1
# Copyright (C) 2007-2011 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
"""Indexing facilities."""
23
'GraphIndexPrefixAdapter',
27
from bisect import bisect_right
28
from cStringIO import StringIO
32
from bzrlib.lazy_import import lazy_import
33
lazy_import(globals(), """
36
revision as _mod_revision,
44
from bzrlib.static_tuple import StaticTuple
46
_HEADER_READV = (0, 200)
47
_OPTION_KEY_ELEMENTS = "key_elements="
49
_OPTION_NODE_REFS = "node_ref_lists="
50
_SIGNATURE = "Bazaar Graph Index 1\n"
53
_whitespace_re = re.compile('[\t\n\x0b\x0c\r\x00 ]')
54
_newline_null_re = re.compile('[\n\0]')
57
def _has_key_from_parent_map(self, key):
58
"""Check if this index has one key.
60
If it's possible to check for multiple keys at once through
61
calling get_parent_map that should be faster.
63
return (key in self.get_parent_map([key]))
66
def _missing_keys_from_parent_map(self, keys):
67
return set(keys) - set(self.get_parent_map(keys))
70
class GraphIndexBuilder(object):
71
"""A builder that can build a GraphIndex.
73
The resulting graph has the structure::
75
_SIGNATURE OPTIONS NODES NEWLINE
76
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
77
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
79
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
80
KEY := Not-whitespace-utf8
82
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
83
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
84
REFERENCE := DIGITS ; digits is the byte offset in the index of the
86
VALUE := no-newline-no-null-bytes
89
def __init__(self, reference_lists=0, key_elements=1):
90
"""Create a GraphIndex builder.
92
:param reference_lists: The number of node references lists for each
94
:param key_elements: The number of bytestrings in each key.
96
self.reference_lists = reference_lists
97
# A dict of {key: (absent, ref_lists, value)}
99
# Keys that are referenced but not actually present in this index
100
self._absent_keys = set()
101
self._nodes_by_key = None
102
self._key_length = key_elements
103
self._optimize_for_size = False
104
self._combine_backing_indices = True
106
def _check_key(self, key):
107
"""Raise BadIndexKey if key is not a valid key for this index."""
108
if type(key) not in (tuple, StaticTuple):
109
raise errors.BadIndexKey(key)
110
if self._key_length != len(key):
111
raise errors.BadIndexKey(key)
113
if not element or _whitespace_re.search(element) is not None:
114
raise errors.BadIndexKey(element)
116
def _external_references(self):
117
"""Return references that are not present in this index.
121
# TODO: JAM 2008-11-21 This makes an assumption about how the reference
122
# lists are used. It is currently correct for pack-0.92 through
123
# 1.9, which use the node references (3rd column) second
124
# reference list as the compression parent. Perhaps this should
125
# be moved into something higher up the stack, since it
126
# makes assumptions about how the index is used.
127
if self.reference_lists > 1:
128
for node in self.iter_all_entries():
130
refs.update(node[3][1])
133
# If reference_lists == 0 there can be no external references, and
134
# if reference_lists == 1, then there isn't a place to store the
138
def _get_nodes_by_key(self):
139
if self._nodes_by_key is None:
141
if self.reference_lists:
142
for key, (absent, references, value) in self._nodes.iteritems():
145
key_dict = nodes_by_key
146
for subkey in key[:-1]:
147
key_dict = key_dict.setdefault(subkey, {})
148
key_dict[key[-1]] = key, value, references
150
for key, (absent, references, value) in self._nodes.iteritems():
153
key_dict = nodes_by_key
154
for subkey in key[:-1]:
155
key_dict = key_dict.setdefault(subkey, {})
156
key_dict[key[-1]] = key, value
157
self._nodes_by_key = nodes_by_key
158
return self._nodes_by_key
160
def _update_nodes_by_key(self, key, value, node_refs):
161
"""Update the _nodes_by_key dict with a new key.
163
For a key of (foo, bar, baz) create
164
_nodes_by_key[foo][bar][baz] = key_value
166
if self._nodes_by_key is None:
168
key_dict = self._nodes_by_key
169
if self.reference_lists:
170
key_value = StaticTuple(key, value, node_refs)
172
key_value = StaticTuple(key, value)
173
for subkey in key[:-1]:
174
key_dict = key_dict.setdefault(subkey, {})
175
key_dict[key[-1]] = key_value
177
def _check_key_ref_value(self, key, references, value):
178
"""Check that 'key' and 'references' are all valid.
180
:param key: A key tuple. Must conform to the key interface (be a tuple,
181
be of the right length, not have any whitespace or nulls in any key
183
:param references: An iterable of reference lists. Something like
184
[[(ref, key)], [(ref, key), (other, key)]]
185
:param value: The value associate with this key. Must not contain
186
newlines or null characters.
187
:return: (node_refs, absent_references)
189
* node_refs: basically a packed form of 'references' where all
191
* absent_references: reference keys that are not in self._nodes.
192
This may contain duplicates if the same key is referenced in
195
as_st = StaticTuple.from_sequence
197
if _newline_null_re.search(value) is not None:
198
raise errors.BadIndexValue(value)
199
if len(references) != self.reference_lists:
200
raise errors.BadIndexValue(references)
202
absent_references = []
203
for reference_list in references:
204
for reference in reference_list:
205
# If reference *is* in self._nodes, then we know it has already
207
if reference not in self._nodes:
208
self._check_key(reference)
209
absent_references.append(reference)
210
reference_list = as_st([as_st(ref).intern()
211
for ref in reference_list])
212
node_refs.append(reference_list)
213
return as_st(node_refs), absent_references
215
def add_node(self, key, value, references=()):
216
"""Add a node to the index.
218
:param key: The key. keys are non-empty tuples containing
219
as many whitespace-free utf8 bytestrings as the key length
220
defined for this index.
221
:param references: An iterable of iterables of keys. Each is a
222
reference to another key.
223
:param value: The value to associate with the key. It may be any
224
bytes as long as it does not contain \\0 or \\n.
227
absent_references) = self._check_key_ref_value(key, references, value)
228
if key in self._nodes and self._nodes[key][0] != 'a':
229
raise errors.BadIndexDuplicateKey(key, self)
230
for reference in absent_references:
231
# There may be duplicates, but I don't think it is worth worrying
233
self._nodes[reference] = ('a', (), '')
234
self._absent_keys.update(absent_references)
235
self._absent_keys.discard(key)
236
self._nodes[key] = ('', node_refs, value)
237
if self._nodes_by_key is not None and self._key_length > 1:
238
self._update_nodes_by_key(key, value, node_refs)
240
def clear_cache(self):
241
"""See GraphIndex.clear_cache()
243
This is a no-op, but we need the api to conform to a generic 'Index'
250
:returns: cStringIO holding the full context of the index as it
251
should be written to disk.
254
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
255
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
256
key_count = len(self._nodes) - len(self._absent_keys)
257
lines.append(_OPTION_LEN + str(key_count) + '\n')
258
prefix_length = sum(len(x) for x in lines)
259
# references are byte offsets. To avoid having to do nasty
260
# polynomial work to resolve offsets (references to later in the
261
# file cannot be determined until all the inbetween references have
262
# been calculated too) we pad the offsets with 0's to make them be
263
# of consistent length. Using binary offsets would break the trivial
265
# to calculate the width of zero's needed we do three passes:
266
# one to gather all the non-reference data and the number of references.
267
# one to pad all the data with reference-length and determine entry
271
# forward sorted by key. In future we may consider topological sorting,
272
# at the cost of table scans for direct lookup, or a second index for
274
nodes = sorted(self._nodes.items())
275
# if we do not prepass, we don't know how long it will be up front.
276
expected_bytes = None
277
# we only need to pre-pass if we have reference lists at all.
278
if self.reference_lists:
280
non_ref_bytes = prefix_length
282
# TODO use simple multiplication for the constants in this loop.
283
for key, (absent, references, value) in nodes:
284
# record the offset known *so far* for this key:
285
# the non reference bytes to date, and the total references to
286
# date - saves reaccumulating on the second pass
287
key_offset_info.append((key, non_ref_bytes, total_references))
288
# key is literal, value is literal, there are 3 null's, 1 NL
289
# key is variable length tuple, \x00 between elements
290
non_ref_bytes += sum(len(element) for element in key)
291
if self._key_length > 1:
292
non_ref_bytes += self._key_length - 1
293
# value is literal bytes, there are 3 null's, 1 NL.
294
non_ref_bytes += len(value) + 3 + 1
295
# one byte for absent if set.
298
elif self.reference_lists:
299
# (ref_lists -1) tabs
300
non_ref_bytes += self.reference_lists - 1
301
# (ref-1 cr's per ref_list)
302
for ref_list in references:
303
# how many references across the whole file?
304
total_references += len(ref_list)
305
# accrue reference separators
307
non_ref_bytes += len(ref_list) - 1
308
# how many digits are needed to represent the total byte count?
310
possible_total_bytes = non_ref_bytes + total_references*digits
311
while 10 ** digits < possible_total_bytes:
313
possible_total_bytes = non_ref_bytes + total_references*digits
314
expected_bytes = possible_total_bytes + 1 # terminating newline
315
# resolve key addresses.
317
for key, non_ref_bytes, total_references in key_offset_info:
318
key_addresses[key] = non_ref_bytes + total_references*digits
320
format_string = '%%0%sd' % digits
321
for key, (absent, references, value) in nodes:
322
flattened_references = []
323
for ref_list in references:
325
for reference in ref_list:
326
ref_addresses.append(format_string % key_addresses[reference])
327
flattened_references.append('\r'.join(ref_addresses))
328
string_key = '\x00'.join(key)
329
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
330
'\t'.join(flattened_references), value))
332
result = StringIO(''.join(lines))
333
if expected_bytes and len(result.getvalue()) != expected_bytes:
334
raise errors.BzrError('Failed index creation. Internal error:'
335
' mismatched output length and expected length: %d %d' %
336
(len(result.getvalue()), expected_bytes))
339
def set_optimize(self, for_size=None, combine_backing_indices=None):
340
"""Change how the builder tries to optimize the result.
342
:param for_size: Tell the builder to try and make the index as small as
344
:param combine_backing_indices: If the builder spills to disk to save
345
memory, should the on-disk indices be combined. Set to True if you
346
are going to be probing the index, but to False if you are not. (If
347
you are not querying, then the time spent combining is wasted.)
350
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
352
if for_size is not None:
353
self._optimize_for_size = for_size
354
if combine_backing_indices is not None:
355
self._combine_backing_indices = combine_backing_indices
357
def find_ancestry(self, keys, ref_list_num):
358
"""See CombinedGraphIndex.find_ancestry()"""
364
for _, key, value, ref_lists in self.iter_entries(pending):
365
parent_keys = ref_lists[ref_list_num]
366
parent_map[key] = parent_keys
367
next_pending.update([p for p in parent_keys if p not in
369
missing_keys.update(pending.difference(parent_map))
370
pending = next_pending
371
return parent_map, missing_keys
374
class GraphIndex(object):
375
"""An index for data with embedded graphs.
377
The index maps keys to a list of key reference lists, and a value.
378
Each node has the same number of key reference lists. Each key reference
379
list can be empty or an arbitrary length. The value is an opaque NULL
380
terminated string without any newlines. The storage of the index is
381
hidden in the interface: keys and key references are always tuples of
382
bytestrings, never the internal representation (e.g. dictionary offsets).
384
It is presumed that the index will not be mutated - it is static data.
386
Successive iter_all_entries calls will read the entire index each time.
387
Additionally, iter_entries calls will read the index linearly until the
388
desired keys are found. XXX: This must be fixed before the index is
389
suitable for production use. :XXX
392
def __init__(self, transport, name, size, unlimited_cache=False, offset=0):
393
"""Open an index called name on transport.
395
:param transport: A bzrlib.transport.Transport.
396
:param name: A path to provide to transport API calls.
397
:param size: The size of the index in bytes. This is used for bisection
398
logic to perform partial index reads. While the size could be
399
obtained by statting the file this introduced an additional round
400
trip as well as requiring stat'able transports, both of which are
401
avoided by having it supplied. If size is None, then bisection
402
support will be disabled and accessing the index will just stream
404
:param offset: Instead of starting the index data at offset 0, start it
405
at an arbitrary offset.
407
self._transport = transport
409
# Becomes a dict of key:(value, reference-list-byte-locations) used by
410
# the bisection interface to store parsed but not resolved keys.
411
self._bisect_nodes = None
412
# Becomes a dict of key:(value, reference-list-keys) which are ready to
413
# be returned directly to callers.
415
# a sorted list of slice-addresses for the parsed bytes of the file.
416
# e.g. (0,1) would mean that byte 0 is parsed.
417
self._parsed_byte_map = []
418
# a sorted list of keys matching each slice address for parsed bytes
419
# e.g. (None, 'foo@bar') would mean that the first byte contained no
420
# key, and the end byte of the slice is the of the data for 'foo@bar'
421
self._parsed_key_map = []
422
self._key_count = None
423
self._keys_by_offset = None
424
self._nodes_by_key = None
426
# The number of bytes we've read so far in trying to process this file
428
self._base_offset = offset
430
def __eq__(self, other):
431
"""Equal when self and other were created with the same parameters."""
433
type(self) == type(other) and
434
self._transport == other._transport and
435
self._name == other._name and
436
self._size == other._size)
438
def __ne__(self, other):
439
return not self.__eq__(other)
442
return "%s(%r)" % (self.__class__.__name__,
443
self._transport.abspath(self._name))
445
def _buffer_all(self, stream=None):
446
"""Buffer all the index data.
448
Mutates self._nodes and self.keys_by_offset.
450
if self._nodes is not None:
451
# We already did this
453
if 'index' in debug.debug_flags:
454
trace.mutter('Reading entire index %s',
455
self._transport.abspath(self._name))
457
stream = self._transport.get(self._name)
458
if self._base_offset != 0:
459
# This is wasteful, but it is better than dealing with
460
# adjusting all the offsets, etc.
461
stream = StringIO(stream.read()[self._base_offset:])
462
self._read_prefix(stream)
463
self._expected_elements = 3 + self._key_length
465
# raw data keyed by offset
466
self._keys_by_offset = {}
467
# ready-to-return key:value or key:value, node_ref_lists
469
self._nodes_by_key = None
472
lines = stream.read().split('\n')
473
# GZ 2009-09-20: Should really use a try/finally block to ensure close
476
_, _, _, trailers = self._parse_lines(lines, pos)
477
for key, absent, references, value in self._keys_by_offset.itervalues():
480
# resolve references:
481
if self.node_ref_lists:
482
node_value = (value, self._resolve_references(references))
485
self._nodes[key] = node_value
486
# cache the keys for quick set intersections
488
# there must be one line - the empty trailer line.
489
raise errors.BadIndexData(self)
491
def clear_cache(self):
492
"""Clear out any cached/memoized values.
494
This can be called at any time, but generally it is used when we have
495
extracted some information, but don't expect to be requesting any more
499
def external_references(self, ref_list_num):
500
"""Return references that are not present in this index.
503
if ref_list_num + 1 > self.node_ref_lists:
504
raise ValueError('No ref list %d, index has %d ref lists'
505
% (ref_list_num, self.node_ref_lists))
508
for key, (value, ref_lists) in nodes.iteritems():
509
ref_list = ref_lists[ref_list_num]
510
refs.update([ref for ref in ref_list if ref not in nodes])
513
def _get_nodes_by_key(self):
514
if self._nodes_by_key is None:
516
if self.node_ref_lists:
517
for key, (value, references) in self._nodes.iteritems():
518
key_dict = nodes_by_key
519
for subkey in key[:-1]:
520
key_dict = key_dict.setdefault(subkey, {})
521
key_dict[key[-1]] = key, value, references
523
for key, value in self._nodes.iteritems():
524
key_dict = nodes_by_key
525
for subkey in key[:-1]:
526
key_dict = key_dict.setdefault(subkey, {})
527
key_dict[key[-1]] = key, value
528
self._nodes_by_key = nodes_by_key
529
return self._nodes_by_key
531
def iter_all_entries(self):
532
"""Iterate over all keys within the index.
534
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
535
The former tuple is used when there are no reference lists in the
536
index, making the API compatible with simple key:value index types.
537
There is no defined order for the result iteration - it will be in
538
the most efficient order for the index.
540
if 'evil' in debug.debug_flags:
541
trace.mutter_callsite(3,
542
"iter_all_entries scales with size of history.")
543
if self._nodes is None:
545
if self.node_ref_lists:
546
for key, (value, node_ref_lists) in self._nodes.iteritems():
547
yield self, key, value, node_ref_lists
549
for key, value in self._nodes.iteritems():
550
yield self, key, value
552
def _read_prefix(self, stream):
553
signature = stream.read(len(self._signature()))
554
if not signature == self._signature():
555
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
556
options_line = stream.readline()
557
if not options_line.startswith(_OPTION_NODE_REFS):
558
raise errors.BadIndexOptions(self)
560
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
562
raise errors.BadIndexOptions(self)
563
options_line = stream.readline()
564
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
565
raise errors.BadIndexOptions(self)
567
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
569
raise errors.BadIndexOptions(self)
570
options_line = stream.readline()
571
if not options_line.startswith(_OPTION_LEN):
572
raise errors.BadIndexOptions(self)
574
self._key_count = int(options_line[len(_OPTION_LEN):-1])
576
raise errors.BadIndexOptions(self)
578
def _resolve_references(self, references):
579
"""Return the resolved key references for references.
581
References are resolved by looking up the location of the key in the
582
_keys_by_offset map and substituting the key name, preserving ordering.
584
:param references: An iterable of iterables of key locations. e.g.
586
:return: A tuple of tuples of keys.
589
for ref_list in references:
590
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
591
return tuple(node_refs)
593
def _find_index(self, range_map, key):
594
"""Helper for the _parsed_*_index calls.
596
Given a range map - [(start, end), ...], finds the index of the range
597
in the map for key if it is in the map, and if it is not there, the
598
immediately preceeding range in the map.
600
result = bisect_right(range_map, key) - 1
601
if result + 1 < len(range_map):
602
# check the border condition, it may be in result + 1
603
if range_map[result + 1][0] == key[0]:
607
def _parsed_byte_index(self, offset):
608
"""Return the index of the entry immediately before offset.
610
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
611
there is one unparsed byte (the 11th, addressed as[10]). then:
612
asking for 0 will return 0
613
asking for 10 will return 0
614
asking for 11 will return 1
615
asking for 12 will return 1
618
return self._find_index(self._parsed_byte_map, key)
620
def _parsed_key_index(self, key):
621
"""Return the index of the entry immediately before key.
623
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
624
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
625
have been parsed, then:
626
asking for '' will return 0
627
asking for 'a' will return 0
628
asking for 'b' will return 1
629
asking for 'e' will return 1
631
search_key = (key, None)
632
return self._find_index(self._parsed_key_map, search_key)
634
def _is_parsed(self, offset):
635
"""Returns True if offset has been parsed."""
636
index = self._parsed_byte_index(offset)
637
if index == len(self._parsed_byte_map):
638
return offset < self._parsed_byte_map[index - 1][1]
639
start, end = self._parsed_byte_map[index]
640
return offset >= start and offset < end
642
def _iter_entries_from_total_buffer(self, keys):
643
"""Iterate over keys when the entire index is parsed."""
644
# Note: See the note in BTreeBuilder.iter_entries for why we don't use
645
# .intersection() here
647
keys = [key for key in keys if key in nodes]
648
if self.node_ref_lists:
650
value, node_refs = nodes[key]
651
yield self, key, value, node_refs
654
yield self, key, nodes[key]
656
def iter_entries(self, keys):
657
"""Iterate over keys within the index.
659
:param keys: An iterable providing the keys to be retrieved.
660
:return: An iterable as per iter_all_entries, but restricted to the
661
keys supplied. No additional keys will be returned, and every
662
key supplied that is in the index will be returned.
667
if self._size is None and self._nodes is None:
670
# We fit about 20 keys per minimum-read (4K), so if we are looking for
671
# more than 1/20th of the index its likely (assuming homogenous key
672
# spread) that we'll read the entire index. If we're going to do that,
673
# buffer the whole thing. A better analysis might take key spread into
674
# account - but B+Tree indices are better anyway.
675
# We could look at all data read, and use a threshold there, which will
676
# trigger on ancestry walks, but that is not yet fully mapped out.
677
if self._nodes is None and len(keys) * 20 > self.key_count():
679
if self._nodes is not None:
680
return self._iter_entries_from_total_buffer(keys)
682
return (result[1] for result in bisect_multi.bisect_multi_bytes(
683
self._lookup_keys_via_location, self._size, keys))
685
def iter_entries_prefix(self, keys):
686
"""Iterate over keys within the index using prefix matching.
688
Prefix matching is applied within the tuple of a key, not to within
689
the bytestring of each key element. e.g. if you have the keys ('foo',
690
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
691
only the former key is returned.
693
WARNING: Note that this method currently causes a full index parse
694
unconditionally (which is reasonably appropriate as it is a means for
695
thunking many small indices into one larger one and still supplies
696
iter_all_entries at the thunk layer).
698
:param keys: An iterable providing the key prefixes to be retrieved.
699
Each key prefix takes the form of a tuple the length of a key, but
700
with the last N elements 'None' rather than a regular bytestring.
701
The first element cannot be 'None'.
702
:return: An iterable as per iter_all_entries, but restricted to the
703
keys with a matching prefix to those supplied. No additional keys
704
will be returned, and every match that is in the index will be
710
# load data - also finds key lengths
711
if self._nodes is None:
713
if self._key_length == 1:
717
raise errors.BadIndexKey(key)
718
if len(key) != self._key_length:
719
raise errors.BadIndexKey(key)
720
if self.node_ref_lists:
721
value, node_refs = self._nodes[key]
722
yield self, key, value, node_refs
724
yield self, key, self._nodes[key]
726
nodes_by_key = self._get_nodes_by_key()
730
raise errors.BadIndexKey(key)
731
if len(key) != self._key_length:
732
raise errors.BadIndexKey(key)
733
# find what it refers to:
734
key_dict = nodes_by_key
736
# find the subdict whose contents should be returned.
738
while len(elements) and elements[0] is not None:
739
key_dict = key_dict[elements[0]]
742
# a non-existant lookup.
747
key_dict = dicts.pop(-1)
748
# can't be empty or would not exist
749
item, value = key_dict.iteritems().next()
750
if type(value) == dict:
752
dicts.extend(key_dict.itervalues())
755
for value in key_dict.itervalues():
756
# each value is the key:value:node refs tuple
758
yield (self, ) + value
760
# the last thing looked up was a terminal element
761
yield (self, ) + key_dict
763
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
764
"""See BTreeIndex._find_ancestors."""
765
# The api can be implemented as a trivial overlay on top of
766
# iter_entries, it is not an efficient implementation, but it at least
770
for index, key, value, refs in self.iter_entries(keys):
771
parent_keys = refs[ref_list_num]
773
parent_map[key] = parent_keys
774
search_keys.update(parent_keys)
775
# Figure out what, if anything, was missing
776
missing_keys.update(set(keys).difference(found_keys))
777
search_keys = search_keys.difference(parent_map)
781
"""Return an estimate of the number of keys in this index.
783
For GraphIndex the estimate is exact.
785
if self._key_count is None:
786
self._read_and_parse([_HEADER_READV])
787
return self._key_count
789
def _lookup_keys_via_location(self, location_keys):
790
"""Public interface for implementing bisection.
792
If _buffer_all has been called, then all the data for the index is in
793
memory, and this method should not be called, as it uses a separate
794
cache because it cannot pre-resolve all indices, which buffer_all does
797
:param location_keys: A list of location(byte offset), key tuples.
798
:return: A list of (location_key, result) tuples as expected by
799
bzrlib.bisect_multi.bisect_multi_bytes.
801
# Possible improvements:
802
# - only bisect lookup each key once
803
# - sort the keys first, and use that to reduce the bisection window
805
# this progresses in three parts:
808
# attempt to answer the question from the now in memory data.
809
# build the readv request
810
# for each location, ask for 800 bytes - much more than rows we've seen
813
for location, key in location_keys:
814
# can we answer from cache?
815
if self._bisect_nodes and key in self._bisect_nodes:
816
# We have the key parsed.
818
index = self._parsed_key_index(key)
819
if (len(self._parsed_key_map) and
820
self._parsed_key_map[index][0] <= key and
821
(self._parsed_key_map[index][1] >= key or
822
# end of the file has been parsed
823
self._parsed_byte_map[index][1] == self._size)):
824
# the key has been parsed, so no lookup is needed even if its
827
# - if we have examined this part of the file already - yes
828
index = self._parsed_byte_index(location)
829
if (len(self._parsed_byte_map) and
830
self._parsed_byte_map[index][0] <= location and
831
self._parsed_byte_map[index][1] > location):
832
# the byte region has been parsed, so no read is needed.
835
if location + length > self._size:
836
length = self._size - location
837
# todo, trim out parsed locations.
839
readv_ranges.append((location, length))
840
# read the header if needed
841
if self._bisect_nodes is None:
842
readv_ranges.append(_HEADER_READV)
843
self._read_and_parse(readv_ranges)
845
if self._nodes is not None:
846
# _read_and_parse triggered a _buffer_all because we requested the
848
for location, key in location_keys:
849
if key not in self._nodes: # not present
850
result.append(((location, key), False))
851
elif self.node_ref_lists:
852
value, refs = self._nodes[key]
853
result.append(((location, key),
854
(self, key, value, refs)))
856
result.append(((location, key),
857
(self, key, self._nodes[key])))
860
# - figure out <, >, missing, present
861
# - result present references so we can return them.
862
# keys that we cannot answer until we resolve references
863
pending_references = []
864
pending_locations = set()
865
for location, key in location_keys:
866
# can we answer from cache?
867
if key in self._bisect_nodes:
868
# the key has been parsed, so no lookup is needed
869
if self.node_ref_lists:
870
# the references may not have been all parsed.
871
value, refs = self._bisect_nodes[key]
872
wanted_locations = []
873
for ref_list in refs:
875
if ref not in self._keys_by_offset:
876
wanted_locations.append(ref)
878
pending_locations.update(wanted_locations)
879
pending_references.append((location, key))
881
result.append(((location, key), (self, key,
882
value, self._resolve_references(refs))))
884
result.append(((location, key),
885
(self, key, self._bisect_nodes[key])))
888
# has the region the key should be in, been parsed?
889
index = self._parsed_key_index(key)
890
if (self._parsed_key_map[index][0] <= key and
891
(self._parsed_key_map[index][1] >= key or
892
# end of the file has been parsed
893
self._parsed_byte_map[index][1] == self._size)):
894
result.append(((location, key), False))
896
# no, is the key above or below the probed location:
897
# get the range of the probed & parsed location
898
index = self._parsed_byte_index(location)
899
# if the key is below the start of the range, its below
900
if key < self._parsed_key_map[index][0]:
904
result.append(((location, key), direction))
906
# lookup data to resolve references
907
for location in pending_locations:
909
if location + length > self._size:
910
length = self._size - location
911
# TODO: trim out parsed locations (e.g. if the 800 is into the
912
# parsed region trim it, and dont use the adjust_for_latency
915
readv_ranges.append((location, length))
916
self._read_and_parse(readv_ranges)
917
if self._nodes is not None:
918
# The _read_and_parse triggered a _buffer_all, grab the data and
920
for location, key in pending_references:
921
value, refs = self._nodes[key]
922
result.append(((location, key), (self, key, value, refs)))
924
for location, key in pending_references:
925
# answer key references we had to look-up-late.
926
value, refs = self._bisect_nodes[key]
927
result.append(((location, key), (self, key,
928
value, self._resolve_references(refs))))
931
def _parse_header_from_bytes(self, bytes):
932
"""Parse the header from a region of bytes.
934
:param bytes: The data to parse.
935
:return: An offset, data tuple such as readv yields, for the unparsed
936
data. (which may length 0).
938
signature = bytes[0:len(self._signature())]
939
if not signature == self._signature():
940
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
941
lines = bytes[len(self._signature()):].splitlines()
942
options_line = lines[0]
943
if not options_line.startswith(_OPTION_NODE_REFS):
944
raise errors.BadIndexOptions(self)
946
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
948
raise errors.BadIndexOptions(self)
949
options_line = lines[1]
950
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
951
raise errors.BadIndexOptions(self)
953
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
955
raise errors.BadIndexOptions(self)
956
options_line = lines[2]
957
if not options_line.startswith(_OPTION_LEN):
958
raise errors.BadIndexOptions(self)
960
self._key_count = int(options_line[len(_OPTION_LEN):])
962
raise errors.BadIndexOptions(self)
963
# calculate the bytes we have processed
964
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
966
self._parsed_bytes(0, None, header_end, None)
967
# setup parsing state
968
self._expected_elements = 3 + self._key_length
969
# raw data keyed by offset
970
self._keys_by_offset = {}
971
# keys with the value and node references
972
self._bisect_nodes = {}
973
return header_end, bytes[header_end:]
975
def _parse_region(self, offset, data):
976
"""Parse node data returned from a readv operation.
978
:param offset: The byte offset the data starts at.
979
:param data: The data to parse.
983
end = offset + len(data)
986
# Trivial test - if the current index's end is within the
987
# low-matching parsed range, we're done.
988
index = self._parsed_byte_index(high_parsed)
989
if end < self._parsed_byte_map[index][1]:
991
# print "[%d:%d]" % (offset, end), \
992
# self._parsed_byte_map[index:index + 2]
993
high_parsed, last_segment = self._parse_segment(
994
offset, data, end, index)
998
def _parse_segment(self, offset, data, end, index):
999
"""Parse one segment of data.
1001
:param offset: Where 'data' begins in the file.
1002
:param data: Some data to parse a segment of.
1003
:param end: Where data ends
1004
:param index: The current index into the parsed bytes map.
1005
:return: True if the parsed segment is the last possible one in the
1007
:return: high_parsed_byte, last_segment.
1008
high_parsed_byte is the location of the highest parsed byte in this
1009
segment, last_segment is True if the parsed segment is the last
1010
possible one in the data block.
1012
# default is to use all data
1014
# accomodate overlap with data before this.
1015
if offset < self._parsed_byte_map[index][1]:
1016
# overlaps the lower parsed region
1017
# skip the parsed data
1018
trim_start = self._parsed_byte_map[index][1] - offset
1019
# don't trim the start for \n
1020
start_adjacent = True
1021
elif offset == self._parsed_byte_map[index][1]:
1022
# abuts the lower parsed region
1025
# do not trim anything
1026
start_adjacent = True
1028
# does not overlap the lower parsed region
1031
# but trim the leading \n
1032
start_adjacent = False
1033
if end == self._size:
1034
# lines up to the end of all data:
1037
# do not strip to the last \n
1040
elif index + 1 == len(self._parsed_byte_map):
1041
# at the end of the parsed data
1044
# but strip to the last \n
1045
end_adjacent = False
1047
elif end == self._parsed_byte_map[index + 1][0]:
1048
# buts up against the next parsed region
1051
# do not strip to the last \n
1054
elif end > self._parsed_byte_map[index + 1][0]:
1055
# overlaps into the next parsed region
1056
# only consider the unparsed data
1057
trim_end = self._parsed_byte_map[index + 1][0] - offset
1058
# do not strip to the last \n as we know its an entire record
1060
last_segment = end < self._parsed_byte_map[index + 1][1]
1062
# does not overlap into the next region
1065
# but strip to the last \n
1066
end_adjacent = False
1068
# now find bytes to discard if needed
1069
if not start_adjacent:
1070
# work around python bug in rfind
1071
if trim_start is None:
1072
trim_start = data.find('\n') + 1
1074
trim_start = data.find('\n', trim_start) + 1
1075
if not (trim_start != 0):
1076
raise AssertionError('no \n was present')
1077
# print 'removing start', offset, trim_start, repr(data[:trim_start])
1078
if not end_adjacent:
1079
# work around python bug in rfind
1080
if trim_end is None:
1081
trim_end = data.rfind('\n') + 1
1083
trim_end = data.rfind('\n', None, trim_end) + 1
1084
if not (trim_end != 0):
1085
raise AssertionError('no \n was present')
1086
# print 'removing end', offset, trim_end, repr(data[trim_end:])
1087
# adjust offset and data to the parseable data.
1088
trimmed_data = data[trim_start:trim_end]
1089
if not (trimmed_data):
1090
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
1091
% (trim_start, trim_end, offset, offset + len(data)))
1093
offset += trim_start
1094
# print "parsing", repr(trimmed_data)
1095
# splitlines mangles the \r delimiters.. don't use it.
1096
lines = trimmed_data.split('\n')
1099
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1100
for key, value in nodes:
1101
self._bisect_nodes[key] = value
1102
self._parsed_bytes(offset, first_key,
1103
offset + len(trimmed_data), last_key)
1104
return offset + len(trimmed_data), last_segment
1106
def _parse_lines(self, lines, pos):
1113
# must be at the end
1115
if not (self._size == pos + 1):
1116
raise AssertionError("%s %s" % (self._size, pos))
1119
elements = line.split('\0')
1120
if len(elements) != self._expected_elements:
1121
raise errors.BadIndexData(self)
1122
# keys are tuples. Each element is a string that may occur many
1123
# times, so we intern them to save space. AB, RC, 200807
1124
key = tuple([intern(element) for element in elements[:self._key_length]])
1125
if first_key is None:
1127
absent, references, value = elements[-3:]
1129
for ref_string in references.split('\t'):
1130
ref_lists.append(tuple([
1131
int(ref) for ref in ref_string.split('\r') if ref
1133
ref_lists = tuple(ref_lists)
1134
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1135
pos += len(line) + 1 # +1 for the \n
1138
if self.node_ref_lists:
1139
node_value = (value, ref_lists)
1142
nodes.append((key, node_value))
1143
# print "parsed ", key
1144
return first_key, key, nodes, trailers
1146
def _parsed_bytes(self, start, start_key, end, end_key):
1147
"""Mark the bytes from start to end as parsed.
1149
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1152
:param start: The start of the parsed region.
1153
:param end: The end of the parsed region.
1155
index = self._parsed_byte_index(start)
1156
new_value = (start, end)
1157
new_key = (start_key, end_key)
1159
# first range parsed is always the beginning.
1160
self._parsed_byte_map.insert(index, new_value)
1161
self._parsed_key_map.insert(index, new_key)
1165
# extend lower region
1166
# extend higher region
1167
# combine two regions
1168
if (index + 1 < len(self._parsed_byte_map) and
1169
self._parsed_byte_map[index][1] == start and
1170
self._parsed_byte_map[index + 1][0] == end):
1171
# combine two regions
1172
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1173
self._parsed_byte_map[index + 1][1])
1174
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1175
self._parsed_key_map[index + 1][1])
1176
del self._parsed_byte_map[index + 1]
1177
del self._parsed_key_map[index + 1]
1178
elif self._parsed_byte_map[index][1] == start:
1179
# extend the lower entry
1180
self._parsed_byte_map[index] = (
1181
self._parsed_byte_map[index][0], end)
1182
self._parsed_key_map[index] = (
1183
self._parsed_key_map[index][0], end_key)
1184
elif (index + 1 < len(self._parsed_byte_map) and
1185
self._parsed_byte_map[index + 1][0] == end):
1186
# extend the higher entry
1187
self._parsed_byte_map[index + 1] = (
1188
start, self._parsed_byte_map[index + 1][1])
1189
self._parsed_key_map[index + 1] = (
1190
start_key, self._parsed_key_map[index + 1][1])
1193
self._parsed_byte_map.insert(index + 1, new_value)
1194
self._parsed_key_map.insert(index + 1, new_key)
1196
def _read_and_parse(self, readv_ranges):
1197
"""Read the ranges and parse the resulting data.
1199
:param readv_ranges: A prepared readv range list.
1201
if not readv_ranges:
1203
if self._nodes is None and self._bytes_read * 2 >= self._size:
1204
# We've already read more than 50% of the file and we are about to
1205
# request more data, just _buffer_all() and be done
1209
base_offset = self._base_offset
1210
if base_offset != 0:
1211
# Rewrite the ranges for the offset
1212
readv_ranges = [(start+base_offset, size)
1213
for start, size in readv_ranges]
1214
readv_data = self._transport.readv(self._name, readv_ranges, True,
1215
self._size + self._base_offset)
1217
for offset, data in readv_data:
1218
offset -= base_offset
1219
self._bytes_read += len(data)
1221
# transport.readv() expanded to extra data which isn't part of
1223
data = data[-offset:]
1225
if offset == 0 and len(data) == self._size:
1226
# We read the whole range, most likely because the
1227
# Transport upcast our readv ranges into one long request
1228
# for enough total data to grab the whole index.
1229
self._buffer_all(StringIO(data))
1231
if self._bisect_nodes is None:
1232
# this must be the start
1233
if not (offset == 0):
1234
raise AssertionError()
1235
offset, data = self._parse_header_from_bytes(data)
1236
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1237
self._parse_region(offset, data)
1239
def _signature(self):
1240
"""The file signature for this index type."""
1244
"""Validate that everything in the index can be accessed."""
1245
# iter_all validates completely at the moment, so just do that.
1246
for node in self.iter_all_entries():
1250
class CombinedGraphIndex(object):
1251
"""A GraphIndex made up from smaller GraphIndices.
1253
The backing indices must implement GraphIndex, and are presumed to be
1256
Queries against the combined index will be made against the first index,
1257
and then the second and so on. The order of indices can thus influence
1258
performance significantly. For example, if one index is on local disk and a
1259
second on a remote server, the local disk index should be before the other
1262
Also, queries tend to need results from the same indices as previous
1263
queries. So the indices will be reordered after every query to put the
1264
indices that had the result(s) of that query first (while otherwise
1265
preserving the relative ordering).
1268
def __init__(self, indices, reload_func=None):
1269
"""Create a CombinedGraphIndex backed by indices.
1271
:param indices: An ordered list of indices to query for data.
1272
:param reload_func: A function to call if we find we are missing an
1273
index. Should have the form reload_func() => True/False to indicate
1274
if reloading actually changed anything.
1276
self._indices = indices
1277
self._reload_func = reload_func
1278
# Sibling indices are other CombinedGraphIndex that we should call
1279
# _move_to_front_by_name on when we auto-reorder ourself.
1280
self._sibling_indices = []
1281
# A list of names that corresponds to the instances in self._indices,
1282
# so _index_names[0] is always the name for _indices[0], etc. Sibling
1283
# indices must all use the same set of names as each other.
1284
self._index_names = [None] * len(self._indices)
1288
self.__class__.__name__,
1289
', '.join(map(repr, self._indices)))
1291
def clear_cache(self):
1292
"""See GraphIndex.clear_cache()"""
1293
for index in self._indices:
1296
def get_parent_map(self, keys):
1297
"""See graph.StackedParentsProvider.get_parent_map"""
1298
search_keys = set(keys)
1299
if _mod_revision.NULL_REVISION in search_keys:
1300
search_keys.discard(_mod_revision.NULL_REVISION)
1301
found_parents = {_mod_revision.NULL_REVISION:[]}
1304
for index, key, value, refs in self.iter_entries(search_keys):
1307
parents = (_mod_revision.NULL_REVISION,)
1308
found_parents[key] = parents
1309
return found_parents
1311
has_key = _has_key_from_parent_map
1313
def insert_index(self, pos, index, name=None):
1314
"""Insert a new index in the list of indices to query.
1316
:param pos: The position to insert the index.
1317
:param index: The index to insert.
1318
:param name: a name for this index, e.g. a pack name. These names can
1319
be used to reflect index reorderings to related CombinedGraphIndex
1320
instances that use the same names. (see set_sibling_indices)
1322
self._indices.insert(pos, index)
1323
self._index_names.insert(pos, name)
1325
def iter_all_entries(self):
1326
"""Iterate over all keys within the index
1328
Duplicate keys across child indices are presumed to have the same
1329
value and are only reported once.
1331
:return: An iterable of (index, key, reference_lists, value).
1332
There is no defined order for the result iteration - it will be in
1333
the most efficient order for the index.
1338
for index in self._indices:
1339
for node in index.iter_all_entries():
1340
if node[1] not in seen_keys:
1342
seen_keys.add(node[1])
1344
except errors.NoSuchFile:
1345
self._reload_or_raise()
1347
def iter_entries(self, keys):
1348
"""Iterate over keys within the index.
1350
Duplicate keys across child indices are presumed to have the same
1351
value and are only reported once.
1353
:param keys: An iterable providing the keys to be retrieved.
1354
:return: An iterable of (index, key, reference_lists, value). There is
1355
no defined order for the result iteration - it will be in the most
1356
efficient order for the index.
1362
for index in self._indices:
1366
for node in index.iter_entries(keys):
1367
keys.remove(node[1])
1371
hit_indices.append(index)
1373
except errors.NoSuchFile:
1374
self._reload_or_raise()
1375
self._move_to_front(hit_indices)
1377
def iter_entries_prefix(self, keys):
1378
"""Iterate over keys within the index using prefix matching.
1380
Duplicate keys across child indices are presumed to have the same
1381
value and are only reported once.
1383
Prefix matching is applied within the tuple of a key, not to within
1384
the bytestring of each key element. e.g. if you have the keys ('foo',
1385
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1386
only the former key is returned.
1388
:param keys: An iterable providing the key prefixes to be retrieved.
1389
Each key prefix takes the form of a tuple the length of a key, but
1390
with the last N elements 'None' rather than a regular bytestring.
1391
The first element cannot be 'None'.
1392
:return: An iterable as per iter_all_entries, but restricted to the
1393
keys with a matching prefix to those supplied. No additional keys
1394
will be returned, and every match that is in the index will be
1404
for index in self._indices:
1406
for node in index.iter_entries_prefix(keys):
1407
if node[1] in seen_keys:
1409
seen_keys.add(node[1])
1413
hit_indices.append(index)
1415
except errors.NoSuchFile:
1416
self._reload_or_raise()
1417
self._move_to_front(hit_indices)
1419
def _move_to_front(self, hit_indices):
1420
"""Rearrange self._indices so that hit_indices are first.
1422
Order is maintained as much as possible, e.g. the first unhit index
1423
will be the first index in _indices after the hit_indices, and the
1424
hit_indices will be present in exactly the order they are passed to
1427
_move_to_front propagates to all objects in self._sibling_indices by
1428
calling _move_to_front_by_name.
1430
if self._indices[:len(hit_indices)] == hit_indices:
1431
# The 'hit_indices' are already at the front (and in the same
1432
# order), no need to re-order
1434
hit_names = self._move_to_front_by_index(hit_indices)
1435
for sibling_idx in self._sibling_indices:
1436
sibling_idx._move_to_front_by_name(hit_names)
1438
def _move_to_front_by_index(self, hit_indices):
1439
"""Core logic for _move_to_front.
1441
Returns a list of names corresponding to the hit_indices param.
1443
indices_info = zip(self._index_names, self._indices)
1444
if 'index' in debug.debug_flags:
1445
trace.mutter('CombinedGraphIndex reordering: currently %r, '
1446
'promoting %r', indices_info, hit_indices)
1449
new_hit_indices = []
1452
for offset, (name, idx) in enumerate(indices_info):
1453
if idx in hit_indices:
1454
hit_names.append(name)
1455
new_hit_indices.append(idx)
1456
if len(new_hit_indices) == len(hit_indices):
1457
# We've found all of the hit entries, everything else is
1459
unhit_names.extend(self._index_names[offset+1:])
1460
unhit_indices.extend(self._indices[offset+1:])
1463
unhit_names.append(name)
1464
unhit_indices.append(idx)
1466
self._indices = new_hit_indices + unhit_indices
1467
self._index_names = hit_names + unhit_names
1468
if 'index' in debug.debug_flags:
1469
trace.mutter('CombinedGraphIndex reordered: %r', self._indices)
1472
def _move_to_front_by_name(self, hit_names):
1473
"""Moves indices named by 'hit_names' to front of the search order, as
1474
described in _move_to_front.
1476
# Translate names to index instances, and then call
1477
# _move_to_front_by_index.
1478
indices_info = zip(self._index_names, self._indices)
1480
for name, idx in indices_info:
1481
if name in hit_names:
1482
hit_indices.append(idx)
1483
self._move_to_front_by_index(hit_indices)
1485
def find_ancestry(self, keys, ref_list_num):
1486
"""Find the complete ancestry for the given set of keys.
1488
Note that this is a whole-ancestry request, so it should be used
1491
:param keys: An iterable of keys to look for
1492
:param ref_list_num: The reference list which references the parents
1494
:return: (parent_map, missing_keys)
1496
# XXX: make this call _move_to_front?
1497
missing_keys = set()
1499
keys_to_lookup = set(keys)
1501
while keys_to_lookup:
1502
# keys that *all* indexes claim are missing, stop searching them
1504
all_index_missing = None
1505
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1506
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1508
# len(missing_keys))
1509
for index_idx, index in enumerate(self._indices):
1510
# TODO: we should probably be doing something with
1511
# 'missing_keys' since we've already determined that
1512
# those revisions have not been found anywhere
1513
index_missing_keys = set()
1514
# Find all of the ancestry we can from this index
1515
# keep looking until the search_keys set is empty, which means
1516
# things we didn't find should be in index_missing_keys
1517
search_keys = keys_to_lookup
1519
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1520
# index_idx, len(search_keys),
1521
# len(parent_map), len(index_missing_keys))
1524
# TODO: ref_list_num should really be a parameter, since
1525
# CombinedGraphIndex does not know what the ref lists
1527
search_keys = index._find_ancestors(search_keys,
1528
ref_list_num, parent_map, index_missing_keys)
1529
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1530
# sub_generation, len(search_keys),
1531
# len(parent_map), len(index_missing_keys))
1532
# Now set whatever was missing to be searched in the next index
1533
keys_to_lookup = index_missing_keys
1534
if all_index_missing is None:
1535
all_index_missing = set(index_missing_keys)
1537
all_index_missing.intersection_update(index_missing_keys)
1538
if not keys_to_lookup:
1540
if all_index_missing is None:
1541
# There were no indexes, so all search keys are 'missing'
1542
missing_keys.update(keys_to_lookup)
1543
keys_to_lookup = None
1545
missing_keys.update(all_index_missing)
1546
keys_to_lookup.difference_update(all_index_missing)
1547
return parent_map, missing_keys
1549
def key_count(self):
1550
"""Return an estimate of the number of keys in this index.
1552
For CombinedGraphIndex this is approximated by the sum of the keys of
1553
the child indices. As child indices may have duplicate keys this can
1554
have a maximum error of the number of child indices * largest number of
1559
return sum((index.key_count() for index in self._indices), 0)
1560
except errors.NoSuchFile:
1561
self._reload_or_raise()
1563
missing_keys = _missing_keys_from_parent_map
1565
def _reload_or_raise(self):
1566
"""We just got a NoSuchFile exception.
1568
Try to reload the indices, if it fails, just raise the current
1571
if self._reload_func is None:
1573
exc_type, exc_value, exc_traceback = sys.exc_info()
1574
trace.mutter('Trying to reload after getting exception: %s',
1576
if not self._reload_func():
1577
# We tried to reload, but nothing changed, so we fail anyway
1578
trace.mutter('_reload_func indicated nothing has changed.'
1579
' Raising original exception.')
1580
raise exc_type, exc_value, exc_traceback
1582
def set_sibling_indices(self, sibling_combined_graph_indices):
1583
"""Set the CombinedGraphIndex objects to reorder after reordering self.
1585
self._sibling_indices = sibling_combined_graph_indices
1588
"""Validate that everything in the index can be accessed."""
1591
for index in self._indices:
1594
except errors.NoSuchFile:
1595
self._reload_or_raise()
1598
class InMemoryGraphIndex(GraphIndexBuilder):
1599
"""A GraphIndex which operates entirely out of memory and is mutable.
1601
This is designed to allow the accumulation of GraphIndex entries during a
1602
single write operation, where the accumulated entries need to be immediately
1603
available - for example via a CombinedGraphIndex.
1606
def add_nodes(self, nodes):
1607
"""Add nodes to the index.
1609
:param nodes: An iterable of (key, node_refs, value) entries to add.
1611
if self.reference_lists:
1612
for (key, value, node_refs) in nodes:
1613
self.add_node(key, value, node_refs)
1615
for (key, value) in nodes:
1616
self.add_node(key, value)
1618
def iter_all_entries(self):
1619
"""Iterate over all keys within the index
1621
:return: An iterable of (index, key, reference_lists, value). There is no
1622
defined order for the result iteration - it will be in the most
1623
efficient order for the index (in this case dictionary hash order).
1625
if 'evil' in debug.debug_flags:
1626
trace.mutter_callsite(3,
1627
"iter_all_entries scales with size of history.")
1628
if self.reference_lists:
1629
for key, (absent, references, value) in self._nodes.iteritems():
1631
yield self, key, value, references
1633
for key, (absent, references, value) in self._nodes.iteritems():
1635
yield self, key, value
1637
def iter_entries(self, keys):
1638
"""Iterate over keys within the index.
1640
:param keys: An iterable providing the keys to be retrieved.
1641
:return: An iterable of (index, key, value, reference_lists). There is no
1642
defined order for the result iteration - it will be in the most
1643
efficient order for the index (keys iteration order in this case).
1645
# Note: See BTreeBuilder.iter_entries for an explanation of why we
1646
# aren't using set().intersection() here
1648
keys = [key for key in keys if key in nodes]
1649
if self.reference_lists:
1653
yield self, key, node[2], node[1]
1658
yield self, key, node[2]
1660
def iter_entries_prefix(self, keys):
1661
"""Iterate over keys within the index using prefix matching.
1663
Prefix matching is applied within the tuple of a key, not to within
1664
the bytestring of each key element. e.g. if you have the keys ('foo',
1665
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1666
only the former key is returned.
1668
:param keys: An iterable providing the key prefixes to be retrieved.
1669
Each key prefix takes the form of a tuple the length of a key, but
1670
with the last N elements 'None' rather than a regular bytestring.
1671
The first element cannot be 'None'.
1672
:return: An iterable as per iter_all_entries, but restricted to the
1673
keys with a matching prefix to those supplied. No additional keys
1674
will be returned, and every match that is in the index will be
1677
# XXX: To much duplication with the GraphIndex class; consider finding
1678
# a good place to pull out the actual common logic.
1682
if self._key_length == 1:
1686
raise errors.BadIndexKey(key)
1687
if len(key) != self._key_length:
1688
raise errors.BadIndexKey(key)
1689
node = self._nodes[key]
1692
if self.reference_lists:
1693
yield self, key, node[2], node[1]
1695
yield self, key, node[2]
1697
nodes_by_key = self._get_nodes_by_key()
1701
raise errors.BadIndexKey(key)
1702
if len(key) != self._key_length:
1703
raise errors.BadIndexKey(key)
1704
# find what it refers to:
1705
key_dict = nodes_by_key
1706
elements = list(key)
1707
# find the subdict to return
1709
while len(elements) and elements[0] is not None:
1710
key_dict = key_dict[elements[0]]
1713
# a non-existant lookup.
1718
key_dict = dicts.pop(-1)
1719
# can't be empty or would not exist
1720
item, value = key_dict.iteritems().next()
1721
if type(value) == dict:
1723
dicts.extend(key_dict.itervalues())
1726
for value in key_dict.itervalues():
1727
yield (self, ) + value
1729
yield (self, ) + key_dict
1731
def key_count(self):
1732
"""Return an estimate of the number of keys in this index.
1734
For InMemoryGraphIndex the estimate is exact.
1736
return len(self._nodes) - len(self._absent_keys)
1739
"""In memory index's have no known corruption at the moment."""
1742
class GraphIndexPrefixAdapter(object):
1743
"""An adapter between GraphIndex with different key lengths.
1745
Queries against this will emit queries against the adapted Graph with the
1746
prefix added, queries for all items use iter_entries_prefix. The returned
1747
nodes will have their keys and node references adjusted to remove the
1748
prefix. Finally, an add_nodes_callback can be supplied - when called the
1749
nodes and references being added will have prefix prepended.
1752
def __init__(self, adapted, prefix, missing_key_length,
1753
add_nodes_callback=None):
1754
"""Construct an adapter against adapted with prefix."""
1755
self.adapted = adapted
1756
self.prefix_key = prefix + (None,)*missing_key_length
1757
self.prefix = prefix
1758
self.prefix_len = len(prefix)
1759
self.add_nodes_callback = add_nodes_callback
1761
def add_nodes(self, nodes):
1762
"""Add nodes to the index.
1764
:param nodes: An iterable of (key, node_refs, value) entries to add.
1766
# save nodes in case its an iterator
1767
nodes = tuple(nodes)
1768
translated_nodes = []
1770
# Add prefix_key to each reference node_refs is a tuple of tuples,
1771
# so split it apart, and add prefix_key to the internal reference
1772
for (key, value, node_refs) in nodes:
1773
adjusted_references = (
1774
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1775
for ref_list in node_refs))
1776
translated_nodes.append((self.prefix + key, value,
1777
adjusted_references))
1779
# XXX: TODO add an explicit interface for getting the reference list
1780
# status, to handle this bit of user-friendliness in the API more
1782
for (key, value) in nodes:
1783
translated_nodes.append((self.prefix + key, value))
1784
self.add_nodes_callback(translated_nodes)
1786
def add_node(self, key, value, references=()):
1787
"""Add a node to the index.
1789
:param key: The key. keys are non-empty tuples containing
1790
as many whitespace-free utf8 bytestrings as the key length
1791
defined for this index.
1792
:param references: An iterable of iterables of keys. Each is a
1793
reference to another key.
1794
:param value: The value to associate with the key. It may be any
1795
bytes as long as it does not contain \0 or \n.
1797
self.add_nodes(((key, value, references), ))
1799
def _strip_prefix(self, an_iter):
1800
"""Strip prefix data from nodes and return it."""
1801
for node in an_iter:
1803
if node[1][:self.prefix_len] != self.prefix:
1804
raise errors.BadIndexData(self)
1805
for ref_list in node[3]:
1806
for ref_node in ref_list:
1807
if ref_node[:self.prefix_len] != self.prefix:
1808
raise errors.BadIndexData(self)
1809
yield node[0], node[1][self.prefix_len:], node[2], (
1810
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1811
for ref_list in node[3]))
1813
def iter_all_entries(self):
1814
"""Iterate over all keys within the index
1816
iter_all_entries is implemented against the adapted index using
1817
iter_entries_prefix.
1819
:return: An iterable of (index, key, reference_lists, value). There is no
1820
defined order for the result iteration - it will be in the most
1821
efficient order for the index (in this case dictionary hash order).
1823
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1825
def iter_entries(self, keys):
1826
"""Iterate over keys within the index.
1828
:param keys: An iterable providing the keys to be retrieved.
1829
:return: An iterable of (index, key, value, reference_lists). There is no
1830
defined order for the result iteration - it will be in the most
1831
efficient order for the index (keys iteration order in this case).
1833
return self._strip_prefix(self.adapted.iter_entries(
1834
self.prefix + key for key in keys))
1836
def iter_entries_prefix(self, keys):
1837
"""Iterate over keys within the index using prefix matching.
1839
Prefix matching is applied within the tuple of a key, not to within
1840
the bytestring of each key element. e.g. if you have the keys ('foo',
1841
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1842
only the former key is returned.
1844
:param keys: An iterable providing the key prefixes to be retrieved.
1845
Each key prefix takes the form of a tuple the length of a key, but
1846
with the last N elements 'None' rather than a regular bytestring.
1847
The first element cannot be 'None'.
1848
:return: An iterable as per iter_all_entries, but restricted to the
1849
keys with a matching prefix to those supplied. No additional keys
1850
will be returned, and every match that is in the index will be
1853
return self._strip_prefix(self.adapted.iter_entries_prefix(
1854
self.prefix + key for key in keys))
1856
def key_count(self):
1857
"""Return an estimate of the number of keys in this index.
1859
For GraphIndexPrefixAdapter this is relatively expensive - key
1860
iteration with the prefix is done.
1862
return len(list(self.iter_all_entries()))
1865
"""Call the adapted's validate."""
1866
self.adapted.validate()
added
removed
bzrlib/index.py
