1
# Copyright (C) 2007, 2008 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Indexing facilities."""
23
'GraphIndexPrefixAdapter',
27
from bisect import bisect_right
28
from cStringIO import StringIO
32
from bzrlib.lazy_import import lazy_import
33
lazy_import(globals(), """
34
from bzrlib import trace
35
from bzrlib.bisect_multi import bisect_multi_bytes
36
from bzrlib.revision import NULL_REVISION
37
from bzrlib.trace import mutter
45
_HEADER_READV = (0, 200)
46
_OPTION_KEY_ELEMENTS = "key_elements="
48
_OPTION_NODE_REFS = "node_ref_lists="
49
_SIGNATURE = "Bazaar Graph Index 1\n"
52
_whitespace_re = re.compile('[\t\n\x0b\x0c\r\x00 ]')
53
_newline_null_re = re.compile('[\n\0]')
56
def _has_key_from_parent_map(self, key):
57
"""Check if this index has one key.
59
If it's possible to check for multiple keys at once through
60
calling get_parent_map that should be faster.
62
return (key in self.get_parent_map([key]))
65
def _missing_keys_from_parent_map(self, keys):
66
return set(keys) - set(self.get_parent_map(keys))
69
class GraphIndexBuilder(object):
70
"""A builder that can build a GraphIndex.
72
The resulting graph has the structure:
74
_SIGNATURE OPTIONS NODES NEWLINE
75
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
76
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
78
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
79
KEY := Not-whitespace-utf8
81
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
82
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
83
REFERENCE := DIGITS ; digits is the byte offset in the index of the
85
VALUE := no-newline-no-null-bytes
88
def __init__(self, reference_lists=0, key_elements=1):
89
"""Create a GraphIndex builder.
91
:param reference_lists: The number of node references lists for each
93
:param key_elements: The number of bytestrings in each key.
95
self.reference_lists = reference_lists
97
# A dict of {key: (absent, ref_lists, value)}
99
self._nodes_by_key = None
100
self._key_length = key_elements
101
self._optimize_for_size = False
103
def _check_key(self, key):
104
"""Raise BadIndexKey if key is not a valid key for this index."""
105
if type(key) != tuple:
106
raise errors.BadIndexKey(key)
107
if self._key_length != len(key):
108
raise errors.BadIndexKey(key)
110
if not element or _whitespace_re.search(element) is not None:
111
raise errors.BadIndexKey(element)
113
def _external_references(self):
114
"""Return references that are not present in this index.
118
# TODO: JAM 2008-11-21 This makes an assumption about how the reference
119
# lists are used. It is currently correct for pack-0.92 through
120
# 1.9, which use the node references (3rd column) second
121
# reference list as the compression parent. Perhaps this should
122
# be moved into something higher up the stack, since it
123
# makes assumptions about how the index is used.
124
if self.reference_lists > 1:
125
for node in self.iter_all_entries():
127
refs.update(node[3][1])
130
# If reference_lists == 0 there can be no external references, and
131
# if reference_lists == 1, then there isn't a place to store the
135
def _get_nodes_by_key(self):
136
if self._nodes_by_key is None:
138
if self.reference_lists:
139
for key, (absent, references, value) in self._nodes.iteritems():
142
key_dict = nodes_by_key
143
for subkey in key[:-1]:
144
key_dict = key_dict.setdefault(subkey, {})
145
key_dict[key[-1]] = key, value, references
147
for key, (absent, references, value) in self._nodes.iteritems():
150
key_dict = nodes_by_key
151
for subkey in key[:-1]:
152
key_dict = key_dict.setdefault(subkey, {})
153
key_dict[key[-1]] = key, value
154
self._nodes_by_key = nodes_by_key
155
return self._nodes_by_key
157
def _update_nodes_by_key(self, key, value, node_refs):
158
"""Update the _nodes_by_key dict with a new key.
160
For a key of (foo, bar, baz) create
161
_nodes_by_key[foo][bar][baz] = key_value
163
if self._nodes_by_key is None:
165
key_dict = self._nodes_by_key
166
if self.reference_lists:
167
key_value = key, value, node_refs
169
key_value = key, value
170
for subkey in key[:-1]:
171
key_dict = key_dict.setdefault(subkey, {})
172
key_dict[key[-1]] = key_value
174
def _check_key_ref_value(self, key, references, value):
175
"""Check that 'key' and 'references' are all valid.
177
:param key: A key tuple. Must conform to the key interface (be a tuple,
178
be of the right length, not have any whitespace or nulls in any key
180
:param references: An iterable of reference lists. Something like
181
[[(ref, key)], [(ref, key), (other, key)]]
182
:param value: The value associate with this key. Must not contain
183
newlines or null characters.
184
:return: (node_refs, absent_references)
185
node_refs basically a packed form of 'references' where all
187
absent_references reference keys that are not in self._nodes.
188
This may contain duplicates if the same key is
189
referenced in multiple lists.
192
if _newline_null_re.search(value) is not None:
193
raise errors.BadIndexValue(value)
194
if len(references) != self.reference_lists:
195
raise errors.BadIndexValue(references)
197
absent_references = []
198
for reference_list in references:
199
for reference in reference_list:
200
# If reference *is* in self._nodes, then we know it has already
202
if reference not in self._nodes:
203
self._check_key(reference)
204
absent_references.append(reference)
205
node_refs.append(tuple(reference_list))
206
return tuple(node_refs), absent_references
208
def add_node(self, key, value, references=()):
209
"""Add a node to the index.
211
:param key: The key. keys are non-empty tuples containing
212
as many whitespace-free utf8 bytestrings as the key length
213
defined for this index.
214
:param references: An iterable of iterables of keys. Each is a
215
reference to another key.
216
:param value: The value to associate with the key. It may be any
217
bytes as long as it does not contain \0 or \n.
220
absent_references) = self._check_key_ref_value(key, references, value)
221
if key in self._nodes and self._nodes[key][0] != 'a':
222
raise errors.BadIndexDuplicateKey(key, self)
223
for reference in absent_references:
224
# There may be duplicates, but I don't think it is worth worrying
226
self._nodes[reference] = ('a', (), '')
227
self._nodes[key] = ('', node_refs, value)
229
if self._nodes_by_key is not None and self._key_length > 1:
230
self._update_nodes_by_key(key, value, node_refs)
234
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
235
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
236
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
237
prefix_length = sum(len(x) for x in lines)
238
# references are byte offsets. To avoid having to do nasty
239
# polynomial work to resolve offsets (references to later in the
240
# file cannot be determined until all the inbetween references have
241
# been calculated too) we pad the offsets with 0's to make them be
242
# of consistent length. Using binary offsets would break the trivial
244
# to calculate the width of zero's needed we do three passes:
245
# one to gather all the non-reference data and the number of references.
246
# one to pad all the data with reference-length and determine entry
250
# forward sorted by key. In future we may consider topological sorting,
251
# at the cost of table scans for direct lookup, or a second index for
253
nodes = sorted(self._nodes.items())
254
# if we do not prepass, we don't know how long it will be up front.
255
expected_bytes = None
256
# we only need to pre-pass if we have reference lists at all.
257
if self.reference_lists:
259
non_ref_bytes = prefix_length
261
# TODO use simple multiplication for the constants in this loop.
262
for key, (absent, references, value) in nodes:
263
# record the offset known *so far* for this key:
264
# the non reference bytes to date, and the total references to
265
# date - saves reaccumulating on the second pass
266
key_offset_info.append((key, non_ref_bytes, total_references))
267
# key is literal, value is literal, there are 3 null's, 1 NL
268
# key is variable length tuple, \x00 between elements
269
non_ref_bytes += sum(len(element) for element in key)
270
if self._key_length > 1:
271
non_ref_bytes += self._key_length - 1
272
# value is literal bytes, there are 3 null's, 1 NL.
273
non_ref_bytes += len(value) + 3 + 1
274
# one byte for absent if set.
277
elif self.reference_lists:
278
# (ref_lists -1) tabs
279
non_ref_bytes += self.reference_lists - 1
280
# (ref-1 cr's per ref_list)
281
for ref_list in references:
282
# how many references across the whole file?
283
total_references += len(ref_list)
284
# accrue reference separators
286
non_ref_bytes += len(ref_list) - 1
287
# how many digits are needed to represent the total byte count?
289
possible_total_bytes = non_ref_bytes + total_references*digits
290
while 10 ** digits < possible_total_bytes:
292
possible_total_bytes = non_ref_bytes + total_references*digits
293
expected_bytes = possible_total_bytes + 1 # terminating newline
294
# resolve key addresses.
296
for key, non_ref_bytes, total_references in key_offset_info:
297
key_addresses[key] = non_ref_bytes + total_references*digits
299
format_string = '%%0%sd' % digits
300
for key, (absent, references, value) in nodes:
301
flattened_references = []
302
for ref_list in references:
304
for reference in ref_list:
305
ref_addresses.append(format_string % key_addresses[reference])
306
flattened_references.append('\r'.join(ref_addresses))
307
string_key = '\x00'.join(key)
308
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
309
'\t'.join(flattened_references), value))
311
result = StringIO(''.join(lines))
312
if expected_bytes and len(result.getvalue()) != expected_bytes:
313
raise errors.BzrError('Failed index creation. Internal error:'
314
' mismatched output length and expected length: %d %d' %
315
(len(result.getvalue()), expected_bytes))
318
def set_optimize(self, for_size=True):
319
"""Change how the builder tries to optimize the result.
321
:param for_size: Tell the builder to try and make the index as small as
325
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
327
self._optimize_for_size = for_size
330
class GraphIndex(object):
331
"""An index for data with embedded graphs.
333
The index maps keys to a list of key reference lists, and a value.
334
Each node has the same number of key reference lists. Each key reference
335
list can be empty or an arbitrary length. The value is an opaque NULL
336
terminated string without any newlines. The storage of the index is
337
hidden in the interface: keys and key references are always tuples of
338
bytestrings, never the internal representation (e.g. dictionary offsets).
340
It is presumed that the index will not be mutated - it is static data.
342
Successive iter_all_entries calls will read the entire index each time.
343
Additionally, iter_entries calls will read the index linearly until the
344
desired keys are found. XXX: This must be fixed before the index is
345
suitable for production use. :XXX
348
def __init__(self, transport, name, size):
349
"""Open an index called name on transport.
351
:param transport: A bzrlib.transport.Transport.
352
:param name: A path to provide to transport API calls.
353
:param size: The size of the index in bytes. This is used for bisection
354
logic to perform partial index reads. While the size could be
355
obtained by statting the file this introduced an additional round
356
trip as well as requiring stat'able transports, both of which are
357
avoided by having it supplied. If size is None, then bisection
358
support will be disabled and accessing the index will just stream
361
self._transport = transport
363
# Becomes a dict of key:(value, reference-list-byte-locations) used by
364
# the bisection interface to store parsed but not resolved keys.
365
self._bisect_nodes = None
366
# Becomes a dict of key:(value, reference-list-keys) which are ready to
367
# be returned directly to callers.
369
# a sorted list of slice-addresses for the parsed bytes of the file.
370
# e.g. (0,1) would mean that byte 0 is parsed.
371
self._parsed_byte_map = []
372
# a sorted list of keys matching each slice address for parsed bytes
373
# e.g. (None, 'foo@bar') would mean that the first byte contained no
374
# key, and the end byte of the slice is the of the data for 'foo@bar'
375
self._parsed_key_map = []
376
self._key_count = None
377
self._keys_by_offset = None
378
self._nodes_by_key = None
380
# The number of bytes we've read so far in trying to process this file
383
def __eq__(self, other):
384
"""Equal when self and other were created with the same parameters."""
386
type(self) == type(other) and
387
self._transport == other._transport and
388
self._name == other._name and
389
self._size == other._size)
391
def __ne__(self, other):
392
return not self.__eq__(other)
395
return "%s(%r)" % (self.__class__.__name__,
396
self._transport.abspath(self._name))
398
def _buffer_all(self, stream=None):
399
"""Buffer all the index data.
401
Mutates self._nodes and self.keys_by_offset.
403
if self._nodes is not None:
404
# We already did this
406
if 'index' in debug.debug_flags:
407
mutter('Reading entire index %s', self._transport.abspath(self._name))
409
stream = self._transport.get(self._name)
410
self._read_prefix(stream)
411
self._expected_elements = 3 + self._key_length
413
# raw data keyed by offset
414
self._keys_by_offset = {}
415
# ready-to-return key:value or key:value, node_ref_lists
417
self._nodes_by_key = None
420
lines = stream.read().split('\n')
422
_, _, _, trailers = self._parse_lines(lines, pos)
423
for key, absent, references, value in self._keys_by_offset.itervalues():
426
# resolve references:
427
if self.node_ref_lists:
428
node_value = (value, self._resolve_references(references))
431
self._nodes[key] = node_value
432
# cache the keys for quick set intersections
433
self._keys = set(self._nodes)
435
# there must be one line - the empty trailer line.
436
raise errors.BadIndexData(self)
438
def external_references(self, ref_list_num):
439
"""Return references that are not present in this index.
442
if ref_list_num + 1 > self.node_ref_lists:
443
raise ValueError('No ref list %d, index has %d ref lists'
444
% (ref_list_num, self.node_ref_lists))
446
for key, (value, ref_lists) in self._nodes.iteritems():
447
ref_list = ref_lists[ref_list_num]
448
refs.update(ref_list)
449
return refs - self._keys
451
def _get_nodes_by_key(self):
452
if self._nodes_by_key is None:
454
if self.node_ref_lists:
455
for key, (value, references) in self._nodes.iteritems():
456
key_dict = nodes_by_key
457
for subkey in key[:-1]:
458
key_dict = key_dict.setdefault(subkey, {})
459
key_dict[key[-1]] = key, value, references
461
for key, value in self._nodes.iteritems():
462
key_dict = nodes_by_key
463
for subkey in key[:-1]:
464
key_dict = key_dict.setdefault(subkey, {})
465
key_dict[key[-1]] = key, value
466
self._nodes_by_key = nodes_by_key
467
return self._nodes_by_key
469
def iter_all_entries(self):
470
"""Iterate over all keys within the index.
472
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
473
The former tuple is used when there are no reference lists in the
474
index, making the API compatible with simple key:value index types.
475
There is no defined order for the result iteration - it will be in
476
the most efficient order for the index.
478
if 'evil' in debug.debug_flags:
479
trace.mutter_callsite(3,
480
"iter_all_entries scales with size of history.")
481
if self._nodes is None:
483
if self.node_ref_lists:
484
for key, (value, node_ref_lists) in self._nodes.iteritems():
485
yield self, key, value, node_ref_lists
487
for key, value in self._nodes.iteritems():
488
yield self, key, value
490
def _read_prefix(self, stream):
491
signature = stream.read(len(self._signature()))
492
if not signature == self._signature():
493
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
494
options_line = stream.readline()
495
if not options_line.startswith(_OPTION_NODE_REFS):
496
raise errors.BadIndexOptions(self)
498
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
500
raise errors.BadIndexOptions(self)
501
options_line = stream.readline()
502
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
503
raise errors.BadIndexOptions(self)
505
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
507
raise errors.BadIndexOptions(self)
508
options_line = stream.readline()
509
if not options_line.startswith(_OPTION_LEN):
510
raise errors.BadIndexOptions(self)
512
self._key_count = int(options_line[len(_OPTION_LEN):-1])
514
raise errors.BadIndexOptions(self)
516
def _resolve_references(self, references):
517
"""Return the resolved key references for references.
519
References are resolved by looking up the location of the key in the
520
_keys_by_offset map and substituting the key name, preserving ordering.
522
:param references: An iterable of iterables of key locations. e.g.
524
:return: A tuple of tuples of keys.
527
for ref_list in references:
528
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
529
return tuple(node_refs)
531
def _find_index(self, range_map, key):
532
"""Helper for the _parsed_*_index calls.
534
Given a range map - [(start, end), ...], finds the index of the range
535
in the map for key if it is in the map, and if it is not there, the
536
immediately preceeding range in the map.
538
result = bisect_right(range_map, key) - 1
539
if result + 1 < len(range_map):
540
# check the border condition, it may be in result + 1
541
if range_map[result + 1][0] == key[0]:
545
def _parsed_byte_index(self, offset):
546
"""Return the index of the entry immediately before offset.
548
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
549
there is one unparsed byte (the 11th, addressed as[10]). then:
550
asking for 0 will return 0
551
asking for 10 will return 0
552
asking for 11 will return 1
553
asking for 12 will return 1
556
return self._find_index(self._parsed_byte_map, key)
558
def _parsed_key_index(self, key):
559
"""Return the index of the entry immediately before key.
561
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
562
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
563
have been parsed, then:
564
asking for '' will return 0
565
asking for 'a' will return 0
566
asking for 'b' will return 1
567
asking for 'e' will return 1
569
search_key = (key, None)
570
return self._find_index(self._parsed_key_map, search_key)
572
def _is_parsed(self, offset):
573
"""Returns True if offset has been parsed."""
574
index = self._parsed_byte_index(offset)
575
if index == len(self._parsed_byte_map):
576
return offset < self._parsed_byte_map[index - 1][1]
577
start, end = self._parsed_byte_map[index]
578
return offset >= start and offset < end
580
def _iter_entries_from_total_buffer(self, keys):
581
"""Iterate over keys when the entire index is parsed."""
582
keys = keys.intersection(self._keys)
583
if self.node_ref_lists:
585
value, node_refs = self._nodes[key]
586
yield self, key, value, node_refs
589
yield self, key, self._nodes[key]
591
def iter_entries(self, keys):
592
"""Iterate over keys within the index.
594
:param keys: An iterable providing the keys to be retrieved.
595
:return: An iterable as per iter_all_entries, but restricted to the
596
keys supplied. No additional keys will be returned, and every
597
key supplied that is in the index will be returned.
602
if self._size is None and self._nodes is None:
605
# We fit about 20 keys per minimum-read (4K), so if we are looking for
606
# more than 1/20th of the index its likely (assuming homogenous key
607
# spread) that we'll read the entire index. If we're going to do that,
608
# buffer the whole thing. A better analysis might take key spread into
609
# account - but B+Tree indices are better anyway.
610
# We could look at all data read, and use a threshold there, which will
611
# trigger on ancestry walks, but that is not yet fully mapped out.
612
if self._nodes is None and len(keys) * 20 > self.key_count():
614
if self._nodes is not None:
615
return self._iter_entries_from_total_buffer(keys)
617
return (result[1] for result in bisect_multi_bytes(
618
self._lookup_keys_via_location, self._size, keys))
620
def iter_entries_prefix(self, keys):
621
"""Iterate over keys within the index using prefix matching.
623
Prefix matching is applied within the tuple of a key, not to within
624
the bytestring of each key element. e.g. if you have the keys ('foo',
625
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
626
only the former key is returned.
628
WARNING: Note that this method currently causes a full index parse
629
unconditionally (which is reasonably appropriate as it is a means for
630
thunking many small indices into one larger one and still supplies
631
iter_all_entries at the thunk layer).
633
:param keys: An iterable providing the key prefixes to be retrieved.
634
Each key prefix takes the form of a tuple the length of a key, but
635
with the last N elements 'None' rather than a regular bytestring.
636
The first element cannot be 'None'.
637
:return: An iterable as per iter_all_entries, but restricted to the
638
keys with a matching prefix to those supplied. No additional keys
639
will be returned, and every match that is in the index will be
645
# load data - also finds key lengths
646
if self._nodes is None:
648
if self._key_length == 1:
652
raise errors.BadIndexKey(key)
653
if len(key) != self._key_length:
654
raise errors.BadIndexKey(key)
655
if self.node_ref_lists:
656
value, node_refs = self._nodes[key]
657
yield self, key, value, node_refs
659
yield self, key, self._nodes[key]
661
nodes_by_key = self._get_nodes_by_key()
665
raise errors.BadIndexKey(key)
666
if len(key) != self._key_length:
667
raise errors.BadIndexKey(key)
668
# find what it refers to:
669
key_dict = nodes_by_key
671
# find the subdict whose contents should be returned.
673
while len(elements) and elements[0] is not None:
674
key_dict = key_dict[elements[0]]
677
# a non-existant lookup.
682
key_dict = dicts.pop(-1)
683
# can't be empty or would not exist
684
item, value = key_dict.iteritems().next()
685
if type(value) == dict:
687
dicts.extend(key_dict.itervalues())
690
for value in key_dict.itervalues():
691
# each value is the key:value:node refs tuple
693
yield (self, ) + value
695
# the last thing looked up was a terminal element
696
yield (self, ) + key_dict
699
"""Return an estimate of the number of keys in this index.
701
For GraphIndex the estimate is exact.
703
if self._key_count is None:
704
self._read_and_parse([_HEADER_READV])
705
return self._key_count
707
def _lookup_keys_via_location(self, location_keys):
708
"""Public interface for implementing bisection.
710
If _buffer_all has been called, then all the data for the index is in
711
memory, and this method should not be called, as it uses a separate
712
cache because it cannot pre-resolve all indices, which buffer_all does
715
:param location_keys: A list of location(byte offset), key tuples.
716
:return: A list of (location_key, result) tuples as expected by
717
bzrlib.bisect_multi.bisect_multi_bytes.
719
# Possible improvements:
720
# - only bisect lookup each key once
721
# - sort the keys first, and use that to reduce the bisection window
723
# this progresses in three parts:
726
# attempt to answer the question from the now in memory data.
727
# build the readv request
728
# for each location, ask for 800 bytes - much more than rows we've seen
731
for location, key in location_keys:
732
# can we answer from cache?
733
if self._bisect_nodes and key in self._bisect_nodes:
734
# We have the key parsed.
736
index = self._parsed_key_index(key)
737
if (len(self._parsed_key_map) and
738
self._parsed_key_map[index][0] <= key and
739
(self._parsed_key_map[index][1] >= key or
740
# end of the file has been parsed
741
self._parsed_byte_map[index][1] == self._size)):
742
# the key has been parsed, so no lookup is needed even if its
745
# - if we have examined this part of the file already - yes
746
index = self._parsed_byte_index(location)
747
if (len(self._parsed_byte_map) and
748
self._parsed_byte_map[index][0] <= location and
749
self._parsed_byte_map[index][1] > location):
750
# the byte region has been parsed, so no read is needed.
753
if location + length > self._size:
754
length = self._size - location
755
# todo, trim out parsed locations.
757
readv_ranges.append((location, length))
758
# read the header if needed
759
if self._bisect_nodes is None:
760
readv_ranges.append(_HEADER_READV)
761
self._read_and_parse(readv_ranges)
763
if self._nodes is not None:
764
# _read_and_parse triggered a _buffer_all because we requested the
766
for location, key in location_keys:
767
if key not in self._nodes: # not present
768
result.append(((location, key), False))
769
elif self.node_ref_lists:
770
value, refs = self._nodes[key]
771
result.append(((location, key),
772
(self, key, value, refs)))
774
result.append(((location, key),
775
(self, key, self._nodes[key])))
778
# - figure out <, >, missing, present
779
# - result present references so we can return them.
780
# keys that we cannot answer until we resolve references
781
pending_references = []
782
pending_locations = set()
783
for location, key in location_keys:
784
# can we answer from cache?
785
if key in self._bisect_nodes:
786
# the key has been parsed, so no lookup is needed
787
if self.node_ref_lists:
788
# the references may not have been all parsed.
789
value, refs = self._bisect_nodes[key]
790
wanted_locations = []
791
for ref_list in refs:
793
if ref not in self._keys_by_offset:
794
wanted_locations.append(ref)
796
pending_locations.update(wanted_locations)
797
pending_references.append((location, key))
799
result.append(((location, key), (self, key,
800
value, self._resolve_references(refs))))
802
result.append(((location, key),
803
(self, key, self._bisect_nodes[key])))
806
# has the region the key should be in, been parsed?
807
index = self._parsed_key_index(key)
808
if (self._parsed_key_map[index][0] <= key and
809
(self._parsed_key_map[index][1] >= key or
810
# end of the file has been parsed
811
self._parsed_byte_map[index][1] == self._size)):
812
result.append(((location, key), False))
814
# no, is the key above or below the probed location:
815
# get the range of the probed & parsed location
816
index = self._parsed_byte_index(location)
817
# if the key is below the start of the range, its below
818
if key < self._parsed_key_map[index][0]:
822
result.append(((location, key), direction))
824
# lookup data to resolve references
825
for location in pending_locations:
827
if location + length > self._size:
828
length = self._size - location
829
# TODO: trim out parsed locations (e.g. if the 800 is into the
830
# parsed region trim it, and dont use the adjust_for_latency
833
readv_ranges.append((location, length))
834
self._read_and_parse(readv_ranges)
835
if self._nodes is not None:
836
# The _read_and_parse triggered a _buffer_all, grab the data and
838
for location, key in pending_references:
839
value, refs = self._nodes[key]
840
result.append(((location, key), (self, key, value, refs)))
842
for location, key in pending_references:
843
# answer key references we had to look-up-late.
844
value, refs = self._bisect_nodes[key]
845
result.append(((location, key), (self, key,
846
value, self._resolve_references(refs))))
849
def _parse_header_from_bytes(self, bytes):
850
"""Parse the header from a region of bytes.
852
:param bytes: The data to parse.
853
:return: An offset, data tuple such as readv yields, for the unparsed
854
data. (which may length 0).
856
signature = bytes[0:len(self._signature())]
857
if not signature == self._signature():
858
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
859
lines = bytes[len(self._signature()):].splitlines()
860
options_line = lines[0]
861
if not options_line.startswith(_OPTION_NODE_REFS):
862
raise errors.BadIndexOptions(self)
864
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
866
raise errors.BadIndexOptions(self)
867
options_line = lines[1]
868
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
869
raise errors.BadIndexOptions(self)
871
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
873
raise errors.BadIndexOptions(self)
874
options_line = lines[2]
875
if not options_line.startswith(_OPTION_LEN):
876
raise errors.BadIndexOptions(self)
878
self._key_count = int(options_line[len(_OPTION_LEN):])
880
raise errors.BadIndexOptions(self)
881
# calculate the bytes we have processed
882
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
884
self._parsed_bytes(0, None, header_end, None)
885
# setup parsing state
886
self._expected_elements = 3 + self._key_length
887
# raw data keyed by offset
888
self._keys_by_offset = {}
889
# keys with the value and node references
890
self._bisect_nodes = {}
891
return header_end, bytes[header_end:]
893
def _parse_region(self, offset, data):
894
"""Parse node data returned from a readv operation.
896
:param offset: The byte offset the data starts at.
897
:param data: The data to parse.
901
end = offset + len(data)
904
# Trivial test - if the current index's end is within the
905
# low-matching parsed range, we're done.
906
index = self._parsed_byte_index(high_parsed)
907
if end < self._parsed_byte_map[index][1]:
909
# print "[%d:%d]" % (offset, end), \
910
# self._parsed_byte_map[index:index + 2]
911
high_parsed, last_segment = self._parse_segment(
912
offset, data, end, index)
916
def _parse_segment(self, offset, data, end, index):
917
"""Parse one segment of data.
919
:param offset: Where 'data' begins in the file.
920
:param data: Some data to parse a segment of.
921
:param end: Where data ends
922
:param index: The current index into the parsed bytes map.
923
:return: True if the parsed segment is the last possible one in the
925
:return: high_parsed_byte, last_segment.
926
high_parsed_byte is the location of the highest parsed byte in this
927
segment, last_segment is True if the parsed segment is the last
928
possible one in the data block.
930
# default is to use all data
932
# accomodate overlap with data before this.
933
if offset < self._parsed_byte_map[index][1]:
934
# overlaps the lower parsed region
935
# skip the parsed data
936
trim_start = self._parsed_byte_map[index][1] - offset
937
# don't trim the start for \n
938
start_adjacent = True
939
elif offset == self._parsed_byte_map[index][1]:
940
# abuts the lower parsed region
943
# do not trim anything
944
start_adjacent = True
946
# does not overlap the lower parsed region
949
# but trim the leading \n
950
start_adjacent = False
951
if end == self._size:
952
# lines up to the end of all data:
955
# do not strip to the last \n
958
elif index + 1 == len(self._parsed_byte_map):
959
# at the end of the parsed data
962
# but strip to the last \n
965
elif end == self._parsed_byte_map[index + 1][0]:
966
# buts up against the next parsed region
969
# do not strip to the last \n
972
elif end > self._parsed_byte_map[index + 1][0]:
973
# overlaps into the next parsed region
974
# only consider the unparsed data
975
trim_end = self._parsed_byte_map[index + 1][0] - offset
976
# do not strip to the last \n as we know its an entire record
978
last_segment = end < self._parsed_byte_map[index + 1][1]
980
# does not overlap into the next region
983
# but strip to the last \n
986
# now find bytes to discard if needed
987
if not start_adjacent:
988
# work around python bug in rfind
989
if trim_start is None:
990
trim_start = data.find('\n') + 1
992
trim_start = data.find('\n', trim_start) + 1
993
if not (trim_start != 0):
994
raise AssertionError('no \n was present')
995
# print 'removing start', offset, trim_start, repr(data[:trim_start])
997
# work around python bug in rfind
999
trim_end = data.rfind('\n') + 1
1001
trim_end = data.rfind('\n', None, trim_end) + 1
1002
if not (trim_end != 0):
1003
raise AssertionError('no \n was present')
1004
# print 'removing end', offset, trim_end, repr(data[trim_end:])
1005
# adjust offset and data to the parseable data.
1006
trimmed_data = data[trim_start:trim_end]
1007
if not (trimmed_data):
1008
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
1009
% (trim_start, trim_end, offset, offset + len(data)))
1011
offset += trim_start
1012
# print "parsing", repr(trimmed_data)
1013
# splitlines mangles the \r delimiters.. don't use it.
1014
lines = trimmed_data.split('\n')
1017
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1018
for key, value in nodes:
1019
self._bisect_nodes[key] = value
1020
self._parsed_bytes(offset, first_key,
1021
offset + len(trimmed_data), last_key)
1022
return offset + len(trimmed_data), last_segment
1024
def _parse_lines(self, lines, pos):
1031
# must be at the end
1033
if not (self._size == pos + 1):
1034
raise AssertionError("%s %s" % (self._size, pos))
1037
elements = line.split('\0')
1038
if len(elements) != self._expected_elements:
1039
raise errors.BadIndexData(self)
1040
# keys are tuples. Each element is a string that may occur many
1041
# times, so we intern them to save space. AB, RC, 200807
1042
key = tuple([intern(element) for element in elements[:self._key_length]])
1043
if first_key is None:
1045
absent, references, value = elements[-3:]
1047
for ref_string in references.split('\t'):
1048
ref_lists.append(tuple([
1049
int(ref) for ref in ref_string.split('\r') if ref
1051
ref_lists = tuple(ref_lists)
1052
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1053
pos += len(line) + 1 # +1 for the \n
1056
if self.node_ref_lists:
1057
node_value = (value, ref_lists)
1060
nodes.append((key, node_value))
1061
# print "parsed ", key
1062
return first_key, key, nodes, trailers
1064
def _parsed_bytes(self, start, start_key, end, end_key):
1065
"""Mark the bytes from start to end as parsed.
1067
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1070
:param start: The start of the parsed region.
1071
:param end: The end of the parsed region.
1073
index = self._parsed_byte_index(start)
1074
new_value = (start, end)
1075
new_key = (start_key, end_key)
1077
# first range parsed is always the beginning.
1078
self._parsed_byte_map.insert(index, new_value)
1079
self._parsed_key_map.insert(index, new_key)
1083
# extend lower region
1084
# extend higher region
1085
# combine two regions
1086
if (index + 1 < len(self._parsed_byte_map) and
1087
self._parsed_byte_map[index][1] == start and
1088
self._parsed_byte_map[index + 1][0] == end):
1089
# combine two regions
1090
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1091
self._parsed_byte_map[index + 1][1])
1092
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1093
self._parsed_key_map[index + 1][1])
1094
del self._parsed_byte_map[index + 1]
1095
del self._parsed_key_map[index + 1]
1096
elif self._parsed_byte_map[index][1] == start:
1097
# extend the lower entry
1098
self._parsed_byte_map[index] = (
1099
self._parsed_byte_map[index][0], end)
1100
self._parsed_key_map[index] = (
1101
self._parsed_key_map[index][0], end_key)
1102
elif (index + 1 < len(self._parsed_byte_map) and
1103
self._parsed_byte_map[index + 1][0] == end):
1104
# extend the higher entry
1105
self._parsed_byte_map[index + 1] = (
1106
start, self._parsed_byte_map[index + 1][1])
1107
self._parsed_key_map[index + 1] = (
1108
start_key, self._parsed_key_map[index + 1][1])
1111
self._parsed_byte_map.insert(index + 1, new_value)
1112
self._parsed_key_map.insert(index + 1, new_key)
1114
def _read_and_parse(self, readv_ranges):
1115
"""Read the the ranges and parse the resulting data.
1117
:param readv_ranges: A prepared readv range list.
1119
if not readv_ranges:
1121
if self._nodes is None and self._bytes_read * 2 >= self._size:
1122
# We've already read more than 50% of the file and we are about to
1123
# request more data, just _buffer_all() and be done
1127
readv_data = self._transport.readv(self._name, readv_ranges, True,
1130
for offset, data in readv_data:
1131
self._bytes_read += len(data)
1132
if offset == 0 and len(data) == self._size:
1133
# We read the whole range, most likely because the
1134
# Transport upcast our readv ranges into one long request
1135
# for enough total data to grab the whole index.
1136
self._buffer_all(StringIO(data))
1138
if self._bisect_nodes is None:
1139
# this must be the start
1140
if not (offset == 0):
1141
raise AssertionError()
1142
offset, data = self._parse_header_from_bytes(data)
1143
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1144
self._parse_region(offset, data)
1146
def _signature(self):
1147
"""The file signature for this index type."""
1151
"""Validate that everything in the index can be accessed."""
1152
# iter_all validates completely at the moment, so just do that.
1153
for node in self.iter_all_entries():
1157
class CombinedGraphIndex(object):
1158
"""A GraphIndex made up from smaller GraphIndices.
1160
The backing indices must implement GraphIndex, and are presumed to be
1163
Queries against the combined index will be made against the first index,
1164
and then the second and so on. The order of index's can thus influence
1165
performance significantly. For example, if one index is on local disk and a
1166
second on a remote server, the local disk index should be before the other
1170
def __init__(self, indices, reload_func=None):
1171
"""Create a CombinedGraphIndex backed by indices.
1173
:param indices: An ordered list of indices to query for data.
1174
:param reload_func: A function to call if we find we are missing an
1175
index. Should have the form reload_func() => True/False to indicate
1176
if reloading actually changed anything.
1178
self._indices = indices
1179
self._reload_func = reload_func
1183
self.__class__.__name__,
1184
', '.join(map(repr, self._indices)))
1186
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1187
def get_parents(self, revision_ids):
1188
"""See graph._StackedParentsProvider.get_parents.
1190
This implementation thunks the graph.Graph.get_parents api across to
1193
:param revision_ids: An iterable of graph keys for this graph.
1194
:return: A list of parent details for each key in revision_ids.
1195
Each parent details will be one of:
1196
* None when the key was missing
1197
* (NULL_REVISION,) when the key has no parents.
1198
* (parent_key, parent_key...) otherwise.
1200
parent_map = self.get_parent_map(revision_ids)
1201
return [parent_map.get(r, None) for r in revision_ids]
1203
def get_parent_map(self, keys):
1204
"""See graph._StackedParentsProvider.get_parent_map"""
1205
search_keys = set(keys)
1206
if NULL_REVISION in search_keys:
1207
search_keys.discard(NULL_REVISION)
1208
found_parents = {NULL_REVISION:[]}
1211
for index, key, value, refs in self.iter_entries(search_keys):
1214
parents = (NULL_REVISION,)
1215
found_parents[key] = parents
1216
return found_parents
1218
has_key = _has_key_from_parent_map
1220
def insert_index(self, pos, index):
1221
"""Insert a new index in the list of indices to query.
1223
:param pos: The position to insert the index.
1224
:param index: The index to insert.
1226
self._indices.insert(pos, index)
1228
def iter_all_entries(self):
1229
"""Iterate over all keys within the index
1231
Duplicate keys across child indices are presumed to have the same
1232
value and are only reported once.
1234
:return: An iterable of (index, key, reference_lists, value).
1235
There is no defined order for the result iteration - it will be in
1236
the most efficient order for the index.
1241
for index in self._indices:
1242
for node in index.iter_all_entries():
1243
if node[1] not in seen_keys:
1245
seen_keys.add(node[1])
1247
except errors.NoSuchFile:
1248
self._reload_or_raise()
1250
def iter_entries(self, keys):
1251
"""Iterate over keys within the index.
1253
Duplicate keys across child indices are presumed to have the same
1254
value and are only reported once.
1256
:param keys: An iterable providing the keys to be retrieved.
1257
:return: An iterable of (index, key, reference_lists, value). There is no
1258
defined order for the result iteration - it will be in the most
1259
efficient order for the index.
1264
for index in self._indices:
1267
for node in index.iter_entries(keys):
1268
keys.remove(node[1])
1271
except errors.NoSuchFile:
1272
self._reload_or_raise()
1274
def iter_entries_prefix(self, keys):
1275
"""Iterate over keys within the index using prefix matching.
1277
Duplicate keys across child indices are presumed to have the same
1278
value and are only reported once.
1280
Prefix matching is applied within the tuple of a key, not to within
1281
the bytestring of each key element. e.g. if you have the keys ('foo',
1282
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1283
only the former key is returned.
1285
:param keys: An iterable providing the key prefixes to be retrieved.
1286
Each key prefix takes the form of a tuple the length of a key, but
1287
with the last N elements 'None' rather than a regular bytestring.
1288
The first element cannot be 'None'.
1289
:return: An iterable as per iter_all_entries, but restricted to the
1290
keys with a matching prefix to those supplied. No additional keys
1291
will be returned, and every match that is in the index will be
1300
for index in self._indices:
1301
for node in index.iter_entries_prefix(keys):
1302
if node[1] in seen_keys:
1304
seen_keys.add(node[1])
1307
except errors.NoSuchFile:
1308
self._reload_or_raise()
1310
def key_count(self):
1311
"""Return an estimate of the number of keys in this index.
1313
For CombinedGraphIndex this is approximated by the sum of the keys of
1314
the child indices. As child indices may have duplicate keys this can
1315
have a maximum error of the number of child indices * largest number of
1320
return sum((index.key_count() for index in self._indices), 0)
1321
except errors.NoSuchFile:
1322
self._reload_or_raise()
1324
missing_keys = _missing_keys_from_parent_map
1326
def _reload_or_raise(self):
1327
"""We just got a NoSuchFile exception.
1329
Try to reload the indices, if it fails, just raise the current
1332
if self._reload_func is None:
1334
exc_type, exc_value, exc_traceback = sys.exc_info()
1335
trace.mutter('Trying to reload after getting exception: %s',
1337
if not self._reload_func():
1338
# We tried to reload, but nothing changed, so we fail anyway
1339
trace.mutter('_reload_func indicated nothing has changed.'
1340
' Raising original exception.')
1341
raise exc_type, exc_value, exc_traceback
1344
"""Validate that everything in the index can be accessed."""
1347
for index in self._indices:
1350
except errors.NoSuchFile:
1351
self._reload_or_raise()
1354
class InMemoryGraphIndex(GraphIndexBuilder):
1355
"""A GraphIndex which operates entirely out of memory and is mutable.
1357
This is designed to allow the accumulation of GraphIndex entries during a
1358
single write operation, where the accumulated entries need to be immediately
1359
available - for example via a CombinedGraphIndex.
1362
def add_nodes(self, nodes):
1363
"""Add nodes to the index.
1365
:param nodes: An iterable of (key, node_refs, value) entries to add.
1367
if self.reference_lists:
1368
for (key, value, node_refs) in nodes:
1369
self.add_node(key, value, node_refs)
1371
for (key, value) in nodes:
1372
self.add_node(key, value)
1374
def iter_all_entries(self):
1375
"""Iterate over all keys within the index
1377
:return: An iterable of (index, key, reference_lists, value). There is no
1378
defined order for the result iteration - it will be in the most
1379
efficient order for the index (in this case dictionary hash order).
1381
if 'evil' in debug.debug_flags:
1382
trace.mutter_callsite(3,
1383
"iter_all_entries scales with size of history.")
1384
if self.reference_lists:
1385
for key, (absent, references, value) in self._nodes.iteritems():
1387
yield self, key, value, references
1389
for key, (absent, references, value) in self._nodes.iteritems():
1391
yield self, key, value
1393
def iter_entries(self, keys):
1394
"""Iterate over keys within the index.
1396
:param keys: An iterable providing the keys to be retrieved.
1397
:return: An iterable of (index, key, value, reference_lists). There is no
1398
defined order for the result iteration - it will be in the most
1399
efficient order for the index (keys iteration order in this case).
1402
if self.reference_lists:
1403
for key in keys.intersection(self._keys):
1404
node = self._nodes[key]
1406
yield self, key, node[2], node[1]
1408
for key in keys.intersection(self._keys):
1409
node = self._nodes[key]
1411
yield self, key, node[2]
1413
def iter_entries_prefix(self, keys):
1414
"""Iterate over keys within the index using prefix matching.
1416
Prefix matching is applied within the tuple of a key, not to within
1417
the bytestring of each key element. e.g. if you have the keys ('foo',
1418
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1419
only the former key is returned.
1421
:param keys: An iterable providing the key prefixes to be retrieved.
1422
Each key prefix takes the form of a tuple the length of a key, but
1423
with the last N elements 'None' rather than a regular bytestring.
1424
The first element cannot be 'None'.
1425
:return: An iterable as per iter_all_entries, but restricted to the
1426
keys with a matching prefix to those supplied. No additional keys
1427
will be returned, and every match that is in the index will be
1430
# XXX: To much duplication with the GraphIndex class; consider finding
1431
# a good place to pull out the actual common logic.
1435
if self._key_length == 1:
1439
raise errors.BadIndexKey(key)
1440
if len(key) != self._key_length:
1441
raise errors.BadIndexKey(key)
1442
node = self._nodes[key]
1445
if self.reference_lists:
1446
yield self, key, node[2], node[1]
1448
yield self, key, node[2]
1450
nodes_by_key = self._get_nodes_by_key()
1454
raise errors.BadIndexKey(key)
1455
if len(key) != self._key_length:
1456
raise errors.BadIndexKey(key)
1457
# find what it refers to:
1458
key_dict = nodes_by_key
1459
elements = list(key)
1460
# find the subdict to return
1462
while len(elements) and elements[0] is not None:
1463
key_dict = key_dict[elements[0]]
1466
# a non-existant lookup.
1471
key_dict = dicts.pop(-1)
1472
# can't be empty or would not exist
1473
item, value = key_dict.iteritems().next()
1474
if type(value) == dict:
1476
dicts.extend(key_dict.itervalues())
1479
for value in key_dict.itervalues():
1480
yield (self, ) + value
1482
yield (self, ) + key_dict
1484
def key_count(self):
1485
"""Return an estimate of the number of keys in this index.
1487
For InMemoryGraphIndex the estimate is exact.
1489
return len(self._keys)
1492
"""In memory index's have no known corruption at the moment."""
1495
class GraphIndexPrefixAdapter(object):
1496
"""An adapter between GraphIndex with different key lengths.
1498
Queries against this will emit queries against the adapted Graph with the
1499
prefix added, queries for all items use iter_entries_prefix. The returned
1500
nodes will have their keys and node references adjusted to remove the
1501
prefix. Finally, an add_nodes_callback can be supplied - when called the
1502
nodes and references being added will have prefix prepended.
1505
def __init__(self, adapted, prefix, missing_key_length,
1506
add_nodes_callback=None):
1507
"""Construct an adapter against adapted with prefix."""
1508
self.adapted = adapted
1509
self.prefix_key = prefix + (None,)*missing_key_length
1510
self.prefix = prefix
1511
self.prefix_len = len(prefix)
1512
self.add_nodes_callback = add_nodes_callback
1514
def add_nodes(self, nodes):
1515
"""Add nodes to the index.
1517
:param nodes: An iterable of (key, node_refs, value) entries to add.
1519
# save nodes in case its an iterator
1520
nodes = tuple(nodes)
1521
translated_nodes = []
1523
# Add prefix_key to each reference node_refs is a tuple of tuples,
1524
# so split it apart, and add prefix_key to the internal reference
1525
for (key, value, node_refs) in nodes:
1526
adjusted_references = (
1527
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1528
for ref_list in node_refs))
1529
translated_nodes.append((self.prefix + key, value,
1530
adjusted_references))
1532
# XXX: TODO add an explicit interface for getting the reference list
1533
# status, to handle this bit of user-friendliness in the API more
1535
for (key, value) in nodes:
1536
translated_nodes.append((self.prefix + key, value))
1537
self.add_nodes_callback(translated_nodes)
1539
def add_node(self, key, value, references=()):
1540
"""Add a node to the index.
1542
:param key: The key. keys are non-empty tuples containing
1543
as many whitespace-free utf8 bytestrings as the key length
1544
defined for this index.
1545
:param references: An iterable of iterables of keys. Each is a
1546
reference to another key.
1547
:param value: The value to associate with the key. It may be any
1548
bytes as long as it does not contain \0 or \n.
1550
self.add_nodes(((key, value, references), ))
1552
def _strip_prefix(self, an_iter):
1553
"""Strip prefix data from nodes and return it."""
1554
for node in an_iter:
1556
if node[1][:self.prefix_len] != self.prefix:
1557
raise errors.BadIndexData(self)
1558
for ref_list in node[3]:
1559
for ref_node in ref_list:
1560
if ref_node[:self.prefix_len] != self.prefix:
1561
raise errors.BadIndexData(self)
1562
yield node[0], node[1][self.prefix_len:], node[2], (
1563
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1564
for ref_list in node[3]))
1566
def iter_all_entries(self):
1567
"""Iterate over all keys within the index
1569
iter_all_entries is implemented against the adapted index using
1570
iter_entries_prefix.
1572
:return: An iterable of (index, key, reference_lists, value). There is no
1573
defined order for the result iteration - it will be in the most
1574
efficient order for the index (in this case dictionary hash order).
1576
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1578
def iter_entries(self, keys):
1579
"""Iterate over keys within the index.
1581
:param keys: An iterable providing the keys to be retrieved.
1582
:return: An iterable of (index, key, value, reference_lists). There is no
1583
defined order for the result iteration - it will be in the most
1584
efficient order for the index (keys iteration order in this case).
1586
return self._strip_prefix(self.adapted.iter_entries(
1587
self.prefix + key for key in keys))
1589
def iter_entries_prefix(self, keys):
1590
"""Iterate over keys within the index using prefix matching.
1592
Prefix matching is applied within the tuple of a key, not to within
1593
the bytestring of each key element. e.g. if you have the keys ('foo',
1594
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1595
only the former key is returned.
1597
:param keys: An iterable providing the key prefixes to be retrieved.
1598
Each key prefix takes the form of a tuple the length of a key, but
1599
with the last N elements 'None' rather than a regular bytestring.
1600
The first element cannot be 'None'.
1601
:return: An iterable as per iter_all_entries, but restricted to the
1602
keys with a matching prefix to those supplied. No additional keys
1603
will be returned, and every match that is in the index will be
1606
return self._strip_prefix(self.adapted.iter_entries_prefix(
1607
self.prefix + key for key in keys))
1609
def key_count(self):
1610
"""Return an estimate of the number of keys in this index.
1612
For GraphIndexPrefixAdapter this is relatively expensive - key
1613
iteration with the prefix is done.
1615
return len(list(self.iter_all_entries()))
1618
"""Call the adapted's validate."""
1619
self.adapted.validate()
added
removed
bzrlib/index.py
