276
268
annotated_kind = ''
277
269
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
278
270
self._raw_record = raw_record
279
self._network_bytes = network_bytes
280
271
self._build_details = build_details
281
272
self._knit = knit
283
def _create_network_bytes(self):
284
"""Create a fully serialised network version for transmission."""
285
# storage_kind, key, parents, Noeol, raw_record
286
key_bytes = '\x00'.join(self.key)
287
if self.parents is None:
288
parent_bytes = 'None:'
290
parent_bytes = '\t'.join('\x00'.join(key) for key in self.parents)
291
if self._build_details[1]:
295
network_bytes = "%s\n%s\n%s\n%s%s" % (self.storage_kind, key_bytes,
296
parent_bytes, noeol, self._raw_record)
297
self._network_bytes = network_bytes
299
def get_bytes_as(self, storage_kind):
300
if storage_kind == self.storage_kind:
301
if self._network_bytes is None:
302
self._create_network_bytes()
303
return self._network_bytes
304
if ('-ft-' in self.storage_kind and
305
storage_kind in ('chunked', 'fulltext')):
306
adapter_key = (self.storage_kind, 'fulltext')
307
adapter_factory = adapter_registry.get(adapter_key)
308
adapter = adapter_factory(None)
309
bytes = adapter.get_bytes(self)
310
if storage_kind == 'chunked':
314
if self._knit is not None:
315
# Not redundant with direct conversion above - that only handles
317
if storage_kind == 'chunked':
318
return self._knit.get_lines(self.key[0])
319
elif storage_kind == 'fulltext':
320
return self._knit.get_text(self.key[0])
321
raise errors.UnavailableRepresentation(self.key, storage_kind,
325
class LazyKnitContentFactory(ContentFactory):
326
"""A ContentFactory which can either generate full text or a wire form.
328
:seealso ContentFactory:
331
def __init__(self, key, parents, generator, first):
332
"""Create a LazyKnitContentFactory.
334
:param key: The key of the record.
335
:param parents: The parents of the record.
336
:param generator: A _ContentMapGenerator containing the record for this
338
:param first: Is this the first content object returned from generator?
339
if it is, its storage kind is knit-delta-closure, otherwise it is
340
knit-delta-closure-ref
343
self.parents = parents
345
self._generator = generator
346
self.storage_kind = "knit-delta-closure"
348
self.storage_kind = self.storage_kind + "-ref"
351
def get_bytes_as(self, storage_kind):
352
if storage_kind == self.storage_kind:
354
return self._generator._wire_bytes()
356
# all the keys etc are contained in the bytes returned in the
359
if storage_kind in ('chunked', 'fulltext'):
360
chunks = self._generator._get_one_work(self.key).text()
361
if storage_kind == 'chunked':
364
return ''.join(chunks)
365
raise errors.UnavailableRepresentation(self.key, storage_kind,
369
def knit_delta_closure_to_records(storage_kind, bytes, line_end):
370
"""Convert a network record to a iterator over stream records.
372
:param storage_kind: The storage kind of the record.
373
Must be 'knit-delta-closure'.
374
:param bytes: The bytes of the record on the network.
376
generator = _NetworkContentMapGenerator(bytes, line_end)
377
return generator.get_record_stream()
380
def knit_network_to_record(storage_kind, bytes, line_end):
381
"""Convert a network record to a record object.
383
:param storage_kind: The storage kind of the record.
384
:param bytes: The bytes of the record on the network.
387
line_end = bytes.find('\n', start)
388
key = tuple(bytes[start:line_end].split('\x00'))
390
line_end = bytes.find('\n', start)
391
parent_line = bytes[start:line_end]
392
if parent_line == 'None:':
396
[tuple(segment.split('\x00')) for segment in parent_line.split('\t')
399
noeol = bytes[start] == 'N'
400
if 'ft' in storage_kind:
403
method = 'line-delta'
404
build_details = (method, noeol)
406
raw_record = bytes[start:]
407
annotated = 'annotated' in storage_kind
408
return [KnitContentFactory(key, parents, build_details, None, raw_record,
409
annotated, network_bytes=bytes)]
274
def get_bytes_as(self, storage_kind):
275
if storage_kind == self.storage_kind:
276
return self._raw_record
277
if storage_kind == 'fulltext' and self._knit is not None:
278
return self._knit.get_text(self.key[0])
280
raise errors.UnavailableRepresentation(self.key, storage_kind,
412
284
class KnitContent(object):
413
285
"""Content of a knit version to which deltas can be applied.
415
287
This is always stored in memory as a list of lines with \n at the end,
416
plus a flag saying if the final ending is really there or not, because that
288
plus a flag saying if the final ending is really there or not, because that
417
289
corresponds to the on-disk knit representation.
1189
961
if not self.get_parent_map([key]):
1190
962
raise RevisionNotPresent(key, self)
1191
963
return cached_version
1192
generator = _VFContentMapGenerator(self, [key])
1193
return generator._get_content(key)
1195
def get_known_graph_ancestry(self, keys):
1196
"""Get a KnownGraph instance with the ancestry of keys."""
1197
parent_map, missing_keys = self._index.find_ancestry(keys)
1198
for fallback in self._fallback_vfs:
964
text_map, contents_map = self._get_content_maps([key])
965
return contents_map[key]
967
def _get_content_maps(self, keys, nonlocal_keys=None):
968
"""Produce maps of text and KnitContents
970
:param keys: The keys to produce content maps for.
971
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
972
which are known to not be in this knit, but rather in one of the
974
:return: (text_map, content_map) where text_map contains the texts for
975
the requested versions and content_map contains the KnitContents.
977
# FUTURE: This function could be improved for the 'extract many' case
978
# by tracking each component and only doing the copy when the number of
979
# children than need to apply delta's to it is > 1 or it is part of the
982
multiple_versions = len(keys) != 1
983
record_map = self._get_record_map(keys, allow_missing=True)
988
if nonlocal_keys is None:
989
nonlocal_keys = set()
991
nonlocal_keys = frozenset(nonlocal_keys)
992
missing_keys = set(nonlocal_keys)
993
for source in self._fallback_vfs:
1199
994
if not missing_keys:
1201
(f_parent_map, f_missing_keys) = fallback._index.find_ancestry(
1203
parent_map.update(f_parent_map)
1204
missing_keys = f_missing_keys
1205
kg = _mod_graph.KnownGraph(parent_map)
996
for record in source.get_record_stream(missing_keys,
998
if record.storage_kind == 'absent':
1000
missing_keys.remove(record.key)
1001
lines = split_lines(record.get_bytes_as('fulltext'))
1002
text_map[record.key] = lines
1003
content_map[record.key] = PlainKnitContent(lines, record.key)
1004
if record.key in keys:
1005
final_content[record.key] = content_map[record.key]
1007
if key in nonlocal_keys:
1012
while cursor is not None:
1014
record, record_details, digest, next = record_map[cursor]
1016
raise RevisionNotPresent(cursor, self)
1017
components.append((cursor, record, record_details, digest))
1019
if cursor in content_map:
1020
# no need to plan further back
1021
components.append((cursor, None, None, None))
1025
for (component_id, record, record_details,
1026
digest) in reversed(components):
1027
if component_id in content_map:
1028
content = content_map[component_id]
1030
content, delta = self._factory.parse_record(key[-1],
1031
record, record_details, content,
1032
copy_base_content=multiple_versions)
1033
if multiple_versions:
1034
content_map[component_id] = content
1036
final_content[key] = content
1038
# digest here is the digest from the last applied component.
1039
text = content.text()
1040
actual_sha = sha_strings(text)
1041
if actual_sha != digest:
1042
raise KnitCorrupt(self,
1044
'\n of reconstructed text does not match'
1046
'\n for version %s' %
1047
(actual_sha, digest, key))
1048
text_map[key] = text
1049
return text_map, final_content
1208
1051
def get_parent_map(self, keys):
1209
1052
"""Get a map of the graph parents of keys.
1251
1094
build-parent of the version, i.e. the leftmost ancestor.
1252
1095
Will be None if the record is not a delta.
1253
1096
:param keys: The keys to build a map for
1254
:param allow_missing: If some records are missing, rather than
1097
:param allow_missing: If some records are missing, rather than
1255
1098
error, just return the data that could be generated.
1257
raw_map = self._get_record_map_unparsed(keys,
1100
position_map = self._get_components_positions(keys,
1258
1101
allow_missing=allow_missing)
1259
return self._raw_map_to_record_map(raw_map)
1261
def _raw_map_to_record_map(self, raw_map):
1262
"""Parse the contents of _get_record_map_unparsed.
1264
:return: see _get_record_map.
1268
data, record_details, next = raw_map[key]
1269
content, digest = self._parse_record(key[-1], data)
1270
result[key] = content, record_details, digest, next
1273
def _get_record_map_unparsed(self, keys, allow_missing=False):
1274
"""Get the raw data for reconstructing keys without parsing it.
1276
:return: A dict suitable for parsing via _raw_map_to_record_map.
1277
key-> raw_bytes, (method, noeol), compression_parent
1279
# This retries the whole request if anything fails. Potentially we
1280
# could be a bit more selective. We could track the keys whose records
1281
# we have successfully found, and then only request the new records
1282
# from there. However, _get_components_positions grabs the whole build
1283
# chain, which means we'll likely try to grab the same records again
1284
# anyway. Also, can the build chains change as part of a pack
1285
# operation? We wouldn't want to end up with a broken chain.
1288
position_map = self._get_components_positions(keys,
1289
allow_missing=allow_missing)
1290
# key = component_id, r = record_details, i_m = index_memo,
1292
records = [(key, i_m) for key, (r, i_m, n)
1293
in position_map.iteritems()]
1294
# Sort by the index memo, so that we request records from the
1295
# same pack file together, and in forward-sorted order
1296
records.sort(key=operator.itemgetter(1))
1298
for key, data in self._read_records_iter_unchecked(records):
1299
(record_details, index_memo, next) = position_map[key]
1300
raw_record_map[key] = data, record_details, next
1301
return raw_record_map
1302
except errors.RetryWithNewPacks, e:
1303
self._access.reload_or_raise(e)
1306
def _split_by_prefix(cls, keys):
1307
"""For the given keys, split them up based on their prefix.
1309
To keep memory pressure somewhat under control, split the
1310
requests back into per-file-id requests, otherwise "bzr co"
1311
extracts the full tree into memory before writing it to disk.
1312
This should be revisited if _get_content_maps() can ever cross
1315
The keys for a given file_id are kept in the same relative order.
1316
Ordering between file_ids is not, though prefix_order will return the
1317
order that the key was first seen.
1319
:param keys: An iterable of key tuples
1320
:return: (split_map, prefix_order)
1321
split_map A dictionary mapping prefix => keys
1322
prefix_order The order that we saw the various prefixes
1324
split_by_prefix = {}
1332
if prefix in split_by_prefix:
1333
split_by_prefix[prefix].append(key)
1335
split_by_prefix[prefix] = [key]
1336
prefix_order.append(prefix)
1337
return split_by_prefix, prefix_order
1339
def _group_keys_for_io(self, keys, non_local_keys, positions,
1340
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1341
"""For the given keys, group them into 'best-sized' requests.
1343
The idea is to avoid making 1 request per file, but to never try to
1344
unpack an entire 1.5GB source tree in a single pass. Also when
1345
possible, we should try to group requests to the same pack file
1348
:return: list of (keys, non_local) tuples that indicate what keys
1349
should be fetched next.
1351
# TODO: Ideally we would group on 2 factors. We want to extract texts
1352
# from the same pack file together, and we want to extract all
1353
# the texts for a given build-chain together. Ultimately it
1354
# probably needs a better global view.
1355
total_keys = len(keys)
1356
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1357
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1359
cur_non_local = set()
1363
for prefix in prefix_order:
1364
keys = prefix_split_keys[prefix]
1365
non_local = prefix_split_non_local_keys.get(prefix, [])
1367
this_size = self._index._get_total_build_size(keys, positions)
1368
cur_size += this_size
1369
cur_keys.extend(keys)
1370
cur_non_local.update(non_local)
1371
if cur_size > _min_buffer_size:
1372
result.append((cur_keys, cur_non_local))
1373
sizes.append(cur_size)
1375
cur_non_local = set()
1378
result.append((cur_keys, cur_non_local))
1379
sizes.append(cur_size)
1102
# key = component_id, r = record_details, i_m = index_memo, n = next
1103
records = [(key, i_m) for key, (r, i_m, n)
1104
in position_map.iteritems()]
1106
for key, record, digest in \
1107
self._read_records_iter(records):
1108
(record_details, index_memo, next) = position_map[key]
1109
record_map[key] = record, record_details, digest, next
1382
1112
def get_record_stream(self, keys, ordering, include_delta_closure):
1383
1113
"""Get a stream of records for keys.
1657
1314
access_memo = self._access.add_raw_records(
1658
1315
[(record.key, len(bytes))], bytes)[0]
1659
1316
index_entry = (record.key, options, access_memo, parents)
1660
1318
if 'fulltext' not in options:
1661
# Not a fulltext, so we need to make sure the compression
1662
# parent will also be present.
1319
basis_parent = parents[0]
1663
1320
# Note that pack backed knits don't need to buffer here
1664
1321
# because they buffer all writes to the transaction level,
1665
1322
# but we don't expose that difference at the index level. If
1666
1323
# the query here has sufficient cost to show up in
1667
1324
# profiling we should do that.
1669
# They're required to be physically in this
1670
# KnitVersionedFiles, not in a fallback.
1671
if not self._index.has_key(compression_parent):
1325
if basis_parent not in self.get_parent_map([basis_parent]):
1672
1326
pending = buffered_index_entries.setdefault(
1673
compression_parent, [])
1674
1328
pending.append(index_entry)
1675
1329
buffered = True
1676
1330
if not buffered:
1677
1331
self._index.add_records([index_entry])
1678
elif record.storage_kind == 'chunked':
1332
elif record.storage_kind == 'fulltext':
1679
1333
self.add_lines(record.key, parents,
1680
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1334
split_lines(record.get_bytes_as('fulltext')))
1682
# Not suitable for direct insertion as a
1683
# delta, either because it's not the right format, or this
1684
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1685
# 0) or because it depends on a base only present in the
1687
self._access.flush()
1689
# Try getting a fulltext directly from the record.
1690
bytes = record.get_bytes_as('fulltext')
1691
except errors.UnavailableRepresentation:
1692
adapter_key = record.storage_kind, 'fulltext'
1693
adapter = get_adapter(adapter_key)
1694
bytes = adapter.get_bytes(record)
1695
lines = split_lines(bytes)
1336
adapter_key = record.storage_kind, 'fulltext'
1337
adapter = get_adapter(adapter_key)
1338
lines = split_lines(adapter.get_bytes(
1339
record, record.get_bytes_as(record.storage_kind)))
1697
1341
self.add_lines(record.key, parents, lines)
1698
1342
except errors.RevisionAlreadyPresent:
1700
1344
# Add any records whose basis parent is now available.
1702
added_keys = [record.key]
1704
key = added_keys.pop(0)
1705
if key in buffered_index_entries:
1706
index_entries = buffered_index_entries[key]
1707
self._index.add_records(index_entries)
1709
[index_entry[0] for index_entry in index_entries])
1710
del buffered_index_entries[key]
1345
added_keys = [record.key]
1347
key = added_keys.pop(0)
1348
if key in buffered_index_entries:
1349
index_entries = buffered_index_entries[key]
1350
self._index.add_records(index_entries)
1352
[index_entry[0] for index_entry in index_entries])
1353
del buffered_index_entries[key]
1354
# If there were any deltas which had a missing basis parent, error.
1711
1355
if buffered_index_entries:
1712
# There were index entries buffered at the end of the stream,
1713
# So these need to be added (if the index supports holding such
1714
# entries for later insertion)
1716
for key in buffered_index_entries:
1717
index_entries = buffered_index_entries[key]
1718
all_entries.extend(index_entries)
1719
self._index.add_records(
1720
all_entries, missing_compression_parents=True)
1722
def get_missing_compression_parent_keys(self):
1723
"""Return an iterable of keys of missing compression parents.
1725
Check this after calling insert_record_stream to find out if there are
1726
any missing compression parents. If there are, the records that
1727
depend on them are not able to be inserted safely. For atomic
1728
KnitVersionedFiles built on packs, the transaction should be aborted or
1729
suspended - commit will fail at this point. Nonatomic knits will error
1730
earlier because they have no staging area to put pending entries into.
1732
return self._index.get_missing_compression_parents()
1356
raise errors.RevisionNotPresent(buffered_index_entries.keys()[0],
1734
1359
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1735
1360
"""Iterate over the lines in the versioned files from keys.
2025
class _ContentMapGenerator(object):
2026
"""Generate texts or expose raw deltas for a set of texts."""
2028
def __init__(self, ordering='unordered'):
2029
self._ordering = ordering
2031
def _get_content(self, key):
2032
"""Get the content object for key."""
2033
# Note that _get_content is only called when the _ContentMapGenerator
2034
# has been constructed with just one key requested for reconstruction.
2035
if key in self.nonlocal_keys:
2036
record = self.get_record_stream().next()
2037
# Create a content object on the fly
2038
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2039
return PlainKnitContent(lines, record.key)
2041
# local keys we can ask for directly
2042
return self._get_one_work(key)
2044
def get_record_stream(self):
2045
"""Get a record stream for the keys requested during __init__."""
2046
for record in self._work():
2050
"""Produce maps of text and KnitContents as dicts.
2052
:return: (text_map, content_map) where text_map contains the texts for
2053
the requested versions and content_map contains the KnitContents.
2055
# NB: By definition we never need to read remote sources unless texts
2056
# are requested from them: we don't delta across stores - and we
2057
# explicitly do not want to to prevent data loss situations.
2058
if self.global_map is None:
2059
self.global_map = self.vf.get_parent_map(self.keys)
2060
nonlocal_keys = self.nonlocal_keys
2062
missing_keys = set(nonlocal_keys)
2063
# Read from remote versioned file instances and provide to our caller.
2064
for source in self.vf._fallback_vfs:
2065
if not missing_keys:
2067
# Loop over fallback repositories asking them for texts - ignore
2068
# any missing from a particular fallback.
2069
for record in source.get_record_stream(missing_keys,
2070
self._ordering, True):
2071
if record.storage_kind == 'absent':
2072
# Not in thie particular stream, may be in one of the
2073
# other fallback vfs objects.
2075
missing_keys.remove(record.key)
2078
if self._raw_record_map is None:
2079
raise AssertionError('_raw_record_map should have been filled')
2081
for key in self.keys:
2082
if key in self.nonlocal_keys:
2084
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2087
def _get_one_work(self, requested_key):
2088
# Now, if we have calculated everything already, just return the
2090
if requested_key in self._contents_map:
2091
return self._contents_map[requested_key]
2092
# To simplify things, parse everything at once - code that wants one text
2093
# probably wants them all.
2094
# FUTURE: This function could be improved for the 'extract many' case
2095
# by tracking each component and only doing the copy when the number of
2096
# children than need to apply delta's to it is > 1 or it is part of the
2098
multiple_versions = len(self.keys) != 1
2099
if self._record_map is None:
2100
self._record_map = self.vf._raw_map_to_record_map(
2101
self._raw_record_map)
2102
record_map = self._record_map
2103
# raw_record_map is key:
2104
# Have read and parsed records at this point.
2105
for key in self.keys:
2106
if key in self.nonlocal_keys:
2111
while cursor is not None:
2113
record, record_details, digest, next = record_map[cursor]
2115
raise RevisionNotPresent(cursor, self)
2116
components.append((cursor, record, record_details, digest))
2118
if cursor in self._contents_map:
2119
# no need to plan further back
2120
components.append((cursor, None, None, None))
2124
for (component_id, record, record_details,
2125
digest) in reversed(components):
2126
if component_id in self._contents_map:
2127
content = self._contents_map[component_id]
2129
content, delta = self._factory.parse_record(key[-1],
2130
record, record_details, content,
2131
copy_base_content=multiple_versions)
2132
if multiple_versions:
2133
self._contents_map[component_id] = content
2135
# digest here is the digest from the last applied component.
2136
text = content.text()
2137
actual_sha = sha_strings(text)
2138
if actual_sha != digest:
2139
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2140
if multiple_versions:
2141
return self._contents_map[requested_key]
2145
def _wire_bytes(self):
2146
"""Get the bytes to put on the wire for 'key'.
2148
The first collection of bytes asked for returns the serialised
2149
raw_record_map and the additional details (key, parent) for key.
2150
Subsequent calls return just the additional details (key, parent).
2151
The wire storage_kind given for the first key is 'knit-delta-closure',
2152
For subsequent keys it is 'knit-delta-closure-ref'.
2154
:param key: A key from the content generator.
2155
:return: Bytes to put on the wire.
2158
# kind marker for dispatch on the far side,
2159
lines.append('knit-delta-closure')
2161
if self.vf._factory.annotated:
2162
lines.append('annotated')
2165
# then the list of keys
2166
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2167
if key not in self.nonlocal_keys]))
2168
# then the _raw_record_map in serialised form:
2170
# for each item in the map:
2172
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2173
# one line with method
2174
# one line with noeol
2175
# one line with next ('' for None)
2176
# one line with byte count of the record bytes
2178
for key, (record_bytes, (method, noeol), next) in \
2179
self._raw_record_map.iteritems():
2180
key_bytes = '\x00'.join(key)
2181
parents = self.global_map.get(key, None)
2183
parent_bytes = 'None:'
2185
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2186
method_bytes = method
2192
next_bytes = '\x00'.join(next)
2195
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2196
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2197
len(record_bytes), record_bytes))
2198
map_bytes = ''.join(map_byte_list)
2199
lines.append(map_bytes)
2200
bytes = '\n'.join(lines)
2204
class _VFContentMapGenerator(_ContentMapGenerator):
2205
"""Content map generator reading from a VersionedFiles object."""
2207
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2208
global_map=None, raw_record_map=None, ordering='unordered'):
2209
"""Create a _ContentMapGenerator.
2211
:param versioned_files: The versioned files that the texts are being
2213
:param keys: The keys to produce content maps for.
2214
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2215
which are known to not be in this knit, but rather in one of the
2217
:param global_map: The result of get_parent_map(keys) (or a supermap).
2218
This is required if get_record_stream() is to be used.
2219
:param raw_record_map: A unparsed raw record map to use for answering
2222
_ContentMapGenerator.__init__(self, ordering=ordering)
2223
# The vf to source data from
2224
self.vf = versioned_files
2226
self.keys = list(keys)
2227
# Keys known to be in fallback vfs objects
2228
if nonlocal_keys is None:
2229
self.nonlocal_keys = set()
2231
self.nonlocal_keys = frozenset(nonlocal_keys)
2232
# Parents data for keys to be returned in get_record_stream
2233
self.global_map = global_map
2234
# The chunked lists for self.keys in text form
2236
# A cache of KnitContent objects used in extracting texts.
2237
self._contents_map = {}
2238
# All the knit records needed to assemble the requested keys as full
2240
self._record_map = None
2241
if raw_record_map is None:
2242
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2245
self._raw_record_map = raw_record_map
2246
# the factory for parsing records
2247
self._factory = self.vf._factory
2250
class _NetworkContentMapGenerator(_ContentMapGenerator):
2251
"""Content map generator sourced from a network stream."""
2253
def __init__(self, bytes, line_end):
2254
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2256
self.global_map = {}
2257
self._raw_record_map = {}
2258
self._contents_map = {}
2259
self._record_map = None
2260
self.nonlocal_keys = []
2261
# Get access to record parsing facilities
2262
self.vf = KnitVersionedFiles(None, None)
2265
line_end = bytes.find('\n', start)
2266
line = bytes[start:line_end]
2267
start = line_end + 1
2268
if line == 'annotated':
2269
self._factory = KnitAnnotateFactory()
2271
self._factory = KnitPlainFactory()
2272
# list of keys to emit in get_record_stream
2273
line_end = bytes.find('\n', start)
2274
line = bytes[start:line_end]
2275
start = line_end + 1
2277
tuple(segment.split('\x00')) for segment in line.split('\t')
2279
# now a loop until the end. XXX: It would be nice if this was just a
2280
# bunch of the same records as get_record_stream(..., False) gives, but
2281
# there is a decent sized gap stopping that at the moment.
2285
line_end = bytes.find('\n', start)
2286
key = tuple(bytes[start:line_end].split('\x00'))
2287
start = line_end + 1
2288
# 1 line with parents (None: for None, '' for ())
2289
line_end = bytes.find('\n', start)
2290
line = bytes[start:line_end]
2295
[tuple(segment.split('\x00')) for segment in line.split('\t')
2297
self.global_map[key] = parents
2298
start = line_end + 1
2299
# one line with method
2300
line_end = bytes.find('\n', start)
2301
line = bytes[start:line_end]
2303
start = line_end + 1
2304
# one line with noeol
2305
line_end = bytes.find('\n', start)
2306
line = bytes[start:line_end]
2308
start = line_end + 1
2309
# one line with next ('' for None)
2310
line_end = bytes.find('\n', start)
2311
line = bytes[start:line_end]
2315
next = tuple(bytes[start:line_end].split('\x00'))
2316
start = line_end + 1
2317
# one line with byte count of the record bytes
2318
line_end = bytes.find('\n', start)
2319
line = bytes[start:line_end]
2321
start = line_end + 1
2323
record_bytes = bytes[start:start+count]
2324
start = start + count
2326
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2328
def get_record_stream(self):
2329
"""Get a record stream for for keys requested by the bytestream."""
2331
for key in self.keys:
2332
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2335
def _wire_bytes(self):
2339
1627
class _KndxIndex(object):
2340
1628
"""Manages knit index files
3431
2417
annotator = _KnitAnnotator(knit)
3432
return iter(annotator.annotate_flat(revision_id))
3435
class _KnitAnnotator(annotate.Annotator):
2418
return iter(annotator.annotate(revision_id))
2421
class _KnitAnnotator(object):
3436
2422
"""Build up the annotations for a text."""
3438
def __init__(self, vf):
3439
annotate.Annotator.__init__(self, vf)
3441
# TODO: handle Nodes which cannot be extracted
3442
# self._ghosts = set()
3444
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3445
self._matching_blocks = {}
3447
# KnitContent objects
3448
self._content_objects = {}
3449
# The number of children that depend on this fulltext content object
3450
self._num_compression_children = {}
3451
# Delta records that need their compression parent before they can be
3453
self._pending_deltas = {}
3454
# Fulltext records that are waiting for their parents fulltexts before
3455
# they can be yielded for annotation
3456
self._pending_annotation = {}
2424
def __init__(self, knit):
2427
# Content objects, differs from fulltexts because of how final newlines
2428
# are treated by knits. the content objects here will always have a
2430
self._fulltext_contents = {}
2432
# Annotated lines of specific revisions
2433
self._annotated_lines = {}
2435
# Track the raw data for nodes that we could not process yet.
2436
# This maps the revision_id of the base to a list of children that will
2437
# annotated from it.
2438
self._pending_children = {}
2440
# Nodes which cannot be extracted
2441
self._ghosts = set()
2443
# Track how many children this node has, so we know if we need to keep
2445
self._annotate_children = {}
2446
self._compression_children = {}
3458
2448
self._all_build_details = {}
2449
# The children => parent revision_id graph
2450
self._revision_id_graph = {}
2452
self._heads_provider = None
2454
self._nodes_to_keep_annotations = set()
2455
self._generations_until_keep = 100
2457
def set_generations_until_keep(self, value):
2458
"""Set the number of generations before caching a node.
2460
Setting this to -1 will cache every merge node, setting this higher
2461
will cache fewer nodes.
2463
self._generations_until_keep = value
2465
def _add_fulltext_content(self, revision_id, content_obj):
2466
self._fulltext_contents[revision_id] = content_obj
2467
# TODO: jam 20080305 It might be good to check the sha1digest here
2468
return content_obj.text()
2470
def _check_parents(self, child, nodes_to_annotate):
2471
"""Check if all parents have been processed.
2473
:param child: A tuple of (rev_id, parents, raw_content)
2474
:param nodes_to_annotate: If child is ready, add it to
2475
nodes_to_annotate, otherwise put it back in self._pending_children
2477
for parent_id in child[1]:
2478
if (parent_id not in self._annotated_lines):
2479
# This parent is present, but another parent is missing
2480
self._pending_children.setdefault(parent_id,
2484
# This one is ready to be processed
2485
nodes_to_annotate.append(child)
2487
def _add_annotation(self, revision_id, fulltext, parent_ids,
2488
left_matching_blocks=None):
2489
"""Add an annotation entry.
2491
All parents should already have been annotated.
2492
:return: A list of children that now have their parents satisfied.
2494
a = self._annotated_lines
2495
annotated_parent_lines = [a[p] for p in parent_ids]
2496
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2497
fulltext, revision_id, left_matching_blocks,
2498
heads_provider=self._get_heads_provider()))
2499
self._annotated_lines[revision_id] = annotated_lines
2500
for p in parent_ids:
2501
ann_children = self._annotate_children[p]
2502
ann_children.remove(revision_id)
2503
if (not ann_children
2504
and p not in self._nodes_to_keep_annotations):
2505
del self._annotated_lines[p]
2506
del self._all_build_details[p]
2507
if p in self._fulltext_contents:
2508
del self._fulltext_contents[p]
2509
# Now that we've added this one, see if there are any pending
2510
# deltas to be done, certainly this parent is finished
2511
nodes_to_annotate = []
2512
for child in self._pending_children.pop(revision_id, []):
2513
self._check_parents(child, nodes_to_annotate)
2514
return nodes_to_annotate
3460
2516
def _get_build_graph(self, key):
3461
2517
"""Get the graphs for building texts and annotations.
3468
2524
:return: A list of (key, index_memo) records, suitable for
3469
passing to read_records_iter to start reading in the raw data from
2525
passing to read_records_iter to start reading in the raw data fro/
2528
if key in self._annotated_lines:
3472
2531
pending = set([key])
3475
self._num_needed_children[key] = 1
3477
2536
# get all pending nodes
3478
2538
this_iteration = pending
3479
build_details = self._vf._index.get_build_details(this_iteration)
2539
build_details = self._knit._index.get_build_details(this_iteration)
3480
2540
self._all_build_details.update(build_details)
3481
# new_nodes = self._vf._index._get_entries(this_iteration)
2541
# new_nodes = self._knit._index._get_entries(this_iteration)
3482
2542
pending = set()
3483
2543
for key, details in build_details.iteritems():
3484
(index_memo, compression_parent, parent_keys,
2544
(index_memo, compression_parent, parents,
3485
2545
record_details) = details
3486
self._parent_map[key] = parent_keys
3487
self._heads_provider = None
2546
self._revision_id_graph[key] = parents
3488
2547
records.append((key, index_memo))
3489
2548
# Do we actually need to check _annotated_lines?
3490
pending.update([p for p in parent_keys
3491
if p not in self._all_build_details])
3493
for parent_key in parent_keys:
3494
if parent_key in self._num_needed_children:
3495
self._num_needed_children[parent_key] += 1
3497
self._num_needed_children[parent_key] = 1
2549
pending.update(p for p in parents
2550
if p not in self._all_build_details)
3498
2551
if compression_parent:
3499
if compression_parent in self._num_compression_children:
3500
self._num_compression_children[compression_parent] += 1
3502
self._num_compression_children[compression_parent] = 1
2552
self._compression_children.setdefault(compression_parent,
2555
for parent in parents:
2556
self._annotate_children.setdefault(parent,
2558
num_gens = generation - kept_generation
2559
if ((num_gens >= self._generations_until_keep)
2560
and len(parents) > 1):
2561
kept_generation = generation
2562
self._nodes_to_keep_annotations.add(key)
3504
2564
missing_versions = this_iteration.difference(build_details.keys())
3505
if missing_versions:
3506
for key in missing_versions:
3507
if key in self._parent_map and key in self._text_cache:
3508
# We already have this text ready, we just need to
3509
# yield it later so we get it annotated
3511
parent_keys = self._parent_map[key]
3512
for parent_key in parent_keys:
3513
if parent_key in self._num_needed_children:
3514
self._num_needed_children[parent_key] += 1
3516
self._num_needed_children[parent_key] = 1
3517
pending.update([p for p in parent_keys
3518
if p not in self._all_build_details])
3520
raise errors.RevisionNotPresent(key, self._vf)
2565
self._ghosts.update(missing_versions)
2566
for missing_version in missing_versions:
2567
# add a key, no parents
2568
self._revision_id_graph[missing_version] = ()
2569
pending.discard(missing_version) # don't look for it
2570
if self._ghosts.intersection(self._compression_children):
2572
"We cannot have nodes which have a ghost compression parent:\n"
2574
"compression children: %r"
2575
% (self._ghosts, self._compression_children))
2576
# Cleanout anything that depends on a ghost so that we don't wait for
2577
# the ghost to show up
2578
for node in self._ghosts:
2579
if node in self._annotate_children:
2580
# We won't be building this node
2581
del self._annotate_children[node]
3521
2582
# Generally we will want to read the records in reverse order, because
3522
2583
# we find the parent nodes after the children
3523
2584
records.reverse()
3524
return records, ann_keys
3526
def _get_needed_texts(self, key, pb=None):
3527
# if True or len(self._vf._fallback_vfs) > 0:
3528
if len(self._vf._fallback_vfs) > 0:
3529
# If we have fallbacks, go to the generic path
3530
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3535
records, ann_keys = self._get_build_graph(key)
3536
for idx, (sub_key, text, num_lines) in enumerate(
3537
self._extract_texts(records)):
3539
pb.update('annotating', idx, len(records))
3540
yield sub_key, text, num_lines
3541
for sub_key in ann_keys:
3542
text = self._text_cache[sub_key]
3543
num_lines = len(text) # bad assumption
3544
yield sub_key, text, num_lines
3546
except errors.RetryWithNewPacks, e:
3547
self._vf._access.reload_or_raise(e)
3548
# The cached build_details are no longer valid
3549
self._all_build_details.clear()
3551
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3552
parent_lines = self._text_cache[compression_parent]
3553
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3554
self._matching_blocks[(key, compression_parent)] = blocks
3556
def _expand_record(self, key, parent_keys, compression_parent, record,
3559
if compression_parent:
3560
if compression_parent not in self._content_objects:
3561
# Waiting for the parent
3562
self._pending_deltas.setdefault(compression_parent, []).append(
3563
(key, parent_keys, record, record_details))
3565
# We have the basis parent, so expand the delta
3566
num = self._num_compression_children[compression_parent]
3569
base_content = self._content_objects.pop(compression_parent)
3570
self._num_compression_children.pop(compression_parent)
3572
self._num_compression_children[compression_parent] = num
3573
base_content = self._content_objects[compression_parent]
3574
# It is tempting to want to copy_base_content=False for the last
3575
# child object. However, whenever noeol=False,
3576
# self._text_cache[parent_key] is content._lines. So mutating it
3577
# gives very bad results.
3578
# The alternative is to copy the lines into text cache, but then we
3579
# are copying anyway, so just do it here.
3580
content, delta = self._vf._factory.parse_record(
3581
key, record, record_details, base_content,
3582
copy_base_content=True)
3585
content, _ = self._vf._factory.parse_record(
3586
key, record, record_details, None)
3587
if self._num_compression_children.get(key, 0) > 0:
3588
self._content_objects[key] = content
3589
lines = content.text()
3590
self._text_cache[key] = lines
3591
if delta is not None:
3592
self._cache_delta_blocks(key, compression_parent, delta, lines)
3595
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3596
"""Get the list of annotations for the parent, and the matching lines.
3598
:param text: The opaque value given by _get_needed_texts
3599
:param parent_key: The key for the parent text
3600
:return: (parent_annotations, matching_blocks)
3601
parent_annotations is a list as long as the number of lines in
3603
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3604
indicating which lines match between the two texts
3606
block_key = (key, parent_key)
3607
if block_key in self._matching_blocks:
3608
blocks = self._matching_blocks.pop(block_key)
3609
parent_annotations = self._annotations_cache[parent_key]
3610
return parent_annotations, blocks
3611
return annotate.Annotator._get_parent_annotations_and_matches(self,
3612
key, text, parent_key)
3614
def _process_pending(self, key):
3615
"""The content for 'key' was just processed.
3617
Determine if there is any more pending work to be processed.
3620
if key in self._pending_deltas:
3621
compression_parent = key
3622
children = self._pending_deltas.pop(key)
3623
for child_key, parent_keys, record, record_details in children:
3624
lines = self._expand_record(child_key, parent_keys,
3626
record, record_details)
3627
if self._check_ready_for_annotations(child_key, parent_keys):
3628
to_return.append(child_key)
3629
# Also check any children that are waiting for this parent to be
3631
if key in self._pending_annotation:
3632
children = self._pending_annotation.pop(key)
3633
to_return.extend([c for c, p_keys in children
3634
if self._check_ready_for_annotations(c, p_keys)])
3637
def _check_ready_for_annotations(self, key, parent_keys):
3638
"""return true if this text is ready to be yielded.
3640
Otherwise, this will return False, and queue the text into
3641
self._pending_annotation
3643
for parent_key in parent_keys:
3644
if parent_key not in self._annotations_cache:
3645
# still waiting on at least one parent text, so queue it up
3646
# Note that if there are multiple parents, we need to wait
3648
self._pending_annotation.setdefault(parent_key,
3649
[]).append((key, parent_keys))
3653
def _extract_texts(self, records):
3654
"""Extract the various texts needed based on records"""
2587
def _annotate_records(self, records):
2588
"""Build the annotations for the listed records."""
3655
2589
# We iterate in the order read, rather than a strict order requested
3656
2590
# However, process what we can, and put off to the side things that
3657
2591
# still need parents, cleaning them up when those parents are
3660
# 1) As 'records' are read, see if we can expand these records into
3661
# Content objects (and thus lines)
3662
# 2) If a given line-delta is waiting on its compression parent, it
3663
# gets queued up into self._pending_deltas, otherwise we expand
3664
# it, and put it into self._text_cache and self._content_objects
3665
# 3) If we expanded the text, we will then check to see if all
3666
# parents have also been processed. If so, this text gets yielded,
3667
# else this record gets set aside into pending_annotation
3668
# 4) Further, if we expanded the text in (2), we will then check to
3669
# see if there are any children in self._pending_deltas waiting to
3670
# also be processed. If so, we go back to (2) for those
3671
# 5) Further again, if we yielded the text, we can then check if that
3672
# 'unlocks' any of the texts in pending_annotations, which should
3673
# then get yielded as well
3674
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3675
# compression child could unlock yet another, and yielding a fulltext
3676
# will also 'unlock' the children that are waiting on that annotation.
3677
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3678
# if other parents are also waiting.)
3679
# We want to yield content before expanding child content objects, so
3680
# that we know when we can re-use the content lines, and the annotation
3681
# code can know when it can stop caching fulltexts, as well.
3683
# Children that are missing their compression parent
3685
for (key, record, digest) in self._vf._read_records_iter(records):
3687
details = self._all_build_details[key]
3688
(_, compression_parent, parent_keys, record_details) = details
3689
lines = self._expand_record(key, parent_keys, compression_parent,
3690
record, record_details)
3692
# Pending delta should be queued up
2593
for (rev_id, record,
2594
digest) in self._knit._read_records_iter(records):
2595
if rev_id in self._annotated_lines:
3694
# At this point, we may be able to yield this content, if all
3695
# parents are also finished
3696
yield_this_text = self._check_ready_for_annotations(key,
3699
# All parents present
3700
yield key, lines, len(lines)
3701
to_process = self._process_pending(key)
3703
this_process = to_process
3705
for key in this_process:
3706
lines = self._text_cache[key]
3707
yield key, lines, len(lines)
3708
to_process.extend(self._process_pending(key))
2597
parent_ids = self._revision_id_graph[rev_id]
2598
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2599
details = self._all_build_details[rev_id]
2600
(index_memo, compression_parent, parents,
2601
record_details) = details
2602
nodes_to_annotate = []
2603
# TODO: Remove the punning between compression parents, and
2604
# parent_ids, we should be able to do this without assuming
2606
if len(parent_ids) == 0:
2607
# There are no parents for this node, so just add it
2608
# TODO: This probably needs to be decoupled
2609
fulltext_content, delta = self._knit._factory.parse_record(
2610
rev_id, record, record_details, None)
2611
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2612
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2613
parent_ids, left_matching_blocks=None))
2615
child = (rev_id, parent_ids, record)
2616
# Check if all the parents are present
2617
self._check_parents(child, nodes_to_annotate)
2618
while nodes_to_annotate:
2619
# Should we use a queue here instead of a stack?
2620
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2621
(index_memo, compression_parent, parents,
2622
record_details) = self._all_build_details[rev_id]
2623
if compression_parent is not None:
2624
comp_children = self._compression_children[compression_parent]
2625
if rev_id not in comp_children:
2626
raise AssertionError("%r not in compression children %r"
2627
% (rev_id, comp_children))
2628
# If there is only 1 child, it is safe to reuse this
2630
reuse_content = (len(comp_children) == 1
2631
and compression_parent not in
2632
self._nodes_to_keep_annotations)
2634
# Remove it from the cache since it will be changing
2635
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
2636
# Make sure to copy the fulltext since it might be
2638
parent_fulltext = list(parent_fulltext_content.text())
2640
parent_fulltext_content = self._fulltext_contents[compression_parent]
2641
parent_fulltext = parent_fulltext_content.text()
2642
comp_children.remove(rev_id)
2643
fulltext_content, delta = self._knit._factory.parse_record(
2644
rev_id, record, record_details,
2645
parent_fulltext_content,
2646
copy_base_content=(not reuse_content))
2647
fulltext = self._add_fulltext_content(rev_id,
2649
blocks = KnitContent.get_line_delta_blocks(delta,
2650
parent_fulltext, fulltext)
2652
fulltext_content = self._knit._factory.parse_fulltext(
2654
fulltext = self._add_fulltext_content(rev_id,
2657
nodes_to_annotate.extend(
2658
self._add_annotation(rev_id, fulltext, parent_ids,
2659
left_matching_blocks=blocks))
2661
def _get_heads_provider(self):
2662
"""Create a heads provider for resolving ancestry issues."""
2663
if self._heads_provider is not None:
2664
return self._heads_provider
2665
parent_provider = _mod_graph.DictParentsProvider(
2666
self._revision_id_graph)
2667
graph_obj = _mod_graph.Graph(parent_provider)
2668
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
2669
self._heads_provider = head_cache
2672
def annotate(self, key):
2673
"""Return the annotated fulltext at the given key.
2675
:param key: The key to annotate.
2677
if True or len(self._knit._fallback_vfs) > 0:
2678
# stacked knits can't use the fast path at present.
2679
return self._simple_annotate(key)
2680
records = self._get_build_graph(key)
2681
if key in self._ghosts:
2682
raise errors.RevisionNotPresent(key, self._knit)
2683
self._annotate_records(records)
2684
return self._annotated_lines[key]
2686
def _simple_annotate(self, key):
2687
"""Return annotated fulltext, rediffing from the full texts.
2689
This is slow but makes no assumptions about the repository
2690
being able to produce line deltas.
2692
# TODO: this code generates a parent maps of present ancestors; it
2693
# could be split out into a separate method, and probably should use
2694
# iter_ancestry instead. -- mbp and robertc 20080704
2695
graph = _mod_graph.Graph(self._knit)
2696
head_cache = _mod_graph.FrozenHeadsCache(graph)
2697
search = graph._make_breadth_first_searcher([key])
2701
present, ghosts = search.next_with_ghosts()
2702
except StopIteration:
2704
keys.update(present)
2705
parent_map = self._knit.get_parent_map(keys)
2707
reannotate = annotate.reannotate
2708
for record in self._knit.get_record_stream(keys, 'topological', True):
2710
fulltext = split_lines(record.get_bytes_as('fulltext'))
2711
parents = parent_map[key]
2712
if parents is not None:
2713
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
2716
parent_cache[key] = list(
2717
reannotate(parent_lines, fulltext, key, None, head_cache))
2719
return parent_cache[key]
2721
raise errors.RevisionNotPresent(key, self._knit)
3711
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3712
except ImportError, e:
3713
osutils.failed_to_load_extension(e)
2725
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3714
2727
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
