85
55
VALUE := no-newline-no-null-bytes
88
def __init__(self, reference_lists=0, key_elements=1):
58
def __init__(self, reference_lists=0):
89
59
"""Create a GraphIndex builder.
91
61
:param reference_lists: The number of node references lists for each
93
:param key_elements: The number of bytestrings in each key.
95
64
self.reference_lists = reference_lists
96
# A dict of {key: (absent, ref_lists, value)}
98
# Keys that are referenced but not actually present in this index
99
self._absent_keys = set()
100
self._nodes_by_key = None
101
self._key_length = key_elements
102
self._optimize_for_size = False
103
self._combine_backing_indices = True
105
def _check_key(self, key):
106
"""Raise BadIndexKey if key is not a valid key for this index."""
107
if type(key) not in (tuple, StaticTuple):
108
raise errors.BadIndexKey(key)
109
if self._key_length != len(key):
110
raise errors.BadIndexKey(key)
112
if not element or _whitespace_re.search(element) is not None:
113
raise errors.BadIndexKey(element)
115
def _external_references(self):
116
"""Return references that are not present in this index.
120
# TODO: JAM 2008-11-21 This makes an assumption about how the reference
121
# lists are used. It is currently correct for pack-0.92 through
122
# 1.9, which use the node references (3rd column) second
123
# reference list as the compression parent. Perhaps this should
124
# be moved into something higher up the stack, since it
125
# makes assumptions about how the index is used.
126
if self.reference_lists > 1:
127
for node in self.iter_all_entries():
129
refs.update(node[3][1])
132
# If reference_lists == 0 there can be no external references, and
133
# if reference_lists == 1, then there isn't a place to store the
137
def _get_nodes_by_key(self):
138
if self._nodes_by_key is None:
140
if self.reference_lists:
141
for key, (absent, references, value) in self._nodes.iteritems():
144
key_dict = nodes_by_key
145
for subkey in key[:-1]:
146
key_dict = key_dict.setdefault(subkey, {})
147
key_dict[key[-1]] = key, value, references
149
for key, (absent, references, value) in self._nodes.iteritems():
152
key_dict = nodes_by_key
153
for subkey in key[:-1]:
154
key_dict = key_dict.setdefault(subkey, {})
155
key_dict[key[-1]] = key, value
156
self._nodes_by_key = nodes_by_key
157
return self._nodes_by_key
159
def _update_nodes_by_key(self, key, value, node_refs):
160
"""Update the _nodes_by_key dict with a new key.
162
For a key of (foo, bar, baz) create
163
_nodes_by_key[foo][bar][baz] = key_value
165
if self._nodes_by_key is None:
167
key_dict = self._nodes_by_key
168
if self.reference_lists:
169
key_value = StaticTuple(key, value, node_refs)
171
key_value = StaticTuple(key, value)
172
for subkey in key[:-1]:
173
key_dict = key_dict.setdefault(subkey, {})
174
key_dict[key[-1]] = key_value
176
def _check_key_ref_value(self, key, references, value):
177
"""Check that 'key' and 'references' are all valid.
179
:param key: A key tuple. Must conform to the key interface (be a tuple,
180
be of the right length, not have any whitespace or nulls in any key
182
:param references: An iterable of reference lists. Something like
183
[[(ref, key)], [(ref, key), (other, key)]]
184
:param value: The value associate with this key. Must not contain
185
newlines or null characters.
186
:return: (node_refs, absent_references)
187
node_refs basically a packed form of 'references' where all
189
absent_references reference keys that are not in self._nodes.
190
This may contain duplicates if the same key is
191
referenced in multiple lists.
193
as_st = StaticTuple.from_sequence
67
def add_node(self, key, value, references=()):
68
"""Add a node to the index.
70
:param key: The key. keys must be whitespace-free utf8.
71
:param references: An iterable of iterables of keys. Each is a
72
reference to another key.
73
:param value: The value to associate with the key. It may be any
74
bytes as long as it does not contain \0 or \n.
76
if not key or _whitespace_re.search(key) is not None:
77
raise errors.BadIndexKey(key)
195
78
if _newline_null_re.search(value) is not None:
196
79
raise errors.BadIndexValue(value)
197
80
if len(references) != self.reference_lists:
198
81
raise errors.BadIndexValue(references)
200
absent_references = []
201
83
for reference_list in references:
202
84
for reference in reference_list:
203
# If reference *is* in self._nodes, then we know it has already
85
if _whitespace_re.search(reference) is not None:
86
raise errors.BadIndexKey(reference)
205
87
if reference not in self._nodes:
206
self._check_key(reference)
207
absent_references.append(reference)
208
reference_list = as_st([as_st(ref).intern()
209
for ref in reference_list])
210
node_refs.append(reference_list)
211
return as_st(node_refs), absent_references
213
def add_node(self, key, value, references=()):
214
"""Add a node to the index.
216
:param key: The key. keys are non-empty tuples containing
217
as many whitespace-free utf8 bytestrings as the key length
218
defined for this index.
219
:param references: An iterable of iterables of keys. Each is a
220
reference to another key.
221
:param value: The value to associate with the key. It may be any
222
bytes as long as it does not contain \0 or \n.
225
absent_references) = self._check_key_ref_value(key, references, value)
226
if key in self._nodes and self._nodes[key][0] != 'a':
88
self._nodes[reference] = ('a', (), '')
89
node_refs.append(tuple(reference_list))
90
if key in self._nodes and self._nodes[key][0] == '':
227
91
raise errors.BadIndexDuplicateKey(key, self)
228
for reference in absent_references:
229
# There may be duplicates, but I don't think it is worth worrying
231
self._nodes[reference] = ('a', (), '')
232
self._absent_keys.update(absent_references)
233
self._absent_keys.discard(key)
234
self._nodes[key] = ('', node_refs, value)
235
if self._nodes_by_key is not None and self._key_length > 1:
236
self._update_nodes_by_key(key, value, node_refs)
238
def clear_cache(self):
239
"""See GraphIndex.clear_cache()
241
This is a no-op, but we need the api to conform to a generic 'Index'
92
self._nodes[key] = ('', tuple(node_refs), value)
246
95
lines = [_SIGNATURE]
247
96
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
248
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
249
key_count = len(self._nodes) - len(self._absent_keys)
250
lines.append(_OPTION_LEN + str(key_count) + '\n')
251
prefix_length = sum(len(x) for x in lines)
97
prefix_length = len(lines[0]) + len(lines[1])
252
98
# references are byte offsets. To avoid having to do nasty
253
# polynomial work to resolve offsets (references to later in the
99
# polynomial work to resolve offsets (references to later in the
254
100
# file cannot be determined until all the inbetween references have
255
101
# been calculated too) we pad the offsets with 0's to make them be
256
102
# of consistent length. Using binary offsets would break the trivial
382
188
suitable for production use. :XXX
385
def __init__(self, transport, name, size, unlimited_cache=False, offset=0):
191
def __init__(self, transport, name):
386
192
"""Open an index called name on transport.
388
194
:param transport: A bzrlib.transport.Transport.
389
195
:param name: A path to provide to transport API calls.
390
:param size: The size of the index in bytes. This is used for bisection
391
logic to perform partial index reads. While the size could be
392
obtained by statting the file this introduced an additional round
393
trip as well as requiring stat'able transports, both of which are
394
avoided by having it supplied. If size is None, then bisection
395
support will be disabled and accessing the index will just stream
397
:param offset: Instead of starting the index data at offset 0, start it
398
at an arbitrary offset.
400
197
self._transport = transport
401
198
self._name = name
402
# Becomes a dict of key:(value, reference-list-byte-locations) used by
403
# the bisection interface to store parsed but not resolved keys.
404
self._bisect_nodes = None
405
# Becomes a dict of key:(value, reference-list-keys) which are ready to
406
# be returned directly to callers.
408
# a sorted list of slice-addresses for the parsed bytes of the file.
409
# e.g. (0,1) would mean that byte 0 is parsed.
410
self._parsed_byte_map = []
411
# a sorted list of keys matching each slice address for parsed bytes
412
# e.g. (None, 'foo@bar') would mean that the first byte contained no
413
# key, and the end byte of the slice is the of the data for 'foo@bar'
414
self._parsed_key_map = []
415
self._key_count = None
416
self._keys_by_offset = None
417
self._nodes_by_key = None
419
# The number of bytes we've read so far in trying to process this file
421
self._base_offset = offset
423
def __eq__(self, other):
424
"""Equal when self and other were created with the same parameters."""
426
type(self) == type(other) and
427
self._transport == other._transport and
428
self._name == other._name and
429
self._size == other._size)
431
def __ne__(self, other):
432
return not self.__eq__(other)
435
return "%s(%r)" % (self.__class__.__name__,
436
self._transport.abspath(self._name))
438
def _buffer_all(self, stream=None):
439
"""Buffer all the index data.
441
Mutates self._nodes and self.keys_by_offset.
200
def iter_all_entries(self):
201
"""Iterate over all keys within the index.
203
:return: An iterable of (key, value) or (key, value, reference_lists).
204
The former tuple is used when there are no reference lists in the
205
index, making the API compatible with simple key:value index types.
206
There is no defined order for the result iteration - it will be in
207
the most efficient order for the index.
443
if self._nodes is not None:
444
# We already did this
446
if 'index' in debug.debug_flags:
447
mutter('Reading entire index %s', self._transport.abspath(self._name))
449
stream = self._transport.get(self._name)
450
if self._base_offset != 0:
451
# This is wasteful, but it is better than dealing with
452
# adjusting all the offsets, etc.
453
stream = StringIO(stream.read()[self._base_offset:])
209
stream = self._transport.get(self._name)
454
210
self._read_prefix(stream)
455
self._expected_elements = 3 + self._key_length
457
# raw data keyed by offset
458
self._keys_by_offset = {}
459
# ready-to-return key:value or key:value, node_ref_lists
461
self._nodes_by_key = None
212
self.keys_by_offset = {}
463
214
pos = stream.tell()
464
lines = stream.read().split('\n')
467
_, _, _, trailers = self._parse_lines(lines, pos)
468
for key, absent, references, value in self._keys_by_offset.itervalues():
215
for line in stream.readlines():
219
key, absent, references, value = line.split('\0')
220
value = value[:-1] # remove the newline
222
for ref_string in references.split('\t'):
223
ref_lists.append(tuple([
224
int(ref) for ref in ref_string.split('\r') if ref
226
ref_lists = tuple(ref_lists)
227
self.keys_by_offset[pos] = (key, absent, ref_lists, value)
229
for key, absent, references, value in self.keys_by_offset.itervalues():
471
232
# resolve references:
472
233
if self.node_ref_lists:
473
node_value = (value, self._resolve_references(references))
235
for ref_list in references:
236
node_refs.append(tuple([self.keys_by_offset[ref][0] for ref in ref_list]))
237
yield (key, value, tuple(node_refs))
476
self._nodes[key] = node_value
477
# cache the keys for quick set intersections
478
240
if trailers != 1:
479
241
# there must be one line - the empty trailer line.
480
242
raise errors.BadIndexData(self)
482
def clear_cache(self):
483
"""Clear out any cached/memoized values.
485
This can be called at any time, but generally it is used when we have
486
extracted some information, but don't expect to be requesting any more
490
def external_references(self, ref_list_num):
491
"""Return references that are not present in this index.
494
if ref_list_num + 1 > self.node_ref_lists:
495
raise ValueError('No ref list %d, index has %d ref lists'
496
% (ref_list_num, self.node_ref_lists))
499
for key, (value, ref_lists) in nodes.iteritems():
500
ref_list = ref_lists[ref_list_num]
501
refs.update([ref for ref in ref_list if ref not in nodes])
504
def _get_nodes_by_key(self):
505
if self._nodes_by_key is None:
507
if self.node_ref_lists:
508
for key, (value, references) in self._nodes.iteritems():
509
key_dict = nodes_by_key
510
for subkey in key[:-1]:
511
key_dict = key_dict.setdefault(subkey, {})
512
key_dict[key[-1]] = key, value, references
514
for key, value in self._nodes.iteritems():
515
key_dict = nodes_by_key
516
for subkey in key[:-1]:
517
key_dict = key_dict.setdefault(subkey, {})
518
key_dict[key[-1]] = key, value
519
self._nodes_by_key = nodes_by_key
520
return self._nodes_by_key
522
def iter_all_entries(self):
523
"""Iterate over all keys within the index.
525
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
526
The former tuple is used when there are no reference lists in the
527
index, making the API compatible with simple key:value index types.
528
There is no defined order for the result iteration - it will be in
529
the most efficient order for the index.
531
if 'evil' in debug.debug_flags:
532
trace.mutter_callsite(3,
533
"iter_all_entries scales with size of history.")
534
if self._nodes is None:
536
if self.node_ref_lists:
537
for key, (value, node_ref_lists) in self._nodes.iteritems():
538
yield self, key, value, node_ref_lists
540
for key, value in self._nodes.iteritems():
541
yield self, key, value
543
244
def _read_prefix(self, stream):
544
245
signature = stream.read(len(self._signature()))
545
246
if not signature == self._signature():
658
if self._size is None and self._nodes is None:
661
# We fit about 20 keys per minimum-read (4K), so if we are looking for
662
# more than 1/20th of the index its likely (assuming homogenous key
663
# spread) that we'll read the entire index. If we're going to do that,
664
# buffer the whole thing. A better analysis might take key spread into
665
# account - but B+Tree indices are better anyway.
666
# We could look at all data read, and use a threshold there, which will
667
# trigger on ancestry walks, but that is not yet fully mapped out.
668
if self._nodes is None and len(keys) * 20 > self.key_count():
670
if self._nodes is not None:
671
return self._iter_entries_from_total_buffer(keys)
673
return (result[1] for result in bisect_multi_bytes(
674
self._lookup_keys_via_location, self._size, keys))
676
def iter_entries_prefix(self, keys):
677
"""Iterate over keys within the index using prefix matching.
679
Prefix matching is applied within the tuple of a key, not to within
680
the bytestring of each key element. e.g. if you have the keys ('foo',
681
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
682
only the former key is returned.
684
WARNING: Note that this method currently causes a full index parse
685
unconditionally (which is reasonably appropriate as it is a means for
686
thunking many small indices into one larger one and still supplies
687
iter_all_entries at the thunk layer).
689
:param keys: An iterable providing the key prefixes to be retrieved.
690
Each key prefix takes the form of a tuple the length of a key, but
691
with the last N elements 'None' rather than a regular bytestring.
692
The first element cannot be 'None'.
693
:return: An iterable as per iter_all_entries, but restricted to the
694
keys with a matching prefix to those supplied. No additional keys
695
will be returned, and every match that is in the index will be
701
# load data - also finds key lengths
702
if self._nodes is None:
704
if self._key_length == 1:
708
raise errors.BadIndexKey(key)
709
if len(key) != self._key_length:
710
raise errors.BadIndexKey(key)
711
if self.node_ref_lists:
712
value, node_refs = self._nodes[key]
713
yield self, key, value, node_refs
715
yield self, key, self._nodes[key]
717
nodes_by_key = self._get_nodes_by_key()
721
raise errors.BadIndexKey(key)
722
if len(key) != self._key_length:
723
raise errors.BadIndexKey(key)
724
# find what it refers to:
725
key_dict = nodes_by_key
727
# find the subdict whose contents should be returned.
729
while len(elements) and elements[0] is not None:
730
key_dict = key_dict[elements[0]]
733
# a non-existant lookup.
738
key_dict = dicts.pop(-1)
739
# can't be empty or would not exist
740
item, value = key_dict.iteritems().next()
741
if type(value) == dict:
743
dicts.extend(key_dict.itervalues())
746
for value in key_dict.itervalues():
747
# each value is the key:value:node refs tuple
749
yield (self, ) + value
751
# the last thing looked up was a terminal element
752
yield (self, ) + key_dict
754
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
755
"""See BTreeIndex._find_ancestors."""
756
# The api can be implemented as a trivial overlay on top of
757
# iter_entries, it is not an efficient implementation, but it at least
761
for index, key, value, refs in self.iter_entries(keys):
762
parent_keys = refs[ref_list_num]
764
parent_map[key] = parent_keys
765
search_keys.update(parent_keys)
766
# Figure out what, if anything, was missing
767
missing_keys.update(set(keys).difference(found_keys))
768
search_keys = search_keys.difference(parent_map)
772
"""Return an estimate of the number of keys in this index.
774
For GraphIndex the estimate is exact.
776
if self._key_count is None:
777
self._read_and_parse([_HEADER_READV])
778
return self._key_count
780
def _lookup_keys_via_location(self, location_keys):
781
"""Public interface for implementing bisection.
783
If _buffer_all has been called, then all the data for the index is in
784
memory, and this method should not be called, as it uses a separate
785
cache because it cannot pre-resolve all indices, which buffer_all does
788
:param location_keys: A list of location(byte offset), key tuples.
789
:return: A list of (location_key, result) tuples as expected by
790
bzrlib.bisect_multi.bisect_multi_bytes.
792
# Possible improvements:
793
# - only bisect lookup each key once
794
# - sort the keys first, and use that to reduce the bisection window
796
# this progresses in three parts:
799
# attempt to answer the question from the now in memory data.
800
# build the readv request
801
# for each location, ask for 800 bytes - much more than rows we've seen
804
for location, key in location_keys:
805
# can we answer from cache?
806
if self._bisect_nodes and key in self._bisect_nodes:
807
# We have the key parsed.
809
index = self._parsed_key_index(key)
810
if (len(self._parsed_key_map) and
811
self._parsed_key_map[index][0] <= key and
812
(self._parsed_key_map[index][1] >= key or
813
# end of the file has been parsed
814
self._parsed_byte_map[index][1] == self._size)):
815
# the key has been parsed, so no lookup is needed even if its
818
# - if we have examined this part of the file already - yes
819
index = self._parsed_byte_index(location)
820
if (len(self._parsed_byte_map) and
821
self._parsed_byte_map[index][0] <= location and
822
self._parsed_byte_map[index][1] > location):
823
# the byte region has been parsed, so no read is needed.
826
if location + length > self._size:
827
length = self._size - location
828
# todo, trim out parsed locations.
830
readv_ranges.append((location, length))
831
# read the header if needed
832
if self._bisect_nodes is None:
833
readv_ranges.append(_HEADER_READV)
834
self._read_and_parse(readv_ranges)
836
if self._nodes is not None:
837
# _read_and_parse triggered a _buffer_all because we requested the
839
for location, key in location_keys:
840
if key not in self._nodes: # not present
841
result.append(((location, key), False))
842
elif self.node_ref_lists:
843
value, refs = self._nodes[key]
844
result.append(((location, key),
845
(self, key, value, refs)))
847
result.append(((location, key),
848
(self, key, self._nodes[key])))
851
# - figure out <, >, missing, present
852
# - result present references so we can return them.
853
# keys that we cannot answer until we resolve references
854
pending_references = []
855
pending_locations = set()
856
for location, key in location_keys:
857
# can we answer from cache?
858
if key in self._bisect_nodes:
859
# the key has been parsed, so no lookup is needed
860
if self.node_ref_lists:
861
# the references may not have been all parsed.
862
value, refs = self._bisect_nodes[key]
863
wanted_locations = []
864
for ref_list in refs:
866
if ref not in self._keys_by_offset:
867
wanted_locations.append(ref)
869
pending_locations.update(wanted_locations)
870
pending_references.append((location, key))
872
result.append(((location, key), (self, key,
873
value, self._resolve_references(refs))))
875
result.append(((location, key),
876
(self, key, self._bisect_nodes[key])))
879
# has the region the key should be in, been parsed?
880
index = self._parsed_key_index(key)
881
if (self._parsed_key_map[index][0] <= key and
882
(self._parsed_key_map[index][1] >= key or
883
# end of the file has been parsed
884
self._parsed_byte_map[index][1] == self._size)):
885
result.append(((location, key), False))
887
# no, is the key above or below the probed location:
888
# get the range of the probed & parsed location
889
index = self._parsed_byte_index(location)
890
# if the key is below the start of the range, its below
891
if key < self._parsed_key_map[index][0]:
895
result.append(((location, key), direction))
897
# lookup data to resolve references
898
for location in pending_locations:
900
if location + length > self._size:
901
length = self._size - location
902
# TODO: trim out parsed locations (e.g. if the 800 is into the
903
# parsed region trim it, and dont use the adjust_for_latency
906
readv_ranges.append((location, length))
907
self._read_and_parse(readv_ranges)
908
if self._nodes is not None:
909
# The _read_and_parse triggered a _buffer_all, grab the data and
911
for location, key in pending_references:
912
value, refs = self._nodes[key]
913
result.append(((location, key), (self, key, value, refs)))
915
for location, key in pending_references:
916
# answer key references we had to look-up-late.
917
value, refs = self._bisect_nodes[key]
918
result.append(((location, key), (self, key,
919
value, self._resolve_references(refs))))
922
def _parse_header_from_bytes(self, bytes):
923
"""Parse the header from a region of bytes.
925
:param bytes: The data to parse.
926
:return: An offset, data tuple such as readv yields, for the unparsed
927
data. (which may length 0).
929
signature = bytes[0:len(self._signature())]
930
if not signature == self._signature():
931
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
932
lines = bytes[len(self._signature()):].splitlines()
933
options_line = lines[0]
934
if not options_line.startswith(_OPTION_NODE_REFS):
935
raise errors.BadIndexOptions(self)
937
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
939
raise errors.BadIndexOptions(self)
940
options_line = lines[1]
941
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
942
raise errors.BadIndexOptions(self)
944
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
946
raise errors.BadIndexOptions(self)
947
options_line = lines[2]
948
if not options_line.startswith(_OPTION_LEN):
949
raise errors.BadIndexOptions(self)
951
self._key_count = int(options_line[len(_OPTION_LEN):])
953
raise errors.BadIndexOptions(self)
954
# calculate the bytes we have processed
955
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
957
self._parsed_bytes(0, None, header_end, None)
958
# setup parsing state
959
self._expected_elements = 3 + self._key_length
960
# raw data keyed by offset
961
self._keys_by_offset = {}
962
# keys with the value and node references
963
self._bisect_nodes = {}
964
return header_end, bytes[header_end:]
966
def _parse_region(self, offset, data):
967
"""Parse node data returned from a readv operation.
969
:param offset: The byte offset the data starts at.
970
:param data: The data to parse.
974
end = offset + len(data)
977
# Trivial test - if the current index's end is within the
978
# low-matching parsed range, we're done.
979
index = self._parsed_byte_index(high_parsed)
980
if end < self._parsed_byte_map[index][1]:
982
# print "[%d:%d]" % (offset, end), \
983
# self._parsed_byte_map[index:index + 2]
984
high_parsed, last_segment = self._parse_segment(
985
offset, data, end, index)
989
def _parse_segment(self, offset, data, end, index):
990
"""Parse one segment of data.
992
:param offset: Where 'data' begins in the file.
993
:param data: Some data to parse a segment of.
994
:param end: Where data ends
995
:param index: The current index into the parsed bytes map.
996
:return: True if the parsed segment is the last possible one in the
998
:return: high_parsed_byte, last_segment.
999
high_parsed_byte is the location of the highest parsed byte in this
1000
segment, last_segment is True if the parsed segment is the last
1001
possible one in the data block.
1003
# default is to use all data
1005
# accomodate overlap with data before this.
1006
if offset < self._parsed_byte_map[index][1]:
1007
# overlaps the lower parsed region
1008
# skip the parsed data
1009
trim_start = self._parsed_byte_map[index][1] - offset
1010
# don't trim the start for \n
1011
start_adjacent = True
1012
elif offset == self._parsed_byte_map[index][1]:
1013
# abuts the lower parsed region
1016
# do not trim anything
1017
start_adjacent = True
1019
# does not overlap the lower parsed region
1022
# but trim the leading \n
1023
start_adjacent = False
1024
if end == self._size:
1025
# lines up to the end of all data:
1028
# do not strip to the last \n
1031
elif index + 1 == len(self._parsed_byte_map):
1032
# at the end of the parsed data
1035
# but strip to the last \n
1036
end_adjacent = False
1038
elif end == self._parsed_byte_map[index + 1][0]:
1039
# buts up against the next parsed region
1042
# do not strip to the last \n
1045
elif end > self._parsed_byte_map[index + 1][0]:
1046
# overlaps into the next parsed region
1047
# only consider the unparsed data
1048
trim_end = self._parsed_byte_map[index + 1][0] - offset
1049
# do not strip to the last \n as we know its an entire record
1051
last_segment = end < self._parsed_byte_map[index + 1][1]
1053
# does not overlap into the next region
1056
# but strip to the last \n
1057
end_adjacent = False
1059
# now find bytes to discard if needed
1060
if not start_adjacent:
1061
# work around python bug in rfind
1062
if trim_start is None:
1063
trim_start = data.find('\n') + 1
1065
trim_start = data.find('\n', trim_start) + 1
1066
if not (trim_start != 0):
1067
raise AssertionError('no \n was present')
1068
# print 'removing start', offset, trim_start, repr(data[:trim_start])
1069
if not end_adjacent:
1070
# work around python bug in rfind
1071
if trim_end is None:
1072
trim_end = data.rfind('\n') + 1
1074
trim_end = data.rfind('\n', None, trim_end) + 1
1075
if not (trim_end != 0):
1076
raise AssertionError('no \n was present')
1077
# print 'removing end', offset, trim_end, repr(data[trim_end:])
1078
# adjust offset and data to the parseable data.
1079
trimmed_data = data[trim_start:trim_end]
1080
if not (trimmed_data):
1081
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
1082
% (trim_start, trim_end, offset, offset + len(data)))
1084
offset += trim_start
1085
# print "parsing", repr(trimmed_data)
1086
# splitlines mangles the \r delimiters.. don't use it.
1087
lines = trimmed_data.split('\n')
1090
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1091
for key, value in nodes:
1092
self._bisect_nodes[key] = value
1093
self._parsed_bytes(offset, first_key,
1094
offset + len(trimmed_data), last_key)
1095
return offset + len(trimmed_data), last_segment
1097
def _parse_lines(self, lines, pos):
1104
# must be at the end
1106
if not (self._size == pos + 1):
1107
raise AssertionError("%s %s" % (self._size, pos))
1110
elements = line.split('\0')
1111
if len(elements) != self._expected_elements:
1112
raise errors.BadIndexData(self)
1113
# keys are tuples. Each element is a string that may occur many
1114
# times, so we intern them to save space. AB, RC, 200807
1115
key = tuple([intern(element) for element in elements[:self._key_length]])
1116
if first_key is None:
1118
absent, references, value = elements[-3:]
1120
for ref_string in references.split('\t'):
1121
ref_lists.append(tuple([
1122
int(ref) for ref in ref_string.split('\r') if ref
1124
ref_lists = tuple(ref_lists)
1125
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1126
pos += len(line) + 1 # +1 for the \n
1129
if self.node_ref_lists:
1130
node_value = (value, ref_lists)
1133
nodes.append((key, node_value))
1134
# print "parsed ", key
1135
return first_key, key, nodes, trailers
1137
def _parsed_bytes(self, start, start_key, end, end_key):
1138
"""Mark the bytes from start to end as parsed.
1140
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1143
:param start: The start of the parsed region.
1144
:param end: The end of the parsed region.
1146
index = self._parsed_byte_index(start)
1147
new_value = (start, end)
1148
new_key = (start_key, end_key)
1150
# first range parsed is always the beginning.
1151
self._parsed_byte_map.insert(index, new_value)
1152
self._parsed_key_map.insert(index, new_key)
1156
# extend lower region
1157
# extend higher region
1158
# combine two regions
1159
if (index + 1 < len(self._parsed_byte_map) and
1160
self._parsed_byte_map[index][1] == start and
1161
self._parsed_byte_map[index + 1][0] == end):
1162
# combine two regions
1163
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1164
self._parsed_byte_map[index + 1][1])
1165
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1166
self._parsed_key_map[index + 1][1])
1167
del self._parsed_byte_map[index + 1]
1168
del self._parsed_key_map[index + 1]
1169
elif self._parsed_byte_map[index][1] == start:
1170
# extend the lower entry
1171
self._parsed_byte_map[index] = (
1172
self._parsed_byte_map[index][0], end)
1173
self._parsed_key_map[index] = (
1174
self._parsed_key_map[index][0], end_key)
1175
elif (index + 1 < len(self._parsed_byte_map) and
1176
self._parsed_byte_map[index + 1][0] == end):
1177
# extend the higher entry
1178
self._parsed_byte_map[index + 1] = (
1179
start, self._parsed_byte_map[index + 1][1])
1180
self._parsed_key_map[index + 1] = (
1181
start_key, self._parsed_key_map[index + 1][1])
1184
self._parsed_byte_map.insert(index + 1, new_value)
1185
self._parsed_key_map.insert(index + 1, new_key)
1187
def _read_and_parse(self, readv_ranges):
1188
"""Read the ranges and parse the resulting data.
1190
:param readv_ranges: A prepared readv range list.
1192
if not readv_ranges:
1194
if self._nodes is None and self._bytes_read * 2 >= self._size:
1195
# We've already read more than 50% of the file and we are about to
1196
# request more data, just _buffer_all() and be done
1200
base_offset = self._base_offset
1201
if base_offset != 0:
1202
# Rewrite the ranges for the offset
1203
readv_ranges = [(start+base_offset, size)
1204
for start, size in readv_ranges]
1205
readv_data = self._transport.readv(self._name, readv_ranges, True,
1206
self._size + self._base_offset)
1208
for offset, data in readv_data:
1209
offset -= base_offset
1210
self._bytes_read += len(data)
1212
# transport.readv() expanded to extra data which isn't part of
1214
data = data[-offset:]
1216
if offset == 0 and len(data) == self._size:
1217
# We read the whole range, most likely because the
1218
# Transport upcast our readv ranges into one long request
1219
# for enough total data to grab the whole index.
1220
self._buffer_all(StringIO(data))
1222
if self._bisect_nodes is None:
1223
# this must be the start
1224
if not (offset == 0):
1225
raise AssertionError()
1226
offset, data = self._parse_header_from_bytes(data)
1227
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1228
self._parse_region(offset, data)
267
for node in self.iter_all_entries():
1230
274
def _signature(self):
1231
275
"""The file signature for this index type."""
1342
334
value and are only reported once.
1344
336
:param keys: An iterable providing the keys to be retrieved.
1345
:return: An iterable of (index, key, reference_lists, value). There is
1346
no defined order for the result iteration - it will be in the most
337
:return: An iterable of (key, reference_lists, value). There is no
338
defined order for the result iteration - it will be in the most
1347
339
efficient order for the index.
1349
341
keys = set(keys)
1353
for index in self._indices:
1357
for node in index.iter_entries(keys):
1358
keys.remove(node[1])
1362
hit_indices.append(index)
1364
except errors.NoSuchFile:
1365
self._reload_or_raise()
1366
self._move_to_front(hit_indices)
1368
def iter_entries_prefix(self, keys):
1369
"""Iterate over keys within the index using prefix matching.
1371
Duplicate keys across child indices are presumed to have the same
1372
value and are only reported once.
1374
Prefix matching is applied within the tuple of a key, not to within
1375
the bytestring of each key element. e.g. if you have the keys ('foo',
1376
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1377
only the former key is returned.
1379
:param keys: An iterable providing the key prefixes to be retrieved.
1380
Each key prefix takes the form of a tuple the length of a key, but
1381
with the last N elements 'None' rather than a regular bytestring.
1382
The first element cannot be 'None'.
1383
:return: An iterable as per iter_all_entries, but restricted to the
1384
keys with a matching prefix to those supplied. No additional keys
1385
will be returned, and every match that is in the index will be
1395
for index in self._indices:
1397
for node in index.iter_entries_prefix(keys):
1398
if node[1] in seen_keys:
1400
seen_keys.add(node[1])
1404
hit_indices.append(index)
1406
except errors.NoSuchFile:
1407
self._reload_or_raise()
1408
self._move_to_front(hit_indices)
1410
def _move_to_front(self, hit_indices):
1411
"""Rearrange self._indices so that hit_indices are first.
1413
Order is maintained as much as possible, e.g. the first unhit index
1414
will be the first index in _indices after the hit_indices, and the
1415
hit_indices will be present in exactly the order they are passed to
1418
_move_to_front propagates to all objects in self._sibling_indices by
1419
calling _move_to_front_by_name.
1421
if self._indices[:len(hit_indices)] == hit_indices:
1422
# The 'hit_indices' are already at the front (and in the same
1423
# order), no need to re-order
1425
hit_names = self._move_to_front_by_index(hit_indices)
1426
for sibling_idx in self._sibling_indices:
1427
sibling_idx._move_to_front_by_name(hit_names)
1429
def _move_to_front_by_index(self, hit_indices):
1430
"""Core logic for _move_to_front.
1432
Returns a list of names corresponding to the hit_indices param.
1434
indices_info = zip(self._index_names, self._indices)
1435
if 'index' in debug.debug_flags:
1436
mutter('CombinedGraphIndex reordering: currently %r, promoting %r',
1437
indices_info, hit_indices)
1440
new_hit_indices = []
1443
for offset, (name, idx) in enumerate(indices_info):
1444
if idx in hit_indices:
1445
hit_names.append(name)
1446
new_hit_indices.append(idx)
1447
if len(new_hit_indices) == len(hit_indices):
1448
# We've found all of the hit entries, everything else is
1450
unhit_names.extend(self._index_names[offset+1:])
1451
unhit_indices.extend(self._indices[offset+1:])
1454
unhit_names.append(name)
1455
unhit_indices.append(idx)
1457
self._indices = new_hit_indices + unhit_indices
1458
self._index_names = hit_names + unhit_names
1459
if 'index' in debug.debug_flags:
1460
mutter('CombinedGraphIndex reordered: %r', self._indices)
1463
def _move_to_front_by_name(self, hit_names):
1464
"""Moves indices named by 'hit_names' to front of the search order, as
1465
described in _move_to_front.
1467
# Translate names to index instances, and then call
1468
# _move_to_front_by_index.
1469
indices_info = zip(self._index_names, self._indices)
1471
for name, idx in indices_info:
1472
if name in hit_names:
1473
hit_indices.append(idx)
1474
self._move_to_front_by_index(hit_indices)
1476
def find_ancestry(self, keys, ref_list_num):
1477
"""Find the complete ancestry for the given set of keys.
1479
Note that this is a whole-ancestry request, so it should be used
1482
:param keys: An iterable of keys to look for
1483
:param ref_list_num: The reference list which references the parents
1485
:return: (parent_map, missing_keys)
1487
# XXX: make this call _move_to_front?
1488
missing_keys = set()
1490
keys_to_lookup = set(keys)
1492
while keys_to_lookup:
1493
# keys that *all* indexes claim are missing, stop searching them
1495
all_index_missing = None
1496
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1497
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1499
# len(missing_keys))
1500
for index_idx, index in enumerate(self._indices):
1501
# TODO: we should probably be doing something with
1502
# 'missing_keys' since we've already determined that
1503
# those revisions have not been found anywhere
1504
index_missing_keys = set()
1505
# Find all of the ancestry we can from this index
1506
# keep looking until the search_keys set is empty, which means
1507
# things we didn't find should be in index_missing_keys
1508
search_keys = keys_to_lookup
1510
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1511
# index_idx, len(search_keys),
1512
# len(parent_map), len(index_missing_keys))
1515
# TODO: ref_list_num should really be a parameter, since
1516
# CombinedGraphIndex does not know what the ref lists
1518
search_keys = index._find_ancestors(search_keys,
1519
ref_list_num, parent_map, index_missing_keys)
1520
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1521
# sub_generation, len(search_keys),
1522
# len(parent_map), len(index_missing_keys))
1523
# Now set whatever was missing to be searched in the next index
1524
keys_to_lookup = index_missing_keys
1525
if all_index_missing is None:
1526
all_index_missing = set(index_missing_keys)
1528
all_index_missing.intersection_update(index_missing_keys)
1529
if not keys_to_lookup:
1531
if all_index_missing is None:
1532
# There were no indexes, so all search keys are 'missing'
1533
missing_keys.update(keys_to_lookup)
1534
keys_to_lookup = None
1536
missing_keys.update(all_index_missing)
1537
keys_to_lookup.difference_update(all_index_missing)
1538
return parent_map, missing_keys
1540
def key_count(self):
1541
"""Return an estimate of the number of keys in this index.
1543
For CombinedGraphIndex this is approximated by the sum of the keys of
1544
the child indices. As child indices may have duplicate keys this can
1545
have a maximum error of the number of child indices * largest number of
1550
return sum((index.key_count() for index in self._indices), 0)
1551
except errors.NoSuchFile:
1552
self._reload_or_raise()
1554
missing_keys = _missing_keys_from_parent_map
1556
def _reload_or_raise(self):
1557
"""We just got a NoSuchFile exception.
1559
Try to reload the indices, if it fails, just raise the current
1562
if self._reload_func is None:
1564
exc_type, exc_value, exc_traceback = sys.exc_info()
1565
trace.mutter('Trying to reload after getting exception: %s',
1567
if not self._reload_func():
1568
# We tried to reload, but nothing changed, so we fail anyway
1569
trace.mutter('_reload_func indicated nothing has changed.'
1570
' Raising original exception.')
1571
raise exc_type, exc_value, exc_traceback
1573
def set_sibling_indices(self, sibling_combined_graph_indices):
1574
"""Set the CombinedGraphIndex objects to reorder after reordering self.
1576
self._sibling_indices = sibling_combined_graph_indices
342
for index in self._indices:
345
for node in index.iter_entries(keys):
1578
349
def validate(self):
1579
350
"""Validate that everything in the index can be accessed."""
1582
for index in self._indices:
1585
except errors.NoSuchFile:
1586
self._reload_or_raise()
351
for index in self._indices:
1589
355
class InMemoryGraphIndex(GraphIndexBuilder):
1600
366
:param nodes: An iterable of (key, node_refs, value) entries to add.
1602
if self.reference_lists:
1603
for (key, value, node_refs) in nodes:
1604
self.add_node(key, value, node_refs)
1606
for (key, value) in nodes:
1607
self.add_node(key, value)
368
for (key, value, node_refs) in nodes:
369
self.add_node(key, value, node_refs)
1609
371
def iter_all_entries(self):
1610
372
"""Iterate over all keys within the index
1612
:return: An iterable of (index, key, reference_lists, value). There is no
374
:return: An iterable of (key, reference_lists, value). There is no
1613
375
defined order for the result iteration - it will be in the most
1614
376
efficient order for the index (in this case dictionary hash order).
1616
if 'evil' in debug.debug_flags:
1617
trace.mutter_callsite(3,
1618
"iter_all_entries scales with size of history.")
1619
378
if self.reference_lists:
1620
379
for key, (absent, references, value) in self._nodes.iteritems():
1622
yield self, key, value, references
381
yield key, value, references
1624
383
for key, (absent, references, value) in self._nodes.iteritems():
1626
yield self, key, value
1628
387
def iter_entries(self, keys):
1629
388
"""Iterate over keys within the index.
1631
390
:param keys: An iterable providing the keys to be retrieved.
1632
:return: An iterable of (index, key, value, reference_lists). There is no
391
:return: An iterable of (key, reference_lists, value). There is no
1633
392
defined order for the result iteration - it will be in the most
1634
393
efficient order for the index (keys iteration order in this case).
1636
# Note: See BTreeBuilder.iter_entries for an explanation of why we
1637
# aren't using set().intersection() here
1639
keys = [key for key in keys if key in nodes]
1640
if self.reference_lists:
1644
yield self, key, node[2], node[1]
1649
yield self, key, node[2]
1651
def iter_entries_prefix(self, keys):
1652
"""Iterate over keys within the index using prefix matching.
1654
Prefix matching is applied within the tuple of a key, not to within
1655
the bytestring of each key element. e.g. if you have the keys ('foo',
1656
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1657
only the former key is returned.
1659
:param keys: An iterable providing the key prefixes to be retrieved.
1660
Each key prefix takes the form of a tuple the length of a key, but
1661
with the last N elements 'None' rather than a regular bytestring.
1662
The first element cannot be 'None'.
1663
:return: An iterable as per iter_all_entries, but restricted to the
1664
keys with a matching prefix to those supplied. No additional keys
1665
will be returned, and every match that is in the index will be
1668
# XXX: To much duplication with the GraphIndex class; consider finding
1669
# a good place to pull out the actual common logic.
1670
395
keys = set(keys)
1673
if self._key_length == 1:
1677
raise errors.BadIndexKey(key)
1678
if len(key) != self._key_length:
1679
raise errors.BadIndexKey(key)
1680
node = self._nodes[key]
1683
if self.reference_lists:
1684
yield self, key, node[2], node[1]
1686
yield self, key, node[2]
1688
nodes_by_key = self._get_nodes_by_key()
1692
raise errors.BadIndexKey(key)
1693
if len(key) != self._key_length:
1694
raise errors.BadIndexKey(key)
1695
# find what it refers to:
1696
key_dict = nodes_by_key
1697
elements = list(key)
1698
# find the subdict to return
1700
while len(elements) and elements[0] is not None:
1701
key_dict = key_dict[elements[0]]
1704
# a non-existant lookup.
1709
key_dict = dicts.pop(-1)
1710
# can't be empty or would not exist
1711
item, value = key_dict.iteritems().next()
1712
if type(value) == dict:
1714
dicts.extend(key_dict.itervalues())
1717
for value in key_dict.itervalues():
1718
yield (self, ) + value
1720
yield (self, ) + key_dict
1722
def key_count(self):
1723
"""Return an estimate of the number of keys in this index.
1725
For InMemoryGraphIndex the estimate is exact.
1727
return len(self._nodes) - len(self._absent_keys)
396
if self.reference_lists:
397
for key in keys.intersection(self._nodes):
398
node = self._nodes[key]
400
yield key, node[2], node[1]
402
for key in keys.intersection(self._nodes):
403
node = self._nodes[key]
1729
407
def validate(self):
1730
408
"""In memory index's have no known corruption at the moment."""
1733
class GraphIndexPrefixAdapter(object):
1734
"""An adapter between GraphIndex with different key lengths.
1736
Queries against this will emit queries against the adapted Graph with the
1737
prefix added, queries for all items use iter_entries_prefix. The returned
1738
nodes will have their keys and node references adjusted to remove the
1739
prefix. Finally, an add_nodes_callback can be supplied - when called the
1740
nodes and references being added will have prefix prepended.
1743
def __init__(self, adapted, prefix, missing_key_length,
1744
add_nodes_callback=None):
1745
"""Construct an adapter against adapted with prefix."""
1746
self.adapted = adapted
1747
self.prefix_key = prefix + (None,)*missing_key_length
1748
self.prefix = prefix
1749
self.prefix_len = len(prefix)
1750
self.add_nodes_callback = add_nodes_callback
1752
def add_nodes(self, nodes):
1753
"""Add nodes to the index.
1755
:param nodes: An iterable of (key, node_refs, value) entries to add.
1757
# save nodes in case its an iterator
1758
nodes = tuple(nodes)
1759
translated_nodes = []
1761
# Add prefix_key to each reference node_refs is a tuple of tuples,
1762
# so split it apart, and add prefix_key to the internal reference
1763
for (key, value, node_refs) in nodes:
1764
adjusted_references = (
1765
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1766
for ref_list in node_refs))
1767
translated_nodes.append((self.prefix + key, value,
1768
adjusted_references))
1770
# XXX: TODO add an explicit interface for getting the reference list
1771
# status, to handle this bit of user-friendliness in the API more
1773
for (key, value) in nodes:
1774
translated_nodes.append((self.prefix + key, value))
1775
self.add_nodes_callback(translated_nodes)
1777
def add_node(self, key, value, references=()):
1778
"""Add a node to the index.
1780
:param key: The key. keys are non-empty tuples containing
1781
as many whitespace-free utf8 bytestrings as the key length
1782
defined for this index.
1783
:param references: An iterable of iterables of keys. Each is a
1784
reference to another key.
1785
:param value: The value to associate with the key. It may be any
1786
bytes as long as it does not contain \0 or \n.
1788
self.add_nodes(((key, value, references), ))
1790
def _strip_prefix(self, an_iter):
1791
"""Strip prefix data from nodes and return it."""
1792
for node in an_iter:
1794
if node[1][:self.prefix_len] != self.prefix:
1795
raise errors.BadIndexData(self)
1796
for ref_list in node[3]:
1797
for ref_node in ref_list:
1798
if ref_node[:self.prefix_len] != self.prefix:
1799
raise errors.BadIndexData(self)
1800
yield node[0], node[1][self.prefix_len:], node[2], (
1801
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1802
for ref_list in node[3]))
1804
def iter_all_entries(self):
1805
"""Iterate over all keys within the index
1807
iter_all_entries is implemented against the adapted index using
1808
iter_entries_prefix.
1810
:return: An iterable of (index, key, reference_lists, value). There is no
1811
defined order for the result iteration - it will be in the most
1812
efficient order for the index (in this case dictionary hash order).
1814
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1816
def iter_entries(self, keys):
1817
"""Iterate over keys within the index.
1819
:param keys: An iterable providing the keys to be retrieved.
1820
:return: An iterable of (index, key, value, reference_lists). There is no
1821
defined order for the result iteration - it will be in the most
1822
efficient order for the index (keys iteration order in this case).
1824
return self._strip_prefix(self.adapted.iter_entries(
1825
self.prefix + key for key in keys))
1827
def iter_entries_prefix(self, keys):
1828
"""Iterate over keys within the index using prefix matching.
1830
Prefix matching is applied within the tuple of a key, not to within
1831
the bytestring of each key element. e.g. if you have the keys ('foo',
1832
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1833
only the former key is returned.
1835
:param keys: An iterable providing the key prefixes to be retrieved.
1836
Each key prefix takes the form of a tuple the length of a key, but
1837
with the last N elements 'None' rather than a regular bytestring.
1838
The first element cannot be 'None'.
1839
:return: An iterable as per iter_all_entries, but restricted to the
1840
keys with a matching prefix to those supplied. No additional keys
1841
will be returned, and every match that is in the index will be
1844
return self._strip_prefix(self.adapted.iter_entries_prefix(
1845
self.prefix + key for key in keys))
1847
def key_count(self):
1848
"""Return an estimate of the number of keys in this index.
1850
For GraphIndexPrefixAdapter this is relatively expensive - key
1851
iteration with the prefix is done.
1853
return len(list(self.iter_all_entries()))
1856
"""Call the adapted's validate."""
1857
self.adapted.validate()
added
removed
bzrlib/index.py
