1
# Copyright (C) 2007-2011 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(), """
36
revision as _mod_revision,
44
from bzrlib.static_tuple import StaticTuple
46
_HEADER_READV = (0, 200)
47
_OPTION_KEY_ELEMENTS = "key_elements="
49
_OPTION_NODE_REFS = "node_ref_lists="
50
_SIGNATURE = "Bazaar Graph Index 1\n"
53
_whitespace_re = re.compile('[\t\n\x0b\x0c\r\x00 ]')
54
_newline_null_re = re.compile('[\n\0]')
57
def _has_key_from_parent_map(self, key):
58
"""Check if this index has one key.
60
If it's possible to check for multiple keys at once through
61
calling get_parent_map that should be faster.
63
return (key in self.get_parent_map([key]))
66
def _missing_keys_from_parent_map(self, keys):
67
return set(keys) - set(self.get_parent_map(keys))
70
class GraphIndexBuilder(object):
71
"""A builder that can build a GraphIndex.
73
The resulting graph has the structure:
75
_SIGNATURE OPTIONS NODES NEWLINE
76
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
77
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
79
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
80
KEY := Not-whitespace-utf8
82
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
83
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
84
REFERENCE := DIGITS ; digits is the byte offset in the index of the
86
VALUE := no-newline-no-null-bytes
89
def __init__(self, reference_lists=0, key_elements=1):
90
"""Create a GraphIndex builder.
92
:param reference_lists: The number of node references lists for each
94
:param key_elements: The number of bytestrings in each key.
96
self.reference_lists = reference_lists
97
# A dict of {key: (absent, ref_lists, value)}
99
# Keys that are referenced but not actually present in this index
100
self._absent_keys = set()
101
self._nodes_by_key = None
102
self._key_length = key_elements
103
self._optimize_for_size = False
104
self._combine_backing_indices = True
106
def _check_key(self, key):
107
"""Raise BadIndexKey if key is not a valid key for this index."""
108
if type(key) not in (tuple, StaticTuple):
109
raise errors.BadIndexKey(key)
110
if self._key_length != len(key):
111
raise errors.BadIndexKey(key)
113
if not element or _whitespace_re.search(element) is not None:
114
raise errors.BadIndexKey(element)
116
def _external_references(self):
117
"""Return references that are not present in this index.
121
# TODO: JAM 2008-11-21 This makes an assumption about how the reference
122
# lists are used. It is currently correct for pack-0.92 through
123
# 1.9, which use the node references (3rd column) second
124
# reference list as the compression parent. Perhaps this should
125
# be moved into something higher up the stack, since it
126
# makes assumptions about how the index is used.
127
if self.reference_lists > 1:
128
for node in self.iter_all_entries():
130
refs.update(node[3][1])
133
# If reference_lists == 0 there can be no external references, and
134
# if reference_lists == 1, then there isn't a place to store the
138
def _get_nodes_by_key(self):
139
if self._nodes_by_key is None:
141
if self.reference_lists:
142
for key, (absent, references, value) in self._nodes.iteritems():
145
key_dict = nodes_by_key
146
for subkey in key[:-1]:
147
key_dict = key_dict.setdefault(subkey, {})
148
key_dict[key[-1]] = key, value, references
150
for key, (absent, references, value) in self._nodes.iteritems():
153
key_dict = nodes_by_key
154
for subkey in key[:-1]:
155
key_dict = key_dict.setdefault(subkey, {})
156
key_dict[key[-1]] = key, value
157
self._nodes_by_key = nodes_by_key
158
return self._nodes_by_key
160
def _update_nodes_by_key(self, key, value, node_refs):
161
"""Update the _nodes_by_key dict with a new key.
163
For a key of (foo, bar, baz) create
164
_nodes_by_key[foo][bar][baz] = key_value
166
if self._nodes_by_key is None:
168
key_dict = self._nodes_by_key
169
if self.reference_lists:
170
key_value = StaticTuple(key, value, node_refs)
172
key_value = StaticTuple(key, value)
173
for subkey in key[:-1]:
174
key_dict = key_dict.setdefault(subkey, {})
175
key_dict[key[-1]] = key_value
177
def _check_key_ref_value(self, key, references, value):
178
"""Check that 'key' and 'references' are all valid.
180
:param key: A key tuple. Must conform to the key interface (be a tuple,
181
be of the right length, not have any whitespace or nulls in any key
183
:param references: An iterable of reference lists. Something like
184
[[(ref, key)], [(ref, key), (other, key)]]
185
:param value: The value associate with this key. Must not contain
186
newlines or null characters.
187
:return: (node_refs, absent_references)
188
node_refs basically a packed form of 'references' where all
190
absent_references reference keys that are not in self._nodes.
191
This may contain duplicates if the same key is
192
referenced in multiple lists.
194
as_st = StaticTuple.from_sequence
196
if _newline_null_re.search(value) is not None:
197
raise errors.BadIndexValue(value)
198
if len(references) != self.reference_lists:
199
raise errors.BadIndexValue(references)
201
absent_references = []
202
for reference_list in references:
203
for reference in reference_list:
204
# If reference *is* in self._nodes, then we know it has already
206
if reference not in self._nodes:
207
self._check_key(reference)
208
absent_references.append(reference)
209
reference_list = as_st([as_st(ref).intern()
210
for ref in reference_list])
211
node_refs.append(reference_list)
212
return as_st(node_refs), absent_references
214
def add_node(self, key, value, references=()):
215
"""Add a node to the index.
217
:param key: The key. keys are non-empty tuples containing
218
as many whitespace-free utf8 bytestrings as the key length
219
defined for this index.
220
:param references: An iterable of iterables of keys. Each is a
221
reference to another key.
222
:param value: The value to associate with the key. It may be any
223
bytes as long as it does not contain \0 or \n.
226
absent_references) = self._check_key_ref_value(key, references, value)
227
if key in self._nodes and self._nodes[key][0] != 'a':
228
raise errors.BadIndexDuplicateKey(key, self)
229
for reference in absent_references:
230
# There may be duplicates, but I don't think it is worth worrying
232
self._nodes[reference] = ('a', (), '')
233
self._absent_keys.update(absent_references)
234
self._absent_keys.discard(key)
235
self._nodes[key] = ('', node_refs, value)
236
if self._nodes_by_key is not None and self._key_length > 1:
237
self._update_nodes_by_key(key, value, node_refs)
239
def clear_cache(self):
240
"""See GraphIndex.clear_cache()
242
This is a no-op, but we need the api to conform to a generic 'Index'
248
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
249
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
250
key_count = len(self._nodes) - len(self._absent_keys)
251
lines.append(_OPTION_LEN + str(key_count) + '\n')
252
prefix_length = sum(len(x) for x in lines)
253
# references are byte offsets. To avoid having to do nasty
254
# polynomial work to resolve offsets (references to later in the
255
# file cannot be determined until all the inbetween references have
256
# been calculated too) we pad the offsets with 0's to make them be
257
# of consistent length. Using binary offsets would break the trivial
259
# to calculate the width of zero's needed we do three passes:
260
# one to gather all the non-reference data and the number of references.
261
# one to pad all the data with reference-length and determine entry
265
# forward sorted by key. In future we may consider topological sorting,
266
# at the cost of table scans for direct lookup, or a second index for
268
nodes = sorted(self._nodes.items())
269
# if we do not prepass, we don't know how long it will be up front.
270
expected_bytes = None
271
# we only need to pre-pass if we have reference lists at all.
272
if self.reference_lists:
274
non_ref_bytes = prefix_length
276
# TODO use simple multiplication for the constants in this loop.
277
for key, (absent, references, value) in nodes:
278
# record the offset known *so far* for this key:
279
# the non reference bytes to date, and the total references to
280
# date - saves reaccumulating on the second pass
281
key_offset_info.append((key, non_ref_bytes, total_references))
282
# key is literal, value is literal, there are 3 null's, 1 NL
283
# key is variable length tuple, \x00 between elements
284
non_ref_bytes += sum(len(element) for element in key)
285
if self._key_length > 1:
286
non_ref_bytes += self._key_length - 1
287
# value is literal bytes, there are 3 null's, 1 NL.
288
non_ref_bytes += len(value) + 3 + 1
289
# one byte for absent if set.
292
elif self.reference_lists:
293
# (ref_lists -1) tabs
294
non_ref_bytes += self.reference_lists - 1
295
# (ref-1 cr's per ref_list)
296
for ref_list in references:
297
# how many references across the whole file?
298
total_references += len(ref_list)
299
# accrue reference separators
301
non_ref_bytes += len(ref_list) - 1
302
# how many digits are needed to represent the total byte count?
304
possible_total_bytes = non_ref_bytes + total_references*digits
305
while 10 ** digits < possible_total_bytes:
307
possible_total_bytes = non_ref_bytes + total_references*digits
308
expected_bytes = possible_total_bytes + 1 # terminating newline
309
# resolve key addresses.
311
for key, non_ref_bytes, total_references in key_offset_info:
312
key_addresses[key] = non_ref_bytes + total_references*digits
314
format_string = '%%0%sd' % digits
315
for key, (absent, references, value) in nodes:
316
flattened_references = []
317
for ref_list in references:
319
for reference in ref_list:
320
ref_addresses.append(format_string % key_addresses[reference])
321
flattened_references.append('\r'.join(ref_addresses))
322
string_key = '\x00'.join(key)
323
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
324
'\t'.join(flattened_references), value))
326
result = StringIO(''.join(lines))
327
if expected_bytes and len(result.getvalue()) != expected_bytes:
328
raise errors.BzrError('Failed index creation. Internal error:'
329
' mismatched output length and expected length: %d %d' %
330
(len(result.getvalue()), expected_bytes))
333
def set_optimize(self, for_size=None, combine_backing_indices=None):
334
"""Change how the builder tries to optimize the result.
336
:param for_size: Tell the builder to try and make the index as small as
338
:param combine_backing_indices: If the builder spills to disk to save
339
memory, should the on-disk indices be combined. Set to True if you
340
are going to be probing the index, but to False if you are not. (If
341
you are not querying, then the time spent combining is wasted.)
344
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
346
if for_size is not None:
347
self._optimize_for_size = for_size
348
if combine_backing_indices is not None:
349
self._combine_backing_indices = combine_backing_indices
351
def find_ancestry(self, keys, ref_list_num):
352
"""See CombinedGraphIndex.find_ancestry()"""
358
for _, key, value, ref_lists in self.iter_entries(pending):
359
parent_keys = ref_lists[ref_list_num]
360
parent_map[key] = parent_keys
361
next_pending.update([p for p in parent_keys if p not in
363
missing_keys.update(pending.difference(parent_map))
364
pending = next_pending
365
return parent_map, missing_keys
368
class GraphIndex(object):
369
"""An index for data with embedded graphs.
371
The index maps keys to a list of key reference lists, and a value.
372
Each node has the same number of key reference lists. Each key reference
373
list can be empty or an arbitrary length. The value is an opaque NULL
374
terminated string without any newlines. The storage of the index is
375
hidden in the interface: keys and key references are always tuples of
376
bytestrings, never the internal representation (e.g. dictionary offsets).
378
It is presumed that the index will not be mutated - it is static data.
380
Successive iter_all_entries calls will read the entire index each time.
381
Additionally, iter_entries calls will read the index linearly until the
382
desired keys are found. XXX: This must be fixed before the index is
383
suitable for production use. :XXX
386
def __init__(self, transport, name, size, unlimited_cache=False, offset=0):
387
"""Open an index called name on transport.
389
:param transport: A bzrlib.transport.Transport.
390
:param name: A path to provide to transport API calls.
391
:param size: The size of the index in bytes. This is used for bisection
392
logic to perform partial index reads. While the size could be
393
obtained by statting the file this introduced an additional round
394
trip as well as requiring stat'able transports, both of which are
395
avoided by having it supplied. If size is None, then bisection
396
support will be disabled and accessing the index will just stream
398
:param offset: Instead of starting the index data at offset 0, start it
399
at an arbitrary offset.
401
self._transport = transport
403
# Becomes a dict of key:(value, reference-list-byte-locations) used by
404
# the bisection interface to store parsed but not resolved keys.
405
self._bisect_nodes = None
406
# Becomes a dict of key:(value, reference-list-keys) which are ready to
407
# be returned directly to callers.
409
# a sorted list of slice-addresses for the parsed bytes of the file.
410
# e.g. (0,1) would mean that byte 0 is parsed.
411
self._parsed_byte_map = []
412
# a sorted list of keys matching each slice address for parsed bytes
413
# e.g. (None, 'foo@bar') would mean that the first byte contained no
414
# key, and the end byte of the slice is the of the data for 'foo@bar'
415
self._parsed_key_map = []
416
self._key_count = None
417
self._keys_by_offset = None
418
self._nodes_by_key = None
420
# The number of bytes we've read so far in trying to process this file
422
self._base_offset = offset
424
def __eq__(self, other):
425
"""Equal when self and other were created with the same parameters."""
427
type(self) == type(other) and
428
self._transport == other._transport and
429
self._name == other._name and
430
self._size == other._size)
432
def __ne__(self, other):
433
return not self.__eq__(other)
436
return "%s(%r)" % (self.__class__.__name__,
437
self._transport.abspath(self._name))
439
def _buffer_all(self, stream=None):
440
"""Buffer all the index data.
442
Mutates self._nodes and self.keys_by_offset.
444
if self._nodes is not None:
445
# We already did this
447
if 'index' in debug.debug_flags:
448
trace.mutter('Reading entire index %s',
449
self._transport.abspath(self._name))
451
stream = self._transport.get(self._name)
452
if self._base_offset != 0:
453
# This is wasteful, but it is better than dealing with
454
# adjusting all the offsets, etc.
455
stream = StringIO(stream.read()[self._base_offset:])
456
self._read_prefix(stream)
457
self._expected_elements = 3 + self._key_length
459
# raw data keyed by offset
460
self._keys_by_offset = {}
461
# ready-to-return key:value or key:value, node_ref_lists
463
self._nodes_by_key = None
466
lines = stream.read().split('\n')
467
# GZ 2009-09-20: Should really use a try/finally block to ensure close
470
_, _, _, trailers = self._parse_lines(lines, pos)
471
for key, absent, references, value in self._keys_by_offset.itervalues():
474
# resolve references:
475
if self.node_ref_lists:
476
node_value = (value, self._resolve_references(references))
479
self._nodes[key] = node_value
480
# cache the keys for quick set intersections
482
# there must be one line - the empty trailer line.
483
raise errors.BadIndexData(self)
485
def clear_cache(self):
486
"""Clear out any cached/memoized values.
488
This can be called at any time, but generally it is used when we have
489
extracted some information, but don't expect to be requesting any more
493
def external_references(self, ref_list_num):
494
"""Return references that are not present in this index.
497
if ref_list_num + 1 > self.node_ref_lists:
498
raise ValueError('No ref list %d, index has %d ref lists'
499
% (ref_list_num, self.node_ref_lists))
502
for key, (value, ref_lists) in nodes.iteritems():
503
ref_list = ref_lists[ref_list_num]
504
refs.update([ref for ref in ref_list if ref not in nodes])
507
def _get_nodes_by_key(self):
508
if self._nodes_by_key is None:
510
if self.node_ref_lists:
511
for key, (value, references) in self._nodes.iteritems():
512
key_dict = nodes_by_key
513
for subkey in key[:-1]:
514
key_dict = key_dict.setdefault(subkey, {})
515
key_dict[key[-1]] = key, value, references
517
for key, value in self._nodes.iteritems():
518
key_dict = nodes_by_key
519
for subkey in key[:-1]:
520
key_dict = key_dict.setdefault(subkey, {})
521
key_dict[key[-1]] = key, value
522
self._nodes_by_key = nodes_by_key
523
return self._nodes_by_key
525
def iter_all_entries(self):
526
"""Iterate over all keys within the index.
528
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
529
The former tuple is used when there are no reference lists in the
530
index, making the API compatible with simple key:value index types.
531
There is no defined order for the result iteration - it will be in
532
the most efficient order for the index.
534
if 'evil' in debug.debug_flags:
535
trace.mutter_callsite(3,
536
"iter_all_entries scales with size of history.")
537
if self._nodes is None:
539
if self.node_ref_lists:
540
for key, (value, node_ref_lists) in self._nodes.iteritems():
541
yield self, key, value, node_ref_lists
543
for key, value in self._nodes.iteritems():
544
yield self, key, value
546
def _read_prefix(self, stream):
547
signature = stream.read(len(self._signature()))
548
if not signature == self._signature():
549
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
550
options_line = stream.readline()
551
if not options_line.startswith(_OPTION_NODE_REFS):
552
raise errors.BadIndexOptions(self)
554
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
556
raise errors.BadIndexOptions(self)
557
options_line = stream.readline()
558
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
559
raise errors.BadIndexOptions(self)
561
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
563
raise errors.BadIndexOptions(self)
564
options_line = stream.readline()
565
if not options_line.startswith(_OPTION_LEN):
566
raise errors.BadIndexOptions(self)
568
self._key_count = int(options_line[len(_OPTION_LEN):-1])
570
raise errors.BadIndexOptions(self)
572
def _resolve_references(self, references):
573
"""Return the resolved key references for references.
575
References are resolved by looking up the location of the key in the
576
_keys_by_offset map and substituting the key name, preserving ordering.
578
:param references: An iterable of iterables of key locations. e.g.
580
:return: A tuple of tuples of keys.
583
for ref_list in references:
584
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
585
return tuple(node_refs)
587
def _find_index(self, range_map, key):
588
"""Helper for the _parsed_*_index calls.
590
Given a range map - [(start, end), ...], finds the index of the range
591
in the map for key if it is in the map, and if it is not there, the
592
immediately preceeding range in the map.
594
result = bisect_right(range_map, key) - 1
595
if result + 1 < len(range_map):
596
# check the border condition, it may be in result + 1
597
if range_map[result + 1][0] == key[0]:
601
def _parsed_byte_index(self, offset):
602
"""Return the index of the entry immediately before offset.
604
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
605
there is one unparsed byte (the 11th, addressed as[10]). then:
606
asking for 0 will return 0
607
asking for 10 will return 0
608
asking for 11 will return 1
609
asking for 12 will return 1
612
return self._find_index(self._parsed_byte_map, key)
614
def _parsed_key_index(self, key):
615
"""Return the index of the entry immediately before key.
617
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
618
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
619
have been parsed, then:
620
asking for '' will return 0
621
asking for 'a' will return 0
622
asking for 'b' will return 1
623
asking for 'e' will return 1
625
search_key = (key, None)
626
return self._find_index(self._parsed_key_map, search_key)
628
def _is_parsed(self, offset):
629
"""Returns True if offset has been parsed."""
630
index = self._parsed_byte_index(offset)
631
if index == len(self._parsed_byte_map):
632
return offset < self._parsed_byte_map[index - 1][1]
633
start, end = self._parsed_byte_map[index]
634
return offset >= start and offset < end
636
def _iter_entries_from_total_buffer(self, keys):
637
"""Iterate over keys when the entire index is parsed."""
638
# Note: See the note in BTreeBuilder.iter_entries for why we don't use
639
# .intersection() here
641
keys = [key for key in keys if key in nodes]
642
if self.node_ref_lists:
644
value, node_refs = nodes[key]
645
yield self, key, value, node_refs
648
yield self, key, nodes[key]
650
def iter_entries(self, keys):
651
"""Iterate over keys within the index.
653
:param keys: An iterable providing the keys to be retrieved.
654
:return: An iterable as per iter_all_entries, but restricted to the
655
keys supplied. No additional keys will be returned, and every
656
key supplied that is in the index will be returned.
661
if self._size is None and self._nodes is None:
664
# We fit about 20 keys per minimum-read (4K), so if we are looking for
665
# more than 1/20th of the index its likely (assuming homogenous key
666
# spread) that we'll read the entire index. If we're going to do that,
667
# buffer the whole thing. A better analysis might take key spread into
668
# account - but B+Tree indices are better anyway.
669
# We could look at all data read, and use a threshold there, which will
670
# trigger on ancestry walks, but that is not yet fully mapped out.
671
if self._nodes is None and len(keys) * 20 > self.key_count():
673
if self._nodes is not None:
674
return self._iter_entries_from_total_buffer(keys)
676
return (result[1] for result in bisect_multi.bisect_multi_bytes(
677
self._lookup_keys_via_location, self._size, keys))
679
def iter_entries_prefix(self, keys):
680
"""Iterate over keys within the index using prefix matching.
682
Prefix matching is applied within the tuple of a key, not to within
683
the bytestring of each key element. e.g. if you have the keys ('foo',
684
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
685
only the former key is returned.
687
WARNING: Note that this method currently causes a full index parse
688
unconditionally (which is reasonably appropriate as it is a means for
689
thunking many small indices into one larger one and still supplies
690
iter_all_entries at the thunk layer).
692
:param keys: An iterable providing the key prefixes to be retrieved.
693
Each key prefix takes the form of a tuple the length of a key, but
694
with the last N elements 'None' rather than a regular bytestring.
695
The first element cannot be 'None'.
696
:return: An iterable as per iter_all_entries, but restricted to the
697
keys with a matching prefix to those supplied. No additional keys
698
will be returned, and every match that is in the index will be
704
# load data - also finds key lengths
705
if self._nodes is None:
707
if self._key_length == 1:
711
raise errors.BadIndexKey(key)
712
if len(key) != self._key_length:
713
raise errors.BadIndexKey(key)
714
if self.node_ref_lists:
715
value, node_refs = self._nodes[key]
716
yield self, key, value, node_refs
718
yield self, key, self._nodes[key]
720
nodes_by_key = self._get_nodes_by_key()
724
raise errors.BadIndexKey(key)
725
if len(key) != self._key_length:
726
raise errors.BadIndexKey(key)
727
# find what it refers to:
728
key_dict = nodes_by_key
730
# find the subdict whose contents should be returned.
732
while len(elements) and elements[0] is not None:
733
key_dict = key_dict[elements[0]]
736
# a non-existant lookup.
741
key_dict = dicts.pop(-1)
742
# can't be empty or would not exist
743
item, value = key_dict.iteritems().next()
744
if type(value) == dict:
746
dicts.extend(key_dict.itervalues())
749
for value in key_dict.itervalues():
750
# each value is the key:value:node refs tuple
752
yield (self, ) + value
754
# the last thing looked up was a terminal element
755
yield (self, ) + key_dict
757
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
758
"""See BTreeIndex._find_ancestors."""
759
# The api can be implemented as a trivial overlay on top of
760
# iter_entries, it is not an efficient implementation, but it at least
764
for index, key, value, refs in self.iter_entries(keys):
765
parent_keys = refs[ref_list_num]
767
parent_map[key] = parent_keys
768
search_keys.update(parent_keys)
769
# Figure out what, if anything, was missing
770
missing_keys.update(set(keys).difference(found_keys))
771
search_keys = search_keys.difference(parent_map)
775
"""Return an estimate of the number of keys in this index.
777
For GraphIndex the estimate is exact.
779
if self._key_count is None:
780
self._read_and_parse([_HEADER_READV])
781
return self._key_count
783
def _lookup_keys_via_location(self, location_keys):
784
"""Public interface for implementing bisection.
786
If _buffer_all has been called, then all the data for the index is in
787
memory, and this method should not be called, as it uses a separate
788
cache because it cannot pre-resolve all indices, which buffer_all does
791
:param location_keys: A list of location(byte offset), key tuples.
792
:return: A list of (location_key, result) tuples as expected by
793
bzrlib.bisect_multi.bisect_multi_bytes.
795
# Possible improvements:
796
# - only bisect lookup each key once
797
# - sort the keys first, and use that to reduce the bisection window
799
# this progresses in three parts:
802
# attempt to answer the question from the now in memory data.
803
# build the readv request
804
# for each location, ask for 800 bytes - much more than rows we've seen
807
for location, key in location_keys:
808
# can we answer from cache?
809
if self._bisect_nodes and key in self._bisect_nodes:
810
# We have the key parsed.
812
index = self._parsed_key_index(key)
813
if (len(self._parsed_key_map) and
814
self._parsed_key_map[index][0] <= key and
815
(self._parsed_key_map[index][1] >= key or
816
# end of the file has been parsed
817
self._parsed_byte_map[index][1] == self._size)):
818
# the key has been parsed, so no lookup is needed even if its
821
# - if we have examined this part of the file already - yes
822
index = self._parsed_byte_index(location)
823
if (len(self._parsed_byte_map) and
824
self._parsed_byte_map[index][0] <= location and
825
self._parsed_byte_map[index][1] > location):
826
# the byte region has been parsed, so no read is needed.
829
if location + length > self._size:
830
length = self._size - location
831
# todo, trim out parsed locations.
833
readv_ranges.append((location, length))
834
# read the header if needed
835
if self._bisect_nodes is None:
836
readv_ranges.append(_HEADER_READV)
837
self._read_and_parse(readv_ranges)
839
if self._nodes is not None:
840
# _read_and_parse triggered a _buffer_all because we requested the
842
for location, key in location_keys:
843
if key not in self._nodes: # not present
844
result.append(((location, key), False))
845
elif self.node_ref_lists:
846
value, refs = self._nodes[key]
847
result.append(((location, key),
848
(self, key, value, refs)))
850
result.append(((location, key),
851
(self, key, self._nodes[key])))
854
# - figure out <, >, missing, present
855
# - result present references so we can return them.
856
# keys that we cannot answer until we resolve references
857
pending_references = []
858
pending_locations = set()
859
for location, key in location_keys:
860
# can we answer from cache?
861
if key in self._bisect_nodes:
862
# the key has been parsed, so no lookup is needed
863
if self.node_ref_lists:
864
# the references may not have been all parsed.
865
value, refs = self._bisect_nodes[key]
866
wanted_locations = []
867
for ref_list in refs:
869
if ref not in self._keys_by_offset:
870
wanted_locations.append(ref)
872
pending_locations.update(wanted_locations)
873
pending_references.append((location, key))
875
result.append(((location, key), (self, key,
876
value, self._resolve_references(refs))))
878
result.append(((location, key),
879
(self, key, self._bisect_nodes[key])))
882
# has the region the key should be in, been parsed?
883
index = self._parsed_key_index(key)
884
if (self._parsed_key_map[index][0] <= key and
885
(self._parsed_key_map[index][1] >= key or
886
# end of the file has been parsed
887
self._parsed_byte_map[index][1] == self._size)):
888
result.append(((location, key), False))
890
# no, is the key above or below the probed location:
891
# get the range of the probed & parsed location
892
index = self._parsed_byte_index(location)
893
# if the key is below the start of the range, its below
894
if key < self._parsed_key_map[index][0]:
898
result.append(((location, key), direction))
900
# lookup data to resolve references
901
for location in pending_locations:
903
if location + length > self._size:
904
length = self._size - location
905
# TODO: trim out parsed locations (e.g. if the 800 is into the
906
# parsed region trim it, and dont use the adjust_for_latency
909
readv_ranges.append((location, length))
910
self._read_and_parse(readv_ranges)
911
if self._nodes is not None:
912
# The _read_and_parse triggered a _buffer_all, grab the data and
914
for location, key in pending_references:
915
value, refs = self._nodes[key]
916
result.append(((location, key), (self, key, value, refs)))
918
for location, key in pending_references:
919
# answer key references we had to look-up-late.
920
value, refs = self._bisect_nodes[key]
921
result.append(((location, key), (self, key,
922
value, self._resolve_references(refs))))
925
def _parse_header_from_bytes(self, bytes):
926
"""Parse the header from a region of bytes.
928
:param bytes: The data to parse.
929
:return: An offset, data tuple such as readv yields, for the unparsed
930
data. (which may length 0).
932
signature = bytes[0:len(self._signature())]
933
if not signature == self._signature():
934
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
935
lines = bytes[len(self._signature()):].splitlines()
936
options_line = lines[0]
937
if not options_line.startswith(_OPTION_NODE_REFS):
938
raise errors.BadIndexOptions(self)
940
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
942
raise errors.BadIndexOptions(self)
943
options_line = lines[1]
944
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
945
raise errors.BadIndexOptions(self)
947
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
949
raise errors.BadIndexOptions(self)
950
options_line = lines[2]
951
if not options_line.startswith(_OPTION_LEN):
952
raise errors.BadIndexOptions(self)
954
self._key_count = int(options_line[len(_OPTION_LEN):])
956
raise errors.BadIndexOptions(self)
957
# calculate the bytes we have processed
958
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
960
self._parsed_bytes(0, None, header_end, None)
961
# setup parsing state
962
self._expected_elements = 3 + self._key_length
963
# raw data keyed by offset
964
self._keys_by_offset = {}
965
# keys with the value and node references
966
self._bisect_nodes = {}
967
return header_end, bytes[header_end:]
969
def _parse_region(self, offset, data):
970
"""Parse node data returned from a readv operation.
972
:param offset: The byte offset the data starts at.
973
:param data: The data to parse.
977
end = offset + len(data)
980
# Trivial test - if the current index's end is within the
981
# low-matching parsed range, we're done.
982
index = self._parsed_byte_index(high_parsed)
983
if end < self._parsed_byte_map[index][1]:
985
# print "[%d:%d]" % (offset, end), \
986
# self._parsed_byte_map[index:index + 2]
987
high_parsed, last_segment = self._parse_segment(
988
offset, data, end, index)
992
def _parse_segment(self, offset, data, end, index):
993
"""Parse one segment of data.
995
:param offset: Where 'data' begins in the file.
996
:param data: Some data to parse a segment of.
997
:param end: Where data ends
998
:param index: The current index into the parsed bytes map.
999
:return: True if the parsed segment is the last possible one in the
1001
:return: high_parsed_byte, last_segment.
1002
high_parsed_byte is the location of the highest parsed byte in this
1003
segment, last_segment is True if the parsed segment is the last
1004
possible one in the data block.
1006
# default is to use all data
1008
# accomodate overlap with data before this.
1009
if offset < self._parsed_byte_map[index][1]:
1010
# overlaps the lower parsed region
1011
# skip the parsed data
1012
trim_start = self._parsed_byte_map[index][1] - offset
1013
# don't trim the start for \n
1014
start_adjacent = True
1015
elif offset == self._parsed_byte_map[index][1]:
1016
# abuts the lower parsed region
1019
# do not trim anything
1020
start_adjacent = True
1022
# does not overlap the lower parsed region
1025
# but trim the leading \n
1026
start_adjacent = False
1027
if end == self._size:
1028
# lines up to the end of all data:
1031
# do not strip to the last \n
1034
elif index + 1 == len(self._parsed_byte_map):
1035
# at the end of the parsed data
1038
# but strip to the last \n
1039
end_adjacent = False
1041
elif end == self._parsed_byte_map[index + 1][0]:
1042
# buts up against the next parsed region
1045
# do not strip to the last \n
1048
elif end > self._parsed_byte_map[index + 1][0]:
1049
# overlaps into the next parsed region
1050
# only consider the unparsed data
1051
trim_end = self._parsed_byte_map[index + 1][0] - offset
1052
# do not strip to the last \n as we know its an entire record
1054
last_segment = end < self._parsed_byte_map[index + 1][1]
1056
# does not overlap into the next region
1059
# but strip to the last \n
1060
end_adjacent = False
1062
# now find bytes to discard if needed
1063
if not start_adjacent:
1064
# work around python bug in rfind
1065
if trim_start is None:
1066
trim_start = data.find('\n') + 1
1068
trim_start = data.find('\n', trim_start) + 1
1069
if not (trim_start != 0):
1070
raise AssertionError('no \n was present')
1071
# print 'removing start', offset, trim_start, repr(data[:trim_start])
1072
if not end_adjacent:
1073
# work around python bug in rfind
1074
if trim_end is None:
1075
trim_end = data.rfind('\n') + 1
1077
trim_end = data.rfind('\n', None, trim_end) + 1
1078
if not (trim_end != 0):
1079
raise AssertionError('no \n was present')
1080
# print 'removing end', offset, trim_end, repr(data[trim_end:])
1081
# adjust offset and data to the parseable data.
1082
trimmed_data = data[trim_start:trim_end]
1083
if not (trimmed_data):
1084
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
1085
% (trim_start, trim_end, offset, offset + len(data)))
1087
offset += trim_start
1088
# print "parsing", repr(trimmed_data)
1089
# splitlines mangles the \r delimiters.. don't use it.
1090
lines = trimmed_data.split('\n')
1093
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1094
for key, value in nodes:
1095
self._bisect_nodes[key] = value
1096
self._parsed_bytes(offset, first_key,
1097
offset + len(trimmed_data), last_key)
1098
return offset + len(trimmed_data), last_segment
1100
def _parse_lines(self, lines, pos):
1107
# must be at the end
1109
if not (self._size == pos + 1):
1110
raise AssertionError("%s %s" % (self._size, pos))
1113
elements = line.split('\0')
1114
if len(elements) != self._expected_elements:
1115
raise errors.BadIndexData(self)
1116
# keys are tuples. Each element is a string that may occur many
1117
# times, so we intern them to save space. AB, RC, 200807
1118
key = tuple([intern(element) for element in elements[:self._key_length]])
1119
if first_key is None:
1121
absent, references, value = elements[-3:]
1123
for ref_string in references.split('\t'):
1124
ref_lists.append(tuple([
1125
int(ref) for ref in ref_string.split('\r') if ref
1127
ref_lists = tuple(ref_lists)
1128
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1129
pos += len(line) + 1 # +1 for the \n
1132
if self.node_ref_lists:
1133
node_value = (value, ref_lists)
1136
nodes.append((key, node_value))
1137
# print "parsed ", key
1138
return first_key, key, nodes, trailers
1140
def _parsed_bytes(self, start, start_key, end, end_key):
1141
"""Mark the bytes from start to end as parsed.
1143
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1146
:param start: The start of the parsed region.
1147
:param end: The end of the parsed region.
1149
index = self._parsed_byte_index(start)
1150
new_value = (start, end)
1151
new_key = (start_key, end_key)
1153
# first range parsed is always the beginning.
1154
self._parsed_byte_map.insert(index, new_value)
1155
self._parsed_key_map.insert(index, new_key)
1159
# extend lower region
1160
# extend higher region
1161
# combine two regions
1162
if (index + 1 < len(self._parsed_byte_map) and
1163
self._parsed_byte_map[index][1] == start and
1164
self._parsed_byte_map[index + 1][0] == end):
1165
# combine two regions
1166
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1167
self._parsed_byte_map[index + 1][1])
1168
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1169
self._parsed_key_map[index + 1][1])
1170
del self._parsed_byte_map[index + 1]
1171
del self._parsed_key_map[index + 1]
1172
elif self._parsed_byte_map[index][1] == start:
1173
# extend the lower entry
1174
self._parsed_byte_map[index] = (
1175
self._parsed_byte_map[index][0], end)
1176
self._parsed_key_map[index] = (
1177
self._parsed_key_map[index][0], end_key)
1178
elif (index + 1 < len(self._parsed_byte_map) and
1179
self._parsed_byte_map[index + 1][0] == end):
1180
# extend the higher entry
1181
self._parsed_byte_map[index + 1] = (
1182
start, self._parsed_byte_map[index + 1][1])
1183
self._parsed_key_map[index + 1] = (
1184
start_key, self._parsed_key_map[index + 1][1])
1187
self._parsed_byte_map.insert(index + 1, new_value)
1188
self._parsed_key_map.insert(index + 1, new_key)
1190
def _read_and_parse(self, readv_ranges):
1191
"""Read the ranges and parse the resulting data.
1193
:param readv_ranges: A prepared readv range list.
1195
if not readv_ranges:
1197
if self._nodes is None and self._bytes_read * 2 >= self._size:
1198
# We've already read more than 50% of the file and we are about to
1199
# request more data, just _buffer_all() and be done
1203
base_offset = self._base_offset
1204
if base_offset != 0:
1205
# Rewrite the ranges for the offset
1206
readv_ranges = [(start+base_offset, size)
1207
for start, size in readv_ranges]
1208
readv_data = self._transport.readv(self._name, readv_ranges, True,
1209
self._size + self._base_offset)
1211
for offset, data in readv_data:
1212
offset -= base_offset
1213
self._bytes_read += len(data)
1215
# transport.readv() expanded to extra data which isn't part of
1217
data = data[-offset:]
1219
if offset == 0 and len(data) == self._size:
1220
# We read the whole range, most likely because the
1221
# Transport upcast our readv ranges into one long request
1222
# for enough total data to grab the whole index.
1223
self._buffer_all(StringIO(data))
1225
if self._bisect_nodes is None:
1226
# this must be the start
1227
if not (offset == 0):
1228
raise AssertionError()
1229
offset, data = self._parse_header_from_bytes(data)
1230
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1231
self._parse_region(offset, data)
1233
def _signature(self):
1234
"""The file signature for this index type."""
1238
"""Validate that everything in the index can be accessed."""
1239
# iter_all validates completely at the moment, so just do that.
1240
for node in self.iter_all_entries():
1244
class CombinedGraphIndex(object):
1245
"""A GraphIndex made up from smaller GraphIndices.
1247
The backing indices must implement GraphIndex, and are presumed to be
1250
Queries against the combined index will be made against the first index,
1251
and then the second and so on. The order of indices can thus influence
1252
performance significantly. For example, if one index is on local disk and a
1253
second on a remote server, the local disk index should be before the other
1256
Also, queries tend to need results from the same indices as previous
1257
queries. So the indices will be reordered after every query to put the
1258
indices that had the result(s) of that query first (while otherwise
1259
preserving the relative ordering).
1262
def __init__(self, indices, reload_func=None):
1263
"""Create a CombinedGraphIndex backed by indices.
1265
:param indices: An ordered list of indices to query for data.
1266
:param reload_func: A function to call if we find we are missing an
1267
index. Should have the form reload_func() => True/False to indicate
1268
if reloading actually changed anything.
1270
self._indices = indices
1271
self._reload_func = reload_func
1272
# Sibling indices are other CombinedGraphIndex that we should call
1273
# _move_to_front_by_name on when we auto-reorder ourself.
1274
self._sibling_indices = []
1275
# A list of names that corresponds to the instances in self._indices,
1276
# so _index_names[0] is always the name for _indices[0], etc. Sibling
1277
# indices must all use the same set of names as each other.
1278
self._index_names = [None] * len(self._indices)
1282
self.__class__.__name__,
1283
', '.join(map(repr, self._indices)))
1285
def clear_cache(self):
1286
"""See GraphIndex.clear_cache()"""
1287
for index in self._indices:
1290
def get_parent_map(self, keys):
1291
"""See graph.StackedParentsProvider.get_parent_map"""
1292
search_keys = set(keys)
1293
if _mod_revision.NULL_REVISION in search_keys:
1294
search_keys.discard(_mod_revision.NULL_REVISION)
1295
found_parents = {_mod_revision.NULL_REVISION:[]}
1298
for index, key, value, refs in self.iter_entries(search_keys):
1301
parents = (_mod_revision.NULL_REVISION,)
1302
found_parents[key] = parents
1303
return found_parents
1305
has_key = _has_key_from_parent_map
1307
def insert_index(self, pos, index, name=None):
1308
"""Insert a new index in the list of indices to query.
1310
:param pos: The position to insert the index.
1311
:param index: The index to insert.
1312
:param name: a name for this index, e.g. a pack name. These names can
1313
be used to reflect index reorderings to related CombinedGraphIndex
1314
instances that use the same names. (see set_sibling_indices)
1316
self._indices.insert(pos, index)
1317
self._index_names.insert(pos, name)
1319
def iter_all_entries(self):
1320
"""Iterate over all keys within the index
1322
Duplicate keys across child indices are presumed to have the same
1323
value and are only reported once.
1325
:return: An iterable of (index, key, reference_lists, value).
1326
There is no defined order for the result iteration - it will be in
1327
the most efficient order for the index.
1332
for index in self._indices:
1333
for node in index.iter_all_entries():
1334
if node[1] not in seen_keys:
1336
seen_keys.add(node[1])
1338
except errors.NoSuchFile:
1339
self._reload_or_raise()
1341
def iter_entries(self, keys):
1342
"""Iterate over keys within the index.
1344
Duplicate keys across child indices are presumed to have the same
1345
value and are only reported once.
1347
:param keys: An iterable providing the keys to be retrieved.
1348
:return: An iterable of (index, key, reference_lists, value). There is
1349
no defined order for the result iteration - it will be in the most
1350
efficient order for the index.
1356
for index in self._indices:
1360
for node in index.iter_entries(keys):
1361
keys.remove(node[1])
1365
hit_indices.append(index)
1367
except errors.NoSuchFile:
1368
self._reload_or_raise()
1369
self._move_to_front(hit_indices)
1371
def iter_entries_prefix(self, keys):
1372
"""Iterate over keys within the index using prefix matching.
1374
Duplicate keys across child indices are presumed to have the same
1375
value and are only reported once.
1377
Prefix matching is applied within the tuple of a key, not to within
1378
the bytestring of each key element. e.g. if you have the keys ('foo',
1379
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1380
only the former key is returned.
1382
:param keys: An iterable providing the key prefixes to be retrieved.
1383
Each key prefix takes the form of a tuple the length of a key, but
1384
with the last N elements 'None' rather than a regular bytestring.
1385
The first element cannot be 'None'.
1386
:return: An iterable as per iter_all_entries, but restricted to the
1387
keys with a matching prefix to those supplied. No additional keys
1388
will be returned, and every match that is in the index will be
1398
for index in self._indices:
1400
for node in index.iter_entries_prefix(keys):
1401
if node[1] in seen_keys:
1403
seen_keys.add(node[1])
1407
hit_indices.append(index)
1409
except errors.NoSuchFile:
1410
self._reload_or_raise()
1411
self._move_to_front(hit_indices)
1413
def _move_to_front(self, hit_indices):
1414
"""Rearrange self._indices so that hit_indices are first.
1416
Order is maintained as much as possible, e.g. the first unhit index
1417
will be the first index in _indices after the hit_indices, and the
1418
hit_indices will be present in exactly the order they are passed to
1421
_move_to_front propagates to all objects in self._sibling_indices by
1422
calling _move_to_front_by_name.
1424
if self._indices[:len(hit_indices)] == hit_indices:
1425
# The 'hit_indices' are already at the front (and in the same
1426
# order), no need to re-order
1428
hit_names = self._move_to_front_by_index(hit_indices)
1429
for sibling_idx in self._sibling_indices:
1430
sibling_idx._move_to_front_by_name(hit_names)
1432
def _move_to_front_by_index(self, hit_indices):
1433
"""Core logic for _move_to_front.
1435
Returns a list of names corresponding to the hit_indices param.
1437
indices_info = zip(self._index_names, self._indices)
1438
if 'index' in debug.debug_flags:
1439
trace.mutter('CombinedGraphIndex reordering: currently %r, '
1440
'promoting %r', indices_info, hit_indices)
1443
new_hit_indices = []
1446
for offset, (name, idx) in enumerate(indices_info):
1447
if idx in hit_indices:
1448
hit_names.append(name)
1449
new_hit_indices.append(idx)
1450
if len(new_hit_indices) == len(hit_indices):
1451
# We've found all of the hit entries, everything else is
1453
unhit_names.extend(self._index_names[offset+1:])
1454
unhit_indices.extend(self._indices[offset+1:])
1457
unhit_names.append(name)
1458
unhit_indices.append(idx)
1460
self._indices = new_hit_indices + unhit_indices
1461
self._index_names = hit_names + unhit_names
1462
if 'index' in debug.debug_flags:
1463
trace.mutter('CombinedGraphIndex reordered: %r', self._indices)
1466
def _move_to_front_by_name(self, hit_names):
1467
"""Moves indices named by 'hit_names' to front of the search order, as
1468
described in _move_to_front.
1470
# Translate names to index instances, and then call
1471
# _move_to_front_by_index.
1472
indices_info = zip(self._index_names, self._indices)
1474
for name, idx in indices_info:
1475
if name in hit_names:
1476
hit_indices.append(idx)
1477
self._move_to_front_by_index(hit_indices)
1479
def find_ancestry(self, keys, ref_list_num):
1480
"""Find the complete ancestry for the given set of keys.
1482
Note that this is a whole-ancestry request, so it should be used
1485
:param keys: An iterable of keys to look for
1486
:param ref_list_num: The reference list which references the parents
1488
:return: (parent_map, missing_keys)
1490
# XXX: make this call _move_to_front?
1491
missing_keys = set()
1493
keys_to_lookup = set(keys)
1495
while keys_to_lookup:
1496
# keys that *all* indexes claim are missing, stop searching them
1498
all_index_missing = None
1499
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1500
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1502
# len(missing_keys))
1503
for index_idx, index in enumerate(self._indices):
1504
# TODO: we should probably be doing something with
1505
# 'missing_keys' since we've already determined that
1506
# those revisions have not been found anywhere
1507
index_missing_keys = set()
1508
# Find all of the ancestry we can from this index
1509
# keep looking until the search_keys set is empty, which means
1510
# things we didn't find should be in index_missing_keys
1511
search_keys = keys_to_lookup
1513
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1514
# index_idx, len(search_keys),
1515
# len(parent_map), len(index_missing_keys))
1518
# TODO: ref_list_num should really be a parameter, since
1519
# CombinedGraphIndex does not know what the ref lists
1521
search_keys = index._find_ancestors(search_keys,
1522
ref_list_num, parent_map, index_missing_keys)
1523
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1524
# sub_generation, len(search_keys),
1525
# len(parent_map), len(index_missing_keys))
1526
# Now set whatever was missing to be searched in the next index
1527
keys_to_lookup = index_missing_keys
1528
if all_index_missing is None:
1529
all_index_missing = set(index_missing_keys)
1531
all_index_missing.intersection_update(index_missing_keys)
1532
if not keys_to_lookup:
1534
if all_index_missing is None:
1535
# There were no indexes, so all search keys are 'missing'
1536
missing_keys.update(keys_to_lookup)
1537
keys_to_lookup = None
1539
missing_keys.update(all_index_missing)
1540
keys_to_lookup.difference_update(all_index_missing)
1541
return parent_map, missing_keys
1543
def key_count(self):
1544
"""Return an estimate of the number of keys in this index.
1546
For CombinedGraphIndex this is approximated by the sum of the keys of
1547
the child indices. As child indices may have duplicate keys this can
1548
have a maximum error of the number of child indices * largest number of
1553
return sum((index.key_count() for index in self._indices), 0)
1554
except errors.NoSuchFile:
1555
self._reload_or_raise()
1557
missing_keys = _missing_keys_from_parent_map
1559
def _reload_or_raise(self):
1560
"""We just got a NoSuchFile exception.
1562
Try to reload the indices, if it fails, just raise the current
1565
if self._reload_func is None:
1567
exc_type, exc_value, exc_traceback = sys.exc_info()
1568
trace.mutter('Trying to reload after getting exception: %s',
1570
if not self._reload_func():
1571
# We tried to reload, but nothing changed, so we fail anyway
1572
trace.mutter('_reload_func indicated nothing has changed.'
1573
' Raising original exception.')
1574
raise exc_type, exc_value, exc_traceback
1576
def set_sibling_indices(self, sibling_combined_graph_indices):
1577
"""Set the CombinedGraphIndex objects to reorder after reordering self.
1579
self._sibling_indices = sibling_combined_graph_indices
1582
"""Validate that everything in the index can be accessed."""
1585
for index in self._indices:
1588
except errors.NoSuchFile:
1589
self._reload_or_raise()
1592
class InMemoryGraphIndex(GraphIndexBuilder):
1593
"""A GraphIndex which operates entirely out of memory and is mutable.
1595
This is designed to allow the accumulation of GraphIndex entries during a
1596
single write operation, where the accumulated entries need to be immediately
1597
available - for example via a CombinedGraphIndex.
1600
def add_nodes(self, nodes):
1601
"""Add nodes to the index.
1603
:param nodes: An iterable of (key, node_refs, value) entries to add.
1605
if self.reference_lists:
1606
for (key, value, node_refs) in nodes:
1607
self.add_node(key, value, node_refs)
1609
for (key, value) in nodes:
1610
self.add_node(key, value)
1612
def iter_all_entries(self):
1613
"""Iterate over all keys within the index
1615
:return: An iterable of (index, key, reference_lists, value). There is no
1616
defined order for the result iteration - it will be in the most
1617
efficient order for the index (in this case dictionary hash order).
1619
if 'evil' in debug.debug_flags:
1620
trace.mutter_callsite(3,
1621
"iter_all_entries scales with size of history.")
1622
if self.reference_lists:
1623
for key, (absent, references, value) in self._nodes.iteritems():
1625
yield self, key, value, references
1627
for key, (absent, references, value) in self._nodes.iteritems():
1629
yield self, key, value
1631
def iter_entries(self, keys):
1632
"""Iterate over keys within the index.
1634
:param keys: An iterable providing the keys to be retrieved.
1635
:return: An iterable of (index, key, value, reference_lists). There is no
1636
defined order for the result iteration - it will be in the most
1637
efficient order for the index (keys iteration order in this case).
1639
# Note: See BTreeBuilder.iter_entries for an explanation of why we
1640
# aren't using set().intersection() here
1642
keys = [key for key in keys if key in nodes]
1643
if self.reference_lists:
1647
yield self, key, node[2], node[1]
1652
yield self, key, node[2]
1654
def iter_entries_prefix(self, keys):
1655
"""Iterate over keys within the index using prefix matching.
1657
Prefix matching is applied within the tuple of a key, not to within
1658
the bytestring of each key element. e.g. if you have the keys ('foo',
1659
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1660
only the former key is returned.
1662
:param keys: An iterable providing the key prefixes to be retrieved.
1663
Each key prefix takes the form of a tuple the length of a key, but
1664
with the last N elements 'None' rather than a regular bytestring.
1665
The first element cannot be 'None'.
1666
:return: An iterable as per iter_all_entries, but restricted to the
1667
keys with a matching prefix to those supplied. No additional keys
1668
will be returned, and every match that is in the index will be
1671
# XXX: To much duplication with the GraphIndex class; consider finding
1672
# a good place to pull out the actual common logic.
1676
if self._key_length == 1:
1680
raise errors.BadIndexKey(key)
1681
if len(key) != self._key_length:
1682
raise errors.BadIndexKey(key)
1683
node = self._nodes[key]
1686
if self.reference_lists:
1687
yield self, key, node[2], node[1]
1689
yield self, key, node[2]
1691
nodes_by_key = self._get_nodes_by_key()
1695
raise errors.BadIndexKey(key)
1696
if len(key) != self._key_length:
1697
raise errors.BadIndexKey(key)
1698
# find what it refers to:
1699
key_dict = nodes_by_key
1700
elements = list(key)
1701
# find the subdict to return
1703
while len(elements) and elements[0] is not None:
1704
key_dict = key_dict[elements[0]]
1707
# a non-existant lookup.
1712
key_dict = dicts.pop(-1)
1713
# can't be empty or would not exist
1714
item, value = key_dict.iteritems().next()
1715
if type(value) == dict:
1717
dicts.extend(key_dict.itervalues())
1720
for value in key_dict.itervalues():
1721
yield (self, ) + value
1723
yield (self, ) + key_dict
1725
def key_count(self):
1726
"""Return an estimate of the number of keys in this index.
1728
For InMemoryGraphIndex the estimate is exact.
1730
return len(self._nodes) - len(self._absent_keys)
1733
"""In memory index's have no known corruption at the moment."""
1736
class GraphIndexPrefixAdapter(object):
1737
"""An adapter between GraphIndex with different key lengths.
1739
Queries against this will emit queries against the adapted Graph with the
1740
prefix added, queries for all items use iter_entries_prefix. The returned
1741
nodes will have their keys and node references adjusted to remove the
1742
prefix. Finally, an add_nodes_callback can be supplied - when called the
1743
nodes and references being added will have prefix prepended.
1746
def __init__(self, adapted, prefix, missing_key_length,
1747
add_nodes_callback=None):
1748
"""Construct an adapter against adapted with prefix."""
1749
self.adapted = adapted
1750
self.prefix_key = prefix + (None,)*missing_key_length
1751
self.prefix = prefix
1752
self.prefix_len = len(prefix)
1753
self.add_nodes_callback = add_nodes_callback
1755
def add_nodes(self, nodes):
1756
"""Add nodes to the index.
1758
:param nodes: An iterable of (key, node_refs, value) entries to add.
1760
# save nodes in case its an iterator
1761
nodes = tuple(nodes)
1762
translated_nodes = []
1764
# Add prefix_key to each reference node_refs is a tuple of tuples,
1765
# so split it apart, and add prefix_key to the internal reference
1766
for (key, value, node_refs) in nodes:
1767
adjusted_references = (
1768
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1769
for ref_list in node_refs))
1770
translated_nodes.append((self.prefix + key, value,
1771
adjusted_references))
1773
# XXX: TODO add an explicit interface for getting the reference list
1774
# status, to handle this bit of user-friendliness in the API more
1776
for (key, value) in nodes:
1777
translated_nodes.append((self.prefix + key, value))
1778
self.add_nodes_callback(translated_nodes)
1780
def add_node(self, key, value, references=()):
1781
"""Add a node to the index.
1783
:param key: The key. keys are non-empty tuples containing
1784
as many whitespace-free utf8 bytestrings as the key length
1785
defined for this index.
1786
:param references: An iterable of iterables of keys. Each is a
1787
reference to another key.
1788
:param value: The value to associate with the key. It may be any
1789
bytes as long as it does not contain \0 or \n.
1791
self.add_nodes(((key, value, references), ))
1793
def _strip_prefix(self, an_iter):
1794
"""Strip prefix data from nodes and return it."""
1795
for node in an_iter:
1797
if node[1][:self.prefix_len] != self.prefix:
1798
raise errors.BadIndexData(self)
1799
for ref_list in node[3]:
1800
for ref_node in ref_list:
1801
if ref_node[:self.prefix_len] != self.prefix:
1802
raise errors.BadIndexData(self)
1803
yield node[0], node[1][self.prefix_len:], node[2], (
1804
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1805
for ref_list in node[3]))
1807
def iter_all_entries(self):
1808
"""Iterate over all keys within the index
1810
iter_all_entries is implemented against the adapted index using
1811
iter_entries_prefix.
1813
:return: An iterable of (index, key, reference_lists, value). There is no
1814
defined order for the result iteration - it will be in the most
1815
efficient order for the index (in this case dictionary hash order).
1817
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1819
def iter_entries(self, keys):
1820
"""Iterate over keys within the index.
1822
:param keys: An iterable providing the keys to be retrieved.
1823
:return: An iterable of (index, key, value, reference_lists). There is no
1824
defined order for the result iteration - it will be in the most
1825
efficient order for the index (keys iteration order in this case).
1827
return self._strip_prefix(self.adapted.iter_entries(
1828
self.prefix + key for key in keys))
1830
def iter_entries_prefix(self, keys):
1831
"""Iterate over keys within the index using prefix matching.
1833
Prefix matching is applied within the tuple of a key, not to within
1834
the bytestring of each key element. e.g. if you have the keys ('foo',
1835
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1836
only the former key is returned.
1838
:param keys: An iterable providing the key prefixes to be retrieved.
1839
Each key prefix takes the form of a tuple the length of a key, but
1840
with the last N elements 'None' rather than a regular bytestring.
1841
The first element cannot be 'None'.
1842
:return: An iterable as per iter_all_entries, but restricted to the
1843
keys with a matching prefix to those supplied. No additional keys
1844
will be returned, and every match that is in the index will be
1847
return self._strip_prefix(self.adapted.iter_entries_prefix(
1848
self.prefix + key for key in keys))
1850
def key_count(self):
1851
"""Return an estimate of the number of keys in this index.
1853
For GraphIndexPrefixAdapter this is relatively expensive - key
1854
iteration with the prefix is done.
1856
return len(list(self.iter_all_entries()))
1859
"""Call the adapted's validate."""
1860
self.adapted.validate()
added
removed
bzrlib/index.py
