1
# Copyright (C) 2007 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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Indexing facilities."""
23
'GraphIndexPrefixAdapter',
27
from bisect import bisect_right
28
from cStringIO import StringIO
31
from bzrlib.lazy_import import lazy_import
32
lazy_import(globals(), """
33
from bzrlib import trace
34
from bzrlib.bisect_multi import bisect_multi_bytes
35
from bzrlib.trace import mutter
37
from bzrlib import debug, errors
39
_OPTION_KEY_ELEMENTS = "key_elements="
41
_OPTION_NODE_REFS = "node_ref_lists="
42
_SIGNATURE = "Bazaar Graph Index 1\n"
45
_whitespace_re = re.compile('[\t\n\x0b\x0c\r\x00 ]')
46
_newline_null_re = re.compile('[\n\0]')
49
class GraphIndexBuilder(object):
50
"""A builder that can build a GraphIndex.
52
The resulting graph has the structure:
54
_SIGNATURE OPTIONS NODES NEWLINE
55
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
56
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
58
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
59
KEY := Not-whitespace-utf8
61
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
62
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
63
REFERENCE := DIGITS ; digits is the byte offset in the index of the
65
VALUE := no-newline-no-null-bytes
68
def __init__(self, reference_lists=0, key_elements=1):
69
"""Create a GraphIndex builder.
71
:param reference_lists: The number of node references lists for each
73
:param key_elements: The number of bytestrings in each key.
75
self.reference_lists = reference_lists
78
self._nodes_by_key = {}
79
self._key_length = key_elements
81
def _check_key(self, key):
82
"""Raise BadIndexKey if key is not a valid key for this index."""
83
if type(key) != tuple:
84
raise errors.BadIndexKey(key)
85
if self._key_length != len(key):
86
raise errors.BadIndexKey(key)
88
if not element or _whitespace_re.search(element) is not None:
89
raise errors.BadIndexKey(element)
91
def add_node(self, key, value, references=()):
92
"""Add a node to the index.
94
:param key: The key. keys are non-empty tuples containing
95
as many whitespace-free utf8 bytestrings as the key length
96
defined for this index.
97
:param references: An iterable of iterables of keys. Each is a
98
reference to another key.
99
:param value: The value to associate with the key. It may be any
100
bytes as long as it does not contain \0 or \n.
103
if _newline_null_re.search(value) is not None:
104
raise errors.BadIndexValue(value)
105
if len(references) != self.reference_lists:
106
raise errors.BadIndexValue(references)
108
for reference_list in references:
109
for reference in reference_list:
110
self._check_key(reference)
111
if reference not in self._nodes:
112
self._nodes[reference] = ('a', (), '')
113
node_refs.append(tuple(reference_list))
114
if key in self._nodes and self._nodes[key][0] == '':
115
raise errors.BadIndexDuplicateKey(key, self)
116
self._nodes[key] = ('', tuple(node_refs), value)
118
if self._key_length > 1:
119
key_dict = self._nodes_by_key
120
if self.reference_lists:
121
key_value = key, value, tuple(node_refs)
123
key_value = key, value
124
# possibly should do this on-demand, but it seems likely it is
126
# For a key of (foo, bar, baz) create
127
# _nodes_by_key[foo][bar][baz] = key_value
128
for subkey in key[:-1]:
129
key_dict = key_dict.setdefault(subkey, {})
130
key_dict[key[-1]] = key_value
134
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
135
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
136
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
137
prefix_length = sum(len(x) for x in lines)
138
# references are byte offsets. To avoid having to do nasty
139
# polynomial work to resolve offsets (references to later in the
140
# file cannot be determined until all the inbetween references have
141
# been calculated too) we pad the offsets with 0's to make them be
142
# of consistent length. Using binary offsets would break the trivial
144
# to calculate the width of zero's needed we do three passes:
145
# one to gather all the non-reference data and the number of references.
146
# one to pad all the data with reference-length and determine entry
150
# forward sorted by key. In future we may consider topological sorting,
151
# at the cost of table scans for direct lookup, or a second index for
153
nodes = sorted(self._nodes.items())
154
# if we do not prepass, we don't know how long it will be up front.
155
expected_bytes = None
156
# we only need to pre-pass if we have reference lists at all.
157
if self.reference_lists:
159
non_ref_bytes = prefix_length
161
# TODO use simple multiplication for the constants in this loop.
162
for key, (absent, references, value) in nodes:
163
# record the offset known *so far* for this key:
164
# the non reference bytes to date, and the total references to
165
# date - saves reaccumulating on the second pass
166
key_offset_info.append((key, non_ref_bytes, total_references))
167
# key is literal, value is literal, there are 3 null's, 1 NL
168
# key is variable length tuple, \x00 between elements
169
non_ref_bytes += sum(len(element) for element in key)
170
if self._key_length > 1:
171
non_ref_bytes += self._key_length - 1
172
# value is literal bytes, there are 3 null's, 1 NL.
173
non_ref_bytes += len(value) + 3 + 1
174
# one byte for absent if set.
177
elif self.reference_lists:
178
# (ref_lists -1) tabs
179
non_ref_bytes += self.reference_lists - 1
180
# (ref-1 cr's per ref_list)
181
for ref_list in references:
182
# how many references across the whole file?
183
total_references += len(ref_list)
184
# accrue reference separators
186
non_ref_bytes += len(ref_list) - 1
187
# how many digits are needed to represent the total byte count?
189
possible_total_bytes = non_ref_bytes + total_references*digits
190
while 10 ** digits < possible_total_bytes:
192
possible_total_bytes = non_ref_bytes + total_references*digits
193
expected_bytes = possible_total_bytes + 1 # terminating newline
194
# resolve key addresses.
196
for key, non_ref_bytes, total_references in key_offset_info:
197
key_addresses[key] = non_ref_bytes + total_references*digits
199
format_string = '%%0%sd' % digits
200
for key, (absent, references, value) in nodes:
201
flattened_references = []
202
for ref_list in references:
204
for reference in ref_list:
205
ref_addresses.append(format_string % key_addresses[reference])
206
flattened_references.append('\r'.join(ref_addresses))
207
string_key = '\x00'.join(key)
208
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
209
'\t'.join(flattened_references), value))
211
result = StringIO(''.join(lines))
212
if expected_bytes and len(result.getvalue()) != expected_bytes:
213
raise errors.BzrError('Failed index creation. Internal error:'
214
' mismatched output length and expected length: %d %d' %
215
(len(result.getvalue()), expected_bytes))
216
return StringIO(''.join(lines))
219
class GraphIndex(object):
220
"""An index for data with embedded graphs.
222
The index maps keys to a list of key reference lists, and a value.
223
Each node has the same number of key reference lists. Each key reference
224
list can be empty or an arbitrary length. The value is an opaque NULL
225
terminated string without any newlines. The storage of the index is
226
hidden in the interface: keys and key references are always tuples of
227
bytestrings, never the internal representation (e.g. dictionary offsets).
229
It is presumed that the index will not be mutated - it is static data.
231
Successive iter_all_entries calls will read the entire index each time.
232
Additionally, iter_entries calls will read the index linearly until the
233
desired keys are found. XXX: This must be fixed before the index is
234
suitable for production use. :XXX
237
def __init__(self, transport, name, size):
238
"""Open an index called name on transport.
240
:param transport: A bzrlib.transport.Transport.
241
:param name: A path to provide to transport API calls.
242
:param size: The size of the index in bytes. This is used for bisection
243
logic to perform partial index reads. While the size could be
244
obtained by statting the file this introduced an additional round
245
trip as well as requiring stat'able transports, both of which are
246
avoided by having it supplied. If size is None, then bisection
247
support will be disabled and accessing the index will just stream
250
self._transport = transport
252
# Becomes a dict of key:(value, reference-list-byte-locations) used by
253
# the bisection interface to store parsed but not resolved keys.
254
self._bisect_nodes = None
255
# Becomes a dict of key:(value, reference-list-keys) which are ready to
256
# be returned directly to callers.
258
# a sorted list of slice-addresses for the parsed bytes of the file.
259
# e.g. (0,1) would mean that byte 0 is parsed.
260
self._parsed_byte_map = []
261
# a sorted list of keys matching each slice address for parsed bytes
262
# e.g. (None, 'foo@bar') would mean that the first byte contained no
263
# key, and the end byte of the slice is the of the data for 'foo@bar'
264
self._parsed_key_map = []
265
self._key_count = None
266
self._keys_by_offset = None
267
self._nodes_by_key = None
270
def _buffer_all(self):
271
"""Buffer all the index data.
273
Mutates self._nodes and self.keys_by_offset.
275
if 'index' in debug.debug_flags:
276
mutter('Reading entire index %s', self._transport.abspath(self._name))
277
stream = self._transport.get(self._name)
278
self._read_prefix(stream)
279
self._expected_elements = 3 + self._key_length
281
# raw data keyed by offset
282
self._keys_by_offset = {}
283
# ready-to-return key:value or key:value, node_ref_lists
285
self._nodes_by_key = {}
288
lines = stream.read().split('\n')
290
_, _, _, trailers = self._parse_lines(lines, pos)
291
for key, absent, references, value in self._keys_by_offset.itervalues():
294
# resolve references:
295
if self.node_ref_lists:
296
node_value = (value, self._resolve_references(references))
299
self._nodes[key] = node_value
300
if self._key_length > 1:
301
subkey = list(reversed(key[:-1]))
302
key_dict = self._nodes_by_key
303
if self.node_ref_lists:
304
key_value = key, node_value[0], node_value[1]
306
key_value = key, node_value
307
# possibly should do this on-demand, but it seems likely it is
309
# For a key of (foo, bar, baz) create
310
# _nodes_by_key[foo][bar][baz] = key_value
311
for subkey in key[:-1]:
312
key_dict = key_dict.setdefault(subkey, {})
313
key_dict[key[-1]] = key_value
314
# cache the keys for quick set intersections
315
self._keys = set(self._nodes)
317
# there must be one line - the empty trailer line.
318
raise errors.BadIndexData(self)
320
def iter_all_entries(self):
321
"""Iterate over all keys within the index.
323
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
324
The former tuple is used when there are no reference lists in the
325
index, making the API compatible with simple key:value index types.
326
There is no defined order for the result iteration - it will be in
327
the most efficient order for the index.
329
if 'evil' in debug.debug_flags:
330
trace.mutter_callsite(3,
331
"iter_all_entries scales with size of history.")
332
if self._nodes is None:
334
if self.node_ref_lists:
335
for key, (value, node_ref_lists) in self._nodes.iteritems():
336
yield self, key, value, node_ref_lists
338
for key, value in self._nodes.iteritems():
339
yield self, key, value
341
def _read_prefix(self, stream):
342
signature = stream.read(len(self._signature()))
343
if not signature == self._signature():
344
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
345
options_line = stream.readline()
346
if not options_line.startswith(_OPTION_NODE_REFS):
347
raise errors.BadIndexOptions(self)
349
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
351
raise errors.BadIndexOptions(self)
352
options_line = stream.readline()
353
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
354
raise errors.BadIndexOptions(self)
356
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
358
raise errors.BadIndexOptions(self)
359
options_line = stream.readline()
360
if not options_line.startswith(_OPTION_LEN):
361
raise errors.BadIndexOptions(self)
363
self._key_count = int(options_line[len(_OPTION_LEN):-1])
365
raise errors.BadIndexOptions(self)
367
def _resolve_references(self, references):
368
"""Return the resolved key references for references.
370
References are resolved by looking up the location of the key in the
371
_keys_by_offset map and substituting the key name, preserving ordering.
373
:param references: An iterable of iterables of key locations. e.g.
375
:return: A tuple of tuples of keys.
378
for ref_list in references:
379
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
380
return tuple(node_refs)
382
def _find_index(self, range_map, key):
383
"""Helper for the _parsed_*_index calls.
385
Given a range map - [(start, end), ...], finds the index of the range
386
in the map for key if it is in the map, and if it is not there, the
387
immediately preceeding range in the map.
389
result = bisect_right(range_map, key) - 1
390
if result + 1 < len(range_map):
391
# check the border condition, it may be in result + 1
392
if range_map[result + 1][0] == key[0]:
396
def _parsed_byte_index(self, offset):
397
"""Return the index of the entry immediately before offset.
399
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
400
there is one unparsed byte (the 11th, addressed as[10]). then:
401
asking for 0 will return 0
402
asking for 10 will return 0
403
asking for 11 will return 1
404
asking for 12 will return 1
407
return self._find_index(self._parsed_byte_map, key)
409
def _parsed_key_index(self, key):
410
"""Return the index of the entry immediately before key.
412
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
413
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
414
have been parsed, then:
415
asking for '' will return 0
416
asking for 'a' will return 0
417
asking for 'b' will return 1
418
asking for 'e' will return 1
420
search_key = (key, None)
421
return self._find_index(self._parsed_key_map, search_key)
423
def _is_parsed(self, offset):
424
"""Returns True if offset has been parsed."""
425
index = self._parsed_byte_index(offset)
426
if index == len(self._parsed_byte_map):
427
return offset < self._parsed_byte_map[index - 1][1]
428
start, end = self._parsed_byte_map[index]
429
return offset >= start and offset < end
431
def _iter_entries_from_total_buffer(self, keys):
432
"""Iterate over keys when the entire index is parsed."""
433
keys = keys.intersection(self._keys)
434
if self.node_ref_lists:
436
value, node_refs = self._nodes[key]
437
yield self, key, value, node_refs
440
yield self, key, self._nodes[key]
442
def iter_entries(self, keys):
443
"""Iterate over keys within the index.
445
:param keys: An iterable providing the keys to be retrieved.
446
:return: An iterable as per iter_all_entries, but restricted to the
447
keys supplied. No additional keys will be returned, and every
448
key supplied that is in the index will be returned.
450
# PERFORMANCE TODO: parse and bisect all remaining data at some
451
# threshold of total-index processing/get calling layers that expect to
452
# read the entire index to use the iter_all_entries method instead.
456
if self._size is None and self._nodes is None:
458
if self._nodes is not None:
459
return self._iter_entries_from_total_buffer(keys)
461
return (result[1] for result in bisect_multi_bytes(
462
self._lookup_keys_via_location, self._size, keys))
464
def iter_entries_prefix(self, keys):
465
"""Iterate over keys within the index using prefix matching.
467
Prefix matching is applied within the tuple of a key, not to within
468
the bytestring of each key element. e.g. if you have the keys ('foo',
469
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
470
only the former key is returned.
472
WARNING: Note that this method currently causes a full index parse
473
unconditionally (which is reasonably appropriate as it is a means for
474
thunking many small indices into one larger one and still supplies
475
iter_all_entries at the thunk layer).
477
:param keys: An iterable providing the key prefixes to be retrieved.
478
Each key prefix takes the form of a tuple the length of a key, but
479
with the last N elements 'None' rather than a regular bytestring.
480
The first element cannot be 'None'.
481
:return: An iterable as per iter_all_entries, but restricted to the
482
keys with a matching prefix to those supplied. No additional keys
483
will be returned, and every match that is in the index will be
489
# load data - also finds key lengths
490
if self._nodes is None:
492
if self._key_length == 1:
496
raise errors.BadIndexKey(key)
497
if len(key) != self._key_length:
498
raise errors.BadIndexKey(key)
499
if self.node_ref_lists:
500
value, node_refs = self._nodes[key]
501
yield self, key, value, node_refs
503
yield self, key, self._nodes[key]
508
raise errors.BadIndexKey(key)
509
if len(key) != self._key_length:
510
raise errors.BadIndexKey(key)
511
# find what it refers to:
512
key_dict = self._nodes_by_key
514
# find the subdict whose contents should be returned.
516
while len(elements) and elements[0] is not None:
517
key_dict = key_dict[elements[0]]
520
# a non-existant lookup.
525
key_dict = dicts.pop(-1)
526
# can't be empty or would not exist
527
item, value = key_dict.iteritems().next()
528
if type(value) == dict:
530
dicts.extend(key_dict.itervalues())
533
for value in key_dict.itervalues():
534
# each value is the key:value:node refs tuple
536
yield (self, ) + value
538
# the last thing looked up was a terminal element
539
yield (self, ) + key_dict
542
"""Return an estimate of the number of keys in this index.
544
For GraphIndex the estimate is exact.
546
if self._key_count is None:
547
# really this should just read the prefix
549
return self._key_count
551
def _lookup_keys_via_location(self, location_keys):
552
"""Public interface for implementing bisection.
554
If _buffer_all has been called, then all the data for the index is in
555
memory, and this method should not be called, as it uses a separate
556
cache because it cannot pre-resolve all indices, which buffer_all does
559
:param location_keys: A list of location(byte offset), key tuples.
560
:return: A list of (location_key, result) tuples as expected by
561
bzrlib.bisect_multi.bisect_multi_bytes.
563
# Possible improvements:
564
# - only bisect lookup each key once
565
# - sort the keys first, and use that to reduce the bisection window
567
# this progresses in three parts:
570
# attempt to answer the question from the now in memory data.
571
# build the readv request
572
# for each location, ask for 800 bytes - much more than rows we've seen
575
for location, key in location_keys:
576
# can we answer from cache?
577
# - if we know the answer - yes
578
index = self._parsed_key_index(key)
579
if (len(self._parsed_key_map) and
580
self._parsed_key_map[index][0] <= key and
581
(self._parsed_key_map[index][1] > key or
582
# end of the file has been parsed
583
self._parsed_byte_map[index][1] == self._size)):
584
# the key has been parsed, so no lookup is needed
586
# - if we have examined this part of the file already - yes
587
index = self._parsed_byte_index(location)
588
if (len(self._parsed_byte_map) and
589
self._parsed_byte_map[index][0] <= location and
590
self._parsed_byte_map[index][1] > location):
591
# the byte region has been parsed, so no read is needed.
594
if location + length > self._size:
595
length = self._size - location
596
# todo, trim out parsed locations.
598
readv_ranges.append((location, length))
599
# read the header if needed
600
if self._bisect_nodes is None:
601
readv_ranges.append((0, 200))
602
self._read_and_parse(readv_ranges)
604
# - figure out <, >, missing, present
605
# - result present references so we can return them.
607
# keys that we cannot answer until we resolve references
608
pending_references = []
609
pending_locations = set()
610
for location, key in location_keys:
611
# can we answer from cache?
612
index = self._parsed_key_index(key)
613
if (self._parsed_key_map[index][0] <= key and
614
(self._parsed_key_map[index][1] > key or
615
# end of the file has been parsed
616
self._parsed_byte_map[index][1] == self._size)):
617
# the key has been parsed, so no lookup is needed
618
if key in self._bisect_nodes:
619
if self.node_ref_lists:
620
# the references may not have been all parsed.
621
value, refs = self._bisect_nodes[key]
622
wanted_locations = []
623
for ref_list in refs:
625
if ref not in self._keys_by_offset:
626
wanted_locations.append(ref)
628
pending_locations.update(wanted_locations)
629
pending_references.append((location, key))
631
result.append(((location, key), (self, key,
632
value, self._resolve_references(refs))))
634
result.append(((location, key),
635
(self, key, self._bisect_nodes[key])))
637
result.append(((location, key), False))
639
# no, is the key above or below the probed location:
640
# get the range of the probed & parsed location
641
index = self._parsed_byte_index(location)
642
# if the key is below the start of the range, its below
643
if key < self._parsed_key_map[index][0]:
647
result.append(((location, key), direction))
649
# lookup data to resolve references
650
for location in pending_locations:
652
if location + length > self._size:
653
length = self._size - location
654
# TODO: trim out parsed locations (e.g. if the 800 is into the
655
# parsed region trim it, and dont use the adjust_for_latency
658
readv_ranges.append((location, length))
659
self._read_and_parse(readv_ranges)
660
for location, key in pending_references:
661
# answer key references we had to look-up-late.
662
index = self._parsed_key_index(key)
663
value, refs = self._bisect_nodes[key]
664
result.append(((location, key), (self, key,
665
value, self._resolve_references(refs))))
668
def _parse_header_from_bytes(self, bytes):
669
"""Parse the header from a region of bytes.
671
:param bytes: The data to parse.
672
:return: An offset, data tuple such as readv yields, for the unparsed
673
data. (which may length 0).
675
signature = bytes[0:len(self._signature())]
676
if not signature == self._signature():
677
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
678
lines = bytes[len(self._signature()):].splitlines()
679
options_line = lines[0]
680
if not options_line.startswith(_OPTION_NODE_REFS):
681
raise errors.BadIndexOptions(self)
683
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
685
raise errors.BadIndexOptions(self)
686
options_line = lines[1]
687
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
688
raise errors.BadIndexOptions(self)
690
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
692
raise errors.BadIndexOptions(self)
693
options_line = lines[2]
694
if not options_line.startswith(_OPTION_LEN):
695
raise errors.BadIndexOptions(self)
697
self._key_count = int(options_line[len(_OPTION_LEN):])
699
raise errors.BadIndexOptions(self)
700
# calculate the bytes we have processed
701
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
703
self._parsed_bytes(0, None, header_end, None)
704
# setup parsing state
705
self._expected_elements = 3 + self._key_length
706
# raw data keyed by offset
707
self._keys_by_offset = {}
708
# keys with the value and node references
709
self._bisect_nodes = {}
710
return header_end, bytes[header_end:]
712
def _parse_region(self, offset, data):
713
"""Parse node data returned from a readv operation.
715
:param offset: The byte offset the data starts at.
716
:param data: The data to parse.
720
end = offset + len(data)
723
# Trivial test - if the current index's end is within the
724
# low-matching parsed range, we're done.
725
index = self._parsed_byte_index(high_parsed)
726
if end < self._parsed_byte_map[index][1]:
728
# print "[%d:%d]" % (offset, end), \
729
# self._parsed_byte_map[index:index + 2]
730
high_parsed, last_segment = self._parse_segment(
731
offset, data, end, index)
735
def _parse_segment(self, offset, data, end, index):
736
"""Parse one segment of data.
738
:param offset: Where 'data' begins in the file.
739
:param data: Some data to parse a segment of.
740
:param end: Where data ends
741
:param index: The current index into the parsed bytes map.
742
:return: True if the parsed segment is the last possible one in the
744
:return: high_parsed_byte, last_segment.
745
high_parsed_byte is the location of the highest parsed byte in this
746
segment, last_segment is True if the parsed segment is the last
747
possible one in the data block.
749
# default is to use all data
751
# accomodate overlap with data before this.
752
if offset < self._parsed_byte_map[index][1]:
753
# overlaps the lower parsed region
754
# skip the parsed data
755
trim_start = self._parsed_byte_map[index][1] - offset
756
# don't trim the start for \n
757
start_adjacent = True
758
elif offset == self._parsed_byte_map[index][1]:
759
# abuts the lower parsed region
762
# do not trim anything
763
start_adjacent = True
765
# does not overlap the lower parsed region
768
# but trim the leading \n
769
start_adjacent = False
770
if end == self._size:
771
# lines up to the end of all data:
774
# do not strip to the last \n
777
elif index + 1 == len(self._parsed_byte_map):
778
# at the end of the parsed data
781
# but strip to the last \n
784
elif end == self._parsed_byte_map[index + 1][0]:
785
# buts up against the next parsed region
788
# do not strip to the last \n
791
elif end > self._parsed_byte_map[index + 1][0]:
792
# overlaps into the next parsed region
793
# only consider the unparsed data
794
trim_end = self._parsed_byte_map[index + 1][0] - offset
795
# do not strip to the last \n as we know its an entire record
797
last_segment = end < self._parsed_byte_map[index + 1][1]
799
# does not overlap into the next region
802
# but strip to the last \n
805
# now find bytes to discard if needed
806
if not start_adjacent:
807
# work around python bug in rfind
808
if trim_start is None:
809
trim_start = data.find('\n') + 1
811
trim_start = data.find('\n', trim_start) + 1
812
assert trim_start != 0, 'no \n was present'
813
# print 'removing start', offset, trim_start, repr(data[:trim_start])
815
# work around python bug in rfind
817
trim_end = data.rfind('\n') + 1
819
trim_end = data.rfind('\n', None, trim_end) + 1
820
assert trim_end != 0, 'no \n was present'
821
# print 'removing end', offset, trim_end, repr(data[trim_end:])
822
# adjust offset and data to the parseable data.
823
trimmed_data = data[trim_start:trim_end]
824
assert trimmed_data, 'read unneeded data [%d:%d] from [%d:%d]' % (
825
trim_start, trim_end, offset, offset + len(data))
828
# print "parsing", repr(trimmed_data)
829
# splitlines mangles the \r delimiters.. don't use it.
830
lines = trimmed_data.split('\n')
833
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
834
for key, value in nodes:
835
self._bisect_nodes[key] = value
836
self._parsed_bytes(offset, first_key,
837
offset + len(trimmed_data), last_key)
838
return offset + len(trimmed_data), last_segment
840
def _parse_lines(self, lines, pos):
849
assert self._size == pos + 1, "%s %s" % (self._size, pos)
852
elements = line.split('\0')
853
if len(elements) != self._expected_elements:
854
raise errors.BadIndexData(self)
856
key = tuple(elements[:self._key_length])
857
if first_key is None:
859
absent, references, value = elements[-3:]
861
for ref_string in references.split('\t'):
862
ref_lists.append(tuple([
863
int(ref) for ref in ref_string.split('\r') if ref
865
ref_lists = tuple(ref_lists)
866
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
867
pos += len(line) + 1 # +1 for the \n
870
if self.node_ref_lists:
871
node_value = (value, ref_lists)
874
nodes.append((key, node_value))
875
# print "parsed ", key
876
return first_key, key, nodes, trailers
878
def _parsed_bytes(self, start, start_key, end, end_key):
879
"""Mark the bytes from start to end as parsed.
881
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
884
:param start: The start of the parsed region.
885
:param end: The end of the parsed region.
887
index = self._parsed_byte_index(start)
888
new_value = (start, end)
889
new_key = (start_key, end_key)
891
# first range parsed is always the beginning.
892
self._parsed_byte_map.insert(index, new_value)
893
self._parsed_key_map.insert(index, new_key)
897
# extend lower region
898
# extend higher region
899
# combine two regions
900
if (index + 1 < len(self._parsed_byte_map) and
901
self._parsed_byte_map[index][1] == start and
902
self._parsed_byte_map[index + 1][0] == end):
903
# combine two regions
904
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
905
self._parsed_byte_map[index + 1][1])
906
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
907
self._parsed_key_map[index + 1][1])
908
del self._parsed_byte_map[index + 1]
909
del self._parsed_key_map[index + 1]
910
elif self._parsed_byte_map[index][1] == start:
911
# extend the lower entry
912
self._parsed_byte_map[index] = (
913
self._parsed_byte_map[index][0], end)
914
self._parsed_key_map[index] = (
915
self._parsed_key_map[index][0], end_key)
916
elif (index + 1 < len(self._parsed_byte_map) and
917
self._parsed_byte_map[index + 1][0] == end):
918
# extend the higher entry
919
self._parsed_byte_map[index + 1] = (
920
start, self._parsed_byte_map[index + 1][1])
921
self._parsed_key_map[index + 1] = (
922
start_key, self._parsed_key_map[index + 1][1])
925
self._parsed_byte_map.insert(index + 1, new_value)
926
self._parsed_key_map.insert(index + 1, new_key)
928
def _read_and_parse(self, readv_ranges):
929
"""Read the the ranges and parse the resulting data.
931
:param readv_ranges: A prepared readv range list.
934
readv_data = self._transport.readv(self._name, readv_ranges, True,
937
for offset, data in readv_data:
938
if self._bisect_nodes is None:
939
# this must be the start
941
offset, data = self._parse_header_from_bytes(data)
942
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
943
self._parse_region(offset, data)
945
def _signature(self):
946
"""The file signature for this index type."""
950
"""Validate that everything in the index can be accessed."""
951
# iter_all validates completely at the moment, so just do that.
952
for node in self.iter_all_entries():
956
class CombinedGraphIndex(object):
957
"""A GraphIndex made up from smaller GraphIndices.
959
The backing indices must implement GraphIndex, and are presumed to be
962
Queries against the combined index will be made against the first index,
963
and then the second and so on. The order of index's can thus influence
964
performance significantly. For example, if one index is on local disk and a
965
second on a remote server, the local disk index should be before the other
969
def __init__(self, indices):
970
"""Create a CombinedGraphIndex backed by indices.
972
:param indices: An ordered list of indices to query for data.
974
self._indices = indices
978
self.__class__.__name__,
979
', '.join(map(repr, self._indices)))
981
def insert_index(self, pos, index):
982
"""Insert a new index in the list of indices to query.
984
:param pos: The position to insert the index.
985
:param index: The index to insert.
987
self._indices.insert(pos, index)
989
def iter_all_entries(self):
990
"""Iterate over all keys within the index
992
Duplicate keys across child indices are presumed to have the same
993
value and are only reported once.
995
:return: An iterable of (index, key, reference_lists, value).
996
There is no defined order for the result iteration - it will be in
997
the most efficient order for the index.
1000
for index in self._indices:
1001
for node in index.iter_all_entries():
1002
if node[1] not in seen_keys:
1004
seen_keys.add(node[1])
1006
def iter_entries(self, keys):
1007
"""Iterate over keys within the index.
1009
Duplicate keys across child indices are presumed to have the same
1010
value and are only reported once.
1012
:param keys: An iterable providing the keys to be retrieved.
1013
:return: An iterable of (index, key, reference_lists, value). There is no
1014
defined order for the result iteration - it will be in the most
1015
efficient order for the index.
1018
for index in self._indices:
1021
for node in index.iter_entries(keys):
1022
keys.remove(node[1])
1025
def iter_entries_prefix(self, keys):
1026
"""Iterate over keys within the index using prefix matching.
1028
Duplicate keys across child indices are presumed to have the same
1029
value and are only reported once.
1031
Prefix matching is applied within the tuple of a key, not to within
1032
the bytestring of each key element. e.g. if you have the keys ('foo',
1033
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1034
only the former key is returned.
1036
:param keys: An iterable providing the key prefixes to be retrieved.
1037
Each key prefix takes the form of a tuple the length of a key, but
1038
with the last N elements 'None' rather than a regular bytestring.
1039
The first element cannot be 'None'.
1040
:return: An iterable as per iter_all_entries, but restricted to the
1041
keys with a matching prefix to those supplied. No additional keys
1042
will be returned, and every match that is in the index will be
1049
for index in self._indices:
1050
for node in index.iter_entries_prefix(keys):
1051
if node[1] in seen_keys:
1053
seen_keys.add(node[1])
1056
def key_count(self):
1057
"""Return an estimate of the number of keys in this index.
1059
For CombinedGraphIndex this is approximated by the sum of the keys of
1060
the child indices. As child indices may have duplicate keys this can
1061
have a maximum error of the number of child indices * largest number of
1064
return sum((index.key_count() for index in self._indices), 0)
1067
"""Validate that everything in the index can be accessed."""
1068
for index in self._indices:
1072
class InMemoryGraphIndex(GraphIndexBuilder):
1073
"""A GraphIndex which operates entirely out of memory and is mutable.
1075
This is designed to allow the accumulation of GraphIndex entries during a
1076
single write operation, where the accumulated entries need to be immediately
1077
available - for example via a CombinedGraphIndex.
1080
def add_nodes(self, nodes):
1081
"""Add nodes to the index.
1083
:param nodes: An iterable of (key, node_refs, value) entries to add.
1085
if self.reference_lists:
1086
for (key, value, node_refs) in nodes:
1087
self.add_node(key, value, node_refs)
1089
for (key, value) in nodes:
1090
self.add_node(key, value)
1092
def iter_all_entries(self):
1093
"""Iterate over all keys within the index
1095
:return: An iterable of (index, key, reference_lists, value). There is no
1096
defined order for the result iteration - it will be in the most
1097
efficient order for the index (in this case dictionary hash order).
1099
if 'evil' in debug.debug_flags:
1100
trace.mutter_callsite(3,
1101
"iter_all_entries scales with size of history.")
1102
if self.reference_lists:
1103
for key, (absent, references, value) in self._nodes.iteritems():
1105
yield self, key, value, references
1107
for key, (absent, references, value) in self._nodes.iteritems():
1109
yield self, key, value
1111
def iter_entries(self, keys):
1112
"""Iterate over keys within the index.
1114
:param keys: An iterable providing the keys to be retrieved.
1115
:return: An iterable of (index, key, reference_lists, value). There is no
1116
defined order for the result iteration - it will be in the most
1117
efficient order for the index (keys iteration order in this case).
1120
if self.reference_lists:
1121
for key in keys.intersection(self._keys):
1122
node = self._nodes[key]
1124
yield self, key, node[2], node[1]
1126
for key in keys.intersection(self._keys):
1127
node = self._nodes[key]
1129
yield self, key, node[2]
1131
def iter_entries_prefix(self, keys):
1132
"""Iterate over keys within the index using prefix matching.
1134
Prefix matching is applied within the tuple of a key, not to within
1135
the bytestring of each key element. e.g. if you have the keys ('foo',
1136
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1137
only the former key is returned.
1139
:param keys: An iterable providing the key prefixes to be retrieved.
1140
Each key prefix takes the form of a tuple the length of a key, but
1141
with the last N elements 'None' rather than a regular bytestring.
1142
The first element cannot be 'None'.
1143
:return: An iterable as per iter_all_entries, but restricted to the
1144
keys with a matching prefix to those supplied. No additional keys
1145
will be returned, and every match that is in the index will be
1148
# XXX: To much duplication with the GraphIndex class; consider finding
1149
# a good place to pull out the actual common logic.
1153
if self._key_length == 1:
1157
raise errors.BadIndexKey(key)
1158
if len(key) != self._key_length:
1159
raise errors.BadIndexKey(key)
1160
node = self._nodes[key]
1163
if self.reference_lists:
1164
yield self, key, node[2], node[1]
1166
yield self, key, node[2]
1171
raise errors.BadIndexKey(key)
1172
if len(key) != self._key_length:
1173
raise errors.BadIndexKey(key)
1174
# find what it refers to:
1175
key_dict = self._nodes_by_key
1176
elements = list(key)
1177
# find the subdict to return
1179
while len(elements) and elements[0] is not None:
1180
key_dict = key_dict[elements[0]]
1183
# a non-existant lookup.
1188
key_dict = dicts.pop(-1)
1189
# can't be empty or would not exist
1190
item, value = key_dict.iteritems().next()
1191
if type(value) == dict:
1193
dicts.extend(key_dict.itervalues())
1196
for value in key_dict.itervalues():
1197
yield (self, ) + value
1199
yield (self, ) + key_dict
1201
def key_count(self):
1202
"""Return an estimate of the number of keys in this index.
1204
For InMemoryGraphIndex the estimate is exact.
1206
return len(self._keys)
1209
"""In memory index's have no known corruption at the moment."""
1212
class GraphIndexPrefixAdapter(object):
1213
"""An adapter between GraphIndex with different key lengths.
1215
Queries against this will emit queries against the adapted Graph with the
1216
prefix added, queries for all items use iter_entries_prefix. The returned
1217
nodes will have their keys and node references adjusted to remove the
1218
prefix. Finally, an add_nodes_callback can be supplied - when called the
1219
nodes and references being added will have prefix prepended.
1222
def __init__(self, adapted, prefix, missing_key_length,
1223
add_nodes_callback=None):
1224
"""Construct an adapter against adapted with prefix."""
1225
self.adapted = adapted
1226
self.prefix_key = prefix + (None,)*missing_key_length
1227
self.prefix = prefix
1228
self.prefix_len = len(prefix)
1229
self.add_nodes_callback = add_nodes_callback
1231
def add_nodes(self, nodes):
1232
"""Add nodes to the index.
1234
:param nodes: An iterable of (key, node_refs, value) entries to add.
1236
# save nodes in case its an iterator
1237
nodes = tuple(nodes)
1238
translated_nodes = []
1240
# Add prefix_key to each reference node_refs is a tuple of tuples,
1241
# so split it apart, and add prefix_key to the internal reference
1242
for (key, value, node_refs) in nodes:
1243
adjusted_references = (
1244
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1245
for ref_list in node_refs))
1246
translated_nodes.append((self.prefix + key, value,
1247
adjusted_references))
1249
# XXX: TODO add an explicit interface for getting the reference list
1250
# status, to handle this bit of user-friendliness in the API more
1252
for (key, value) in nodes:
1253
translated_nodes.append((self.prefix + key, value))
1254
self.add_nodes_callback(translated_nodes)
1256
def add_node(self, key, value, references=()):
1257
"""Add a node to the index.
1259
:param key: The key. keys are non-empty tuples containing
1260
as many whitespace-free utf8 bytestrings as the key length
1261
defined for this index.
1262
:param references: An iterable of iterables of keys. Each is a
1263
reference to another key.
1264
:param value: The value to associate with the key. It may be any
1265
bytes as long as it does not contain \0 or \n.
1267
self.add_nodes(((key, value, references), ))
1269
def _strip_prefix(self, an_iter):
1270
"""Strip prefix data from nodes and return it."""
1271
for node in an_iter:
1273
if node[1][:self.prefix_len] != self.prefix:
1274
raise errors.BadIndexData(self)
1275
for ref_list in node[3]:
1276
for ref_node in ref_list:
1277
if ref_node[:self.prefix_len] != self.prefix:
1278
raise errors.BadIndexData(self)
1279
yield node[0], node[1][self.prefix_len:], node[2], (
1280
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1281
for ref_list in node[3]))
1283
def iter_all_entries(self):
1284
"""Iterate over all keys within the index
1286
iter_all_entries is implemented against the adapted index using
1287
iter_entries_prefix.
1289
:return: An iterable of (index, key, reference_lists, value). There is no
1290
defined order for the result iteration - it will be in the most
1291
efficient order for the index (in this case dictionary hash order).
1293
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1295
def iter_entries(self, keys):
1296
"""Iterate over keys within the index.
1298
:param keys: An iterable providing the keys to be retrieved.
1299
:return: An iterable of (key, reference_lists, value). There is no
1300
defined order for the result iteration - it will be in the most
1301
efficient order for the index (keys iteration order in this case).
1303
return self._strip_prefix(self.adapted.iter_entries(
1304
self.prefix + key for key in keys))
1306
def iter_entries_prefix(self, keys):
1307
"""Iterate over keys within the index using prefix matching.
1309
Prefix matching is applied within the tuple of a key, not to within
1310
the bytestring of each key element. e.g. if you have the keys ('foo',
1311
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1312
only the former key is returned.
1314
:param keys: An iterable providing the key prefixes to be retrieved.
1315
Each key prefix takes the form of a tuple the length of a key, but
1316
with the last N elements 'None' rather than a regular bytestring.
1317
The first element cannot be 'None'.
1318
:return: An iterable as per iter_all_entries, but restricted to the
1319
keys with a matching prefix to those supplied. No additional keys
1320
will be returned, and every match that is in the index will be
1323
return self._strip_prefix(self.adapted.iter_entries_prefix(
1324
self.prefix + key for key in keys))
1326
def key_count(self):
1327
"""Return an estimate of the number of keys in this index.
1329
For GraphIndexPrefixAdapter this is relatively expensive - key
1330
iteration with the prefix is done.
1332
return len(list(self.iter_all_entries()))
1335
"""Call the adapted's validate."""
1336
self.adapted.validate()
added
removed
bzrlib/index.py
