1
# Copyright (C) 2007-2010 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
43
from bzrlib.static_tuple import StaticTuple
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
96
# A dict of {key: (absent, ref_lists, value)}
98
# Keys that are referenced but not actually present in this index
99
self._absent_keys = set()
100
self._nodes_by_key = None
101
self._key_length = key_elements
102
self._optimize_for_size = False
103
self._combine_backing_indices = True
105
def _check_key(self, key):
106
"""Raise BadIndexKey if key is not a valid key for this index."""
107
if type(key) not in (tuple, StaticTuple):
108
raise errors.BadIndexKey(key)
109
if self._key_length != len(key):
110
raise errors.BadIndexKey(key)
112
if not element or _whitespace_re.search(element) is not None:
113
raise errors.BadIndexKey(element)
115
def _external_references(self):
116
"""Return references that are not present in this index.
120
# TODO: JAM 2008-11-21 This makes an assumption about how the reference
121
# lists are used. It is currently correct for pack-0.92 through
122
# 1.9, which use the node references (3rd column) second
123
# reference list as the compression parent. Perhaps this should
124
# be moved into something higher up the stack, since it
125
# makes assumptions about how the index is used.
126
if self.reference_lists > 1:
127
for node in self.iter_all_entries():
129
refs.update(node[3][1])
132
# If reference_lists == 0 there can be no external references, and
133
# if reference_lists == 1, then there isn't a place to store the
137
def _get_nodes_by_key(self):
138
if self._nodes_by_key is None:
140
if self.reference_lists:
141
for key, (absent, references, value) in self._nodes.iteritems():
144
key_dict = nodes_by_key
145
for subkey in key[:-1]:
146
key_dict = key_dict.setdefault(subkey, {})
147
key_dict[key[-1]] = key, value, references
149
for key, (absent, references, value) in self._nodes.iteritems():
152
key_dict = nodes_by_key
153
for subkey in key[:-1]:
154
key_dict = key_dict.setdefault(subkey, {})
155
key_dict[key[-1]] = key, value
156
self._nodes_by_key = nodes_by_key
157
return self._nodes_by_key
159
def _update_nodes_by_key(self, key, value, node_refs):
160
"""Update the _nodes_by_key dict with a new key.
162
For a key of (foo, bar, baz) create
163
_nodes_by_key[foo][bar][baz] = key_value
165
if self._nodes_by_key is None:
167
key_dict = self._nodes_by_key
168
if self.reference_lists:
169
key_value = StaticTuple(key, value, node_refs)
171
key_value = StaticTuple(key, value)
172
for subkey in key[:-1]:
173
key_dict = key_dict.setdefault(subkey, {})
174
key_dict[key[-1]] = key_value
176
def _check_key_ref_value(self, key, references, value):
177
"""Check that 'key' and 'references' are all valid.
179
:param key: A key tuple. Must conform to the key interface (be a tuple,
180
be of the right length, not have any whitespace or nulls in any key
182
:param references: An iterable of reference lists. Something like
183
[[(ref, key)], [(ref, key), (other, key)]]
184
:param value: The value associate with this key. Must not contain
185
newlines or null characters.
186
:return: (node_refs, absent_references)
187
node_refs basically a packed form of 'references' where all
189
absent_references reference keys that are not in self._nodes.
190
This may contain duplicates if the same key is
191
referenced in multiple lists.
193
as_st = StaticTuple.from_sequence
195
if _newline_null_re.search(value) is not None:
196
raise errors.BadIndexValue(value)
197
if len(references) != self.reference_lists:
198
raise errors.BadIndexValue(references)
200
absent_references = []
201
for reference_list in references:
202
for reference in reference_list:
203
# If reference *is* in self._nodes, then we know it has already
205
if reference not in self._nodes:
206
self._check_key(reference)
207
absent_references.append(reference)
208
reference_list = as_st([as_st(ref).intern()
209
for ref in reference_list])
210
node_refs.append(reference_list)
211
return as_st(node_refs), absent_references
213
def add_node(self, key, value, references=()):
214
"""Add a node to the index.
216
:param key: The key. keys are non-empty tuples containing
217
as many whitespace-free utf8 bytestrings as the key length
218
defined for this index.
219
:param references: An iterable of iterables of keys. Each is a
220
reference to another key.
221
:param value: The value to associate with the key. It may be any
222
bytes as long as it does not contain \0 or \n.
225
absent_references) = self._check_key_ref_value(key, references, value)
226
if key in self._nodes and self._nodes[key][0] != 'a':
227
raise errors.BadIndexDuplicateKey(key, self)
228
for reference in absent_references:
229
# There may be duplicates, but I don't think it is worth worrying
231
self._nodes[reference] = ('a', (), '')
232
self._absent_keys.update(absent_references)
233
self._absent_keys.discard(key)
234
self._nodes[key] = ('', node_refs, value)
235
if self._nodes_by_key is not None and self._key_length > 1:
236
self._update_nodes_by_key(key, value, node_refs)
238
def clear_cache(self):
239
"""See GraphIndex.clear_cache()
241
This is a no-op, but we need the api to conform to a generic 'Index'
247
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
248
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
249
key_count = len(self._nodes) - len(self._absent_keys)
250
lines.append(_OPTION_LEN + str(key_count) + '\n')
251
prefix_length = sum(len(x) for x in lines)
252
# references are byte offsets. To avoid having to do nasty
253
# polynomial work to resolve offsets (references to later in the
254
# file cannot be determined until all the inbetween references have
255
# been calculated too) we pad the offsets with 0's to make them be
256
# of consistent length. Using binary offsets would break the trivial
258
# to calculate the width of zero's needed we do three passes:
259
# one to gather all the non-reference data and the number of references.
260
# one to pad all the data with reference-length and determine entry
264
# forward sorted by key. In future we may consider topological sorting,
265
# at the cost of table scans for direct lookup, or a second index for
267
nodes = sorted(self._nodes.items())
268
# if we do not prepass, we don't know how long it will be up front.
269
expected_bytes = None
270
# we only need to pre-pass if we have reference lists at all.
271
if self.reference_lists:
273
non_ref_bytes = prefix_length
275
# TODO use simple multiplication for the constants in this loop.
276
for key, (absent, references, value) in nodes:
277
# record the offset known *so far* for this key:
278
# the non reference bytes to date, and the total references to
279
# date - saves reaccumulating on the second pass
280
key_offset_info.append((key, non_ref_bytes, total_references))
281
# key is literal, value is literal, there are 3 null's, 1 NL
282
# key is variable length tuple, \x00 between elements
283
non_ref_bytes += sum(len(element) for element in key)
284
if self._key_length > 1:
285
non_ref_bytes += self._key_length - 1
286
# value is literal bytes, there are 3 null's, 1 NL.
287
non_ref_bytes += len(value) + 3 + 1
288
# one byte for absent if set.
291
elif self.reference_lists:
292
# (ref_lists -1) tabs
293
non_ref_bytes += self.reference_lists - 1
294
# (ref-1 cr's per ref_list)
295
for ref_list in references:
296
# how many references across the whole file?
297
total_references += len(ref_list)
298
# accrue reference separators
300
non_ref_bytes += len(ref_list) - 1
301
# how many digits are needed to represent the total byte count?
303
possible_total_bytes = non_ref_bytes + total_references*digits
304
while 10 ** digits < possible_total_bytes:
306
possible_total_bytes = non_ref_bytes + total_references*digits
307
expected_bytes = possible_total_bytes + 1 # terminating newline
308
# resolve key addresses.
310
for key, non_ref_bytes, total_references in key_offset_info:
311
key_addresses[key] = non_ref_bytes + total_references*digits
313
format_string = '%%0%sd' % digits
314
for key, (absent, references, value) in nodes:
315
flattened_references = []
316
for ref_list in references:
318
for reference in ref_list:
319
ref_addresses.append(format_string % key_addresses[reference])
320
flattened_references.append('\r'.join(ref_addresses))
321
string_key = '\x00'.join(key)
322
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
323
'\t'.join(flattened_references), value))
325
result = StringIO(''.join(lines))
326
if expected_bytes and len(result.getvalue()) != expected_bytes:
327
raise errors.BzrError('Failed index creation. Internal error:'
328
' mismatched output length and expected length: %d %d' %
329
(len(result.getvalue()), expected_bytes))
332
def set_optimize(self, for_size=None, combine_backing_indices=None):
333
"""Change how the builder tries to optimize the result.
335
:param for_size: Tell the builder to try and make the index as small as
337
:param combine_backing_indices: If the builder spills to disk to save
338
memory, should the on-disk indices be combined. Set to True if you
339
are going to be probing the index, but to False if you are not. (If
340
you are not querying, then the time spent combining is wasted.)
343
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
345
if for_size is not None:
346
self._optimize_for_size = for_size
347
if combine_backing_indices is not None:
348
self._combine_backing_indices = combine_backing_indices
350
def find_ancestry(self, keys, ref_list_num):
351
"""See CombinedGraphIndex.find_ancestry()"""
357
for _, key, value, ref_lists in self.iter_entries(pending):
358
parent_keys = ref_lists[ref_list_num]
359
parent_map[key] = parent_keys
360
next_pending.update([p for p in parent_keys if p not in
362
missing_keys.update(pending.difference(parent_map))
363
pending = next_pending
364
return parent_map, missing_keys
367
class GraphIndex(object):
368
"""An index for data with embedded graphs.
370
The index maps keys to a list of key reference lists, and a value.
371
Each node has the same number of key reference lists. Each key reference
372
list can be empty or an arbitrary length. The value is an opaque NULL
373
terminated string without any newlines. The storage of the index is
374
hidden in the interface: keys and key references are always tuples of
375
bytestrings, never the internal representation (e.g. dictionary offsets).
377
It is presumed that the index will not be mutated - it is static data.
379
Successive iter_all_entries calls will read the entire index each time.
380
Additionally, iter_entries calls will read the index linearly until the
381
desired keys are found. XXX: This must be fixed before the index is
382
suitable for production use. :XXX
385
def __init__(self, transport, name, size, unlimited_cache=False):
386
"""Open an index called name on transport.
388
:param transport: A bzrlib.transport.Transport.
389
:param name: A path to provide to transport API calls.
390
:param size: The size of the index in bytes. This is used for bisection
391
logic to perform partial index reads. While the size could be
392
obtained by statting the file this introduced an additional round
393
trip as well as requiring stat'able transports, both of which are
394
avoided by having it supplied. If size is None, then bisection
395
support will be disabled and accessing the index will just stream
398
self._transport = transport
400
# Becomes a dict of key:(value, reference-list-byte-locations) used by
401
# the bisection interface to store parsed but not resolved keys.
402
self._bisect_nodes = None
403
# Becomes a dict of key:(value, reference-list-keys) which are ready to
404
# be returned directly to callers.
406
# a sorted list of slice-addresses for the parsed bytes of the file.
407
# e.g. (0,1) would mean that byte 0 is parsed.
408
self._parsed_byte_map = []
409
# a sorted list of keys matching each slice address for parsed bytes
410
# e.g. (None, 'foo@bar') would mean that the first byte contained no
411
# key, and the end byte of the slice is the of the data for 'foo@bar'
412
self._parsed_key_map = []
413
self._key_count = None
414
self._keys_by_offset = None
415
self._nodes_by_key = None
417
# The number of bytes we've read so far in trying to process this file
420
def __eq__(self, other):
421
"""Equal when self and other were created with the same parameters."""
423
type(self) == type(other) and
424
self._transport == other._transport and
425
self._name == other._name and
426
self._size == other._size)
428
def __ne__(self, other):
429
return not self.__eq__(other)
432
return "%s(%r)" % (self.__class__.__name__,
433
self._transport.abspath(self._name))
435
def _buffer_all(self, stream=None):
436
"""Buffer all the index data.
438
Mutates self._nodes and self.keys_by_offset.
440
if self._nodes is not None:
441
# We already did this
443
if 'index' in debug.debug_flags:
444
mutter('Reading entire index %s', self._transport.abspath(self._name))
446
stream = self._transport.get(self._name)
447
self._read_prefix(stream)
448
self._expected_elements = 3 + self._key_length
450
# raw data keyed by offset
451
self._keys_by_offset = {}
452
# ready-to-return key:value or key:value, node_ref_lists
454
self._nodes_by_key = None
457
lines = stream.read().split('\n')
460
_, _, _, trailers = self._parse_lines(lines, pos)
461
for key, absent, references, value in self._keys_by_offset.itervalues():
464
# resolve references:
465
if self.node_ref_lists:
466
node_value = (value, self._resolve_references(references))
469
self._nodes[key] = node_value
470
# cache the keys for quick set intersections
472
# there must be one line - the empty trailer line.
473
raise errors.BadIndexData(self)
475
def clear_cache(self):
476
"""Clear out any cached/memoized values.
478
This can be called at any time, but generally it is used when we have
479
extracted some information, but don't expect to be requesting any more
483
def external_references(self, ref_list_num):
484
"""Return references that are not present in this index.
487
if ref_list_num + 1 > self.node_ref_lists:
488
raise ValueError('No ref list %d, index has %d ref lists'
489
% (ref_list_num, self.node_ref_lists))
492
for key, (value, ref_lists) in nodes.iteritems():
493
ref_list = ref_lists[ref_list_num]
494
refs.update([ref for ref in ref_list if ref not in nodes])
497
def _get_nodes_by_key(self):
498
if self._nodes_by_key is None:
500
if self.node_ref_lists:
501
for key, (value, references) in self._nodes.iteritems():
502
key_dict = nodes_by_key
503
for subkey in key[:-1]:
504
key_dict = key_dict.setdefault(subkey, {})
505
key_dict[key[-1]] = key, value, references
507
for key, value in self._nodes.iteritems():
508
key_dict = nodes_by_key
509
for subkey in key[:-1]:
510
key_dict = key_dict.setdefault(subkey, {})
511
key_dict[key[-1]] = key, value
512
self._nodes_by_key = nodes_by_key
513
return self._nodes_by_key
515
def iter_all_entries(self):
516
"""Iterate over all keys within the index.
518
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
519
The former tuple is used when there are no reference lists in the
520
index, making the API compatible with simple key:value index types.
521
There is no defined order for the result iteration - it will be in
522
the most efficient order for the index.
524
if 'evil' in debug.debug_flags:
525
trace.mutter_callsite(3,
526
"iter_all_entries scales with size of history.")
527
if self._nodes is None:
529
if self.node_ref_lists:
530
for key, (value, node_ref_lists) in self._nodes.iteritems():
531
yield self, key, value, node_ref_lists
533
for key, value in self._nodes.iteritems():
534
yield self, key, value
536
def _read_prefix(self, stream):
537
signature = stream.read(len(self._signature()))
538
if not signature == self._signature():
539
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
540
options_line = stream.readline()
541
if not options_line.startswith(_OPTION_NODE_REFS):
542
raise errors.BadIndexOptions(self)
544
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
546
raise errors.BadIndexOptions(self)
547
options_line = stream.readline()
548
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
549
raise errors.BadIndexOptions(self)
551
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
553
raise errors.BadIndexOptions(self)
554
options_line = stream.readline()
555
if not options_line.startswith(_OPTION_LEN):
556
raise errors.BadIndexOptions(self)
558
self._key_count = int(options_line[len(_OPTION_LEN):-1])
560
raise errors.BadIndexOptions(self)
562
def _resolve_references(self, references):
563
"""Return the resolved key references for references.
565
References are resolved by looking up the location of the key in the
566
_keys_by_offset map and substituting the key name, preserving ordering.
568
:param references: An iterable of iterables of key locations. e.g.
570
:return: A tuple of tuples of keys.
573
for ref_list in references:
574
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
575
return tuple(node_refs)
577
def _find_index(self, range_map, key):
578
"""Helper for the _parsed_*_index calls.
580
Given a range map - [(start, end), ...], finds the index of the range
581
in the map for key if it is in the map, and if it is not there, the
582
immediately preceeding range in the map.
584
result = bisect_right(range_map, key) - 1
585
if result + 1 < len(range_map):
586
# check the border condition, it may be in result + 1
587
if range_map[result + 1][0] == key[0]:
591
def _parsed_byte_index(self, offset):
592
"""Return the index of the entry immediately before offset.
594
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
595
there is one unparsed byte (the 11th, addressed as[10]). then:
596
asking for 0 will return 0
597
asking for 10 will return 0
598
asking for 11 will return 1
599
asking for 12 will return 1
602
return self._find_index(self._parsed_byte_map, key)
604
def _parsed_key_index(self, key):
605
"""Return the index of the entry immediately before key.
607
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
608
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
609
have been parsed, then:
610
asking for '' will return 0
611
asking for 'a' will return 0
612
asking for 'b' will return 1
613
asking for 'e' will return 1
615
search_key = (key, None)
616
return self._find_index(self._parsed_key_map, search_key)
618
def _is_parsed(self, offset):
619
"""Returns True if offset has been parsed."""
620
index = self._parsed_byte_index(offset)
621
if index == len(self._parsed_byte_map):
622
return offset < self._parsed_byte_map[index - 1][1]
623
start, end = self._parsed_byte_map[index]
624
return offset >= start and offset < end
626
def _iter_entries_from_total_buffer(self, keys):
627
"""Iterate over keys when the entire index is parsed."""
628
# Note: See the note in BTreeBuilder.iter_entries for why we don't use
629
# .intersection() here
631
keys = [key for key in keys if key in nodes]
632
if self.node_ref_lists:
634
value, node_refs = nodes[key]
635
yield self, key, value, node_refs
638
yield self, key, nodes[key]
640
def iter_entries(self, keys):
641
"""Iterate over keys within the index.
643
:param keys: An iterable providing the keys to be retrieved.
644
:return: An iterable as per iter_all_entries, but restricted to the
645
keys supplied. No additional keys will be returned, and every
646
key supplied that is in the index will be returned.
651
if self._size is None and self._nodes is None:
654
# We fit about 20 keys per minimum-read (4K), so if we are looking for
655
# more than 1/20th of the index its likely (assuming homogenous key
656
# spread) that we'll read the entire index. If we're going to do that,
657
# buffer the whole thing. A better analysis might take key spread into
658
# account - but B+Tree indices are better anyway.
659
# We could look at all data read, and use a threshold there, which will
660
# trigger on ancestry walks, but that is not yet fully mapped out.
661
if self._nodes is None and len(keys) * 20 > self.key_count():
663
if self._nodes is not None:
664
return self._iter_entries_from_total_buffer(keys)
666
return (result[1] for result in bisect_multi_bytes(
667
self._lookup_keys_via_location, self._size, keys))
669
def iter_entries_prefix(self, keys):
670
"""Iterate over keys within the index using prefix matching.
672
Prefix matching is applied within the tuple of a key, not to within
673
the bytestring of each key element. e.g. if you have the keys ('foo',
674
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
675
only the former key is returned.
677
WARNING: Note that this method currently causes a full index parse
678
unconditionally (which is reasonably appropriate as it is a means for
679
thunking many small indices into one larger one and still supplies
680
iter_all_entries at the thunk layer).
682
:param keys: An iterable providing the key prefixes to be retrieved.
683
Each key prefix takes the form of a tuple the length of a key, but
684
with the last N elements 'None' rather than a regular bytestring.
685
The first element cannot be 'None'.
686
:return: An iterable as per iter_all_entries, but restricted to the
687
keys with a matching prefix to those supplied. No additional keys
688
will be returned, and every match that is in the index will be
694
# load data - also finds key lengths
695
if self._nodes is None:
697
if self._key_length == 1:
701
raise errors.BadIndexKey(key)
702
if len(key) != self._key_length:
703
raise errors.BadIndexKey(key)
704
if self.node_ref_lists:
705
value, node_refs = self._nodes[key]
706
yield self, key, value, node_refs
708
yield self, key, self._nodes[key]
710
nodes_by_key = self._get_nodes_by_key()
714
raise errors.BadIndexKey(key)
715
if len(key) != self._key_length:
716
raise errors.BadIndexKey(key)
717
# find what it refers to:
718
key_dict = nodes_by_key
720
# find the subdict whose contents should be returned.
722
while len(elements) and elements[0] is not None:
723
key_dict = key_dict[elements[0]]
726
# a non-existant lookup.
731
key_dict = dicts.pop(-1)
732
# can't be empty or would not exist
733
item, value = key_dict.iteritems().next()
734
if type(value) == dict:
736
dicts.extend(key_dict.itervalues())
739
for value in key_dict.itervalues():
740
# each value is the key:value:node refs tuple
742
yield (self, ) + value
744
# the last thing looked up was a terminal element
745
yield (self, ) + key_dict
747
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
748
"""See BTreeIndex._find_ancestors."""
749
# The api can be implemented as a trivial overlay on top of
750
# iter_entries, it is not an efficient implementation, but it at least
754
for index, key, value, refs in self.iter_entries(keys):
755
parent_keys = refs[ref_list_num]
757
parent_map[key] = parent_keys
758
search_keys.update(parent_keys)
759
# Figure out what, if anything, was missing
760
missing_keys.update(set(keys).difference(found_keys))
761
search_keys = search_keys.difference(parent_map)
765
"""Return an estimate of the number of keys in this index.
767
For GraphIndex the estimate is exact.
769
if self._key_count is None:
770
self._read_and_parse([_HEADER_READV])
771
return self._key_count
773
def _lookup_keys_via_location(self, location_keys):
774
"""Public interface for implementing bisection.
776
If _buffer_all has been called, then all the data for the index is in
777
memory, and this method should not be called, as it uses a separate
778
cache because it cannot pre-resolve all indices, which buffer_all does
781
:param location_keys: A list of location(byte offset), key tuples.
782
:return: A list of (location_key, result) tuples as expected by
783
bzrlib.bisect_multi.bisect_multi_bytes.
785
# Possible improvements:
786
# - only bisect lookup each key once
787
# - sort the keys first, and use that to reduce the bisection window
789
# this progresses in three parts:
792
# attempt to answer the question from the now in memory data.
793
# build the readv request
794
# for each location, ask for 800 bytes - much more than rows we've seen
797
for location, key in location_keys:
798
# can we answer from cache?
799
if self._bisect_nodes and key in self._bisect_nodes:
800
# We have the key parsed.
802
index = self._parsed_key_index(key)
803
if (len(self._parsed_key_map) and
804
self._parsed_key_map[index][0] <= key and
805
(self._parsed_key_map[index][1] >= key or
806
# end of the file has been parsed
807
self._parsed_byte_map[index][1] == self._size)):
808
# the key has been parsed, so no lookup is needed even if its
811
# - if we have examined this part of the file already - yes
812
index = self._parsed_byte_index(location)
813
if (len(self._parsed_byte_map) and
814
self._parsed_byte_map[index][0] <= location and
815
self._parsed_byte_map[index][1] > location):
816
# the byte region has been parsed, so no read is needed.
819
if location + length > self._size:
820
length = self._size - location
821
# todo, trim out parsed locations.
823
readv_ranges.append((location, length))
824
# read the header if needed
825
if self._bisect_nodes is None:
826
readv_ranges.append(_HEADER_READV)
827
self._read_and_parse(readv_ranges)
829
if self._nodes is not None:
830
# _read_and_parse triggered a _buffer_all because we requested the
832
for location, key in location_keys:
833
if key not in self._nodes: # not present
834
result.append(((location, key), False))
835
elif self.node_ref_lists:
836
value, refs = self._nodes[key]
837
result.append(((location, key),
838
(self, key, value, refs)))
840
result.append(((location, key),
841
(self, key, self._nodes[key])))
844
# - figure out <, >, missing, present
845
# - result present references so we can return them.
846
# keys that we cannot answer until we resolve references
847
pending_references = []
848
pending_locations = set()
849
for location, key in location_keys:
850
# can we answer from cache?
851
if key in self._bisect_nodes:
852
# the key has been parsed, so no lookup is needed
853
if self.node_ref_lists:
854
# the references may not have been all parsed.
855
value, refs = self._bisect_nodes[key]
856
wanted_locations = []
857
for ref_list in refs:
859
if ref not in self._keys_by_offset:
860
wanted_locations.append(ref)
862
pending_locations.update(wanted_locations)
863
pending_references.append((location, key))
865
result.append(((location, key), (self, key,
866
value, self._resolve_references(refs))))
868
result.append(((location, key),
869
(self, key, self._bisect_nodes[key])))
872
# has the region the key should be in, been parsed?
873
index = self._parsed_key_index(key)
874
if (self._parsed_key_map[index][0] <= key and
875
(self._parsed_key_map[index][1] >= key or
876
# end of the file has been parsed
877
self._parsed_byte_map[index][1] == self._size)):
878
result.append(((location, key), False))
880
# no, is the key above or below the probed location:
881
# get the range of the probed & parsed location
882
index = self._parsed_byte_index(location)
883
# if the key is below the start of the range, its below
884
if key < self._parsed_key_map[index][0]:
888
result.append(((location, key), direction))
890
# lookup data to resolve references
891
for location in pending_locations:
893
if location + length > self._size:
894
length = self._size - location
895
# TODO: trim out parsed locations (e.g. if the 800 is into the
896
# parsed region trim it, and dont use the adjust_for_latency
899
readv_ranges.append((location, length))
900
self._read_and_parse(readv_ranges)
901
if self._nodes is not None:
902
# The _read_and_parse triggered a _buffer_all, grab the data and
904
for location, key in pending_references:
905
value, refs = self._nodes[key]
906
result.append(((location, key), (self, key, value, refs)))
908
for location, key in pending_references:
909
# answer key references we had to look-up-late.
910
value, refs = self._bisect_nodes[key]
911
result.append(((location, key), (self, key,
912
value, self._resolve_references(refs))))
915
def _parse_header_from_bytes(self, bytes):
916
"""Parse the header from a region of bytes.
918
:param bytes: The data to parse.
919
:return: An offset, data tuple such as readv yields, for the unparsed
920
data. (which may length 0).
922
signature = bytes[0:len(self._signature())]
923
if not signature == self._signature():
924
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
925
lines = bytes[len(self._signature()):].splitlines()
926
options_line = lines[0]
927
if not options_line.startswith(_OPTION_NODE_REFS):
928
raise errors.BadIndexOptions(self)
930
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
932
raise errors.BadIndexOptions(self)
933
options_line = lines[1]
934
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
935
raise errors.BadIndexOptions(self)
937
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
939
raise errors.BadIndexOptions(self)
940
options_line = lines[2]
941
if not options_line.startswith(_OPTION_LEN):
942
raise errors.BadIndexOptions(self)
944
self._key_count = int(options_line[len(_OPTION_LEN):])
946
raise errors.BadIndexOptions(self)
947
# calculate the bytes we have processed
948
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
950
self._parsed_bytes(0, None, header_end, None)
951
# setup parsing state
952
self._expected_elements = 3 + self._key_length
953
# raw data keyed by offset
954
self._keys_by_offset = {}
955
# keys with the value and node references
956
self._bisect_nodes = {}
957
return header_end, bytes[header_end:]
959
def _parse_region(self, offset, data):
960
"""Parse node data returned from a readv operation.
962
:param offset: The byte offset the data starts at.
963
:param data: The data to parse.
967
end = offset + len(data)
970
# Trivial test - if the current index's end is within the
971
# low-matching parsed range, we're done.
972
index = self._parsed_byte_index(high_parsed)
973
if end < self._parsed_byte_map[index][1]:
975
# print "[%d:%d]" % (offset, end), \
976
# self._parsed_byte_map[index:index + 2]
977
high_parsed, last_segment = self._parse_segment(
978
offset, data, end, index)
982
def _parse_segment(self, offset, data, end, index):
983
"""Parse one segment of data.
985
:param offset: Where 'data' begins in the file.
986
:param data: Some data to parse a segment of.
987
:param end: Where data ends
988
:param index: The current index into the parsed bytes map.
989
:return: True if the parsed segment is the last possible one in the
991
:return: high_parsed_byte, last_segment.
992
high_parsed_byte is the location of the highest parsed byte in this
993
segment, last_segment is True if the parsed segment is the last
994
possible one in the data block.
996
# default is to use all data
998
# accomodate overlap with data before this.
999
if offset < self._parsed_byte_map[index][1]:
1000
# overlaps the lower parsed region
1001
# skip the parsed data
1002
trim_start = self._parsed_byte_map[index][1] - offset
1003
# don't trim the start for \n
1004
start_adjacent = True
1005
elif offset == self._parsed_byte_map[index][1]:
1006
# abuts the lower parsed region
1009
# do not trim anything
1010
start_adjacent = True
1012
# does not overlap the lower parsed region
1015
# but trim the leading \n
1016
start_adjacent = False
1017
if end == self._size:
1018
# lines up to the end of all data:
1021
# do not strip to the last \n
1024
elif index + 1 == len(self._parsed_byte_map):
1025
# at the end of the parsed data
1028
# but strip to the last \n
1029
end_adjacent = False
1031
elif end == self._parsed_byte_map[index + 1][0]:
1032
# buts up against the next parsed region
1035
# do not strip to the last \n
1038
elif end > self._parsed_byte_map[index + 1][0]:
1039
# overlaps into the next parsed region
1040
# only consider the unparsed data
1041
trim_end = self._parsed_byte_map[index + 1][0] - offset
1042
# do not strip to the last \n as we know its an entire record
1044
last_segment = end < self._parsed_byte_map[index + 1][1]
1046
# does not overlap into the next region
1049
# but strip to the last \n
1050
end_adjacent = False
1052
# now find bytes to discard if needed
1053
if not start_adjacent:
1054
# work around python bug in rfind
1055
if trim_start is None:
1056
trim_start = data.find('\n') + 1
1058
trim_start = data.find('\n', trim_start) + 1
1059
if not (trim_start != 0):
1060
raise AssertionError('no \n was present')
1061
# print 'removing start', offset, trim_start, repr(data[:trim_start])
1062
if not end_adjacent:
1063
# work around python bug in rfind
1064
if trim_end is None:
1065
trim_end = data.rfind('\n') + 1
1067
trim_end = data.rfind('\n', None, trim_end) + 1
1068
if not (trim_end != 0):
1069
raise AssertionError('no \n was present')
1070
# print 'removing end', offset, trim_end, repr(data[trim_end:])
1071
# adjust offset and data to the parseable data.
1072
trimmed_data = data[trim_start:trim_end]
1073
if not (trimmed_data):
1074
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
1075
% (trim_start, trim_end, offset, offset + len(data)))
1077
offset += trim_start
1078
# print "parsing", repr(trimmed_data)
1079
# splitlines mangles the \r delimiters.. don't use it.
1080
lines = trimmed_data.split('\n')
1083
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1084
for key, value in nodes:
1085
self._bisect_nodes[key] = value
1086
self._parsed_bytes(offset, first_key,
1087
offset + len(trimmed_data), last_key)
1088
return offset + len(trimmed_data), last_segment
1090
def _parse_lines(self, lines, pos):
1097
# must be at the end
1099
if not (self._size == pos + 1):
1100
raise AssertionError("%s %s" % (self._size, pos))
1103
elements = line.split('\0')
1104
if len(elements) != self._expected_elements:
1105
raise errors.BadIndexData(self)
1106
# keys are tuples. Each element is a string that may occur many
1107
# times, so we intern them to save space. AB, RC, 200807
1108
key = tuple([intern(element) for element in elements[:self._key_length]])
1109
if first_key is None:
1111
absent, references, value = elements[-3:]
1113
for ref_string in references.split('\t'):
1114
ref_lists.append(tuple([
1115
int(ref) for ref in ref_string.split('\r') if ref
1117
ref_lists = tuple(ref_lists)
1118
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1119
pos += len(line) + 1 # +1 for the \n
1122
if self.node_ref_lists:
1123
node_value = (value, ref_lists)
1126
nodes.append((key, node_value))
1127
# print "parsed ", key
1128
return first_key, key, nodes, trailers
1130
def _parsed_bytes(self, start, start_key, end, end_key):
1131
"""Mark the bytes from start to end as parsed.
1133
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1136
:param start: The start of the parsed region.
1137
:param end: The end of the parsed region.
1139
index = self._parsed_byte_index(start)
1140
new_value = (start, end)
1141
new_key = (start_key, end_key)
1143
# first range parsed is always the beginning.
1144
self._parsed_byte_map.insert(index, new_value)
1145
self._parsed_key_map.insert(index, new_key)
1149
# extend lower region
1150
# extend higher region
1151
# combine two regions
1152
if (index + 1 < len(self._parsed_byte_map) and
1153
self._parsed_byte_map[index][1] == start and
1154
self._parsed_byte_map[index + 1][0] == end):
1155
# combine two regions
1156
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1157
self._parsed_byte_map[index + 1][1])
1158
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1159
self._parsed_key_map[index + 1][1])
1160
del self._parsed_byte_map[index + 1]
1161
del self._parsed_key_map[index + 1]
1162
elif self._parsed_byte_map[index][1] == start:
1163
# extend the lower entry
1164
self._parsed_byte_map[index] = (
1165
self._parsed_byte_map[index][0], end)
1166
self._parsed_key_map[index] = (
1167
self._parsed_key_map[index][0], end_key)
1168
elif (index + 1 < len(self._parsed_byte_map) and
1169
self._parsed_byte_map[index + 1][0] == end):
1170
# extend the higher entry
1171
self._parsed_byte_map[index + 1] = (
1172
start, self._parsed_byte_map[index + 1][1])
1173
self._parsed_key_map[index + 1] = (
1174
start_key, self._parsed_key_map[index + 1][1])
1177
self._parsed_byte_map.insert(index + 1, new_value)
1178
self._parsed_key_map.insert(index + 1, new_key)
1180
def _read_and_parse(self, readv_ranges):
1181
"""Read the ranges and parse the resulting data.
1183
:param readv_ranges: A prepared readv range list.
1185
if not readv_ranges:
1187
if self._nodes is None and self._bytes_read * 2 >= self._size:
1188
# We've already read more than 50% of the file and we are about to
1189
# request more data, just _buffer_all() and be done
1193
readv_data = self._transport.readv(self._name, readv_ranges, True,
1196
for offset, data in readv_data:
1197
self._bytes_read += len(data)
1198
if offset == 0 and len(data) == self._size:
1199
# We read the whole range, most likely because the
1200
# Transport upcast our readv ranges into one long request
1201
# for enough total data to grab the whole index.
1202
self._buffer_all(StringIO(data))
1204
if self._bisect_nodes is None:
1205
# this must be the start
1206
if not (offset == 0):
1207
raise AssertionError()
1208
offset, data = self._parse_header_from_bytes(data)
1209
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1210
self._parse_region(offset, data)
1212
def _signature(self):
1213
"""The file signature for this index type."""
1217
"""Validate that everything in the index can be accessed."""
1218
# iter_all validates completely at the moment, so just do that.
1219
for node in self.iter_all_entries():
1223
class CombinedGraphIndex(object):
1224
"""A GraphIndex made up from smaller GraphIndices.
1226
The backing indices must implement GraphIndex, and are presumed to be
1229
Queries against the combined index will be made against the first index,
1230
and then the second and so on. The order of index's can thus influence
1231
performance significantly. For example, if one index is on local disk and a
1232
second on a remote server, the local disk index should be before the other
1236
def __init__(self, indices, reload_func=None):
1237
"""Create a CombinedGraphIndex backed by indices.
1239
:param indices: An ordered list of indices to query for data.
1240
:param reload_func: A function to call if we find we are missing an
1241
index. Should have the form reload_func() => True/False to indicate
1242
if reloading actually changed anything.
1244
self._indices = indices
1245
self._reload_func = reload_func
1249
self.__class__.__name__,
1250
', '.join(map(repr, self._indices)))
1252
def clear_cache(self):
1253
"""See GraphIndex.clear_cache()"""
1254
for index in self._indices:
1257
def get_parent_map(self, keys):
1258
"""See graph.StackedParentsProvider.get_parent_map"""
1259
search_keys = set(keys)
1260
if NULL_REVISION in search_keys:
1261
search_keys.discard(NULL_REVISION)
1262
found_parents = {NULL_REVISION:[]}
1265
for index, key, value, refs in self.iter_entries(search_keys):
1268
parents = (NULL_REVISION,)
1269
found_parents[key] = parents
1270
return found_parents
1272
has_key = _has_key_from_parent_map
1274
def insert_index(self, pos, index):
1275
"""Insert a new index in the list of indices to query.
1277
:param pos: The position to insert the index.
1278
:param index: The index to insert.
1280
self._indices.insert(pos, index)
1282
def iter_all_entries(self):
1283
"""Iterate over all keys within the index
1285
Duplicate keys across child indices are presumed to have the same
1286
value and are only reported once.
1288
:return: An iterable of (index, key, reference_lists, value).
1289
There is no defined order for the result iteration - it will be in
1290
the most efficient order for the index.
1295
for index in self._indices:
1296
for node in index.iter_all_entries():
1297
if node[1] not in seen_keys:
1299
seen_keys.add(node[1])
1301
except errors.NoSuchFile:
1302
self._reload_or_raise()
1304
def iter_entries(self, keys):
1305
"""Iterate over keys within the index.
1307
Duplicate keys across child indices are presumed to have the same
1308
value and are only reported once.
1310
:param keys: An iterable providing the keys to be retrieved.
1311
:return: An iterable of (index, key, reference_lists, value). There is no
1312
defined order for the result iteration - it will be in the most
1313
efficient order for the index.
1318
for index in self._indices:
1321
for node in index.iter_entries(keys):
1322
keys.remove(node[1])
1325
except errors.NoSuchFile:
1326
self._reload_or_raise()
1328
def iter_entries_prefix(self, keys):
1329
"""Iterate over keys within the index using prefix matching.
1331
Duplicate keys across child indices are presumed to have the same
1332
value and are only reported once.
1334
Prefix matching is applied within the tuple of a key, not to within
1335
the bytestring of each key element. e.g. if you have the keys ('foo',
1336
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1337
only the former key is returned.
1339
:param keys: An iterable providing the key prefixes to be retrieved.
1340
Each key prefix takes the form of a tuple the length of a key, but
1341
with the last N elements 'None' rather than a regular bytestring.
1342
The first element cannot be 'None'.
1343
:return: An iterable as per iter_all_entries, but restricted to the
1344
keys with a matching prefix to those supplied. No additional keys
1345
will be returned, and every match that is in the index will be
1354
for index in self._indices:
1355
for node in index.iter_entries_prefix(keys):
1356
if node[1] in seen_keys:
1358
seen_keys.add(node[1])
1361
except errors.NoSuchFile:
1362
self._reload_or_raise()
1364
def find_ancestry(self, keys, ref_list_num):
1365
"""Find the complete ancestry for the given set of keys.
1367
Note that this is a whole-ancestry request, so it should be used
1370
:param keys: An iterable of keys to look for
1371
:param ref_list_num: The reference list which references the parents
1373
:return: (parent_map, missing_keys)
1375
missing_keys = set()
1377
keys_to_lookup = set(keys)
1379
while keys_to_lookup:
1380
# keys that *all* indexes claim are missing, stop searching them
1382
all_index_missing = None
1383
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1384
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1386
# len(missing_keys))
1387
for index_idx, index in enumerate(self._indices):
1388
# TODO: we should probably be doing something with
1389
# 'missing_keys' since we've already determined that
1390
# those revisions have not been found anywhere
1391
index_missing_keys = set()
1392
# Find all of the ancestry we can from this index
1393
# keep looking until the search_keys set is empty, which means
1394
# things we didn't find should be in index_missing_keys
1395
search_keys = keys_to_lookup
1397
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1398
# index_idx, len(search_keys),
1399
# len(parent_map), len(index_missing_keys))
1402
# TODO: ref_list_num should really be a parameter, since
1403
# CombinedGraphIndex does not know what the ref lists
1405
search_keys = index._find_ancestors(search_keys,
1406
ref_list_num, parent_map, index_missing_keys)
1407
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1408
# sub_generation, len(search_keys),
1409
# len(parent_map), len(index_missing_keys))
1410
# Now set whatever was missing to be searched in the next index
1411
keys_to_lookup = index_missing_keys
1412
if all_index_missing is None:
1413
all_index_missing = set(index_missing_keys)
1415
all_index_missing.intersection_update(index_missing_keys)
1416
if not keys_to_lookup:
1418
if all_index_missing is None:
1419
# There were no indexes, so all search keys are 'missing'
1420
missing_keys.update(keys_to_lookup)
1421
keys_to_lookup = None
1423
missing_keys.update(all_index_missing)
1424
keys_to_lookup.difference_update(all_index_missing)
1425
return parent_map, missing_keys
1427
def key_count(self):
1428
"""Return an estimate of the number of keys in this index.
1430
For CombinedGraphIndex this is approximated by the sum of the keys of
1431
the child indices. As child indices may have duplicate keys this can
1432
have a maximum error of the number of child indices * largest number of
1437
return sum((index.key_count() for index in self._indices), 0)
1438
except errors.NoSuchFile:
1439
self._reload_or_raise()
1441
missing_keys = _missing_keys_from_parent_map
1443
def _reload_or_raise(self):
1444
"""We just got a NoSuchFile exception.
1446
Try to reload the indices, if it fails, just raise the current
1449
if self._reload_func is None:
1451
exc_type, exc_value, exc_traceback = sys.exc_info()
1452
trace.mutter('Trying to reload after getting exception: %s',
1454
if not self._reload_func():
1455
# We tried to reload, but nothing changed, so we fail anyway
1456
trace.mutter('_reload_func indicated nothing has changed.'
1457
' Raising original exception.')
1458
raise exc_type, exc_value, exc_traceback
1461
"""Validate that everything in the index can be accessed."""
1464
for index in self._indices:
1467
except errors.NoSuchFile:
1468
self._reload_or_raise()
1471
class InMemoryGraphIndex(GraphIndexBuilder):
1472
"""A GraphIndex which operates entirely out of memory and is mutable.
1474
This is designed to allow the accumulation of GraphIndex entries during a
1475
single write operation, where the accumulated entries need to be immediately
1476
available - for example via a CombinedGraphIndex.
1479
def add_nodes(self, nodes):
1480
"""Add nodes to the index.
1482
:param nodes: An iterable of (key, node_refs, value) entries to add.
1484
if self.reference_lists:
1485
for (key, value, node_refs) in nodes:
1486
self.add_node(key, value, node_refs)
1488
for (key, value) in nodes:
1489
self.add_node(key, value)
1491
def iter_all_entries(self):
1492
"""Iterate over all keys within the index
1494
:return: An iterable of (index, key, reference_lists, value). There is no
1495
defined order for the result iteration - it will be in the most
1496
efficient order for the index (in this case dictionary hash order).
1498
if 'evil' in debug.debug_flags:
1499
trace.mutter_callsite(3,
1500
"iter_all_entries scales with size of history.")
1501
if self.reference_lists:
1502
for key, (absent, references, value) in self._nodes.iteritems():
1504
yield self, key, value, references
1506
for key, (absent, references, value) in self._nodes.iteritems():
1508
yield self, key, value
1510
def iter_entries(self, keys):
1511
"""Iterate over keys within the index.
1513
:param keys: An iterable providing the keys to be retrieved.
1514
:return: An iterable of (index, key, value, reference_lists). There is no
1515
defined order for the result iteration - it will be in the most
1516
efficient order for the index (keys iteration order in this case).
1518
# Note: See BTreeBuilder.iter_entries for an explanation of why we
1519
# aren't using set().intersection() here
1521
keys = [key for key in keys if key in nodes]
1522
if self.reference_lists:
1526
yield self, key, node[2], node[1]
1531
yield self, key, node[2]
1533
def iter_entries_prefix(self, keys):
1534
"""Iterate over keys within the index using prefix matching.
1536
Prefix matching is applied within the tuple of a key, not to within
1537
the bytestring of each key element. e.g. if you have the keys ('foo',
1538
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1539
only the former key is returned.
1541
:param keys: An iterable providing the key prefixes to be retrieved.
1542
Each key prefix takes the form of a tuple the length of a key, but
1543
with the last N elements 'None' rather than a regular bytestring.
1544
The first element cannot be 'None'.
1545
:return: An iterable as per iter_all_entries, but restricted to the
1546
keys with a matching prefix to those supplied. No additional keys
1547
will be returned, and every match that is in the index will be
1550
# XXX: To much duplication with the GraphIndex class; consider finding
1551
# a good place to pull out the actual common logic.
1555
if self._key_length == 1:
1559
raise errors.BadIndexKey(key)
1560
if len(key) != self._key_length:
1561
raise errors.BadIndexKey(key)
1562
node = self._nodes[key]
1565
if self.reference_lists:
1566
yield self, key, node[2], node[1]
1568
yield self, key, node[2]
1570
nodes_by_key = self._get_nodes_by_key()
1574
raise errors.BadIndexKey(key)
1575
if len(key) != self._key_length:
1576
raise errors.BadIndexKey(key)
1577
# find what it refers to:
1578
key_dict = nodes_by_key
1579
elements = list(key)
1580
# find the subdict to return
1582
while len(elements) and elements[0] is not None:
1583
key_dict = key_dict[elements[0]]
1586
# a non-existant lookup.
1591
key_dict = dicts.pop(-1)
1592
# can't be empty or would not exist
1593
item, value = key_dict.iteritems().next()
1594
if type(value) == dict:
1596
dicts.extend(key_dict.itervalues())
1599
for value in key_dict.itervalues():
1600
yield (self, ) + value
1602
yield (self, ) + key_dict
1604
def key_count(self):
1605
"""Return an estimate of the number of keys in this index.
1607
For InMemoryGraphIndex the estimate is exact.
1609
return len(self._nodes) - len(self._absent_keys)
1612
"""In memory index's have no known corruption at the moment."""
1615
class GraphIndexPrefixAdapter(object):
1616
"""An adapter between GraphIndex with different key lengths.
1618
Queries against this will emit queries against the adapted Graph with the
1619
prefix added, queries for all items use iter_entries_prefix. The returned
1620
nodes will have their keys and node references adjusted to remove the
1621
prefix. Finally, an add_nodes_callback can be supplied - when called the
1622
nodes and references being added will have prefix prepended.
1625
def __init__(self, adapted, prefix, missing_key_length,
1626
add_nodes_callback=None):
1627
"""Construct an adapter against adapted with prefix."""
1628
self.adapted = adapted
1629
self.prefix_key = prefix + (None,)*missing_key_length
1630
self.prefix = prefix
1631
self.prefix_len = len(prefix)
1632
self.add_nodes_callback = add_nodes_callback
1634
def add_nodes(self, nodes):
1635
"""Add nodes to the index.
1637
:param nodes: An iterable of (key, node_refs, value) entries to add.
1639
# save nodes in case its an iterator
1640
nodes = tuple(nodes)
1641
translated_nodes = []
1643
# Add prefix_key to each reference node_refs is a tuple of tuples,
1644
# so split it apart, and add prefix_key to the internal reference
1645
for (key, value, node_refs) in nodes:
1646
adjusted_references = (
1647
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1648
for ref_list in node_refs))
1649
translated_nodes.append((self.prefix + key, value,
1650
adjusted_references))
1652
# XXX: TODO add an explicit interface for getting the reference list
1653
# status, to handle this bit of user-friendliness in the API more
1655
for (key, value) in nodes:
1656
translated_nodes.append((self.prefix + key, value))
1657
self.add_nodes_callback(translated_nodes)
1659
def add_node(self, key, value, references=()):
1660
"""Add a node to the index.
1662
:param key: The key. keys are non-empty tuples containing
1663
as many whitespace-free utf8 bytestrings as the key length
1664
defined for this index.
1665
:param references: An iterable of iterables of keys. Each is a
1666
reference to another key.
1667
:param value: The value to associate with the key. It may be any
1668
bytes as long as it does not contain \0 or \n.
1670
self.add_nodes(((key, value, references), ))
1672
def _strip_prefix(self, an_iter):
1673
"""Strip prefix data from nodes and return it."""
1674
for node in an_iter:
1676
if node[1][:self.prefix_len] != self.prefix:
1677
raise errors.BadIndexData(self)
1678
for ref_list in node[3]:
1679
for ref_node in ref_list:
1680
if ref_node[:self.prefix_len] != self.prefix:
1681
raise errors.BadIndexData(self)
1682
yield node[0], node[1][self.prefix_len:], node[2], (
1683
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1684
for ref_list in node[3]))
1686
def iter_all_entries(self):
1687
"""Iterate over all keys within the index
1689
iter_all_entries is implemented against the adapted index using
1690
iter_entries_prefix.
1692
:return: An iterable of (index, key, reference_lists, value). There is no
1693
defined order for the result iteration - it will be in the most
1694
efficient order for the index (in this case dictionary hash order).
1696
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1698
def iter_entries(self, keys):
1699
"""Iterate over keys within the index.
1701
:param keys: An iterable providing the keys to be retrieved.
1702
:return: An iterable of (index, key, value, reference_lists). There is no
1703
defined order for the result iteration - it will be in the most
1704
efficient order for the index (keys iteration order in this case).
1706
return self._strip_prefix(self.adapted.iter_entries(
1707
self.prefix + key for key in keys))
1709
def iter_entries_prefix(self, keys):
1710
"""Iterate over keys within the index using prefix matching.
1712
Prefix matching is applied within the tuple of a key, not to within
1713
the bytestring of each key element. e.g. if you have the keys ('foo',
1714
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1715
only the former key is returned.
1717
:param keys: An iterable providing the key prefixes to be retrieved.
1718
Each key prefix takes the form of a tuple the length of a key, but
1719
with the last N elements 'None' rather than a regular bytestring.
1720
The first element cannot be 'None'.
1721
:return: An iterable as per iter_all_entries, but restricted to the
1722
keys with a matching prefix to those supplied. No additional keys
1723
will be returned, and every match that is in the index will be
1726
return self._strip_prefix(self.adapted.iter_entries_prefix(
1727
self.prefix + key for key in keys))
1729
def key_count(self):
1730
"""Return an estimate of the number of keys in this index.
1732
For GraphIndexPrefixAdapter this is relatively expensive - key
1733
iteration with the prefix is done.
1735
return len(list(self.iter_all_entries()))
1738
"""Call the adapted's validate."""
1739
self.adapted.validate()
added
removed
bzrlib/index.py
