = self.get_parent_map(keys)
1780
1692
result = {}
1781
for version_id in version_ids:
1782
if version_id not in self._cache:
1783
# ghosts are omitted
1784
continue
1785
method = self.get_method(version_id)
1786
parents = self.get_parents_with_ghosts(version_id)
1693
for key in keys:
1694
if key not in parent_map:
1695
continue # Ghost
1696
method = self.get_method(key)
1697
parents = parent_map[key]
1787
1698
if method == 'fulltext':
1788
1699
compression_parent = None
1789
1700
else:
1790
1701
compression_parent = parents[0]
1791
noeol = 'no-eol' in self.get_options(version_id)
1792
index_memo = self.get_position(version_id)
1793
result[version_id] = (index_memo, compression_parent,
1702
noeol = 'no-eol' in self.get_options(key)
1703
index_memo = self.get_position(key)
1704
result[key] = (index_memo, compression_parent,
1794
1705
parents, (method, noeol))
1795
1706
return result
1796
1707
1797
def num_versions(self):
1798
return len(self._history)
1799
1800
__len__ = num_versions
1801
1802
def get_versions(self):
1803
"""Get all the versions in the file. not topologically sorted."""
1804
return self._history
1805
1806
def _version_list_to_index(self, versions):
1708
def get_method(self, key):
1709
"""Return compression method of specified key."""
1710
options = self.get_options(key)
1711
if 'fulltext' in options:
1712
return 'fulltext'
1713
elif 'line-delta' in options:
1714
return 'line-delta'
1715
else:
1716
raise errors.KnitIndexUnknownMethod(self, options)
1717
1718
def get_options(self, key):
1719
"""Return a list representing options.
1720
1721
e.g. ['foo', 'bar']
1722
"""
1723
prefix, suffix = self._split_key(key)
1724
self._load_prefixes([prefix])
1725
return self._kndx_cache[prefix][0][suffix][1]
1726
1727
def get_parent_map(self, keys):
1728
"""Get a map of the parents of keys.
1729
1730
:param keys: The keys to look up parents for.
1731
:return: A mapping from keys to parents. Absent keys are absent from
1732
the mapping.
1733
"""
1734
# Parse what we need to up front, this potentially trades off I/O
1735
# locality (.kndx and .knit in the same block group for the same file
1736
# id) for less checking in inner loops.
1737
prefixes = set()
1738
prefixes.update(key[:-1] for key in keys)
1739
self._load_prefixes(prefixes)
1740
result = {}
1741
for key in keys:
1742
prefix = key[:-1]
1743
try:
1744
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
1745
except KeyError:
1746
pass
1747
else:
1748
result[key] = tuple(prefix + (suffix,) for
1749
suffix in suffix_parents)
1750
return result
1751
1752
def get_position(self, key):
1753
"""Return details needed to access the version.
1754
1755
:return: a tuple (key, data position, size) to hand to the access
1756
logic to get the record.
1757
"""
1758
prefix, suffix = self._split_key(key)
1759
self._load_prefixes([prefix])
1760
entry = self._kndx_cache[prefix][0][suffix]
1761
return key, entry[2], entry[3]
1762
1763
def _init_index(self, path, extra_lines=[]):
1764
"""Initialize an index."""
1765
sio = StringIO()
1766
sio.write(self.HEADER)
1767
sio.writelines(extra_lines)
1768
sio.seek(0)
1769
self._transport.put_file_non_atomic(path, sio,
1770
create_parent_dir=True)
1771
# self._create_parent_dir)
1772
# mode=self._file_mode,
1773
# dir_mode=self._dir_mode)
1774
1775
def keys(self):
1776
"""Get all the keys in the collection.
1777
1778
The keys are not ordered.
1779
"""
1780
result = set()
1781
# Identify all key prefixes.
1782
# XXX: A bit hacky, needs polish.
1783
if type(self._mapper) == ConstantMapper:
1784
prefixes = [()]
1785
else:
1786
relpaths = set()
1787
for quoted_relpath in self._transport.iter_files_recursive():
1788
path, ext = os.path.splitext(quoted_relpath)
1789
relpaths.add(path)
1790
prefixes = [self._mapper.unmap(path) for path in relpaths]
1791
self._load_prefixes(prefixes)
1792
for prefix in prefixes:
1793
for suffix in self._kndx_cache[prefix][1]:
1794
result.add(prefix + (suffix,))
1795
return result
1796
1797
def _load_prefixes(self, prefixes):
1798
"""Load the indices for prefixes."""
1799
self._check_read()
1800
for prefix in prefixes:
1801
if prefix not in self._kndx_cache:
1802
# the load_data interface writes to these variables.
1803
self._cache = {}
1804
self._history = []
1805
self._filename = prefix
1806
try:
1807
path = self._mapper.map(prefix) + '.kndx'
1808
fp = self._transport.get(path)
1809
try:
1810
# _load_data may raise NoSuchFile if the target knit is
1811
# completely empty.
1812
_load_data(self, fp)
1813
finally:
1814
fp.close()
1815
self._kndx_cache[prefix] = (self._cache, self._history)
1816
del self._cache
1817
del self._filename
1818
del self._history
1819
except NoSuchFile:
1820
self._kndx_cache[prefix] = ({}, [])
1821
if type(self._mapper) == ConstantMapper:
1822
# preserve behaviour for revisions.kndx etc.
1823
self._init_index(path)
1824
del self._cache
1825
del self._filename
1826
del self._history
1827
1828
def _partition_keys(self, keys):
1829
"""Turn keys into a dict of prefix:suffix_list."""
1830
result = {}
1831
for key in keys:
1832
prefix_keys = result.setdefault(key[:-1], [])
1833
prefix_keys.append(key[-1])
1834
return result
1835
1836
def _dictionary_compress(self, keys):
1837
"""Dictionary compress keys.
1838
1839
:param keys: The keys to generate references to.
1840
:return: A string representation of keys. keys which are present are
1841
dictionary compressed, and others are emitted as fulltext with a
1842
'.' prefix.
1843
"""
1844
if not keys:
1845
return ''
1807
1846
result_list = []
1808
cache = self._cache
1809
for version in versions:
1810
if version in cache:
1847
prefix = keys[0][:-1]
1848
cache = self._kndx_cache[prefix][0]
1849
for key in keys:
1850
if key[:-1] != prefix:
1851
# kndx indices cannot refer across partitioned storage.
1852
raise ValueError("mismatched prefixes for %r" % keys)
1853
if key[-1] in cache:
1811
1854
# -- inlined lookup() --
1812
result_list.append(str(cache[version][5]))
1855
result_list.append(str(cache[key[-1]][5]))
1813
1856
# -- end lookup () --
1814
1857
else:
1815
result_list.append('.' + version)
1858
result_list.append('.' + key[-1])
1816
1859
return ' '.join(result_list)
1817
1860
1818
def add_version(self, version_id, options, index_memo, parents):
1819
"""Add a version record to the index."""
1820
self.add_versions(((version_id, options, index_memo, parents),))
1821
1822
def add_versions(self, versions, random_id=False):
1823
"""Add multiple versions to the index.
1824
1825
:param versions: a list of tuples:
1826
(version_id, options, pos, size, parents).
1827
:param random_id: If True the ids being added were randomly generated
1828
and no check for existence will be performed.
1829
"""
1830
lines = []
1831
orig_history = self._history[:]
1832
orig_cache = self._cache.copy()
1833
1834
try:
1835
for version_id, options, (index, pos, size), parents in versions:
1836
line = "\n%s %s %s %s %s :" % (version_id,
1837
','.join(options),
1838
pos,
1839
size,
1840
self._version_list_to_index(parents))
1841
lines.append(line)
1842
self._cache_version(version_id, options, pos, size, tuple(parents))
1843
if not self._need_to_create:
1844
self._transport.append_bytes(self._filename, ''.join(lines))
1845
else:
1846
sio = StringIO()
1847
sio.write(self.HEADER)
1848
sio.writelines(lines)
1849
sio.seek(0)
1850
self._transport.put_file_non_atomic(self._filename, sio,
1851
create_parent_dir=self._create_parent_dir,
1852
mode=self._file_mode,
1853
dir_mode=self._dir_mode)
1854
self._need_to_create = False
1855
except:
1856
# If any problems happen, restore the original values and re-raise
1857
self._history = orig_history
1858
self._cache = orig_cache
1859
raise
1860
1861
def has_version(self, version_id):
1862
"""True if the version is in the index."""
1863
return version_id in self._cache
1864
1865
def get_position(self, version_id):
1866
"""Return details needed to access the version.
1867
1868
.kndx indices do not support split-out data, so return None for the
1869
index field.
1870
1871
:return: a tuple (None, data position, size) to hand to the access
1872
logic to get the record.
1873
"""
1874
entry = self._cache[version_id]
1875
return None, entry[2], entry[3]
1876
1877
def get_method(self, version_id):
1878
"""Return compression method of specified version."""
1879
try:
1880
options = self._cache[version_id][1]
1881
except KeyError:
1882
raise RevisionNotPresent(version_id, self._filename)
1883
if 'fulltext' in options:
1884
return 'fulltext'
1861
def _reset_cache(self):
1862
# Possibly this should be a LRU cache. A dictionary from key_prefix to
1863
# (cache_dict, history_vector) for parsed kndx files.
1864
self._kndx_cache = {}
1865
self._scope = self._get_scope()
1866
allow_writes = self._allow_writes()
1867
if allow_writes:
1868
self._mode = 'w'
1885
1869
else:
1886
if 'line-delta' not in options:
1887
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1888
return 'line-delta'
1889
1890
def get_options(self, version_id):
1891
"""Return a list representing options.
1892
1893
e.g. ['foo', 'bar']
1894
"""
1895
return self._cache[version_id][1]
1896
1897
def get_parent_map(self, version_ids):
1898
"""Passed through to by KnitVersionedFile.get_parent_map."""
1899
result = {}
1900
for version_id in version_ids:
1901
try:
1902
result[version_id] = tuple(self._cache[version_id][4])
1903
except KeyError:
1904
pass
1905
return result
1906
1907
def get_parents_with_ghosts(self, version_id):
1908
"""Return parents of specified version with ghosts."""
1909
try:
1910
return self.get_parent_map([version_id])[version_id]
1911
except KeyError:
1912
raise RevisionNotPresent(version_id, self)
1913
1914
def check_versions_present(self, version_ids):
1915
"""Check that all specified versions are present."""
1916
cache = self._cache
1917
for version_id in version_ids:
1918
if version_id not in cache:
1919
raise RevisionNotPresent(version_id, self._filename)
1920
1921
1922
class KnitGraphIndex(object):
1923
"""A knit index that builds on GraphIndex."""
1924
1925
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1870
self._mode = 'r'
1871
1872
def _split_key(self, key):
1873
"""Split key into a prefix and suffix."""
1874
return key[:-1], key[-1]
1875
1876
1877
class _KnitGraphIndex(object):
1878
"""A KnitVersionedFiles index layered on GraphIndex."""
1879
1880
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
1881
add_callback=None):
1926
1882
"""Construct a KnitGraphIndex on a graph_index.
1927
1883
1928
1884
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1885
:param is_locked: A callback to check whether the object should answer
1886
queries.
1929
1887
:param deltas: Allow delta-compressed records.
1888
:param parents: If True, record knits parents, if not do not record
1889
parents.
1930
1890
:param add_callback: If not None, allow additions to the index and call
1931
1891
this callback with a list of added GraphIndex nodes:
1932
1892
[(node, value, node_refs), ...]
1933
:param parents: If True, record knits parents, if not do not record
1934
parents.
1893
:param is_locked: A callback, returns True if the index is locked and
1894
thus usable.
1935
1895
"""
1896
self._add_callback = add_callback
1936
1897
self._graph_index = graph_index
1937
1898
self._deltas = deltas
1938
self._add_callback = add_callback
1939
1899
self._parents = parents
1940
1900
if deltas and not parents:
1901
# XXX: TODO: Delta tree and parent graph should be conceptually
1902
# separate.
1941
1903
raise KnitCorrupt(self, "Cannot do delta compression without "
1942
1904
"parent tracking.")
1943
1944
def _check_write_ok(self):
1945
pass
1946
1947
def _get_entries(self, keys, check_present=False):
1948
"""Get the entries for keys.
1949
1950
:param keys: An iterable of index keys, - 1-tuples.
1951
"""
1952
keys = set(keys)
1953
found_keys = set()
1954
if self._parents:
1955
for node in self._graph_index.iter_entries(keys):
1956
yield node
1957
found_keys.add(node[1])
1958
else:
1959
# adapt parentless index to the rest of the code.
1960
for node in self._graph_index.iter_entries(keys):
1961
yield node[0], node[1], node[2], ()
1962
found_keys.add(node[1])
1963
if check_present:
1964
missing_keys = keys.difference(found_keys)
1965
if missing_keys:
1966
raise RevisionNotPresent(missing_keys.pop(), self)
1967
1968
def _present_keys(self, version_ids):
1969
return set([
1970
node[1] for node in self._get_entries(version_ids)])
1971
1972
def _parentless_ancestry(self, versions):
1973
"""Honour the get_ancestry API for parentless knit indices."""
1974
wanted_keys = self._version_ids_to_keys(versions)
1975
present_keys = self._present_keys(wanted_keys)
1976
missing = set(wanted_keys).difference(present_keys)
1977
if missing:
1978
raise RevisionNotPresent(missing.pop(), self)
1979
return list(self._keys_to_version_ids(present_keys))
1980
1981
def get_ancestry(self, versions, topo_sorted=True):
1982
"""See VersionedFile.get_ancestry."""
1983
if not self._parents:
1984
return self._parentless_ancestry(versions)
1985
# XXX: This will do len(history) index calls - perhaps
1986
# it should be altered to be a index core feature?
1987
# get a graph of all the mentioned versions:
1988
graph = {}
1989
ghosts = set()
1990
versions = self._version_ids_to_keys(versions)
1991
pending = set(versions)
1992
while pending:
1993
# get all pending nodes
1994
this_iteration = pending
1995
new_nodes = self._get_entries(this_iteration)
1996
found = set()
1997
pending = set()
1998
for (index, key, value, node_refs) in new_nodes:
1999
# dont ask for ghosties - otherwise
2000
# we we can end up looping with pending
2001
# being entirely ghosted.
2002
graph[key] = [parent for parent in node_refs[0]
2003
if parent not in ghosts]
2004
# queue parents
2005
for parent in graph[key]:
2006
# dont examine known nodes again
2007
if parent in graph:
2008
continue
2009
pending.add(parent)
2010
found.add(key)
2011
ghosts.update(this_iteration.difference(found))
2012
if versions.difference(graph):
2013
raise RevisionNotPresent(versions.difference(graph).pop(), self)
2014
if topo_sorted:
2015
result_keys = topo_sort(graph.items())
2016
else:
2017
result_keys = graph.iterkeys()
2018
return [key[0] for key in result_keys]
2019
2020
def get_ancestry_with_ghosts(self, versions):
2021
"""See VersionedFile.get_ancestry."""
2022
if not self._parents:
2023
return self._parentless_ancestry(versions)
2024
# XXX: This will do len(history) index calls - perhaps
2025
# it should be altered to be a index core feature?
2026
# get a graph of all the mentioned versions:
2027
graph = {}
2028
versions = self._version_ids_to_keys(versions)
2029
pending = set(versions)
2030
while pending:
2031
# get all pending nodes
2032
this_iteration = pending
2033
new_nodes = self._get_entries(this_iteration)
2034
pending = set()
2035
for (index, key, value, node_refs) in new_nodes:
2036
graph[key] = node_refs[0]
2037
# queue parents
2038
for parent in graph[key]:
2039
# dont examine known nodes again
2040
if parent in graph:
2041
continue
2042
pending.add(parent)
2043
missing_versions = this_iteration.difference(graph)
2044
missing_needed = versions.intersection(missing_versions)
2045
if missing_needed:
2046
raise RevisionNotPresent(missing_needed.pop(), self)
2047
for missing_version in missing_versions:
2048
# add a key, no parents
2049
graph[missing_version] = []
2050
pending.discard(missing_version) # don't look for it
2051
result_keys = topo_sort(graph.items())
2052
return [key[0] for key in result_keys]
2053
2054
def get_build_details(self, version_ids):
2055
"""Get the method, index_memo and compression parent for version_ids.
2056
2057
Ghosts are omitted from the result.
2058
2059
:param version_ids: An iterable of version_ids.
2060
:return: A dict of version_id:(index_memo, compression_parent,
2061
parents, record_details).
2062
index_memo
2063
opaque structure to pass to read_records to extract the raw
2064
data
2065
compression_parent
2066
Content that this record is built upon, may be None
2067
parents
2068
Logical parents of this node
2069
record_details
2070
extra information about the content which needs to be passed to
2071
Factory.parse_record
2072
"""
2073
result = {}
2074
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
2075
for entry in entries:
2076
version_id = self._keys_to_version_ids((entry[1],))[0]
2077
if not self._parents:
2078
parents = ()
2079
else:
2080
parents = self._keys_to_version_ids(entry[3][0])
2081
if not self._deltas:
2082
compression_parent = None
2083
else:
2084
compression_parent_key = self._compression_parent(entry)
2085
if compression_parent_key:
2086
compression_parent = self._keys_to_version_ids(
2087
(compression_parent_key,))[0]
2088
else:
2089
compression_parent = None
2090
noeol = (entry[2][0] == 'N')
2091
if compression_parent:
2092
method = 'line-delta'
2093
else:
2094
method = 'fulltext'
2095
result[version_id] = (self._node_to_position(entry),
2096
compression_parent, parents,
2097
(method, noeol))
2098
return result
2099
2100
def _compression_parent(self, an_entry):
2101
# return the key that an_entry is compressed against, or None
2102
# Grab the second parent list (as deltas implies parents currently)
2103
compression_parents = an_entry[3][1]
2104
if not compression_parents:
2105
return None
2106
return compression_parents[0]
2107
2108
def _get_method(self, node):
2109
if not self._deltas:
2110
return 'fulltext'
2111
if self._compression_parent(node):
2112
return 'line-delta'
2113
else:
2114
return 'fulltext'
2115
2116
def num_versions(self):
2117
return len(list(self._graph_index.iter_all_entries()))
2118
2119
__len__ = num_versions
2120
2121
def get_versions(self):
2122
"""Get all the versions in the file. not topologically sorted."""
2123
return [node[1][0] for node in self._graph_index.iter_all_entries()]
2124
2125
def has_version(self, version_id):
2126
"""True if the version is in the index."""
2127
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
2128
2129
def _keys_to_version_ids(self, keys):
2130
return tuple(key[0] for key in keys)
2131
2132
def get_position(self, version_id):
2133
"""Return details needed to access the version.
2134
2135
:return: a tuple (index, data position, size) to hand to the access
2136
logic to get the record.
2137
"""
2138
node = self._get_node(version_id)
2139
return self._node_to_position(node)
2140
2141
def _node_to_position(self, node):
2142
"""Convert an index value to position details."""
2143
bits = node[2][1:].split(' ')
2144
return node[0], int(bits[0]), int(bits[1])
2145
2146
def get_method(self, version_id):
2147
"""Return compression method of specified version."""
2148
return self._get_method(self._get_node(version_id))
2149
2150
def _get_node(self, version_id):
2151
try:
2152
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
2153
except IndexError:
2154
raise RevisionNotPresent(version_id, self)
2155
2156
def get_options(self, version_id):
2157
"""Return a list representing options.
2158
2159
e.g. ['foo', 'bar']
2160
"""
2161
node = self._get_node(version_id)
2162
options = [self._get_method(node)]
2163
if node[2][0] == 'N':
2164
options.append('no-eol')
2165
return options
2166
2167
def get_parent_map(self, version_ids):
2168
"""Passed through to by KnitVersionedFile.get_parent_map."""
2169
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
2170
result = {}
2171
if self._parents:
2172
for node in nodes:
2173
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
2174
else:
2175
for node in nodes:
2176
result[node[1][0]] = ()
2177
return result
2178
2179
def get_parents_with_ghosts(self, version_id):
2180
"""Return parents of specified version with ghosts."""
2181
try:
2182
return self.get_parent_map([version_id])[version_id]
2183
except KeyError:
2184
raise RevisionNotPresent(version_id, self)
2185
2186
def check_versions_present(self, version_ids):
2187
"""Check that all specified versions are present."""
2188
keys = self._version_ids_to_keys(version_ids)
2189
present = self._present_keys(keys)
2190
missing = keys.difference(present)
2191
if missing:
2192
raise RevisionNotPresent(missing.pop(), self)
2193
2194
def add_version(self, version_id, options, access_memo, parents):
2195
"""Add a version record to the index."""
2196
return self.add_versions(((version_id, options, access_memo, parents),))
2197
2198
def add_versions(self, versions, random_id=False):
2199
"""Add multiple versions to the index.
1905
self.has_graph = parents
1906
self._is_locked = is_locked
1907
1908
def add_records(self, records, random_id=False):
1909
"""Add multiple records to the index.
2200
1910
2201
1911
This function does not insert data into the Immutable GraphIndex
2202
1912
backing the KnitGraphIndex, instead it prepares data for insertion by
2203
1913
the caller and checks that it is safe to insert then calls
2204
1914
self._add_callback with the prepared GraphIndex nodes.
2205
1915
2206
:param versions: a list of tuples:
2207
(version_id, options, pos, size, parents).
1916
:param records: a list of tuples:
1917
(key, options, access_memo, parents).
2208
1918
:param random_id: If True the ids being added were randomly generated
2209
1919
and no check for existence will be performed.
2210
1920
"""
2212
1922
raise errors.ReadOnlyError(self)
2213
1923
# we hope there are no repositories with inconsistent parentage
2214
1924
# anymore.
2215
# check for dups
2216
1925
2217
1926
keys = {}
2218
for (version_id, options, access_memo, parents) in versions:
1927
for (key, options, access_memo, parents) in records:
1928
if self._parents:
1929
parents = tuple(parents)
2219
1930
index, pos, size = access_memo
2220
key = (version_id, )
2221
parents = tuple((parent, ) for parent in parents)
2222
1931
if 'no-eol' in options:
2223
1932
value = 'N'
2224
1933
else:
2241
1950
"in parentless index.")
2242
1951
node_refs = ()
2243
1952
keys[key] = (value, node_refs)
1953
# check for dups
2244
1954
if not random_id:
2245
1955
present_nodes = self._get_entries(keys)
2246
1956
for (index, key, value, node_refs) in present_nodes:
2247
if (value, node_refs) != keys[key]:
2248
raise KnitCorrupt(self, "inconsistent details in add_versions"
1957
if (value[0] != keys[key][0][0] or
1958
node_refs != keys[key][1]):
1959
raise KnitCorrupt(self, "inconsistent details in add_records"
2249
1960
": %s %s" % ((value, node_refs), keys[key]))
2250
1961
del keys[key]
2251
1962
result = []
2257
1968
result.append((key, value))
2258
1969
self._add_callback(result)
2259
1970
2260
def _version_ids_to_keys(self, version_ids):
2261
return set((version_id, ) for version_id in version_ids)
2262
2263
2264
class _KnitAccess(object):
2265
"""Access to knit records in a .knit file."""
2266
2267
def __init__(self, transport, filename, _file_mode, _dir_mode,
2268
_need_to_create, _create_parent_dir):
2269
"""Create a _KnitAccess for accessing and inserting data.
2270
2271
:param transport: The transport the .knit is located on.
2272
:param filename: The filename of the .knit.
1971
def _check_read(self):
1972
"""raise if reads are not permitted."""
1973
if not self._is_locked():
1974
raise errors.ObjectNotLocked(self)
1975
1976
def _check_write_ok(self):
1977
"""Assert if writes are not permitted."""
1978
if not self._is_locked():
1979
raise errors.ObjectNotLocked(self)
1980
1981
def _compression_parent(self, an_entry):
1982
# return the key that an_entry is compressed against, or None
1983
# Grab the second parent list (as deltas implies parents currently)
1984
compression_parents = an_entry[3][1]
1985
if not compression_parents:
1986
return None
1987
if len(compression_parents) != 1:
1988
raise AssertionError(
1989
"Too many compression parents: %r" % compression_parents)
1990
return compression_parents[0]
1991
1992
def get_build_details(self, keys):
1993
"""Get the method, index_memo and compression parent for version_ids.
1994
1995
Ghosts are omitted from the result.
1996
1997
:param keys: An iterable of keys.
1998
:return: A dict of key:
1999
(index_memo, compression_parent, parents, record_details).
2000
index_memo
2001
opaque structure to pass to read_records to extract the raw
2002
data
2003
compression_parent
2004
Content that this record is built upon, may be None
2005
parents
2006
Logical parents of this node
2007
record_details
2008
extra information about the content which needs to be passed to
2009
Factory.parse_record
2010
"""
2011
self._check_read()
2012
result = {}
2013
entries = self._get_entries(keys, False)
2014
for entry in entries:
2015
key = entry[1]
2016
if not self._parents:
2017
parents = ()
2018
else:
2019
parents = entry[3][0]
2020
if not self._deltas:
2021
compression_parent_key = None
2022
else:
2023
compression_parent_key = self._compression_parent(entry)
2024
noeol = (entry[2][0] == 'N')
2025
if compression_parent_key:
2026
method = 'line-delta'
2027
else:
2028
method = 'fulltext'
2029
result[key] = (self._node_to_position(entry),
2030
compression_parent_key, parents,
2031
(method, noeol))
2032
return result
2033
2034
def _get_entries(self, keys, check_present=False):
2035
"""Get the entries for keys.
2036
2037
:param keys: An iterable of index key tuples.
2038
"""
2039
keys = set(keys)
2040
found_keys = set()
2041
if self._parents:
2042
for node in self._graph_index.iter_entries(keys):
2043
yield node
2044
found_keys.add(node[1])
2045
else:
2046
# adapt parentless index to the rest of the code.
2047
for node in self._graph_index.iter_entries(keys):
2048
yield node[0], node[1], node[2], ()
2049
found_keys.add(node[1])
2050
if check_present:
2051
missing_keys = keys.difference(found_keys)
2052
if missing_keys:
2053
raise RevisionNotPresent(missing_keys.pop(), self)
2054
2055
def get_method(self, key):
2056
"""Return compression method of specified key."""
2057
return self._get_method(self._get_node(key))
2058
2059
def _get_method(self, node):
2060
if not self._deltas:
2061
return 'fulltext'
2062
if self._compression_parent(node):
2063
return 'line-delta'
2064
else:
2065
return 'fulltext'
2066
2067
def _get_node(self, key):
2068
try:
2069
return list(self._get_entries([key]))[0]
2070
except IndexError:
2071
raise RevisionNotPresent(key, self)
2072
2073
def get_options(self, key):
2074
"""Return a list representing options.
2075
2076
e.g. ['foo', 'bar']
2077
"""
2078
node = self._get_node(key)
2079
options = [self._get_method(node)]
2080
if node[2][0] == 'N':
2081
options.append('no-eol')
2082
return options
2083
2084
def get_parent_map(self, keys):
2085
"""Get a map of the parents of keys.
2086
2087
:param keys: The keys to look up parents for.
2088
:return: A mapping from keys to parents. Absent keys are absent from
2089
the mapping.
2090
"""
2091
self._check_read()
2092
nodes = self._get_entries(keys)
2093
result = {}
2094
if self._parents:
2095
for node in nodes:
2096
result[node[1]] = node[3][0]
2097
else:
2098
for node in nodes:
2099
result[node[1]] = None
2100
return result
2101
2102
def get_position(self, key):
2103
"""Return details needed to access the version.
2104
2105
:return: a tuple (index, data position, size) to hand to the access
2106
logic to get the record.
2107
"""
2108
node = self._get_node(key)
2109
return self._node_to_position(node)
2110
2111
def keys(self):
2112
"""Get all the keys in the collection.
2113
2114
The keys are not ordered.
2115
"""
2116
self._check_read()
2117
return [node[1] for node in self._graph_index.iter_all_entries()]
2118
2119
def _node_to_position(self, node):
2120
"""Convert an index value to position details."""
2121
bits = node[2][1:].split(' ')
2122
return node[0], int(bits[0]), int(bits[1])
2123
2124
2125
class _KnitKeyAccess(object):
2126
"""Access to records in .knit files."""
2127
2128
def __init__(self, transport, mapper):
2129
"""Create a _KnitKeyAccess with transport and mapper.
2130
2131
:param transport: The transport the access object is rooted at.
2132
:param mapper: The mapper used to map keys to .knit files.
2273
2133
"""
2274
2134
self._transport = transport
2275
self._filename = filename
2276
self._file_mode = _file_mode
2277
self._dir_mode = _dir_mode
2278
self._need_to_create = _need_to_create
2279
self._create_parent_dir = _create_parent_dir
2135
self._mapper = mapper
2280
2136
2281
def add_raw_records(self, sizes, raw_data):
2137
def add_raw_records(self, key_sizes, raw_data):
2282
2138
"""Add raw knit bytes to a storage area.
2283
2139
2284
The data is spooled to whereever the access method is storing data.
2140
The data is spooled to the container writer in one bytes-record per
2141
raw data item.
2285
2142
2286
:param sizes: An iterable containing the size of each raw data segment.
2143
:param sizes: An iterable of tuples containing the key and size of each
2144
raw data segment.
2287
2145
:param raw_data: A bytestring containing the data.
2288
:return: A list of memos to retrieve the record later. Each memo is a
2289
tuple - (index, pos, length), where the index field is always None
2290
for the .knit access method.
2146
:return: A list of memos to retrieve the record later. Each memo is an
2147
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
2148
length), where the key is the record key.
2291
2149
"""
2292
if not self._need_to_create:
2293
base = self._transport.append_bytes(self._filename, raw_data)
2294
else:
2295
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2296
create_parent_dir=self._create_parent_dir,
2297
mode=self._file_mode,
2298
dir_mode=self._dir_mode)
2299
self._need_to_create = False
2300
base = 0
2150
if type(raw_data) != str:
2151
raise AssertionError(
2152
'data must be plain bytes was %s' % type(raw_data))
2301
2153
result = []
2302
for size in sizes:
2303
result.append((None, base, size))
2304
base += size
2154
offset = 0
2155
# TODO: This can be tuned for writing to sftp and other servers where
2156
# append() is relatively expensive by grouping the writes to each key
2157
# prefix.
2158
for key, size in key_sizes:
2159
path = self._mapper.map(key)
2160
try:
2161
base = self._transport.append_bytes(path + '.knit',
2162
raw_data[offset:offset+size])
2163
except errors.NoSuchFile:
2164
self._transport.mkdir(osutils.dirname(path))
2165
base = self._transport.append_bytes(path + '.knit',
2166
raw_data[offset:offset+size])
2167
# if base == 0:
2168
# chmod.
2169
offset += size
2170
result.append((key, base, size))
2305
2171
return result
2306
2172
2307
def create(self):
2308
"""IFF this data access has its own storage area, initialise it.
2309
2310
:return: None.
2311
"""
2312
self._transport.put_bytes_non_atomic(self._filename, '',
2313
mode=self._file_mode)
2314
2315
def open_file(self):
2316
"""IFF this data access can be represented as a single file, open it.
2317
2318
For knits that are not mapped to a single file on disk this will
2319
always return None.
2320
2321
:return: None or a file handle.
2322
"""
2323
try:
2324
return self._transport.get(self._filename)
2325
except NoSuchFile:
2326
pass
2327
return None
2328
2329
2173
def get_raw_records(self, memos_for_retrieval):
2330
2174
"""Get the raw bytes for a records.
2331
2175
2332
:param memos_for_retrieval: An iterable containing the (index, pos,
2333
length) memo for retrieving the bytes. The .knit method ignores
2334
the index as there is always only a single file.
2176
:param memos_for_retrieval: An iterable containing the access memo for
2177
retrieving the bytes.
2335
2178
:return: An iterator over the bytes of the records.
2336
2179
"""
2337
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2338
for pos, data in self._transport.readv(self._filename, read_vector):
2339
yield data
2340
2341
2342
class _PackAccess(object):
2343
"""Access to knit records via a collection of packs."""
2344
2345
def __init__(self, index_to_packs, writer=None):
2346
"""Create a _PackAccess object.
2180
# first pass, group into same-index request to minimise readv's issued.
2181
request_lists = []
2182
current_prefix = None
2183
for (key, offset, length) in memos_for_retrieval:
2184
if current_prefix == key[:-1]:
2185
current_list.append((offset, length))
2186
else:
2187
if current_prefix is not None:
2188
request_lists.append((current_prefix, current_list))
2189
current_prefix = key[:-1]
2190
current_list = [(offset, length)]
2191
# handle the last entry
2192
if current_prefix is not None:
2193
request_lists.append((current_prefix, current_list))
2194
for prefix, read_vector in request_lists:
2195
path = self._mapper.map(prefix) + '.knit'
2196
for pos, data in self._transport.readv(path, read_vector):
2197
yield data
2198
2199
2200
class _DirectPackAccess(object):
2201
"""Access to data in one or more packs with less translation."""
2202
2203
def __init__(self, index_to_packs):
2204
"""Create a _DirectPackAccess object.
2347
2205
2348
2206
:param index_to_packs: A dict mapping index objects to the transport
2349
2207
and file names for obtaining data.
2350
:param writer: A tuple (pack.ContainerWriter, write_index) which
2351
contains the pack to write, and the index that reads from it will
2352
be associated with.
2353
2208
"""
2354
if writer:
2355
self.container_writer = writer[0]
2356
self.write_index = writer[1]
2357
else:
2358
self.container_writer = None
2359
self.write_index = None
2360
self.indices = index_to_packs
2209
self._container_writer = None
2210
self._write_index = None
2211
self._indices = index_to_packs
2361
2212
2362
def add_raw_records(self, sizes, raw_data):
2213
def add_raw_records(self, key_sizes, raw_data):
2363
2214
"""Add raw knit bytes to a storage area.
2364
2215
2365
2216
The data is spooled to the container writer in one bytes-record per
2366
2217
raw data item.
2367
2218
2368
:param sizes: An iterable containing the size of each raw data segment.
2219
:param sizes: An iterable of tuples containing the key and size of each
2220
raw data segment.
2369
2221
:param raw_data: A bytestring containing the data.
2370
:return: A list of memos to retrieve the record later. Each memo is a
2371
tuple - (index, pos, length), where the index field is the
2372
write_index object supplied to the PackAccess object.
2222
:return: A list of memos to retrieve the record later. Each memo is an
2223
opaque index memo. For _DirectPackAccess the memo is (index, pos,
2224
length), where the index field is the write_index object supplied
2225
to the PackAccess object.
2373
2226
"""
2227
if type(raw_data) != str:
2228
raise AssertionError(
2229
'data must be plain bytes was %s' % type(raw_data))
2374
2230
result = []
2375
2231
offset = 0
2376
for size in sizes:
2377
p_offset, p_length = self.container_writer.add_bytes_record(
2232
for key, size in key_sizes:
2233
p_offset, p_length = self._container_writer.add_bytes_record(
2378
2234
raw_data[offset:offset+size], [])
2379
2235
offset += size
2380
result.append((self.write_index, p_offset, p_length))
2236
result.append((self._write_index, p_offset, p_length))
2381
2237
return result
2382
2238
2383
def create(self):
2384
"""Pack based knits do not get individually created."""
2385
2386
2239
def get_raw_records(self, memos_for_retrieval):
2387
2240
"""Get the raw bytes for a records.
2388
2241
2407
2260
if current_index is not None:
2408
2261
request_lists.append((current_index, current_list))
2409
2262
for index, offsets in request_lists:
2410
transport, path = self.indices[index]
2263
transport, path = self._indices[index]
2411
2264
reader = pack.make_readv_reader(transport, path, offsets)
2412
2265
for names, read_func in reader.iter_records():
2413
2266
yield read_func(None)
2414
2267
2415
def open_file(self):
2416
"""Pack based knits have no single file."""
2417
return None
2418
2419
def set_writer(self, writer, index, (transport, packname)):
2268
def set_writer(self, writer, index, transport_packname):
2420
2269
"""Set a writer to use for adding data."""
2421
2270
if index is not None:
2422
self.indices[index] = (transport, packname)
2423
self.container_writer = writer
2424
self.write_index = index
2425
2426
2427
class _StreamAccess(object):
2428
"""A Knit Access object that provides data from a datastream.
2429
2430
It also provides a fallback to present as unannotated data, annotated data
2431
from a *backing* access object.
2432
2433
This is triggered by a index_memo which is pointing to a different index
2434
than this was constructed with, and is used to allow extracting full
2435
unannotated texts for insertion into annotated knits.
2436
"""
2437
2438
def __init__(self, reader_callable, stream_index, backing_knit,
2439
orig_factory):
2440
"""Create a _StreamAccess object.
2441
2442
:param reader_callable: The reader_callable from the datastream.
2443
This is called to buffer all the data immediately, for
2444
random access.
2445
:param stream_index: The index the data stream this provides access to
2446
which will be present in native index_memo's.
2447
:param backing_knit: The knit object that will provide access to
2448
annotated texts which are not available in the stream, so as to
2449
create unannotated texts.
2450
:param orig_factory: The original content factory used to generate the
2451
stream. This is used for checking whether the thunk code for
2452
supporting _copy_texts will generate the correct form of data.
2453
"""
2454
self.data = reader_callable(None)
2455
self.stream_index = stream_index
2456
self.backing_knit = backing_knit
2457
self.orig_factory = orig_factory
2458
2459
def get_raw_records(self, memos_for_retrieval):
2460
"""Get the raw bytes for a records.
2461
2462
:param memos_for_retrieval: An iterable of memos from the
2463
_StreamIndex object identifying bytes to read; for these classes
2464
they are (from_backing_knit, index, start, end) and can point to
2465
either the backing knit or streamed data.
2466
:return: An iterator yielding a byte string for each record in
2467
memos_for_retrieval.
2468
"""
2469
# use a generator for memory friendliness
2470
for from_backing_knit, version_id, start, end in memos_for_retrieval:
2471
if not from_backing_knit:
2472
if version_id is not self.stream_index:
2473
raise AssertionError()
2474
yield self.data[start:end]
2475
continue
2476
# we have been asked to thunk. This thunking only occurs when
2477
# we are obtaining plain texts from an annotated backing knit
2478
# so that _copy_texts will work.
2479
# We could improve performance here by scanning for where we need
2480
# to do this and using get_line_list, then interleaving the output
2481
# as desired. However, for now, this is sufficient.
2482
if self.orig_factory.__class__ != KnitPlainFactory:
2483
raise errors.KnitCorrupt(
2484
self, 'Bad thunk request %r cannot be backed by %r' %
2485
(version_id, self.orig_factory))
2486
lines = self.backing_knit.get_lines(version_id)
2487
line_bytes = ''.join(lines)
2488
digest = sha_string(line_bytes)
2489
# the packed form of the fulltext always has a trailing newline,
2490
# even if the actual text does not, unless the file is empty. the
2491
# record options including the noeol flag are passed through by
2492
# _StreamIndex, so this is safe.
2493
if lines:
2494
if lines[-1][-1] != '\n':
2495
lines[-1] = lines[-1] + '\n'
2496
line_bytes += '\n'
2497
# We want plain data, because we expect to thunk only to allow text
2498
# extraction.
2499
size, bytes = self.backing_knit._data._record_to_data(version_id,
2500
digest, lines, line_bytes)
2501
yield bytes
2502
2503
2504
class _StreamIndex(object):
2505
"""A Knit Index object that uses the data map from a datastream."""
2506
2507
def __init__(self, data_list, backing_index):
2508
"""Create a _StreamIndex object.
2509
2510
:param data_list: The data_list from the datastream.
2511
:param backing_index: The index which will supply values for nodes
2512
referenced outside of this stream.
2513
"""
2514
self.data_list = data_list
2515
self.backing_index = backing_index
2516
self._by_version = {}
2517
pos = 0
2518
for key, options, length, parents in data_list:
2519
self._by_version[key] = options, (pos, pos + length), parents
2520
pos += length
2521
2522
def get_ancestry(self, versions, topo_sorted):
2523
"""Get an ancestry list for versions."""
2524
if topo_sorted:
2525
# Not needed for basic joins
2526
raise NotImplementedError(self.get_ancestry)
2527
# get a graph of all the mentioned versions:
2528
# Little ugly - basically copied from KnitIndex, but don't want to
2529
# accidentally incorporate too much of that index's code.
2530
ancestry = set()
2531
pending = set(versions)
2532
cache = self._by_version
2533
while pending:
2534
version = pending.pop()
2535
# trim ghosts
2536
try:
2537
parents = [p for p in cache[version][2] if p in cache]
2538
except KeyError:
2539
raise RevisionNotPresent(version, self)
2540
# if not completed and not a ghost
2541
pending.update([p for p in parents if p not in ancestry])
2542
ancestry.add(version)
2543
return list(ancestry)
2544
2545
def get_build_details(self, version_ids):
2546
"""Get the method, index_memo and compression parent for version_ids.
2547
2548
Ghosts are omitted from the result.
2549
2550
:param version_ids: An iterable of version_ids.
2551
:return: A dict of version_id:(index_memo, compression_parent,
2552
parents, record_details).
2553
index_memo
2554
opaque memo that can be passed to _StreamAccess.read_records
2555
to extract the raw data; for these classes it is
2556
(from_backing_knit, index, start, end)
2557
compression_parent
2558
Content that this record is built upon, may be None
2559
parents
2560
Logical parents of this node
2561
record_details
2562
extra information about the content which needs to be passed to
2563
Factory.parse_record
2564
"""
2565
result = {}
2566
for version_id in version_ids:
2567
try:
2568
method = self.get_method(version_id)
2569
except errors.RevisionNotPresent:
2570
# ghosts are omitted
2571
continue
2572
parent_ids = self.get_parents_with_ghosts(version_id)
2573
noeol = ('no-eol' in self.get_options(version_id))
2574
index_memo = self.get_position(version_id)
2575
from_backing_knit = index_memo[0]
2576
if from_backing_knit:
2577
# texts retrieved from the backing knit are always full texts
2578
method = 'fulltext'
2579
if method == 'fulltext':
2580
compression_parent = None
2581
else:
2582
compression_parent = parent_ids[0]
2583
result[version_id] = (index_memo, compression_parent,
2584
parent_ids, (method, noeol))
2585
return result
2586
2587
def get_method(self, version_id):
2588
"""Return compression method of specified version."""
2589
options = self.get_options(version_id)
2590
if 'fulltext' in options:
2591
return 'fulltext'
2592
elif 'line-delta' in options:
2593
return 'line-delta'
2594
else:
2595
raise errors.KnitIndexUnknownMethod(self, options)
2596
2597
def get_options(self, version_id):
2598
"""Return a list representing options.
2599
2600
e.g. ['foo', 'bar']
2601
"""
2602
try:
2603
return self._by_version[version_id][0]
2604
except KeyError:
2605
options = list(self.backing_index.get_options(version_id))
2606
if 'fulltext' in options:
2607
pass
2608
elif 'line-delta' in options:
2609
# Texts from the backing knit are always returned from the stream
2610
# as full texts
2611
options.remove('line-delta')
2612
options.append('fulltext')
2613
else:
2614
raise errors.KnitIndexUnknownMethod(self, options)
2615
return tuple(options)
2616
2617
def get_parent_map(self, version_ids):
2618
"""Passed through to by KnitVersionedFile.get_parent_map."""
2619
result = {}
2620
pending_ids = set()
2621
for version_id in version_ids:
2622
try:
2623
result[version_id] = self._by_version[version_id][2]
2624
except KeyError:
2625
pending_ids.add(version_id)
2626
result.update(self.backing_index.get_parent_map(pending_ids))
2627
return result
2628
2629
def get_parents_with_ghosts(self, version_id):
2630
"""Return parents of specified version with ghosts."""
2631
try:
2632
return self.get_parent_map([version_id])[version_id]
2633
except KeyError:
2634
raise RevisionNotPresent(version_id, self)
2635
2636
def get_position(self, version_id):
2637
"""Return details needed to access the version.
2638
2639
_StreamAccess has the data as a big array, so we return slice
2640
coordinates into that (as index_memo's are opaque outside the
2641
index and matching access class).
2642
2643
:return: a tuple (from_backing_knit, index, start, end) that can
2644
be passed e.g. to get_raw_records.
2645
If from_backing_knit is False, index will be self, otherwise it
2646
will be a version id.
2647
"""
2648
try:
2649
start, end = self._by_version[version_id][1]
2650
return False, self, start, end
2651
except KeyError:
2652
# Signal to the access object to handle this from the backing knit.
2653
return (True, version_id, None, None)
2654
2655
def get_versions(self):
2656
"""Get all the versions in the stream."""
2657
return self._by_version.keys()
2658
2659
2660
class _KnitData(object):
2661
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2662
2663
The KnitData class provides the logic for parsing and using knit records,
2664
making use of an access method for the low level read and write operations.
2665
"""
2666
2667
def __init__(self, access):
2668
"""Create a KnitData object.
2669
2670
:param access: The access method to use. Access methods such as
2671
_KnitAccess manage the insertion of raw records and the subsequent
2672
retrieval of the same.
2673
"""
2674
self._access = access
2675
self._checked = False
2676
2677
def _open_file(self):
2678
return self._access.open_file()
2679
2680
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2681
"""Convert version_id, digest, lines into a raw data block.
2682
2683
:param dense_lines: The bytes of lines but in a denser form. For
2684
instance, if lines is a list of 1000 bytestrings each ending in \n,
2685
dense_lines may be a list with one line in it, containing all the
2686
1000's lines and their \n's. Using dense_lines if it is already
2687
known is a win because the string join to create bytes in this
2688
function spends less time resizing the final string.
2689
:return: (len, a StringIO instance with the raw data ready to read.)
2690
"""
2691
# Note: using a string copy here increases memory pressure with e.g.
2692
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2693
# when doing the initial commit of a mozilla tree. RBC 20070921
2694
bytes = ''.join(chain(
2695
["version %s %d %s\n" % (version_id,
2696
len(lines),
2697
digest)],
2698
dense_lines or lines,
2699
["end %s\n" % version_id]))
2700
compressed_bytes = bytes_to_gzip(bytes)
2701
return len(compressed_bytes), compressed_bytes
2702
2703
def add_raw_records(self, sizes, raw_data):
2704
"""Append a prepared record to the data file.
2705
2706
:param sizes: An iterable containing the size of each raw data segment.
2707
:param raw_data: A bytestring containing the data.
2708
:return: a list of index data for the way the data was stored.
2709
See the access method add_raw_records documentation for more
2710
details.
2711
"""
2712
return self._access.add_raw_records(sizes, raw_data)
2713
2714
def _parse_record_header(self, version_id, raw_data):
2715
"""Parse a record header for consistency.
2716
2717
:return: the header and the decompressor stream.
2718
as (stream, header_record)
2719
"""
2720
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2721
try:
2722
rec = self._check_header(version_id, df.readline())
2723
except Exception, e:
2724
raise KnitCorrupt(self._access,
2725
"While reading {%s} got %s(%s)"
2726
% (version_id, e.__class__.__name__, str(e)))
2727
return df, rec
2728
2729
def _split_header(self, line):
2730
rec = line.split()
2731
if len(rec) != 4:
2732
raise KnitCorrupt(self._access,
2733
'unexpected number of elements in record header')
2734
return rec
2735
2736
def _check_header_version(self, rec, version_id):
2737
if rec[1] != version_id:
2738
raise KnitCorrupt(self._access,
2739
'unexpected version, wanted %r, got %r'
2740
% (version_id, rec[1]))
2741
2742
def _check_header(self, version_id, line):
2743
rec = self._split_header(line)
2744
self._check_header_version(rec, version_id)
2745
return rec
2746
2747
def _parse_record_unchecked(self, data):
2748
# profiling notes:
2749
# 4168 calls in 2880 217 internal
2750
# 4168 calls to _parse_record_header in 2121
2751
# 4168 calls to readlines in 330
2752
df = GzipFile(mode='rb', fileobj=StringIO(data))
2753
try:
2754
record_contents = df.readlines()
2755
except Exception, e:
2756
raise KnitCorrupt(self._access, "Corrupt compressed record %r, got %s(%s)" %
2757
(data, e.__class__.__name__, str(e)))
2758
header = record_contents.pop(0)
2759
rec = self._split_header(header)
2760
last_line = record_contents.pop()
2761
if len(record_contents) != int(rec[2]):
2762
raise KnitCorrupt(self._access,
2763
'incorrect number of lines %s != %s'
2764
' for version {%s}'
2765
% (len(record_contents), int(rec[2]),
2766
rec[1]))
2767
if last_line != 'end %s\n' % rec[1]:
2768
raise KnitCorrupt(self._access,
2769
'unexpected version end line %r, wanted %r'
2770
% (last_line, rec[1]))
2771
df.close()
2772
return rec, record_contents
2773
2774
def _parse_record(self, version_id, data):
2775
rec, record_contents = self._parse_record_unchecked(data)
2776
self._check_header_version(rec, version_id)
2777
return record_contents, rec[3]
2778
2779
def read_records_iter_raw(self, records):
2780
"""Read text records from data file and yield raw data.
2781
2782
This unpacks enough of the text record to validate the id is
2783
as expected but thats all.
2784
2785
Each item the iterator yields is (version_id, bytes,
2786
sha1_of_full_text).
2787
"""
2788
# setup an iterator of the external records:
2789
# uses readv so nice and fast we hope.
2790
if len(records):
2791
# grab the disk data needed.
2792
needed_offsets = [index_memo for version_id, index_memo
2793
in records]
2794
raw_records = self._access.get_raw_records(needed_offsets)
2795
2796
for version_id, index_memo in records:
2797
data = raw_records.next()
2798
# validate the header
2799
df, rec = self._parse_record_header(version_id, data)
2800
df.close()
2801
yield version_id, data, rec[3]
2802
2803
def read_records_iter(self, records):
2804
"""Read text records from data file and yield result.
2805
2806
The result will be returned in whatever is the fastest to read.
2807
Not by the order requested. Also, multiple requests for the same
2808
record will only yield 1 response.
2809
:param records: A list of (version_id, pos, len) entries
2810
:return: Yields (version_id, contents, digest) in the order
2811
read, not the order requested
2812
"""
2813
if not records:
2814
return
2815
2816
needed_records = sorted(set(records), key=operator.itemgetter(1))
2817
if not needed_records:
2818
return
2819
2820
# The transport optimizes the fetching as well
2821
# (ie, reads continuous ranges.)
2822
raw_data = self._access.get_raw_records(
2823
[index_memo for version_id, index_memo in needed_records])
2824
2825
for (version_id, index_memo), data in \
2826
izip(iter(needed_records), raw_data):
2827
content, digest = self._parse_record(version_id, data)
2828
yield version_id, content, digest
2829
2830
def read_records(self, records):
2831
"""Read records into a dictionary."""
2832
components = {}
2833
for record_id, content, digest in \
2834
self.read_records_iter(records):
2835
components[record_id] = (content, digest)
2836
return components
2837
2838
2839
class InterKnit(InterVersionedFile):
2840
"""Optimised code paths for knit to knit operations."""
2841
2842
_matching_file_from_factory = staticmethod(make_file_knit)
2843
_matching_file_to_factory = staticmethod(make_file_knit)
2844
2845
@staticmethod
2846
def is_compatible(source, target):
2847
"""Be compatible with knits. """
2848
try:
2849
return (isinstance(source, KnitVersionedFile) and
2850
isinstance(target, KnitVersionedFile))
2851
except AttributeError:
2852
return False
2853
2854
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2855
"""Copy texts to the target by extracting and adding them one by one.
2856
2857
see join() for the parameter definitions.
2858
"""
2859
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2860
# --- the below is factorable out with VersionedFile.join, but wait for
2861
# VersionedFiles, it may all be simpler then.
2862
graph = Graph(self.source)
2863
search = graph._make_breadth_first_searcher(version_ids)
2864
transitive_ids = set()
2865
map(transitive_ids.update, list(search))
2866
parent_map = self.source.get_parent_map(transitive_ids)
2867
order = topo_sort(parent_map.items())
2868
2869
def size_of_content(content):
2870
return sum(len(line) for line in content.text())
2871
# Cache at most 10MB of parent texts
2872
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2873
compute_size=size_of_content)
2874
# TODO: jam 20071116 It would be nice to have a streaming interface to
2875
# get multiple texts from a source. The source could be smarter
2876
# about how it handled intermediate stages.
2877
# get_line_list() or make_mpdiffs() seem like a possibility, but
2878
# at the moment they extract all full texts into memory, which
2879
# causes us to store more than our 3x fulltext goal.
2880
# Repository.iter_files_bytes() may be another possibility
2881
to_process = [version for version in order
2882
if version not in self.target]
2883
total = len(to_process)
2884
pb = ui.ui_factory.nested_progress_bar()
2885
try:
2886
for index, version in enumerate(to_process):
2887
pb.update('Converting versioned data', index, total)
2888
sha1, num_bytes, parent_text = self.target.add_lines(version,
2889
self.source.get_parents_with_ghosts(version),
2890
self.source.get_lines(version),
2891
parent_texts=parent_cache)
2892
parent_cache[version] = parent_text
2893
finally:
2894
pb.finished()
2895
return total
2896
2897
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2898
"""See InterVersionedFile.join."""
2899
# If the source and target are mismatched w.r.t. annotations vs
2900
# plain, the data needs to be converted accordingly
2901
if self.source.factory.annotated == self.target.factory.annotated:
2902
converter = None
2903
elif self.source.factory.annotated:
2904
converter = self._anno_to_plain_converter
2905
else:
2906
# We're converting from a plain to an annotated knit. Copy them
2907
# across by full texts.
2908
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2909
2910
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2911
if not version_ids:
2912
return 0
2913
2914
pb = ui.ui_factory.nested_progress_bar()
2915
try:
2916
version_ids = list(version_ids)
2917
if None in version_ids:
2918
version_ids.remove(None)
2919
2920
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2921
topo_sorted=False))
2922
this_versions = set(self.target._index.get_versions())
2923
# XXX: For efficiency we should not look at the whole index,
2924
# we only need to consider the referenced revisions - they
2925
# must all be present, or the method must be full-text.
2926
# TODO, RBC 20070919
2927
needed_versions = self.source_ancestry - this_versions
2928
2929
if not needed_versions:
2930
return 0
2931
full_list = topo_sort(
2932
self.source.get_parent_map(self.source.versions()))
2933
2934
version_list = [i for i in full_list if (not self.target.has_version(i)
2935
and i in needed_versions)]
2936
2937
# plan the join:
2938
copy_queue = []
2939
copy_queue_records = []
2940
copy_set = set()
2941
for version_id in version_list:
2942
options = self.source._index.get_options(version_id)
2943
parents = self.source._index.get_parents_with_ghosts(version_id)
2944
# check that its will be a consistent copy:
2945
for parent in parents:
2946
# if source has the parent, we must :
2947
# * already have it or
2948
# * have it scheduled already
2949
# otherwise we don't care
2950
if not (self.target.has_version(parent) or
2951
parent in copy_set or
2952
not self.source.has_version(parent)):
2953
raise AssertionError("problem joining parent %r "
2954
"from %r to %r"
2955
% (parent, self.source, self.target))
2956
index_memo = self.source._index.get_position(version_id)
2957
copy_queue_records.append((version_id, index_memo))
2958
copy_queue.append((version_id, options, parents))
2959
copy_set.add(version_id)
2960
2961
# data suck the join:
2962
count = 0
2963
total = len(version_list)
2964
raw_datum = []
2965
raw_records = []
2966
for (version_id, raw_data, _), \
2967
(version_id2, options, parents) in \
2968
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2969
copy_queue):
2970
if not (version_id == version_id2):
2971
raise AssertionError('logic error, inconsistent results')
2972
count = count + 1
2973
pb.update("Joining knit", count, total)
2974
if converter:
2975
size, raw_data = converter(raw_data, version_id, options,
2976
parents)
2977
else:
2978
size = len(raw_data)
2979
raw_records.append((version_id, options, parents, size))
2980
raw_datum.append(raw_data)
2981
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2982
return count
2983
finally:
2984
pb.finished()
2985
2986
def _anno_to_plain_converter(self, raw_data, version_id, options,
2987
parents):
2988
"""Convert annotated content to plain content."""
2989
data, digest = self.source._data._parse_record(version_id, raw_data)
2990
if 'fulltext' in options:
2991
content = self.source.factory.parse_fulltext(data, version_id)
2992
lines = self.target.factory.lower_fulltext(content)
2993
else:
2994
delta = self.source.factory.parse_line_delta(data, version_id,
2995
plain=True)
2996
lines = self.target.factory.lower_line_delta(delta)
2997
return self.target._data._record_to_data(version_id, digest, lines)
2998
2999
3000
InterVersionedFile.register_optimiser(InterKnit)
3001
3002
3003
class WeaveToKnit(InterVersionedFile):
3004
"""Optimised code paths for weave to knit operations."""
3005
3006
_matching_file_from_factory = bzrlib.weave.WeaveFile
3007
_matching_file_to_factory = staticmethod(make_file_knit)
3008
3009
@staticmethod
3010
def is_compatible(source, target):
3011
"""Be compatible with weaves to knits."""
3012
try:
3013
return (isinstance(source, bzrlib.weave.Weave) and
3014
isinstance(target, KnitVersionedFile))
3015
except AttributeError:
3016
return False
3017
3018
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
3019
"""See InterVersionedFile.join."""
3020
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
3021
3022
if not version_ids:
3023
return 0
3024
3025
pb = ui.ui_factory.nested_progress_bar()
3026
try:
3027
version_ids = list(version_ids)
3028
3029
self.source_ancestry = set(self.source.get_ancestry(version_ids))
3030
this_versions = set(self.target._index.get_versions())
3031
needed_versions = self.source_ancestry - this_versions
3032
3033
if not needed_versions:
3034
return 0
3035
full_list = topo_sort(
3036
self.source.get_parent_map(self.source.versions()))
3037
3038
version_list = [i for i in full_list if (not self.target.has_version(i)
3039
and i in needed_versions)]
3040
3041
# do the join:
3042
count = 0
3043
total = len(version_list)
3044
parent_map = self.source.get_parent_map(version_list)
3045
for version_id in version_list:
3046
pb.update("Converting to knit", count, total)
3047
parents = parent_map[version_id]
3048
# check that its will be a consistent copy:
3049
for parent in parents:
3050
# if source has the parent, we must already have it
3051
if not self.target.has_version(parent):
3052
raise AssertionError("%r does not have parent %r"
3053
% (self.target, parent))
3054
self.target.add_lines(
3055
version_id, parents, self.source.get_lines(version_id))
3056
count = count + 1
3057
return count
3058
finally:
3059
pb.finished()
3060
3061
3062
InterVersionedFile.register_optimiser(WeaveToKnit)
2271
self._indices[index] = transport_packname
2272
self._container_writer = writer
2273
self._write_index = index
3063
2274
3064
2275
3065
2276
# Deprecated, use PatienceSequenceMatcher instead
3172
2383
self._check_parents(child, nodes_to_annotate)
3173
2384
return nodes_to_annotate
3174
2385
3175
def _get_build_graph(self, revision_id):
2386
def _get_build_graph(self, key):
3176
2387
"""Get the graphs for building texts and annotations.
3177
2388
3178
2389
The data you need for creating a full text may be different than the
3180
2391
parents to create an annotation, but only need 1 parent to generate the
3181
2392
fulltext.)
3182
2393
3183
:return: A list of (revision_id, index_memo) records, suitable for
2394
:return: A list of (key, index_memo) records, suitable for
3184
2395
passing to read_records_iter to start reading in the raw data fro/
3185
2396
the pack file.
3186
2397
"""
3187
if revision_id in self._annotated_lines:
2398
if key in self._annotated_lines:
3188
2399
# Nothing to do
3189
2400
return []
3190
pending = set([revision_id])
2401
pending = set([key])
3191
2402
records = []
3192
2403
generation = 0
3193
2404
kept_generation = 0
3199
2410
self._all_build_details.update(build_details)
3200
2411
# new_nodes = self._knit._index._get_entries(this_iteration)
3201
2412
pending = set()
3202
for rev_id, details in build_details.iteritems():
2413
for key, details in build_details.iteritems():
3203
2414
(index_memo, compression_parent, parents,
3204
2415
record_details) = details
3205
self._revision_id_graph[rev_id] = parents
3206
records.append((rev_id, index_memo))
2416
self._revision_id_graph[key] = parents
2417
records.append((key, index_memo))
3207
2418
# Do we actually need to check _annotated_lines?
3208
2419
pending.update(p for p in parents
3209
2420
if p not in self._all_build_details)
3210
2421
if compression_parent:
3211
2422
self._compression_children.setdefault(compression_parent,
3212
[]).append(rev_id)
2423
[]).append(key)
3213
2424
if parents:
3214
2425
for parent in parents:
3215
2426
self._annotate_children.setdefault(parent,
3216
[]).append(rev_id)
2427
[]).append(key)
3217
2428
num_gens = generation - kept_generation
3218
2429
if ((num_gens >= self._generations_until_keep)
3219
2430
and len(parents) > 1):
3220
2431
kept_generation = generation
3221
self._nodes_to_keep_annotations.add(rev_id)
2432
self._nodes_to_keep_annotations.add(key)
3222
2433
3223
2434
missing_versions = this_iteration.difference(build_details.keys())
3224
2435
self._ghosts.update(missing_versions)
3250
2461
# still need parents, cleaning them up when those parents are
3251
2462
# processed.
3252
2463
for (rev_id, record,
3253
digest) in self._knit._data.read_records_iter(records):
2464
digest) in self._knit._read_records_iter(records):
3254
2465
if rev_id in self._annotated_lines:
3255
2466
continue
3256
2467
parent_ids = self._revision_id_graph[rev_id]
3265
2476
if len(parent_ids) == 0:
3266
2477
# There are no parents for this node, so just add it
3267
2478
# TODO: This probably needs to be decoupled
3268
fulltext_content, delta = self._knit.factory.parse_record(
2479
fulltext_content, delta = self._knit._factory.parse_record(
3269
2480
rev_id, record, record_details, None)
3270
2481
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3271
2482
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3299
2510
parent_fulltext_content = self._fulltext_contents[compression_parent]
3300
2511
parent_fulltext = parent_fulltext_content.text()
3301
2512
comp_children.remove(rev_id)
3302
fulltext_content, delta = self._knit.factory.parse_record(
2513
fulltext_content, delta = self._knit._factory.parse_record(
3303
2514
rev_id, record, record_details,
3304
2515
parent_fulltext_content,
3305
2516
copy_base_content=(not reuse_content))
3308
2519
blocks = KnitContent.get_line_delta_blocks(delta,
3309
2520
parent_fulltext, fulltext)
3310
2521
else:
3311
fulltext_content = self._knit.factory.parse_fulltext(
2522
fulltext_content = self._knit._factory.parse_fulltext(
3312
2523
record, rev_id)
3313
2524
fulltext = self._add_fulltext_content(rev_id,
3314
2525
fulltext_content)
3328
2539
self._heads_provider = head_cache
3329
2540
return head_cache
3330
2541
3331
def annotate(self, revision_id):
3332
"""Return the annotated fulltext at the given revision.
2542
def annotate(self, key):
2543
"""Return the annotated fulltext at the given key.
3333
2544
3334
:param revision_id: The revision id for this file
2545
:param key: The key to annotate.
3335
2546
"""
3336
records = self._get_build_graph(revision_id)
3337
if revision_id in self._ghosts:
3338
raise errors.RevisionNotPresent(revision_id, self._knit)
2547
records = self._get_build_graph(key)
2548
if key in self._ghosts:
2549
raise errors.RevisionNotPresent(key, self._knit)
3339
2550
self._annotate_records(records)
3340
return self._annotated_lines[revision_id]
2551
return self._annotated_lines[key]
3341
2552
3342
2553
3343
2554
try:
Loggerhead is a web-based interface for Breezy
Version: 2.0.1