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
node_refs.append(as_st(reference_list))
209
return as_st(node_refs), absent_references
211
def add_node(self, key, value, references=()):
212
"""Add a node to the index.
214
:param key: The key. keys are non-empty tuples containing
215
as many whitespace-free utf8 bytestrings as the key length
216
defined for this index.
217
:param references: An iterable of iterables of keys. Each is a
218
reference to another key.
219
:param value: The value to associate with the key. It may be any
220
bytes as long as it does not contain \0 or \n.
223
absent_references) = self._check_key_ref_value(key, references, value)
224
if key in self._nodes and self._nodes[key][0] != 'a':
225
raise errors.BadIndexDuplicateKey(key, self)
226
for reference in absent_references:
227
# There may be duplicates, but I don't think it is worth worrying
229
self._nodes[reference] = ('a', (), '')
230
self._absent_keys.update(absent_references)
231
self._absent_keys.discard(key)
232
self._nodes[key] = ('', node_refs, value)
233
if self._nodes_by_key is not None and self._key_length > 1:
234
self._update_nodes_by_key(key, value, node_refs)
236
def clear_cache(self):
237
"""See GraphIndex.clear_cache()
239
This is a no-op, but we need the api to conform to a generic 'Index'
245
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
246
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
247
key_count = len(self._nodes) - len(self._absent_keys)
248
lines.append(_OPTION_LEN + str(key_count) + '\n')
249
prefix_length = sum(len(x) for x in lines)
250
# references are byte offsets. To avoid having to do nasty
251
# polynomial work to resolve offsets (references to later in the
252
# file cannot be determined until all the inbetween references have
253
# been calculated too) we pad the offsets with 0's to make them be
254
# of consistent length. Using binary offsets would break the trivial
256
# to calculate the width of zero's needed we do three passes:
257
# one to gather all the non-reference data and the number of references.
258
# one to pad all the data with reference-length and determine entry
262
# forward sorted by key. In future we may consider topological sorting,
263
# at the cost of table scans for direct lookup, or a second index for
265
nodes = sorted(self._nodes.items())
266
# if we do not prepass, we don't know how long it will be up front.
267
expected_bytes = None
268
# we only need to pre-pass if we have reference lists at all.
269
if self.reference_lists:
271
non_ref_bytes = prefix_length
273
# TODO use simple multiplication for the constants in this loop.
274
for key, (absent, references, value) in nodes:
275
# record the offset known *so far* for this key:
276
# the non reference bytes to date, and the total references to
277
# date - saves reaccumulating on the second pass
278
key_offset_info.append((key, non_ref_bytes, total_references))
279
# key is literal, value is literal, there are 3 null's, 1 NL
280
# key is variable length tuple, \x00 between elements
281
non_ref_bytes += sum(len(element) for element in key)
282
if self._key_length > 1:
283
non_ref_bytes += self._key_length - 1
284
# value is literal bytes, there are 3 null's, 1 NL.
285
non_ref_bytes += len(value) + 3 + 1
286
# one byte for absent if set.
289
elif self.reference_lists:
290
# (ref_lists -1) tabs
291
non_ref_bytes += self.reference_lists - 1
292
# (ref-1 cr's per ref_list)
293
for ref_list in references:
294
# how many references across the whole file?
295
total_references += len(ref_list)
296
# accrue reference separators
298
non_ref_bytes += len(ref_list) - 1
299
# how many digits are needed to represent the total byte count?
301
possible_total_bytes = non_ref_bytes + total_references*digits
302
while 10 ** digits < possible_total_bytes:
304
possible_total_bytes = non_ref_bytes + total_references*digits
305
expected_bytes = possible_total_bytes + 1 # terminating newline
306
# resolve key addresses.
308
for key, non_ref_bytes, total_references in key_offset_info:
309
key_addresses[key] = non_ref_bytes + total_references*digits
311
format_string = '%%0%sd' % digits
312
for key, (absent, references, value) in nodes:
313
flattened_references = []
314
for ref_list in references:
316
for reference in ref_list:
317
ref_addresses.append(format_string % key_addresses[reference])
318
flattened_references.append('\r'.join(ref_addresses))
319
string_key = '\x00'.join(key)
320
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
321
'\t'.join(flattened_references), value))
323
result = StringIO(''.join(lines))
324
if expected_bytes and len(result.getvalue()) != expected_bytes:
325
raise errors.BzrError('Failed index creation. Internal error:'
326
' mismatched output length and expected length: %d %d' %
327
(len(result.getvalue()), expected_bytes))
330
def set_optimize(self, for_size=None, combine_backing_indices=None):
331
"""Change how the builder tries to optimize the result.
333
:param for_size: Tell the builder to try and make the index as small as
335
:param combine_backing_indices: If the builder spills to disk to save
336
memory, should the on-disk indices be combined. Set to True if you
337
are going to be probing the index, but to False if you are not. (If
338
you are not querying, then the time spent combining is wasted.)
341
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
343
if for_size is not None:
344
self._optimize_for_size = for_size
345
if combine_backing_indices is not None:
346
self._combine_backing_indices = combine_backing_indices
348
def find_ancestry(self, keys, ref_list_num):
349
"""See CombinedGraphIndex.find_ancestry()"""
355
for _, key, value, ref_lists in self.iter_entries(pending):
356
parent_keys = ref_lists[ref_list_num]
357
parent_map[key] = parent_keys
358
next_pending.update([p for p in parent_keys if p not in
360
missing_keys.update(pending.difference(parent_map))
361
pending = next_pending
362
return parent_map, missing_keys
365
class GraphIndex(object):
366
"""An index for data with embedded graphs.
368
The index maps keys to a list of key reference lists, and a value.
369
Each node has the same number of key reference lists. Each key reference
370
list can be empty or an arbitrary length. The value is an opaque NULL
371
terminated string without any newlines. The storage of the index is
372
hidden in the interface: keys and key references are always tuples of
373
bytestrings, never the internal representation (e.g. dictionary offsets).
375
It is presumed that the index will not be mutated - it is static data.
377
Successive iter_all_entries calls will read the entire index each time.
378
Additionally, iter_entries calls will read the index linearly until the
379
desired keys are found. XXX: This must be fixed before the index is
380
suitable for production use. :XXX
383
def __init__(self, transport, name, size, unlimited_cache=False):
384
"""Open an index called name on transport.
386
:param transport: A bzrlib.transport.Transport.
387
:param name: A path to provide to transport API calls.
388
:param size: The size of the index in bytes. This is used for bisection
389
logic to perform partial index reads. While the size could be
390
obtained by statting the file this introduced an additional round
391
trip as well as requiring stat'able transports, both of which are
392
avoided by having it supplied. If size is None, then bisection
393
support will be disabled and accessing the index will just stream
396
self._transport = transport
398
# Becomes a dict of key:(value, reference-list-byte-locations) used by
399
# the bisection interface to store parsed but not resolved keys.
400
self._bisect_nodes = None
401
# Becomes a dict of key:(value, reference-list-keys) which are ready to
402
# be returned directly to callers.
404
# a sorted list of slice-addresses for the parsed bytes of the file.
405
# e.g. (0,1) would mean that byte 0 is parsed.
406
self._parsed_byte_map = []
407
# a sorted list of keys matching each slice address for parsed bytes
408
# e.g. (None, 'foo@bar') would mean that the first byte contained no
409
# key, and the end byte of the slice is the of the data for 'foo@bar'
410
self._parsed_key_map = []
411
self._key_count = None
412
self._keys_by_offset = None
413
self._nodes_by_key = None
415
# The number of bytes we've read so far in trying to process this file
418
def __eq__(self, other):
419
"""Equal when self and other were created with the same parameters."""
421
type(self) == type(other) and
422
self._transport == other._transport and
423
self._name == other._name and
424
self._size == other._size)
426
def __ne__(self, other):
427
return not self.__eq__(other)
430
return "%s(%r)" % (self.__class__.__name__,
431
self._transport.abspath(self._name))
433
def _buffer_all(self, stream=None):
434
"""Buffer all the index data.
436
Mutates self._nodes and self.keys_by_offset.
438
if self._nodes is not None:
439
# We already did this
441
if 'index' in debug.debug_flags:
442
mutter('Reading entire index %s', self._transport.abspath(self._name))
444
stream = self._transport.get(self._name)
445
self._read_prefix(stream)
446
self._expected_elements = 3 + self._key_length
448
# raw data keyed by offset
449
self._keys_by_offset = {}
450
# ready-to-return key:value or key:value, node_ref_lists
452
self._nodes_by_key = None
455
lines = stream.read().split('\n')
457
_, _, _, trailers = self._parse_lines(lines, pos)
458
for key, absent, references, value in self._keys_by_offset.itervalues():
461
# resolve references:
462
if self.node_ref_lists:
463
node_value = (value, self._resolve_references(references))
466
self._nodes[key] = node_value
467
# cache the keys for quick set intersections
469
# there must be one line - the empty trailer line.
470
raise errors.BadIndexData(self)
472
def clear_cache(self):
473
"""Clear out any cached/memoized values.
475
This can be called at any time, but generally it is used when we have
476
extracted some information, but don't expect to be requesting any more
480
def external_references(self, ref_list_num):
481
"""Return references that are not present in this index.
484
if ref_list_num + 1 > self.node_ref_lists:
485
raise ValueError('No ref list %d, index has %d ref lists'
486
% (ref_list_num, self.node_ref_lists))
489
for key, (value, ref_lists) in nodes.iteritems():
490
ref_list = ref_lists[ref_list_num]
491
refs.update([ref for ref in ref_list if ref not in nodes])
494
def _get_nodes_by_key(self):
495
if self._nodes_by_key is None:
497
if self.node_ref_lists:
498
for key, (value, references) in self._nodes.iteritems():
499
key_dict = nodes_by_key
500
for subkey in key[:-1]:
501
key_dict = key_dict.setdefault(subkey, {})
502
key_dict[key[-1]] = key, value, references
504
for key, value in self._nodes.iteritems():
505
key_dict = nodes_by_key
506
for subkey in key[:-1]:
507
key_dict = key_dict.setdefault(subkey, {})
508
key_dict[key[-1]] = key, value
509
self._nodes_by_key = nodes_by_key
510
return self._nodes_by_key
512
def iter_all_entries(self):
513
"""Iterate over all keys within the index.
515
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
516
The former tuple is used when there are no reference lists in the
517
index, making the API compatible with simple key:value index types.
518
There is no defined order for the result iteration - it will be in
519
the most efficient order for the index.
521
if 'evil' in debug.debug_flags:
522
trace.mutter_callsite(3,
523
"iter_all_entries scales with size of history.")
524
if self._nodes is None:
526
if self.node_ref_lists:
527
for key, (value, node_ref_lists) in self._nodes.iteritems():
528
yield self, key, value, node_ref_lists
530
for key, value in self._nodes.iteritems():
531
yield self, key, value
533
def _read_prefix(self, stream):
534
signature = stream.read(len(self._signature()))
535
if not signature == self._signature():
536
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
537
options_line = stream.readline()
538
if not options_line.startswith(_OPTION_NODE_REFS):
539
raise errors.BadIndexOptions(self)
541
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
543
raise errors.BadIndexOptions(self)
544
options_line = stream.readline()
545
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
546
raise errors.BadIndexOptions(self)
548
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
550
raise errors.BadIndexOptions(self)
551
options_line = stream.readline()
552
if not options_line.startswith(_OPTION_LEN):
553
raise errors.BadIndexOptions(self)
555
self._key_count = int(options_line[len(_OPTION_LEN):-1])
557
raise errors.BadIndexOptions(self)
559
def _resolve_references(self, references):
560
"""Return the resolved key references for references.
562
References are resolved by looking up the location of the key in the
563
_keys_by_offset map and substituting the key name, preserving ordering.
565
:param references: An iterable of iterables of key locations. e.g.
567
:return: A tuple of tuples of keys.
570
for ref_list in references:
571
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
572
return tuple(node_refs)
574
def _find_index(self, range_map, key):
575
"""Helper for the _parsed_*_index calls.
577
Given a range map - [(start, end), ...], finds the index of the range
578
in the map for key if it is in the map, and if it is not there, the
579
immediately preceeding range in the map.
581
result = bisect_right(range_map, key) - 1
582
if result + 1 < len(range_map):
583
# check the border condition, it may be in result + 1
584
if range_map[result + 1][0] == key[0]:
588
def _parsed_byte_index(self, offset):
589
"""Return the index of the entry immediately before offset.
591
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
592
there is one unparsed byte (the 11th, addressed as[10]). then:
593
asking for 0 will return 0
594
asking for 10 will return 0
595
asking for 11 will return 1
596
asking for 12 will return 1
599
return self._find_index(self._parsed_byte_map, key)
601
def _parsed_key_index(self, key):
602
"""Return the index of the entry immediately before key.
604
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
605
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
606
have been parsed, then:
607
asking for '' will return 0
608
asking for 'a' will return 0
609
asking for 'b' will return 1
610
asking for 'e' will return 1
612
search_key = (key, None)
613
return self._find_index(self._parsed_key_map, search_key)
615
def _is_parsed(self, offset):
616
"""Returns True if offset has been parsed."""
617
index = self._parsed_byte_index(offset)
618
if index == len(self._parsed_byte_map):
619
return offset < self._parsed_byte_map[index - 1][1]
620
start, end = self._parsed_byte_map[index]
621
return offset >= start and offset < end
623
def _iter_entries_from_total_buffer(self, keys):
624
"""Iterate over keys when the entire index is parsed."""
625
# Note: See the note in BTreeBuilder.iter_entries for why we don't use
626
# .intersection() here
628
keys = [key for key in keys if key in nodes]
629
if self.node_ref_lists:
631
value, node_refs = nodes[key]
632
yield self, key, value, node_refs
635
yield self, key, nodes[key]
637
def iter_entries(self, keys):
638
"""Iterate over keys within the index.
640
:param keys: An iterable providing the keys to be retrieved.
641
:return: An iterable as per iter_all_entries, but restricted to the
642
keys supplied. No additional keys will be returned, and every
643
key supplied that is in the index will be returned.
648
if self._size is None and self._nodes is None:
651
# We fit about 20 keys per minimum-read (4K), so if we are looking for
652
# more than 1/20th of the index its likely (assuming homogenous key
653
# spread) that we'll read the entire index. If we're going to do that,
654
# buffer the whole thing. A better analysis might take key spread into
655
# account - but B+Tree indices are better anyway.
656
# We could look at all data read, and use a threshold there, which will
657
# trigger on ancestry walks, but that is not yet fully mapped out.
658
if self._nodes is None and len(keys) * 20 > self.key_count():
660
if self._nodes is not None:
661
return self._iter_entries_from_total_buffer(keys)
663
return (result[1] for result in bisect_multi_bytes(
664
self._lookup_keys_via_location, self._size, keys))
666
def iter_entries_prefix(self, keys):
667
"""Iterate over keys within the index using prefix matching.
669
Prefix matching is applied within the tuple of a key, not to within
670
the bytestring of each key element. e.g. if you have the keys ('foo',
671
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
672
only the former key is returned.
674
WARNING: Note that this method currently causes a full index parse
675
unconditionally (which is reasonably appropriate as it is a means for
676
thunking many small indices into one larger one and still supplies
677
iter_all_entries at the thunk layer).
679
:param keys: An iterable providing the key prefixes to be retrieved.
680
Each key prefix takes the form of a tuple the length of a key, but
681
with the last N elements 'None' rather than a regular bytestring.
682
The first element cannot be 'None'.
683
:return: An iterable as per iter_all_entries, but restricted to the
684
keys with a matching prefix to those supplied. No additional keys
685
will be returned, and every match that is in the index will be
691
# load data - also finds key lengths
692
if self._nodes is None:
694
if self._key_length == 1:
698
raise errors.BadIndexKey(key)
699
if len(key) != self._key_length:
700
raise errors.BadIndexKey(key)
701
if self.node_ref_lists:
702
value, node_refs = self._nodes[key]
703
yield self, key, value, node_refs
705
yield self, key, self._nodes[key]
707
nodes_by_key = self._get_nodes_by_key()
711
raise errors.BadIndexKey(key)
712
if len(key) != self._key_length:
713
raise errors.BadIndexKey(key)
714
# find what it refers to:
715
key_dict = nodes_by_key
717
# find the subdict whose contents should be returned.
719
while len(elements) and elements[0] is not None:
720
key_dict = key_dict[elements[0]]
723
# a non-existant lookup.
728
key_dict = dicts.pop(-1)
729
# can't be empty or would not exist
730
item, value = key_dict.iteritems().next()
731
if type(value) == dict:
733
dicts.extend(key_dict.itervalues())
736
for value in key_dict.itervalues():
737
# each value is the key:value:node refs tuple
739
yield (self, ) + value
741
# the last thing looked up was a terminal element
742
yield (self, ) + key_dict
744
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
745
"""See BTreeIndex._find_ancestors."""
746
# The api can be implemented as a trivial overlay on top of
747
# iter_entries, it is not an efficient implementation, but it at least
751
for index, key, value, refs in self.iter_entries(keys):
752
parent_keys = refs[ref_list_num]
754
parent_map[key] = parent_keys
755
search_keys.update(parent_keys)
756
# Figure out what, if anything, was missing
757
missing_keys.update(set(keys).difference(found_keys))
758
search_keys = search_keys.difference(parent_map)
762
"""Return an estimate of the number of keys in this index.
764
For GraphIndex the estimate is exact.
766
if self._key_count is None:
767
self._read_and_parse([_HEADER_READV])
768
return self._key_count
770
def _lookup_keys_via_location(self, location_keys):
771
"""Public interface for implementing bisection.
773
If _buffer_all has been called, then all the data for the index is in
774
memory, and this method should not be called, as it uses a separate
775
cache because it cannot pre-resolve all indices, which buffer_all does
778
:param location_keys: A list of location(byte offset), key tuples.
779
:return: A list of (location_key, result) tuples as expected by
780
bzrlib.bisect_multi.bisect_multi_bytes.
782
# Possible improvements:
783
# - only bisect lookup each key once
784
# - sort the keys first, and use that to reduce the bisection window
786
# this progresses in three parts:
789
# attempt to answer the question from the now in memory data.
790
# build the readv request
791
# for each location, ask for 800 bytes - much more than rows we've seen
794
for location, key in location_keys:
795
# can we answer from cache?
796
if self._bisect_nodes and key in self._bisect_nodes:
797
# We have the key parsed.
799
index = self._parsed_key_index(key)
800
if (len(self._parsed_key_map) and
801
self._parsed_key_map[index][0] <= key and
802
(self._parsed_key_map[index][1] >= key or
803
# end of the file has been parsed
804
self._parsed_byte_map[index][1] == self._size)):
805
# the key has been parsed, so no lookup is needed even if its
808
# - if we have examined this part of the file already - yes
809
index = self._parsed_byte_index(location)
810
if (len(self._parsed_byte_map) and
811
self._parsed_byte_map[index][0] <= location and
812
self._parsed_byte_map[index][1] > location):
813
# the byte region has been parsed, so no read is needed.
816
if location + length > self._size:
817
length = self._size - location
818
# todo, trim out parsed locations.
820
readv_ranges.append((location, length))
821
# read the header if needed
822
if self._bisect_nodes is None:
823
readv_ranges.append(_HEADER_READV)
824
self._read_and_parse(readv_ranges)
826
if self._nodes is not None:
827
# _read_and_parse triggered a _buffer_all because we requested the
829
for location, key in location_keys:
830
if key not in self._nodes: # not present
831
result.append(((location, key), False))
832
elif self.node_ref_lists:
833
value, refs = self._nodes[key]
834
result.append(((location, key),
835
(self, key, value, refs)))
837
result.append(((location, key),
838
(self, key, self._nodes[key])))
841
# - figure out <, >, missing, present
842
# - result present references so we can return them.
843
# keys that we cannot answer until we resolve references
844
pending_references = []
845
pending_locations = set()
846
for location, key in location_keys:
847
# can we answer from cache?
848
if key in self._bisect_nodes:
849
# the key has been parsed, so no lookup is needed
850
if self.node_ref_lists:
851
# the references may not have been all parsed.
852
value, refs = self._bisect_nodes[key]
853
wanted_locations = []
854
for ref_list in refs:
856
if ref not in self._keys_by_offset:
857
wanted_locations.append(ref)
859
pending_locations.update(wanted_locations)
860
pending_references.append((location, key))
862
result.append(((location, key), (self, key,
863
value, self._resolve_references(refs))))
865
result.append(((location, key),
866
(self, key, self._bisect_nodes[key])))
869
# has the region the key should be in, been parsed?
870
index = self._parsed_key_index(key)
871
if (self._parsed_key_map[index][0] <= key and
872
(self._parsed_key_map[index][1] >= key or
873
# end of the file has been parsed
874
self._parsed_byte_map[index][1] == self._size)):
875
result.append(((location, key), False))
877
# no, is the key above or below the probed location:
878
# get the range of the probed & parsed location
879
index = self._parsed_byte_index(location)
880
# if the key is below the start of the range, its below
881
if key < self._parsed_key_map[index][0]:
885
result.append(((location, key), direction))
887
# lookup data to resolve references
888
for location in pending_locations:
890
if location + length > self._size:
891
length = self._size - location
892
# TODO: trim out parsed locations (e.g. if the 800 is into the
893
# parsed region trim it, and dont use the adjust_for_latency
896
readv_ranges.append((location, length))
897
self._read_and_parse(readv_ranges)
898
if self._nodes is not None:
899
# The _read_and_parse triggered a _buffer_all, grab the data and
901
for location, key in pending_references:
902
value, refs = self._nodes[key]
903
result.append(((location, key), (self, key, value, refs)))
905
for location, key in pending_references:
906
# answer key references we had to look-up-late.
907
value, refs = self._bisect_nodes[key]
908
result.append(((location, key), (self, key,
909
value, self._resolve_references(refs))))
912
def _parse_header_from_bytes(self, bytes):
913
"""Parse the header from a region of bytes.
915
:param bytes: The data to parse.
916
:return: An offset, data tuple such as readv yields, for the unparsed
917
data. (which may length 0).
919
signature = bytes[0:len(self._signature())]
920
if not signature == self._signature():
921
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
922
lines = bytes[len(self._signature()):].splitlines()
923
options_line = lines[0]
924
if not options_line.startswith(_OPTION_NODE_REFS):
925
raise errors.BadIndexOptions(self)
927
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
929
raise errors.BadIndexOptions(self)
930
options_line = lines[1]
931
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
932
raise errors.BadIndexOptions(self)
934
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
936
raise errors.BadIndexOptions(self)
937
options_line = lines[2]
938
if not options_line.startswith(_OPTION_LEN):
939
raise errors.BadIndexOptions(self)
941
self._key_count = int(options_line[len(_OPTION_LEN):])
943
raise errors.BadIndexOptions(self)
944
# calculate the bytes we have processed
945
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
947
self._parsed_bytes(0, None, header_end, None)
948
# setup parsing state
949
self._expected_elements = 3 + self._key_length
950
# raw data keyed by offset
951
self._keys_by_offset = {}
952
# keys with the value and node references
953
self._bisect_nodes = {}
954
return header_end, bytes[header_end:]
956
def _parse_region(self, offset, data):
957
"""Parse node data returned from a readv operation.
959
:param offset: The byte offset the data starts at.
960
:param data: The data to parse.
964
end = offset + len(data)
967
# Trivial test - if the current index's end is within the
968
# low-matching parsed range, we're done.
969
index = self._parsed_byte_index(high_parsed)
970
if end < self._parsed_byte_map[index][1]:
972
# print "[%d:%d]" % (offset, end), \
973
# self._parsed_byte_map[index:index + 2]
974
high_parsed, last_segment = self._parse_segment(
975
offset, data, end, index)
979
def _parse_segment(self, offset, data, end, index):
980
"""Parse one segment of data.
982
:param offset: Where 'data' begins in the file.
983
:param data: Some data to parse a segment of.
984
:param end: Where data ends
985
:param index: The current index into the parsed bytes map.
986
:return: True if the parsed segment is the last possible one in the
988
:return: high_parsed_byte, last_segment.
989
high_parsed_byte is the location of the highest parsed byte in this
990
segment, last_segment is True if the parsed segment is the last
991
possible one in the data block.
993
# default is to use all data
995
# accomodate overlap with data before this.
996
if offset < self._parsed_byte_map[index][1]:
997
# overlaps the lower parsed region
998
# skip the parsed data
999
trim_start = self._parsed_byte_map[index][1] - offset
1000
# don't trim the start for \n
1001
start_adjacent = True
1002
elif offset == self._parsed_byte_map[index][1]:
1003
# abuts the lower parsed region
1006
# do not trim anything
1007
start_adjacent = True
1009
# does not overlap the lower parsed region
1012
# but trim the leading \n
1013
start_adjacent = False
1014
if end == self._size:
1015
# lines up to the end of all data:
1018
# do not strip to the last \n
1021
elif index + 1 == len(self._parsed_byte_map):
1022
# at the end of the parsed data
1025
# but strip to the last \n
1026
end_adjacent = False
1028
elif end == self._parsed_byte_map[index + 1][0]:
1029
# buts up against the next parsed region
1032
# do not strip to the last \n
1035
elif end > self._parsed_byte_map[index + 1][0]:
1036
# overlaps into the next parsed region
1037
# only consider the unparsed data
1038
trim_end = self._parsed_byte_map[index + 1][0] - offset
1039
# do not strip to the last \n as we know its an entire record
1041
last_segment = end < self._parsed_byte_map[index + 1][1]
1043
# does not overlap into the next region
1046
# but strip to the last \n
1047
end_adjacent = False
1049
# now find bytes to discard if needed
1050
if not start_adjacent:
1051
# work around python bug in rfind
1052
if trim_start is None:
1053
trim_start = data.find('\n') + 1
1055
trim_start = data.find('\n', trim_start) + 1
1056
if not (trim_start != 0):
1057
raise AssertionError('no \n was present')
1058
# print 'removing start', offset, trim_start, repr(data[:trim_start])
1059
if not end_adjacent:
1060
# work around python bug in rfind
1061
if trim_end is None:
1062
trim_end = data.rfind('\n') + 1
1064
trim_end = data.rfind('\n', None, trim_end) + 1
1065
if not (trim_end != 0):
1066
raise AssertionError('no \n was present')
1067
# print 'removing end', offset, trim_end, repr(data[trim_end:])
1068
# adjust offset and data to the parseable data.
1069
trimmed_data = data[trim_start:trim_end]
1070
if not (trimmed_data):
1071
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
1072
% (trim_start, trim_end, offset, offset + len(data)))
1074
offset += trim_start
1075
# print "parsing", repr(trimmed_data)
1076
# splitlines mangles the \r delimiters.. don't use it.
1077
lines = trimmed_data.split('\n')
1080
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1081
for key, value in nodes:
1082
self._bisect_nodes[key] = value
1083
self._parsed_bytes(offset, first_key,
1084
offset + len(trimmed_data), last_key)
1085
return offset + len(trimmed_data), last_segment
1087
def _parse_lines(self, lines, pos):
1094
# must be at the end
1096
if not (self._size == pos + 1):
1097
raise AssertionError("%s %s" % (self._size, pos))
1100
elements = line.split('\0')
1101
if len(elements) != self._expected_elements:
1102
raise errors.BadIndexData(self)
1103
# keys are tuples. Each element is a string that may occur many
1104
# times, so we intern them to save space. AB, RC, 200807
1105
key = tuple([intern(element) for element in elements[:self._key_length]])
1106
if first_key is None:
1108
absent, references, value = elements[-3:]
1110
for ref_string in references.split('\t'):
1111
ref_lists.append(tuple([
1112
int(ref) for ref in ref_string.split('\r') if ref
1114
ref_lists = tuple(ref_lists)
1115
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1116
pos += len(line) + 1 # +1 for the \n
1119
if self.node_ref_lists:
1120
node_value = (value, ref_lists)
1123
nodes.append((key, node_value))
1124
# print "parsed ", key
1125
return first_key, key, nodes, trailers
1127
def _parsed_bytes(self, start, start_key, end, end_key):
1128
"""Mark the bytes from start to end as parsed.
1130
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1133
:param start: The start of the parsed region.
1134
:param end: The end of the parsed region.
1136
index = self._parsed_byte_index(start)
1137
new_value = (start, end)
1138
new_key = (start_key, end_key)
1140
# first range parsed is always the beginning.
1141
self._parsed_byte_map.insert(index, new_value)
1142
self._parsed_key_map.insert(index, new_key)
1146
# extend lower region
1147
# extend higher region
1148
# combine two regions
1149
if (index + 1 < len(self._parsed_byte_map) and
1150
self._parsed_byte_map[index][1] == start and
1151
self._parsed_byte_map[index + 1][0] == end):
1152
# combine two regions
1153
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1154
self._parsed_byte_map[index + 1][1])
1155
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1156
self._parsed_key_map[index + 1][1])
1157
del self._parsed_byte_map[index + 1]
1158
del self._parsed_key_map[index + 1]
1159
elif self._parsed_byte_map[index][1] == start:
1160
# extend the lower entry
1161
self._parsed_byte_map[index] = (
1162
self._parsed_byte_map[index][0], end)
1163
self._parsed_key_map[index] = (
1164
self._parsed_key_map[index][0], end_key)
1165
elif (index + 1 < len(self._parsed_byte_map) and
1166
self._parsed_byte_map[index + 1][0] == end):
1167
# extend the higher entry
1168
self._parsed_byte_map[index + 1] = (
1169
start, self._parsed_byte_map[index + 1][1])
1170
self._parsed_key_map[index + 1] = (
1171
start_key, self._parsed_key_map[index + 1][1])
1174
self._parsed_byte_map.insert(index + 1, new_value)
1175
self._parsed_key_map.insert(index + 1, new_key)
1177
def _read_and_parse(self, readv_ranges):
1178
"""Read the ranges and parse the resulting data.
1180
:param readv_ranges: A prepared readv range list.
1182
if not readv_ranges:
1184
if self._nodes is None and self._bytes_read * 2 >= self._size:
1185
# We've already read more than 50% of the file and we are about to
1186
# request more data, just _buffer_all() and be done
1190
readv_data = self._transport.readv(self._name, readv_ranges, True,
1193
for offset, data in readv_data:
1194
self._bytes_read += len(data)
1195
if offset == 0 and len(data) == self._size:
1196
# We read the whole range, most likely because the
1197
# Transport upcast our readv ranges into one long request
1198
# for enough total data to grab the whole index.
1199
self._buffer_all(StringIO(data))
1201
if self._bisect_nodes is None:
1202
# this must be the start
1203
if not (offset == 0):
1204
raise AssertionError()
1205
offset, data = self._parse_header_from_bytes(data)
1206
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1207
self._parse_region(offset, data)
1209
def _signature(self):
1210
"""The file signature for this index type."""
1214
"""Validate that everything in the index can be accessed."""
1215
# iter_all validates completely at the moment, so just do that.
1216
for node in self.iter_all_entries():
1220
class CombinedGraphIndex(object):
1221
"""A GraphIndex made up from smaller GraphIndices.
1223
The backing indices must implement GraphIndex, and are presumed to be
1226
Queries against the combined index will be made against the first index,
1227
and then the second and so on. The order of index's can thus influence
1228
performance significantly. For example, if one index is on local disk and a
1229
second on a remote server, the local disk index should be before the other
1233
def __init__(self, indices, reload_func=None):
1234
"""Create a CombinedGraphIndex backed by indices.
1236
:param indices: An ordered list of indices to query for data.
1237
:param reload_func: A function to call if we find we are missing an
1238
index. Should have the form reload_func() => True/False to indicate
1239
if reloading actually changed anything.
1241
self._indices = indices
1242
self._reload_func = reload_func
1246
self.__class__.__name__,
1247
', '.join(map(repr, self._indices)))
1249
def clear_cache(self):
1250
"""See GraphIndex.clear_cache()"""
1251
for index in self._indices:
1254
def get_parent_map(self, keys):
1255
"""See graph.StackedParentsProvider.get_parent_map"""
1256
search_keys = set(keys)
1257
if NULL_REVISION in search_keys:
1258
search_keys.discard(NULL_REVISION)
1259
found_parents = {NULL_REVISION:[]}
1262
for index, key, value, refs in self.iter_entries(search_keys):
1265
parents = (NULL_REVISION,)
1266
found_parents[key] = parents
1267
return found_parents
1269
has_key = _has_key_from_parent_map
1271
def insert_index(self, pos, index):
1272
"""Insert a new index in the list of indices to query.
1274
:param pos: The position to insert the index.
1275
:param index: The index to insert.
1277
self._indices.insert(pos, index)
1279
def iter_all_entries(self):
1280
"""Iterate over all keys within the index
1282
Duplicate keys across child indices are presumed to have the same
1283
value and are only reported once.
1285
:return: An iterable of (index, key, reference_lists, value).
1286
There is no defined order for the result iteration - it will be in
1287
the most efficient order for the index.
1292
for index in self._indices:
1293
for node in index.iter_all_entries():
1294
if node[1] not in seen_keys:
1296
seen_keys.add(node[1])
1298
except errors.NoSuchFile:
1299
self._reload_or_raise()
1301
def iter_entries(self, keys):
1302
"""Iterate over keys within the index.
1304
Duplicate keys across child indices are presumed to have the same
1305
value and are only reported once.
1307
:param keys: An iterable providing the keys to be retrieved.
1308
:return: An iterable of (index, key, reference_lists, value). There is no
1309
defined order for the result iteration - it will be in the most
1310
efficient order for the index.
1315
for index in self._indices:
1318
for node in index.iter_entries(keys):
1319
keys.remove(node[1])
1322
except errors.NoSuchFile:
1323
self._reload_or_raise()
1325
def iter_entries_prefix(self, keys):
1326
"""Iterate over keys within the index using prefix matching.
1328
Duplicate keys across child indices are presumed to have the same
1329
value and are only reported once.
1331
Prefix matching is applied within the tuple of a key, not to within
1332
the bytestring of each key element. e.g. if you have the keys ('foo',
1333
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1334
only the former key is returned.
1336
:param keys: An iterable providing the key prefixes to be retrieved.
1337
Each key prefix takes the form of a tuple the length of a key, but
1338
with the last N elements 'None' rather than a regular bytestring.
1339
The first element cannot be 'None'.
1340
:return: An iterable as per iter_all_entries, but restricted to the
1341
keys with a matching prefix to those supplied. No additional keys
1342
will be returned, and every match that is in the index will be
1351
for index in self._indices:
1352
for node in index.iter_entries_prefix(keys):
1353
if node[1] in seen_keys:
1355
seen_keys.add(node[1])
1358
except errors.NoSuchFile:
1359
self._reload_or_raise()
1361
def find_ancestry(self, keys, ref_list_num):
1362
"""Find the complete ancestry for the given set of keys.
1364
Note that this is a whole-ancestry request, so it should be used
1367
:param keys: An iterable of keys to look for
1368
:param ref_list_num: The reference list which references the parents
1370
:return: (parent_map, missing_keys)
1372
missing_keys = set()
1374
keys_to_lookup = set(keys)
1376
while keys_to_lookup:
1377
# keys that *all* indexes claim are missing, stop searching them
1379
all_index_missing = None
1380
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1381
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1383
# len(missing_keys))
1384
for index_idx, index in enumerate(self._indices):
1385
# TODO: we should probably be doing something with
1386
# 'missing_keys' since we've already determined that
1387
# those revisions have not been found anywhere
1388
index_missing_keys = set()
1389
# Find all of the ancestry we can from this index
1390
# keep looking until the search_keys set is empty, which means
1391
# things we didn't find should be in index_missing_keys
1392
search_keys = keys_to_lookup
1394
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1395
# index_idx, len(search_keys),
1396
# len(parent_map), len(index_missing_keys))
1399
# TODO: ref_list_num should really be a parameter, since
1400
# CombinedGraphIndex does not know what the ref lists
1402
search_keys = index._find_ancestors(search_keys,
1403
ref_list_num, parent_map, index_missing_keys)
1404
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1405
# sub_generation, len(search_keys),
1406
# len(parent_map), len(index_missing_keys))
1407
# Now set whatever was missing to be searched in the next index
1408
keys_to_lookup = index_missing_keys
1409
if all_index_missing is None:
1410
all_index_missing = set(index_missing_keys)
1412
all_index_missing.intersection_update(index_missing_keys)
1413
if not keys_to_lookup:
1415
if all_index_missing is None:
1416
# There were no indexes, so all search keys are 'missing'
1417
missing_keys.update(keys_to_lookup)
1418
keys_to_lookup = None
1420
missing_keys.update(all_index_missing)
1421
keys_to_lookup.difference_update(all_index_missing)
1422
return parent_map, missing_keys
1424
def key_count(self):
1425
"""Return an estimate of the number of keys in this index.
1427
For CombinedGraphIndex this is approximated by the sum of the keys of
1428
the child indices. As child indices may have duplicate keys this can
1429
have a maximum error of the number of child indices * largest number of
1434
return sum((index.key_count() for index in self._indices), 0)
1435
except errors.NoSuchFile:
1436
self._reload_or_raise()
1438
missing_keys = _missing_keys_from_parent_map
1440
def _reload_or_raise(self):
1441
"""We just got a NoSuchFile exception.
1443
Try to reload the indices, if it fails, just raise the current
1446
if self._reload_func is None:
1448
exc_type, exc_value, exc_traceback = sys.exc_info()
1449
trace.mutter('Trying to reload after getting exception: %s',
1451
if not self._reload_func():
1452
# We tried to reload, but nothing changed, so we fail anyway
1453
trace.mutter('_reload_func indicated nothing has changed.'
1454
' Raising original exception.')
1455
raise exc_type, exc_value, exc_traceback
1458
"""Validate that everything in the index can be accessed."""
1461
for index in self._indices:
1464
except errors.NoSuchFile:
1465
self._reload_or_raise()
1468
class InMemoryGraphIndex(GraphIndexBuilder):
1469
"""A GraphIndex which operates entirely out of memory and is mutable.
1471
This is designed to allow the accumulation of GraphIndex entries during a
1472
single write operation, where the accumulated entries need to be immediately
1473
available - for example via a CombinedGraphIndex.
1476
def add_nodes(self, nodes):
1477
"""Add nodes to the index.
1479
:param nodes: An iterable of (key, node_refs, value) entries to add.
1481
if self.reference_lists:
1482
for (key, value, node_refs) in nodes:
1483
self.add_node(key, value, node_refs)
1485
for (key, value) in nodes:
1486
self.add_node(key, value)
1488
def iter_all_entries(self):
1489
"""Iterate over all keys within the index
1491
:return: An iterable of (index, key, reference_lists, value). There is no
1492
defined order for the result iteration - it will be in the most
1493
efficient order for the index (in this case dictionary hash order).
1495
if 'evil' in debug.debug_flags:
1496
trace.mutter_callsite(3,
1497
"iter_all_entries scales with size of history.")
1498
if self.reference_lists:
1499
for key, (absent, references, value) in self._nodes.iteritems():
1501
yield self, key, value, references
1503
for key, (absent, references, value) in self._nodes.iteritems():
1505
yield self, key, value
1507
def iter_entries(self, keys):
1508
"""Iterate over keys within the index.
1510
:param keys: An iterable providing the keys to be retrieved.
1511
:return: An iterable of (index, key, value, reference_lists). There is no
1512
defined order for the result iteration - it will be in the most
1513
efficient order for the index (keys iteration order in this case).
1515
# Note: See BTreeBuilder.iter_entries for an explanation of why we
1516
# aren't using set().intersection() here
1518
keys = [key for key in keys if key in nodes]
1519
if self.reference_lists:
1523
yield self, key, node[2], node[1]
1528
yield self, key, node[2]
1530
def iter_entries_prefix(self, keys):
1531
"""Iterate over keys within the index using prefix matching.
1533
Prefix matching is applied within the tuple of a key, not to within
1534
the bytestring of each key element. e.g. if you have the keys ('foo',
1535
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1536
only the former key is returned.
1538
:param keys: An iterable providing the key prefixes to be retrieved.
1539
Each key prefix takes the form of a tuple the length of a key, but
1540
with the last N elements 'None' rather than a regular bytestring.
1541
The first element cannot be 'None'.
1542
:return: An iterable as per iter_all_entries, but restricted to the
1543
keys with a matching prefix to those supplied. No additional keys
1544
will be returned, and every match that is in the index will be
1547
# XXX: To much duplication with the GraphIndex class; consider finding
1548
# a good place to pull out the actual common logic.
1552
if self._key_length == 1:
1556
raise errors.BadIndexKey(key)
1557
if len(key) != self._key_length:
1558
raise errors.BadIndexKey(key)
1559
node = self._nodes[key]
1562
if self.reference_lists:
1563
yield self, key, node[2], node[1]
1565
yield self, key, node[2]
1567
nodes_by_key = self._get_nodes_by_key()
1571
raise errors.BadIndexKey(key)
1572
if len(key) != self._key_length:
1573
raise errors.BadIndexKey(key)
1574
# find what it refers to:
1575
key_dict = nodes_by_key
1576
elements = list(key)
1577
# find the subdict to return
1579
while len(elements) and elements[0] is not None:
1580
key_dict = key_dict[elements[0]]
1583
# a non-existant lookup.
1588
key_dict = dicts.pop(-1)
1589
# can't be empty or would not exist
1590
item, value = key_dict.iteritems().next()
1591
if type(value) == dict:
1593
dicts.extend(key_dict.itervalues())
1596
for value in key_dict.itervalues():
1597
yield (self, ) + value
1599
yield (self, ) + key_dict
1601
def key_count(self):
1602
"""Return an estimate of the number of keys in this index.
1604
For InMemoryGraphIndex the estimate is exact.
1606
return len(self._nodes) - len(self._absent_keys)
1609
"""In memory index's have no known corruption at the moment."""
1612
class GraphIndexPrefixAdapter(object):
1613
"""An adapter between GraphIndex with different key lengths.
1615
Queries against this will emit queries against the adapted Graph with the
1616
prefix added, queries for all items use iter_entries_prefix. The returned
1617
nodes will have their keys and node references adjusted to remove the
1618
prefix. Finally, an add_nodes_callback can be supplied - when called the
1619
nodes and references being added will have prefix prepended.
1622
def __init__(self, adapted, prefix, missing_key_length,
1623
add_nodes_callback=None):
1624
"""Construct an adapter against adapted with prefix."""
1625
self.adapted = adapted
1626
self.prefix_key = prefix + (None,)*missing_key_length
1627
self.prefix = prefix
1628
self.prefix_len = len(prefix)
1629
self.add_nodes_callback = add_nodes_callback
1631
def add_nodes(self, nodes):
1632
"""Add nodes to the index.
1634
:param nodes: An iterable of (key, node_refs, value) entries to add.
1636
# save nodes in case its an iterator
1637
nodes = tuple(nodes)
1638
translated_nodes = []
1640
# Add prefix_key to each reference node_refs is a tuple of tuples,
1641
# so split it apart, and add prefix_key to the internal reference
1642
for (key, value, node_refs) in nodes:
1643
adjusted_references = (
1644
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1645
for ref_list in node_refs))
1646
translated_nodes.append((self.prefix + key, value,
1647
adjusted_references))
1649
# XXX: TODO add an explicit interface for getting the reference list
1650
# status, to handle this bit of user-friendliness in the API more
1652
for (key, value) in nodes:
1653
translated_nodes.append((self.prefix + key, value))
1654
self.add_nodes_callback(translated_nodes)
1656
def add_node(self, key, value, references=()):
1657
"""Add a node to the index.
1659
:param key: The key. keys are non-empty tuples containing
1660
as many whitespace-free utf8 bytestrings as the key length
1661
defined for this index.
1662
:param references: An iterable of iterables of keys. Each is a
1663
reference to another key.
1664
:param value: The value to associate with the key. It may be any
1665
bytes as long as it does not contain \0 or \n.
1667
self.add_nodes(((key, value, references), ))
1669
def _strip_prefix(self, an_iter):
1670
"""Strip prefix data from nodes and return it."""
1671
for node in an_iter:
1673
if node[1][:self.prefix_len] != self.prefix:
1674
raise errors.BadIndexData(self)
1675
for ref_list in node[3]:
1676
for ref_node in ref_list:
1677
if ref_node[:self.prefix_len] != self.prefix:
1678
raise errors.BadIndexData(self)
1679
yield node[0], node[1][self.prefix_len:], node[2], (
1680
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1681
for ref_list in node[3]))
1683
def iter_all_entries(self):
1684
"""Iterate over all keys within the index
1686
iter_all_entries is implemented against the adapted index using
1687
iter_entries_prefix.
1689
:return: An iterable of (index, key, reference_lists, value). There is no
1690
defined order for the result iteration - it will be in the most
1691
efficient order for the index (in this case dictionary hash order).
1693
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1695
def iter_entries(self, keys):
1696
"""Iterate over keys within the index.
1698
:param keys: An iterable providing the keys to be retrieved.
1699
:return: An iterable of (index, key, value, reference_lists). There is no
1700
defined order for the result iteration - it will be in the most
1701
efficient order for the index (keys iteration order in this case).
1703
return self._strip_prefix(self.adapted.iter_entries(
1704
self.prefix + key for key in keys))
1706
def iter_entries_prefix(self, keys):
1707
"""Iterate over keys within the index using prefix matching.
1709
Prefix matching is applied within the tuple of a key, not to within
1710
the bytestring of each key element. e.g. if you have the keys ('foo',
1711
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1712
only the former key is returned.
1714
:param keys: An iterable providing the key prefixes to be retrieved.
1715
Each key prefix takes the form of a tuple the length of a key, but
1716
with the last N elements 'None' rather than a regular bytestring.
1717
The first element cannot be 'None'.
1718
:return: An iterable as per iter_all_entries, but restricted to the
1719
keys with a matching prefix to those supplied. No additional keys
1720
will be returned, and every match that is in the index will be
1723
return self._strip_prefix(self.adapted.iter_entries_prefix(
1724
self.prefix + key for key in keys))
1726
def key_count(self):
1727
"""Return an estimate of the number of keys in this index.
1729
For GraphIndexPrefixAdapter this is relatively expensive - key
1730
iteration with the prefix is done.
1732
return len(list(self.iter_all_entries()))
1735
"""Call the adapted's validate."""
1736
self.adapted.validate()
added
removed
bzrlib/index.py
