1
# Copyright (C) 2007-2010 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
"""Indexing facilities."""
23
'GraphIndexPrefixAdapter',
27
from bisect import bisect_right
28
from cStringIO import StringIO
32
from bzrlib.lazy_import import lazy_import
33
lazy_import(globals(), """
34
from bzrlib import trace
35
from bzrlib.bisect_multi import bisect_multi_bytes
36
from bzrlib.revision import NULL_REVISION
37
from bzrlib.trace import mutter
43
from bzrlib.static_tuple import StaticTuple
45
_HEADER_READV = (0, 200)
46
_OPTION_KEY_ELEMENTS = "key_elements="
48
_OPTION_NODE_REFS = "node_ref_lists="
49
_SIGNATURE = "Bazaar Graph Index 1\n"
52
_whitespace_re = re.compile('[\t\n\x0b\x0c\r\x00 ]')
53
_newline_null_re = re.compile('[\n\0]')
56
def _has_key_from_parent_map(self, key):
57
"""Check if this index has one key.
59
If it's possible to check for multiple keys at once through
60
calling get_parent_map that should be faster.
62
return (key in self.get_parent_map([key]))
65
def _missing_keys_from_parent_map(self, keys):
66
return set(keys) - set(self.get_parent_map(keys))
69
class GraphIndexBuilder(object):
70
"""A builder that can build a GraphIndex.
72
The resulting graph has the structure:
74
_SIGNATURE OPTIONS NODES NEWLINE
75
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
76
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
78
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
79
KEY := Not-whitespace-utf8
81
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
82
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
83
REFERENCE := DIGITS ; digits is the byte offset in the index of the
85
VALUE := no-newline-no-null-bytes
88
def __init__(self, reference_lists=0, key_elements=1):
89
"""Create a GraphIndex builder.
91
:param reference_lists: The number of node references lists for each
93
:param key_elements: The number of bytestrings in each key.
95
self.reference_lists = reference_lists
96
# A dict of {key: (absent, ref_lists, value)}
98
# Keys that are referenced but not actually present in this index
99
self._absent_keys = set()
100
self._nodes_by_key = None
101
self._key_length = key_elements
102
self._optimize_for_size = False
103
self._combine_backing_indices = True
105
def _check_key(self, key):
106
"""Raise BadIndexKey if key is not a valid key for this index."""
107
if type(key) not in (tuple, StaticTuple):
108
raise errors.BadIndexKey(key)
109
if self._key_length != len(key):
110
raise errors.BadIndexKey(key)
112
if not element or _whitespace_re.search(element) is not None:
113
raise errors.BadIndexKey(element)
115
def _external_references(self):
116
"""Return references that are not present in this index.
120
# TODO: JAM 2008-11-21 This makes an assumption about how the reference
121
# lists are used. It is currently correct for pack-0.92 through
122
# 1.9, which use the node references (3rd column) second
123
# reference list as the compression parent. Perhaps this should
124
# be moved into something higher up the stack, since it
125
# makes assumptions about how the index is used.
126
if self.reference_lists > 1:
127
for node in self.iter_all_entries():
129
refs.update(node[3][1])
132
# If reference_lists == 0 there can be no external references, and
133
# if reference_lists == 1, then there isn't a place to store the
137
def _get_nodes_by_key(self):
138
if self._nodes_by_key is None:
140
if self.reference_lists:
141
for key, (absent, references, value) in self._nodes.iteritems():
144
key_dict = nodes_by_key
145
for subkey in key[:-1]:
146
key_dict = key_dict.setdefault(subkey, {})
147
key_dict[key[-1]] = key, value, references
149
for key, (absent, references, value) in self._nodes.iteritems():
152
key_dict = nodes_by_key
153
for subkey in key[:-1]:
154
key_dict = key_dict.setdefault(subkey, {})
155
key_dict[key[-1]] = key, value
156
self._nodes_by_key = nodes_by_key
157
return self._nodes_by_key
159
def _update_nodes_by_key(self, key, value, node_refs):
160
"""Update the _nodes_by_key dict with a new key.
162
For a key of (foo, bar, baz) create
163
_nodes_by_key[foo][bar][baz] = key_value
165
if self._nodes_by_key is None:
167
key_dict = self._nodes_by_key
168
if self.reference_lists:
169
key_value = StaticTuple(key, value, node_refs)
171
key_value = StaticTuple(key, value)
172
for subkey in key[:-1]:
173
key_dict = key_dict.setdefault(subkey, {})
174
key_dict[key[-1]] = key_value
176
def _check_key_ref_value(self, key, references, value):
177
"""Check that 'key' and 'references' are all valid.
179
:param key: A key tuple. Must conform to the key interface (be a tuple,
180
be of the right length, not have any whitespace or nulls in any key
182
:param references: An iterable of reference lists. Something like
183
[[(ref, key)], [(ref, key), (other, key)]]
184
:param value: The value associate with this key. Must not contain
185
newlines or null characters.
186
:return: (node_refs, absent_references)
187
node_refs basically a packed form of 'references' where all
189
absent_references reference keys that are not in self._nodes.
190
This may contain duplicates if the same key is
191
referenced in multiple lists.
193
as_st = StaticTuple.from_sequence
195
if _newline_null_re.search(value) is not None:
196
raise errors.BadIndexValue(value)
197
if len(references) != self.reference_lists:
198
raise errors.BadIndexValue(references)
200
absent_references = []
201
for reference_list in references:
202
for reference in reference_list:
203
# If reference *is* in self._nodes, then we know it has already
205
if reference not in self._nodes:
206
self._check_key(reference)
207
absent_references.append(reference)
208
reference_list = as_st([as_st(ref).intern()
209
for ref in reference_list])
210
node_refs.append(reference_list)
211
return as_st(node_refs), absent_references
213
def add_node(self, key, value, references=()):
214
"""Add a node to the index.
216
:param key: The key. keys are non-empty tuples containing
217
as many whitespace-free utf8 bytestrings as the key length
218
defined for this index.
219
:param references: An iterable of iterables of keys. Each is a
220
reference to another key.
221
:param value: The value to associate with the key. It may be any
222
bytes as long as it does not contain \0 or \n.
225
absent_references) = self._check_key_ref_value(key, references, value)
226
if key in self._nodes and self._nodes[key][0] != 'a':
227
raise errors.BadIndexDuplicateKey(key, self)
228
for reference in absent_references:
229
# There may be duplicates, but I don't think it is worth worrying
231
self._nodes[reference] = ('a', (), '')
232
self._absent_keys.update(absent_references)
233
self._absent_keys.discard(key)
234
self._nodes[key] = ('', node_refs, value)
235
if self._nodes_by_key is not None and self._key_length > 1:
236
self._update_nodes_by_key(key, value, node_refs)
238
def clear_cache(self):
239
"""See GraphIndex.clear_cache()
241
This is a no-op, but we need the api to conform to a generic 'Index'
247
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
248
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
249
key_count = len(self._nodes) - len(self._absent_keys)
250
lines.append(_OPTION_LEN + str(key_count) + '\n')
251
prefix_length = sum(len(x) for x in lines)
252
# references are byte offsets. To avoid having to do nasty
253
# polynomial work to resolve offsets (references to later in the
254
# file cannot be determined until all the inbetween references have
255
# been calculated too) we pad the offsets with 0's to make them be
256
# of consistent length. Using binary offsets would break the trivial
258
# to calculate the width of zero's needed we do three passes:
259
# one to gather all the non-reference data and the number of references.
260
# one to pad all the data with reference-length and determine entry
264
# forward sorted by key. In future we may consider topological sorting,
265
# at the cost of table scans for direct lookup, or a second index for
267
nodes = sorted(self._nodes.items())
268
# if we do not prepass, we don't know how long it will be up front.
269
expected_bytes = None
270
# we only need to pre-pass if we have reference lists at all.
271
if self.reference_lists:
273
non_ref_bytes = prefix_length
275
# TODO use simple multiplication for the constants in this loop.
276
for key, (absent, references, value) in nodes:
277
# record the offset known *so far* for this key:
278
# the non reference bytes to date, and the total references to
279
# date - saves reaccumulating on the second pass
280
key_offset_info.append((key, non_ref_bytes, total_references))
281
# key is literal, value is literal, there are 3 null's, 1 NL
282
# key is variable length tuple, \x00 between elements
283
non_ref_bytes += sum(len(element) for element in key)
284
if self._key_length > 1:
285
non_ref_bytes += self._key_length - 1
286
# value is literal bytes, there are 3 null's, 1 NL.
287
non_ref_bytes += len(value) + 3 + 1
288
# one byte for absent if set.
291
elif self.reference_lists:
292
# (ref_lists -1) tabs
293
non_ref_bytes += self.reference_lists - 1
294
# (ref-1 cr's per ref_list)
295
for ref_list in references:
296
# how many references across the whole file?
297
total_references += len(ref_list)
298
# accrue reference separators
300
non_ref_bytes += len(ref_list) - 1
301
# how many digits are needed to represent the total byte count?
303
possible_total_bytes = non_ref_bytes + total_references*digits
304
while 10 ** digits < possible_total_bytes:
306
possible_total_bytes = non_ref_bytes + total_references*digits
307
expected_bytes = possible_total_bytes + 1 # terminating newline
308
# resolve key addresses.
310
for key, non_ref_bytes, total_references in key_offset_info:
311
key_addresses[key] = non_ref_bytes + total_references*digits
313
format_string = '%%0%sd' % digits
314
for key, (absent, references, value) in nodes:
315
flattened_references = []
316
for ref_list in references:
318
for reference in ref_list:
319
ref_addresses.append(format_string % key_addresses[reference])
320
flattened_references.append('\r'.join(ref_addresses))
321
string_key = '\x00'.join(key)
322
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
323
'\t'.join(flattened_references), value))
325
result = StringIO(''.join(lines))
326
if expected_bytes and len(result.getvalue()) != expected_bytes:
327
raise errors.BzrError('Failed index creation. Internal error:'
328
' mismatched output length and expected length: %d %d' %
329
(len(result.getvalue()), expected_bytes))
332
def set_optimize(self, for_size=None, combine_backing_indices=None):
333
"""Change how the builder tries to optimize the result.
335
:param for_size: Tell the builder to try and make the index as small as
337
:param combine_backing_indices: If the builder spills to disk to save
338
memory, should the on-disk indices be combined. Set to True if you
339
are going to be probing the index, but to False if you are not. (If
340
you are not querying, then the time spent combining is wasted.)
343
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
345
if for_size is not None:
346
self._optimize_for_size = for_size
347
if combine_backing_indices is not None:
348
self._combine_backing_indices = combine_backing_indices
350
def find_ancestry(self, keys, ref_list_num):
351
"""See CombinedGraphIndex.find_ancestry()"""
357
for _, key, value, ref_lists in self.iter_entries(pending):
358
parent_keys = ref_lists[ref_list_num]
359
parent_map[key] = parent_keys
360
next_pending.update([p for p in parent_keys if p not in
362
missing_keys.update(pending.difference(parent_map))
363
pending = next_pending
364
return parent_map, missing_keys
367
class GraphIndex(object):
368
"""An index for data with embedded graphs.
370
The index maps keys to a list of key reference lists, and a value.
371
Each node has the same number of key reference lists. Each key reference
372
list can be empty or an arbitrary length. The value is an opaque NULL
373
terminated string without any newlines. The storage of the index is
374
hidden in the interface: keys and key references are always tuples of
375
bytestrings, never the internal representation (e.g. dictionary offsets).
377
It is presumed that the index will not be mutated - it is static data.
379
Successive iter_all_entries calls will read the entire index each time.
380
Additionally, iter_entries calls will read the index linearly until the
381
desired keys are found. XXX: This must be fixed before the index is
382
suitable for production use. :XXX
385
def __init__(self, transport, name, size, unlimited_cache=False, offset=0):
386
"""Open an index called name on transport.
388
:param transport: A bzrlib.transport.Transport.
389
:param name: A path to provide to transport API calls.
390
:param size: The size of the index in bytes. This is used for bisection
391
logic to perform partial index reads. While the size could be
392
obtained by statting the file this introduced an additional round
393
trip as well as requiring stat'able transports, both of which are
394
avoided by having it supplied. If size is None, then bisection
395
support will be disabled and accessing the index will just stream
397
:param offset: Instead of starting the index data at offset 0, start it
398
at an arbitrary offset.
400
self._transport = transport
402
# Becomes a dict of key:(value, reference-list-byte-locations) used by
403
# the bisection interface to store parsed but not resolved keys.
404
self._bisect_nodes = None
405
# Becomes a dict of key:(value, reference-list-keys) which are ready to
406
# be returned directly to callers.
408
# a sorted list of slice-addresses for the parsed bytes of the file.
409
# e.g. (0,1) would mean that byte 0 is parsed.
410
self._parsed_byte_map = []
411
# a sorted list of keys matching each slice address for parsed bytes
412
# e.g. (None, 'foo@bar') would mean that the first byte contained no
413
# key, and the end byte of the slice is the of the data for 'foo@bar'
414
self._parsed_key_map = []
415
self._key_count = None
416
self._keys_by_offset = None
417
self._nodes_by_key = None
419
# The number of bytes we've read so far in trying to process this file
421
self._base_offset = offset
423
def __eq__(self, other):
424
"""Equal when self and other were created with the same parameters."""
426
type(self) == type(other) and
427
self._transport == other._transport and
428
self._name == other._name and
429
self._size == other._size)
431
def __ne__(self, other):
432
return not self.__eq__(other)
435
return "%s(%r)" % (self.__class__.__name__,
436
self._transport.abspath(self._name))
438
def _buffer_all(self, stream=None):
439
"""Buffer all the index data.
441
Mutates self._nodes and self.keys_by_offset.
443
if self._nodes is not None:
444
# We already did this
446
if 'index' in debug.debug_flags:
447
mutter('Reading entire index %s', self._transport.abspath(self._name))
449
stream = self._transport.get(self._name)
450
if self._base_offset != 0:
451
# This is wasteful, but it is better than dealing with
452
# adjusting all the offsets, etc.
453
stream = StringIO(stream.read()[self._base_offset:])
454
self._read_prefix(stream)
455
self._expected_elements = 3 + self._key_length
457
# raw data keyed by offset
458
self._keys_by_offset = {}
459
# ready-to-return key:value or key:value, node_ref_lists
461
self._nodes_by_key = None
464
lines = stream.read().split('\n')
465
# GZ 2009-09-20: Should really use a try/finally block to ensure close
468
_, _, _, trailers = self._parse_lines(lines, pos)
469
for key, absent, references, value in self._keys_by_offset.itervalues():
472
# resolve references:
473
if self.node_ref_lists:
474
node_value = (value, self._resolve_references(references))
477
self._nodes[key] = node_value
478
# cache the keys for quick set intersections
480
# there must be one line - the empty trailer line.
481
raise errors.BadIndexData(self)
483
def clear_cache(self):
484
"""Clear out any cached/memoized values.
486
This can be called at any time, but generally it is used when we have
487
extracted some information, but don't expect to be requesting any more
491
def external_references(self, ref_list_num):
492
"""Return references that are not present in this index.
495
if ref_list_num + 1 > self.node_ref_lists:
496
raise ValueError('No ref list %d, index has %d ref lists'
497
% (ref_list_num, self.node_ref_lists))
500
for key, (value, ref_lists) in nodes.iteritems():
501
ref_list = ref_lists[ref_list_num]
502
refs.update([ref for ref in ref_list if ref not in nodes])
505
def _get_nodes_by_key(self):
506
if self._nodes_by_key is None:
508
if self.node_ref_lists:
509
for key, (value, references) in self._nodes.iteritems():
510
key_dict = nodes_by_key
511
for subkey in key[:-1]:
512
key_dict = key_dict.setdefault(subkey, {})
513
key_dict[key[-1]] = key, value, references
515
for key, value in self._nodes.iteritems():
516
key_dict = nodes_by_key
517
for subkey in key[:-1]:
518
key_dict = key_dict.setdefault(subkey, {})
519
key_dict[key[-1]] = key, value
520
self._nodes_by_key = nodes_by_key
521
return self._nodes_by_key
523
def iter_all_entries(self):
524
"""Iterate over all keys within the index.
526
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
527
The former tuple is used when there are no reference lists in the
528
index, making the API compatible with simple key:value index types.
529
There is no defined order for the result iteration - it will be in
530
the most efficient order for the index.
532
if 'evil' in debug.debug_flags:
533
trace.mutter_callsite(3,
534
"iter_all_entries scales with size of history.")
535
if self._nodes is None:
537
if self.node_ref_lists:
538
for key, (value, node_ref_lists) in self._nodes.iteritems():
539
yield self, key, value, node_ref_lists
541
for key, value in self._nodes.iteritems():
542
yield self, key, value
544
def _read_prefix(self, stream):
545
signature = stream.read(len(self._signature()))
546
if not signature == self._signature():
547
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
548
options_line = stream.readline()
549
if not options_line.startswith(_OPTION_NODE_REFS):
550
raise errors.BadIndexOptions(self)
552
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
554
raise errors.BadIndexOptions(self)
555
options_line = stream.readline()
556
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
557
raise errors.BadIndexOptions(self)
559
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
561
raise errors.BadIndexOptions(self)
562
options_line = stream.readline()
563
if not options_line.startswith(_OPTION_LEN):
564
raise errors.BadIndexOptions(self)
566
self._key_count = int(options_line[len(_OPTION_LEN):-1])
568
raise errors.BadIndexOptions(self)
570
def _resolve_references(self, references):
571
"""Return the resolved key references for references.
573
References are resolved by looking up the location of the key in the
574
_keys_by_offset map and substituting the key name, preserving ordering.
576
:param references: An iterable of iterables of key locations. e.g.
578
:return: A tuple of tuples of keys.
581
for ref_list in references:
582
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
583
return tuple(node_refs)
585
def _find_index(self, range_map, key):
586
"""Helper for the _parsed_*_index calls.
588
Given a range map - [(start, end), ...], finds the index of the range
589
in the map for key if it is in the map, and if it is not there, the
590
immediately preceeding range in the map.
592
result = bisect_right(range_map, key) - 1
593
if result + 1 < len(range_map):
594
# check the border condition, it may be in result + 1
595
if range_map[result + 1][0] == key[0]:
599
def _parsed_byte_index(self, offset):
600
"""Return the index of the entry immediately before offset.
602
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
603
there is one unparsed byte (the 11th, addressed as[10]). then:
604
asking for 0 will return 0
605
asking for 10 will return 0
606
asking for 11 will return 1
607
asking for 12 will return 1
610
return self._find_index(self._parsed_byte_map, key)
612
def _parsed_key_index(self, key):
613
"""Return the index of the entry immediately before key.
615
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
616
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
617
have been parsed, then:
618
asking for '' will return 0
619
asking for 'a' will return 0
620
asking for 'b' will return 1
621
asking for 'e' will return 1
623
search_key = (key, None)
624
return self._find_index(self._parsed_key_map, search_key)
626
def _is_parsed(self, offset):
627
"""Returns True if offset has been parsed."""
628
index = self._parsed_byte_index(offset)
629
if index == len(self._parsed_byte_map):
630
return offset < self._parsed_byte_map[index - 1][1]
631
start, end = self._parsed_byte_map[index]
632
return offset >= start and offset < end
634
def _iter_entries_from_total_buffer(self, keys):
635
"""Iterate over keys when the entire index is parsed."""
636
# Note: See the note in BTreeBuilder.iter_entries for why we don't use
637
# .intersection() here
639
keys = [key for key in keys if key in nodes]
640
if self.node_ref_lists:
642
value, node_refs = nodes[key]
643
yield self, key, value, node_refs
646
yield self, key, nodes[key]
648
def iter_entries(self, keys):
649
"""Iterate over keys within the index.
651
:param keys: An iterable providing the keys to be retrieved.
652
:return: An iterable as per iter_all_entries, but restricted to the
653
keys supplied. No additional keys will be returned, and every
654
key supplied that is in the index will be returned.
659
if self._size is None and self._nodes is None:
662
# We fit about 20 keys per minimum-read (4K), so if we are looking for
663
# more than 1/20th of the index its likely (assuming homogenous key
664
# spread) that we'll read the entire index. If we're going to do that,
665
# buffer the whole thing. A better analysis might take key spread into
666
# account - but B+Tree indices are better anyway.
667
# We could look at all data read, and use a threshold there, which will
668
# trigger on ancestry walks, but that is not yet fully mapped out.
669
if self._nodes is None and len(keys) * 20 > self.key_count():
671
if self._nodes is not None:
672
return self._iter_entries_from_total_buffer(keys)
674
return (result[1] for result in bisect_multi_bytes(
675
self._lookup_keys_via_location, self._size, keys))
677
def iter_entries_prefix(self, keys):
678
"""Iterate over keys within the index using prefix matching.
680
Prefix matching is applied within the tuple of a key, not to within
681
the bytestring of each key element. e.g. if you have the keys ('foo',
682
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
683
only the former key is returned.
685
WARNING: Note that this method currently causes a full index parse
686
unconditionally (which is reasonably appropriate as it is a means for
687
thunking many small indices into one larger one and still supplies
688
iter_all_entries at the thunk layer).
690
:param keys: An iterable providing the key prefixes to be retrieved.
691
Each key prefix takes the form of a tuple the length of a key, but
692
with the last N elements 'None' rather than a regular bytestring.
693
The first element cannot be 'None'.
694
:return: An iterable as per iter_all_entries, but restricted to the
695
keys with a matching prefix to those supplied. No additional keys
696
will be returned, and every match that is in the index will be
702
# load data - also finds key lengths
703
if self._nodes is None:
705
if self._key_length == 1:
709
raise errors.BadIndexKey(key)
710
if len(key) != self._key_length:
711
raise errors.BadIndexKey(key)
712
if self.node_ref_lists:
713
value, node_refs = self._nodes[key]
714
yield self, key, value, node_refs
716
yield self, key, self._nodes[key]
718
nodes_by_key = self._get_nodes_by_key()
722
raise errors.BadIndexKey(key)
723
if len(key) != self._key_length:
724
raise errors.BadIndexKey(key)
725
# find what it refers to:
726
key_dict = nodes_by_key
728
# find the subdict whose contents should be returned.
730
while len(elements) and elements[0] is not None:
731
key_dict = key_dict[elements[0]]
734
# a non-existant lookup.
739
key_dict = dicts.pop(-1)
740
# can't be empty or would not exist
741
item, value = key_dict.iteritems().next()
742
if type(value) == dict:
744
dicts.extend(key_dict.itervalues())
747
for value in key_dict.itervalues():
748
# each value is the key:value:node refs tuple
750
yield (self, ) + value
752
# the last thing looked up was a terminal element
753
yield (self, ) + key_dict
755
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
756
"""See BTreeIndex._find_ancestors."""
757
# The api can be implemented as a trivial overlay on top of
758
# iter_entries, it is not an efficient implementation, but it at least
762
for index, key, value, refs in self.iter_entries(keys):
763
parent_keys = refs[ref_list_num]
765
parent_map[key] = parent_keys
766
search_keys.update(parent_keys)
767
# Figure out what, if anything, was missing
768
missing_keys.update(set(keys).difference(found_keys))
769
search_keys = search_keys.difference(parent_map)
773
"""Return an estimate of the number of keys in this index.
775
For GraphIndex the estimate is exact.
777
if self._key_count is None:
778
self._read_and_parse([_HEADER_READV])
779
return self._key_count
781
def _lookup_keys_via_location(self, location_keys):
782
"""Public interface for implementing bisection.
784
If _buffer_all has been called, then all the data for the index is in
785
memory, and this method should not be called, as it uses a separate
786
cache because it cannot pre-resolve all indices, which buffer_all does
789
:param location_keys: A list of location(byte offset), key tuples.
790
:return: A list of (location_key, result) tuples as expected by
791
bzrlib.bisect_multi.bisect_multi_bytes.
793
# Possible improvements:
794
# - only bisect lookup each key once
795
# - sort the keys first, and use that to reduce the bisection window
797
# this progresses in three parts:
800
# attempt to answer the question from the now in memory data.
801
# build the readv request
802
# for each location, ask for 800 bytes - much more than rows we've seen
805
for location, key in location_keys:
806
# can we answer from cache?
807
if self._bisect_nodes and key in self._bisect_nodes:
808
# We have the key parsed.
810
index = self._parsed_key_index(key)
811
if (len(self._parsed_key_map) and
812
self._parsed_key_map[index][0] <= key and
813
(self._parsed_key_map[index][1] >= key or
814
# end of the file has been parsed
815
self._parsed_byte_map[index][1] == self._size)):
816
# the key has been parsed, so no lookup is needed even if its
819
# - if we have examined this part of the file already - yes
820
index = self._parsed_byte_index(location)
821
if (len(self._parsed_byte_map) and
822
self._parsed_byte_map[index][0] <= location and
823
self._parsed_byte_map[index][1] > location):
824
# the byte region has been parsed, so no read is needed.
827
if location + length > self._size:
828
length = self._size - location
829
# todo, trim out parsed locations.
831
readv_ranges.append((location, length))
832
# read the header if needed
833
if self._bisect_nodes is None:
834
readv_ranges.append(_HEADER_READV)
835
self._read_and_parse(readv_ranges)
837
if self._nodes is not None:
838
# _read_and_parse triggered a _buffer_all because we requested the
840
for location, key in location_keys:
841
if key not in self._nodes: # not present
842
result.append(((location, key), False))
843
elif self.node_ref_lists:
844
value, refs = self._nodes[key]
845
result.append(((location, key),
846
(self, key, value, refs)))
848
result.append(((location, key),
849
(self, key, self._nodes[key])))
852
# - figure out <, >, missing, present
853
# - result present references so we can return them.
854
# keys that we cannot answer until we resolve references
855
pending_references = []
856
pending_locations = set()
857
for location, key in location_keys:
858
# can we answer from cache?
859
if key in self._bisect_nodes:
860
# the key has been parsed, so no lookup is needed
861
if self.node_ref_lists:
862
# the references may not have been all parsed.
863
value, refs = self._bisect_nodes[key]
864
wanted_locations = []
865
for ref_list in refs:
867
if ref not in self._keys_by_offset:
868
wanted_locations.append(ref)
870
pending_locations.update(wanted_locations)
871
pending_references.append((location, key))
873
result.append(((location, key), (self, key,
874
value, self._resolve_references(refs))))
876
result.append(((location, key),
877
(self, key, self._bisect_nodes[key])))
880
# has the region the key should be in, been parsed?
881
index = self._parsed_key_index(key)
882
if (self._parsed_key_map[index][0] <= key and
883
(self._parsed_key_map[index][1] >= key or
884
# end of the file has been parsed
885
self._parsed_byte_map[index][1] == self._size)):
886
result.append(((location, key), False))
888
# no, is the key above or below the probed location:
889
# get the range of the probed & parsed location
890
index = self._parsed_byte_index(location)
891
# if the key is below the start of the range, its below
892
if key < self._parsed_key_map[index][0]:
896
result.append(((location, key), direction))
898
# lookup data to resolve references
899
for location in pending_locations:
901
if location + length > self._size:
902
length = self._size - location
903
# TODO: trim out parsed locations (e.g. if the 800 is into the
904
# parsed region trim it, and dont use the adjust_for_latency
907
readv_ranges.append((location, length))
908
self._read_and_parse(readv_ranges)
909
if self._nodes is not None:
910
# The _read_and_parse triggered a _buffer_all, grab the data and
912
for location, key in pending_references:
913
value, refs = self._nodes[key]
914
result.append(((location, key), (self, key, value, refs)))
916
for location, key in pending_references:
917
# answer key references we had to look-up-late.
918
value, refs = self._bisect_nodes[key]
919
result.append(((location, key), (self, key,
920
value, self._resolve_references(refs))))
923
def _parse_header_from_bytes(self, bytes):
924
"""Parse the header from a region of bytes.
926
:param bytes: The data to parse.
927
:return: An offset, data tuple such as readv yields, for the unparsed
928
data. (which may length 0).
930
signature = bytes[0:len(self._signature())]
931
if not signature == self._signature():
932
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
933
lines = bytes[len(self._signature()):].splitlines()
934
options_line = lines[0]
935
if not options_line.startswith(_OPTION_NODE_REFS):
936
raise errors.BadIndexOptions(self)
938
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
940
raise errors.BadIndexOptions(self)
941
options_line = lines[1]
942
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
943
raise errors.BadIndexOptions(self)
945
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
947
raise errors.BadIndexOptions(self)
948
options_line = lines[2]
949
if not options_line.startswith(_OPTION_LEN):
950
raise errors.BadIndexOptions(self)
952
self._key_count = int(options_line[len(_OPTION_LEN):])
954
raise errors.BadIndexOptions(self)
955
# calculate the bytes we have processed
956
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
958
self._parsed_bytes(0, None, header_end, None)
959
# setup parsing state
960
self._expected_elements = 3 + self._key_length
961
# raw data keyed by offset
962
self._keys_by_offset = {}
963
# keys with the value and node references
964
self._bisect_nodes = {}
965
return header_end, bytes[header_end:]
967
def _parse_region(self, offset, data):
968
"""Parse node data returned from a readv operation.
970
:param offset: The byte offset the data starts at.
971
:param data: The data to parse.
975
end = offset + len(data)
978
# Trivial test - if the current index's end is within the
979
# low-matching parsed range, we're done.
980
index = self._parsed_byte_index(high_parsed)
981
if end < self._parsed_byte_map[index][1]:
983
# print "[%d:%d]" % (offset, end), \
984
# self._parsed_byte_map[index:index + 2]
985
high_parsed, last_segment = self._parse_segment(
986
offset, data, end, index)
990
def _parse_segment(self, offset, data, end, index):
991
"""Parse one segment of data.
993
:param offset: Where 'data' begins in the file.
994
:param data: Some data to parse a segment of.
995
:param end: Where data ends
996
:param index: The current index into the parsed bytes map.
997
:return: True if the parsed segment is the last possible one in the
999
:return: high_parsed_byte, last_segment.
1000
high_parsed_byte is the location of the highest parsed byte in this
1001
segment, last_segment is True if the parsed segment is the last
1002
possible one in the data block.
1004
# default is to use all data
1006
# accomodate overlap with data before this.
1007
if offset < self._parsed_byte_map[index][1]:
1008
# overlaps the lower parsed region
1009
# skip the parsed data
1010
trim_start = self._parsed_byte_map[index][1] - offset
1011
# don't trim the start for \n
1012
start_adjacent = True
1013
elif offset == self._parsed_byte_map[index][1]:
1014
# abuts the lower parsed region
1017
# do not trim anything
1018
start_adjacent = True
1020
# does not overlap the lower parsed region
1023
# but trim the leading \n
1024
start_adjacent = False
1025
if end == self._size:
1026
# lines up to the end of all data:
1029
# do not strip to the last \n
1032
elif index + 1 == len(self._parsed_byte_map):
1033
# at the end of the parsed data
1036
# but strip to the last \n
1037
end_adjacent = False
1039
elif end == self._parsed_byte_map[index + 1][0]:
1040
# buts up against the next parsed region
1043
# do not strip to the last \n
1046
elif end > self._parsed_byte_map[index + 1][0]:
1047
# overlaps into the next parsed region
1048
# only consider the unparsed data
1049
trim_end = self._parsed_byte_map[index + 1][0] - offset
1050
# do not strip to the last \n as we know its an entire record
1052
last_segment = end < self._parsed_byte_map[index + 1][1]
1054
# does not overlap into the next region
1057
# but strip to the last \n
1058
end_adjacent = False
1060
# now find bytes to discard if needed
1061
if not start_adjacent:
1062
# work around python bug in rfind
1063
if trim_start is None:
1064
trim_start = data.find('\n') + 1
1066
trim_start = data.find('\n', trim_start) + 1
1067
if not (trim_start != 0):
1068
raise AssertionError('no \n was present')
1069
# print 'removing start', offset, trim_start, repr(data[:trim_start])
1070
if not end_adjacent:
1071
# work around python bug in rfind
1072
if trim_end is None:
1073
trim_end = data.rfind('\n') + 1
1075
trim_end = data.rfind('\n', None, trim_end) + 1
1076
if not (trim_end != 0):
1077
raise AssertionError('no \n was present')
1078
# print 'removing end', offset, trim_end, repr(data[trim_end:])
1079
# adjust offset and data to the parseable data.
1080
trimmed_data = data[trim_start:trim_end]
1081
if not (trimmed_data):
1082
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
1083
% (trim_start, trim_end, offset, offset + len(data)))
1085
offset += trim_start
1086
# print "parsing", repr(trimmed_data)
1087
# splitlines mangles the \r delimiters.. don't use it.
1088
lines = trimmed_data.split('\n')
1091
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1092
for key, value in nodes:
1093
self._bisect_nodes[key] = value
1094
self._parsed_bytes(offset, first_key,
1095
offset + len(trimmed_data), last_key)
1096
return offset + len(trimmed_data), last_segment
1098
def _parse_lines(self, lines, pos):
1105
# must be at the end
1107
if not (self._size == pos + 1):
1108
raise AssertionError("%s %s" % (self._size, pos))
1111
elements = line.split('\0')
1112
if len(elements) != self._expected_elements:
1113
raise errors.BadIndexData(self)
1114
# keys are tuples. Each element is a string that may occur many
1115
# times, so we intern them to save space. AB, RC, 200807
1116
key = tuple([intern(element) for element in elements[:self._key_length]])
1117
if first_key is None:
1119
absent, references, value = elements[-3:]
1121
for ref_string in references.split('\t'):
1122
ref_lists.append(tuple([
1123
int(ref) for ref in ref_string.split('\r') if ref
1125
ref_lists = tuple(ref_lists)
1126
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1127
pos += len(line) + 1 # +1 for the \n
1130
if self.node_ref_lists:
1131
node_value = (value, ref_lists)
1134
nodes.append((key, node_value))
1135
# print "parsed ", key
1136
return first_key, key, nodes, trailers
1138
def _parsed_bytes(self, start, start_key, end, end_key):
1139
"""Mark the bytes from start to end as parsed.
1141
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1144
:param start: The start of the parsed region.
1145
:param end: The end of the parsed region.
1147
index = self._parsed_byte_index(start)
1148
new_value = (start, end)
1149
new_key = (start_key, end_key)
1151
# first range parsed is always the beginning.
1152
self._parsed_byte_map.insert(index, new_value)
1153
self._parsed_key_map.insert(index, new_key)
1157
# extend lower region
1158
# extend higher region
1159
# combine two regions
1160
if (index + 1 < len(self._parsed_byte_map) and
1161
self._parsed_byte_map[index][1] == start and
1162
self._parsed_byte_map[index + 1][0] == end):
1163
# combine two regions
1164
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1165
self._parsed_byte_map[index + 1][1])
1166
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1167
self._parsed_key_map[index + 1][1])
1168
del self._parsed_byte_map[index + 1]
1169
del self._parsed_key_map[index + 1]
1170
elif self._parsed_byte_map[index][1] == start:
1171
# extend the lower entry
1172
self._parsed_byte_map[index] = (
1173
self._parsed_byte_map[index][0], end)
1174
self._parsed_key_map[index] = (
1175
self._parsed_key_map[index][0], end_key)
1176
elif (index + 1 < len(self._parsed_byte_map) and
1177
self._parsed_byte_map[index + 1][0] == end):
1178
# extend the higher entry
1179
self._parsed_byte_map[index + 1] = (
1180
start, self._parsed_byte_map[index + 1][1])
1181
self._parsed_key_map[index + 1] = (
1182
start_key, self._parsed_key_map[index + 1][1])
1185
self._parsed_byte_map.insert(index + 1, new_value)
1186
self._parsed_key_map.insert(index + 1, new_key)
1188
def _read_and_parse(self, readv_ranges):
1189
"""Read the ranges and parse the resulting data.
1191
:param readv_ranges: A prepared readv range list.
1193
if not readv_ranges:
1195
if self._nodes is None and self._bytes_read * 2 >= self._size:
1196
# We've already read more than 50% of the file and we are about to
1197
# request more data, just _buffer_all() and be done
1201
base_offset = self._base_offset
1202
if base_offset != 0:
1203
# Rewrite the ranges for the offset
1204
readv_ranges = [(start+base_offset, size)
1205
for start, size in readv_ranges]
1206
readv_data = self._transport.readv(self._name, readv_ranges, True,
1207
self._size + self._base_offset)
1209
for offset, data in readv_data:
1210
offset -= base_offset
1211
self._bytes_read += len(data)
1213
# transport.readv() expanded to extra data which isn't part of
1215
data = data[-offset:]
1217
if offset == 0 and len(data) == self._size:
1218
# We read the whole range, most likely because the
1219
# Transport upcast our readv ranges into one long request
1220
# for enough total data to grab the whole index.
1221
self._buffer_all(StringIO(data))
1223
if self._bisect_nodes is None:
1224
# this must be the start
1225
if not (offset == 0):
1226
raise AssertionError()
1227
offset, data = self._parse_header_from_bytes(data)
1228
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1229
self._parse_region(offset, data)
1231
def _signature(self):
1232
"""The file signature for this index type."""
1236
"""Validate that everything in the index can be accessed."""
1237
# iter_all validates completely at the moment, so just do that.
1238
for node in self.iter_all_entries():
1242
class CombinedGraphIndex(object):
1243
"""A GraphIndex made up from smaller GraphIndices.
1245
The backing indices must implement GraphIndex, and are presumed to be
1248
Queries against the combined index will be made against the first index,
1249
and then the second and so on. The order of indices can thus influence
1250
performance significantly. For example, if one index is on local disk and a
1251
second on a remote server, the local disk index should be before the other
1254
Also, queries tend to need results from the same indices as previous
1255
queries. So the indices will be reordered after every query to put the
1256
indices that had the result(s) of that query first (while otherwise
1257
preserving the relative ordering).
1260
def __init__(self, indices, reload_func=None):
1261
"""Create a CombinedGraphIndex backed by indices.
1263
:param indices: An ordered list of indices to query for data.
1264
:param reload_func: A function to call if we find we are missing an
1265
index. Should have the form reload_func() => True/False to indicate
1266
if reloading actually changed anything.
1268
self._indices = indices
1269
self._reload_func = reload_func
1270
# Sibling indices are other CombinedGraphIndex that we should call
1271
# _move_to_front_by_name on when we auto-reorder ourself.
1272
self._sibling_indices = []
1273
# A list of names that corresponds to the instances in self._indices,
1274
# so _index_names[0] is always the name for _indices[0], etc. Sibling
1275
# indices must all use the same set of names as each other.
1276
self._index_names = [None] * len(self._indices)
1280
self.__class__.__name__,
1281
', '.join(map(repr, self._indices)))
1283
def clear_cache(self):
1284
"""See GraphIndex.clear_cache()"""
1285
for index in self._indices:
1288
def get_parent_map(self, keys):
1289
"""See graph.StackedParentsProvider.get_parent_map"""
1290
search_keys = set(keys)
1291
if NULL_REVISION in search_keys:
1292
search_keys.discard(NULL_REVISION)
1293
found_parents = {NULL_REVISION:[]}
1296
for index, key, value, refs in self.iter_entries(search_keys):
1299
parents = (NULL_REVISION,)
1300
found_parents[key] = parents
1301
return found_parents
1303
has_key = _has_key_from_parent_map
1305
def insert_index(self, pos, index, name=None):
1306
"""Insert a new index in the list of indices to query.
1308
:param pos: The position to insert the index.
1309
:param index: The index to insert.
1310
:param name: a name for this index, e.g. a pack name. These names can
1311
be used to reflect index reorderings to related CombinedGraphIndex
1312
instances that use the same names. (see set_sibling_indices)
1314
self._indices.insert(pos, index)
1315
self._index_names.insert(pos, name)
1317
def iter_all_entries(self):
1318
"""Iterate over all keys within the index
1320
Duplicate keys across child indices are presumed to have the same
1321
value and are only reported once.
1323
:return: An iterable of (index, key, reference_lists, value).
1324
There is no defined order for the result iteration - it will be in
1325
the most efficient order for the index.
1330
for index in self._indices:
1331
for node in index.iter_all_entries():
1332
if node[1] not in seen_keys:
1334
seen_keys.add(node[1])
1336
except errors.NoSuchFile:
1337
self._reload_or_raise()
1339
def iter_entries(self, keys):
1340
"""Iterate over keys within the index.
1342
Duplicate keys across child indices are presumed to have the same
1343
value and are only reported once.
1345
:param keys: An iterable providing the keys to be retrieved.
1346
:return: An iterable of (index, key, reference_lists, value). There is
1347
no defined order for the result iteration - it will be in the most
1348
efficient order for the index.
1354
for index in self._indices:
1358
for node in index.iter_entries(keys):
1359
keys.remove(node[1])
1363
hit_indices.append(index)
1365
except errors.NoSuchFile:
1366
self._reload_or_raise()
1367
self._move_to_front(hit_indices)
1369
def iter_entries_prefix(self, keys):
1370
"""Iterate over keys within the index using prefix matching.
1372
Duplicate keys across child indices are presumed to have the same
1373
value and are only reported once.
1375
Prefix matching is applied within the tuple of a key, not to within
1376
the bytestring of each key element. e.g. if you have the keys ('foo',
1377
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1378
only the former key is returned.
1380
:param keys: An iterable providing the key prefixes to be retrieved.
1381
Each key prefix takes the form of a tuple the length of a key, but
1382
with the last N elements 'None' rather than a regular bytestring.
1383
The first element cannot be 'None'.
1384
:return: An iterable as per iter_all_entries, but restricted to the
1385
keys with a matching prefix to those supplied. No additional keys
1386
will be returned, and every match that is in the index will be
1396
for index in self._indices:
1398
for node in index.iter_entries_prefix(keys):
1399
if node[1] in seen_keys:
1401
seen_keys.add(node[1])
1405
hit_indices.append(index)
1407
except errors.NoSuchFile:
1408
self._reload_or_raise()
1409
self._move_to_front(hit_indices)
1411
def _move_to_front(self, hit_indices):
1412
"""Rearrange self._indices so that hit_indices are first.
1414
Order is maintained as much as possible, e.g. the first unhit index
1415
will be the first index in _indices after the hit_indices, and the
1416
hit_indices will be present in exactly the order they are passed to
1419
_move_to_front propagates to all objects in self._sibling_indices by
1420
calling _move_to_front_by_name.
1422
if self._indices[:len(hit_indices)] == hit_indices:
1423
# The 'hit_indices' are already at the front (and in the same
1424
# order), no need to re-order
1426
hit_names = self._move_to_front_by_index(hit_indices)
1427
for sibling_idx in self._sibling_indices:
1428
sibling_idx._move_to_front_by_name(hit_names)
1430
def _move_to_front_by_index(self, hit_indices):
1431
"""Core logic for _move_to_front.
1433
Returns a list of names corresponding to the hit_indices param.
1435
indices_info = zip(self._index_names, self._indices)
1436
if 'index' in debug.debug_flags:
1437
mutter('CombinedGraphIndex reordering: currently %r, promoting %r',
1438
indices_info, hit_indices)
1441
new_hit_indices = []
1444
for offset, (name, idx) in enumerate(indices_info):
1445
if idx in hit_indices:
1446
hit_names.append(name)
1447
new_hit_indices.append(idx)
1448
if len(new_hit_indices) == len(hit_indices):
1449
# We've found all of the hit entries, everything else is
1451
unhit_names.extend(self._index_names[offset+1:])
1452
unhit_indices.extend(self._indices[offset+1:])
1455
unhit_names.append(name)
1456
unhit_indices.append(idx)
1458
self._indices = new_hit_indices + unhit_indices
1459
self._index_names = hit_names + unhit_names
1460
if 'index' in debug.debug_flags:
1461
mutter('CombinedGraphIndex reordered: %r', self._indices)
1464
def _move_to_front_by_name(self, hit_names):
1465
"""Moves indices named by 'hit_names' to front of the search order, as
1466
described in _move_to_front.
1468
# Translate names to index instances, and then call
1469
# _move_to_front_by_index.
1470
indices_info = zip(self._index_names, self._indices)
1472
for name, idx in indices_info:
1473
if name in hit_names:
1474
hit_indices.append(idx)
1475
self._move_to_front_by_index(hit_indices)
1477
def find_ancestry(self, keys, ref_list_num):
1478
"""Find the complete ancestry for the given set of keys.
1480
Note that this is a whole-ancestry request, so it should be used
1483
:param keys: An iterable of keys to look for
1484
:param ref_list_num: The reference list which references the parents
1486
:return: (parent_map, missing_keys)
1488
# XXX: make this call _move_to_front?
1489
missing_keys = set()
1491
keys_to_lookup = set(keys)
1493
while keys_to_lookup:
1494
# keys that *all* indexes claim are missing, stop searching them
1496
all_index_missing = None
1497
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1498
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1500
# len(missing_keys))
1501
for index_idx, index in enumerate(self._indices):
1502
# TODO: we should probably be doing something with
1503
# 'missing_keys' since we've already determined that
1504
# those revisions have not been found anywhere
1505
index_missing_keys = set()
1506
# Find all of the ancestry we can from this index
1507
# keep looking until the search_keys set is empty, which means
1508
# things we didn't find should be in index_missing_keys
1509
search_keys = keys_to_lookup
1511
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1512
# index_idx, len(search_keys),
1513
# len(parent_map), len(index_missing_keys))
1516
# TODO: ref_list_num should really be a parameter, since
1517
# CombinedGraphIndex does not know what the ref lists
1519
search_keys = index._find_ancestors(search_keys,
1520
ref_list_num, parent_map, index_missing_keys)
1521
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1522
# sub_generation, len(search_keys),
1523
# len(parent_map), len(index_missing_keys))
1524
# Now set whatever was missing to be searched in the next index
1525
keys_to_lookup = index_missing_keys
1526
if all_index_missing is None:
1527
all_index_missing = set(index_missing_keys)
1529
all_index_missing.intersection_update(index_missing_keys)
1530
if not keys_to_lookup:
1532
if all_index_missing is None:
1533
# There were no indexes, so all search keys are 'missing'
1534
missing_keys.update(keys_to_lookup)
1535
keys_to_lookup = None
1537
missing_keys.update(all_index_missing)
1538
keys_to_lookup.difference_update(all_index_missing)
1539
return parent_map, missing_keys
1541
def key_count(self):
1542
"""Return an estimate of the number of keys in this index.
1544
For CombinedGraphIndex this is approximated by the sum of the keys of
1545
the child indices. As child indices may have duplicate keys this can
1546
have a maximum error of the number of child indices * largest number of
1551
return sum((index.key_count() for index in self._indices), 0)
1552
except errors.NoSuchFile:
1553
self._reload_or_raise()
1555
missing_keys = _missing_keys_from_parent_map
1557
def _reload_or_raise(self):
1558
"""We just got a NoSuchFile exception.
1560
Try to reload the indices, if it fails, just raise the current
1563
if self._reload_func is None:
1565
exc_type, exc_value, exc_traceback = sys.exc_info()
1566
trace.mutter('Trying to reload after getting exception: %s',
1568
if not self._reload_func():
1569
# We tried to reload, but nothing changed, so we fail anyway
1570
trace.mutter('_reload_func indicated nothing has changed.'
1571
' Raising original exception.')
1572
raise exc_type, exc_value, exc_traceback
1574
def set_sibling_indices(self, sibling_combined_graph_indices):
1575
"""Set the CombinedGraphIndex objects to reorder after reordering self.
1577
self._sibling_indices = sibling_combined_graph_indices
1580
"""Validate that everything in the index can be accessed."""
1583
for index in self._indices:
1586
except errors.NoSuchFile:
1587
self._reload_or_raise()
1590
class InMemoryGraphIndex(GraphIndexBuilder):
1591
"""A GraphIndex which operates entirely out of memory and is mutable.
1593
This is designed to allow the accumulation of GraphIndex entries during a
1594
single write operation, where the accumulated entries need to be immediately
1595
available - for example via a CombinedGraphIndex.
1598
def add_nodes(self, nodes):
1599
"""Add nodes to the index.
1601
:param nodes: An iterable of (key, node_refs, value) entries to add.
1603
if self.reference_lists:
1604
for (key, value, node_refs) in nodes:
1605
self.add_node(key, value, node_refs)
1607
for (key, value) in nodes:
1608
self.add_node(key, value)
1610
def iter_all_entries(self):
1611
"""Iterate over all keys within the index
1613
:return: An iterable of (index, key, reference_lists, value). There is no
1614
defined order for the result iteration - it will be in the most
1615
efficient order for the index (in this case dictionary hash order).
1617
if 'evil' in debug.debug_flags:
1618
trace.mutter_callsite(3,
1619
"iter_all_entries scales with size of history.")
1620
if self.reference_lists:
1621
for key, (absent, references, value) in self._nodes.iteritems():
1623
yield self, key, value, references
1625
for key, (absent, references, value) in self._nodes.iteritems():
1627
yield self, key, value
1629
def iter_entries(self, keys):
1630
"""Iterate over keys within the index.
1632
:param keys: An iterable providing the keys to be retrieved.
1633
:return: An iterable of (index, key, value, reference_lists). There is no
1634
defined order for the result iteration - it will be in the most
1635
efficient order for the index (keys iteration order in this case).
1637
# Note: See BTreeBuilder.iter_entries for an explanation of why we
1638
# aren't using set().intersection() here
1640
keys = [key for key in keys if key in nodes]
1641
if self.reference_lists:
1645
yield self, key, node[2], node[1]
1650
yield self, key, node[2]
1652
def iter_entries_prefix(self, keys):
1653
"""Iterate over keys within the index using prefix matching.
1655
Prefix matching is applied within the tuple of a key, not to within
1656
the bytestring of each key element. e.g. if you have the keys ('foo',
1657
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1658
only the former key is returned.
1660
:param keys: An iterable providing the key prefixes to be retrieved.
1661
Each key prefix takes the form of a tuple the length of a key, but
1662
with the last N elements 'None' rather than a regular bytestring.
1663
The first element cannot be 'None'.
1664
:return: An iterable as per iter_all_entries, but restricted to the
1665
keys with a matching prefix to those supplied. No additional keys
1666
will be returned, and every match that is in the index will be
1669
# XXX: To much duplication with the GraphIndex class; consider finding
1670
# a good place to pull out the actual common logic.
1674
if self._key_length == 1:
1678
raise errors.BadIndexKey(key)
1679
if len(key) != self._key_length:
1680
raise errors.BadIndexKey(key)
1681
node = self._nodes[key]
1684
if self.reference_lists:
1685
yield self, key, node[2], node[1]
1687
yield self, key, node[2]
1689
nodes_by_key = self._get_nodes_by_key()
1693
raise errors.BadIndexKey(key)
1694
if len(key) != self._key_length:
1695
raise errors.BadIndexKey(key)
1696
# find what it refers to:
1697
key_dict = nodes_by_key
1698
elements = list(key)
1699
# find the subdict to return
1701
while len(elements) and elements[0] is not None:
1702
key_dict = key_dict[elements[0]]
1705
# a non-existant lookup.
1710
key_dict = dicts.pop(-1)
1711
# can't be empty or would not exist
1712
item, value = key_dict.iteritems().next()
1713
if type(value) == dict:
1715
dicts.extend(key_dict.itervalues())
1718
for value in key_dict.itervalues():
1719
yield (self, ) + value
1721
yield (self, ) + key_dict
1723
def key_count(self):
1724
"""Return an estimate of the number of keys in this index.
1726
For InMemoryGraphIndex the estimate is exact.
1728
return len(self._nodes) - len(self._absent_keys)
1731
"""In memory index's have no known corruption at the moment."""
1734
class GraphIndexPrefixAdapter(object):
1735
"""An adapter between GraphIndex with different key lengths.
1737
Queries against this will emit queries against the adapted Graph with the
1738
prefix added, queries for all items use iter_entries_prefix. The returned
1739
nodes will have their keys and node references adjusted to remove the
1740
prefix. Finally, an add_nodes_callback can be supplied - when called the
1741
nodes and references being added will have prefix prepended.
1744
def __init__(self, adapted, prefix, missing_key_length,
1745
add_nodes_callback=None):
1746
"""Construct an adapter against adapted with prefix."""
1747
self.adapted = adapted
1748
self.prefix_key = prefix + (None,)*missing_key_length
1749
self.prefix = prefix
1750
self.prefix_len = len(prefix)
1751
self.add_nodes_callback = add_nodes_callback
1753
def add_nodes(self, nodes):
1754
"""Add nodes to the index.
1756
:param nodes: An iterable of (key, node_refs, value) entries to add.
1758
# save nodes in case its an iterator
1759
nodes = tuple(nodes)
1760
translated_nodes = []
1762
# Add prefix_key to each reference node_refs is a tuple of tuples,
1763
# so split it apart, and add prefix_key to the internal reference
1764
for (key, value, node_refs) in nodes:
1765
adjusted_references = (
1766
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1767
for ref_list in node_refs))
1768
translated_nodes.append((self.prefix + key, value,
1769
adjusted_references))
1771
# XXX: TODO add an explicit interface for getting the reference list
1772
# status, to handle this bit of user-friendliness in the API more
1774
for (key, value) in nodes:
1775
translated_nodes.append((self.prefix + key, value))
1776
self.add_nodes_callback(translated_nodes)
1778
def add_node(self, key, value, references=()):
1779
"""Add a node to the index.
1781
:param key: The key. keys are non-empty tuples containing
1782
as many whitespace-free utf8 bytestrings as the key length
1783
defined for this index.
1784
:param references: An iterable of iterables of keys. Each is a
1785
reference to another key.
1786
:param value: The value to associate with the key. It may be any
1787
bytes as long as it does not contain \0 or \n.
1789
self.add_nodes(((key, value, references), ))
1791
def _strip_prefix(self, an_iter):
1792
"""Strip prefix data from nodes and return it."""
1793
for node in an_iter:
1795
if node[1][:self.prefix_len] != self.prefix:
1796
raise errors.BadIndexData(self)
1797
for ref_list in node[3]:
1798
for ref_node in ref_list:
1799
if ref_node[:self.prefix_len] != self.prefix:
1800
raise errors.BadIndexData(self)
1801
yield node[0], node[1][self.prefix_len:], node[2], (
1802
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1803
for ref_list in node[3]))
1805
def iter_all_entries(self):
1806
"""Iterate over all keys within the index
1808
iter_all_entries is implemented against the adapted index using
1809
iter_entries_prefix.
1811
:return: An iterable of (index, key, reference_lists, value). There is no
1812
defined order for the result iteration - it will be in the most
1813
efficient order for the index (in this case dictionary hash order).
1815
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1817
def iter_entries(self, keys):
1818
"""Iterate over keys within the index.
1820
:param keys: An iterable providing the keys to be retrieved.
1821
:return: An iterable of (index, key, value, reference_lists). There is no
1822
defined order for the result iteration - it will be in the most
1823
efficient order for the index (keys iteration order in this case).
1825
return self._strip_prefix(self.adapted.iter_entries(
1826
self.prefix + key for key in keys))
1828
def iter_entries_prefix(self, keys):
1829
"""Iterate over keys within the index using prefix matching.
1831
Prefix matching is applied within the tuple of a key, not to within
1832
the bytestring of each key element. e.g. if you have the keys ('foo',
1833
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1834
only the former key is returned.
1836
:param keys: An iterable providing the key prefixes to be retrieved.
1837
Each key prefix takes the form of a tuple the length of a key, but
1838
with the last N elements 'None' rather than a regular bytestring.
1839
The first element cannot be 'None'.
1840
:return: An iterable as per iter_all_entries, but restricted to the
1841
keys with a matching prefix to those supplied. No additional keys
1842
will be returned, and every match that is in the index will be
1845
return self._strip_prefix(self.adapted.iter_entries_prefix(
1846
self.prefix + key for key in keys))
1848
def key_count(self):
1849
"""Return an estimate of the number of keys in this index.
1851
For GraphIndexPrefixAdapter this is relatively expensive - key
1852
iteration with the prefix is done.
1854
return len(list(self.iter_all_entries()))
1857
"""Call the adapted's validate."""
1858
self.adapted.validate()
added
removed
bzrlib/index.py
