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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 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
102
self._combine_backing_indices = True
104
def _check_key(self, key):
105
"""Raise BadIndexKey if key is not a valid key for this index."""
106
if type(key) != tuple:
107
raise errors.BadIndexKey(key)
108
if self._key_length != len(key):
109
raise errors.BadIndexKey(key)
111
if not element or _whitespace_re.search(element) is not None:
112
raise errors.BadIndexKey(element)
114
def _external_references(self):
115
"""Return references that are not present in this index.
119
# TODO: JAM 2008-11-21 This makes an assumption about how the reference
120
# lists are used. It is currently correct for pack-0.92 through
121
# 1.9, which use the node references (3rd column) second
122
# reference list as the compression parent. Perhaps this should
123
# be moved into something higher up the stack, since it
124
# makes assumptions about how the index is used.
125
if self.reference_lists > 1:
126
for node in self.iter_all_entries():
128
refs.update(node[3][1])
131
# If reference_lists == 0 there can be no external references, and
132
# if reference_lists == 1, then there isn't a place to store the
136
def _get_nodes_by_key(self):
137
if self._nodes_by_key is None:
139
if self.reference_lists:
140
for key, (absent, references, value) in self._nodes.iteritems():
143
key_dict = nodes_by_key
144
for subkey in key[:-1]:
145
key_dict = key_dict.setdefault(subkey, {})
146
key_dict[key[-1]] = key, value, references
148
for key, (absent, references, value) in self._nodes.iteritems():
151
key_dict = nodes_by_key
152
for subkey in key[:-1]:
153
key_dict = key_dict.setdefault(subkey, {})
154
key_dict[key[-1]] = key, value
155
self._nodes_by_key = nodes_by_key
156
return self._nodes_by_key
158
def _update_nodes_by_key(self, key, value, node_refs):
159
"""Update the _nodes_by_key dict with a new key.
161
For a key of (foo, bar, baz) create
162
_nodes_by_key[foo][bar][baz] = key_value
164
if self._nodes_by_key is None:
166
key_dict = self._nodes_by_key
167
if self.reference_lists:
168
key_value = key, value, node_refs
170
key_value = key, value
171
for subkey in key[:-1]:
172
key_dict = key_dict.setdefault(subkey, {})
173
key_dict[key[-1]] = key_value
175
def _check_key_ref_value(self, key, references, value):
176
"""Check that 'key' and 'references' are all valid.
178
:param key: A key tuple. Must conform to the key interface (be a tuple,
179
be of the right length, not have any whitespace or nulls in any key
181
:param references: An iterable of reference lists. Something like
182
[[(ref, key)], [(ref, key), (other, key)]]
183
:param value: The value associate with this key. Must not contain
184
newlines or null characters.
185
:return: (node_refs, absent_references)
186
node_refs basically a packed form of 'references' where all
188
absent_references reference keys that are not in self._nodes.
189
This may contain duplicates if the same key is
190
referenced in multiple lists.
193
if _newline_null_re.search(value) is not None:
194
raise errors.BadIndexValue(value)
195
if len(references) != self.reference_lists:
196
raise errors.BadIndexValue(references)
198
absent_references = []
199
for reference_list in references:
200
for reference in reference_list:
201
# If reference *is* in self._nodes, then we know it has already
203
if reference not in self._nodes:
204
self._check_key(reference)
205
absent_references.append(reference)
206
node_refs.append(tuple(reference_list))
207
return tuple(node_refs), absent_references
209
def add_node(self, key, value, references=()):
210
"""Add a node to the index.
212
:param key: The key. keys are non-empty tuples containing
213
as many whitespace-free utf8 bytestrings as the key length
214
defined for this index.
215
:param references: An iterable of iterables of keys. Each is a
216
reference to another key.
217
:param value: The value to associate with the key. It may be any
218
bytes as long as it does not contain \0 or \n.
221
absent_references) = self._check_key_ref_value(key, references, value)
222
if key in self._nodes and self._nodes[key][0] != 'a':
223
raise errors.BadIndexDuplicateKey(key, self)
224
for reference in absent_references:
225
# There may be duplicates, but I don't think it is worth worrying
227
self._nodes[reference] = ('a', (), '')
228
self._nodes[key] = ('', node_refs, value)
230
if self._nodes_by_key is not None and self._key_length > 1:
231
self._update_nodes_by_key(key, value, node_refs)
235
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
236
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
237
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
238
prefix_length = sum(len(x) for x in lines)
239
# references are byte offsets. To avoid having to do nasty
240
# polynomial work to resolve offsets (references to later in the
241
# file cannot be determined until all the inbetween references have
242
# been calculated too) we pad the offsets with 0's to make them be
243
# of consistent length. Using binary offsets would break the trivial
245
# to calculate the width of zero's needed we do three passes:
246
# one to gather all the non-reference data and the number of references.
247
# one to pad all the data with reference-length and determine entry
251
# forward sorted by key. In future we may consider topological sorting,
252
# at the cost of table scans for direct lookup, or a second index for
254
nodes = sorted(self._nodes.items())
255
# if we do not prepass, we don't know how long it will be up front.
256
expected_bytes = None
257
# we only need to pre-pass if we have reference lists at all.
258
if self.reference_lists:
260
non_ref_bytes = prefix_length
262
# TODO use simple multiplication for the constants in this loop.
263
for key, (absent, references, value) in nodes:
264
# record the offset known *so far* for this key:
265
# the non reference bytes to date, and the total references to
266
# date - saves reaccumulating on the second pass
267
key_offset_info.append((key, non_ref_bytes, total_references))
268
# key is literal, value is literal, there are 3 null's, 1 NL
269
# key is variable length tuple, \x00 between elements
270
non_ref_bytes += sum(len(element) for element in key)
271
if self._key_length > 1:
272
non_ref_bytes += self._key_length - 1
273
# value is literal bytes, there are 3 null's, 1 NL.
274
non_ref_bytes += len(value) + 3 + 1
275
# one byte for absent if set.
278
elif self.reference_lists:
279
# (ref_lists -1) tabs
280
non_ref_bytes += self.reference_lists - 1
281
# (ref-1 cr's per ref_list)
282
for ref_list in references:
283
# how many references across the whole file?
284
total_references += len(ref_list)
285
# accrue reference separators
287
non_ref_bytes += len(ref_list) - 1
288
# how many digits are needed to represent the total byte count?
290
possible_total_bytes = non_ref_bytes + total_references*digits
291
while 10 ** digits < possible_total_bytes:
293
possible_total_bytes = non_ref_bytes + total_references*digits
294
expected_bytes = possible_total_bytes + 1 # terminating newline
295
# resolve key addresses.
297
for key, non_ref_bytes, total_references in key_offset_info:
298
key_addresses[key] = non_ref_bytes + total_references*digits
300
format_string = '%%0%sd' % digits
301
for key, (absent, references, value) in nodes:
302
flattened_references = []
303
for ref_list in references:
305
for reference in ref_list:
306
ref_addresses.append(format_string % key_addresses[reference])
307
flattened_references.append('\r'.join(ref_addresses))
308
string_key = '\x00'.join(key)
309
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
310
'\t'.join(flattened_references), value))
312
result = StringIO(''.join(lines))
313
if expected_bytes and len(result.getvalue()) != expected_bytes:
314
raise errors.BzrError('Failed index creation. Internal error:'
315
' mismatched output length and expected length: %d %d' %
316
(len(result.getvalue()), expected_bytes))
319
def set_optimize(self, for_size=None, combine_backing_indices=None):
320
"""Change how the builder tries to optimize the result.
322
:param for_size: Tell the builder to try and make the index as small as
324
:param combine_backing_indices: If the builder spills to disk to save
325
memory, should the on-disk indices be combined. Set to True if you
326
are going to be probing the index, but to False if you are not. (If
327
you are not querying, then the time spent combining is wasted.)
330
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
332
if for_size is not None:
333
self._optimize_for_size = for_size
334
if combine_backing_indices is not None:
335
self._combine_backing_indices = combine_backing_indices
338
class GraphIndex(object):
339
"""An index for data with embedded graphs.
341
The index maps keys to a list of key reference lists, and a value.
342
Each node has the same number of key reference lists. Each key reference
343
list can be empty or an arbitrary length. The value is an opaque NULL
344
terminated string without any newlines. The storage of the index is
345
hidden in the interface: keys and key references are always tuples of
346
bytestrings, never the internal representation (e.g. dictionary offsets).
348
It is presumed that the index will not be mutated - it is static data.
350
Successive iter_all_entries calls will read the entire index each time.
351
Additionally, iter_entries calls will read the index linearly until the
352
desired keys are found. XXX: This must be fixed before the index is
353
suitable for production use. :XXX
356
def __init__(self, transport, name, size):
357
"""Open an index called name on transport.
359
:param transport: A bzrlib.transport.Transport.
360
:param name: A path to provide to transport API calls.
361
:param size: The size of the index in bytes. This is used for bisection
362
logic to perform partial index reads. While the size could be
363
obtained by statting the file this introduced an additional round
364
trip as well as requiring stat'able transports, both of which are
365
avoided by having it supplied. If size is None, then bisection
366
support will be disabled and accessing the index will just stream
369
self._transport = transport
371
# Becomes a dict of key:(value, reference-list-byte-locations) used by
372
# the bisection interface to store parsed but not resolved keys.
373
self._bisect_nodes = None
374
# Becomes a dict of key:(value, reference-list-keys) which are ready to
375
# be returned directly to callers.
377
# a sorted list of slice-addresses for the parsed bytes of the file.
378
# e.g. (0,1) would mean that byte 0 is parsed.
379
self._parsed_byte_map = []
380
# a sorted list of keys matching each slice address for parsed bytes
381
# e.g. (None, 'foo@bar') would mean that the first byte contained no
382
# key, and the end byte of the slice is the of the data for 'foo@bar'
383
self._parsed_key_map = []
384
self._key_count = None
385
self._keys_by_offset = None
386
self._nodes_by_key = None
388
# The number of bytes we've read so far in trying to process this file
391
def __eq__(self, other):
392
"""Equal when self and other were created with the same parameters."""
394
type(self) == type(other) and
395
self._transport == other._transport and
396
self._name == other._name and
397
self._size == other._size)
399
def __ne__(self, other):
400
return not self.__eq__(other)
403
return "%s(%r)" % (self.__class__.__name__,
404
self._transport.abspath(self._name))
406
def _buffer_all(self, stream=None):
407
"""Buffer all the index data.
409
Mutates self._nodes and self.keys_by_offset.
411
if self._nodes is not None:
412
# We already did this
414
if 'index' in debug.debug_flags:
415
mutter('Reading entire index %s', self._transport.abspath(self._name))
417
stream = self._transport.get(self._name)
418
self._read_prefix(stream)
419
self._expected_elements = 3 + self._key_length
421
# raw data keyed by offset
422
self._keys_by_offset = {}
423
# ready-to-return key:value or key:value, node_ref_lists
425
self._nodes_by_key = None
428
lines = stream.read().split('\n')
430
_, _, _, trailers = self._parse_lines(lines, pos)
431
for key, absent, references, value in self._keys_by_offset.itervalues():
434
# resolve references:
435
if self.node_ref_lists:
436
node_value = (value, self._resolve_references(references))
439
self._nodes[key] = node_value
440
# cache the keys for quick set intersections
441
self._keys = set(self._nodes)
443
# there must be one line - the empty trailer line.
444
raise errors.BadIndexData(self)
446
def external_references(self, ref_list_num):
447
"""Return references that are not present in this index.
450
if ref_list_num + 1 > self.node_ref_lists:
451
raise ValueError('No ref list %d, index has %d ref lists'
452
% (ref_list_num, self.node_ref_lists))
454
for key, (value, ref_lists) in self._nodes.iteritems():
455
ref_list = ref_lists[ref_list_num]
456
refs.update(ref_list)
457
return refs - self._keys
459
def _get_nodes_by_key(self):
460
if self._nodes_by_key is None:
462
if self.node_ref_lists:
463
for key, (value, references) in self._nodes.iteritems():
464
key_dict = nodes_by_key
465
for subkey in key[:-1]:
466
key_dict = key_dict.setdefault(subkey, {})
467
key_dict[key[-1]] = key, value, references
469
for key, value in self._nodes.iteritems():
470
key_dict = nodes_by_key
471
for subkey in key[:-1]:
472
key_dict = key_dict.setdefault(subkey, {})
473
key_dict[key[-1]] = key, value
474
self._nodes_by_key = nodes_by_key
475
return self._nodes_by_key
477
def iter_all_entries(self):
478
"""Iterate over all keys within the index.
480
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
481
The former tuple is used when there are no reference lists in the
482
index, making the API compatible with simple key:value index types.
483
There is no defined order for the result iteration - it will be in
484
the most efficient order for the index.
486
if 'evil' in debug.debug_flags:
487
trace.mutter_callsite(3,
488
"iter_all_entries scales with size of history.")
489
if self._nodes is None:
491
if self.node_ref_lists:
492
for key, (value, node_ref_lists) in self._nodes.iteritems():
493
yield self, key, value, node_ref_lists
495
for key, value in self._nodes.iteritems():
496
yield self, key, value
498
def _read_prefix(self, stream):
499
signature = stream.read(len(self._signature()))
500
if not signature == self._signature():
501
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
502
options_line = stream.readline()
503
if not options_line.startswith(_OPTION_NODE_REFS):
504
raise errors.BadIndexOptions(self)
506
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
508
raise errors.BadIndexOptions(self)
509
options_line = stream.readline()
510
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
511
raise errors.BadIndexOptions(self)
513
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
515
raise errors.BadIndexOptions(self)
516
options_line = stream.readline()
517
if not options_line.startswith(_OPTION_LEN):
518
raise errors.BadIndexOptions(self)
520
self._key_count = int(options_line[len(_OPTION_LEN):-1])
522
raise errors.BadIndexOptions(self)
524
def _resolve_references(self, references):
525
"""Return the resolved key references for references.
527
References are resolved by looking up the location of the key in the
528
_keys_by_offset map and substituting the key name, preserving ordering.
530
:param references: An iterable of iterables of key locations. e.g.
532
:return: A tuple of tuples of keys.
535
for ref_list in references:
536
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
537
return tuple(node_refs)
539
def _find_index(self, range_map, key):
540
"""Helper for the _parsed_*_index calls.
542
Given a range map - [(start, end), ...], finds the index of the range
543
in the map for key if it is in the map, and if it is not there, the
544
immediately preceeding range in the map.
546
result = bisect_right(range_map, key) - 1
547
if result + 1 < len(range_map):
548
# check the border condition, it may be in result + 1
549
if range_map[result + 1][0] == key[0]:
553
def _parsed_byte_index(self, offset):
554
"""Return the index of the entry immediately before offset.
556
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
557
there is one unparsed byte (the 11th, addressed as[10]). then:
558
asking for 0 will return 0
559
asking for 10 will return 0
560
asking for 11 will return 1
561
asking for 12 will return 1
564
return self._find_index(self._parsed_byte_map, key)
566
def _parsed_key_index(self, key):
567
"""Return the index of the entry immediately before key.
569
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
570
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
571
have been parsed, then:
572
asking for '' will return 0
573
asking for 'a' will return 0
574
asking for 'b' will return 1
575
asking for 'e' will return 1
577
search_key = (key, None)
578
return self._find_index(self._parsed_key_map, search_key)
580
def _is_parsed(self, offset):
581
"""Returns True if offset has been parsed."""
582
index = self._parsed_byte_index(offset)
583
if index == len(self._parsed_byte_map):
584
return offset < self._parsed_byte_map[index - 1][1]
585
start, end = self._parsed_byte_map[index]
586
return offset >= start and offset < end
588
def _iter_entries_from_total_buffer(self, keys):
589
"""Iterate over keys when the entire index is parsed."""
590
keys = keys.intersection(self._keys)
591
if self.node_ref_lists:
593
value, node_refs = self._nodes[key]
594
yield self, key, value, node_refs
597
yield self, key, self._nodes[key]
599
def iter_entries(self, keys):
600
"""Iterate over keys within the index.
602
:param keys: An iterable providing the keys to be retrieved.
603
:return: An iterable as per iter_all_entries, but restricted to the
604
keys supplied. No additional keys will be returned, and every
605
key supplied that is in the index will be returned.
610
if self._size is None and self._nodes is None:
613
# We fit about 20 keys per minimum-read (4K), so if we are looking for
614
# more than 1/20th of the index its likely (assuming homogenous key
615
# spread) that we'll read the entire index. If we're going to do that,
616
# buffer the whole thing. A better analysis might take key spread into
617
# account - but B+Tree indices are better anyway.
618
# We could look at all data read, and use a threshold there, which will
619
# trigger on ancestry walks, but that is not yet fully mapped out.
620
if self._nodes is None and len(keys) * 20 > self.key_count():
622
if self._nodes is not None:
623
return self._iter_entries_from_total_buffer(keys)
625
return (result[1] for result in bisect_multi_bytes(
626
self._lookup_keys_via_location, self._size, keys))
628
def iter_entries_prefix(self, keys):
629
"""Iterate over keys within the index using prefix matching.
631
Prefix matching is applied within the tuple of a key, not to within
632
the bytestring of each key element. e.g. if you have the keys ('foo',
633
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
634
only the former key is returned.
636
WARNING: Note that this method currently causes a full index parse
637
unconditionally (which is reasonably appropriate as it is a means for
638
thunking many small indices into one larger one and still supplies
639
iter_all_entries at the thunk layer).
641
:param keys: An iterable providing the key prefixes to be retrieved.
642
Each key prefix takes the form of a tuple the length of a key, but
643
with the last N elements 'None' rather than a regular bytestring.
644
The first element cannot be 'None'.
645
:return: An iterable as per iter_all_entries, but restricted to the
646
keys with a matching prefix to those supplied. No additional keys
647
will be returned, and every match that is in the index will be
653
# load data - also finds key lengths
654
if self._nodes is None:
656
if self._key_length == 1:
660
raise errors.BadIndexKey(key)
661
if len(key) != self._key_length:
662
raise errors.BadIndexKey(key)
663
if self.node_ref_lists:
664
value, node_refs = self._nodes[key]
665
yield self, key, value, node_refs
667
yield self, key, self._nodes[key]
669
nodes_by_key = self._get_nodes_by_key()
673
raise errors.BadIndexKey(key)
674
if len(key) != self._key_length:
675
raise errors.BadIndexKey(key)
676
# find what it refers to:
677
key_dict = nodes_by_key
679
# find the subdict whose contents should be returned.
681
while len(elements) and elements[0] is not None:
682
key_dict = key_dict[elements[0]]
685
# a non-existant lookup.
690
key_dict = dicts.pop(-1)
691
# can't be empty or would not exist
692
item, value = key_dict.iteritems().next()
693
if type(value) == dict:
695
dicts.extend(key_dict.itervalues())
698
for value in key_dict.itervalues():
699
# each value is the key:value:node refs tuple
701
yield (self, ) + value
703
# the last thing looked up was a terminal element
704
yield (self, ) + key_dict
707
"""Return an estimate of the number of keys in this index.
709
For GraphIndex the estimate is exact.
711
if self._key_count is None:
712
self._read_and_parse([_HEADER_READV])
713
return self._key_count
715
def _lookup_keys_via_location(self, location_keys):
716
"""Public interface for implementing bisection.
718
If _buffer_all has been called, then all the data for the index is in
719
memory, and this method should not be called, as it uses a separate
720
cache because it cannot pre-resolve all indices, which buffer_all does
723
:param location_keys: A list of location(byte offset), key tuples.
724
:return: A list of (location_key, result) tuples as expected by
725
bzrlib.bisect_multi.bisect_multi_bytes.
727
# Possible improvements:
728
# - only bisect lookup each key once
729
# - sort the keys first, and use that to reduce the bisection window
731
# this progresses in three parts:
734
# attempt to answer the question from the now in memory data.
735
# build the readv request
736
# for each location, ask for 800 bytes - much more than rows we've seen
739
for location, key in location_keys:
740
# can we answer from cache?
741
if self._bisect_nodes and key in self._bisect_nodes:
742
# We have the key parsed.
744
index = self._parsed_key_index(key)
745
if (len(self._parsed_key_map) and
746
self._parsed_key_map[index][0] <= key and
747
(self._parsed_key_map[index][1] >= key or
748
# end of the file has been parsed
749
self._parsed_byte_map[index][1] == self._size)):
750
# the key has been parsed, so no lookup is needed even if its
753
# - if we have examined this part of the file already - yes
754
index = self._parsed_byte_index(location)
755
if (len(self._parsed_byte_map) and
756
self._parsed_byte_map[index][0] <= location and
757
self._parsed_byte_map[index][1] > location):
758
# the byte region has been parsed, so no read is needed.
761
if location + length > self._size:
762
length = self._size - location
763
# todo, trim out parsed locations.
765
readv_ranges.append((location, length))
766
# read the header if needed
767
if self._bisect_nodes is None:
768
readv_ranges.append(_HEADER_READV)
769
self._read_and_parse(readv_ranges)
771
if self._nodes is not None:
772
# _read_and_parse triggered a _buffer_all because we requested the
774
for location, key in location_keys:
775
if key not in self._nodes: # not present
776
result.append(((location, key), False))
777
elif self.node_ref_lists:
778
value, refs = self._nodes[key]
779
result.append(((location, key),
780
(self, key, value, refs)))
782
result.append(((location, key),
783
(self, key, self._nodes[key])))
786
# - figure out <, >, missing, present
787
# - result present references so we can return them.
788
# keys that we cannot answer until we resolve references
789
pending_references = []
790
pending_locations = set()
791
for location, key in location_keys:
792
# can we answer from cache?
793
if key in self._bisect_nodes:
794
# the key has been parsed, so no lookup is needed
795
if self.node_ref_lists:
796
# the references may not have been all parsed.
797
value, refs = self._bisect_nodes[key]
798
wanted_locations = []
799
for ref_list in refs:
801
if ref not in self._keys_by_offset:
802
wanted_locations.append(ref)
804
pending_locations.update(wanted_locations)
805
pending_references.append((location, key))
807
result.append(((location, key), (self, key,
808
value, self._resolve_references(refs))))
810
result.append(((location, key),
811
(self, key, self._bisect_nodes[key])))
814
# has the region the key should be in, been parsed?
815
index = self._parsed_key_index(key)
816
if (self._parsed_key_map[index][0] <= key and
817
(self._parsed_key_map[index][1] >= key or
818
# end of the file has been parsed
819
self._parsed_byte_map[index][1] == self._size)):
820
result.append(((location, key), False))
822
# no, is the key above or below the probed location:
823
# get the range of the probed & parsed location
824
index = self._parsed_byte_index(location)
825
# if the key is below the start of the range, its below
826
if key < self._parsed_key_map[index][0]:
830
result.append(((location, key), direction))
832
# lookup data to resolve references
833
for location in pending_locations:
835
if location + length > self._size:
836
length = self._size - location
837
# TODO: trim out parsed locations (e.g. if the 800 is into the
838
# parsed region trim it, and dont use the adjust_for_latency
841
readv_ranges.append((location, length))
842
self._read_and_parse(readv_ranges)
843
if self._nodes is not None:
844
# The _read_and_parse triggered a _buffer_all, grab the data and
846
for location, key in pending_references:
847
value, refs = self._nodes[key]
848
result.append(((location, key), (self, key, value, refs)))
850
for location, key in pending_references:
851
# answer key references we had to look-up-late.
852
value, refs = self._bisect_nodes[key]
853
result.append(((location, key), (self, key,
854
value, self._resolve_references(refs))))
857
def _parse_header_from_bytes(self, bytes):
858
"""Parse the header from a region of bytes.
860
:param bytes: The data to parse.
861
:return: An offset, data tuple such as readv yields, for the unparsed
862
data. (which may length 0).
864
signature = bytes[0:len(self._signature())]
865
if not signature == self._signature():
866
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
867
lines = bytes[len(self._signature()):].splitlines()
868
options_line = lines[0]
869
if not options_line.startswith(_OPTION_NODE_REFS):
870
raise errors.BadIndexOptions(self)
872
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
874
raise errors.BadIndexOptions(self)
875
options_line = lines[1]
876
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
877
raise errors.BadIndexOptions(self)
879
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
881
raise errors.BadIndexOptions(self)
882
options_line = lines[2]
883
if not options_line.startswith(_OPTION_LEN):
884
raise errors.BadIndexOptions(self)
886
self._key_count = int(options_line[len(_OPTION_LEN):])
888
raise errors.BadIndexOptions(self)
889
# calculate the bytes we have processed
890
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
892
self._parsed_bytes(0, None, header_end, None)
893
# setup parsing state
894
self._expected_elements = 3 + self._key_length
895
# raw data keyed by offset
896
self._keys_by_offset = {}
897
# keys with the value and node references
898
self._bisect_nodes = {}
899
return header_end, bytes[header_end:]
901
def _parse_region(self, offset, data):
902
"""Parse node data returned from a readv operation.
904
:param offset: The byte offset the data starts at.
905
:param data: The data to parse.
909
end = offset + len(data)
912
# Trivial test - if the current index's end is within the
913
# low-matching parsed range, we're done.
914
index = self._parsed_byte_index(high_parsed)
915
if end < self._parsed_byte_map[index][1]:
917
# print "[%d:%d]" % (offset, end), \
918
# self._parsed_byte_map[index:index + 2]
919
high_parsed, last_segment = self._parse_segment(
920
offset, data, end, index)
924
def _parse_segment(self, offset, data, end, index):
925
"""Parse one segment of data.
927
:param offset: Where 'data' begins in the file.
928
:param data: Some data to parse a segment of.
929
:param end: Where data ends
930
:param index: The current index into the parsed bytes map.
931
:return: True if the parsed segment is the last possible one in the
933
:return: high_parsed_byte, last_segment.
934
high_parsed_byte is the location of the highest parsed byte in this
935
segment, last_segment is True if the parsed segment is the last
936
possible one in the data block.
938
# default is to use all data
940
# accomodate overlap with data before this.
941
if offset < self._parsed_byte_map[index][1]:
942
# overlaps the lower parsed region
943
# skip the parsed data
944
trim_start = self._parsed_byte_map[index][1] - offset
945
# don't trim the start for \n
946
start_adjacent = True
947
elif offset == self._parsed_byte_map[index][1]:
948
# abuts the lower parsed region
951
# do not trim anything
952
start_adjacent = True
954
# does not overlap the lower parsed region
957
# but trim the leading \n
958
start_adjacent = False
959
if end == self._size:
960
# lines up to the end of all data:
963
# do not strip to the last \n
966
elif index + 1 == len(self._parsed_byte_map):
967
# at the end of the parsed data
970
# but strip to the last \n
973
elif end == self._parsed_byte_map[index + 1][0]:
974
# buts up against the next parsed region
977
# do not strip to the last \n
980
elif end > self._parsed_byte_map[index + 1][0]:
981
# overlaps into the next parsed region
982
# only consider the unparsed data
983
trim_end = self._parsed_byte_map[index + 1][0] - offset
984
# do not strip to the last \n as we know its an entire record
986
last_segment = end < self._parsed_byte_map[index + 1][1]
988
# does not overlap into the next region
991
# but strip to the last \n
994
# now find bytes to discard if needed
995
if not start_adjacent:
996
# work around python bug in rfind
997
if trim_start is None:
998
trim_start = data.find('\n') + 1
1000
trim_start = data.find('\n', trim_start) + 1
1001
if not (trim_start != 0):
1002
raise AssertionError('no \n was present')
1003
# print 'removing start', offset, trim_start, repr(data[:trim_start])
1004
if not end_adjacent:
1005
# work around python bug in rfind
1006
if trim_end is None:
1007
trim_end = data.rfind('\n') + 1
1009
trim_end = data.rfind('\n', None, trim_end) + 1
1010
if not (trim_end != 0):
1011
raise AssertionError('no \n was present')
1012
# print 'removing end', offset, trim_end, repr(data[trim_end:])
1013
# adjust offset and data to the parseable data.
1014
trimmed_data = data[trim_start:trim_end]
1015
if not (trimmed_data):
1016
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
1017
% (trim_start, trim_end, offset, offset + len(data)))
1019
offset += trim_start
1020
# print "parsing", repr(trimmed_data)
1021
# splitlines mangles the \r delimiters.. don't use it.
1022
lines = trimmed_data.split('\n')
1025
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1026
for key, value in nodes:
1027
self._bisect_nodes[key] = value
1028
self._parsed_bytes(offset, first_key,
1029
offset + len(trimmed_data), last_key)
1030
return offset + len(trimmed_data), last_segment
1032
def _parse_lines(self, lines, pos):
1039
# must be at the end
1041
if not (self._size == pos + 1):
1042
raise AssertionError("%s %s" % (self._size, pos))
1045
elements = line.split('\0')
1046
if len(elements) != self._expected_elements:
1047
raise errors.BadIndexData(self)
1048
# keys are tuples. Each element is a string that may occur many
1049
# times, so we intern them to save space. AB, RC, 200807
1050
key = tuple([intern(element) for element in elements[:self._key_length]])
1051
if first_key is None:
1053
absent, references, value = elements[-3:]
1055
for ref_string in references.split('\t'):
1056
ref_lists.append(tuple([
1057
int(ref) for ref in ref_string.split('\r') if ref
1059
ref_lists = tuple(ref_lists)
1060
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1061
pos += len(line) + 1 # +1 for the \n
1064
if self.node_ref_lists:
1065
node_value = (value, ref_lists)
1068
nodes.append((key, node_value))
1069
# print "parsed ", key
1070
return first_key, key, nodes, trailers
1072
def _parsed_bytes(self, start, start_key, end, end_key):
1073
"""Mark the bytes from start to end as parsed.
1075
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1078
:param start: The start of the parsed region.
1079
:param end: The end of the parsed region.
1081
index = self._parsed_byte_index(start)
1082
new_value = (start, end)
1083
new_key = (start_key, end_key)
1085
# first range parsed is always the beginning.
1086
self._parsed_byte_map.insert(index, new_value)
1087
self._parsed_key_map.insert(index, new_key)
1091
# extend lower region
1092
# extend higher region
1093
# combine two regions
1094
if (index + 1 < len(self._parsed_byte_map) and
1095
self._parsed_byte_map[index][1] == start and
1096
self._parsed_byte_map[index + 1][0] == end):
1097
# combine two regions
1098
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1099
self._parsed_byte_map[index + 1][1])
1100
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1101
self._parsed_key_map[index + 1][1])
1102
del self._parsed_byte_map[index + 1]
1103
del self._parsed_key_map[index + 1]
1104
elif self._parsed_byte_map[index][1] == start:
1105
# extend the lower entry
1106
self._parsed_byte_map[index] = (
1107
self._parsed_byte_map[index][0], end)
1108
self._parsed_key_map[index] = (
1109
self._parsed_key_map[index][0], end_key)
1110
elif (index + 1 < len(self._parsed_byte_map) and
1111
self._parsed_byte_map[index + 1][0] == end):
1112
# extend the higher entry
1113
self._parsed_byte_map[index + 1] = (
1114
start, self._parsed_byte_map[index + 1][1])
1115
self._parsed_key_map[index + 1] = (
1116
start_key, self._parsed_key_map[index + 1][1])
1119
self._parsed_byte_map.insert(index + 1, new_value)
1120
self._parsed_key_map.insert(index + 1, new_key)
1122
def _read_and_parse(self, readv_ranges):
1123
"""Read the the ranges and parse the resulting data.
1125
:param readv_ranges: A prepared readv range list.
1127
if not readv_ranges:
1129
if self._nodes is None and self._bytes_read * 2 >= self._size:
1130
# We've already read more than 50% of the file and we are about to
1131
# request more data, just _buffer_all() and be done
1135
readv_data = self._transport.readv(self._name, readv_ranges, True,
1138
for offset, data in readv_data:
1139
self._bytes_read += len(data)
1140
if offset == 0 and len(data) == self._size:
1141
# We read the whole range, most likely because the
1142
# Transport upcast our readv ranges into one long request
1143
# for enough total data to grab the whole index.
1144
self._buffer_all(StringIO(data))
1146
if self._bisect_nodes is None:
1147
# this must be the start
1148
if not (offset == 0):
1149
raise AssertionError()
1150
offset, data = self._parse_header_from_bytes(data)
1151
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1152
self._parse_region(offset, data)
1154
def _signature(self):
1155
"""The file signature for this index type."""
1159
"""Validate that everything in the index can be accessed."""
1160
# iter_all validates completely at the moment, so just do that.
1161
for node in self.iter_all_entries():
1165
class CombinedGraphIndex(object):
1166
"""A GraphIndex made up from smaller GraphIndices.
1168
The backing indices must implement GraphIndex, and are presumed to be
1171
Queries against the combined index will be made against the first index,
1172
and then the second and so on. The order of index's can thus influence
1173
performance significantly. For example, if one index is on local disk and a
1174
second on a remote server, the local disk index should be before the other
1178
def __init__(self, indices, reload_func=None):
1179
"""Create a CombinedGraphIndex backed by indices.
1181
:param indices: An ordered list of indices to query for data.
1182
:param reload_func: A function to call if we find we are missing an
1183
index. Should have the form reload_func() => True/False to indicate
1184
if reloading actually changed anything.
1186
self._indices = indices
1187
self._reload_func = reload_func
1191
self.__class__.__name__,
1192
', '.join(map(repr, self._indices)))
1194
def get_parent_map(self, keys):
1195
"""See graph.StackedParentsProvider.get_parent_map"""
1196
search_keys = set(keys)
1197
if NULL_REVISION in search_keys:
1198
search_keys.discard(NULL_REVISION)
1199
found_parents = {NULL_REVISION:[]}
1202
for index, key, value, refs in self.iter_entries(search_keys):
1205
parents = (NULL_REVISION,)
1206
found_parents[key] = parents
1207
return found_parents
1209
has_key = _has_key_from_parent_map
1211
def insert_index(self, pos, index):
1212
"""Insert a new index in the list of indices to query.
1214
:param pos: The position to insert the index.
1215
:param index: The index to insert.
1217
self._indices.insert(pos, index)
1219
def iter_all_entries(self):
1220
"""Iterate over all keys within the index
1222
Duplicate keys across child indices are presumed to have the same
1223
value and are only reported once.
1225
:return: An iterable of (index, key, reference_lists, value).
1226
There is no defined order for the result iteration - it will be in
1227
the most efficient order for the index.
1232
for index in self._indices:
1233
for node in index.iter_all_entries():
1234
if node[1] not in seen_keys:
1236
seen_keys.add(node[1])
1238
except errors.NoSuchFile:
1239
self._reload_or_raise()
1241
def iter_entries(self, keys):
1242
"""Iterate over keys within the index.
1244
Duplicate keys across child indices are presumed to have the same
1245
value and are only reported once.
1247
:param keys: An iterable providing the keys to be retrieved.
1248
:return: An iterable of (index, key, reference_lists, value). There is no
1249
defined order for the result iteration - it will be in the most
1250
efficient order for the index.
1255
for index in self._indices:
1258
for node in index.iter_entries(keys):
1259
keys.remove(node[1])
1262
except errors.NoSuchFile:
1263
self._reload_or_raise()
1265
def iter_entries_prefix(self, keys):
1266
"""Iterate over keys within the index using prefix matching.
1268
Duplicate keys across child indices are presumed to have the same
1269
value and are only reported once.
1271
Prefix matching is applied within the tuple of a key, not to within
1272
the bytestring of each key element. e.g. if you have the keys ('foo',
1273
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1274
only the former key is returned.
1276
:param keys: An iterable providing the key prefixes to be retrieved.
1277
Each key prefix takes the form of a tuple the length of a key, but
1278
with the last N elements 'None' rather than a regular bytestring.
1279
The first element cannot be 'None'.
1280
:return: An iterable as per iter_all_entries, but restricted to the
1281
keys with a matching prefix to those supplied. No additional keys
1282
will be returned, and every match that is in the index will be
1291
for index in self._indices:
1292
for node in index.iter_entries_prefix(keys):
1293
if node[1] in seen_keys:
1295
seen_keys.add(node[1])
1298
except errors.NoSuchFile:
1299
self._reload_or_raise()
1301
def key_count(self):
1302
"""Return an estimate of the number of keys in this index.
1304
For CombinedGraphIndex this is approximated by the sum of the keys of
1305
the child indices. As child indices may have duplicate keys this can
1306
have a maximum error of the number of child indices * largest number of
1311
return sum((index.key_count() for index in self._indices), 0)
1312
except errors.NoSuchFile:
1313
self._reload_or_raise()
1315
missing_keys = _missing_keys_from_parent_map
1317
def _reload_or_raise(self):
1318
"""We just got a NoSuchFile exception.
1320
Try to reload the indices, if it fails, just raise the current
1323
if self._reload_func is None:
1325
exc_type, exc_value, exc_traceback = sys.exc_info()
1326
trace.mutter('Trying to reload after getting exception: %s',
1328
if not self._reload_func():
1329
# We tried to reload, but nothing changed, so we fail anyway
1330
trace.mutter('_reload_func indicated nothing has changed.'
1331
' Raising original exception.')
1332
raise exc_type, exc_value, exc_traceback
1335
"""Validate that everything in the index can be accessed."""
1338
for index in self._indices:
1341
except errors.NoSuchFile:
1342
self._reload_or_raise()
1345
class InMemoryGraphIndex(GraphIndexBuilder):
1346
"""A GraphIndex which operates entirely out of memory and is mutable.
1348
This is designed to allow the accumulation of GraphIndex entries during a
1349
single write operation, where the accumulated entries need to be immediately
1350
available - for example via a CombinedGraphIndex.
1353
def add_nodes(self, nodes):
1354
"""Add nodes to the index.
1356
:param nodes: An iterable of (key, node_refs, value) entries to add.
1358
if self.reference_lists:
1359
for (key, value, node_refs) in nodes:
1360
self.add_node(key, value, node_refs)
1362
for (key, value) in nodes:
1363
self.add_node(key, value)
1365
def iter_all_entries(self):
1366
"""Iterate over all keys within the index
1368
:return: An iterable of (index, key, reference_lists, value). There is no
1369
defined order for the result iteration - it will be in the most
1370
efficient order for the index (in this case dictionary hash order).
1372
if 'evil' in debug.debug_flags:
1373
trace.mutter_callsite(3,
1374
"iter_all_entries scales with size of history.")
1375
if self.reference_lists:
1376
for key, (absent, references, value) in self._nodes.iteritems():
1378
yield self, key, value, references
1380
for key, (absent, references, value) in self._nodes.iteritems():
1382
yield self, key, value
1384
def iter_entries(self, keys):
1385
"""Iterate over keys within the index.
1387
:param keys: An iterable providing the keys to be retrieved.
1388
:return: An iterable of (index, key, value, reference_lists). There is no
1389
defined order for the result iteration - it will be in the most
1390
efficient order for the index (keys iteration order in this case).
1393
if self.reference_lists:
1394
for key in keys.intersection(self._keys):
1395
node = self._nodes[key]
1397
yield self, key, node[2], node[1]
1399
for key in keys.intersection(self._keys):
1400
node = self._nodes[key]
1402
yield self, key, node[2]
1404
def iter_entries_prefix(self, keys):
1405
"""Iterate over keys within the index using prefix matching.
1407
Prefix matching is applied within the tuple of a key, not to within
1408
the bytestring of each key element. e.g. if you have the keys ('foo',
1409
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1410
only the former key is returned.
1412
:param keys: An iterable providing the key prefixes to be retrieved.
1413
Each key prefix takes the form of a tuple the length of a key, but
1414
with the last N elements 'None' rather than a regular bytestring.
1415
The first element cannot be 'None'.
1416
:return: An iterable as per iter_all_entries, but restricted to the
1417
keys with a matching prefix to those supplied. No additional keys
1418
will be returned, and every match that is in the index will be
1421
# XXX: To much duplication with the GraphIndex class; consider finding
1422
# a good place to pull out the actual common logic.
1426
if self._key_length == 1:
1430
raise errors.BadIndexKey(key)
1431
if len(key) != self._key_length:
1432
raise errors.BadIndexKey(key)
1433
node = self._nodes[key]
1436
if self.reference_lists:
1437
yield self, key, node[2], node[1]
1439
yield self, key, node[2]
1441
nodes_by_key = self._get_nodes_by_key()
1445
raise errors.BadIndexKey(key)
1446
if len(key) != self._key_length:
1447
raise errors.BadIndexKey(key)
1448
# find what it refers to:
1449
key_dict = nodes_by_key
1450
elements = list(key)
1451
# find the subdict to return
1453
while len(elements) and elements[0] is not None:
1454
key_dict = key_dict[elements[0]]
1457
# a non-existant lookup.
1462
key_dict = dicts.pop(-1)
1463
# can't be empty or would not exist
1464
item, value = key_dict.iteritems().next()
1465
if type(value) == dict:
1467
dicts.extend(key_dict.itervalues())
1470
for value in key_dict.itervalues():
1471
yield (self, ) + value
1473
yield (self, ) + key_dict
1475
def key_count(self):
1476
"""Return an estimate of the number of keys in this index.
1478
For InMemoryGraphIndex the estimate is exact.
1480
return len(self._keys)
1483
"""In memory index's have no known corruption at the moment."""
1486
class GraphIndexPrefixAdapter(object):
1487
"""An adapter between GraphIndex with different key lengths.
1489
Queries against this will emit queries against the adapted Graph with the
1490
prefix added, queries for all items use iter_entries_prefix. The returned
1491
nodes will have their keys and node references adjusted to remove the
1492
prefix. Finally, an add_nodes_callback can be supplied - when called the
1493
nodes and references being added will have prefix prepended.
1496
def __init__(self, adapted, prefix, missing_key_length,
1497
add_nodes_callback=None):
1498
"""Construct an adapter against adapted with prefix."""
1499
self.adapted = adapted
1500
self.prefix_key = prefix + (None,)*missing_key_length
1501
self.prefix = prefix
1502
self.prefix_len = len(prefix)
1503
self.add_nodes_callback = add_nodes_callback
1505
def add_nodes(self, nodes):
1506
"""Add nodes to the index.
1508
:param nodes: An iterable of (key, node_refs, value) entries to add.
1510
# save nodes in case its an iterator
1511
nodes = tuple(nodes)
1512
translated_nodes = []
1514
# Add prefix_key to each reference node_refs is a tuple of tuples,
1515
# so split it apart, and add prefix_key to the internal reference
1516
for (key, value, node_refs) in nodes:
1517
adjusted_references = (
1518
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1519
for ref_list in node_refs))
1520
translated_nodes.append((self.prefix + key, value,
1521
adjusted_references))
1523
# XXX: TODO add an explicit interface for getting the reference list
1524
# status, to handle this bit of user-friendliness in the API more
1526
for (key, value) in nodes:
1527
translated_nodes.append((self.prefix + key, value))
1528
self.add_nodes_callback(translated_nodes)
1530
def add_node(self, key, value, references=()):
1531
"""Add a node to the index.
1533
:param key: The key. keys are non-empty tuples containing
1534
as many whitespace-free utf8 bytestrings as the key length
1535
defined for this index.
1536
:param references: An iterable of iterables of keys. Each is a
1537
reference to another key.
1538
:param value: The value to associate with the key. It may be any
1539
bytes as long as it does not contain \0 or \n.
1541
self.add_nodes(((key, value, references), ))
1543
def _strip_prefix(self, an_iter):
1544
"""Strip prefix data from nodes and return it."""
1545
for node in an_iter:
1547
if node[1][:self.prefix_len] != self.prefix:
1548
raise errors.BadIndexData(self)
1549
for ref_list in node[3]:
1550
for ref_node in ref_list:
1551
if ref_node[:self.prefix_len] != self.prefix:
1552
raise errors.BadIndexData(self)
1553
yield node[0], node[1][self.prefix_len:], node[2], (
1554
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1555
for ref_list in node[3]))
1557
def iter_all_entries(self):
1558
"""Iterate over all keys within the index
1560
iter_all_entries is implemented against the adapted index using
1561
iter_entries_prefix.
1563
:return: An iterable of (index, key, reference_lists, value). There is no
1564
defined order for the result iteration - it will be in the most
1565
efficient order for the index (in this case dictionary hash order).
1567
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1569
def iter_entries(self, keys):
1570
"""Iterate over keys within the index.
1572
:param keys: An iterable providing the keys to be retrieved.
1573
:return: An iterable of (index, key, value, reference_lists). There is no
1574
defined order for the result iteration - it will be in the most
1575
efficient order for the index (keys iteration order in this case).
1577
return self._strip_prefix(self.adapted.iter_entries(
1578
self.prefix + key for key in keys))
1580
def iter_entries_prefix(self, keys):
1581
"""Iterate over keys within the index using prefix matching.
1583
Prefix matching is applied within the tuple of a key, not to within
1584
the bytestring of each key element. e.g. if you have the keys ('foo',
1585
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1586
only the former key is returned.
1588
:param keys: An iterable providing the key prefixes to be retrieved.
1589
Each key prefix takes the form of a tuple the length of a key, but
1590
with the last N elements 'None' rather than a regular bytestring.
1591
The first element cannot be 'None'.
1592
:return: An iterable as per iter_all_entries, but restricted to the
1593
keys with a matching prefix to those supplied. No additional keys
1594
will be returned, and every match that is in the index will be
1597
return self._strip_prefix(self.adapted.iter_entries_prefix(
1598
self.prefix + key for key in keys))
1600
def key_count(self):
1601
"""Return an estimate of the number of keys in this index.
1603
For GraphIndexPrefixAdapter this is relatively expensive - key
1604
iteration with the prefix is done.
1606
return len(list(self.iter_all_entries()))
1609
"""Call the adapted's validate."""
1610
self.adapted.validate()