1
# Copyright (C) 2007, 2008, 2009 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')
459
_, _, _, trailers = self._parse_lines(lines, pos)
460
for key, absent, references, value in self._keys_by_offset.itervalues():
463
# resolve references:
464
if self.node_ref_lists:
465
node_value = (value, self._resolve_references(references))
468
self._nodes[key] = node_value
469
# cache the keys for quick set intersections
471
# there must be one line - the empty trailer line.
472
raise errors.BadIndexData(self)
474
def clear_cache(self):
475
"""Clear out any cached/memoized values.
477
This can be called at any time, but generally it is used when we have
478
extracted some information, but don't expect to be requesting any more
482
def external_references(self, ref_list_num):
483
"""Return references that are not present in this index.
486
if ref_list_num + 1 > self.node_ref_lists:
487
raise ValueError('No ref list %d, index has %d ref lists'
488
% (ref_list_num, self.node_ref_lists))
491
for key, (value, ref_lists) in nodes.iteritems():
492
ref_list = ref_lists[ref_list_num]
493
refs.update([ref for ref in ref_list if ref not in nodes])
496
def _get_nodes_by_key(self):
497
if self._nodes_by_key is None:
499
if self.node_ref_lists:
500
for key, (value, references) in self._nodes.iteritems():
501
key_dict = nodes_by_key
502
for subkey in key[:-1]:
503
key_dict = key_dict.setdefault(subkey, {})
504
key_dict[key[-1]] = key, value, references
506
for key, value in self._nodes.iteritems():
507
key_dict = nodes_by_key
508
for subkey in key[:-1]:
509
key_dict = key_dict.setdefault(subkey, {})
510
key_dict[key[-1]] = key, value
511
self._nodes_by_key = nodes_by_key
512
return self._nodes_by_key
514
def iter_all_entries(self):
515
"""Iterate over all keys within the index.
517
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
518
The former tuple is used when there are no reference lists in the
519
index, making the API compatible with simple key:value index types.
520
There is no defined order for the result iteration - it will be in
521
the most efficient order for the index.
523
if 'evil' in debug.debug_flags:
524
trace.mutter_callsite(3,
525
"iter_all_entries scales with size of history.")
526
if self._nodes is None:
528
if self.node_ref_lists:
529
for key, (value, node_ref_lists) in self._nodes.iteritems():
530
yield self, key, value, node_ref_lists
532
for key, value in self._nodes.iteritems():
533
yield self, key, value
535
def _read_prefix(self, stream):
536
signature = stream.read(len(self._signature()))
537
if not signature == self._signature():
538
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
539
options_line = stream.readline()
540
if not options_line.startswith(_OPTION_NODE_REFS):
541
raise errors.BadIndexOptions(self)
543
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
545
raise errors.BadIndexOptions(self)
546
options_line = stream.readline()
547
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
548
raise errors.BadIndexOptions(self)
550
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
552
raise errors.BadIndexOptions(self)
553
options_line = stream.readline()
554
if not options_line.startswith(_OPTION_LEN):
555
raise errors.BadIndexOptions(self)
557
self._key_count = int(options_line[len(_OPTION_LEN):-1])
559
raise errors.BadIndexOptions(self)
561
def _resolve_references(self, references):
562
"""Return the resolved key references for references.
564
References are resolved by looking up the location of the key in the
565
_keys_by_offset map and substituting the key name, preserving ordering.
567
:param references: An iterable of iterables of key locations. e.g.
569
:return: A tuple of tuples of keys.
572
for ref_list in references:
573
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
574
return tuple(node_refs)
576
def _find_index(self, range_map, key):
577
"""Helper for the _parsed_*_index calls.
579
Given a range map - [(start, end), ...], finds the index of the range
580
in the map for key if it is in the map, and if it is not there, the
581
immediately preceeding range in the map.
583
result = bisect_right(range_map, key) - 1
584
if result + 1 < len(range_map):
585
# check the border condition, it may be in result + 1
586
if range_map[result + 1][0] == key[0]:
590
def _parsed_byte_index(self, offset):
591
"""Return the index of the entry immediately before offset.
593
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
594
there is one unparsed byte (the 11th, addressed as[10]). then:
595
asking for 0 will return 0
596
asking for 10 will return 0
597
asking for 11 will return 1
598
asking for 12 will return 1
601
return self._find_index(self._parsed_byte_map, key)
603
def _parsed_key_index(self, key):
604
"""Return the index of the entry immediately before key.
606
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
607
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
608
have been parsed, then:
609
asking for '' will return 0
610
asking for 'a' will return 0
611
asking for 'b' will return 1
612
asking for 'e' will return 1
614
search_key = (key, None)
615
return self._find_index(self._parsed_key_map, search_key)
617
def _is_parsed(self, offset):
618
"""Returns True if offset has been parsed."""
619
index = self._parsed_byte_index(offset)
620
if index == len(self._parsed_byte_map):
621
return offset < self._parsed_byte_map[index - 1][1]
622
start, end = self._parsed_byte_map[index]
623
return offset >= start and offset < end
625
def _iter_entries_from_total_buffer(self, keys):
626
"""Iterate over keys when the entire index is parsed."""
627
# Note: See the note in BTreeBuilder.iter_entries for why we don't use
628
# .intersection() here
630
keys = [key for key in keys if key in nodes]
631
if self.node_ref_lists:
633
value, node_refs = nodes[key]
634
yield self, key, value, node_refs
637
yield self, key, nodes[key]
639
def iter_entries(self, keys):
640
"""Iterate over keys within the index.
642
:param keys: An iterable providing the keys to be retrieved.
643
:return: An iterable as per iter_all_entries, but restricted to the
644
keys supplied. No additional keys will be returned, and every
645
key supplied that is in the index will be returned.
650
if self._size is None and self._nodes is None:
653
# We fit about 20 keys per minimum-read (4K), so if we are looking for
654
# more than 1/20th of the index its likely (assuming homogenous key
655
# spread) that we'll read the entire index. If we're going to do that,
656
# buffer the whole thing. A better analysis might take key spread into
657
# account - but B+Tree indices are better anyway.
658
# We could look at all data read, and use a threshold there, which will
659
# trigger on ancestry walks, but that is not yet fully mapped out.
660
if self._nodes is None and len(keys) * 20 > self.key_count():
662
if self._nodes is not None:
663
return self._iter_entries_from_total_buffer(keys)
665
return (result[1] for result in bisect_multi_bytes(
666
self._lookup_keys_via_location, self._size, keys))
668
def iter_entries_prefix(self, keys):
669
"""Iterate over keys within the index using prefix matching.
671
Prefix matching is applied within the tuple of a key, not to within
672
the bytestring of each key element. e.g. if you have the keys ('foo',
673
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
674
only the former key is returned.
676
WARNING: Note that this method currently causes a full index parse
677
unconditionally (which is reasonably appropriate as it is a means for
678
thunking many small indices into one larger one and still supplies
679
iter_all_entries at the thunk layer).
681
:param keys: An iterable providing the key prefixes to be retrieved.
682
Each key prefix takes the form of a tuple the length of a key, but
683
with the last N elements 'None' rather than a regular bytestring.
684
The first element cannot be 'None'.
685
:return: An iterable as per iter_all_entries, but restricted to the
686
keys with a matching prefix to those supplied. No additional keys
687
will be returned, and every match that is in the index will be
693
# load data - also finds key lengths
694
if self._nodes is None:
696
if self._key_length == 1:
700
raise errors.BadIndexKey(key)
701
if len(key) != self._key_length:
702
raise errors.BadIndexKey(key)
703
if self.node_ref_lists:
704
value, node_refs = self._nodes[key]
705
yield self, key, value, node_refs
707
yield self, key, self._nodes[key]
709
nodes_by_key = self._get_nodes_by_key()
713
raise errors.BadIndexKey(key)
714
if len(key) != self._key_length:
715
raise errors.BadIndexKey(key)
716
# find what it refers to:
717
key_dict = nodes_by_key
719
# find the subdict whose contents should be returned.
721
while len(elements) and elements[0] is not None:
722
key_dict = key_dict[elements[0]]
725
# a non-existant lookup.
730
key_dict = dicts.pop(-1)
731
# can't be empty or would not exist
732
item, value = key_dict.iteritems().next()
733
if type(value) == dict:
735
dicts.extend(key_dict.itervalues())
738
for value in key_dict.itervalues():
739
# each value is the key:value:node refs tuple
741
yield (self, ) + value
743
# the last thing looked up was a terminal element
744
yield (self, ) + key_dict
746
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
747
"""See BTreeIndex._find_ancestors."""
748
# The api can be implemented as a trivial overlay on top of
749
# iter_entries, it is not an efficient implementation, but it at least
753
for index, key, value, refs in self.iter_entries(keys):
754
parent_keys = refs[ref_list_num]
756
parent_map[key] = parent_keys
757
search_keys.update(parent_keys)
758
# Figure out what, if anything, was missing
759
missing_keys.update(set(keys).difference(found_keys))
760
search_keys = search_keys.difference(parent_map)
764
"""Return an estimate of the number of keys in this index.
766
For GraphIndex the estimate is exact.
768
if self._key_count is None:
769
self._read_and_parse([_HEADER_READV])
770
return self._key_count
772
def _lookup_keys_via_location(self, location_keys):
773
"""Public interface for implementing bisection.
775
If _buffer_all has been called, then all the data for the index is in
776
memory, and this method should not be called, as it uses a separate
777
cache because it cannot pre-resolve all indices, which buffer_all does
780
:param location_keys: A list of location(byte offset), key tuples.
781
:return: A list of (location_key, result) tuples as expected by
782
bzrlib.bisect_multi.bisect_multi_bytes.
784
# Possible improvements:
785
# - only bisect lookup each key once
786
# - sort the keys first, and use that to reduce the bisection window
788
# this progresses in three parts:
791
# attempt to answer the question from the now in memory data.
792
# build the readv request
793
# for each location, ask for 800 bytes - much more than rows we've seen
796
for location, key in location_keys:
797
# can we answer from cache?
798
if self._bisect_nodes and key in self._bisect_nodes:
799
# We have the key parsed.
801
index = self._parsed_key_index(key)
802
if (len(self._parsed_key_map) and
803
self._parsed_key_map[index][0] <= key and
804
(self._parsed_key_map[index][1] >= key or
805
# end of the file has been parsed
806
self._parsed_byte_map[index][1] == self._size)):
807
# the key has been parsed, so no lookup is needed even if its
810
# - if we have examined this part of the file already - yes
811
index = self._parsed_byte_index(location)
812
if (len(self._parsed_byte_map) and
813
self._parsed_byte_map[index][0] <= location and
814
self._parsed_byte_map[index][1] > location):
815
# the byte region has been parsed, so no read is needed.
818
if location + length > self._size:
819
length = self._size - location
820
# todo, trim out parsed locations.
822
readv_ranges.append((location, length))
823
# read the header if needed
824
if self._bisect_nodes is None:
825
readv_ranges.append(_HEADER_READV)
826
self._read_and_parse(readv_ranges)
828
if self._nodes is not None:
829
# _read_and_parse triggered a _buffer_all because we requested the
831
for location, key in location_keys:
832
if key not in self._nodes: # not present
833
result.append(((location, key), False))
834
elif self.node_ref_lists:
835
value, refs = self._nodes[key]
836
result.append(((location, key),
837
(self, key, value, refs)))
839
result.append(((location, key),
840
(self, key, self._nodes[key])))
843
# - figure out <, >, missing, present
844
# - result present references so we can return them.
845
# keys that we cannot answer until we resolve references
846
pending_references = []
847
pending_locations = set()
848
for location, key in location_keys:
849
# can we answer from cache?
850
if key in self._bisect_nodes:
851
# the key has been parsed, so no lookup is needed
852
if self.node_ref_lists:
853
# the references may not have been all parsed.
854
value, refs = self._bisect_nodes[key]
855
wanted_locations = []
856
for ref_list in refs:
858
if ref not in self._keys_by_offset:
859
wanted_locations.append(ref)
861
pending_locations.update(wanted_locations)
862
pending_references.append((location, key))
864
result.append(((location, key), (self, key,
865
value, self._resolve_references(refs))))
867
result.append(((location, key),
868
(self, key, self._bisect_nodes[key])))
871
# has the region the key should be in, been parsed?
872
index = self._parsed_key_index(key)
873
if (self._parsed_key_map[index][0] <= key and
874
(self._parsed_key_map[index][1] >= key or
875
# end of the file has been parsed
876
self._parsed_byte_map[index][1] == self._size)):
877
result.append(((location, key), False))
879
# no, is the key above or below the probed location:
880
# get the range of the probed & parsed location
881
index = self._parsed_byte_index(location)
882
# if the key is below the start of the range, its below
883
if key < self._parsed_key_map[index][0]:
887
result.append(((location, key), direction))
889
# lookup data to resolve references
890
for location in pending_locations:
892
if location + length > self._size:
893
length = self._size - location
894
# TODO: trim out parsed locations (e.g. if the 800 is into the
895
# parsed region trim it, and dont use the adjust_for_latency
898
readv_ranges.append((location, length))
899
self._read_and_parse(readv_ranges)
900
if self._nodes is not None:
901
# The _read_and_parse triggered a _buffer_all, grab the data and
903
for location, key in pending_references:
904
value, refs = self._nodes[key]
905
result.append(((location, key), (self, key, value, refs)))
907
for location, key in pending_references:
908
# answer key references we had to look-up-late.
909
value, refs = self._bisect_nodes[key]
910
result.append(((location, key), (self, key,
911
value, self._resolve_references(refs))))
914
def _parse_header_from_bytes(self, bytes):
915
"""Parse the header from a region of bytes.
917
:param bytes: The data to parse.
918
:return: An offset, data tuple such as readv yields, for the unparsed
919
data. (which may length 0).
921
signature = bytes[0:len(self._signature())]
922
if not signature == self._signature():
923
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
924
lines = bytes[len(self._signature()):].splitlines()
925
options_line = lines[0]
926
if not options_line.startswith(_OPTION_NODE_REFS):
927
raise errors.BadIndexOptions(self)
929
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
931
raise errors.BadIndexOptions(self)
932
options_line = lines[1]
933
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
934
raise errors.BadIndexOptions(self)
936
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
938
raise errors.BadIndexOptions(self)
939
options_line = lines[2]
940
if not options_line.startswith(_OPTION_LEN):
941
raise errors.BadIndexOptions(self)
943
self._key_count = int(options_line[len(_OPTION_LEN):])
945
raise errors.BadIndexOptions(self)
946
# calculate the bytes we have processed
947
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
949
self._parsed_bytes(0, None, header_end, None)
950
# setup parsing state
951
self._expected_elements = 3 + self._key_length
952
# raw data keyed by offset
953
self._keys_by_offset = {}
954
# keys with the value and node references
955
self._bisect_nodes = {}
956
return header_end, bytes[header_end:]
958
def _parse_region(self, offset, data):
959
"""Parse node data returned from a readv operation.
961
:param offset: The byte offset the data starts at.
962
:param data: The data to parse.
966
end = offset + len(data)
969
# Trivial test - if the current index's end is within the
970
# low-matching parsed range, we're done.
971
index = self._parsed_byte_index(high_parsed)
972
if end < self._parsed_byte_map[index][1]:
974
# print "[%d:%d]" % (offset, end), \
975
# self._parsed_byte_map[index:index + 2]
976
high_parsed, last_segment = self._parse_segment(
977
offset, data, end, index)
981
def _parse_segment(self, offset, data, end, index):
982
"""Parse one segment of data.
984
:param offset: Where 'data' begins in the file.
985
:param data: Some data to parse a segment of.
986
:param end: Where data ends
987
:param index: The current index into the parsed bytes map.
988
:return: True if the parsed segment is the last possible one in the
990
:return: high_parsed_byte, last_segment.
991
high_parsed_byte is the location of the highest parsed byte in this
992
segment, last_segment is True if the parsed segment is the last
993
possible one in the data block.
995
# default is to use all data
997
# accomodate overlap with data before this.
998
if offset < self._parsed_byte_map[index][1]:
999
# overlaps the lower parsed region
1000
# skip the parsed data
1001
trim_start = self._parsed_byte_map[index][1] - offset
1002
# don't trim the start for \n
1003
start_adjacent = True
1004
elif offset == self._parsed_byte_map[index][1]:
1005
# abuts the lower parsed region
1008
# do not trim anything
1009
start_adjacent = True
1011
# does not overlap the lower parsed region
1014
# but trim the leading \n
1015
start_adjacent = False
1016
if end == self._size:
1017
# lines up to the end of all data:
1020
# do not strip to the last \n
1023
elif index + 1 == len(self._parsed_byte_map):
1024
# at the end of the parsed data
1027
# but strip to the last \n
1028
end_adjacent = False
1030
elif end == self._parsed_byte_map[index + 1][0]:
1031
# buts up against the next parsed region
1034
# do not strip to the last \n
1037
elif end > self._parsed_byte_map[index + 1][0]:
1038
# overlaps into the next parsed region
1039
# only consider the unparsed data
1040
trim_end = self._parsed_byte_map[index + 1][0] - offset
1041
# do not strip to the last \n as we know its an entire record
1043
last_segment = end < self._parsed_byte_map[index + 1][1]
1045
# does not overlap into the next region
1048
# but strip to the last \n
1049
end_adjacent = False
1051
# now find bytes to discard if needed
1052
if not start_adjacent:
1053
# work around python bug in rfind
1054
if trim_start is None:
1055
trim_start = data.find('\n') + 1
1057
trim_start = data.find('\n', trim_start) + 1
1058
if not (trim_start != 0):
1059
raise AssertionError('no \n was present')
1060
# print 'removing start', offset, trim_start, repr(data[:trim_start])
1061
if not end_adjacent:
1062
# work around python bug in rfind
1063
if trim_end is None:
1064
trim_end = data.rfind('\n') + 1
1066
trim_end = data.rfind('\n', None, trim_end) + 1
1067
if not (trim_end != 0):
1068
raise AssertionError('no \n was present')
1069
# print 'removing end', offset, trim_end, repr(data[trim_end:])
1070
# adjust offset and data to the parseable data.
1071
trimmed_data = data[trim_start:trim_end]
1072
if not (trimmed_data):
1073
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
1074
% (trim_start, trim_end, offset, offset + len(data)))
1076
offset += trim_start
1077
# print "parsing", repr(trimmed_data)
1078
# splitlines mangles the \r delimiters.. don't use it.
1079
lines = trimmed_data.split('\n')
1082
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1083
for key, value in nodes:
1084
self._bisect_nodes[key] = value
1085
self._parsed_bytes(offset, first_key,
1086
offset + len(trimmed_data), last_key)
1087
return offset + len(trimmed_data), last_segment
1089
def _parse_lines(self, lines, pos):
1096
# must be at the end
1098
if not (self._size == pos + 1):
1099
raise AssertionError("%s %s" % (self._size, pos))
1102
elements = line.split('\0')
1103
if len(elements) != self._expected_elements:
1104
raise errors.BadIndexData(self)
1105
# keys are tuples. Each element is a string that may occur many
1106
# times, so we intern them to save space. AB, RC, 200807
1107
key = tuple([intern(element) for element in elements[:self._key_length]])
1108
if first_key is None:
1110
absent, references, value = elements[-3:]
1112
for ref_string in references.split('\t'):
1113
ref_lists.append(tuple([
1114
int(ref) for ref in ref_string.split('\r') if ref
1116
ref_lists = tuple(ref_lists)
1117
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1118
pos += len(line) + 1 # +1 for the \n
1121
if self.node_ref_lists:
1122
node_value = (value, ref_lists)
1125
nodes.append((key, node_value))
1126
# print "parsed ", key
1127
return first_key, key, nodes, trailers
1129
def _parsed_bytes(self, start, start_key, end, end_key):
1130
"""Mark the bytes from start to end as parsed.
1132
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1135
:param start: The start of the parsed region.
1136
:param end: The end of the parsed region.
1138
index = self._parsed_byte_index(start)
1139
new_value = (start, end)
1140
new_key = (start_key, end_key)
1142
# first range parsed is always the beginning.
1143
self._parsed_byte_map.insert(index, new_value)
1144
self._parsed_key_map.insert(index, new_key)
1148
# extend lower region
1149
# extend higher region
1150
# combine two regions
1151
if (index + 1 < len(self._parsed_byte_map) and
1152
self._parsed_byte_map[index][1] == start and
1153
self._parsed_byte_map[index + 1][0] == end):
1154
# combine two regions
1155
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1156
self._parsed_byte_map[index + 1][1])
1157
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1158
self._parsed_key_map[index + 1][1])
1159
del self._parsed_byte_map[index + 1]
1160
del self._parsed_key_map[index + 1]
1161
elif self._parsed_byte_map[index][1] == start:
1162
# extend the lower entry
1163
self._parsed_byte_map[index] = (
1164
self._parsed_byte_map[index][0], end)
1165
self._parsed_key_map[index] = (
1166
self._parsed_key_map[index][0], end_key)
1167
elif (index + 1 < len(self._parsed_byte_map) and
1168
self._parsed_byte_map[index + 1][0] == end):
1169
# extend the higher entry
1170
self._parsed_byte_map[index + 1] = (
1171
start, self._parsed_byte_map[index + 1][1])
1172
self._parsed_key_map[index + 1] = (
1173
start_key, self._parsed_key_map[index + 1][1])
1176
self._parsed_byte_map.insert(index + 1, new_value)
1177
self._parsed_key_map.insert(index + 1, new_key)
1179
def _read_and_parse(self, readv_ranges):
1180
"""Read the ranges and parse the resulting data.
1182
:param readv_ranges: A prepared readv range list.
1184
if not readv_ranges:
1186
if self._nodes is None and self._bytes_read * 2 >= self._size:
1187
# We've already read more than 50% of the file and we are about to
1188
# request more data, just _buffer_all() and be done
1192
readv_data = self._transport.readv(self._name, readv_ranges, True,
1195
for offset, data in readv_data:
1196
self._bytes_read += len(data)
1197
if offset == 0 and len(data) == self._size:
1198
# We read the whole range, most likely because the
1199
# Transport upcast our readv ranges into one long request
1200
# for enough total data to grab the whole index.
1201
self._buffer_all(StringIO(data))
1203
if self._bisect_nodes is None:
1204
# this must be the start
1205
if not (offset == 0):
1206
raise AssertionError()
1207
offset, data = self._parse_header_from_bytes(data)
1208
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1209
self._parse_region(offset, data)
1211
def _signature(self):
1212
"""The file signature for this index type."""
1216
"""Validate that everything in the index can be accessed."""
1217
# iter_all validates completely at the moment, so just do that.
1218
for node in self.iter_all_entries():
1222
class CombinedGraphIndex(object):
1223
"""A GraphIndex made up from smaller GraphIndices.
1225
The backing indices must implement GraphIndex, and are presumed to be
1228
Queries against the combined index will be made against the first index,
1229
and then the second and so on. The order of index's can thus influence
1230
performance significantly. For example, if one index is on local disk and a
1231
second on a remote server, the local disk index should be before the other
1235
def __init__(self, indices, reload_func=None):
1236
"""Create a CombinedGraphIndex backed by indices.
1238
:param indices: An ordered list of indices to query for data.
1239
:param reload_func: A function to call if we find we are missing an
1240
index. Should have the form reload_func() => True/False to indicate
1241
if reloading actually changed anything.
1243
self._indices = indices
1244
self._reload_func = reload_func
1248
self.__class__.__name__,
1249
', '.join(map(repr, self._indices)))
1251
def clear_cache(self):
1252
"""See GraphIndex.clear_cache()"""
1253
for index in self._indices:
1256
def get_parent_map(self, keys):
1257
"""See graph.StackedParentsProvider.get_parent_map"""
1258
search_keys = set(keys)
1259
if NULL_REVISION in search_keys:
1260
search_keys.discard(NULL_REVISION)
1261
found_parents = {NULL_REVISION:[]}
1264
for index, key, value, refs in self.iter_entries(search_keys):
1267
parents = (NULL_REVISION,)
1268
found_parents[key] = parents
1269
return found_parents
1271
has_key = _has_key_from_parent_map
1273
def insert_index(self, pos, index):
1274
"""Insert a new index in the list of indices to query.
1276
:param pos: The position to insert the index.
1277
:param index: The index to insert.
1279
self._indices.insert(pos, index)
1281
def iter_all_entries(self):
1282
"""Iterate over all keys within the index
1284
Duplicate keys across child indices are presumed to have the same
1285
value and are only reported once.
1287
:return: An iterable of (index, key, reference_lists, value).
1288
There is no defined order for the result iteration - it will be in
1289
the most efficient order for the index.
1294
for index in self._indices:
1295
for node in index.iter_all_entries():
1296
if node[1] not in seen_keys:
1298
seen_keys.add(node[1])
1300
except errors.NoSuchFile:
1301
self._reload_or_raise()
1303
def iter_entries(self, keys):
1304
"""Iterate over keys within the index.
1306
Duplicate keys across child indices are presumed to have the same
1307
value and are only reported once.
1309
:param keys: An iterable providing the keys to be retrieved.
1310
:return: An iterable of (index, key, reference_lists, value). There is no
1311
defined order for the result iteration - it will be in the most
1312
efficient order for the index.
1317
for index in self._indices:
1320
for node in index.iter_entries(keys):
1321
keys.remove(node[1])
1324
except errors.NoSuchFile:
1325
self._reload_or_raise()
1327
def iter_entries_prefix(self, keys):
1328
"""Iterate over keys within the index using prefix matching.
1330
Duplicate keys across child indices are presumed to have the same
1331
value and are only reported once.
1333
Prefix matching is applied within the tuple of a key, not to within
1334
the bytestring of each key element. e.g. if you have the keys ('foo',
1335
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1336
only the former key is returned.
1338
:param keys: An iterable providing the key prefixes to be retrieved.
1339
Each key prefix takes the form of a tuple the length of a key, but
1340
with the last N elements 'None' rather than a regular bytestring.
1341
The first element cannot be 'None'.
1342
:return: An iterable as per iter_all_entries, but restricted to the
1343
keys with a matching prefix to those supplied. No additional keys
1344
will be returned, and every match that is in the index will be
1353
for index in self._indices:
1354
for node in index.iter_entries_prefix(keys):
1355
if node[1] in seen_keys:
1357
seen_keys.add(node[1])
1360
except errors.NoSuchFile:
1361
self._reload_or_raise()
1363
def find_ancestry(self, keys, ref_list_num):
1364
"""Find the complete ancestry for the given set of keys.
1366
Note that this is a whole-ancestry request, so it should be used
1369
:param keys: An iterable of keys to look for
1370
:param ref_list_num: The reference list which references the parents
1372
:return: (parent_map, missing_keys)
1374
missing_keys = set()
1376
keys_to_lookup = set(keys)
1378
while keys_to_lookup:
1379
# keys that *all* indexes claim are missing, stop searching them
1381
all_index_missing = None
1382
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1383
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1385
# len(missing_keys))
1386
for index_idx, index in enumerate(self._indices):
1387
# TODO: we should probably be doing something with
1388
# 'missing_keys' since we've already determined that
1389
# those revisions have not been found anywhere
1390
index_missing_keys = set()
1391
# Find all of the ancestry we can from this index
1392
# keep looking until the search_keys set is empty, which means
1393
# things we didn't find should be in index_missing_keys
1394
search_keys = keys_to_lookup
1396
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1397
# index_idx, len(search_keys),
1398
# len(parent_map), len(index_missing_keys))
1401
# TODO: ref_list_num should really be a parameter, since
1402
# CombinedGraphIndex does not know what the ref lists
1404
search_keys = index._find_ancestors(search_keys,
1405
ref_list_num, parent_map, index_missing_keys)
1406
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1407
# sub_generation, len(search_keys),
1408
# len(parent_map), len(index_missing_keys))
1409
# Now set whatever was missing to be searched in the next index
1410
keys_to_lookup = index_missing_keys
1411
if all_index_missing is None:
1412
all_index_missing = set(index_missing_keys)
1414
all_index_missing.intersection_update(index_missing_keys)
1415
if not keys_to_lookup:
1417
if all_index_missing is None:
1418
# There were no indexes, so all search keys are 'missing'
1419
missing_keys.update(keys_to_lookup)
1420
keys_to_lookup = None
1422
missing_keys.update(all_index_missing)
1423
keys_to_lookup.difference_update(all_index_missing)
1424
return parent_map, missing_keys
1426
def key_count(self):
1427
"""Return an estimate of the number of keys in this index.
1429
For CombinedGraphIndex this is approximated by the sum of the keys of
1430
the child indices. As child indices may have duplicate keys this can
1431
have a maximum error of the number of child indices * largest number of
1436
return sum((index.key_count() for index in self._indices), 0)
1437
except errors.NoSuchFile:
1438
self._reload_or_raise()
1440
missing_keys = _missing_keys_from_parent_map
1442
def _reload_or_raise(self):
1443
"""We just got a NoSuchFile exception.
1445
Try to reload the indices, if it fails, just raise the current
1448
if self._reload_func is None:
1450
exc_type, exc_value, exc_traceback = sys.exc_info()
1451
trace.mutter('Trying to reload after getting exception: %s',
1453
if not self._reload_func():
1454
# We tried to reload, but nothing changed, so we fail anyway
1455
trace.mutter('_reload_func indicated nothing has changed.'
1456
' Raising original exception.')
1457
raise exc_type, exc_value, exc_traceback
1460
"""Validate that everything in the index can be accessed."""
1463
for index in self._indices:
1466
except errors.NoSuchFile:
1467
self._reload_or_raise()
1470
class InMemoryGraphIndex(GraphIndexBuilder):
1471
"""A GraphIndex which operates entirely out of memory and is mutable.
1473
This is designed to allow the accumulation of GraphIndex entries during a
1474
single write operation, where the accumulated entries need to be immediately
1475
available - for example via a CombinedGraphIndex.
1478
def add_nodes(self, nodes):
1479
"""Add nodes to the index.
1481
:param nodes: An iterable of (key, node_refs, value) entries to add.
1483
if self.reference_lists:
1484
for (key, value, node_refs) in nodes:
1485
self.add_node(key, value, node_refs)
1487
for (key, value) in nodes:
1488
self.add_node(key, value)
1490
def iter_all_entries(self):
1491
"""Iterate over all keys within the index
1493
:return: An iterable of (index, key, reference_lists, value). There is no
1494
defined order for the result iteration - it will be in the most
1495
efficient order for the index (in this case dictionary hash order).
1497
if 'evil' in debug.debug_flags:
1498
trace.mutter_callsite(3,
1499
"iter_all_entries scales with size of history.")
1500
if self.reference_lists:
1501
for key, (absent, references, value) in self._nodes.iteritems():
1503
yield self, key, value, references
1505
for key, (absent, references, value) in self._nodes.iteritems():
1507
yield self, key, value
1509
def iter_entries(self, keys):
1510
"""Iterate over keys within the index.
1512
:param keys: An iterable providing the keys to be retrieved.
1513
:return: An iterable of (index, key, value, reference_lists). There is no
1514
defined order for the result iteration - it will be in the most
1515
efficient order for the index (keys iteration order in this case).
1517
# Note: See BTreeBuilder.iter_entries for an explanation of why we
1518
# aren't using set().intersection() here
1520
keys = [key for key in keys if key in nodes]
1521
if self.reference_lists:
1525
yield self, key, node[2], node[1]
1530
yield self, key, node[2]
1532
def iter_entries_prefix(self, keys):
1533
"""Iterate over keys within the index using prefix matching.
1535
Prefix matching is applied within the tuple of a key, not to within
1536
the bytestring of each key element. e.g. if you have the keys ('foo',
1537
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1538
only the former key is returned.
1540
:param keys: An iterable providing the key prefixes to be retrieved.
1541
Each key prefix takes the form of a tuple the length of a key, but
1542
with the last N elements 'None' rather than a regular bytestring.
1543
The first element cannot be 'None'.
1544
:return: An iterable as per iter_all_entries, but restricted to the
1545
keys with a matching prefix to those supplied. No additional keys
1546
will be returned, and every match that is in the index will be
1549
# XXX: To much duplication with the GraphIndex class; consider finding
1550
# a good place to pull out the actual common logic.
1554
if self._key_length == 1:
1558
raise errors.BadIndexKey(key)
1559
if len(key) != self._key_length:
1560
raise errors.BadIndexKey(key)
1561
node = self._nodes[key]
1564
if self.reference_lists:
1565
yield self, key, node[2], node[1]
1567
yield self, key, node[2]
1569
nodes_by_key = self._get_nodes_by_key()
1573
raise errors.BadIndexKey(key)
1574
if len(key) != self._key_length:
1575
raise errors.BadIndexKey(key)
1576
# find what it refers to:
1577
key_dict = nodes_by_key
1578
elements = list(key)
1579
# find the subdict to return
1581
while len(elements) and elements[0] is not None:
1582
key_dict = key_dict[elements[0]]
1585
# a non-existant lookup.
1590
key_dict = dicts.pop(-1)
1591
# can't be empty or would not exist
1592
item, value = key_dict.iteritems().next()
1593
if type(value) == dict:
1595
dicts.extend(key_dict.itervalues())
1598
for value in key_dict.itervalues():
1599
yield (self, ) + value
1601
yield (self, ) + key_dict
1603
def key_count(self):
1604
"""Return an estimate of the number of keys in this index.
1606
For InMemoryGraphIndex the estimate is exact.
1608
return len(self._nodes) - len(self._absent_keys)
1611
"""In memory index's have no known corruption at the moment."""
1614
class GraphIndexPrefixAdapter(object):
1615
"""An adapter between GraphIndex with different key lengths.
1617
Queries against this will emit queries against the adapted Graph with the
1618
prefix added, queries for all items use iter_entries_prefix. The returned
1619
nodes will have their keys and node references adjusted to remove the
1620
prefix. Finally, an add_nodes_callback can be supplied - when called the
1621
nodes and references being added will have prefix prepended.
1624
def __init__(self, adapted, prefix, missing_key_length,
1625
add_nodes_callback=None):
1626
"""Construct an adapter against adapted with prefix."""
1627
self.adapted = adapted
1628
self.prefix_key = prefix + (None,)*missing_key_length
1629
self.prefix = prefix
1630
self.prefix_len = len(prefix)
1631
self.add_nodes_callback = add_nodes_callback
1633
def add_nodes(self, nodes):
1634
"""Add nodes to the index.
1636
:param nodes: An iterable of (key, node_refs, value) entries to add.
1638
# save nodes in case its an iterator
1639
nodes = tuple(nodes)
1640
translated_nodes = []
1642
# Add prefix_key to each reference node_refs is a tuple of tuples,
1643
# so split it apart, and add prefix_key to the internal reference
1644
for (key, value, node_refs) in nodes:
1645
adjusted_references = (
1646
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1647
for ref_list in node_refs))
1648
translated_nodes.append((self.prefix + key, value,
1649
adjusted_references))
1651
# XXX: TODO add an explicit interface for getting the reference list
1652
# status, to handle this bit of user-friendliness in the API more
1654
for (key, value) in nodes:
1655
translated_nodes.append((self.prefix + key, value))
1656
self.add_nodes_callback(translated_nodes)
1658
def add_node(self, key, value, references=()):
1659
"""Add a node to the index.
1661
:param key: The key. keys are non-empty tuples containing
1662
as many whitespace-free utf8 bytestrings as the key length
1663
defined for this index.
1664
:param references: An iterable of iterables of keys. Each is a
1665
reference to another key.
1666
:param value: The value to associate with the key. It may be any
1667
bytes as long as it does not contain \0 or \n.
1669
self.add_nodes(((key, value, references), ))
1671
def _strip_prefix(self, an_iter):
1672
"""Strip prefix data from nodes and return it."""
1673
for node in an_iter:
1675
if node[1][:self.prefix_len] != self.prefix:
1676
raise errors.BadIndexData(self)
1677
for ref_list in node[3]:
1678
for ref_node in ref_list:
1679
if ref_node[:self.prefix_len] != self.prefix:
1680
raise errors.BadIndexData(self)
1681
yield node[0], node[1][self.prefix_len:], node[2], (
1682
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1683
for ref_list in node[3]))
1685
def iter_all_entries(self):
1686
"""Iterate over all keys within the index
1688
iter_all_entries is implemented against the adapted index using
1689
iter_entries_prefix.
1691
:return: An iterable of (index, key, reference_lists, value). There is no
1692
defined order for the result iteration - it will be in the most
1693
efficient order for the index (in this case dictionary hash order).
1695
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1697
def iter_entries(self, keys):
1698
"""Iterate over keys within the index.
1700
:param keys: An iterable providing the keys to be retrieved.
1701
:return: An iterable of (index, key, value, reference_lists). There is no
1702
defined order for the result iteration - it will be in the most
1703
efficient order for the index (keys iteration order in this case).
1705
return self._strip_prefix(self.adapted.iter_entries(
1706
self.prefix + key for key in keys))
1708
def iter_entries_prefix(self, keys):
1709
"""Iterate over keys within the index using prefix matching.
1711
Prefix matching is applied within the tuple of a key, not to within
1712
the bytestring of each key element. e.g. if you have the keys ('foo',
1713
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1714
only the former key is returned.
1716
:param keys: An iterable providing the key prefixes to be retrieved.
1717
Each key prefix takes the form of a tuple the length of a key, but
1718
with the last N elements 'None' rather than a regular bytestring.
1719
The first element cannot be 'None'.
1720
:return: An iterable as per iter_all_entries, but restricted to the
1721
keys with a matching prefix to those supplied. No additional keys
1722
will be returned, and every match that is in the index will be
1725
return self._strip_prefix(self.adapted.iter_entries_prefix(
1726
self.prefix + key for key in keys))
1728
def key_count(self):
1729
"""Return an estimate of the number of keys in this index.
1731
For GraphIndexPrefixAdapter this is relatively expensive - key
1732
iteration with the prefix is done.
1734
return len(list(self.iter_all_entries()))
1737
"""Call the adapted's validate."""
1738
self.adapted.validate()
added
removed
bzrlib/index.py
