1
# Copyright (C) 2007, 2008, 2009 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
"""Indexing facilities."""
23
'GraphIndexPrefixAdapter',
27
from bisect import bisect_right
28
from cStringIO import StringIO
32
from bzrlib.lazy_import import lazy_import
33
lazy_import(globals(), """
34
from bzrlib import trace
35
from bzrlib.bisect_multi import bisect_multi_bytes
36
from bzrlib.revision import NULL_REVISION
37
from bzrlib.trace import mutter
43
from bzrlib.static_tuple import StaticTuple
45
_HEADER_READV = (0, 200)
46
_OPTION_KEY_ELEMENTS = "key_elements="
48
_OPTION_NODE_REFS = "node_ref_lists="
49
_SIGNATURE = "Bazaar Graph Index 1\n"
52
_whitespace_re = re.compile('[\t\n\x0b\x0c\r\x00 ]')
53
_newline_null_re = re.compile('[\n\0]')
56
def _has_key_from_parent_map(self, key):
57
"""Check if this index has one key.
59
If it's possible to check for multiple keys at once through
60
calling get_parent_map that should be faster.
62
return (key in self.get_parent_map([key]))
65
def _missing_keys_from_parent_map(self, keys):
66
return set(keys) - set(self.get_parent_map(keys))
69
class GraphIndexBuilder(object):
70
"""A builder that can build a GraphIndex.
72
The resulting graph has the structure:
74
_SIGNATURE OPTIONS NODES NEWLINE
75
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
76
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
78
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
79
KEY := Not-whitespace-utf8
81
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
82
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
83
REFERENCE := DIGITS ; digits is the byte offset in the index of the
85
VALUE := no-newline-no-null-bytes
88
def __init__(self, reference_lists=0, key_elements=1):
89
"""Create a GraphIndex builder.
91
:param reference_lists: The number of node references lists for each
93
:param key_elements: The number of bytestrings in each key.
95
self.reference_lists = reference_lists
97
# A dict of {key: (absent, ref_lists, value)}
99
self._nodes_by_key = None
100
self._key_length = key_elements
101
self._optimize_for_size = False
102
self._combine_backing_indices = True
104
def _check_key(self, key):
105
"""Raise BadIndexKey if key is not a valid key for this index."""
106
if type(key) not in (tuple, StaticTuple):
107
raise errors.BadIndexKey(key)
108
if self._key_length != len(key):
109
raise errors.BadIndexKey(key)
111
if not element or _whitespace_re.search(element) is not None:
112
raise errors.BadIndexKey(element)
114
def _external_references(self):
115
"""Return references that are not present in this index.
119
# TODO: JAM 2008-11-21 This makes an assumption about how the reference
120
# lists are used. It is currently correct for pack-0.92 through
121
# 1.9, which use the node references (3rd column) second
122
# reference list as the compression parent. Perhaps this should
123
# be moved into something higher up the stack, since it
124
# makes assumptions about how the index is used.
125
if self.reference_lists > 1:
126
for node in self.iter_all_entries():
128
refs.update(node[3][1])
131
# If reference_lists == 0 there can be no external references, and
132
# if reference_lists == 1, then there isn't a place to store the
136
def _get_nodes_by_key(self):
137
if self._nodes_by_key is None:
139
if self.reference_lists:
140
for key, (absent, references, value) in self._nodes.iteritems():
143
key_dict = nodes_by_key
144
for subkey in key[:-1]:
145
key_dict = key_dict.setdefault(subkey, {})
146
key_dict[key[-1]] = key, value, references
148
for key, (absent, references, value) in self._nodes.iteritems():
151
key_dict = nodes_by_key
152
for subkey in key[:-1]:
153
key_dict = key_dict.setdefault(subkey, {})
154
key_dict[key[-1]] = key, value
155
self._nodes_by_key = nodes_by_key
156
return self._nodes_by_key
158
def _update_nodes_by_key(self, key, value, node_refs):
159
"""Update the _nodes_by_key dict with a new key.
161
For a key of (foo, bar, baz) create
162
_nodes_by_key[foo][bar][baz] = key_value
164
if self._nodes_by_key is None:
166
key_dict = self._nodes_by_key
167
if self.reference_lists:
168
key_value = key, value, node_refs
170
key_value = key, value
171
for subkey in key[:-1]:
172
key_dict = key_dict.setdefault(subkey, {})
173
key_dict[key[-1]] = key_value
175
def _check_key_ref_value(self, key, references, value):
176
"""Check that 'key' and 'references' are all valid.
178
:param key: A key tuple. Must conform to the key interface (be a tuple,
179
be of the right length, not have any whitespace or nulls in any key
181
:param references: An iterable of reference lists. Something like
182
[[(ref, key)], [(ref, key), (other, key)]]
183
:param value: The value associate with this key. Must not contain
184
newlines or null characters.
185
:return: (node_refs, absent_references)
186
node_refs basically a packed form of 'references' where all
188
absent_references reference keys that are not in self._nodes.
189
This may contain duplicates if the same key is
190
referenced in multiple lists.
193
if _newline_null_re.search(value) is not None:
194
raise errors.BadIndexValue(value)
195
if len(references) != self.reference_lists:
196
raise errors.BadIndexValue(references)
198
absent_references = []
199
for reference_list in references:
200
for reference in reference_list:
201
# If reference *is* in self._nodes, then we know it has already
203
if reference not in self._nodes:
204
self._check_key(reference)
205
absent_references.append(reference)
207
node_refs.append(tuple(reference_list))
209
return tuple(node_refs), absent_references
211
def add_node(self, key, value, references=()):
212
"""Add a node to the index.
214
:param key: The key. keys are non-empty tuples containing
215
as many whitespace-free utf8 bytestrings as the key length
216
defined for this index.
217
:param references: An iterable of iterables of keys. Each is a
218
reference to another key.
219
:param value: The value to associate with the key. It may be any
220
bytes as long as it does not contain \0 or \n.
223
absent_references) = self._check_key_ref_value(key, references, value)
224
if key in self._nodes and self._nodes[key][0] != 'a':
225
raise errors.BadIndexDuplicateKey(key, self)
226
for reference in absent_references:
227
# There may be duplicates, but I don't think it is worth worrying
229
self._nodes[reference] = ('a', (), '')
230
self._nodes[key] = ('', node_refs, value)
232
if self._nodes_by_key is not None and self._key_length > 1:
233
self._update_nodes_by_key(key, value, node_refs)
237
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
238
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
239
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
240
prefix_length = sum(len(x) for x in lines)
241
# references are byte offsets. To avoid having to do nasty
242
# polynomial work to resolve offsets (references to later in the
243
# file cannot be determined until all the inbetween references have
244
# been calculated too) we pad the offsets with 0's to make them be
245
# of consistent length. Using binary offsets would break the trivial
247
# to calculate the width of zero's needed we do three passes:
248
# one to gather all the non-reference data and the number of references.
249
# one to pad all the data with reference-length and determine entry
253
# forward sorted by key. In future we may consider topological sorting,
254
# at the cost of table scans for direct lookup, or a second index for
256
nodes = sorted(self._nodes.items())
257
# if we do not prepass, we don't know how long it will be up front.
258
expected_bytes = None
259
# we only need to pre-pass if we have reference lists at all.
260
if self.reference_lists:
262
non_ref_bytes = prefix_length
264
# TODO use simple multiplication for the constants in this loop.
265
for key, (absent, references, value) in nodes:
266
# record the offset known *so far* for this key:
267
# the non reference bytes to date, and the total references to
268
# date - saves reaccumulating on the second pass
269
key_offset_info.append((key, non_ref_bytes, total_references))
270
# key is literal, value is literal, there are 3 null's, 1 NL
271
# key is variable length tuple, \x00 between elements
272
non_ref_bytes += sum(len(element) for element in key)
273
if self._key_length > 1:
274
non_ref_bytes += self._key_length - 1
275
# value is literal bytes, there are 3 null's, 1 NL.
276
non_ref_bytes += len(value) + 3 + 1
277
# one byte for absent if set.
280
elif self.reference_lists:
281
# (ref_lists -1) tabs
282
non_ref_bytes += self.reference_lists - 1
283
# (ref-1 cr's per ref_list)
284
for ref_list in references:
285
# how many references across the whole file?
286
total_references += len(ref_list)
287
# accrue reference separators
289
non_ref_bytes += len(ref_list) - 1
290
# how many digits are needed to represent the total byte count?
292
possible_total_bytes = non_ref_bytes + total_references*digits
293
while 10 ** digits < possible_total_bytes:
295
possible_total_bytes = non_ref_bytes + total_references*digits
296
expected_bytes = possible_total_bytes + 1 # terminating newline
297
# resolve key addresses.
299
for key, non_ref_bytes, total_references in key_offset_info:
300
key_addresses[key] = non_ref_bytes + total_references*digits
302
format_string = '%%0%sd' % digits
303
for key, (absent, references, value) in nodes:
304
flattened_references = []
305
for ref_list in references:
307
for reference in ref_list:
308
ref_addresses.append(format_string % key_addresses[reference])
309
flattened_references.append('\r'.join(ref_addresses))
310
string_key = '\x00'.join(key)
311
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
312
'\t'.join(flattened_references), value))
314
result = StringIO(''.join(lines))
315
if expected_bytes and len(result.getvalue()) != expected_bytes:
316
raise errors.BzrError('Failed index creation. Internal error:'
317
' mismatched output length and expected length: %d %d' %
318
(len(result.getvalue()), expected_bytes))
321
def set_optimize(self, for_size=None, combine_backing_indices=None):
322
"""Change how the builder tries to optimize the result.
324
:param for_size: Tell the builder to try and make the index as small as
326
:param combine_backing_indices: If the builder spills to disk to save
327
memory, should the on-disk indices be combined. Set to True if you
328
are going to be probing the index, but to False if you are not. (If
329
you are not querying, then the time spent combining is wasted.)
332
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
334
if for_size is not None:
335
self._optimize_for_size = for_size
336
if combine_backing_indices is not None:
337
self._combine_backing_indices = combine_backing_indices
339
def find_ancestry(self, keys, ref_list_num):
340
"""See CombinedGraphIndex.find_ancestry()"""
346
for _, key, value, ref_lists in self.iter_entries(pending):
347
parent_keys = ref_lists[ref_list_num]
348
parent_map[key] = parent_keys
349
next_pending.update([p for p in parent_keys if p not in
351
missing_keys.update(pending.difference(parent_map))
352
pending = next_pending
353
return parent_map, missing_keys
356
class GraphIndex(object):
357
"""An index for data with embedded graphs.
359
The index maps keys to a list of key reference lists, and a value.
360
Each node has the same number of key reference lists. Each key reference
361
list can be empty or an arbitrary length. The value is an opaque NULL
362
terminated string without any newlines. The storage of the index is
363
hidden in the interface: keys and key references are always tuples of
364
bytestrings, never the internal representation (e.g. dictionary offsets).
366
It is presumed that the index will not be mutated - it is static data.
368
Successive iter_all_entries calls will read the entire index each time.
369
Additionally, iter_entries calls will read the index linearly until the
370
desired keys are found. XXX: This must be fixed before the index is
371
suitable for production use. :XXX
374
def __init__(self, transport, name, size, unlimited_cache=False):
375
"""Open an index called name on transport.
377
:param transport: A bzrlib.transport.Transport.
378
:param name: A path to provide to transport API calls.
379
:param size: The size of the index in bytes. This is used for bisection
380
logic to perform partial index reads. While the size could be
381
obtained by statting the file this introduced an additional round
382
trip as well as requiring stat'able transports, both of which are
383
avoided by having it supplied. If size is None, then bisection
384
support will be disabled and accessing the index will just stream
387
self._transport = transport
389
# Becomes a dict of key:(value, reference-list-byte-locations) used by
390
# the bisection interface to store parsed but not resolved keys.
391
self._bisect_nodes = None
392
# Becomes a dict of key:(value, reference-list-keys) which are ready to
393
# be returned directly to callers.
395
# a sorted list of slice-addresses for the parsed bytes of the file.
396
# e.g. (0,1) would mean that byte 0 is parsed.
397
self._parsed_byte_map = []
398
# a sorted list of keys matching each slice address for parsed bytes
399
# e.g. (None, 'foo@bar') would mean that the first byte contained no
400
# key, and the end byte of the slice is the of the data for 'foo@bar'
401
self._parsed_key_map = []
402
self._key_count = None
403
self._keys_by_offset = None
404
self._nodes_by_key = None
406
# The number of bytes we've read so far in trying to process this file
409
def __eq__(self, other):
410
"""Equal when self and other were created with the same parameters."""
412
type(self) == type(other) and
413
self._transport == other._transport and
414
self._name == other._name and
415
self._size == other._size)
417
def __ne__(self, other):
418
return not self.__eq__(other)
421
return "%s(%r)" % (self.__class__.__name__,
422
self._transport.abspath(self._name))
424
def _buffer_all(self, stream=None):
425
"""Buffer all the index data.
427
Mutates self._nodes and self.keys_by_offset.
429
if self._nodes is not None:
430
# We already did this
432
if 'index' in debug.debug_flags:
433
mutter('Reading entire index %s', self._transport.abspath(self._name))
435
stream = self._transport.get(self._name)
436
self._read_prefix(stream)
437
self._expected_elements = 3 + self._key_length
439
# raw data keyed by offset
440
self._keys_by_offset = {}
441
# ready-to-return key:value or key:value, node_ref_lists
443
self._nodes_by_key = None
446
lines = stream.read().split('\n')
448
_, _, _, trailers = self._parse_lines(lines, pos)
449
for key, absent, references, value in self._keys_by_offset.itervalues():
452
# resolve references:
453
if self.node_ref_lists:
454
node_value = (value, self._resolve_references(references))
457
self._nodes[key] = node_value
458
# cache the keys for quick set intersections
459
self._keys = set(self._nodes)
461
# there must be one line - the empty trailer line.
462
raise errors.BadIndexData(self)
464
def clear_cache(self):
465
"""Clear out any cached/memoized values.
467
This can be called at any time, but generally it is used when we have
468
extracted some information, but don't expect to be requesting any more
472
def external_references(self, ref_list_num):
473
"""Return references that are not present in this index.
476
if ref_list_num + 1 > self.node_ref_lists:
477
raise ValueError('No ref list %d, index has %d ref lists'
478
% (ref_list_num, self.node_ref_lists))
480
for key, (value, ref_lists) in self._nodes.iteritems():
481
ref_list = ref_lists[ref_list_num]
482
refs.update(ref_list)
483
return refs - self._keys
485
def _get_nodes_by_key(self):
486
if self._nodes_by_key is None:
488
if self.node_ref_lists:
489
for key, (value, references) in self._nodes.iteritems():
490
key_dict = nodes_by_key
491
for subkey in key[:-1]:
492
key_dict = key_dict.setdefault(subkey, {})
493
key_dict[key[-1]] = key, value, references
495
for key, value in self._nodes.iteritems():
496
key_dict = nodes_by_key
497
for subkey in key[:-1]:
498
key_dict = key_dict.setdefault(subkey, {})
499
key_dict[key[-1]] = key, value
500
self._nodes_by_key = nodes_by_key
501
return self._nodes_by_key
503
def iter_all_entries(self):
504
"""Iterate over all keys within the index.
506
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
507
The former tuple is used when there are no reference lists in the
508
index, making the API compatible with simple key:value index types.
509
There is no defined order for the result iteration - it will be in
510
the most efficient order for the index.
512
if 'evil' in debug.debug_flags:
513
trace.mutter_callsite(3,
514
"iter_all_entries scales with size of history.")
515
if self._nodes is None:
517
if self.node_ref_lists:
518
for key, (value, node_ref_lists) in self._nodes.iteritems():
519
yield self, key, value, node_ref_lists
521
for key, value in self._nodes.iteritems():
522
yield self, key, value
524
def _read_prefix(self, stream):
525
signature = stream.read(len(self._signature()))
526
if not signature == self._signature():
527
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
528
options_line = stream.readline()
529
if not options_line.startswith(_OPTION_NODE_REFS):
530
raise errors.BadIndexOptions(self)
532
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
534
raise errors.BadIndexOptions(self)
535
options_line = stream.readline()
536
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
537
raise errors.BadIndexOptions(self)
539
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
541
raise errors.BadIndexOptions(self)
542
options_line = stream.readline()
543
if not options_line.startswith(_OPTION_LEN):
544
raise errors.BadIndexOptions(self)
546
self._key_count = int(options_line[len(_OPTION_LEN):-1])
548
raise errors.BadIndexOptions(self)
550
def _resolve_references(self, references):
551
"""Return the resolved key references for references.
553
References are resolved by looking up the location of the key in the
554
_keys_by_offset map and substituting the key name, preserving ordering.
556
:param references: An iterable of iterables of key locations. e.g.
558
:return: A tuple of tuples of keys.
561
for ref_list in references:
562
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
563
return tuple(node_refs)
565
def _find_index(self, range_map, key):
566
"""Helper for the _parsed_*_index calls.
568
Given a range map - [(start, end), ...], finds the index of the range
569
in the map for key if it is in the map, and if it is not there, the
570
immediately preceeding range in the map.
572
result = bisect_right(range_map, key) - 1
573
if result + 1 < len(range_map):
574
# check the border condition, it may be in result + 1
575
if range_map[result + 1][0] == key[0]:
579
def _parsed_byte_index(self, offset):
580
"""Return the index of the entry immediately before offset.
582
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
583
there is one unparsed byte (the 11th, addressed as[10]). then:
584
asking for 0 will return 0
585
asking for 10 will return 0
586
asking for 11 will return 1
587
asking for 12 will return 1
590
return self._find_index(self._parsed_byte_map, key)
592
def _parsed_key_index(self, key):
593
"""Return the index of the entry immediately before key.
595
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
596
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
597
have been parsed, then:
598
asking for '' will return 0
599
asking for 'a' will return 0
600
asking for 'b' will return 1
601
asking for 'e' will return 1
603
search_key = (key, None)
604
return self._find_index(self._parsed_key_map, search_key)
606
def _is_parsed(self, offset):
607
"""Returns True if offset has been parsed."""
608
index = self._parsed_byte_index(offset)
609
if index == len(self._parsed_byte_map):
610
return offset < self._parsed_byte_map[index - 1][1]
611
start, end = self._parsed_byte_map[index]
612
return offset >= start and offset < end
614
def _iter_entries_from_total_buffer(self, keys):
615
"""Iterate over keys when the entire index is parsed."""
616
keys = keys.intersection(self._keys)
617
if self.node_ref_lists:
619
value, node_refs = self._nodes[key]
620
yield self, key, value, node_refs
623
yield self, key, self._nodes[key]
625
def iter_entries(self, keys):
626
"""Iterate over keys within the index.
628
:param keys: An iterable providing the keys to be retrieved.
629
:return: An iterable as per iter_all_entries, but restricted to the
630
keys supplied. No additional keys will be returned, and every
631
key supplied that is in the index will be returned.
636
if self._size is None and self._nodes is None:
639
# We fit about 20 keys per minimum-read (4K), so if we are looking for
640
# more than 1/20th of the index its likely (assuming homogenous key
641
# spread) that we'll read the entire index. If we're going to do that,
642
# buffer the whole thing. A better analysis might take key spread into
643
# account - but B+Tree indices are better anyway.
644
# We could look at all data read, and use a threshold there, which will
645
# trigger on ancestry walks, but that is not yet fully mapped out.
646
if self._nodes is None and len(keys) * 20 > self.key_count():
648
if self._nodes is not None:
649
return self._iter_entries_from_total_buffer(keys)
651
return (result[1] for result in bisect_multi_bytes(
652
self._lookup_keys_via_location, self._size, keys))
654
def iter_entries_prefix(self, keys):
655
"""Iterate over keys within the index using prefix matching.
657
Prefix matching is applied within the tuple of a key, not to within
658
the bytestring of each key element. e.g. if you have the keys ('foo',
659
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
660
only the former key is returned.
662
WARNING: Note that this method currently causes a full index parse
663
unconditionally (which is reasonably appropriate as it is a means for
664
thunking many small indices into one larger one and still supplies
665
iter_all_entries at the thunk layer).
667
:param keys: An iterable providing the key prefixes to be retrieved.
668
Each key prefix takes the form of a tuple the length of a key, but
669
with the last N elements 'None' rather than a regular bytestring.
670
The first element cannot be 'None'.
671
:return: An iterable as per iter_all_entries, but restricted to the
672
keys with a matching prefix to those supplied. No additional keys
673
will be returned, and every match that is in the index will be
679
# load data - also finds key lengths
680
if self._nodes is None:
682
if self._key_length == 1:
686
raise errors.BadIndexKey(key)
687
if len(key) != self._key_length:
688
raise errors.BadIndexKey(key)
689
if self.node_ref_lists:
690
value, node_refs = self._nodes[key]
691
yield self, key, value, node_refs
693
yield self, key, self._nodes[key]
695
nodes_by_key = self._get_nodes_by_key()
699
raise errors.BadIndexKey(key)
700
if len(key) != self._key_length:
701
raise errors.BadIndexKey(key)
702
# find what it refers to:
703
key_dict = nodes_by_key
705
# find the subdict whose contents should be returned.
707
while len(elements) and elements[0] is not None:
708
key_dict = key_dict[elements[0]]
711
# a non-existant lookup.
716
key_dict = dicts.pop(-1)
717
# can't be empty or would not exist
718
item, value = key_dict.iteritems().next()
719
if type(value) == dict:
721
dicts.extend(key_dict.itervalues())
724
for value in key_dict.itervalues():
725
# each value is the key:value:node refs tuple
727
yield (self, ) + value
729
# the last thing looked up was a terminal element
730
yield (self, ) + key_dict
732
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
733
"""See BTreeIndex._find_ancestors."""
734
# The api can be implemented as a trivial overlay on top of
735
# iter_entries, it is not an efficient implementation, but it at least
739
for index, key, value, refs in self.iter_entries(keys):
740
parent_keys = refs[ref_list_num]
742
parent_map[key] = parent_keys
743
search_keys.update(parent_keys)
744
# Figure out what, if anything, was missing
745
missing_keys.update(set(keys).difference(found_keys))
746
search_keys = search_keys.difference(parent_map)
750
"""Return an estimate of the number of keys in this index.
752
For GraphIndex the estimate is exact.
754
if self._key_count is None:
755
self._read_and_parse([_HEADER_READV])
756
return self._key_count
758
def _lookup_keys_via_location(self, location_keys):
759
"""Public interface for implementing bisection.
761
If _buffer_all has been called, then all the data for the index is in
762
memory, and this method should not be called, as it uses a separate
763
cache because it cannot pre-resolve all indices, which buffer_all does
766
:param location_keys: A list of location(byte offset), key tuples.
767
:return: A list of (location_key, result) tuples as expected by
768
bzrlib.bisect_multi.bisect_multi_bytes.
770
# Possible improvements:
771
# - only bisect lookup each key once
772
# - sort the keys first, and use that to reduce the bisection window
774
# this progresses in three parts:
777
# attempt to answer the question from the now in memory data.
778
# build the readv request
779
# for each location, ask for 800 bytes - much more than rows we've seen
782
for location, key in location_keys:
783
# can we answer from cache?
784
if self._bisect_nodes and key in self._bisect_nodes:
785
# We have the key parsed.
787
index = self._parsed_key_index(key)
788
if (len(self._parsed_key_map) and
789
self._parsed_key_map[index][0] <= key and
790
(self._parsed_key_map[index][1] >= key or
791
# end of the file has been parsed
792
self._parsed_byte_map[index][1] == self._size)):
793
# the key has been parsed, so no lookup is needed even if its
796
# - if we have examined this part of the file already - yes
797
index = self._parsed_byte_index(location)
798
if (len(self._parsed_byte_map) and
799
self._parsed_byte_map[index][0] <= location and
800
self._parsed_byte_map[index][1] > location):
801
# the byte region has been parsed, so no read is needed.
804
if location + length > self._size:
805
length = self._size - location
806
# todo, trim out parsed locations.
808
readv_ranges.append((location, length))
809
# read the header if needed
810
if self._bisect_nodes is None:
811
readv_ranges.append(_HEADER_READV)
812
self._read_and_parse(readv_ranges)
814
if self._nodes is not None:
815
# _read_and_parse triggered a _buffer_all because we requested the
817
for location, key in location_keys:
818
if key not in self._nodes: # not present
819
result.append(((location, key), False))
820
elif self.node_ref_lists:
821
value, refs = self._nodes[key]
822
result.append(((location, key),
823
(self, key, value, refs)))
825
result.append(((location, key),
826
(self, key, self._nodes[key])))
829
# - figure out <, >, missing, present
830
# - result present references so we can return them.
831
# keys that we cannot answer until we resolve references
832
pending_references = []
833
pending_locations = set()
834
for location, key in location_keys:
835
# can we answer from cache?
836
if key in self._bisect_nodes:
837
# the key has been parsed, so no lookup is needed
838
if self.node_ref_lists:
839
# the references may not have been all parsed.
840
value, refs = self._bisect_nodes[key]
841
wanted_locations = []
842
for ref_list in refs:
844
if ref not in self._keys_by_offset:
845
wanted_locations.append(ref)
847
pending_locations.update(wanted_locations)
848
pending_references.append((location, key))
850
result.append(((location, key), (self, key,
851
value, self._resolve_references(refs))))
853
result.append(((location, key),
854
(self, key, self._bisect_nodes[key])))
857
# has the region the key should be in, been parsed?
858
index = self._parsed_key_index(key)
859
if (self._parsed_key_map[index][0] <= key and
860
(self._parsed_key_map[index][1] >= key or
861
# end of the file has been parsed
862
self._parsed_byte_map[index][1] == self._size)):
863
result.append(((location, key), False))
865
# no, is the key above or below the probed location:
866
# get the range of the probed & parsed location
867
index = self._parsed_byte_index(location)
868
# if the key is below the start of the range, its below
869
if key < self._parsed_key_map[index][0]:
873
result.append(((location, key), direction))
875
# lookup data to resolve references
876
for location in pending_locations:
878
if location + length > self._size:
879
length = self._size - location
880
# TODO: trim out parsed locations (e.g. if the 800 is into the
881
# parsed region trim it, and dont use the adjust_for_latency
884
readv_ranges.append((location, length))
885
self._read_and_parse(readv_ranges)
886
if self._nodes is not None:
887
# The _read_and_parse triggered a _buffer_all, grab the data and
889
for location, key in pending_references:
890
value, refs = self._nodes[key]
891
result.append(((location, key), (self, key, value, refs)))
893
for location, key in pending_references:
894
# answer key references we had to look-up-late.
895
value, refs = self._bisect_nodes[key]
896
result.append(((location, key), (self, key,
897
value, self._resolve_references(refs))))
900
def _parse_header_from_bytes(self, bytes):
901
"""Parse the header from a region of bytes.
903
:param bytes: The data to parse.
904
:return: An offset, data tuple such as readv yields, for the unparsed
905
data. (which may length 0).
907
signature = bytes[0:len(self._signature())]
908
if not signature == self._signature():
909
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
910
lines = bytes[len(self._signature()):].splitlines()
911
options_line = lines[0]
912
if not options_line.startswith(_OPTION_NODE_REFS):
913
raise errors.BadIndexOptions(self)
915
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
917
raise errors.BadIndexOptions(self)
918
options_line = lines[1]
919
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
920
raise errors.BadIndexOptions(self)
922
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
924
raise errors.BadIndexOptions(self)
925
options_line = lines[2]
926
if not options_line.startswith(_OPTION_LEN):
927
raise errors.BadIndexOptions(self)
929
self._key_count = int(options_line[len(_OPTION_LEN):])
931
raise errors.BadIndexOptions(self)
932
# calculate the bytes we have processed
933
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
935
self._parsed_bytes(0, None, header_end, None)
936
# setup parsing state
937
self._expected_elements = 3 + self._key_length
938
# raw data keyed by offset
939
self._keys_by_offset = {}
940
# keys with the value and node references
941
self._bisect_nodes = {}
942
return header_end, bytes[header_end:]
944
def _parse_region(self, offset, data):
945
"""Parse node data returned from a readv operation.
947
:param offset: The byte offset the data starts at.
948
:param data: The data to parse.
952
end = offset + len(data)
955
# Trivial test - if the current index's end is within the
956
# low-matching parsed range, we're done.
957
index = self._parsed_byte_index(high_parsed)
958
if end < self._parsed_byte_map[index][1]:
960
# print "[%d:%d]" % (offset, end), \
961
# self._parsed_byte_map[index:index + 2]
962
high_parsed, last_segment = self._parse_segment(
963
offset, data, end, index)
967
def _parse_segment(self, offset, data, end, index):
968
"""Parse one segment of data.
970
:param offset: Where 'data' begins in the file.
971
:param data: Some data to parse a segment of.
972
:param end: Where data ends
973
:param index: The current index into the parsed bytes map.
974
:return: True if the parsed segment is the last possible one in the
976
:return: high_parsed_byte, last_segment.
977
high_parsed_byte is the location of the highest parsed byte in this
978
segment, last_segment is True if the parsed segment is the last
979
possible one in the data block.
981
# default is to use all data
983
# accomodate overlap with data before this.
984
if offset < self._parsed_byte_map[index][1]:
985
# overlaps the lower parsed region
986
# skip the parsed data
987
trim_start = self._parsed_byte_map[index][1] - offset
988
# don't trim the start for \n
989
start_adjacent = True
990
elif offset == self._parsed_byte_map[index][1]:
991
# abuts the lower parsed region
994
# do not trim anything
995
start_adjacent = True
997
# does not overlap the lower parsed region
1000
# but trim the leading \n
1001
start_adjacent = False
1002
if end == self._size:
1003
# lines up to the end of all data:
1006
# do not strip to the last \n
1009
elif index + 1 == len(self._parsed_byte_map):
1010
# at the end of the parsed data
1013
# but strip to the last \n
1014
end_adjacent = False
1016
elif end == self._parsed_byte_map[index + 1][0]:
1017
# buts up against the next parsed region
1020
# do not strip to the last \n
1023
elif end > self._parsed_byte_map[index + 1][0]:
1024
# overlaps into the next parsed region
1025
# only consider the unparsed data
1026
trim_end = self._parsed_byte_map[index + 1][0] - offset
1027
# do not strip to the last \n as we know its an entire record
1029
last_segment = end < self._parsed_byte_map[index + 1][1]
1031
# does not overlap into the next region
1034
# but strip to the last \n
1035
end_adjacent = False
1037
# now find bytes to discard if needed
1038
if not start_adjacent:
1039
# work around python bug in rfind
1040
if trim_start is None:
1041
trim_start = data.find('\n') + 1
1043
trim_start = data.find('\n', trim_start) + 1
1044
if not (trim_start != 0):
1045
raise AssertionError('no \n was present')
1046
# print 'removing start', offset, trim_start, repr(data[:trim_start])
1047
if not end_adjacent:
1048
# work around python bug in rfind
1049
if trim_end is None:
1050
trim_end = data.rfind('\n') + 1
1052
trim_end = data.rfind('\n', None, trim_end) + 1
1053
if not (trim_end != 0):
1054
raise AssertionError('no \n was present')
1055
# print 'removing end', offset, trim_end, repr(data[trim_end:])
1056
# adjust offset and data to the parseable data.
1057
trimmed_data = data[trim_start:trim_end]
1058
if not (trimmed_data):
1059
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
1060
% (trim_start, trim_end, offset, offset + len(data)))
1062
offset += trim_start
1063
# print "parsing", repr(trimmed_data)
1064
# splitlines mangles the \r delimiters.. don't use it.
1065
lines = trimmed_data.split('\n')
1068
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1069
for key, value in nodes:
1070
self._bisect_nodes[key] = value
1071
self._parsed_bytes(offset, first_key,
1072
offset + len(trimmed_data), last_key)
1073
return offset + len(trimmed_data), last_segment
1075
def _parse_lines(self, lines, pos):
1082
# must be at the end
1084
if not (self._size == pos + 1):
1085
raise AssertionError("%s %s" % (self._size, pos))
1088
elements = line.split('\0')
1089
if len(elements) != self._expected_elements:
1090
raise errors.BadIndexData(self)
1091
# keys are tuples. Each element is a string that may occur many
1092
# times, so we intern them to save space. AB, RC, 200807
1093
key = tuple([intern(element) for element in elements[:self._key_length]])
1094
if first_key is None:
1096
absent, references, value = elements[-3:]
1098
for ref_string in references.split('\t'):
1099
ref_lists.append(tuple([
1100
int(ref) for ref in ref_string.split('\r') if ref
1102
ref_lists = tuple(ref_lists)
1103
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1104
pos += len(line) + 1 # +1 for the \n
1107
if self.node_ref_lists:
1108
node_value = (value, ref_lists)
1111
nodes.append((key, node_value))
1112
# print "parsed ", key
1113
return first_key, key, nodes, trailers
1115
def _parsed_bytes(self, start, start_key, end, end_key):
1116
"""Mark the bytes from start to end as parsed.
1118
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1121
:param start: The start of the parsed region.
1122
:param end: The end of the parsed region.
1124
index = self._parsed_byte_index(start)
1125
new_value = (start, end)
1126
new_key = (start_key, end_key)
1128
# first range parsed is always the beginning.
1129
self._parsed_byte_map.insert(index, new_value)
1130
self._parsed_key_map.insert(index, new_key)
1134
# extend lower region
1135
# extend higher region
1136
# combine two regions
1137
if (index + 1 < len(self._parsed_byte_map) and
1138
self._parsed_byte_map[index][1] == start and
1139
self._parsed_byte_map[index + 1][0] == end):
1140
# combine two regions
1141
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1142
self._parsed_byte_map[index + 1][1])
1143
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1144
self._parsed_key_map[index + 1][1])
1145
del self._parsed_byte_map[index + 1]
1146
del self._parsed_key_map[index + 1]
1147
elif self._parsed_byte_map[index][1] == start:
1148
# extend the lower entry
1149
self._parsed_byte_map[index] = (
1150
self._parsed_byte_map[index][0], end)
1151
self._parsed_key_map[index] = (
1152
self._parsed_key_map[index][0], end_key)
1153
elif (index + 1 < len(self._parsed_byte_map) and
1154
self._parsed_byte_map[index + 1][0] == end):
1155
# extend the higher entry
1156
self._parsed_byte_map[index + 1] = (
1157
start, self._parsed_byte_map[index + 1][1])
1158
self._parsed_key_map[index + 1] = (
1159
start_key, self._parsed_key_map[index + 1][1])
1162
self._parsed_byte_map.insert(index + 1, new_value)
1163
self._parsed_key_map.insert(index + 1, new_key)
1165
def _read_and_parse(self, readv_ranges):
1166
"""Read the the ranges and parse the resulting data.
1168
:param readv_ranges: A prepared readv range list.
1170
if not readv_ranges:
1172
if self._nodes is None and self._bytes_read * 2 >= self._size:
1173
# We've already read more than 50% of the file and we are about to
1174
# request more data, just _buffer_all() and be done
1178
readv_data = self._transport.readv(self._name, readv_ranges, True,
1181
for offset, data in readv_data:
1182
self._bytes_read += len(data)
1183
if offset == 0 and len(data) == self._size:
1184
# We read the whole range, most likely because the
1185
# Transport upcast our readv ranges into one long request
1186
# for enough total data to grab the whole index.
1187
self._buffer_all(StringIO(data))
1189
if self._bisect_nodes is None:
1190
# this must be the start
1191
if not (offset == 0):
1192
raise AssertionError()
1193
offset, data = self._parse_header_from_bytes(data)
1194
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1195
self._parse_region(offset, data)
1197
def _signature(self):
1198
"""The file signature for this index type."""
1202
"""Validate that everything in the index can be accessed."""
1203
# iter_all validates completely at the moment, so just do that.
1204
for node in self.iter_all_entries():
1208
class CombinedGraphIndex(object):
1209
"""A GraphIndex made up from smaller GraphIndices.
1211
The backing indices must implement GraphIndex, and are presumed to be
1214
Queries against the combined index will be made against the first index,
1215
and then the second and so on. The order of index's can thus influence
1216
performance significantly. For example, if one index is on local disk and a
1217
second on a remote server, the local disk index should be before the other
1221
def __init__(self, indices, reload_func=None):
1222
"""Create a CombinedGraphIndex backed by indices.
1224
:param indices: An ordered list of indices to query for data.
1225
:param reload_func: A function to call if we find we are missing an
1226
index. Should have the form reload_func() => True/False to indicate
1227
if reloading actually changed anything.
1229
self._indices = indices
1230
self._reload_func = reload_func
1234
self.__class__.__name__,
1235
', '.join(map(repr, self._indices)))
1237
def clear_cache(self):
1238
"""See GraphIndex.clear_cache()"""
1239
for index in self._indices:
1242
def get_parent_map(self, keys):
1243
"""See graph.StackedParentsProvider.get_parent_map"""
1244
search_keys = set(keys)
1245
if NULL_REVISION in search_keys:
1246
search_keys.discard(NULL_REVISION)
1247
found_parents = {NULL_REVISION:[]}
1250
for index, key, value, refs in self.iter_entries(search_keys):
1253
parents = (NULL_REVISION,)
1254
found_parents[key] = parents
1255
return found_parents
1257
has_key = _has_key_from_parent_map
1259
def insert_index(self, pos, index):
1260
"""Insert a new index in the list of indices to query.
1262
:param pos: The position to insert the index.
1263
:param index: The index to insert.
1265
self._indices.insert(pos, index)
1267
def iter_all_entries(self):
1268
"""Iterate over all keys within the index
1270
Duplicate keys across child indices are presumed to have the same
1271
value and are only reported once.
1273
:return: An iterable of (index, key, reference_lists, value).
1274
There is no defined order for the result iteration - it will be in
1275
the most efficient order for the index.
1280
for index in self._indices:
1281
for node in index.iter_all_entries():
1282
if node[1] not in seen_keys:
1284
seen_keys.add(node[1])
1286
except errors.NoSuchFile:
1287
self._reload_or_raise()
1289
def iter_entries(self, keys):
1290
"""Iterate over keys within the index.
1292
Duplicate keys across child indices are presumed to have the same
1293
value and are only reported once.
1295
:param keys: An iterable providing the keys to be retrieved.
1296
:return: An iterable of (index, key, reference_lists, value). There is no
1297
defined order for the result iteration - it will be in the most
1298
efficient order for the index.
1303
for index in self._indices:
1306
for node in index.iter_entries(keys):
1307
keys.remove(node[1])
1310
except errors.NoSuchFile:
1311
self._reload_or_raise()
1313
def iter_entries_prefix(self, keys):
1314
"""Iterate over keys within the index using prefix matching.
1316
Duplicate keys across child indices are presumed to have the same
1317
value and are only reported once.
1319
Prefix matching is applied within the tuple of a key, not to within
1320
the bytestring of each key element. e.g. if you have the keys ('foo',
1321
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1322
only the former key is returned.
1324
:param keys: An iterable providing the key prefixes to be retrieved.
1325
Each key prefix takes the form of a tuple the length of a key, but
1326
with the last N elements 'None' rather than a regular bytestring.
1327
The first element cannot be 'None'.
1328
:return: An iterable as per iter_all_entries, but restricted to the
1329
keys with a matching prefix to those supplied. No additional keys
1330
will be returned, and every match that is in the index will be
1339
for index in self._indices:
1340
for node in index.iter_entries_prefix(keys):
1341
if node[1] in seen_keys:
1343
seen_keys.add(node[1])
1346
except errors.NoSuchFile:
1347
self._reload_or_raise()
1349
def find_ancestry(self, keys, ref_list_num):
1350
"""Find the complete ancestry for the given set of keys.
1352
Note that this is a whole-ancestry request, so it should be used
1355
:param keys: An iterable of keys to look for
1356
:param ref_list_num: The reference list which references the parents
1358
:return: (parent_map, missing_keys)
1360
missing_keys = set()
1362
keys_to_lookup = set(keys)
1364
while keys_to_lookup:
1365
# keys that *all* indexes claim are missing, stop searching them
1367
all_index_missing = None
1368
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1369
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1371
# len(missing_keys))
1372
for index_idx, index in enumerate(self._indices):
1373
# TODO: we should probably be doing something with
1374
# 'missing_keys' since we've already determined that
1375
# those revisions have not been found anywhere
1376
index_missing_keys = set()
1377
# Find all of the ancestry we can from this index
1378
# keep looking until the search_keys set is empty, which means
1379
# things we didn't find should be in index_missing_keys
1380
search_keys = keys_to_lookup
1382
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1383
# index_idx, len(search_keys),
1384
# len(parent_map), len(index_missing_keys))
1387
# TODO: ref_list_num should really be a parameter, since
1388
# CombinedGraphIndex does not know what the ref lists
1390
search_keys = index._find_ancestors(search_keys,
1391
ref_list_num, parent_map, index_missing_keys)
1392
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1393
# sub_generation, len(search_keys),
1394
# len(parent_map), len(index_missing_keys))
1395
# Now set whatever was missing to be searched in the next index
1396
keys_to_lookup = index_missing_keys
1397
if all_index_missing is None:
1398
all_index_missing = set(index_missing_keys)
1400
all_index_missing.intersection_update(index_missing_keys)
1401
if not keys_to_lookup:
1403
if all_index_missing is None:
1404
# There were no indexes, so all search keys are 'missing'
1405
missing_keys.update(keys_to_lookup)
1406
keys_to_lookup = None
1408
missing_keys.update(all_index_missing)
1409
keys_to_lookup.difference_update(all_index_missing)
1410
return parent_map, missing_keys
1412
def key_count(self):
1413
"""Return an estimate of the number of keys in this index.
1415
For CombinedGraphIndex this is approximated by the sum of the keys of
1416
the child indices. As child indices may have duplicate keys this can
1417
have a maximum error of the number of child indices * largest number of
1422
return sum((index.key_count() for index in self._indices), 0)
1423
except errors.NoSuchFile:
1424
self._reload_or_raise()
1426
missing_keys = _missing_keys_from_parent_map
1428
def _reload_or_raise(self):
1429
"""We just got a NoSuchFile exception.
1431
Try to reload the indices, if it fails, just raise the current
1434
if self._reload_func is None:
1436
exc_type, exc_value, exc_traceback = sys.exc_info()
1437
trace.mutter('Trying to reload after getting exception: %s',
1439
if not self._reload_func():
1440
# We tried to reload, but nothing changed, so we fail anyway
1441
trace.mutter('_reload_func indicated nothing has changed.'
1442
' Raising original exception.')
1443
raise exc_type, exc_value, exc_traceback
1446
"""Validate that everything in the index can be accessed."""
1449
for index in self._indices:
1452
except errors.NoSuchFile:
1453
self._reload_or_raise()
1456
class InMemoryGraphIndex(GraphIndexBuilder):
1457
"""A GraphIndex which operates entirely out of memory and is mutable.
1459
This is designed to allow the accumulation of GraphIndex entries during a
1460
single write operation, where the accumulated entries need to be immediately
1461
available - for example via a CombinedGraphIndex.
1464
def add_nodes(self, nodes):
1465
"""Add nodes to the index.
1467
:param nodes: An iterable of (key, node_refs, value) entries to add.
1469
if self.reference_lists:
1470
for (key, value, node_refs) in nodes:
1471
self.add_node(key, value, node_refs)
1473
for (key, value) in nodes:
1474
self.add_node(key, value)
1476
def iter_all_entries(self):
1477
"""Iterate over all keys within the index
1479
:return: An iterable of (index, key, reference_lists, value). There is no
1480
defined order for the result iteration - it will be in the most
1481
efficient order for the index (in this case dictionary hash order).
1483
if 'evil' in debug.debug_flags:
1484
trace.mutter_callsite(3,
1485
"iter_all_entries scales with size of history.")
1486
if self.reference_lists:
1487
for key, (absent, references, value) in self._nodes.iteritems():
1489
yield self, key, value, references
1491
for key, (absent, references, value) in self._nodes.iteritems():
1493
yield self, key, value
1495
def iter_entries(self, keys):
1496
"""Iterate over keys within the index.
1498
:param keys: An iterable providing the keys to be retrieved.
1499
:return: An iterable of (index, key, value, reference_lists). There is no
1500
defined order for the result iteration - it will be in the most
1501
efficient order for the index (keys iteration order in this case).
1504
if self.reference_lists:
1505
for key in keys.intersection(self._keys):
1506
node = self._nodes[key]
1508
yield self, key, node[2], node[1]
1510
for key in keys.intersection(self._keys):
1511
node = self._nodes[key]
1513
yield self, key, node[2]
1515
def iter_entries_prefix(self, keys):
1516
"""Iterate over keys within the index using prefix matching.
1518
Prefix matching is applied within the tuple of a key, not to within
1519
the bytestring of each key element. e.g. if you have the keys ('foo',
1520
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1521
only the former key is returned.
1523
:param keys: An iterable providing the key prefixes to be retrieved.
1524
Each key prefix takes the form of a tuple the length of a key, but
1525
with the last N elements 'None' rather than a regular bytestring.
1526
The first element cannot be 'None'.
1527
:return: An iterable as per iter_all_entries, but restricted to the
1528
keys with a matching prefix to those supplied. No additional keys
1529
will be returned, and every match that is in the index will be
1532
# XXX: To much duplication with the GraphIndex class; consider finding
1533
# a good place to pull out the actual common logic.
1537
if self._key_length == 1:
1541
raise errors.BadIndexKey(key)
1542
if len(key) != self._key_length:
1543
raise errors.BadIndexKey(key)
1544
node = self._nodes[key]
1547
if self.reference_lists:
1548
yield self, key, node[2], node[1]
1550
yield self, key, node[2]
1552
nodes_by_key = self._get_nodes_by_key()
1556
raise errors.BadIndexKey(key)
1557
if len(key) != self._key_length:
1558
raise errors.BadIndexKey(key)
1559
# find what it refers to:
1560
key_dict = nodes_by_key
1561
elements = list(key)
1562
# find the subdict to return
1564
while len(elements) and elements[0] is not None:
1565
key_dict = key_dict[elements[0]]
1568
# a non-existant lookup.
1573
key_dict = dicts.pop(-1)
1574
# can't be empty or would not exist
1575
item, value = key_dict.iteritems().next()
1576
if type(value) == dict:
1578
dicts.extend(key_dict.itervalues())
1581
for value in key_dict.itervalues():
1582
yield (self, ) + value
1584
yield (self, ) + key_dict
1586
def key_count(self):
1587
"""Return an estimate of the number of keys in this index.
1589
For InMemoryGraphIndex the estimate is exact.
1591
return len(self._keys)
1594
"""In memory index's have no known corruption at the moment."""
1597
class GraphIndexPrefixAdapter(object):
1598
"""An adapter between GraphIndex with different key lengths.
1600
Queries against this will emit queries against the adapted Graph with the
1601
prefix added, queries for all items use iter_entries_prefix. The returned
1602
nodes will have their keys and node references adjusted to remove the
1603
prefix. Finally, an add_nodes_callback can be supplied - when called the
1604
nodes and references being added will have prefix prepended.
1607
def __init__(self, adapted, prefix, missing_key_length,
1608
add_nodes_callback=None):
1609
"""Construct an adapter against adapted with prefix."""
1610
self.adapted = adapted
1611
self.prefix_key = prefix + (None,)*missing_key_length
1612
self.prefix = prefix
1613
self.prefix_len = len(prefix)
1614
self.add_nodes_callback = add_nodes_callback
1616
def add_nodes(self, nodes):
1617
"""Add nodes to the index.
1619
:param nodes: An iterable of (key, node_refs, value) entries to add.
1621
# save nodes in case its an iterator
1622
nodes = tuple(nodes)
1623
translated_nodes = []
1625
# Add prefix_key to each reference node_refs is a tuple of tuples,
1626
# so split it apart, and add prefix_key to the internal reference
1627
for (key, value, node_refs) in nodes:
1628
adjusted_references = (
1629
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1630
for ref_list in node_refs))
1631
translated_nodes.append((self.prefix + key, value,
1632
adjusted_references))
1634
# XXX: TODO add an explicit interface for getting the reference list
1635
# status, to handle this bit of user-friendliness in the API more
1637
for (key, value) in nodes:
1638
translated_nodes.append((self.prefix + key, value))
1639
self.add_nodes_callback(translated_nodes)
1641
def add_node(self, key, value, references=()):
1642
"""Add a node to the index.
1644
:param key: The key. keys are non-empty tuples containing
1645
as many whitespace-free utf8 bytestrings as the key length
1646
defined for this index.
1647
:param references: An iterable of iterables of keys. Each is a
1648
reference to another key.
1649
:param value: The value to associate with the key. It may be any
1650
bytes as long as it does not contain \0 or \n.
1652
self.add_nodes(((key, value, references), ))
1654
def _strip_prefix(self, an_iter):
1655
"""Strip prefix data from nodes and return it."""
1656
for node in an_iter:
1658
if node[1][:self.prefix_len] != self.prefix:
1659
raise errors.BadIndexData(self)
1660
for ref_list in node[3]:
1661
for ref_node in ref_list:
1662
if ref_node[:self.prefix_len] != self.prefix:
1663
raise errors.BadIndexData(self)
1664
yield node[0], node[1][self.prefix_len:], node[2], (
1665
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1666
for ref_list in node[3]))
1668
def iter_all_entries(self):
1669
"""Iterate over all keys within the index
1671
iter_all_entries is implemented against the adapted index using
1672
iter_entries_prefix.
1674
:return: An iterable of (index, key, reference_lists, value). There is no
1675
defined order for the result iteration - it will be in the most
1676
efficient order for the index (in this case dictionary hash order).
1678
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1680
def iter_entries(self, keys):
1681
"""Iterate over keys within the index.
1683
:param keys: An iterable providing the keys to be retrieved.
1684
:return: An iterable of (index, key, value, reference_lists). There is no
1685
defined order for the result iteration - it will be in the most
1686
efficient order for the index (keys iteration order in this case).
1688
return self._strip_prefix(self.adapted.iter_entries(
1689
self.prefix + key for key in keys))
1691
def iter_entries_prefix(self, keys):
1692
"""Iterate over keys within the index using prefix matching.
1694
Prefix matching is applied within the tuple of a key, not to within
1695
the bytestring of each key element. e.g. if you have the keys ('foo',
1696
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1697
only the former key is returned.
1699
:param keys: An iterable providing the key prefixes to be retrieved.
1700
Each key prefix takes the form of a tuple the length of a key, but
1701
with the last N elements 'None' rather than a regular bytestring.
1702
The first element cannot be 'None'.
1703
:return: An iterable as per iter_all_entries, but restricted to the
1704
keys with a matching prefix to those supplied. No additional keys
1705
will be returned, and every match that is in the index will be
1708
return self._strip_prefix(self.adapted.iter_entries_prefix(
1709
self.prefix + key for key in keys))
1711
def key_count(self):
1712
"""Return an estimate of the number of keys in this index.
1714
For GraphIndexPrefixAdapter this is relatively expensive - key
1715
iteration with the prefix is done.
1717
return len(list(self.iter_all_entries()))
1720
"""Call the adapted's validate."""
1721
self.adapted.validate()
added
removed
bzrlib/index.py
