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 _get_nodes_by_key(self):
439
if self._nodes_by_key is None:
441
if self.node_ref_lists:
442
for key, (value, references) in self._nodes.iteritems():
443
key_dict = nodes_by_key
444
for subkey in key[:-1]:
445
key_dict = key_dict.setdefault(subkey, {})
446
key_dict[key[-1]] = key, value, references
448
for key, value in self._nodes.iteritems():
449
key_dict = nodes_by_key
450
for subkey in key[:-1]:
451
key_dict = key_dict.setdefault(subkey, {})
452
key_dict[key[-1]] = key, value
453
self._nodes_by_key = nodes_by_key
454
return self._nodes_by_key
456
def iter_all_entries(self):
457
"""Iterate over all keys within the index.
459
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
460
The former tuple is used when there are no reference lists in the
461
index, making the API compatible with simple key:value index types.
462
There is no defined order for the result iteration - it will be in
463
the most efficient order for the index.
465
if 'evil' in debug.debug_flags:
466
trace.mutter_callsite(3,
467
"iter_all_entries scales with size of history.")
468
if self._nodes is None:
470
if self.node_ref_lists:
471
for key, (value, node_ref_lists) in self._nodes.iteritems():
472
yield self, key, value, node_ref_lists
474
for key, value in self._nodes.iteritems():
475
yield self, key, value
477
def _read_prefix(self, stream):
478
signature = stream.read(len(self._signature()))
479
if not signature == self._signature():
480
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
481
options_line = stream.readline()
482
if not options_line.startswith(_OPTION_NODE_REFS):
483
raise errors.BadIndexOptions(self)
485
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
487
raise errors.BadIndexOptions(self)
488
options_line = stream.readline()
489
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
490
raise errors.BadIndexOptions(self)
492
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
494
raise errors.BadIndexOptions(self)
495
options_line = stream.readline()
496
if not options_line.startswith(_OPTION_LEN):
497
raise errors.BadIndexOptions(self)
499
self._key_count = int(options_line[len(_OPTION_LEN):-1])
501
raise errors.BadIndexOptions(self)
503
def _resolve_references(self, references):
504
"""Return the resolved key references for references.
506
References are resolved by looking up the location of the key in the
507
_keys_by_offset map and substituting the key name, preserving ordering.
509
:param references: An iterable of iterables of key locations. e.g.
511
:return: A tuple of tuples of keys.
514
for ref_list in references:
515
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
516
return tuple(node_refs)
518
def _find_index(self, range_map, key):
519
"""Helper for the _parsed_*_index calls.
521
Given a range map - [(start, end), ...], finds the index of the range
522
in the map for key if it is in the map, and if it is not there, the
523
immediately preceeding range in the map.
525
result = bisect_right(range_map, key) - 1
526
if result + 1 < len(range_map):
527
# check the border condition, it may be in result + 1
528
if range_map[result + 1][0] == key[0]:
532
def _parsed_byte_index(self, offset):
533
"""Return the index of the entry immediately before offset.
535
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
536
there is one unparsed byte (the 11th, addressed as[10]). then:
537
asking for 0 will return 0
538
asking for 10 will return 0
539
asking for 11 will return 1
540
asking for 12 will return 1
543
return self._find_index(self._parsed_byte_map, key)
545
def _parsed_key_index(self, key):
546
"""Return the index of the entry immediately before key.
548
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
549
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
550
have been parsed, then:
551
asking for '' will return 0
552
asking for 'a' will return 0
553
asking for 'b' will return 1
554
asking for 'e' will return 1
556
search_key = (key, None)
557
return self._find_index(self._parsed_key_map, search_key)
559
def _is_parsed(self, offset):
560
"""Returns True if offset has been parsed."""
561
index = self._parsed_byte_index(offset)
562
if index == len(self._parsed_byte_map):
563
return offset < self._parsed_byte_map[index - 1][1]
564
start, end = self._parsed_byte_map[index]
565
return offset >= start and offset < end
567
def _iter_entries_from_total_buffer(self, keys):
568
"""Iterate over keys when the entire index is parsed."""
569
keys = keys.intersection(self._keys)
570
if self.node_ref_lists:
572
value, node_refs = self._nodes[key]
573
yield self, key, value, node_refs
576
yield self, key, self._nodes[key]
578
def iter_entries(self, keys):
579
"""Iterate over keys within the index.
581
:param keys: An iterable providing the keys to be retrieved.
582
:return: An iterable as per iter_all_entries, but restricted to the
583
keys supplied. No additional keys will be returned, and every
584
key supplied that is in the index will be returned.
589
if self._size is None and self._nodes is None:
592
# We fit about 20 keys per minimum-read (4K), so if we are looking for
593
# more than 1/20th of the index its likely (assuming homogenous key
594
# spread) that we'll read the entire index. If we're going to do that,
595
# buffer the whole thing. A better analysis might take key spread into
596
# account - but B+Tree indices are better anyway.
597
# We could look at all data read, and use a threshold there, which will
598
# trigger on ancestry walks, but that is not yet fully mapped out.
599
if self._nodes is None and len(keys) * 20 > self.key_count():
601
if self._nodes is not None:
602
return self._iter_entries_from_total_buffer(keys)
604
return (result[1] for result in bisect_multi_bytes(
605
self._lookup_keys_via_location, self._size, keys))
607
def iter_entries_prefix(self, keys):
608
"""Iterate over keys within the index using prefix matching.
610
Prefix matching is applied within the tuple of a key, not to within
611
the bytestring of each key element. e.g. if you have the keys ('foo',
612
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
613
only the former key is returned.
615
WARNING: Note that this method currently causes a full index parse
616
unconditionally (which is reasonably appropriate as it is a means for
617
thunking many small indices into one larger one and still supplies
618
iter_all_entries at the thunk layer).
620
:param keys: An iterable providing the key prefixes to be retrieved.
621
Each key prefix takes the form of a tuple the length of a key, but
622
with the last N elements 'None' rather than a regular bytestring.
623
The first element cannot be 'None'.
624
:return: An iterable as per iter_all_entries, but restricted to the
625
keys with a matching prefix to those supplied. No additional keys
626
will be returned, and every match that is in the index will be
632
# load data - also finds key lengths
633
if self._nodes is None:
635
if self._key_length == 1:
639
raise errors.BadIndexKey(key)
640
if len(key) != self._key_length:
641
raise errors.BadIndexKey(key)
642
if self.node_ref_lists:
643
value, node_refs = self._nodes[key]
644
yield self, key, value, node_refs
646
yield self, key, self._nodes[key]
648
nodes_by_key = self._get_nodes_by_key()
652
raise errors.BadIndexKey(key)
653
if len(key) != self._key_length:
654
raise errors.BadIndexKey(key)
655
# find what it refers to:
656
key_dict = nodes_by_key
658
# find the subdict whose contents should be returned.
660
while len(elements) and elements[0] is not None:
661
key_dict = key_dict[elements[0]]
664
# a non-existant lookup.
669
key_dict = dicts.pop(-1)
670
# can't be empty or would not exist
671
item, value = key_dict.iteritems().next()
672
if type(value) == dict:
674
dicts.extend(key_dict.itervalues())
677
for value in key_dict.itervalues():
678
# each value is the key:value:node refs tuple
680
yield (self, ) + value
682
# the last thing looked up was a terminal element
683
yield (self, ) + key_dict
686
"""Return an estimate of the number of keys in this index.
688
For GraphIndex the estimate is exact.
690
if self._key_count is None:
691
self._read_and_parse([_HEADER_READV])
692
return self._key_count
694
def _lookup_keys_via_location(self, location_keys):
695
"""Public interface for implementing bisection.
697
If _buffer_all has been called, then all the data for the index is in
698
memory, and this method should not be called, as it uses a separate
699
cache because it cannot pre-resolve all indices, which buffer_all does
702
:param location_keys: A list of location(byte offset), key tuples.
703
:return: A list of (location_key, result) tuples as expected by
704
bzrlib.bisect_multi.bisect_multi_bytes.
706
# Possible improvements:
707
# - only bisect lookup each key once
708
# - sort the keys first, and use that to reduce the bisection window
710
# this progresses in three parts:
713
# attempt to answer the question from the now in memory data.
714
# build the readv request
715
# for each location, ask for 800 bytes - much more than rows we've seen
718
for location, key in location_keys:
719
# can we answer from cache?
720
if self._bisect_nodes and key in self._bisect_nodes:
721
# We have the key parsed.
723
index = self._parsed_key_index(key)
724
if (len(self._parsed_key_map) and
725
self._parsed_key_map[index][0] <= key and
726
(self._parsed_key_map[index][1] >= key or
727
# end of the file has been parsed
728
self._parsed_byte_map[index][1] == self._size)):
729
# the key has been parsed, so no lookup is needed even if its
732
# - if we have examined this part of the file already - yes
733
index = self._parsed_byte_index(location)
734
if (len(self._parsed_byte_map) and
735
self._parsed_byte_map[index][0] <= location and
736
self._parsed_byte_map[index][1] > location):
737
# the byte region has been parsed, so no read is needed.
740
if location + length > self._size:
741
length = self._size - location
742
# todo, trim out parsed locations.
744
readv_ranges.append((location, length))
745
# read the header if needed
746
if self._bisect_nodes is None:
747
readv_ranges.append(_HEADER_READV)
748
self._read_and_parse(readv_ranges)
750
if self._nodes is not None:
751
# _read_and_parse triggered a _buffer_all because we requested the
753
for location, key in location_keys:
754
if key not in self._nodes: # not present
755
result.append(((location, key), False))
756
elif self.node_ref_lists:
757
value, refs = self._nodes[key]
758
result.append(((location, key),
759
(self, key, value, refs)))
761
result.append(((location, key),
762
(self, key, self._nodes[key])))
765
# - figure out <, >, missing, present
766
# - result present references so we can return them.
767
# keys that we cannot answer until we resolve references
768
pending_references = []
769
pending_locations = set()
770
for location, key in location_keys:
771
# can we answer from cache?
772
if key in self._bisect_nodes:
773
# the key has been parsed, so no lookup is needed
774
if self.node_ref_lists:
775
# the references may not have been all parsed.
776
value, refs = self._bisect_nodes[key]
777
wanted_locations = []
778
for ref_list in refs:
780
if ref not in self._keys_by_offset:
781
wanted_locations.append(ref)
783
pending_locations.update(wanted_locations)
784
pending_references.append((location, key))
786
result.append(((location, key), (self, key,
787
value, self._resolve_references(refs))))
789
result.append(((location, key),
790
(self, key, self._bisect_nodes[key])))
793
# has the region the key should be in, been parsed?
794
index = self._parsed_key_index(key)
795
if (self._parsed_key_map[index][0] <= key and
796
(self._parsed_key_map[index][1] >= key or
797
# end of the file has been parsed
798
self._parsed_byte_map[index][1] == self._size)):
799
result.append(((location, key), False))
801
# no, is the key above or below the probed location:
802
# get the range of the probed & parsed location
803
index = self._parsed_byte_index(location)
804
# if the key is below the start of the range, its below
805
if key < self._parsed_key_map[index][0]:
809
result.append(((location, key), direction))
811
# lookup data to resolve references
812
for location in pending_locations:
814
if location + length > self._size:
815
length = self._size - location
816
# TODO: trim out parsed locations (e.g. if the 800 is into the
817
# parsed region trim it, and dont use the adjust_for_latency
820
readv_ranges.append((location, length))
821
self._read_and_parse(readv_ranges)
822
if self._nodes is not None:
823
# The _read_and_parse triggered a _buffer_all, grab the data and
825
for location, key in pending_references:
826
value, refs = self._nodes[key]
827
result.append(((location, key), (self, key, value, refs)))
829
for location, key in pending_references:
830
# answer key references we had to look-up-late.
831
value, refs = self._bisect_nodes[key]
832
result.append(((location, key), (self, key,
833
value, self._resolve_references(refs))))
836
def _parse_header_from_bytes(self, bytes):
837
"""Parse the header from a region of bytes.
839
:param bytes: The data to parse.
840
:return: An offset, data tuple such as readv yields, for the unparsed
841
data. (which may length 0).
843
signature = bytes[0:len(self._signature())]
844
if not signature == self._signature():
845
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
846
lines = bytes[len(self._signature()):].splitlines()
847
options_line = lines[0]
848
if not options_line.startswith(_OPTION_NODE_REFS):
849
raise errors.BadIndexOptions(self)
851
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
853
raise errors.BadIndexOptions(self)
854
options_line = lines[1]
855
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
856
raise errors.BadIndexOptions(self)
858
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
860
raise errors.BadIndexOptions(self)
861
options_line = lines[2]
862
if not options_line.startswith(_OPTION_LEN):
863
raise errors.BadIndexOptions(self)
865
self._key_count = int(options_line[len(_OPTION_LEN):])
867
raise errors.BadIndexOptions(self)
868
# calculate the bytes we have processed
869
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
871
self._parsed_bytes(0, None, header_end, None)
872
# setup parsing state
873
self._expected_elements = 3 + self._key_length
874
# raw data keyed by offset
875
self._keys_by_offset = {}
876
# keys with the value and node references
877
self._bisect_nodes = {}
878
return header_end, bytes[header_end:]
880
def _parse_region(self, offset, data):
881
"""Parse node data returned from a readv operation.
883
:param offset: The byte offset the data starts at.
884
:param data: The data to parse.
888
end = offset + len(data)
891
# Trivial test - if the current index's end is within the
892
# low-matching parsed range, we're done.
893
index = self._parsed_byte_index(high_parsed)
894
if end < self._parsed_byte_map[index][1]:
896
# print "[%d:%d]" % (offset, end), \
897
# self._parsed_byte_map[index:index + 2]
898
high_parsed, last_segment = self._parse_segment(
899
offset, data, end, index)
903
def _parse_segment(self, offset, data, end, index):
904
"""Parse one segment of data.
906
:param offset: Where 'data' begins in the file.
907
:param data: Some data to parse a segment of.
908
:param end: Where data ends
909
:param index: The current index into the parsed bytes map.
910
:return: True if the parsed segment is the last possible one in the
912
:return: high_parsed_byte, last_segment.
913
high_parsed_byte is the location of the highest parsed byte in this
914
segment, last_segment is True if the parsed segment is the last
915
possible one in the data block.
917
# default is to use all data
919
# accomodate overlap with data before this.
920
if offset < self._parsed_byte_map[index][1]:
921
# overlaps the lower parsed region
922
# skip the parsed data
923
trim_start = self._parsed_byte_map[index][1] - offset
924
# don't trim the start for \n
925
start_adjacent = True
926
elif offset == self._parsed_byte_map[index][1]:
927
# abuts the lower parsed region
930
# do not trim anything
931
start_adjacent = True
933
# does not overlap the lower parsed region
936
# but trim the leading \n
937
start_adjacent = False
938
if end == self._size:
939
# lines up to the end of all data:
942
# do not strip to the last \n
945
elif index + 1 == len(self._parsed_byte_map):
946
# at the end of the parsed data
949
# but strip to the last \n
952
elif end == self._parsed_byte_map[index + 1][0]:
953
# buts up against the next parsed region
956
# do not strip to the last \n
959
elif end > self._parsed_byte_map[index + 1][0]:
960
# overlaps into the next parsed region
961
# only consider the unparsed data
962
trim_end = self._parsed_byte_map[index + 1][0] - offset
963
# do not strip to the last \n as we know its an entire record
965
last_segment = end < self._parsed_byte_map[index + 1][1]
967
# does not overlap into the next region
970
# but strip to the last \n
973
# now find bytes to discard if needed
974
if not start_adjacent:
975
# work around python bug in rfind
976
if trim_start is None:
977
trim_start = data.find('\n') + 1
979
trim_start = data.find('\n', trim_start) + 1
980
if not (trim_start != 0):
981
raise AssertionError('no \n was present')
982
# print 'removing start', offset, trim_start, repr(data[:trim_start])
984
# work around python bug in rfind
986
trim_end = data.rfind('\n') + 1
988
trim_end = data.rfind('\n', None, trim_end) + 1
989
if not (trim_end != 0):
990
raise AssertionError('no \n was present')
991
# print 'removing end', offset, trim_end, repr(data[trim_end:])
992
# adjust offset and data to the parseable data.
993
trimmed_data = data[trim_start:trim_end]
994
if not (trimmed_data):
995
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
996
% (trim_start, trim_end, offset, offset + len(data)))
999
# print "parsing", repr(trimmed_data)
1000
# splitlines mangles the \r delimiters.. don't use it.
1001
lines = trimmed_data.split('\n')
1004
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1005
for key, value in nodes:
1006
self._bisect_nodes[key] = value
1007
self._parsed_bytes(offset, first_key,
1008
offset + len(trimmed_data), last_key)
1009
return offset + len(trimmed_data), last_segment
1011
def _parse_lines(self, lines, pos):
1018
# must be at the end
1020
if not (self._size == pos + 1):
1021
raise AssertionError("%s %s" % (self._size, pos))
1024
elements = line.split('\0')
1025
if len(elements) != self._expected_elements:
1026
raise errors.BadIndexData(self)
1027
# keys are tuples. Each element is a string that may occur many
1028
# times, so we intern them to save space. AB, RC, 200807
1029
key = tuple([intern(element) for element in elements[:self._key_length]])
1030
if first_key is None:
1032
absent, references, value = elements[-3:]
1034
for ref_string in references.split('\t'):
1035
ref_lists.append(tuple([
1036
int(ref) for ref in ref_string.split('\r') if ref
1038
ref_lists = tuple(ref_lists)
1039
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1040
pos += len(line) + 1 # +1 for the \n
1043
if self.node_ref_lists:
1044
node_value = (value, ref_lists)
1047
nodes.append((key, node_value))
1048
# print "parsed ", key
1049
return first_key, key, nodes, trailers
1051
def _parsed_bytes(self, start, start_key, end, end_key):
1052
"""Mark the bytes from start to end as parsed.
1054
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1057
:param start: The start of the parsed region.
1058
:param end: The end of the parsed region.
1060
index = self._parsed_byte_index(start)
1061
new_value = (start, end)
1062
new_key = (start_key, end_key)
1064
# first range parsed is always the beginning.
1065
self._parsed_byte_map.insert(index, new_value)
1066
self._parsed_key_map.insert(index, new_key)
1070
# extend lower region
1071
# extend higher region
1072
# combine two regions
1073
if (index + 1 < len(self._parsed_byte_map) and
1074
self._parsed_byte_map[index][1] == start and
1075
self._parsed_byte_map[index + 1][0] == end):
1076
# combine two regions
1077
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1078
self._parsed_byte_map[index + 1][1])
1079
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1080
self._parsed_key_map[index + 1][1])
1081
del self._parsed_byte_map[index + 1]
1082
del self._parsed_key_map[index + 1]
1083
elif self._parsed_byte_map[index][1] == start:
1084
# extend the lower entry
1085
self._parsed_byte_map[index] = (
1086
self._parsed_byte_map[index][0], end)
1087
self._parsed_key_map[index] = (
1088
self._parsed_key_map[index][0], end_key)
1089
elif (index + 1 < len(self._parsed_byte_map) and
1090
self._parsed_byte_map[index + 1][0] == end):
1091
# extend the higher entry
1092
self._parsed_byte_map[index + 1] = (
1093
start, self._parsed_byte_map[index + 1][1])
1094
self._parsed_key_map[index + 1] = (
1095
start_key, self._parsed_key_map[index + 1][1])
1098
self._parsed_byte_map.insert(index + 1, new_value)
1099
self._parsed_key_map.insert(index + 1, new_key)
1101
def _read_and_parse(self, readv_ranges):
1102
"""Read the the ranges and parse the resulting data.
1104
:param readv_ranges: A prepared readv range list.
1106
if not readv_ranges:
1108
if self._nodes is None and self._bytes_read * 2 >= self._size:
1109
# We've already read more than 50% of the file and we are about to
1110
# request more data, just _buffer_all() and be done
1114
readv_data = self._transport.readv(self._name, readv_ranges, True,
1117
for offset, data in readv_data:
1118
self._bytes_read += len(data)
1119
if offset == 0 and len(data) == self._size:
1120
# We read the whole range, most likely because the
1121
# Transport upcast our readv ranges into one long request
1122
# for enough total data to grab the whole index.
1123
self._buffer_all(StringIO(data))
1125
if self._bisect_nodes is None:
1126
# this must be the start
1127
if not (offset == 0):
1128
raise AssertionError()
1129
offset, data = self._parse_header_from_bytes(data)
1130
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1131
self._parse_region(offset, data)
1133
def _signature(self):
1134
"""The file signature for this index type."""
1138
"""Validate that everything in the index can be accessed."""
1139
# iter_all validates completely at the moment, so just do that.
1140
for node in self.iter_all_entries():
1144
class CombinedGraphIndex(object):
1145
"""A GraphIndex made up from smaller GraphIndices.
1147
The backing indices must implement GraphIndex, and are presumed to be
1150
Queries against the combined index will be made against the first index,
1151
and then the second and so on. The order of index's can thus influence
1152
performance significantly. For example, if one index is on local disk and a
1153
second on a remote server, the local disk index should be before the other
1157
def __init__(self, indices, reload_func=None):
1158
"""Create a CombinedGraphIndex backed by indices.
1160
:param indices: An ordered list of indices to query for data.
1161
:param reload_func: A function to call if we find we are missing an
1162
index. Should have the form reload_func() => True/False to indicate
1163
if reloading actually changed anything.
1165
self._indices = indices
1166
self._reload_func = reload_func
1170
self.__class__.__name__,
1171
', '.join(map(repr, self._indices)))
1173
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1174
def get_parents(self, revision_ids):
1175
"""See graph._StackedParentsProvider.get_parents.
1177
This implementation thunks the graph.Graph.get_parents api across to
1180
:param revision_ids: An iterable of graph keys for this graph.
1181
:return: A list of parent details for each key in revision_ids.
1182
Each parent details will be one of:
1183
* None when the key was missing
1184
* (NULL_REVISION,) when the key has no parents.
1185
* (parent_key, parent_key...) otherwise.
1187
parent_map = self.get_parent_map(revision_ids)
1188
return [parent_map.get(r, None) for r in revision_ids]
1190
def get_parent_map(self, keys):
1191
"""See graph._StackedParentsProvider.get_parent_map"""
1192
search_keys = set(keys)
1193
if NULL_REVISION in search_keys:
1194
search_keys.discard(NULL_REVISION)
1195
found_parents = {NULL_REVISION:[]}
1198
for index, key, value, refs in self.iter_entries(search_keys):
1201
parents = (NULL_REVISION,)
1202
found_parents[key] = parents
1203
return found_parents
1205
has_key = _has_key_from_parent_map
1207
def insert_index(self, pos, index):
1208
"""Insert a new index in the list of indices to query.
1210
:param pos: The position to insert the index.
1211
:param index: The index to insert.
1213
self._indices.insert(pos, index)
1215
def iter_all_entries(self):
1216
"""Iterate over all keys within the index
1218
Duplicate keys across child indices are presumed to have the same
1219
value and are only reported once.
1221
:return: An iterable of (index, key, reference_lists, value).
1222
There is no defined order for the result iteration - it will be in
1223
the most efficient order for the index.
1228
for index in self._indices:
1229
for node in index.iter_all_entries():
1230
if node[1] not in seen_keys:
1232
seen_keys.add(node[1])
1234
except errors.NoSuchFile:
1235
self._reload_or_raise()
1237
def iter_entries(self, keys):
1238
"""Iterate over keys within the index.
1240
Duplicate keys across child indices are presumed to have the same
1241
value and are only reported once.
1243
:param keys: An iterable providing the keys to be retrieved.
1244
:return: An iterable of (index, key, reference_lists, value). There is no
1245
defined order for the result iteration - it will be in the most
1246
efficient order for the index.
1251
for index in self._indices:
1254
for node in index.iter_entries(keys):
1255
keys.remove(node[1])
1258
except errors.NoSuchFile:
1259
self._reload_or_raise()
1261
def iter_entries_prefix(self, keys):
1262
"""Iterate over keys within the index using prefix matching.
1264
Duplicate keys across child indices are presumed to have the same
1265
value and are only reported once.
1267
Prefix matching is applied within the tuple of a key, not to within
1268
the bytestring of each key element. e.g. if you have the keys ('foo',
1269
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1270
only the former key is returned.
1272
:param keys: An iterable providing the key prefixes to be retrieved.
1273
Each key prefix takes the form of a tuple the length of a key, but
1274
with the last N elements 'None' rather than a regular bytestring.
1275
The first element cannot be 'None'.
1276
:return: An iterable as per iter_all_entries, but restricted to the
1277
keys with a matching prefix to those supplied. No additional keys
1278
will be returned, and every match that is in the index will be
1287
for index in self._indices:
1288
for node in index.iter_entries_prefix(keys):
1289
if node[1] in seen_keys:
1291
seen_keys.add(node[1])
1294
except errors.NoSuchFile:
1295
self._reload_or_raise()
1297
def key_count(self):
1298
"""Return an estimate of the number of keys in this index.
1300
For CombinedGraphIndex this is approximated by the sum of the keys of
1301
the child indices. As child indices may have duplicate keys this can
1302
have a maximum error of the number of child indices * largest number of
1307
return sum((index.key_count() for index in self._indices), 0)
1308
except errors.NoSuchFile:
1309
self._reload_or_raise()
1311
missing_keys = _missing_keys_from_parent_map
1313
def _reload_or_raise(self):
1314
"""We just got a NoSuchFile exception.
1316
Try to reload the indices, if it fails, just raise the current
1319
if self._reload_func is None:
1321
exc_type, exc_value, exc_traceback = sys.exc_info()
1322
trace.mutter('Trying to reload after getting exception: %s',
1324
if not self._reload_func():
1325
# We tried to reload, but nothing changed, so we fail anyway
1326
trace.mutter('_reload_func indicated nothing has changed.'
1327
' Raising original exception.')
1328
raise exc_type, exc_value, exc_traceback
1331
"""Validate that everything in the index can be accessed."""
1334
for index in self._indices:
1337
except errors.NoSuchFile:
1338
self._reload_or_raise()
1341
class InMemoryGraphIndex(GraphIndexBuilder):
1342
"""A GraphIndex which operates entirely out of memory and is mutable.
1344
This is designed to allow the accumulation of GraphIndex entries during a
1345
single write operation, where the accumulated entries need to be immediately
1346
available - for example via a CombinedGraphIndex.
1349
def add_nodes(self, nodes):
1350
"""Add nodes to the index.
1352
:param nodes: An iterable of (key, node_refs, value) entries to add.
1354
if self.reference_lists:
1355
for (key, value, node_refs) in nodes:
1356
self.add_node(key, value, node_refs)
1358
for (key, value) in nodes:
1359
self.add_node(key, value)
1361
def iter_all_entries(self):
1362
"""Iterate over all keys within the index
1364
:return: An iterable of (index, key, reference_lists, value). There is no
1365
defined order for the result iteration - it will be in the most
1366
efficient order for the index (in this case dictionary hash order).
1368
if 'evil' in debug.debug_flags:
1369
trace.mutter_callsite(3,
1370
"iter_all_entries scales with size of history.")
1371
if self.reference_lists:
1372
for key, (absent, references, value) in self._nodes.iteritems():
1374
yield self, key, value, references
1376
for key, (absent, references, value) in self._nodes.iteritems():
1378
yield self, key, value
1380
def iter_entries(self, keys):
1381
"""Iterate over keys within the index.
1383
:param keys: An iterable providing the keys to be retrieved.
1384
:return: An iterable of (index, key, value, reference_lists). There is no
1385
defined order for the result iteration - it will be in the most
1386
efficient order for the index (keys iteration order in this case).
1389
if self.reference_lists:
1390
for key in keys.intersection(self._keys):
1391
node = self._nodes[key]
1393
yield self, key, node[2], node[1]
1395
for key in keys.intersection(self._keys):
1396
node = self._nodes[key]
1398
yield self, key, node[2]
1400
def iter_entries_prefix(self, keys):
1401
"""Iterate over keys within the index using prefix matching.
1403
Prefix matching is applied within the tuple of a key, not to within
1404
the bytestring of each key element. e.g. if you have the keys ('foo',
1405
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1406
only the former key is returned.
1408
:param keys: An iterable providing the key prefixes to be retrieved.
1409
Each key prefix takes the form of a tuple the length of a key, but
1410
with the last N elements 'None' rather than a regular bytestring.
1411
The first element cannot be 'None'.
1412
:return: An iterable as per iter_all_entries, but restricted to the
1413
keys with a matching prefix to those supplied. No additional keys
1414
will be returned, and every match that is in the index will be
1417
# XXX: To much duplication with the GraphIndex class; consider finding
1418
# a good place to pull out the actual common logic.
1422
if self._key_length == 1:
1426
raise errors.BadIndexKey(key)
1427
if len(key) != self._key_length:
1428
raise errors.BadIndexKey(key)
1429
node = self._nodes[key]
1432
if self.reference_lists:
1433
yield self, key, node[2], node[1]
1435
yield self, key, node[2]
1437
nodes_by_key = self._get_nodes_by_key()
1441
raise errors.BadIndexKey(key)
1442
if len(key) != self._key_length:
1443
raise errors.BadIndexKey(key)
1444
# find what it refers to:
1445
key_dict = nodes_by_key
1446
elements = list(key)
1447
# find the subdict to return
1449
while len(elements) and elements[0] is not None:
1450
key_dict = key_dict[elements[0]]
1453
# a non-existant lookup.
1458
key_dict = dicts.pop(-1)
1459
# can't be empty or would not exist
1460
item, value = key_dict.iteritems().next()
1461
if type(value) == dict:
1463
dicts.extend(key_dict.itervalues())
1466
for value in key_dict.itervalues():
1467
yield (self, ) + value
1469
yield (self, ) + key_dict
1471
def key_count(self):
1472
"""Return an estimate of the number of keys in this index.
1474
For InMemoryGraphIndex the estimate is exact.
1476
return len(self._keys)
1479
"""In memory index's have no known corruption at the moment."""
1482
class GraphIndexPrefixAdapter(object):
1483
"""An adapter between GraphIndex with different key lengths.
1485
Queries against this will emit queries against the adapted Graph with the
1486
prefix added, queries for all items use iter_entries_prefix. The returned
1487
nodes will have their keys and node references adjusted to remove the
1488
prefix. Finally, an add_nodes_callback can be supplied - when called the
1489
nodes and references being added will have prefix prepended.
1492
def __init__(self, adapted, prefix, missing_key_length,
1493
add_nodes_callback=None):
1494
"""Construct an adapter against adapted with prefix."""
1495
self.adapted = adapted
1496
self.prefix_key = prefix + (None,)*missing_key_length
1497
self.prefix = prefix
1498
self.prefix_len = len(prefix)
1499
self.add_nodes_callback = add_nodes_callback
1501
def add_nodes(self, nodes):
1502
"""Add nodes to the index.
1504
:param nodes: An iterable of (key, node_refs, value) entries to add.
1506
# save nodes in case its an iterator
1507
nodes = tuple(nodes)
1508
translated_nodes = []
1510
# Add prefix_key to each reference node_refs is a tuple of tuples,
1511
# so split it apart, and add prefix_key to the internal reference
1512
for (key, value, node_refs) in nodes:
1513
adjusted_references = (
1514
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1515
for ref_list in node_refs))
1516
translated_nodes.append((self.prefix + key, value,
1517
adjusted_references))
1519
# XXX: TODO add an explicit interface for getting the reference list
1520
# status, to handle this bit of user-friendliness in the API more
1522
for (key, value) in nodes:
1523
translated_nodes.append((self.prefix + key, value))
1524
self.add_nodes_callback(translated_nodes)
1526
def add_node(self, key, value, references=()):
1527
"""Add a node to the index.
1529
:param key: The key. keys are non-empty tuples containing
1530
as many whitespace-free utf8 bytestrings as the key length
1531
defined for this index.
1532
:param references: An iterable of iterables of keys. Each is a
1533
reference to another key.
1534
:param value: The value to associate with the key. It may be any
1535
bytes as long as it does not contain \0 or \n.
1537
self.add_nodes(((key, value, references), ))
1539
def _strip_prefix(self, an_iter):
1540
"""Strip prefix data from nodes and return it."""
1541
for node in an_iter:
1543
if node[1][:self.prefix_len] != self.prefix:
1544
raise errors.BadIndexData(self)
1545
for ref_list in node[3]:
1546
for ref_node in ref_list:
1547
if ref_node[:self.prefix_len] != self.prefix:
1548
raise errors.BadIndexData(self)
1549
yield node[0], node[1][self.prefix_len:], node[2], (
1550
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1551
for ref_list in node[3]))
1553
def iter_all_entries(self):
1554
"""Iterate over all keys within the index
1556
iter_all_entries is implemented against the adapted index using
1557
iter_entries_prefix.
1559
:return: An iterable of (index, key, reference_lists, value). There is no
1560
defined order for the result iteration - it will be in the most
1561
efficient order for the index (in this case dictionary hash order).
1563
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1565
def iter_entries(self, keys):
1566
"""Iterate over keys within the index.
1568
:param keys: An iterable providing the keys to be retrieved.
1569
:return: An iterable of (index, key, value, reference_lists). There is no
1570
defined order for the result iteration - it will be in the most
1571
efficient order for the index (keys iteration order in this case).
1573
return self._strip_prefix(self.adapted.iter_entries(
1574
self.prefix + key for key in keys))
1576
def iter_entries_prefix(self, keys):
1577
"""Iterate over keys within the index using prefix matching.
1579
Prefix matching is applied within the tuple of a key, not to within
1580
the bytestring of each key element. e.g. if you have the keys ('foo',
1581
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1582
only the former key is returned.
1584
:param keys: An iterable providing the key prefixes to be retrieved.
1585
Each key prefix takes the form of a tuple the length of a key, but
1586
with the last N elements 'None' rather than a regular bytestring.
1587
The first element cannot be 'None'.
1588
:return: An iterable as per iter_all_entries, but restricted to the
1589
keys with a matching prefix to those supplied. No additional keys
1590
will be returned, and every match that is in the index will be
1593
return self._strip_prefix(self.adapted.iter_entries_prefix(
1594
self.prefix + key for key in keys))
1596
def key_count(self):
1597
"""Return an estimate of the number of keys in this index.
1599
For GraphIndexPrefixAdapter this is relatively expensive - key
1600
iteration with the prefix is done.
1602
return len(list(self.iter_all_entries()))
1605
"""Call the adapted's validate."""
1606
self.adapted.validate()
added
removed
bzrlib/index.py
