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
33
from .lazy_import import lazy_import
34
lazy_import(globals(), """
37
revision as _mod_revision,
50
from .static_tuple import StaticTuple
52
_HEADER_READV = (0, 200)
53
_OPTION_KEY_ELEMENTS = "key_elements="
55
_OPTION_NODE_REFS = "node_ref_lists="
56
_SIGNATURE = "Bazaar Graph Index 1\n"
59
_whitespace_re = re.compile('[\t\n\x0b\x0c\r\x00 ]')
60
_newline_null_re = re.compile('[\n\0]')
63
def _has_key_from_parent_map(self, key):
64
"""Check if this index has one key.
66
If it's possible to check for multiple keys at once through
67
calling get_parent_map that should be faster.
69
return (key in self.get_parent_map([key]))
72
def _missing_keys_from_parent_map(self, keys):
73
return set(keys) - set(self.get_parent_map(keys))
76
class GraphIndexBuilder(object):
77
"""A builder that can build a GraphIndex.
79
The resulting graph has the structure::
81
_SIGNATURE OPTIONS NODES NEWLINE
82
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
83
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
85
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
86
KEY := Not-whitespace-utf8
88
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
89
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
90
REFERENCE := DIGITS ; digits is the byte offset in the index of the
92
VALUE := no-newline-no-null-bytes
95
def __init__(self, reference_lists=0, key_elements=1):
96
"""Create a GraphIndex builder.
98
:param reference_lists: The number of node references lists for each
100
:param key_elements: The number of bytestrings in each key.
102
self.reference_lists = reference_lists
103
# A dict of {key: (absent, ref_lists, value)}
105
# Keys that are referenced but not actually present in this index
106
self._absent_keys = set()
107
self._nodes_by_key = None
108
self._key_length = key_elements
109
self._optimize_for_size = False
110
self._combine_backing_indices = True
112
def _check_key(self, key):
113
"""Raise BadIndexKey if key is not a valid key for this index."""
114
if type(key) not in (tuple, StaticTuple):
115
raise errors.BadIndexKey(key)
116
if self._key_length != len(key):
117
raise errors.BadIndexKey(key)
119
if not element or _whitespace_re.search(element) is not None:
120
raise errors.BadIndexKey(element)
122
def _external_references(self):
123
"""Return references that are not present in this index.
127
# TODO: JAM 2008-11-21 This makes an assumption about how the reference
128
# lists are used. It is currently correct for pack-0.92 through
129
# 1.9, which use the node references (3rd column) second
130
# reference list as the compression parent. Perhaps this should
131
# be moved into something higher up the stack, since it
132
# makes assumptions about how the index is used.
133
if self.reference_lists > 1:
134
for node in self.iter_all_entries():
136
refs.update(node[3][1])
139
# If reference_lists == 0 there can be no external references, and
140
# if reference_lists == 1, then there isn't a place to store the
144
def _get_nodes_by_key(self):
145
if self._nodes_by_key is None:
147
if self.reference_lists:
148
for key, (absent, references, value) in viewitems(self._nodes):
151
key_dict = nodes_by_key
152
for subkey in key[:-1]:
153
key_dict = key_dict.setdefault(subkey, {})
154
key_dict[key[-1]] = key, value, references
156
for key, (absent, references, value) in viewitems(self._nodes):
159
key_dict = nodes_by_key
160
for subkey in key[:-1]:
161
key_dict = key_dict.setdefault(subkey, {})
162
key_dict[key[-1]] = key, value
163
self._nodes_by_key = nodes_by_key
164
return self._nodes_by_key
166
def _update_nodes_by_key(self, key, value, node_refs):
167
"""Update the _nodes_by_key dict with a new key.
169
For a key of (foo, bar, baz) create
170
_nodes_by_key[foo][bar][baz] = key_value
172
if self._nodes_by_key is None:
174
key_dict = self._nodes_by_key
175
if self.reference_lists:
176
key_value = StaticTuple(key, value, node_refs)
178
key_value = StaticTuple(key, value)
179
for subkey in key[:-1]:
180
key_dict = key_dict.setdefault(subkey, {})
181
key_dict[key[-1]] = key_value
183
def _check_key_ref_value(self, key, references, value):
184
"""Check that 'key' and 'references' are all valid.
186
:param key: A key tuple. Must conform to the key interface (be a tuple,
187
be of the right length, not have any whitespace or nulls in any key
189
:param references: An iterable of reference lists. Something like
190
[[(ref, key)], [(ref, key), (other, key)]]
191
:param value: The value associate with this key. Must not contain
192
newlines or null characters.
193
:return: (node_refs, absent_references)
195
* node_refs: basically a packed form of 'references' where all
197
* absent_references: reference keys that are not in self._nodes.
198
This may contain duplicates if the same key is referenced in
201
as_st = StaticTuple.from_sequence
203
if _newline_null_re.search(value) is not None:
204
raise errors.BadIndexValue(value)
205
if len(references) != self.reference_lists:
206
raise errors.BadIndexValue(references)
208
absent_references = []
209
for reference_list in references:
210
for reference in reference_list:
211
# If reference *is* in self._nodes, then we know it has already
213
if reference not in self._nodes:
214
self._check_key(reference)
215
absent_references.append(reference)
216
reference_list = as_st([as_st(ref).intern()
217
for ref in reference_list])
218
node_refs.append(reference_list)
219
return as_st(node_refs), absent_references
221
def add_node(self, key, value, references=()):
222
"""Add a node to the index.
224
:param key: The key. keys are non-empty tuples containing
225
as many whitespace-free utf8 bytestrings as the key length
226
defined for this index.
227
:param references: An iterable of iterables of keys. Each is a
228
reference to another key.
229
:param value: The value to associate with the key. It may be any
230
bytes as long as it does not contain \\0 or \\n.
233
absent_references) = self._check_key_ref_value(key, references, value)
234
if key in self._nodes and self._nodes[key][0] != 'a':
235
raise errors.BadIndexDuplicateKey(key, self)
236
for reference in absent_references:
237
# There may be duplicates, but I don't think it is worth worrying
239
self._nodes[reference] = ('a', (), '')
240
self._absent_keys.update(absent_references)
241
self._absent_keys.discard(key)
242
self._nodes[key] = ('', node_refs, value)
243
if self._nodes_by_key is not None and self._key_length > 1:
244
self._update_nodes_by_key(key, value, node_refs)
246
def clear_cache(self):
247
"""See GraphIndex.clear_cache()
249
This is a no-op, but we need the api to conform to a generic 'Index'
256
:returns: cBytesIO holding the full context of the index as it
257
should be written to disk.
260
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
261
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
262
key_count = len(self._nodes) - len(self._absent_keys)
263
lines.append(_OPTION_LEN + str(key_count) + '\n')
264
prefix_length = sum(len(x) for x in lines)
265
# references are byte offsets. To avoid having to do nasty
266
# polynomial work to resolve offsets (references to later in the
267
# file cannot be determined until all the inbetween references have
268
# been calculated too) we pad the offsets with 0's to make them be
269
# of consistent length. Using binary offsets would break the trivial
271
# to calculate the width of zero's needed we do three passes:
272
# one to gather all the non-reference data and the number of references.
273
# one to pad all the data with reference-length and determine entry
277
# forward sorted by key. In future we may consider topological sorting,
278
# at the cost of table scans for direct lookup, or a second index for
280
nodes = sorted(viewitems(self._nodes))
281
# if we do not prepass, we don't know how long it will be up front.
282
expected_bytes = None
283
# we only need to pre-pass if we have reference lists at all.
284
if self.reference_lists:
286
non_ref_bytes = prefix_length
288
# TODO use simple multiplication for the constants in this loop.
289
for key, (absent, references, value) in nodes:
290
# record the offset known *so far* for this key:
291
# the non reference bytes to date, and the total references to
292
# date - saves reaccumulating on the second pass
293
key_offset_info.append((key, non_ref_bytes, total_references))
294
# key is literal, value is literal, there are 3 null's, 1 NL
295
# key is variable length tuple, \x00 between elements
296
non_ref_bytes += sum(len(element) for element in key)
297
if self._key_length > 1:
298
non_ref_bytes += self._key_length - 1
299
# value is literal bytes, there are 3 null's, 1 NL.
300
non_ref_bytes += len(value) + 3 + 1
301
# one byte for absent if set.
304
elif self.reference_lists:
305
# (ref_lists -1) tabs
306
non_ref_bytes += self.reference_lists - 1
307
# (ref-1 cr's per ref_list)
308
for ref_list in references:
309
# how many references across the whole file?
310
total_references += len(ref_list)
311
# accrue reference separators
313
non_ref_bytes += len(ref_list) - 1
314
# how many digits are needed to represent the total byte count?
316
possible_total_bytes = non_ref_bytes + total_references*digits
317
while 10 ** digits < possible_total_bytes:
319
possible_total_bytes = non_ref_bytes + total_references*digits
320
expected_bytes = possible_total_bytes + 1 # terminating newline
321
# resolve key addresses.
323
for key, non_ref_bytes, total_references in key_offset_info:
324
key_addresses[key] = non_ref_bytes + total_references*digits
326
format_string = '%%0%sd' % digits
327
for key, (absent, references, value) in nodes:
328
flattened_references = []
329
for ref_list in references:
331
for reference in ref_list:
332
ref_addresses.append(format_string % key_addresses[reference])
333
flattened_references.append('\r'.join(ref_addresses))
334
string_key = '\x00'.join(key)
335
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
336
'\t'.join(flattened_references), value))
338
result = BytesIO(''.join(lines))
339
if expected_bytes and len(result.getvalue()) != expected_bytes:
340
raise errors.BzrError('Failed index creation. Internal error:'
341
' mismatched output length and expected length: %d %d' %
342
(len(result.getvalue()), expected_bytes))
345
def set_optimize(self, for_size=None, combine_backing_indices=None):
346
"""Change how the builder tries to optimize the result.
348
:param for_size: Tell the builder to try and make the index as small as
350
:param combine_backing_indices: If the builder spills to disk to save
351
memory, should the on-disk indices be combined. Set to True if you
352
are going to be probing the index, but to False if you are not. (If
353
you are not querying, then the time spent combining is wasted.)
356
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
358
if for_size is not None:
359
self._optimize_for_size = for_size
360
if combine_backing_indices is not None:
361
self._combine_backing_indices = combine_backing_indices
363
def find_ancestry(self, keys, ref_list_num):
364
"""See CombinedGraphIndex.find_ancestry()"""
370
for _, key, value, ref_lists in self.iter_entries(pending):
371
parent_keys = ref_lists[ref_list_num]
372
parent_map[key] = parent_keys
373
next_pending.update([p for p in parent_keys if p not in
375
missing_keys.update(pending.difference(parent_map))
376
pending = next_pending
377
return parent_map, missing_keys
380
class GraphIndex(object):
381
"""An index for data with embedded graphs.
383
The index maps keys to a list of key reference lists, and a value.
384
Each node has the same number of key reference lists. Each key reference
385
list can be empty or an arbitrary length. The value is an opaque NULL
386
terminated string without any newlines. The storage of the index is
387
hidden in the interface: keys and key references are always tuples of
388
bytestrings, never the internal representation (e.g. dictionary offsets).
390
It is presumed that the index will not be mutated - it is static data.
392
Successive iter_all_entries calls will read the entire index each time.
393
Additionally, iter_entries calls will read the index linearly until the
394
desired keys are found. XXX: This must be fixed before the index is
395
suitable for production use. :XXX
398
def __init__(self, transport, name, size, unlimited_cache=False, offset=0):
399
"""Open an index called name on transport.
401
:param transport: A breezy.transport.Transport.
402
:param name: A path to provide to transport API calls.
403
:param size: The size of the index in bytes. This is used for bisection
404
logic to perform partial index reads. While the size could be
405
obtained by statting the file this introduced an additional round
406
trip as well as requiring stat'able transports, both of which are
407
avoided by having it supplied. If size is None, then bisection
408
support will be disabled and accessing the index will just stream
410
:param offset: Instead of starting the index data at offset 0, start it
411
at an arbitrary offset.
413
self._transport = transport
415
# Becomes a dict of key:(value, reference-list-byte-locations) used by
416
# the bisection interface to store parsed but not resolved keys.
417
self._bisect_nodes = None
418
# Becomes a dict of key:(value, reference-list-keys) which are ready to
419
# be returned directly to callers.
421
# a sorted list of slice-addresses for the parsed bytes of the file.
422
# e.g. (0,1) would mean that byte 0 is parsed.
423
self._parsed_byte_map = []
424
# a sorted list of keys matching each slice address for parsed bytes
425
# e.g. (None, 'foo@bar') would mean that the first byte contained no
426
# key, and the end byte of the slice is the of the data for 'foo@bar'
427
self._parsed_key_map = []
428
self._key_count = None
429
self._keys_by_offset = None
430
self._nodes_by_key = None
432
# The number of bytes we've read so far in trying to process this file
434
self._base_offset = offset
436
def __eq__(self, other):
437
"""Equal when self and other were created with the same parameters."""
439
isinstance(self, type(other)) and
440
self._transport == other._transport and
441
self._name == other._name and
442
self._size == other._size)
444
def __ne__(self, other):
445
return not self.__eq__(other)
448
return "%s(%r)" % (self.__class__.__name__,
449
self._transport.abspath(self._name))
451
def _buffer_all(self, stream=None):
452
"""Buffer all the index data.
454
Mutates self._nodes and self.keys_by_offset.
456
if self._nodes is not None:
457
# We already did this
459
if 'index' in debug.debug_flags:
460
trace.mutter('Reading entire index %s',
461
self._transport.abspath(self._name))
463
stream = self._transport.get(self._name)
464
if self._base_offset != 0:
465
# This is wasteful, but it is better than dealing with
466
# adjusting all the offsets, etc.
467
stream = BytesIO(stream.read()[self._base_offset:])
468
self._read_prefix(stream)
469
self._expected_elements = 3 + self._key_length
471
# raw data keyed by offset
472
self._keys_by_offset = {}
473
# ready-to-return key:value or key:value, node_ref_lists
475
self._nodes_by_key = None
478
lines = stream.read().split('\n')
479
# GZ 2009-09-20: Should really use a try/finally block to ensure close
482
_, _, _, trailers = self._parse_lines(lines, pos)
483
for key, absent, references, value in viewvalues(self._keys_by_offset):
486
# resolve references:
487
if self.node_ref_lists:
488
node_value = (value, self._resolve_references(references))
491
self._nodes[key] = node_value
492
# cache the keys for quick set intersections
494
# there must be one line - the empty trailer line.
495
raise errors.BadIndexData(self)
497
def clear_cache(self):
498
"""Clear out any cached/memoized values.
500
This can be called at any time, but generally it is used when we have
501
extracted some information, but don't expect to be requesting any more
505
def external_references(self, ref_list_num):
506
"""Return references that are not present in this index.
509
if ref_list_num + 1 > self.node_ref_lists:
510
raise ValueError('No ref list %d, index has %d ref lists'
511
% (ref_list_num, self.node_ref_lists))
514
for key, (value, ref_lists) in viewitems(nodes):
515
ref_list = ref_lists[ref_list_num]
516
refs.update([ref for ref in ref_list if ref not in nodes])
519
def _get_nodes_by_key(self):
520
if self._nodes_by_key is None:
522
if self.node_ref_lists:
523
for key, (value, references) in viewitems(self._nodes):
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, references
529
for key, value in viewitems(self._nodes):
530
key_dict = nodes_by_key
531
for subkey in key[:-1]:
532
key_dict = key_dict.setdefault(subkey, {})
533
key_dict[key[-1]] = key, value
534
self._nodes_by_key = nodes_by_key
535
return self._nodes_by_key
537
def iter_all_entries(self):
538
"""Iterate over all keys within the index.
540
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
541
The former tuple is used when there are no reference lists in the
542
index, making the API compatible with simple key:value index types.
543
There is no defined order for the result iteration - it will be in
544
the most efficient order for the index.
546
if 'evil' in debug.debug_flags:
547
trace.mutter_callsite(3,
548
"iter_all_entries scales with size of history.")
549
if self._nodes is None:
551
if self.node_ref_lists:
552
for key, (value, node_ref_lists) in viewitems(self._nodes):
553
yield self, key, value, node_ref_lists
555
for key, value in viewitems(self._nodes):
556
yield self, key, value
558
def _read_prefix(self, stream):
559
signature = stream.read(len(self._signature()))
560
if not signature == self._signature():
561
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
562
options_line = stream.readline()
563
if not options_line.startswith(_OPTION_NODE_REFS):
564
raise errors.BadIndexOptions(self)
566
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
568
raise errors.BadIndexOptions(self)
569
options_line = stream.readline()
570
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
571
raise errors.BadIndexOptions(self)
573
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
575
raise errors.BadIndexOptions(self)
576
options_line = stream.readline()
577
if not options_line.startswith(_OPTION_LEN):
578
raise errors.BadIndexOptions(self)
580
self._key_count = int(options_line[len(_OPTION_LEN):-1])
582
raise errors.BadIndexOptions(self)
584
def _resolve_references(self, references):
585
"""Return the resolved key references for references.
587
References are resolved by looking up the location of the key in the
588
_keys_by_offset map and substituting the key name, preserving ordering.
590
:param references: An iterable of iterables of key locations. e.g.
592
:return: A tuple of tuples of keys.
595
for ref_list in references:
596
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
597
return tuple(node_refs)
599
def _find_index(self, range_map, key):
600
"""Helper for the _parsed_*_index calls.
602
Given a range map - [(start, end), ...], finds the index of the range
603
in the map for key if it is in the map, and if it is not there, the
604
immediately preceeding range in the map.
606
result = bisect_right(range_map, key) - 1
607
if result + 1 < len(range_map):
608
# check the border condition, it may be in result + 1
609
if range_map[result + 1][0] == key[0]:
613
def _parsed_byte_index(self, offset):
614
"""Return the index of the entry immediately before offset.
616
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
617
there is one unparsed byte (the 11th, addressed as[10]). then:
618
asking for 0 will return 0
619
asking for 10 will return 0
620
asking for 11 will return 1
621
asking for 12 will return 1
624
return self._find_index(self._parsed_byte_map, key)
626
def _parsed_key_index(self, key):
627
"""Return the index of the entry immediately before key.
629
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
630
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
631
have been parsed, then:
632
asking for '' will return 0
633
asking for 'a' will return 0
634
asking for 'b' will return 1
635
asking for 'e' will return 1
637
search_key = (key, None)
638
return self._find_index(self._parsed_key_map, search_key)
640
def _is_parsed(self, offset):
641
"""Returns True if offset has been parsed."""
642
index = self._parsed_byte_index(offset)
643
if index == len(self._parsed_byte_map):
644
return offset < self._parsed_byte_map[index - 1][1]
645
start, end = self._parsed_byte_map[index]
646
return offset >= start and offset < end
648
def _iter_entries_from_total_buffer(self, keys):
649
"""Iterate over keys when the entire index is parsed."""
650
# Note: See the note in BTreeBuilder.iter_entries for why we don't use
651
# .intersection() here
653
keys = [key for key in keys if key in nodes]
654
if self.node_ref_lists:
656
value, node_refs = nodes[key]
657
yield self, key, value, node_refs
660
yield self, key, nodes[key]
662
def iter_entries(self, keys):
663
"""Iterate over keys within the index.
665
:param keys: An iterable providing the keys to be retrieved.
666
:return: An iterable as per iter_all_entries, but restricted to the
667
keys supplied. No additional keys will be returned, and every
668
key supplied that is in the index will be returned.
673
if self._size is None and self._nodes is None:
676
# We fit about 20 keys per minimum-read (4K), so if we are looking for
677
# more than 1/20th of the index its likely (assuming homogenous key
678
# spread) that we'll read the entire index. If we're going to do that,
679
# buffer the whole thing. A better analysis might take key spread into
680
# account - but B+Tree indices are better anyway.
681
# We could look at all data read, and use a threshold there, which will
682
# trigger on ancestry walks, but that is not yet fully mapped out.
683
if self._nodes is None and len(keys) * 20 > self.key_count():
685
if self._nodes is not None:
686
return self._iter_entries_from_total_buffer(keys)
688
return (result[1] for result in bisect_multi.bisect_multi_bytes(
689
self._lookup_keys_via_location, self._size, keys))
691
def iter_entries_prefix(self, keys):
692
"""Iterate over keys within the index using prefix matching.
694
Prefix matching is applied within the tuple of a key, not to within
695
the bytestring of each key element. e.g. if you have the keys ('foo',
696
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
697
only the former key is returned.
699
WARNING: Note that this method currently causes a full index parse
700
unconditionally (which is reasonably appropriate as it is a means for
701
thunking many small indices into one larger one and still supplies
702
iter_all_entries at the thunk layer).
704
:param keys: An iterable providing the key prefixes to be retrieved.
705
Each key prefix takes the form of a tuple the length of a key, but
706
with the last N elements 'None' rather than a regular bytestring.
707
The first element cannot be 'None'.
708
:return: An iterable as per iter_all_entries, but restricted to the
709
keys with a matching prefix to those supplied. No additional keys
710
will be returned, and every match that is in the index will be
716
# load data - also finds key lengths
717
if self._nodes is None:
719
if self._key_length == 1:
721
_sanity_check_key(self, 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()
729
for entry in _iter_entries_prefix(self, nodes_by_key, keys):
732
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
733
"""See BTreeIndex._find_ancestors."""
734
# The api can be implemented as a trivial overlay on top of
735
# iter_entries, it is not an efficient implementation, but it at least
739
for index, key, value, refs in self.iter_entries(keys):
740
parent_keys = refs[ref_list_num]
742
parent_map[key] = parent_keys
743
search_keys.update(parent_keys)
744
# Figure out what, if anything, was missing
745
missing_keys.update(set(keys).difference(found_keys))
746
search_keys = search_keys.difference(parent_map)
750
"""Return an estimate of the number of keys in this index.
752
For GraphIndex the estimate is exact.
754
if self._key_count is None:
755
self._read_and_parse([_HEADER_READV])
756
return self._key_count
758
def _lookup_keys_via_location(self, location_keys):
759
"""Public interface for implementing bisection.
761
If _buffer_all has been called, then all the data for the index is in
762
memory, and this method should not be called, as it uses a separate
763
cache because it cannot pre-resolve all indices, which buffer_all does
766
:param location_keys: A list of location(byte offset), key tuples.
767
:return: A list of (location_key, result) tuples as expected by
768
breezy.bisect_multi.bisect_multi_bytes.
770
# Possible improvements:
771
# - only bisect lookup each key once
772
# - sort the keys first, and use that to reduce the bisection window
774
# this progresses in three parts:
777
# attempt to answer the question from the now in memory data.
778
# build the readv request
779
# for each location, ask for 800 bytes - much more than rows we've seen
782
for location, key in location_keys:
783
# can we answer from cache?
784
if self._bisect_nodes and key in self._bisect_nodes:
785
# We have the key parsed.
787
index = self._parsed_key_index(key)
788
if (len(self._parsed_key_map) and
789
self._parsed_key_map[index][0] <= key and
790
(self._parsed_key_map[index][1] >= key or
791
# end of the file has been parsed
792
self._parsed_byte_map[index][1] == self._size)):
793
# the key has been parsed, so no lookup is needed even if its
796
# - if we have examined this part of the file already - yes
797
index = self._parsed_byte_index(location)
798
if (len(self._parsed_byte_map) and
799
self._parsed_byte_map[index][0] <= location and
800
self._parsed_byte_map[index][1] > location):
801
# the byte region has been parsed, so no read is needed.
804
if location + length > self._size:
805
length = self._size - location
806
# todo, trim out parsed locations.
808
readv_ranges.append((location, length))
809
# read the header if needed
810
if self._bisect_nodes is None:
811
readv_ranges.append(_HEADER_READV)
812
self._read_and_parse(readv_ranges)
814
if self._nodes is not None:
815
# _read_and_parse triggered a _buffer_all because we requested the
817
for location, key in location_keys:
818
if key not in self._nodes: # not present
819
result.append(((location, key), False))
820
elif self.node_ref_lists:
821
value, refs = self._nodes[key]
822
result.append(((location, key),
823
(self, key, value, refs)))
825
result.append(((location, key),
826
(self, key, self._nodes[key])))
829
# - figure out <, >, missing, present
830
# - result present references so we can return them.
831
# keys that we cannot answer until we resolve references
832
pending_references = []
833
pending_locations = set()
834
for location, key in location_keys:
835
# can we answer from cache?
836
if key in self._bisect_nodes:
837
# the key has been parsed, so no lookup is needed
838
if self.node_ref_lists:
839
# the references may not have been all parsed.
840
value, refs = self._bisect_nodes[key]
841
wanted_locations = []
842
for ref_list in refs:
844
if ref not in self._keys_by_offset:
845
wanted_locations.append(ref)
847
pending_locations.update(wanted_locations)
848
pending_references.append((location, key))
850
result.append(((location, key), (self, key,
851
value, self._resolve_references(refs))))
853
result.append(((location, key),
854
(self, key, self._bisect_nodes[key])))
857
# has the region the key should be in, been parsed?
858
index = self._parsed_key_index(key)
859
if (self._parsed_key_map[index][0] <= key and
860
(self._parsed_key_map[index][1] >= key or
861
# end of the file has been parsed
862
self._parsed_byte_map[index][1] == self._size)):
863
result.append(((location, key), False))
865
# no, is the key above or below the probed location:
866
# get the range of the probed & parsed location
867
index = self._parsed_byte_index(location)
868
# if the key is below the start of the range, its below
869
if key < self._parsed_key_map[index][0]:
873
result.append(((location, key), direction))
875
# lookup data to resolve references
876
for location in pending_locations:
878
if location + length > self._size:
879
length = self._size - location
880
# TODO: trim out parsed locations (e.g. if the 800 is into the
881
# parsed region trim it, and dont use the adjust_for_latency
884
readv_ranges.append((location, length))
885
self._read_and_parse(readv_ranges)
886
if self._nodes is not None:
887
# The _read_and_parse triggered a _buffer_all, grab the data and
889
for location, key in pending_references:
890
value, refs = self._nodes[key]
891
result.append(((location, key), (self, key, value, refs)))
893
for location, key in pending_references:
894
# answer key references we had to look-up-late.
895
value, refs = self._bisect_nodes[key]
896
result.append(((location, key), (self, key,
897
value, self._resolve_references(refs))))
900
def _parse_header_from_bytes(self, bytes):
901
"""Parse the header from a region of bytes.
903
:param bytes: The data to parse.
904
:return: An offset, data tuple such as readv yields, for the unparsed
905
data. (which may length 0).
907
signature = bytes[0:len(self._signature())]
908
if not signature == self._signature():
909
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
910
lines = bytes[len(self._signature()):].splitlines()
911
options_line = lines[0]
912
if not options_line.startswith(_OPTION_NODE_REFS):
913
raise errors.BadIndexOptions(self)
915
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
917
raise errors.BadIndexOptions(self)
918
options_line = lines[1]
919
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
920
raise errors.BadIndexOptions(self)
922
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
924
raise errors.BadIndexOptions(self)
925
options_line = lines[2]
926
if not options_line.startswith(_OPTION_LEN):
927
raise errors.BadIndexOptions(self)
929
self._key_count = int(options_line[len(_OPTION_LEN):])
931
raise errors.BadIndexOptions(self)
932
# calculate the bytes we have processed
933
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
935
self._parsed_bytes(0, None, header_end, None)
936
# setup parsing state
937
self._expected_elements = 3 + self._key_length
938
# raw data keyed by offset
939
self._keys_by_offset = {}
940
# keys with the value and node references
941
self._bisect_nodes = {}
942
return header_end, bytes[header_end:]
944
def _parse_region(self, offset, data):
945
"""Parse node data returned from a readv operation.
947
:param offset: The byte offset the data starts at.
948
:param data: The data to parse.
952
end = offset + len(data)
955
# Trivial test - if the current index's end is within the
956
# low-matching parsed range, we're done.
957
index = self._parsed_byte_index(high_parsed)
958
if end < self._parsed_byte_map[index][1]:
960
# print "[%d:%d]" % (offset, end), \
961
# self._parsed_byte_map[index:index + 2]
962
high_parsed, last_segment = self._parse_segment(
963
offset, data, end, index)
967
def _parse_segment(self, offset, data, end, index):
968
"""Parse one segment of data.
970
:param offset: Where 'data' begins in the file.
971
:param data: Some data to parse a segment of.
972
:param end: Where data ends
973
:param index: The current index into the parsed bytes map.
974
:return: True if the parsed segment is the last possible one in the
976
:return: high_parsed_byte, last_segment.
977
high_parsed_byte is the location of the highest parsed byte in this
978
segment, last_segment is True if the parsed segment is the last
979
possible one in the data block.
981
# default is to use all data
983
# accomodate overlap with data before this.
984
if offset < self._parsed_byte_map[index][1]:
985
# overlaps the lower parsed region
986
# skip the parsed data
987
trim_start = self._parsed_byte_map[index][1] - offset
988
# don't trim the start for \n
989
start_adjacent = True
990
elif offset == self._parsed_byte_map[index][1]:
991
# abuts the lower parsed region
994
# do not trim anything
995
start_adjacent = True
997
# does not overlap the lower parsed region
1000
# but trim the leading \n
1001
start_adjacent = False
1002
if end == self._size:
1003
# lines up to the end of all data:
1006
# do not strip to the last \n
1009
elif index + 1 == len(self._parsed_byte_map):
1010
# at the end of the parsed data
1013
# but strip to the last \n
1014
end_adjacent = False
1016
elif end == self._parsed_byte_map[index + 1][0]:
1017
# buts up against the next parsed region
1020
# do not strip to the last \n
1023
elif end > self._parsed_byte_map[index + 1][0]:
1024
# overlaps into the next parsed region
1025
# only consider the unparsed data
1026
trim_end = self._parsed_byte_map[index + 1][0] - offset
1027
# do not strip to the last \n as we know its an entire record
1029
last_segment = end < self._parsed_byte_map[index + 1][1]
1031
# does not overlap into the next region
1034
# but strip to the last \n
1035
end_adjacent = False
1037
# now find bytes to discard if needed
1038
if not start_adjacent:
1039
# work around python bug in rfind
1040
if trim_start is None:
1041
trim_start = data.find('\n') + 1
1043
trim_start = data.find('\n', trim_start) + 1
1044
if not (trim_start != 0):
1045
raise AssertionError('no \n was present')
1046
# print 'removing start', offset, trim_start, repr(data[:trim_start])
1047
if not end_adjacent:
1048
# work around python bug in rfind
1049
if trim_end is None:
1050
trim_end = data.rfind('\n') + 1
1052
trim_end = data.rfind('\n', None, trim_end) + 1
1053
if not (trim_end != 0):
1054
raise AssertionError('no \n was present')
1055
# print 'removing end', offset, trim_end, repr(data[trim_end:])
1056
# adjust offset and data to the parseable data.
1057
trimmed_data = data[trim_start:trim_end]
1058
if not (trimmed_data):
1059
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
1060
% (trim_start, trim_end, offset, offset + len(data)))
1062
offset += trim_start
1063
# print "parsing", repr(trimmed_data)
1064
# splitlines mangles the \r delimiters.. don't use it.
1065
lines = trimmed_data.split('\n')
1068
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1069
for key, value in nodes:
1070
self._bisect_nodes[key] = value
1071
self._parsed_bytes(offset, first_key,
1072
offset + len(trimmed_data), last_key)
1073
return offset + len(trimmed_data), last_segment
1075
def _parse_lines(self, lines, pos):
1082
# must be at the end
1084
if not (self._size == pos + 1):
1085
raise AssertionError("%s %s" % (self._size, pos))
1088
elements = line.split('\0')
1089
if len(elements) != self._expected_elements:
1090
raise errors.BadIndexData(self)
1091
# keys are tuples. Each element is a string that may occur many
1092
# times, so we intern them to save space. AB, RC, 200807
1093
key = tuple([intern(element) for element in elements[:self._key_length]])
1094
if first_key is None:
1096
absent, references, value = elements[-3:]
1098
for ref_string in references.split('\t'):
1099
ref_lists.append(tuple([
1100
int(ref) for ref in ref_string.split('\r') if ref
1102
ref_lists = tuple(ref_lists)
1103
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1104
pos += len(line) + 1 # +1 for the \n
1107
if self.node_ref_lists:
1108
node_value = (value, ref_lists)
1111
nodes.append((key, node_value))
1112
# print "parsed ", key
1113
return first_key, key, nodes, trailers
1115
def _parsed_bytes(self, start, start_key, end, end_key):
1116
"""Mark the bytes from start to end as parsed.
1118
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1121
:param start: The start of the parsed region.
1122
:param end: The end of the parsed region.
1124
index = self._parsed_byte_index(start)
1125
new_value = (start, end)
1126
new_key = (start_key, end_key)
1128
# first range parsed is always the beginning.
1129
self._parsed_byte_map.insert(index, new_value)
1130
self._parsed_key_map.insert(index, new_key)
1134
# extend lower region
1135
# extend higher region
1136
# combine two regions
1137
if (index + 1 < len(self._parsed_byte_map) and
1138
self._parsed_byte_map[index][1] == start and
1139
self._parsed_byte_map[index + 1][0] == end):
1140
# combine two regions
1141
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1142
self._parsed_byte_map[index + 1][1])
1143
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1144
self._parsed_key_map[index + 1][1])
1145
del self._parsed_byte_map[index + 1]
1146
del self._parsed_key_map[index + 1]
1147
elif self._parsed_byte_map[index][1] == start:
1148
# extend the lower entry
1149
self._parsed_byte_map[index] = (
1150
self._parsed_byte_map[index][0], end)
1151
self._parsed_key_map[index] = (
1152
self._parsed_key_map[index][0], end_key)
1153
elif (index + 1 < len(self._parsed_byte_map) and
1154
self._parsed_byte_map[index + 1][0] == end):
1155
# extend the higher entry
1156
self._parsed_byte_map[index + 1] = (
1157
start, self._parsed_byte_map[index + 1][1])
1158
self._parsed_key_map[index + 1] = (
1159
start_key, self._parsed_key_map[index + 1][1])
1162
self._parsed_byte_map.insert(index + 1, new_value)
1163
self._parsed_key_map.insert(index + 1, new_key)
1165
def _read_and_parse(self, readv_ranges):
1166
"""Read the ranges and parse the resulting data.
1168
:param readv_ranges: A prepared readv range list.
1170
if not readv_ranges:
1172
if self._nodes is None and self._bytes_read * 2 >= self._size:
1173
# We've already read more than 50% of the file and we are about to
1174
# request more data, just _buffer_all() and be done
1178
base_offset = self._base_offset
1179
if base_offset != 0:
1180
# Rewrite the ranges for the offset
1181
readv_ranges = [(start+base_offset, size)
1182
for start, size in readv_ranges]
1183
readv_data = self._transport.readv(self._name, readv_ranges, True,
1184
self._size + self._base_offset)
1186
for offset, data in readv_data:
1187
offset -= base_offset
1188
self._bytes_read += len(data)
1190
# transport.readv() expanded to extra data which isn't part of
1192
data = data[-offset:]
1194
if offset == 0 and len(data) == self._size:
1195
# We read the whole range, most likely because the
1196
# Transport upcast our readv ranges into one long request
1197
# for enough total data to grab the whole index.
1198
self._buffer_all(BytesIO(data))
1200
if self._bisect_nodes is None:
1201
# this must be the start
1202
if not (offset == 0):
1203
raise AssertionError()
1204
offset, data = self._parse_header_from_bytes(data)
1205
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1206
self._parse_region(offset, data)
1208
def _signature(self):
1209
"""The file signature for this index type."""
1213
"""Validate that everything in the index can be accessed."""
1214
# iter_all validates completely at the moment, so just do that.
1215
for node in self.iter_all_entries():
1219
class CombinedGraphIndex(object):
1220
"""A GraphIndex made up from smaller GraphIndices.
1222
The backing indices must implement GraphIndex, and are presumed to be
1225
Queries against the combined index will be made against the first index,
1226
and then the second and so on. The order of indices can thus influence
1227
performance significantly. For example, if one index is on local disk and a
1228
second on a remote server, the local disk index should be before the other
1231
Also, queries tend to need results from the same indices as previous
1232
queries. So the indices will be reordered after every query to put the
1233
indices that had the result(s) of that query first (while otherwise
1234
preserving the relative ordering).
1237
def __init__(self, indices, reload_func=None):
1238
"""Create a CombinedGraphIndex backed by indices.
1240
:param indices: An ordered list of indices to query for data.
1241
:param reload_func: A function to call if we find we are missing an
1242
index. Should have the form reload_func() => True/False to indicate
1243
if reloading actually changed anything.
1245
self._indices = indices
1246
self._reload_func = reload_func
1247
# Sibling indices are other CombinedGraphIndex that we should call
1248
# _move_to_front_by_name on when we auto-reorder ourself.
1249
self._sibling_indices = []
1250
# A list of names that corresponds to the instances in self._indices,
1251
# so _index_names[0] is always the name for _indices[0], etc. Sibling
1252
# indices must all use the same set of names as each other.
1253
self._index_names = [None] * len(self._indices)
1257
self.__class__.__name__,
1258
', '.join(map(repr, self._indices)))
1260
def clear_cache(self):
1261
"""See GraphIndex.clear_cache()"""
1262
for index in self._indices:
1265
def get_parent_map(self, keys):
1266
"""See graph.StackedParentsProvider.get_parent_map"""
1267
search_keys = set(keys)
1268
if _mod_revision.NULL_REVISION in search_keys:
1269
search_keys.discard(_mod_revision.NULL_REVISION)
1270
found_parents = {_mod_revision.NULL_REVISION:[]}
1273
for index, key, value, refs in self.iter_entries(search_keys):
1276
parents = (_mod_revision.NULL_REVISION,)
1277
found_parents[key] = parents
1278
return found_parents
1280
__contains__ = _has_key_from_parent_map
1282
def insert_index(self, pos, index, name=None):
1283
"""Insert a new index in the list of indices to query.
1285
:param pos: The position to insert the index.
1286
:param index: The index to insert.
1287
:param name: a name for this index, e.g. a pack name. These names can
1288
be used to reflect index reorderings to related CombinedGraphIndex
1289
instances that use the same names. (see set_sibling_indices)
1291
self._indices.insert(pos, index)
1292
self._index_names.insert(pos, name)
1294
def iter_all_entries(self):
1295
"""Iterate over all keys within the index
1297
Duplicate keys across child indices are presumed to have the same
1298
value and are only reported once.
1300
:return: An iterable of (index, key, reference_lists, value).
1301
There is no defined order for the result iteration - it will be in
1302
the most efficient order for the index.
1307
for index in self._indices:
1308
for node in index.iter_all_entries():
1309
if node[1] not in seen_keys:
1311
seen_keys.add(node[1])
1313
except errors.NoSuchFile as e:
1314
if not self._try_reload(e):
1317
def iter_entries(self, keys):
1318
"""Iterate over keys within the index.
1320
Duplicate keys across child indices are presumed to have the same
1321
value and are only reported once.
1323
:param keys: An iterable providing the keys to be retrieved.
1324
:return: An iterable of (index, key, reference_lists, value). There is
1325
no defined order for the result iteration - it will be in the most
1326
efficient order for the index.
1332
for index in self._indices:
1336
for node in index.iter_entries(keys):
1337
keys.remove(node[1])
1341
hit_indices.append(index)
1343
except errors.NoSuchFile as e:
1344
if not self._try_reload(e):
1346
self._move_to_front(hit_indices)
1348
def iter_entries_prefix(self, keys):
1349
"""Iterate over keys within the index using prefix matching.
1351
Duplicate keys across child indices are presumed to have the same
1352
value and are only reported once.
1354
Prefix matching is applied within the tuple of a key, not to within
1355
the bytestring of each key element. e.g. if you have the keys ('foo',
1356
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1357
only the former key is returned.
1359
:param keys: An iterable providing the key prefixes to be retrieved.
1360
Each key prefix takes the form of a tuple the length of a key, but
1361
with the last N elements 'None' rather than a regular bytestring.
1362
The first element cannot be 'None'.
1363
:return: An iterable as per iter_all_entries, but restricted to the
1364
keys with a matching prefix to those supplied. No additional keys
1365
will be returned, and every match that is in the index will be
1375
for index in self._indices:
1377
for node in index.iter_entries_prefix(keys):
1378
if node[1] in seen_keys:
1380
seen_keys.add(node[1])
1384
hit_indices.append(index)
1386
except errors.NoSuchFile as e:
1387
if not self._try_reload(e):
1389
self._move_to_front(hit_indices)
1391
def _move_to_front(self, hit_indices):
1392
"""Rearrange self._indices so that hit_indices are first.
1394
Order is maintained as much as possible, e.g. the first unhit index
1395
will be the first index in _indices after the hit_indices, and the
1396
hit_indices will be present in exactly the order they are passed to
1399
_move_to_front propagates to all objects in self._sibling_indices by
1400
calling _move_to_front_by_name.
1402
if self._indices[:len(hit_indices)] == hit_indices:
1403
# The 'hit_indices' are already at the front (and in the same
1404
# order), no need to re-order
1406
hit_names = self._move_to_front_by_index(hit_indices)
1407
for sibling_idx in self._sibling_indices:
1408
sibling_idx._move_to_front_by_name(hit_names)
1410
def _move_to_front_by_index(self, hit_indices):
1411
"""Core logic for _move_to_front.
1413
Returns a list of names corresponding to the hit_indices param.
1415
indices_info = zip(self._index_names, self._indices)
1416
if 'index' in debug.debug_flags:
1417
indices_info = list(indices_info)
1418
trace.mutter('CombinedGraphIndex reordering: currently %r, '
1419
'promoting %r', indices_info, hit_indices)
1422
new_hit_indices = []
1425
for offset, (name, idx) in enumerate(indices_info):
1426
if idx in hit_indices:
1427
hit_names.append(name)
1428
new_hit_indices.append(idx)
1429
if len(new_hit_indices) == len(hit_indices):
1430
# We've found all of the hit entries, everything else is
1432
unhit_names.extend(self._index_names[offset+1:])
1433
unhit_indices.extend(self._indices[offset+1:])
1436
unhit_names.append(name)
1437
unhit_indices.append(idx)
1439
self._indices = new_hit_indices + unhit_indices
1440
self._index_names = hit_names + unhit_names
1441
if 'index' in debug.debug_flags:
1442
trace.mutter('CombinedGraphIndex reordered: %r', self._indices)
1445
def _move_to_front_by_name(self, hit_names):
1446
"""Moves indices named by 'hit_names' to front of the search order, as
1447
described in _move_to_front.
1449
# Translate names to index instances, and then call
1450
# _move_to_front_by_index.
1451
indices_info = zip(self._index_names, self._indices)
1453
for name, idx in indices_info:
1454
if name in hit_names:
1455
hit_indices.append(idx)
1456
self._move_to_front_by_index(hit_indices)
1458
def find_ancestry(self, keys, ref_list_num):
1459
"""Find the complete ancestry for the given set of keys.
1461
Note that this is a whole-ancestry request, so it should be used
1464
:param keys: An iterable of keys to look for
1465
:param ref_list_num: The reference list which references the parents
1467
:return: (parent_map, missing_keys)
1469
# XXX: make this call _move_to_front?
1470
missing_keys = set()
1472
keys_to_lookup = set(keys)
1474
while keys_to_lookup:
1475
# keys that *all* indexes claim are missing, stop searching them
1477
all_index_missing = None
1478
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1479
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1481
# len(missing_keys))
1482
for index_idx, index in enumerate(self._indices):
1483
# TODO: we should probably be doing something with
1484
# 'missing_keys' since we've already determined that
1485
# those revisions have not been found anywhere
1486
index_missing_keys = set()
1487
# Find all of the ancestry we can from this index
1488
# keep looking until the search_keys set is empty, which means
1489
# things we didn't find should be in index_missing_keys
1490
search_keys = keys_to_lookup
1492
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1493
# index_idx, len(search_keys),
1494
# len(parent_map), len(index_missing_keys))
1497
# TODO: ref_list_num should really be a parameter, since
1498
# CombinedGraphIndex does not know what the ref lists
1500
search_keys = index._find_ancestors(search_keys,
1501
ref_list_num, parent_map, index_missing_keys)
1502
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1503
# sub_generation, len(search_keys),
1504
# len(parent_map), len(index_missing_keys))
1505
# Now set whatever was missing to be searched in the next index
1506
keys_to_lookup = index_missing_keys
1507
if all_index_missing is None:
1508
all_index_missing = set(index_missing_keys)
1510
all_index_missing.intersection_update(index_missing_keys)
1511
if not keys_to_lookup:
1513
if all_index_missing is None:
1514
# There were no indexes, so all search keys are 'missing'
1515
missing_keys.update(keys_to_lookup)
1516
keys_to_lookup = None
1518
missing_keys.update(all_index_missing)
1519
keys_to_lookup.difference_update(all_index_missing)
1520
return parent_map, missing_keys
1522
def key_count(self):
1523
"""Return an estimate of the number of keys in this index.
1525
For CombinedGraphIndex this is approximated by the sum of the keys of
1526
the child indices. As child indices may have duplicate keys this can
1527
have a maximum error of the number of child indices * largest number of
1532
return sum((index.key_count() for index in self._indices), 0)
1533
except errors.NoSuchFile as e:
1534
if not self._try_reload(e):
1537
missing_keys = _missing_keys_from_parent_map
1539
def _try_reload(self, error):
1540
"""We just got a NoSuchFile exception.
1542
Try to reload the indices, if it fails, just raise the current
1545
if self._reload_func is None:
1547
trace.mutter('Trying to reload after getting exception: %s', error)
1548
if not self._reload_func():
1549
# We tried to reload, but nothing changed, so we fail anyway
1550
trace.mutter('_reload_func indicated nothing has changed.'
1551
' Raising original exception.')
1555
def set_sibling_indices(self, sibling_combined_graph_indices):
1556
"""Set the CombinedGraphIndex objects to reorder after reordering self.
1558
self._sibling_indices = sibling_combined_graph_indices
1561
"""Validate that everything in the index can be accessed."""
1564
for index in self._indices:
1567
except errors.NoSuchFile as e:
1568
if not self._try_reload(e):
1572
class InMemoryGraphIndex(GraphIndexBuilder):
1573
"""A GraphIndex which operates entirely out of memory and is mutable.
1575
This is designed to allow the accumulation of GraphIndex entries during a
1576
single write operation, where the accumulated entries need to be immediately
1577
available - for example via a CombinedGraphIndex.
1580
def add_nodes(self, nodes):
1581
"""Add nodes to the index.
1583
:param nodes: An iterable of (key, node_refs, value) entries to add.
1585
if self.reference_lists:
1586
for (key, value, node_refs) in nodes:
1587
self.add_node(key, value, node_refs)
1589
for (key, value) in nodes:
1590
self.add_node(key, value)
1592
def iter_all_entries(self):
1593
"""Iterate over all keys within the index
1595
:return: An iterable of (index, key, reference_lists, value). There is no
1596
defined order for the result iteration - it will be in the most
1597
efficient order for the index (in this case dictionary hash order).
1599
if 'evil' in debug.debug_flags:
1600
trace.mutter_callsite(3,
1601
"iter_all_entries scales with size of history.")
1602
if self.reference_lists:
1603
for key, (absent, references, value) in viewitems(self._nodes):
1605
yield self, key, value, references
1607
for key, (absent, references, value) in viewitems(self._nodes):
1609
yield self, key, value
1611
def iter_entries(self, keys):
1612
"""Iterate over keys within the index.
1614
:param keys: An iterable providing the keys to be retrieved.
1615
:return: An iterable of (index, key, value, reference_lists). There is no
1616
defined order for the result iteration - it will be in the most
1617
efficient order for the index (keys iteration order in this case).
1619
# Note: See BTreeBuilder.iter_entries for an explanation of why we
1620
# aren't using set().intersection() here
1622
keys = [key for key in keys if key in nodes]
1623
if self.reference_lists:
1627
yield self, key, node[2], node[1]
1632
yield self, key, node[2]
1634
def iter_entries_prefix(self, keys):
1635
"""Iterate over keys within the index using prefix matching.
1637
Prefix matching is applied within the tuple of a key, not to within
1638
the bytestring of each key element. e.g. if you have the keys ('foo',
1639
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1640
only the former key is returned.
1642
:param keys: An iterable providing the key prefixes to be retrieved.
1643
Each key prefix takes the form of a tuple the length of a key, but
1644
with the last N elements 'None' rather than a regular bytestring.
1645
The first element cannot be 'None'.
1646
:return: An iterable as per iter_all_entries, but restricted to the
1647
keys with a matching prefix to those supplied. No additional keys
1648
will be returned, and every match that is in the index will be
1654
if self._key_length == 1:
1656
_sanity_check_key(self, key)
1657
node = self._nodes[key]
1660
if self.reference_lists:
1661
yield self, key, node[2], node[1]
1663
yield self, key, node[2]
1665
nodes_by_key = self._get_nodes_by_key()
1666
for entry in _iter_entries_prefix(self, nodes_by_key, keys):
1669
def key_count(self):
1670
"""Return an estimate of the number of keys in this index.
1672
For InMemoryGraphIndex the estimate is exact.
1674
return len(self._nodes) - len(self._absent_keys)
1677
"""In memory index's have no known corruption at the moment."""
1680
class GraphIndexPrefixAdapter(object):
1681
"""An adapter between GraphIndex with different key lengths.
1683
Queries against this will emit queries against the adapted Graph with the
1684
prefix added, queries for all items use iter_entries_prefix. The returned
1685
nodes will have their keys and node references adjusted to remove the
1686
prefix. Finally, an add_nodes_callback can be supplied - when called the
1687
nodes and references being added will have prefix prepended.
1690
def __init__(self, adapted, prefix, missing_key_length,
1691
add_nodes_callback=None):
1692
"""Construct an adapter against adapted with prefix."""
1693
self.adapted = adapted
1694
self.prefix_key = prefix + (None,)*missing_key_length
1695
self.prefix = prefix
1696
self.prefix_len = len(prefix)
1697
self.add_nodes_callback = add_nodes_callback
1699
def add_nodes(self, nodes):
1700
"""Add nodes to the index.
1702
:param nodes: An iterable of (key, node_refs, value) entries to add.
1704
# save nodes in case its an iterator
1705
nodes = tuple(nodes)
1706
translated_nodes = []
1708
# Add prefix_key to each reference node_refs is a tuple of tuples,
1709
# so split it apart, and add prefix_key to the internal reference
1710
for (key, value, node_refs) in nodes:
1711
adjusted_references = (
1712
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1713
for ref_list in node_refs))
1714
translated_nodes.append((self.prefix + key, value,
1715
adjusted_references))
1717
# XXX: TODO add an explicit interface for getting the reference list
1718
# status, to handle this bit of user-friendliness in the API more
1720
for (key, value) in nodes:
1721
translated_nodes.append((self.prefix + key, value))
1722
self.add_nodes_callback(translated_nodes)
1724
def add_node(self, key, value, references=()):
1725
"""Add a node to the index.
1727
:param key: The key. keys are non-empty tuples containing
1728
as many whitespace-free utf8 bytestrings as the key length
1729
defined for this index.
1730
:param references: An iterable of iterables of keys. Each is a
1731
reference to another key.
1732
:param value: The value to associate with the key. It may be any
1733
bytes as long as it does not contain \0 or \n.
1735
self.add_nodes(((key, value, references), ))
1737
def _strip_prefix(self, an_iter):
1738
"""Strip prefix data from nodes and return it."""
1739
for node in an_iter:
1741
if node[1][:self.prefix_len] != self.prefix:
1742
raise errors.BadIndexData(self)
1743
for ref_list in node[3]:
1744
for ref_node in ref_list:
1745
if ref_node[:self.prefix_len] != self.prefix:
1746
raise errors.BadIndexData(self)
1747
yield node[0], node[1][self.prefix_len:], node[2], (
1748
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1749
for ref_list in node[3]))
1751
def iter_all_entries(self):
1752
"""Iterate over all keys within the index
1754
iter_all_entries is implemented against the adapted index using
1755
iter_entries_prefix.
1757
:return: An iterable of (index, key, reference_lists, value). There is no
1758
defined order for the result iteration - it will be in the most
1759
efficient order for the index (in this case dictionary hash order).
1761
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1763
def iter_entries(self, keys):
1764
"""Iterate over keys within the index.
1766
:param keys: An iterable providing the keys to be retrieved.
1767
:return: An iterable of (index, key, value, reference_lists). There is no
1768
defined order for the result iteration - it will be in the most
1769
efficient order for the index (keys iteration order in this case).
1771
return self._strip_prefix(self.adapted.iter_entries(
1772
self.prefix + key for key in keys))
1774
def iter_entries_prefix(self, keys):
1775
"""Iterate over keys within the index using prefix matching.
1777
Prefix matching is applied within the tuple of a key, not to within
1778
the bytestring of each key element. e.g. if you have the keys ('foo',
1779
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1780
only the former key is returned.
1782
:param keys: An iterable providing the key prefixes to be retrieved.
1783
Each key prefix takes the form of a tuple the length of a key, but
1784
with the last N elements 'None' rather than a regular bytestring.
1785
The first element cannot be 'None'.
1786
:return: An iterable as per iter_all_entries, but restricted to the
1787
keys with a matching prefix to those supplied. No additional keys
1788
will be returned, and every match that is in the index will be
1791
return self._strip_prefix(self.adapted.iter_entries_prefix(
1792
self.prefix + key for key in keys))
1794
def key_count(self):
1795
"""Return an estimate of the number of keys in this index.
1797
For GraphIndexPrefixAdapter this is relatively expensive - key
1798
iteration with the prefix is done.
1800
return len(list(self.iter_all_entries()))
1803
"""Call the adapted's validate."""
1804
self.adapted.validate()
1807
def _sanity_check_key(index_or_builder, key):
1808
"""Raise BadIndexKey if key cannot be used for prefix matching."""
1810
raise errors.BadIndexKey(key)
1811
if len(key) != index_or_builder._key_length:
1812
raise errors.BadIndexKey(key)
1815
def _iter_entries_prefix(index_or_builder, nodes_by_key, keys):
1816
"""Helper for implementing prefix matching iterators."""
1818
_sanity_check_key(index_or_builder, key)
1819
# find what it refers to:
1820
key_dict = nodes_by_key
1821
elements = list(key)
1822
# find the subdict whose contents should be returned.
1824
while len(elements) and elements[0] is not None:
1825
key_dict = key_dict[elements[0]]
1828
# a non-existant lookup.
1833
values_view = viewvalues(dicts.pop())
1834
# can't be empty or would not exist
1835
value = next(iter(values_view))
1836
if isinstance(value, dict):
1837
# still descending, push values
1838
dicts.extend(values_view)
1840
# at leaf tuples, yield values
1841
for value in values_view:
1842
# each value is the key:value:node refs tuple
1844
yield (index_or_builder, ) + value
1846
# the last thing looked up was a terminal element
1847
yield (index_or_builder, ) + key_dict
added
removed
breezy/index.py
