1
# Copyright (C) 2007-2011 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
"""Indexing facilities."""
19
from __future__ import absolute_import
25
'GraphIndexPrefixAdapter',
29
from bisect import bisect_right
30
from cStringIO import StringIO
34
from bzrlib.lazy_import import lazy_import
35
lazy_import(globals(), """
38
revision as _mod_revision,
46
from bzrlib.static_tuple import StaticTuple
48
_HEADER_READV = (0, 200)
49
_OPTION_KEY_ELEMENTS = "key_elements="
51
_OPTION_NODE_REFS = "node_ref_lists="
52
_SIGNATURE = "Bazaar Graph Index 1\n"
55
_whitespace_re = re.compile('[\t\n\x0b\x0c\r\x00 ]')
56
_newline_null_re = re.compile('[\n\0]')
59
def _has_key_from_parent_map(self, key):
60
"""Check if this index has one key.
62
If it's possible to check for multiple keys at once through
63
calling get_parent_map that should be faster.
65
return (key in self.get_parent_map([key]))
68
def _missing_keys_from_parent_map(self, keys):
69
return set(keys) - set(self.get_parent_map(keys))
72
class GraphIndexBuilder(object):
73
"""A builder that can build a GraphIndex.
75
The resulting graph has the structure::
77
_SIGNATURE OPTIONS NODES NEWLINE
78
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
79
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
81
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
82
KEY := Not-whitespace-utf8
84
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
85
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
86
REFERENCE := DIGITS ; digits is the byte offset in the index of the
88
VALUE := no-newline-no-null-bytes
91
def __init__(self, reference_lists=0, key_elements=1):
92
"""Create a GraphIndex builder.
94
:param reference_lists: The number of node references lists for each
96
:param key_elements: The number of bytestrings in each key.
98
self.reference_lists = reference_lists
99
# A dict of {key: (absent, ref_lists, value)}
101
# Keys that are referenced but not actually present in this index
102
self._absent_keys = set()
103
self._nodes_by_key = None
104
self._key_length = key_elements
105
self._optimize_for_size = False
106
self._combine_backing_indices = True
108
def _check_key(self, key):
109
"""Raise BadIndexKey if key is not a valid key for this index."""
110
if type(key) not in (tuple, StaticTuple):
111
raise errors.BadIndexKey(key)
112
if self._key_length != len(key):
113
raise errors.BadIndexKey(key)
115
if not element or _whitespace_re.search(element) is not None:
116
raise errors.BadIndexKey(element)
118
def _external_references(self):
119
"""Return references that are not present in this index.
123
# TODO: JAM 2008-11-21 This makes an assumption about how the reference
124
# lists are used. It is currently correct for pack-0.92 through
125
# 1.9, which use the node references (3rd column) second
126
# reference list as the compression parent. Perhaps this should
127
# be moved into something higher up the stack, since it
128
# makes assumptions about how the index is used.
129
if self.reference_lists > 1:
130
for node in self.iter_all_entries():
132
refs.update(node[3][1])
135
# If reference_lists == 0 there can be no external references, and
136
# if reference_lists == 1, then there isn't a place to store the
140
def _get_nodes_by_key(self):
141
if self._nodes_by_key is None:
143
if self.reference_lists:
144
for key, (absent, references, value) in self._nodes.iteritems():
147
key_dict = nodes_by_key
148
for subkey in key[:-1]:
149
key_dict = key_dict.setdefault(subkey, {})
150
key_dict[key[-1]] = key, value, references
152
for key, (absent, references, value) in self._nodes.iteritems():
155
key_dict = nodes_by_key
156
for subkey in key[:-1]:
157
key_dict = key_dict.setdefault(subkey, {})
158
key_dict[key[-1]] = key, value
159
self._nodes_by_key = nodes_by_key
160
return self._nodes_by_key
162
def _update_nodes_by_key(self, key, value, node_refs):
163
"""Update the _nodes_by_key dict with a new key.
165
For a key of (foo, bar, baz) create
166
_nodes_by_key[foo][bar][baz] = key_value
168
if self._nodes_by_key is None:
170
key_dict = self._nodes_by_key
171
if self.reference_lists:
172
key_value = StaticTuple(key, value, node_refs)
174
key_value = StaticTuple(key, value)
175
for subkey in key[:-1]:
176
key_dict = key_dict.setdefault(subkey, {})
177
key_dict[key[-1]] = key_value
179
def _check_key_ref_value(self, key, references, value):
180
"""Check that 'key' and 'references' are all valid.
182
:param key: A key tuple. Must conform to the key interface (be a tuple,
183
be of the right length, not have any whitespace or nulls in any key
185
:param references: An iterable of reference lists. Something like
186
[[(ref, key)], [(ref, key), (other, key)]]
187
:param value: The value associate with this key. Must not contain
188
newlines or null characters.
189
:return: (node_refs, absent_references)
191
* node_refs: basically a packed form of 'references' where all
193
* absent_references: reference keys that are not in self._nodes.
194
This may contain duplicates if the same key is referenced in
197
as_st = StaticTuple.from_sequence
199
if _newline_null_re.search(value) is not None:
200
raise errors.BadIndexValue(value)
201
if len(references) != self.reference_lists:
202
raise errors.BadIndexValue(references)
204
absent_references = []
205
for reference_list in references:
206
for reference in reference_list:
207
# If reference *is* in self._nodes, then we know it has already
209
if reference not in self._nodes:
210
self._check_key(reference)
211
absent_references.append(reference)
212
reference_list = as_st([as_st(ref).intern()
213
for ref in reference_list])
214
node_refs.append(reference_list)
215
return as_st(node_refs), absent_references
217
def add_node(self, key, value, references=()):
218
"""Add a node to the index.
220
:param key: The key. keys are non-empty tuples containing
221
as many whitespace-free utf8 bytestrings as the key length
222
defined for this index.
223
:param references: An iterable of iterables of keys. Each is a
224
reference to another key.
225
:param value: The value to associate with the key. It may be any
226
bytes as long as it does not contain \\0 or \\n.
229
absent_references) = self._check_key_ref_value(key, references, value)
230
if key in self._nodes and self._nodes[key][0] != 'a':
231
raise errors.BadIndexDuplicateKey(key, self)
232
for reference in absent_references:
233
# There may be duplicates, but I don't think it is worth worrying
235
self._nodes[reference] = ('a', (), '')
236
self._absent_keys.update(absent_references)
237
self._absent_keys.discard(key)
238
self._nodes[key] = ('', node_refs, value)
239
if self._nodes_by_key is not None and self._key_length > 1:
240
self._update_nodes_by_key(key, value, node_refs)
242
def clear_cache(self):
243
"""See GraphIndex.clear_cache()
245
This is a no-op, but we need the api to conform to a generic 'Index'
252
:returns: cStringIO holding the full context of the index as it
253
should be written to disk.
256
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
257
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
258
key_count = len(self._nodes) - len(self._absent_keys)
259
lines.append(_OPTION_LEN + str(key_count) + '\n')
260
prefix_length = sum(len(x) for x in lines)
261
# references are byte offsets. To avoid having to do nasty
262
# polynomial work to resolve offsets (references to later in the
263
# file cannot be determined until all the inbetween references have
264
# been calculated too) we pad the offsets with 0's to make them be
265
# of consistent length. Using binary offsets would break the trivial
267
# to calculate the width of zero's needed we do three passes:
268
# one to gather all the non-reference data and the number of references.
269
# one to pad all the data with reference-length and determine entry
273
# forward sorted by key. In future we may consider topological sorting,
274
# at the cost of table scans for direct lookup, or a second index for
276
nodes = sorted(self._nodes.items())
277
# if we do not prepass, we don't know how long it will be up front.
278
expected_bytes = None
279
# we only need to pre-pass if we have reference lists at all.
280
if self.reference_lists:
282
non_ref_bytes = prefix_length
284
# TODO use simple multiplication for the constants in this loop.
285
for key, (absent, references, value) in nodes:
286
# record the offset known *so far* for this key:
287
# the non reference bytes to date, and the total references to
288
# date - saves reaccumulating on the second pass
289
key_offset_info.append((key, non_ref_bytes, total_references))
290
# key is literal, value is literal, there are 3 null's, 1 NL
291
# key is variable length tuple, \x00 between elements
292
non_ref_bytes += sum(len(element) for element in key)
293
if self._key_length > 1:
294
non_ref_bytes += self._key_length - 1
295
# value is literal bytes, there are 3 null's, 1 NL.
296
non_ref_bytes += len(value) + 3 + 1
297
# one byte for absent if set.
300
elif self.reference_lists:
301
# (ref_lists -1) tabs
302
non_ref_bytes += self.reference_lists - 1
303
# (ref-1 cr's per ref_list)
304
for ref_list in references:
305
# how many references across the whole file?
306
total_references += len(ref_list)
307
# accrue reference separators
309
non_ref_bytes += len(ref_list) - 1
310
# how many digits are needed to represent the total byte count?
312
possible_total_bytes = non_ref_bytes + total_references*digits
313
while 10 ** digits < possible_total_bytes:
315
possible_total_bytes = non_ref_bytes + total_references*digits
316
expected_bytes = possible_total_bytes + 1 # terminating newline
317
# resolve key addresses.
319
for key, non_ref_bytes, total_references in key_offset_info:
320
key_addresses[key] = non_ref_bytes + total_references*digits
322
format_string = '%%0%sd' % digits
323
for key, (absent, references, value) in nodes:
324
flattened_references = []
325
for ref_list in references:
327
for reference in ref_list:
328
ref_addresses.append(format_string % key_addresses[reference])
329
flattened_references.append('\r'.join(ref_addresses))
330
string_key = '\x00'.join(key)
331
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
332
'\t'.join(flattened_references), value))
334
result = StringIO(''.join(lines))
335
if expected_bytes and len(result.getvalue()) != expected_bytes:
336
raise errors.BzrError('Failed index creation. Internal error:'
337
' mismatched output length and expected length: %d %d' %
338
(len(result.getvalue()), expected_bytes))
341
def set_optimize(self, for_size=None, combine_backing_indices=None):
342
"""Change how the builder tries to optimize the result.
344
:param for_size: Tell the builder to try and make the index as small as
346
:param combine_backing_indices: If the builder spills to disk to save
347
memory, should the on-disk indices be combined. Set to True if you
348
are going to be probing the index, but to False if you are not. (If
349
you are not querying, then the time spent combining is wasted.)
352
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
354
if for_size is not None:
355
self._optimize_for_size = for_size
356
if combine_backing_indices is not None:
357
self._combine_backing_indices = combine_backing_indices
359
def find_ancestry(self, keys, ref_list_num):
360
"""See CombinedGraphIndex.find_ancestry()"""
366
for _, key, value, ref_lists in self.iter_entries(pending):
367
parent_keys = ref_lists[ref_list_num]
368
parent_map[key] = parent_keys
369
next_pending.update([p for p in parent_keys if p not in
371
missing_keys.update(pending.difference(parent_map))
372
pending = next_pending
373
return parent_map, missing_keys
376
class GraphIndex(object):
377
"""An index for data with embedded graphs.
379
The index maps keys to a list of key reference lists, and a value.
380
Each node has the same number of key reference lists. Each key reference
381
list can be empty or an arbitrary length. The value is an opaque NULL
382
terminated string without any newlines. The storage of the index is
383
hidden in the interface: keys and key references are always tuples of
384
bytestrings, never the internal representation (e.g. dictionary offsets).
386
It is presumed that the index will not be mutated - it is static data.
388
Successive iter_all_entries calls will read the entire index each time.
389
Additionally, iter_entries calls will read the index linearly until the
390
desired keys are found. XXX: This must be fixed before the index is
391
suitable for production use. :XXX
394
def __init__(self, transport, name, size, unlimited_cache=False, offset=0):
395
"""Open an index called name on transport.
397
:param transport: A bzrlib.transport.Transport.
398
:param name: A path to provide to transport API calls.
399
:param size: The size of the index in bytes. This is used for bisection
400
logic to perform partial index reads. While the size could be
401
obtained by statting the file this introduced an additional round
402
trip as well as requiring stat'able transports, both of which are
403
avoided by having it supplied. If size is None, then bisection
404
support will be disabled and accessing the index will just stream
406
:param offset: Instead of starting the index data at offset 0, start it
407
at an arbitrary offset.
409
self._transport = transport
411
# Becomes a dict of key:(value, reference-list-byte-locations) used by
412
# the bisection interface to store parsed but not resolved keys.
413
self._bisect_nodes = None
414
# Becomes a dict of key:(value, reference-list-keys) which are ready to
415
# be returned directly to callers.
417
# a sorted list of slice-addresses for the parsed bytes of the file.
418
# e.g. (0,1) would mean that byte 0 is parsed.
419
self._parsed_byte_map = []
420
# a sorted list of keys matching each slice address for parsed bytes
421
# e.g. (None, 'foo@bar') would mean that the first byte contained no
422
# key, and the end byte of the slice is the of the data for 'foo@bar'
423
self._parsed_key_map = []
424
self._key_count = None
425
self._keys_by_offset = None
426
self._nodes_by_key = None
428
# The number of bytes we've read so far in trying to process this file
430
self._base_offset = offset
432
def __eq__(self, other):
433
"""Equal when self and other were created with the same parameters."""
435
type(self) == type(other) and
436
self._transport == other._transport and
437
self._name == other._name and
438
self._size == other._size)
440
def __ne__(self, other):
441
return not self.__eq__(other)
444
return "%s(%r)" % (self.__class__.__name__,
445
self._transport.abspath(self._name))
447
def _buffer_all(self, stream=None):
448
"""Buffer all the index data.
450
Mutates self._nodes and self.keys_by_offset.
452
if self._nodes is not None:
453
# We already did this
455
if 'index' in debug.debug_flags:
456
trace.mutter('Reading entire index %s',
457
self._transport.abspath(self._name))
459
stream = self._transport.get(self._name)
460
if self._base_offset != 0:
461
# This is wasteful, but it is better than dealing with
462
# adjusting all the offsets, etc.
463
stream = StringIO(stream.read()[self._base_offset:])
464
self._read_prefix(stream)
465
self._expected_elements = 3 + self._key_length
467
# raw data keyed by offset
468
self._keys_by_offset = {}
469
# ready-to-return key:value or key:value, node_ref_lists
471
self._nodes_by_key = None
474
lines = stream.read().split('\n')
475
# GZ 2009-09-20: Should really use a try/finally block to ensure close
478
_, _, _, trailers = self._parse_lines(lines, pos)
479
for key, absent, references, value in self._keys_by_offset.itervalues():
482
# resolve references:
483
if self.node_ref_lists:
484
node_value = (value, self._resolve_references(references))
487
self._nodes[key] = node_value
488
# cache the keys for quick set intersections
490
# there must be one line - the empty trailer line.
491
raise errors.BadIndexData(self)
493
def clear_cache(self):
494
"""Clear out any cached/memoized values.
496
This can be called at any time, but generally it is used when we have
497
extracted some information, but don't expect to be requesting any more
501
def external_references(self, ref_list_num):
502
"""Return references that are not present in this index.
505
if ref_list_num + 1 > self.node_ref_lists:
506
raise ValueError('No ref list %d, index has %d ref lists'
507
% (ref_list_num, self.node_ref_lists))
510
for key, (value, ref_lists) in nodes.iteritems():
511
ref_list = ref_lists[ref_list_num]
512
refs.update([ref for ref in ref_list if ref not in nodes])
515
def _get_nodes_by_key(self):
516
if self._nodes_by_key is None:
518
if self.node_ref_lists:
519
for key, (value, references) in self._nodes.iteritems():
520
key_dict = nodes_by_key
521
for subkey in key[:-1]:
522
key_dict = key_dict.setdefault(subkey, {})
523
key_dict[key[-1]] = key, value, references
525
for key, value in self._nodes.iteritems():
526
key_dict = nodes_by_key
527
for subkey in key[:-1]:
528
key_dict = key_dict.setdefault(subkey, {})
529
key_dict[key[-1]] = key, value
530
self._nodes_by_key = nodes_by_key
531
return self._nodes_by_key
533
def iter_all_entries(self):
534
"""Iterate over all keys within the index.
536
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
537
The former tuple is used when there are no reference lists in the
538
index, making the API compatible with simple key:value index types.
539
There is no defined order for the result iteration - it will be in
540
the most efficient order for the index.
542
if 'evil' in debug.debug_flags:
543
trace.mutter_callsite(3,
544
"iter_all_entries scales with size of history.")
545
if self._nodes is None:
547
if self.node_ref_lists:
548
for key, (value, node_ref_lists) in self._nodes.iteritems():
549
yield self, key, value, node_ref_lists
551
for key, value in self._nodes.iteritems():
552
yield self, key, value
554
def _read_prefix(self, stream):
555
signature = stream.read(len(self._signature()))
556
if not signature == self._signature():
557
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
558
options_line = stream.readline()
559
if not options_line.startswith(_OPTION_NODE_REFS):
560
raise errors.BadIndexOptions(self)
562
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
564
raise errors.BadIndexOptions(self)
565
options_line = stream.readline()
566
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
567
raise errors.BadIndexOptions(self)
569
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
571
raise errors.BadIndexOptions(self)
572
options_line = stream.readline()
573
if not options_line.startswith(_OPTION_LEN):
574
raise errors.BadIndexOptions(self)
576
self._key_count = int(options_line[len(_OPTION_LEN):-1])
578
raise errors.BadIndexOptions(self)
580
def _resolve_references(self, references):
581
"""Return the resolved key references for references.
583
References are resolved by looking up the location of the key in the
584
_keys_by_offset map and substituting the key name, preserving ordering.
586
:param references: An iterable of iterables of key locations. e.g.
588
:return: A tuple of tuples of keys.
591
for ref_list in references:
592
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
593
return tuple(node_refs)
595
def _find_index(self, range_map, key):
596
"""Helper for the _parsed_*_index calls.
598
Given a range map - [(start, end), ...], finds the index of the range
599
in the map for key if it is in the map, and if it is not there, the
600
immediately preceeding range in the map.
602
result = bisect_right(range_map, key) - 1
603
if result + 1 < len(range_map):
604
# check the border condition, it may be in result + 1
605
if range_map[result + 1][0] == key[0]:
609
def _parsed_byte_index(self, offset):
610
"""Return the index of the entry immediately before offset.
612
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
613
there is one unparsed byte (the 11th, addressed as[10]). then:
614
asking for 0 will return 0
615
asking for 10 will return 0
616
asking for 11 will return 1
617
asking for 12 will return 1
620
return self._find_index(self._parsed_byte_map, key)
622
def _parsed_key_index(self, key):
623
"""Return the index of the entry immediately before key.
625
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
626
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
627
have been parsed, then:
628
asking for '' will return 0
629
asking for 'a' will return 0
630
asking for 'b' will return 1
631
asking for 'e' will return 1
633
search_key = (key, None)
634
return self._find_index(self._parsed_key_map, search_key)
636
def _is_parsed(self, offset):
637
"""Returns True if offset has been parsed."""
638
index = self._parsed_byte_index(offset)
639
if index == len(self._parsed_byte_map):
640
return offset < self._parsed_byte_map[index - 1][1]
641
start, end = self._parsed_byte_map[index]
642
return offset >= start and offset < end
644
def _iter_entries_from_total_buffer(self, keys):
645
"""Iterate over keys when the entire index is parsed."""
646
# Note: See the note in BTreeBuilder.iter_entries for why we don't use
647
# .intersection() here
649
keys = [key for key in keys if key in nodes]
650
if self.node_ref_lists:
652
value, node_refs = nodes[key]
653
yield self, key, value, node_refs
656
yield self, key, nodes[key]
658
def iter_entries(self, keys):
659
"""Iterate over keys within the index.
661
:param keys: An iterable providing the keys to be retrieved.
662
:return: An iterable as per iter_all_entries, but restricted to the
663
keys supplied. No additional keys will be returned, and every
664
key supplied that is in the index will be returned.
669
if self._size is None and self._nodes is None:
672
# We fit about 20 keys per minimum-read (4K), so if we are looking for
673
# more than 1/20th of the index its likely (assuming homogenous key
674
# spread) that we'll read the entire index. If we're going to do that,
675
# buffer the whole thing. A better analysis might take key spread into
676
# account - but B+Tree indices are better anyway.
677
# We could look at all data read, and use a threshold there, which will
678
# trigger on ancestry walks, but that is not yet fully mapped out.
679
if self._nodes is None and len(keys) * 20 > self.key_count():
681
if self._nodes is not None:
682
return self._iter_entries_from_total_buffer(keys)
684
return (result[1] for result in bisect_multi.bisect_multi_bytes(
685
self._lookup_keys_via_location, self._size, keys))
687
def iter_entries_prefix(self, keys):
688
"""Iterate over keys within the index using prefix matching.
690
Prefix matching is applied within the tuple of a key, not to within
691
the bytestring of each key element. e.g. if you have the keys ('foo',
692
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
693
only the former key is returned.
695
WARNING: Note that this method currently causes a full index parse
696
unconditionally (which is reasonably appropriate as it is a means for
697
thunking many small indices into one larger one and still supplies
698
iter_all_entries at the thunk layer).
700
:param keys: An iterable providing the key prefixes to be retrieved.
701
Each key prefix takes the form of a tuple the length of a key, but
702
with the last N elements 'None' rather than a regular bytestring.
703
The first element cannot be 'None'.
704
:return: An iterable as per iter_all_entries, but restricted to the
705
keys with a matching prefix to those supplied. No additional keys
706
will be returned, and every match that is in the index will be
712
# load data - also finds key lengths
713
if self._nodes is None:
715
if self._key_length == 1:
719
raise errors.BadIndexKey(key)
720
if len(key) != self._key_length:
721
raise errors.BadIndexKey(key)
722
if self.node_ref_lists:
723
value, node_refs = self._nodes[key]
724
yield self, key, value, node_refs
726
yield self, key, self._nodes[key]
728
nodes_by_key = self._get_nodes_by_key()
732
raise errors.BadIndexKey(key)
733
if len(key) != self._key_length:
734
raise errors.BadIndexKey(key)
735
# find what it refers to:
736
key_dict = nodes_by_key
738
# find the subdict whose contents should be returned.
740
while len(elements) and elements[0] is not None:
741
key_dict = key_dict[elements[0]]
744
# a non-existant lookup.
749
key_dict = dicts.pop(-1)
750
# can't be empty or would not exist
751
item, value = key_dict.iteritems().next()
752
if type(value) == dict:
754
dicts.extend(key_dict.itervalues())
757
for value in key_dict.itervalues():
758
# each value is the key:value:node refs tuple
760
yield (self, ) + value
762
# the last thing looked up was a terminal element
763
yield (self, ) + key_dict
765
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
766
"""See BTreeIndex._find_ancestors."""
767
# The api can be implemented as a trivial overlay on top of
768
# iter_entries, it is not an efficient implementation, but it at least
772
for index, key, value, refs in self.iter_entries(keys):
773
parent_keys = refs[ref_list_num]
775
parent_map[key] = parent_keys
776
search_keys.update(parent_keys)
777
# Figure out what, if anything, was missing
778
missing_keys.update(set(keys).difference(found_keys))
779
search_keys = search_keys.difference(parent_map)
783
"""Return an estimate of the number of keys in this index.
785
For GraphIndex the estimate is exact.
787
if self._key_count is None:
788
self._read_and_parse([_HEADER_READV])
789
return self._key_count
791
def _lookup_keys_via_location(self, location_keys):
792
"""Public interface for implementing bisection.
794
If _buffer_all has been called, then all the data for the index is in
795
memory, and this method should not be called, as it uses a separate
796
cache because it cannot pre-resolve all indices, which buffer_all does
799
:param location_keys: A list of location(byte offset), key tuples.
800
:return: A list of (location_key, result) tuples as expected by
801
bzrlib.bisect_multi.bisect_multi_bytes.
803
# Possible improvements:
804
# - only bisect lookup each key once
805
# - sort the keys first, and use that to reduce the bisection window
807
# this progresses in three parts:
810
# attempt to answer the question from the now in memory data.
811
# build the readv request
812
# for each location, ask for 800 bytes - much more than rows we've seen
815
for location, key in location_keys:
816
# can we answer from cache?
817
if self._bisect_nodes and key in self._bisect_nodes:
818
# We have the key parsed.
820
index = self._parsed_key_index(key)
821
if (len(self._parsed_key_map) and
822
self._parsed_key_map[index][0] <= key and
823
(self._parsed_key_map[index][1] >= key or
824
# end of the file has been parsed
825
self._parsed_byte_map[index][1] == self._size)):
826
# the key has been parsed, so no lookup is needed even if its
829
# - if we have examined this part of the file already - yes
830
index = self._parsed_byte_index(location)
831
if (len(self._parsed_byte_map) and
832
self._parsed_byte_map[index][0] <= location and
833
self._parsed_byte_map[index][1] > location):
834
# the byte region has been parsed, so no read is needed.
837
if location + length > self._size:
838
length = self._size - location
839
# todo, trim out parsed locations.
841
readv_ranges.append((location, length))
842
# read the header if needed
843
if self._bisect_nodes is None:
844
readv_ranges.append(_HEADER_READV)
845
self._read_and_parse(readv_ranges)
847
if self._nodes is not None:
848
# _read_and_parse triggered a _buffer_all because we requested the
850
for location, key in location_keys:
851
if key not in self._nodes: # not present
852
result.append(((location, key), False))
853
elif self.node_ref_lists:
854
value, refs = self._nodes[key]
855
result.append(((location, key),
856
(self, key, value, refs)))
858
result.append(((location, key),
859
(self, key, self._nodes[key])))
862
# - figure out <, >, missing, present
863
# - result present references so we can return them.
864
# keys that we cannot answer until we resolve references
865
pending_references = []
866
pending_locations = set()
867
for location, key in location_keys:
868
# can we answer from cache?
869
if key in self._bisect_nodes:
870
# the key has been parsed, so no lookup is needed
871
if self.node_ref_lists:
872
# the references may not have been all parsed.
873
value, refs = self._bisect_nodes[key]
874
wanted_locations = []
875
for ref_list in refs:
877
if ref not in self._keys_by_offset:
878
wanted_locations.append(ref)
880
pending_locations.update(wanted_locations)
881
pending_references.append((location, key))
883
result.append(((location, key), (self, key,
884
value, self._resolve_references(refs))))
886
result.append(((location, key),
887
(self, key, self._bisect_nodes[key])))
890
# has the region the key should be in, been parsed?
891
index = self._parsed_key_index(key)
892
if (self._parsed_key_map[index][0] <= key and
893
(self._parsed_key_map[index][1] >= key or
894
# end of the file has been parsed
895
self._parsed_byte_map[index][1] == self._size)):
896
result.append(((location, key), False))
898
# no, is the key above or below the probed location:
899
# get the range of the probed & parsed location
900
index = self._parsed_byte_index(location)
901
# if the key is below the start of the range, its below
902
if key < self._parsed_key_map[index][0]:
906
result.append(((location, key), direction))
908
# lookup data to resolve references
909
for location in pending_locations:
911
if location + length > self._size:
912
length = self._size - location
913
# TODO: trim out parsed locations (e.g. if the 800 is into the
914
# parsed region trim it, and dont use the adjust_for_latency
917
readv_ranges.append((location, length))
918
self._read_and_parse(readv_ranges)
919
if self._nodes is not None:
920
# The _read_and_parse triggered a _buffer_all, grab the data and
922
for location, key in pending_references:
923
value, refs = self._nodes[key]
924
result.append(((location, key), (self, key, value, refs)))
926
for location, key in pending_references:
927
# answer key references we had to look-up-late.
928
value, refs = self._bisect_nodes[key]
929
result.append(((location, key), (self, key,
930
value, self._resolve_references(refs))))
933
def _parse_header_from_bytes(self, bytes):
934
"""Parse the header from a region of bytes.
936
:param bytes: The data to parse.
937
:return: An offset, data tuple such as readv yields, for the unparsed
938
data. (which may length 0).
940
signature = bytes[0:len(self._signature())]
941
if not signature == self._signature():
942
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
943
lines = bytes[len(self._signature()):].splitlines()
944
options_line = lines[0]
945
if not options_line.startswith(_OPTION_NODE_REFS):
946
raise errors.BadIndexOptions(self)
948
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
950
raise errors.BadIndexOptions(self)
951
options_line = lines[1]
952
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
953
raise errors.BadIndexOptions(self)
955
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
957
raise errors.BadIndexOptions(self)
958
options_line = lines[2]
959
if not options_line.startswith(_OPTION_LEN):
960
raise errors.BadIndexOptions(self)
962
self._key_count = int(options_line[len(_OPTION_LEN):])
964
raise errors.BadIndexOptions(self)
965
# calculate the bytes we have processed
966
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
968
self._parsed_bytes(0, None, header_end, None)
969
# setup parsing state
970
self._expected_elements = 3 + self._key_length
971
# raw data keyed by offset
972
self._keys_by_offset = {}
973
# keys with the value and node references
974
self._bisect_nodes = {}
975
return header_end, bytes[header_end:]
977
def _parse_region(self, offset, data):
978
"""Parse node data returned from a readv operation.
980
:param offset: The byte offset the data starts at.
981
:param data: The data to parse.
985
end = offset + len(data)
988
# Trivial test - if the current index's end is within the
989
# low-matching parsed range, we're done.
990
index = self._parsed_byte_index(high_parsed)
991
if end < self._parsed_byte_map[index][1]:
993
# print "[%d:%d]" % (offset, end), \
994
# self._parsed_byte_map[index:index + 2]
995
high_parsed, last_segment = self._parse_segment(
996
offset, data, end, index)
1000
def _parse_segment(self, offset, data, end, index):
1001
"""Parse one segment of data.
1003
:param offset: Where 'data' begins in the file.
1004
:param data: Some data to parse a segment of.
1005
:param end: Where data ends
1006
:param index: The current index into the parsed bytes map.
1007
:return: True if the parsed segment is the last possible one in the
1009
:return: high_parsed_byte, last_segment.
1010
high_parsed_byte is the location of the highest parsed byte in this
1011
segment, last_segment is True if the parsed segment is the last
1012
possible one in the data block.
1014
# default is to use all data
1016
# accomodate overlap with data before this.
1017
if offset < self._parsed_byte_map[index][1]:
1018
# overlaps the lower parsed region
1019
# skip the parsed data
1020
trim_start = self._parsed_byte_map[index][1] - offset
1021
# don't trim the start for \n
1022
start_adjacent = True
1023
elif offset == self._parsed_byte_map[index][1]:
1024
# abuts the lower parsed region
1027
# do not trim anything
1028
start_adjacent = True
1030
# does not overlap the lower parsed region
1033
# but trim the leading \n
1034
start_adjacent = False
1035
if end == self._size:
1036
# lines up to the end of all data:
1039
# do not strip to the last \n
1042
elif index + 1 == len(self._parsed_byte_map):
1043
# at the end of the parsed data
1046
# but strip to the last \n
1047
end_adjacent = False
1049
elif end == self._parsed_byte_map[index + 1][0]:
1050
# buts up against the next parsed region
1053
# do not strip to the last \n
1056
elif end > self._parsed_byte_map[index + 1][0]:
1057
# overlaps into the next parsed region
1058
# only consider the unparsed data
1059
trim_end = self._parsed_byte_map[index + 1][0] - offset
1060
# do not strip to the last \n as we know its an entire record
1062
last_segment = end < self._parsed_byte_map[index + 1][1]
1064
# does not overlap into the next region
1067
# but strip to the last \n
1068
end_adjacent = False
1070
# now find bytes to discard if needed
1071
if not start_adjacent:
1072
# work around python bug in rfind
1073
if trim_start is None:
1074
trim_start = data.find('\n') + 1
1076
trim_start = data.find('\n', trim_start) + 1
1077
if not (trim_start != 0):
1078
raise AssertionError('no \n was present')
1079
# print 'removing start', offset, trim_start, repr(data[:trim_start])
1080
if not end_adjacent:
1081
# work around python bug in rfind
1082
if trim_end is None:
1083
trim_end = data.rfind('\n') + 1
1085
trim_end = data.rfind('\n', None, trim_end) + 1
1086
if not (trim_end != 0):
1087
raise AssertionError('no \n was present')
1088
# print 'removing end', offset, trim_end, repr(data[trim_end:])
1089
# adjust offset and data to the parseable data.
1090
trimmed_data = data[trim_start:trim_end]
1091
if not (trimmed_data):
1092
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
1093
% (trim_start, trim_end, offset, offset + len(data)))
1095
offset += trim_start
1096
# print "parsing", repr(trimmed_data)
1097
# splitlines mangles the \r delimiters.. don't use it.
1098
lines = trimmed_data.split('\n')
1101
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1102
for key, value in nodes:
1103
self._bisect_nodes[key] = value
1104
self._parsed_bytes(offset, first_key,
1105
offset + len(trimmed_data), last_key)
1106
return offset + len(trimmed_data), last_segment
1108
def _parse_lines(self, lines, pos):
1115
# must be at the end
1117
if not (self._size == pos + 1):
1118
raise AssertionError("%s %s" % (self._size, pos))
1121
elements = line.split('\0')
1122
if len(elements) != self._expected_elements:
1123
raise errors.BadIndexData(self)
1124
# keys are tuples. Each element is a string that may occur many
1125
# times, so we intern them to save space. AB, RC, 200807
1126
key = tuple([intern(element) for element in elements[:self._key_length]])
1127
if first_key is None:
1129
absent, references, value = elements[-3:]
1131
for ref_string in references.split('\t'):
1132
ref_lists.append(tuple([
1133
int(ref) for ref in ref_string.split('\r') if ref
1135
ref_lists = tuple(ref_lists)
1136
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1137
pos += len(line) + 1 # +1 for the \n
1140
if self.node_ref_lists:
1141
node_value = (value, ref_lists)
1144
nodes.append((key, node_value))
1145
# print "parsed ", key
1146
return first_key, key, nodes, trailers
1148
def _parsed_bytes(self, start, start_key, end, end_key):
1149
"""Mark the bytes from start to end as parsed.
1151
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1154
:param start: The start of the parsed region.
1155
:param end: The end of the parsed region.
1157
index = self._parsed_byte_index(start)
1158
new_value = (start, end)
1159
new_key = (start_key, end_key)
1161
# first range parsed is always the beginning.
1162
self._parsed_byte_map.insert(index, new_value)
1163
self._parsed_key_map.insert(index, new_key)
1167
# extend lower region
1168
# extend higher region
1169
# combine two regions
1170
if (index + 1 < len(self._parsed_byte_map) and
1171
self._parsed_byte_map[index][1] == start and
1172
self._parsed_byte_map[index + 1][0] == end):
1173
# combine two regions
1174
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1175
self._parsed_byte_map[index + 1][1])
1176
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1177
self._parsed_key_map[index + 1][1])
1178
del self._parsed_byte_map[index + 1]
1179
del self._parsed_key_map[index + 1]
1180
elif self._parsed_byte_map[index][1] == start:
1181
# extend the lower entry
1182
self._parsed_byte_map[index] = (
1183
self._parsed_byte_map[index][0], end)
1184
self._parsed_key_map[index] = (
1185
self._parsed_key_map[index][0], end_key)
1186
elif (index + 1 < len(self._parsed_byte_map) and
1187
self._parsed_byte_map[index + 1][0] == end):
1188
# extend the higher entry
1189
self._parsed_byte_map[index + 1] = (
1190
start, self._parsed_byte_map[index + 1][1])
1191
self._parsed_key_map[index + 1] = (
1192
start_key, self._parsed_key_map[index + 1][1])
1195
self._parsed_byte_map.insert(index + 1, new_value)
1196
self._parsed_key_map.insert(index + 1, new_key)
1198
def _read_and_parse(self, readv_ranges):
1199
"""Read the ranges and parse the resulting data.
1201
:param readv_ranges: A prepared readv range list.
1203
if not readv_ranges:
1205
if self._nodes is None and self._bytes_read * 2 >= self._size:
1206
# We've already read more than 50% of the file and we are about to
1207
# request more data, just _buffer_all() and be done
1211
base_offset = self._base_offset
1212
if base_offset != 0:
1213
# Rewrite the ranges for the offset
1214
readv_ranges = [(start+base_offset, size)
1215
for start, size in readv_ranges]
1216
readv_data = self._transport.readv(self._name, readv_ranges, True,
1217
self._size + self._base_offset)
1219
for offset, data in readv_data:
1220
offset -= base_offset
1221
self._bytes_read += len(data)
1223
# transport.readv() expanded to extra data which isn't part of
1225
data = data[-offset:]
1227
if offset == 0 and len(data) == self._size:
1228
# We read the whole range, most likely because the
1229
# Transport upcast our readv ranges into one long request
1230
# for enough total data to grab the whole index.
1231
self._buffer_all(StringIO(data))
1233
if self._bisect_nodes is None:
1234
# this must be the start
1235
if not (offset == 0):
1236
raise AssertionError()
1237
offset, data = self._parse_header_from_bytes(data)
1238
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1239
self._parse_region(offset, data)
1241
def _signature(self):
1242
"""The file signature for this index type."""
1246
"""Validate that everything in the index can be accessed."""
1247
# iter_all validates completely at the moment, so just do that.
1248
for node in self.iter_all_entries():
1252
class CombinedGraphIndex(object):
1253
"""A GraphIndex made up from smaller GraphIndices.
1255
The backing indices must implement GraphIndex, and are presumed to be
1258
Queries against the combined index will be made against the first index,
1259
and then the second and so on. The order of indices can thus influence
1260
performance significantly. For example, if one index is on local disk and a
1261
second on a remote server, the local disk index should be before the other
1264
Also, queries tend to need results from the same indices as previous
1265
queries. So the indices will be reordered after every query to put the
1266
indices that had the result(s) of that query first (while otherwise
1267
preserving the relative ordering).
1270
def __init__(self, indices, reload_func=None):
1271
"""Create a CombinedGraphIndex backed by indices.
1273
:param indices: An ordered list of indices to query for data.
1274
:param reload_func: A function to call if we find we are missing an
1275
index. Should have the form reload_func() => True/False to indicate
1276
if reloading actually changed anything.
1278
self._indices = indices
1279
self._reload_func = reload_func
1280
# Sibling indices are other CombinedGraphIndex that we should call
1281
# _move_to_front_by_name on when we auto-reorder ourself.
1282
self._sibling_indices = []
1283
# A list of names that corresponds to the instances in self._indices,
1284
# so _index_names[0] is always the name for _indices[0], etc. Sibling
1285
# indices must all use the same set of names as each other.
1286
self._index_names = [None] * len(self._indices)
1290
self.__class__.__name__,
1291
', '.join(map(repr, self._indices)))
1293
def clear_cache(self):
1294
"""See GraphIndex.clear_cache()"""
1295
for index in self._indices:
1298
def get_parent_map(self, keys):
1299
"""See graph.StackedParentsProvider.get_parent_map"""
1300
search_keys = set(keys)
1301
if _mod_revision.NULL_REVISION in search_keys:
1302
search_keys.discard(_mod_revision.NULL_REVISION)
1303
found_parents = {_mod_revision.NULL_REVISION:[]}
1306
for index, key, value, refs in self.iter_entries(search_keys):
1309
parents = (_mod_revision.NULL_REVISION,)
1310
found_parents[key] = parents
1311
return found_parents
1313
has_key = _has_key_from_parent_map
1315
def insert_index(self, pos, index, name=None):
1316
"""Insert a new index in the list of indices to query.
1318
:param pos: The position to insert the index.
1319
:param index: The index to insert.
1320
:param name: a name for this index, e.g. a pack name. These names can
1321
be used to reflect index reorderings to related CombinedGraphIndex
1322
instances that use the same names. (see set_sibling_indices)
1324
self._indices.insert(pos, index)
1325
self._index_names.insert(pos, name)
1327
def iter_all_entries(self):
1328
"""Iterate over all keys within the index
1330
Duplicate keys across child indices are presumed to have the same
1331
value and are only reported once.
1333
:return: An iterable of (index, key, reference_lists, value).
1334
There is no defined order for the result iteration - it will be in
1335
the most efficient order for the index.
1340
for index in self._indices:
1341
for node in index.iter_all_entries():
1342
if node[1] not in seen_keys:
1344
seen_keys.add(node[1])
1346
except errors.NoSuchFile:
1347
self._reload_or_raise()
1349
def iter_entries(self, keys):
1350
"""Iterate over keys within the index.
1352
Duplicate keys across child indices are presumed to have the same
1353
value and are only reported once.
1355
:param keys: An iterable providing the keys to be retrieved.
1356
:return: An iterable of (index, key, reference_lists, value). There is
1357
no defined order for the result iteration - it will be in the most
1358
efficient order for the index.
1364
for index in self._indices:
1368
for node in index.iter_entries(keys):
1369
keys.remove(node[1])
1373
hit_indices.append(index)
1375
except errors.NoSuchFile:
1376
self._reload_or_raise()
1377
self._move_to_front(hit_indices)
1379
def iter_entries_prefix(self, keys):
1380
"""Iterate over keys within the index using prefix matching.
1382
Duplicate keys across child indices are presumed to have the same
1383
value and are only reported once.
1385
Prefix matching is applied within the tuple of a key, not to within
1386
the bytestring of each key element. e.g. if you have the keys ('foo',
1387
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1388
only the former key is returned.
1390
:param keys: An iterable providing the key prefixes to be retrieved.
1391
Each key prefix takes the form of a tuple the length of a key, but
1392
with the last N elements 'None' rather than a regular bytestring.
1393
The first element cannot be 'None'.
1394
:return: An iterable as per iter_all_entries, but restricted to the
1395
keys with a matching prefix to those supplied. No additional keys
1396
will be returned, and every match that is in the index will be
1406
for index in self._indices:
1408
for node in index.iter_entries_prefix(keys):
1409
if node[1] in seen_keys:
1411
seen_keys.add(node[1])
1415
hit_indices.append(index)
1417
except errors.NoSuchFile:
1418
self._reload_or_raise()
1419
self._move_to_front(hit_indices)
1421
def _move_to_front(self, hit_indices):
1422
"""Rearrange self._indices so that hit_indices are first.
1424
Order is maintained as much as possible, e.g. the first unhit index
1425
will be the first index in _indices after the hit_indices, and the
1426
hit_indices will be present in exactly the order they are passed to
1429
_move_to_front propagates to all objects in self._sibling_indices by
1430
calling _move_to_front_by_name.
1432
if self._indices[:len(hit_indices)] == hit_indices:
1433
# The 'hit_indices' are already at the front (and in the same
1434
# order), no need to re-order
1436
hit_names = self._move_to_front_by_index(hit_indices)
1437
for sibling_idx in self._sibling_indices:
1438
sibling_idx._move_to_front_by_name(hit_names)
1440
def _move_to_front_by_index(self, hit_indices):
1441
"""Core logic for _move_to_front.
1443
Returns a list of names corresponding to the hit_indices param.
1445
indices_info = zip(self._index_names, self._indices)
1446
if 'index' in debug.debug_flags:
1447
trace.mutter('CombinedGraphIndex reordering: currently %r, '
1448
'promoting %r', indices_info, hit_indices)
1451
new_hit_indices = []
1454
for offset, (name, idx) in enumerate(indices_info):
1455
if idx in hit_indices:
1456
hit_names.append(name)
1457
new_hit_indices.append(idx)
1458
if len(new_hit_indices) == len(hit_indices):
1459
# We've found all of the hit entries, everything else is
1461
unhit_names.extend(self._index_names[offset+1:])
1462
unhit_indices.extend(self._indices[offset+1:])
1465
unhit_names.append(name)
1466
unhit_indices.append(idx)
1468
self._indices = new_hit_indices + unhit_indices
1469
self._index_names = hit_names + unhit_names
1470
if 'index' in debug.debug_flags:
1471
trace.mutter('CombinedGraphIndex reordered: %r', self._indices)
1474
def _move_to_front_by_name(self, hit_names):
1475
"""Moves indices named by 'hit_names' to front of the search order, as
1476
described in _move_to_front.
1478
# Translate names to index instances, and then call
1479
# _move_to_front_by_index.
1480
indices_info = zip(self._index_names, self._indices)
1482
for name, idx in indices_info:
1483
if name in hit_names:
1484
hit_indices.append(idx)
1485
self._move_to_front_by_index(hit_indices)
1487
def find_ancestry(self, keys, ref_list_num):
1488
"""Find the complete ancestry for the given set of keys.
1490
Note that this is a whole-ancestry request, so it should be used
1493
:param keys: An iterable of keys to look for
1494
:param ref_list_num: The reference list which references the parents
1496
:return: (parent_map, missing_keys)
1498
# XXX: make this call _move_to_front?
1499
missing_keys = set()
1501
keys_to_lookup = set(keys)
1503
while keys_to_lookup:
1504
# keys that *all* indexes claim are missing, stop searching them
1506
all_index_missing = None
1507
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1508
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1510
# len(missing_keys))
1511
for index_idx, index in enumerate(self._indices):
1512
# TODO: we should probably be doing something with
1513
# 'missing_keys' since we've already determined that
1514
# those revisions have not been found anywhere
1515
index_missing_keys = set()
1516
# Find all of the ancestry we can from this index
1517
# keep looking until the search_keys set is empty, which means
1518
# things we didn't find should be in index_missing_keys
1519
search_keys = keys_to_lookup
1521
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1522
# index_idx, len(search_keys),
1523
# len(parent_map), len(index_missing_keys))
1526
# TODO: ref_list_num should really be a parameter, since
1527
# CombinedGraphIndex does not know what the ref lists
1529
search_keys = index._find_ancestors(search_keys,
1530
ref_list_num, parent_map, index_missing_keys)
1531
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1532
# sub_generation, len(search_keys),
1533
# len(parent_map), len(index_missing_keys))
1534
# Now set whatever was missing to be searched in the next index
1535
keys_to_lookup = index_missing_keys
1536
if all_index_missing is None:
1537
all_index_missing = set(index_missing_keys)
1539
all_index_missing.intersection_update(index_missing_keys)
1540
if not keys_to_lookup:
1542
if all_index_missing is None:
1543
# There were no indexes, so all search keys are 'missing'
1544
missing_keys.update(keys_to_lookup)
1545
keys_to_lookup = None
1547
missing_keys.update(all_index_missing)
1548
keys_to_lookup.difference_update(all_index_missing)
1549
return parent_map, missing_keys
1551
def key_count(self):
1552
"""Return an estimate of the number of keys in this index.
1554
For CombinedGraphIndex this is approximated by the sum of the keys of
1555
the child indices. As child indices may have duplicate keys this can
1556
have a maximum error of the number of child indices * largest number of
1561
return sum((index.key_count() for index in self._indices), 0)
1562
except errors.NoSuchFile:
1563
self._reload_or_raise()
1565
missing_keys = _missing_keys_from_parent_map
1567
def _reload_or_raise(self):
1568
"""We just got a NoSuchFile exception.
1570
Try to reload the indices, if it fails, just raise the current
1573
if self._reload_func is None:
1575
exc_type, exc_value, exc_traceback = sys.exc_info()
1576
trace.mutter('Trying to reload after getting exception: %s',
1578
if not self._reload_func():
1579
# We tried to reload, but nothing changed, so we fail anyway
1580
trace.mutter('_reload_func indicated nothing has changed.'
1581
' Raising original exception.')
1582
raise exc_type, exc_value, exc_traceback
1584
def set_sibling_indices(self, sibling_combined_graph_indices):
1585
"""Set the CombinedGraphIndex objects to reorder after reordering self.
1587
self._sibling_indices = sibling_combined_graph_indices
1590
"""Validate that everything in the index can be accessed."""
1593
for index in self._indices:
1596
except errors.NoSuchFile:
1597
self._reload_or_raise()
1600
class InMemoryGraphIndex(GraphIndexBuilder):
1601
"""A GraphIndex which operates entirely out of memory and is mutable.
1603
This is designed to allow the accumulation of GraphIndex entries during a
1604
single write operation, where the accumulated entries need to be immediately
1605
available - for example via a CombinedGraphIndex.
1608
def add_nodes(self, nodes):
1609
"""Add nodes to the index.
1611
:param nodes: An iterable of (key, node_refs, value) entries to add.
1613
if self.reference_lists:
1614
for (key, value, node_refs) in nodes:
1615
self.add_node(key, value, node_refs)
1617
for (key, value) in nodes:
1618
self.add_node(key, value)
1620
def iter_all_entries(self):
1621
"""Iterate over all keys within the index
1623
:return: An iterable of (index, key, reference_lists, value). There is no
1624
defined order for the result iteration - it will be in the most
1625
efficient order for the index (in this case dictionary hash order).
1627
if 'evil' in debug.debug_flags:
1628
trace.mutter_callsite(3,
1629
"iter_all_entries scales with size of history.")
1630
if self.reference_lists:
1631
for key, (absent, references, value) in self._nodes.iteritems():
1633
yield self, key, value, references
1635
for key, (absent, references, value) in self._nodes.iteritems():
1637
yield self, key, value
1639
def iter_entries(self, keys):
1640
"""Iterate over keys within the index.
1642
:param keys: An iterable providing the keys to be retrieved.
1643
:return: An iterable of (index, key, value, reference_lists). There is no
1644
defined order for the result iteration - it will be in the most
1645
efficient order for the index (keys iteration order in this case).
1647
# Note: See BTreeBuilder.iter_entries for an explanation of why we
1648
# aren't using set().intersection() here
1650
keys = [key for key in keys if key in nodes]
1651
if self.reference_lists:
1655
yield self, key, node[2], node[1]
1660
yield self, key, node[2]
1662
def iter_entries_prefix(self, keys):
1663
"""Iterate over keys within the index using prefix matching.
1665
Prefix matching is applied within the tuple of a key, not to within
1666
the bytestring of each key element. e.g. if you have the keys ('foo',
1667
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1668
only the former key is returned.
1670
:param keys: An iterable providing the key prefixes to be retrieved.
1671
Each key prefix takes the form of a tuple the length of a key, but
1672
with the last N elements 'None' rather than a regular bytestring.
1673
The first element cannot be 'None'.
1674
:return: An iterable as per iter_all_entries, but restricted to the
1675
keys with a matching prefix to those supplied. No additional keys
1676
will be returned, and every match that is in the index will be
1679
# XXX: To much duplication with the GraphIndex class; consider finding
1680
# a good place to pull out the actual common logic.
1684
if self._key_length == 1:
1688
raise errors.BadIndexKey(key)
1689
if len(key) != self._key_length:
1690
raise errors.BadIndexKey(key)
1691
node = self._nodes[key]
1694
if self.reference_lists:
1695
yield self, key, node[2], node[1]
1697
yield self, key, node[2]
1699
nodes_by_key = self._get_nodes_by_key()
1703
raise errors.BadIndexKey(key)
1704
if len(key) != self._key_length:
1705
raise errors.BadIndexKey(key)
1706
# find what it refers to:
1707
key_dict = nodes_by_key
1708
elements = list(key)
1709
# find the subdict to return
1711
while len(elements) and elements[0] is not None:
1712
key_dict = key_dict[elements[0]]
1715
# a non-existant lookup.
1720
key_dict = dicts.pop(-1)
1721
# can't be empty or would not exist
1722
item, value = key_dict.iteritems().next()
1723
if type(value) == dict:
1725
dicts.extend(key_dict.itervalues())
1728
for value in key_dict.itervalues():
1729
yield (self, ) + value
1731
yield (self, ) + key_dict
1733
def key_count(self):
1734
"""Return an estimate of the number of keys in this index.
1736
For InMemoryGraphIndex the estimate is exact.
1738
return len(self._nodes) - len(self._absent_keys)
1741
"""In memory index's have no known corruption at the moment."""
1744
class GraphIndexPrefixAdapter(object):
1745
"""An adapter between GraphIndex with different key lengths.
1747
Queries against this will emit queries against the adapted Graph with the
1748
prefix added, queries for all items use iter_entries_prefix. The returned
1749
nodes will have their keys and node references adjusted to remove the
1750
prefix. Finally, an add_nodes_callback can be supplied - when called the
1751
nodes and references being added will have prefix prepended.
1754
def __init__(self, adapted, prefix, missing_key_length,
1755
add_nodes_callback=None):
1756
"""Construct an adapter against adapted with prefix."""
1757
self.adapted = adapted
1758
self.prefix_key = prefix + (None,)*missing_key_length
1759
self.prefix = prefix
1760
self.prefix_len = len(prefix)
1761
self.add_nodes_callback = add_nodes_callback
1763
def add_nodes(self, nodes):
1764
"""Add nodes to the index.
1766
:param nodes: An iterable of (key, node_refs, value) entries to add.
1768
# save nodes in case its an iterator
1769
nodes = tuple(nodes)
1770
translated_nodes = []
1772
# Add prefix_key to each reference node_refs is a tuple of tuples,
1773
# so split it apart, and add prefix_key to the internal reference
1774
for (key, value, node_refs) in nodes:
1775
adjusted_references = (
1776
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1777
for ref_list in node_refs))
1778
translated_nodes.append((self.prefix + key, value,
1779
adjusted_references))
1781
# XXX: TODO add an explicit interface for getting the reference list
1782
# status, to handle this bit of user-friendliness in the API more
1784
for (key, value) in nodes:
1785
translated_nodes.append((self.prefix + key, value))
1786
self.add_nodes_callback(translated_nodes)
1788
def add_node(self, key, value, references=()):
1789
"""Add a node to the index.
1791
:param key: The key. keys are non-empty tuples containing
1792
as many whitespace-free utf8 bytestrings as the key length
1793
defined for this index.
1794
:param references: An iterable of iterables of keys. Each is a
1795
reference to another key.
1796
:param value: The value to associate with the key. It may be any
1797
bytes as long as it does not contain \0 or \n.
1799
self.add_nodes(((key, value, references), ))
1801
def _strip_prefix(self, an_iter):
1802
"""Strip prefix data from nodes and return it."""
1803
for node in an_iter:
1805
if node[1][:self.prefix_len] != self.prefix:
1806
raise errors.BadIndexData(self)
1807
for ref_list in node[3]:
1808
for ref_node in ref_list:
1809
if ref_node[:self.prefix_len] != self.prefix:
1810
raise errors.BadIndexData(self)
1811
yield node[0], node[1][self.prefix_len:], node[2], (
1812
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1813
for ref_list in node[3]))
1815
def iter_all_entries(self):
1816
"""Iterate over all keys within the index
1818
iter_all_entries is implemented against the adapted index using
1819
iter_entries_prefix.
1821
:return: An iterable of (index, key, reference_lists, value). There is no
1822
defined order for the result iteration - it will be in the most
1823
efficient order for the index (in this case dictionary hash order).
1825
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1827
def iter_entries(self, keys):
1828
"""Iterate over keys within the index.
1830
:param keys: An iterable providing the keys to be retrieved.
1831
:return: An iterable of (index, key, value, reference_lists). There is no
1832
defined order for the result iteration - it will be in the most
1833
efficient order for the index (keys iteration order in this case).
1835
return self._strip_prefix(self.adapted.iter_entries(
1836
self.prefix + key for key in keys))
1838
def iter_entries_prefix(self, keys):
1839
"""Iterate over keys within the index using prefix matching.
1841
Prefix matching is applied within the tuple of a key, not to within
1842
the bytestring of each key element. e.g. if you have the keys ('foo',
1843
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1844
only the former key is returned.
1846
:param keys: An iterable providing the key prefixes to be retrieved.
1847
Each key prefix takes the form of a tuple the length of a key, but
1848
with the last N elements 'None' rather than a regular bytestring.
1849
The first element cannot be 'None'.
1850
:return: An iterable as per iter_all_entries, but restricted to the
1851
keys with a matching prefix to those supplied. No additional keys
1852
will be returned, and every match that is in the index will be
1855
return self._strip_prefix(self.adapted.iter_entries_prefix(
1856
self.prefix + key for key in keys))
1858
def key_count(self):
1859
"""Return an estimate of the number of keys in this index.
1861
For GraphIndexPrefixAdapter this is relatively expensive - key
1862
iteration with the prefix is done.
1864
return len(list(self.iter_all_entries()))
1867
"""Call the adapted's validate."""
1868
self.adapted.validate()
added
removed
bzrlib/index.py
