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.revision import NULL_REVISION
36
from bzrlib.trace import mutter
44
_HEADER_READV = (0, 200)
45
_OPTION_KEY_ELEMENTS = "key_elements="
47
_OPTION_NODE_REFS = "node_ref_lists="
48
_SIGNATURE = "Bazaar Graph Index 1\n"
51
_whitespace_re = re.compile('[\t\n\x0b\x0c\r\x00 ]')
52
_newline_null_re = re.compile('[\n\0]')
55
class GraphIndexBuilder(object):
56
"""A builder that can build a GraphIndex.
58
The resulting graph has the structure:
60
_SIGNATURE OPTIONS NODES NEWLINE
61
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
62
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
64
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
65
KEY := Not-whitespace-utf8
67
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
68
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
69
REFERENCE := DIGITS ; digits is the byte offset in the index of the
71
VALUE := no-newline-no-null-bytes
74
def __init__(self, reference_lists=0, key_elements=1):
75
"""Create a GraphIndex builder.
77
:param reference_lists: The number of node references lists for each
79
:param key_elements: The number of bytestrings in each key.
81
self.reference_lists = reference_lists
84
self._nodes_by_key = {}
85
self._key_length = key_elements
87
def _check_key(self, key):
88
"""Raise BadIndexKey if key is not a valid key for this index."""
89
if type(key) != tuple:
90
raise errors.BadIndexKey(key)
91
if self._key_length != len(key):
92
raise errors.BadIndexKey(key)
94
if not element or _whitespace_re.search(element) is not None:
95
raise errors.BadIndexKey(element)
97
def add_node(self, key, value, references=()):
98
"""Add a node to the index.
100
:param key: The key. keys are non-empty tuples containing
101
as many whitespace-free utf8 bytestrings as the key length
102
defined for this index.
103
:param references: An iterable of iterables of keys. Each is a
104
reference to another key.
105
:param value: The value to associate with the key. It may be any
106
bytes as long as it does not contain \0 or \n.
109
if _newline_null_re.search(value) is not None:
110
raise errors.BadIndexValue(value)
111
if len(references) != self.reference_lists:
112
raise errors.BadIndexValue(references)
114
for reference_list in references:
115
for reference in reference_list:
116
self._check_key(reference)
117
if reference not in self._nodes:
118
self._nodes[reference] = ('a', (), '')
119
node_refs.append(tuple(reference_list))
120
if key in self._nodes and self._nodes[key][0] == '':
121
raise errors.BadIndexDuplicateKey(key, self)
122
self._nodes[key] = ('', tuple(node_refs), value)
124
if self._key_length > 1:
125
key_dict = self._nodes_by_key
126
if self.reference_lists:
127
key_value = key, value, tuple(node_refs)
129
key_value = key, value
130
# possibly should do this on-demand, but it seems likely it is
132
# For a key of (foo, bar, baz) create
133
# _nodes_by_key[foo][bar][baz] = key_value
134
for subkey in key[:-1]:
135
key_dict = key_dict.setdefault(subkey, {})
136
key_dict[key[-1]] = key_value
140
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
141
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
142
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
143
prefix_length = sum(len(x) for x in lines)
144
# references are byte offsets. To avoid having to do nasty
145
# polynomial work to resolve offsets (references to later in the
146
# file cannot be determined until all the inbetween references have
147
# been calculated too) we pad the offsets with 0's to make them be
148
# of consistent length. Using binary offsets would break the trivial
150
# to calculate the width of zero's needed we do three passes:
151
# one to gather all the non-reference data and the number of references.
152
# one to pad all the data with reference-length and determine entry
156
# forward sorted by key. In future we may consider topological sorting,
157
# at the cost of table scans for direct lookup, or a second index for
159
nodes = sorted(self._nodes.items())
160
# if we do not prepass, we don't know how long it will be up front.
161
expected_bytes = None
162
# we only need to pre-pass if we have reference lists at all.
163
if self.reference_lists:
165
non_ref_bytes = prefix_length
167
# TODO use simple multiplication for the constants in this loop.
168
for key, (absent, references, value) in nodes:
169
# record the offset known *so far* for this key:
170
# the non reference bytes to date, and the total references to
171
# date - saves reaccumulating on the second pass
172
key_offset_info.append((key, non_ref_bytes, total_references))
173
# key is literal, value is literal, there are 3 null's, 1 NL
174
# key is variable length tuple, \x00 between elements
175
non_ref_bytes += sum(len(element) for element in key)
176
if self._key_length > 1:
177
non_ref_bytes += self._key_length - 1
178
# value is literal bytes, there are 3 null's, 1 NL.
179
non_ref_bytes += len(value) + 3 + 1
180
# one byte for absent if set.
183
elif self.reference_lists:
184
# (ref_lists -1) tabs
185
non_ref_bytes += self.reference_lists - 1
186
# (ref-1 cr's per ref_list)
187
for ref_list in references:
188
# how many references across the whole file?
189
total_references += len(ref_list)
190
# accrue reference separators
192
non_ref_bytes += len(ref_list) - 1
193
# how many digits are needed to represent the total byte count?
195
possible_total_bytes = non_ref_bytes + total_references*digits
196
while 10 ** digits < possible_total_bytes:
198
possible_total_bytes = non_ref_bytes + total_references*digits
199
expected_bytes = possible_total_bytes + 1 # terminating newline
200
# resolve key addresses.
202
for key, non_ref_bytes, total_references in key_offset_info:
203
key_addresses[key] = non_ref_bytes + total_references*digits
205
format_string = '%%0%sd' % digits
206
for key, (absent, references, value) in nodes:
207
flattened_references = []
208
for ref_list in references:
210
for reference in ref_list:
211
ref_addresses.append(format_string % key_addresses[reference])
212
flattened_references.append('\r'.join(ref_addresses))
213
string_key = '\x00'.join(key)
214
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
215
'\t'.join(flattened_references), value))
217
result = StringIO(''.join(lines))
218
if expected_bytes and len(result.getvalue()) != expected_bytes:
219
raise errors.BzrError('Failed index creation. Internal error:'
220
' mismatched output length and expected length: %d %d' %
221
(len(result.getvalue()), expected_bytes))
225
class GraphIndex(object):
226
"""An index for data with embedded graphs.
228
The index maps keys to a list of key reference lists, and a value.
229
Each node has the same number of key reference lists. Each key reference
230
list can be empty or an arbitrary length. The value is an opaque NULL
231
terminated string without any newlines. The storage of the index is
232
hidden in the interface: keys and key references are always tuples of
233
bytestrings, never the internal representation (e.g. dictionary offsets).
235
It is presumed that the index will not be mutated - it is static data.
237
Successive iter_all_entries calls will read the entire index each time.
238
Additionally, iter_entries calls will read the index linearly until the
239
desired keys are found. XXX: This must be fixed before the index is
240
suitable for production use. :XXX
243
def __init__(self, transport, name, size):
244
"""Open an index called name on transport.
246
:param transport: A bzrlib.transport.Transport.
247
:param name: A path to provide to transport API calls.
248
:param size: The size of the index in bytes. This is used for bisection
249
logic to perform partial index reads. While the size could be
250
obtained by statting the file this introduced an additional round
251
trip as well as requiring stat'able transports, both of which are
252
avoided by having it supplied. If size is None, then bisection
253
support will be disabled and accessing the index will just stream
256
self._transport = transport
258
# Becomes a dict of key:(value, reference-list-byte-locations) used by
259
# the bisection interface to store parsed but not resolved keys.
260
self._bisect_nodes = None
261
# Becomes a dict of key:(value, reference-list-keys) which are ready to
262
# be returned directly to callers.
264
# a sorted list of slice-addresses for the parsed bytes of the file.
265
# e.g. (0,1) would mean that byte 0 is parsed.
266
self._parsed_byte_map = []
267
# a sorted list of keys matching each slice address for parsed bytes
268
# e.g. (None, 'foo@bar') would mean that the first byte contained no
269
# key, and the end byte of the slice is the of the data for 'foo@bar'
270
self._parsed_key_map = []
271
self._key_count = None
272
self._keys_by_offset = None
273
self._nodes_by_key = None
276
def __eq__(self, other):
277
"""Equal when self and other were created with the same parameters."""
279
type(self) == type(other) and
280
self._transport == other._transport and
281
self._name == other._name and
282
self._size == other._size)
284
def __ne__(self, other):
285
return not self.__eq__(other)
288
return "%s(%r)" % (self.__class__.__name__,
289
self._transport.abspath(self._name))
291
def _buffer_all(self):
292
"""Buffer all the index data.
294
Mutates self._nodes and self.keys_by_offset.
296
if 'index' in debug.debug_flags:
297
mutter('Reading entire index %s', self._transport.abspath(self._name))
298
stream = self._transport.get(self._name)
299
self._read_prefix(stream)
300
self._expected_elements = 3 + self._key_length
302
# raw data keyed by offset
303
self._keys_by_offset = {}
304
# ready-to-return key:value or key:value, node_ref_lists
306
self._nodes_by_key = {}
309
lines = stream.read().split('\n')
311
_, _, _, trailers = self._parse_lines(lines, pos)
312
for key, absent, references, value in self._keys_by_offset.itervalues():
315
# resolve references:
316
if self.node_ref_lists:
317
node_value = (value, self._resolve_references(references))
320
self._nodes[key] = node_value
321
if self._key_length > 1:
322
subkey = list(reversed(key[:-1]))
323
key_dict = self._nodes_by_key
324
if self.node_ref_lists:
325
key_value = key, node_value[0], node_value[1]
327
key_value = key, node_value
328
# possibly should do this on-demand, but it seems likely it is
330
# For a key of (foo, bar, baz) create
331
# _nodes_by_key[foo][bar][baz] = key_value
332
for subkey in key[:-1]:
333
key_dict = key_dict.setdefault(subkey, {})
334
key_dict[key[-1]] = key_value
335
# cache the keys for quick set intersections
336
self._keys = set(self._nodes)
338
# there must be one line - the empty trailer line.
339
raise errors.BadIndexData(self)
341
def iter_all_entries(self):
342
"""Iterate over all keys within the index.
344
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
345
The former tuple is used when there are no reference lists in the
346
index, making the API compatible with simple key:value index types.
347
There is no defined order for the result iteration - it will be in
348
the most efficient order for the index.
350
if 'evil' in debug.debug_flags:
351
trace.mutter_callsite(3,
352
"iter_all_entries scales with size of history.")
353
if self._nodes is None:
355
if self.node_ref_lists:
356
for key, (value, node_ref_lists) in self._nodes.iteritems():
357
yield self, key, value, node_ref_lists
359
for key, value in self._nodes.iteritems():
360
yield self, key, value
362
def _read_prefix(self, stream):
363
signature = stream.read(len(self._signature()))
364
if not signature == self._signature():
365
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
366
options_line = stream.readline()
367
if not options_line.startswith(_OPTION_NODE_REFS):
368
raise errors.BadIndexOptions(self)
370
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
372
raise errors.BadIndexOptions(self)
373
options_line = stream.readline()
374
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
375
raise errors.BadIndexOptions(self)
377
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
379
raise errors.BadIndexOptions(self)
380
options_line = stream.readline()
381
if not options_line.startswith(_OPTION_LEN):
382
raise errors.BadIndexOptions(self)
384
self._key_count = int(options_line[len(_OPTION_LEN):-1])
386
raise errors.BadIndexOptions(self)
388
def _resolve_references(self, references):
389
"""Return the resolved key references for references.
391
References are resolved by looking up the location of the key in the
392
_keys_by_offset map and substituting the key name, preserving ordering.
394
:param references: An iterable of iterables of key locations. e.g.
396
:return: A tuple of tuples of keys.
399
for ref_list in references:
400
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
401
return tuple(node_refs)
403
def _find_index(self, range_map, key):
404
"""Helper for the _parsed_*_index calls.
406
Given a range map - [(start, end), ...], finds the index of the range
407
in the map for key if it is in the map, and if it is not there, the
408
immediately preceeding range in the map.
410
result = bisect_right(range_map, key) - 1
411
if result + 1 < len(range_map):
412
# check the border condition, it may be in result + 1
413
if range_map[result + 1][0] == key[0]:
417
def _parsed_byte_index(self, offset):
418
"""Return the index of the entry immediately before offset.
420
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
421
there is one unparsed byte (the 11th, addressed as[10]). then:
422
asking for 0 will return 0
423
asking for 10 will return 0
424
asking for 11 will return 1
425
asking for 12 will return 1
428
return self._find_index(self._parsed_byte_map, key)
430
def _parsed_key_index(self, key):
431
"""Return the index of the entry immediately before key.
433
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
434
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
435
have been parsed, then:
436
asking for '' will return 0
437
asking for 'a' will return 0
438
asking for 'b' will return 1
439
asking for 'e' will return 1
441
search_key = (key, None)
442
return self._find_index(self._parsed_key_map, search_key)
444
def _is_parsed(self, offset):
445
"""Returns True if offset has been parsed."""
446
index = self._parsed_byte_index(offset)
447
if index == len(self._parsed_byte_map):
448
return offset < self._parsed_byte_map[index - 1][1]
449
start, end = self._parsed_byte_map[index]
450
return offset >= start and offset < end
452
def _iter_entries_from_total_buffer(self, keys):
453
"""Iterate over keys when the entire index is parsed."""
454
keys = keys.intersection(self._keys)
455
if self.node_ref_lists:
457
value, node_refs = self._nodes[key]
458
yield self, key, value, node_refs
461
yield self, key, self._nodes[key]
463
def iter_entries(self, keys):
464
"""Iterate over keys within the index.
466
:param keys: An iterable providing the keys to be retrieved.
467
:return: An iterable as per iter_all_entries, but restricted to the
468
keys supplied. No additional keys will be returned, and every
469
key supplied that is in the index will be returned.
474
if self._size is None and self._nodes is None:
476
# We fit about 20 keys per minimum-read (4K), so if we are looking for
477
# more than 1/20th of the index its likely (assuming homogenous key
478
# spread) that we'll read the entire index. If we're going to do that,
479
# buffer the whole thing. A better analysis might take key spread into
480
# account - but B+Tree indices are better anyway.
481
# We could look at all data read, and use a threshold there, which will
482
# trigger on ancestry walks, but that is not yet fully mapped out.
483
if self._nodes is None and len(keys) * 20 > self.key_count():
485
if self._nodes is not None:
486
return self._iter_entries_from_total_buffer(keys)
488
return (result[1] for result in bisect_multi_bytes(
489
self._lookup_keys_via_location, self._size, keys))
491
def iter_entries_prefix(self, keys):
492
"""Iterate over keys within the index using prefix matching.
494
Prefix matching is applied within the tuple of a key, not to within
495
the bytestring of each key element. e.g. if you have the keys ('foo',
496
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
497
only the former key is returned.
499
WARNING: Note that this method currently causes a full index parse
500
unconditionally (which is reasonably appropriate as it is a means for
501
thunking many small indices into one larger one and still supplies
502
iter_all_entries at the thunk layer).
504
:param keys: An iterable providing the key prefixes to be retrieved.
505
Each key prefix takes the form of a tuple the length of a key, but
506
with the last N elements 'None' rather than a regular bytestring.
507
The first element cannot be 'None'.
508
:return: An iterable as per iter_all_entries, but restricted to the
509
keys with a matching prefix to those supplied. No additional keys
510
will be returned, and every match that is in the index will be
516
# load data - also finds key lengths
517
if self._nodes is None:
519
if self._key_length == 1:
523
raise errors.BadIndexKey(key)
524
if len(key) != self._key_length:
525
raise errors.BadIndexKey(key)
526
if self.node_ref_lists:
527
value, node_refs = self._nodes[key]
528
yield self, key, value, node_refs
530
yield self, key, self._nodes[key]
535
raise errors.BadIndexKey(key)
536
if len(key) != self._key_length:
537
raise errors.BadIndexKey(key)
538
# find what it refers to:
539
key_dict = self._nodes_by_key
541
# find the subdict whose contents should be returned.
543
while len(elements) and elements[0] is not None:
544
key_dict = key_dict[elements[0]]
547
# a non-existant lookup.
552
key_dict = dicts.pop(-1)
553
# can't be empty or would not exist
554
item, value = key_dict.iteritems().next()
555
if type(value) == dict:
557
dicts.extend(key_dict.itervalues())
560
for value in key_dict.itervalues():
561
# each value is the key:value:node refs tuple
563
yield (self, ) + value
565
# the last thing looked up was a terminal element
566
yield (self, ) + key_dict
569
"""Return an estimate of the number of keys in this index.
571
For GraphIndex the estimate is exact.
573
if self._key_count is None:
574
self._read_and_parse([_HEADER_READV])
575
return self._key_count
577
def _lookup_keys_via_location(self, location_keys):
578
"""Public interface for implementing bisection.
580
If _buffer_all has been called, then all the data for the index is in
581
memory, and this method should not be called, as it uses a separate
582
cache because it cannot pre-resolve all indices, which buffer_all does
585
:param location_keys: A list of location(byte offset), key tuples.
586
:return: A list of (location_key, result) tuples as expected by
587
bzrlib.bisect_multi.bisect_multi_bytes.
589
# Possible improvements:
590
# - only bisect lookup each key once
591
# - sort the keys first, and use that to reduce the bisection window
593
# this progresses in three parts:
596
# attempt to answer the question from the now in memory data.
597
# build the readv request
598
# for each location, ask for 800 bytes - much more than rows we've seen
601
for location, key in location_keys:
602
# can we answer from cache?
603
if self._bisect_nodes and key in self._bisect_nodes:
604
# We have the key parsed.
606
index = self._parsed_key_index(key)
607
if (len(self._parsed_key_map) and
608
self._parsed_key_map[index][0] <= key and
609
(self._parsed_key_map[index][1] >= key or
610
# end of the file has been parsed
611
self._parsed_byte_map[index][1] == self._size)):
612
# the key has been parsed, so no lookup is needed even if its
615
# - if we have examined this part of the file already - yes
616
index = self._parsed_byte_index(location)
617
if (len(self._parsed_byte_map) and
618
self._parsed_byte_map[index][0] <= location and
619
self._parsed_byte_map[index][1] > location):
620
# the byte region has been parsed, so no read is needed.
623
if location + length > self._size:
624
length = self._size - location
625
# todo, trim out parsed locations.
627
readv_ranges.append((location, length))
628
# read the header if needed
629
if self._bisect_nodes is None:
630
readv_ranges.append(_HEADER_READV)
631
self._read_and_parse(readv_ranges)
633
# - figure out <, >, missing, present
634
# - result present references so we can return them.
636
# keys that we cannot answer until we resolve references
637
pending_references = []
638
pending_locations = set()
639
for location, key in location_keys:
640
# can we answer from cache?
641
if key in self._bisect_nodes:
642
# the key has been parsed, so no lookup is needed
643
if self.node_ref_lists:
644
# the references may not have been all parsed.
645
value, refs = self._bisect_nodes[key]
646
wanted_locations = []
647
for ref_list in refs:
649
if ref not in self._keys_by_offset:
650
wanted_locations.append(ref)
652
pending_locations.update(wanted_locations)
653
pending_references.append((location, key))
655
result.append(((location, key), (self, key,
656
value, self._resolve_references(refs))))
658
result.append(((location, key),
659
(self, key, self._bisect_nodes[key])))
662
# has the region the key should be in, been parsed?
663
index = self._parsed_key_index(key)
664
if (self._parsed_key_map[index][0] <= key and
665
(self._parsed_key_map[index][1] >= key or
666
# end of the file has been parsed
667
self._parsed_byte_map[index][1] == self._size)):
668
result.append(((location, key), False))
670
# no, is the key above or below the probed location:
671
# get the range of the probed & parsed location
672
index = self._parsed_byte_index(location)
673
# if the key is below the start of the range, its below
674
if key < self._parsed_key_map[index][0]:
678
result.append(((location, key), direction))
680
# lookup data to resolve references
681
for location in pending_locations:
683
if location + length > self._size:
684
length = self._size - location
685
# TODO: trim out parsed locations (e.g. if the 800 is into the
686
# parsed region trim it, and dont use the adjust_for_latency
689
readv_ranges.append((location, length))
690
self._read_and_parse(readv_ranges)
691
for location, key in pending_references:
692
# answer key references we had to look-up-late.
693
index = self._parsed_key_index(key)
694
value, refs = self._bisect_nodes[key]
695
result.append(((location, key), (self, key,
696
value, self._resolve_references(refs))))
699
def _parse_header_from_bytes(self, bytes):
700
"""Parse the header from a region of bytes.
702
:param bytes: The data to parse.
703
:return: An offset, data tuple such as readv yields, for the unparsed
704
data. (which may length 0).
706
signature = bytes[0:len(self._signature())]
707
if not signature == self._signature():
708
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
709
lines = bytes[len(self._signature()):].splitlines()
710
options_line = lines[0]
711
if not options_line.startswith(_OPTION_NODE_REFS):
712
raise errors.BadIndexOptions(self)
714
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
716
raise errors.BadIndexOptions(self)
717
options_line = lines[1]
718
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
719
raise errors.BadIndexOptions(self)
721
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
723
raise errors.BadIndexOptions(self)
724
options_line = lines[2]
725
if not options_line.startswith(_OPTION_LEN):
726
raise errors.BadIndexOptions(self)
728
self._key_count = int(options_line[len(_OPTION_LEN):])
730
raise errors.BadIndexOptions(self)
731
# calculate the bytes we have processed
732
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
734
self._parsed_bytes(0, None, header_end, None)
735
# setup parsing state
736
self._expected_elements = 3 + self._key_length
737
# raw data keyed by offset
738
self._keys_by_offset = {}
739
# keys with the value and node references
740
self._bisect_nodes = {}
741
return header_end, bytes[header_end:]
743
def _parse_region(self, offset, data):
744
"""Parse node data returned from a readv operation.
746
:param offset: The byte offset the data starts at.
747
:param data: The data to parse.
751
end = offset + len(data)
754
# Trivial test - if the current index's end is within the
755
# low-matching parsed range, we're done.
756
index = self._parsed_byte_index(high_parsed)
757
if end < self._parsed_byte_map[index][1]:
759
# print "[%d:%d]" % (offset, end), \
760
# self._parsed_byte_map[index:index + 2]
761
high_parsed, last_segment = self._parse_segment(
762
offset, data, end, index)
766
def _parse_segment(self, offset, data, end, index):
767
"""Parse one segment of data.
769
:param offset: Where 'data' begins in the file.
770
:param data: Some data to parse a segment of.
771
:param end: Where data ends
772
:param index: The current index into the parsed bytes map.
773
:return: True if the parsed segment is the last possible one in the
775
:return: high_parsed_byte, last_segment.
776
high_parsed_byte is the location of the highest parsed byte in this
777
segment, last_segment is True if the parsed segment is the last
778
possible one in the data block.
780
# default is to use all data
782
# accomodate overlap with data before this.
783
if offset < self._parsed_byte_map[index][1]:
784
# overlaps the lower parsed region
785
# skip the parsed data
786
trim_start = self._parsed_byte_map[index][1] - offset
787
# don't trim the start for \n
788
start_adjacent = True
789
elif offset == self._parsed_byte_map[index][1]:
790
# abuts the lower parsed region
793
# do not trim anything
794
start_adjacent = True
796
# does not overlap the lower parsed region
799
# but trim the leading \n
800
start_adjacent = False
801
if end == self._size:
802
# lines up to the end of all data:
805
# do not strip to the last \n
808
elif index + 1 == len(self._parsed_byte_map):
809
# at the end of the parsed data
812
# but strip to the last \n
815
elif end == self._parsed_byte_map[index + 1][0]:
816
# buts up against the next parsed region
819
# do not strip to the last \n
822
elif end > self._parsed_byte_map[index + 1][0]:
823
# overlaps into the next parsed region
824
# only consider the unparsed data
825
trim_end = self._parsed_byte_map[index + 1][0] - offset
826
# do not strip to the last \n as we know its an entire record
828
last_segment = end < self._parsed_byte_map[index + 1][1]
830
# does not overlap into the next region
833
# but strip to the last \n
836
# now find bytes to discard if needed
837
if not start_adjacent:
838
# work around python bug in rfind
839
if trim_start is None:
840
trim_start = data.find('\n') + 1
842
trim_start = data.find('\n', trim_start) + 1
843
if not (trim_start != 0):
844
raise AssertionError('no \n was present')
845
# print 'removing start', offset, trim_start, repr(data[:trim_start])
847
# work around python bug in rfind
849
trim_end = data.rfind('\n') + 1
851
trim_end = data.rfind('\n', None, trim_end) + 1
852
if not (trim_end != 0):
853
raise AssertionError('no \n was present')
854
# print 'removing end', offset, trim_end, repr(data[trim_end:])
855
# adjust offset and data to the parseable data.
856
trimmed_data = data[trim_start:trim_end]
857
if not (trimmed_data):
858
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
859
% (trim_start, trim_end, offset, offset + len(data)))
862
# print "parsing", repr(trimmed_data)
863
# splitlines mangles the \r delimiters.. don't use it.
864
lines = trimmed_data.split('\n')
867
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
868
for key, value in nodes:
869
self._bisect_nodes[key] = value
870
self._parsed_bytes(offset, first_key,
871
offset + len(trimmed_data), last_key)
872
return offset + len(trimmed_data), last_segment
874
def _parse_lines(self, lines, pos):
883
if not (self._size == pos + 1):
884
raise AssertionError("%s %s" % (self._size, pos))
887
elements = line.split('\0')
888
if len(elements) != self._expected_elements:
889
raise errors.BadIndexData(self)
890
# keys are tuples. Each element is a string that may occur many
891
# times, so we intern them to save space. AB, RC, 200807
892
key = tuple(intern(element) for element in elements[:self._key_length])
893
if first_key is None:
895
absent, references, value = elements[-3:]
897
for ref_string in references.split('\t'):
898
ref_lists.append(tuple([
899
int(ref) for ref in ref_string.split('\r') if ref
901
ref_lists = tuple(ref_lists)
902
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
903
pos += len(line) + 1 # +1 for the \n
906
if self.node_ref_lists:
907
node_value = (value, ref_lists)
910
nodes.append((key, node_value))
911
# print "parsed ", key
912
return first_key, key, nodes, trailers
914
def _parsed_bytes(self, start, start_key, end, end_key):
915
"""Mark the bytes from start to end as parsed.
917
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
920
:param start: The start of the parsed region.
921
:param end: The end of the parsed region.
923
index = self._parsed_byte_index(start)
924
new_value = (start, end)
925
new_key = (start_key, end_key)
927
# first range parsed is always the beginning.
928
self._parsed_byte_map.insert(index, new_value)
929
self._parsed_key_map.insert(index, new_key)
933
# extend lower region
934
# extend higher region
935
# combine two regions
936
if (index + 1 < len(self._parsed_byte_map) and
937
self._parsed_byte_map[index][1] == start and
938
self._parsed_byte_map[index + 1][0] == end):
939
# combine two regions
940
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
941
self._parsed_byte_map[index + 1][1])
942
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
943
self._parsed_key_map[index + 1][1])
944
del self._parsed_byte_map[index + 1]
945
del self._parsed_key_map[index + 1]
946
elif self._parsed_byte_map[index][1] == start:
947
# extend the lower entry
948
self._parsed_byte_map[index] = (
949
self._parsed_byte_map[index][0], end)
950
self._parsed_key_map[index] = (
951
self._parsed_key_map[index][0], end_key)
952
elif (index + 1 < len(self._parsed_byte_map) and
953
self._parsed_byte_map[index + 1][0] == end):
954
# extend the higher entry
955
self._parsed_byte_map[index + 1] = (
956
start, self._parsed_byte_map[index + 1][1])
957
self._parsed_key_map[index + 1] = (
958
start_key, self._parsed_key_map[index + 1][1])
961
self._parsed_byte_map.insert(index + 1, new_value)
962
self._parsed_key_map.insert(index + 1, new_key)
964
def _read_and_parse(self, readv_ranges):
965
"""Read the the ranges and parse the resulting data.
967
:param readv_ranges: A prepared readv range list.
970
readv_data = self._transport.readv(self._name, readv_ranges, True,
973
for offset, data in readv_data:
974
if self._bisect_nodes is None:
975
# this must be the start
976
if not (offset == 0):
977
raise AssertionError()
978
offset, data = self._parse_header_from_bytes(data)
979
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
980
self._parse_region(offset, data)
982
def _signature(self):
983
"""The file signature for this index type."""
987
"""Validate that everything in the index can be accessed."""
988
# iter_all validates completely at the moment, so just do that.
989
for node in self.iter_all_entries():
993
class CombinedGraphIndex(object):
994
"""A GraphIndex made up from smaller GraphIndices.
996
The backing indices must implement GraphIndex, and are presumed to be
999
Queries against the combined index will be made against the first index,
1000
and then the second and so on. The order of index's can thus influence
1001
performance significantly. For example, if one index is on local disk and a
1002
second on a remote server, the local disk index should be before the other
1006
def __init__(self, indices):
1007
"""Create a CombinedGraphIndex backed by indices.
1009
:param indices: An ordered list of indices to query for data.
1011
self._indices = indices
1015
self.__class__.__name__,
1016
', '.join(map(repr, self._indices)))
1018
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1019
def get_parents(self, revision_ids):
1020
"""See graph._StackedParentsProvider.get_parents.
1022
This implementation thunks the graph.Graph.get_parents api across to
1025
:param revision_ids: An iterable of graph keys for this graph.
1026
:return: A list of parent details for each key in revision_ids.
1027
Each parent details will be one of:
1028
* None when the key was missing
1029
* (NULL_REVISION,) when the key has no parents.
1030
* (parent_key, parent_key...) otherwise.
1032
parent_map = self.get_parent_map(revision_ids)
1033
return [parent_map.get(r, None) for r in revision_ids]
1035
def get_parent_map(self, keys):
1036
"""See graph._StackedParentsProvider.get_parent_map"""
1037
search_keys = set(keys)
1038
if NULL_REVISION in search_keys:
1039
search_keys.discard(NULL_REVISION)
1040
found_parents = {NULL_REVISION:[]}
1043
for index, key, value, refs in self.iter_entries(search_keys):
1046
parents = (NULL_REVISION,)
1047
found_parents[key] = parents
1048
return found_parents
1050
def insert_index(self, pos, index):
1051
"""Insert a new index in the list of indices to query.
1053
:param pos: The position to insert the index.
1054
:param index: The index to insert.
1056
self._indices.insert(pos, index)
1058
def iter_all_entries(self):
1059
"""Iterate over all keys within the index
1061
Duplicate keys across child indices are presumed to have the same
1062
value and are only reported once.
1064
:return: An iterable of (index, key, reference_lists, value).
1065
There is no defined order for the result iteration - it will be in
1066
the most efficient order for the index.
1069
for index in self._indices:
1070
for node in index.iter_all_entries():
1071
if node[1] not in seen_keys:
1073
seen_keys.add(node[1])
1075
def iter_entries(self, keys):
1076
"""Iterate over keys within the index.
1078
Duplicate keys across child indices are presumed to have the same
1079
value and are only reported once.
1081
:param keys: An iterable providing the keys to be retrieved.
1082
:return: An iterable of (index, key, reference_lists, value). There is no
1083
defined order for the result iteration - it will be in the most
1084
efficient order for the index.
1087
for index in self._indices:
1090
for node in index.iter_entries(keys):
1091
keys.remove(node[1])
1094
def iter_entries_prefix(self, keys):
1095
"""Iterate over keys within the index using prefix matching.
1097
Duplicate keys across child indices are presumed to have the same
1098
value and are only reported once.
1100
Prefix matching is applied within the tuple of a key, not to within
1101
the bytestring of each key element. e.g. if you have the keys ('foo',
1102
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1103
only the former key is returned.
1105
:param keys: An iterable providing the key prefixes to be retrieved.
1106
Each key prefix takes the form of a tuple the length of a key, but
1107
with the last N elements 'None' rather than a regular bytestring.
1108
The first element cannot be 'None'.
1109
:return: An iterable as per iter_all_entries, but restricted to the
1110
keys with a matching prefix to those supplied. No additional keys
1111
will be returned, and every match that is in the index will be
1118
for index in self._indices:
1119
for node in index.iter_entries_prefix(keys):
1120
if node[1] in seen_keys:
1122
seen_keys.add(node[1])
1125
def key_count(self):
1126
"""Return an estimate of the number of keys in this index.
1128
For CombinedGraphIndex this is approximated by the sum of the keys of
1129
the child indices. As child indices may have duplicate keys this can
1130
have a maximum error of the number of child indices * largest number of
1133
return sum((index.key_count() for index in self._indices), 0)
1136
"""Validate that everything in the index can be accessed."""
1137
for index in self._indices:
1141
class InMemoryGraphIndex(GraphIndexBuilder):
1142
"""A GraphIndex which operates entirely out of memory and is mutable.
1144
This is designed to allow the accumulation of GraphIndex entries during a
1145
single write operation, where the accumulated entries need to be immediately
1146
available - for example via a CombinedGraphIndex.
1149
def add_nodes(self, nodes):
1150
"""Add nodes to the index.
1152
:param nodes: An iterable of (key, node_refs, value) entries to add.
1154
if self.reference_lists:
1155
for (key, value, node_refs) in nodes:
1156
self.add_node(key, value, node_refs)
1158
for (key, value) in nodes:
1159
self.add_node(key, value)
1161
def iter_all_entries(self):
1162
"""Iterate over all keys within the index
1164
:return: An iterable of (index, key, reference_lists, value). There is no
1165
defined order for the result iteration - it will be in the most
1166
efficient order for the index (in this case dictionary hash order).
1168
if 'evil' in debug.debug_flags:
1169
trace.mutter_callsite(3,
1170
"iter_all_entries scales with size of history.")
1171
if self.reference_lists:
1172
for key, (absent, references, value) in self._nodes.iteritems():
1174
yield self, key, value, references
1176
for key, (absent, references, value) in self._nodes.iteritems():
1178
yield self, key, value
1180
def iter_entries(self, keys):
1181
"""Iterate over keys within the index.
1183
:param keys: An iterable providing the keys to be retrieved.
1184
:return: An iterable of (index, key, value, reference_lists). There is no
1185
defined order for the result iteration - it will be in the most
1186
efficient order for the index (keys iteration order in this case).
1189
if self.reference_lists:
1190
for key in keys.intersection(self._keys):
1191
node = self._nodes[key]
1193
yield self, key, node[2], node[1]
1195
for key in keys.intersection(self._keys):
1196
node = self._nodes[key]
1198
yield self, key, node[2]
1200
def iter_entries_prefix(self, keys):
1201
"""Iterate over keys within the index using prefix matching.
1203
Prefix matching is applied within the tuple of a key, not to within
1204
the bytestring of each key element. e.g. if you have the keys ('foo',
1205
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1206
only the former key is returned.
1208
:param keys: An iterable providing the key prefixes to be retrieved.
1209
Each key prefix takes the form of a tuple the length of a key, but
1210
with the last N elements 'None' rather than a regular bytestring.
1211
The first element cannot be 'None'.
1212
:return: An iterable as per iter_all_entries, but restricted to the
1213
keys with a matching prefix to those supplied. No additional keys
1214
will be returned, and every match that is in the index will be
1217
# XXX: To much duplication with the GraphIndex class; consider finding
1218
# a good place to pull out the actual common logic.
1222
if self._key_length == 1:
1226
raise errors.BadIndexKey(key)
1227
if len(key) != self._key_length:
1228
raise errors.BadIndexKey(key)
1229
node = self._nodes[key]
1232
if self.reference_lists:
1233
yield self, key, node[2], node[1]
1235
yield self, key, node[2]
1240
raise errors.BadIndexKey(key)
1241
if len(key) != self._key_length:
1242
raise errors.BadIndexKey(key)
1243
# find what it refers to:
1244
key_dict = self._nodes_by_key
1245
elements = list(key)
1246
# find the subdict to return
1248
while len(elements) and elements[0] is not None:
1249
key_dict = key_dict[elements[0]]
1252
# a non-existant lookup.
1257
key_dict = dicts.pop(-1)
1258
# can't be empty or would not exist
1259
item, value = key_dict.iteritems().next()
1260
if type(value) == dict:
1262
dicts.extend(key_dict.itervalues())
1265
for value in key_dict.itervalues():
1266
yield (self, ) + value
1268
yield (self, ) + key_dict
1270
def key_count(self):
1271
"""Return an estimate of the number of keys in this index.
1273
For InMemoryGraphIndex the estimate is exact.
1275
return len(self._keys)
1278
"""In memory index's have no known corruption at the moment."""
1281
class GraphIndexPrefixAdapter(object):
1282
"""An adapter between GraphIndex with different key lengths.
1284
Queries against this will emit queries against the adapted Graph with the
1285
prefix added, queries for all items use iter_entries_prefix. The returned
1286
nodes will have their keys and node references adjusted to remove the
1287
prefix. Finally, an add_nodes_callback can be supplied - when called the
1288
nodes and references being added will have prefix prepended.
1291
def __init__(self, adapted, prefix, missing_key_length,
1292
add_nodes_callback=None):
1293
"""Construct an adapter against adapted with prefix."""
1294
self.adapted = adapted
1295
self.prefix_key = prefix + (None,)*missing_key_length
1296
self.prefix = prefix
1297
self.prefix_len = len(prefix)
1298
self.add_nodes_callback = add_nodes_callback
1300
def add_nodes(self, nodes):
1301
"""Add nodes to the index.
1303
:param nodes: An iterable of (key, node_refs, value) entries to add.
1305
# save nodes in case its an iterator
1306
nodes = tuple(nodes)
1307
translated_nodes = []
1309
# Add prefix_key to each reference node_refs is a tuple of tuples,
1310
# so split it apart, and add prefix_key to the internal reference
1311
for (key, value, node_refs) in nodes:
1312
adjusted_references = (
1313
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1314
for ref_list in node_refs))
1315
translated_nodes.append((self.prefix + key, value,
1316
adjusted_references))
1318
# XXX: TODO add an explicit interface for getting the reference list
1319
# status, to handle this bit of user-friendliness in the API more
1321
for (key, value) in nodes:
1322
translated_nodes.append((self.prefix + key, value))
1323
self.add_nodes_callback(translated_nodes)
1325
def add_node(self, key, value, references=()):
1326
"""Add a node to the index.
1328
:param key: The key. keys are non-empty tuples containing
1329
as many whitespace-free utf8 bytestrings as the key length
1330
defined for this index.
1331
:param references: An iterable of iterables of keys. Each is a
1332
reference to another key.
1333
:param value: The value to associate with the key. It may be any
1334
bytes as long as it does not contain \0 or \n.
1336
self.add_nodes(((key, value, references), ))
1338
def _strip_prefix(self, an_iter):
1339
"""Strip prefix data from nodes and return it."""
1340
for node in an_iter:
1342
if node[1][:self.prefix_len] != self.prefix:
1343
raise errors.BadIndexData(self)
1344
for ref_list in node[3]:
1345
for ref_node in ref_list:
1346
if ref_node[:self.prefix_len] != self.prefix:
1347
raise errors.BadIndexData(self)
1348
yield node[0], node[1][self.prefix_len:], node[2], (
1349
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1350
for ref_list in node[3]))
1352
def iter_all_entries(self):
1353
"""Iterate over all keys within the index
1355
iter_all_entries is implemented against the adapted index using
1356
iter_entries_prefix.
1358
:return: An iterable of (index, key, reference_lists, value). There is no
1359
defined order for the result iteration - it will be in the most
1360
efficient order for the index (in this case dictionary hash order).
1362
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1364
def iter_entries(self, keys):
1365
"""Iterate over keys within the index.
1367
:param keys: An iterable providing the keys to be retrieved.
1368
:return: An iterable of (index, key, value, reference_lists). There is no
1369
defined order for the result iteration - it will be in the most
1370
efficient order for the index (keys iteration order in this case).
1372
return self._strip_prefix(self.adapted.iter_entries(
1373
self.prefix + key for key in keys))
1375
def iter_entries_prefix(self, keys):
1376
"""Iterate over keys within the index using prefix matching.
1378
Prefix matching is applied within the tuple of a key, not to within
1379
the bytestring of each key element. e.g. if you have the keys ('foo',
1380
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1381
only the former key is returned.
1383
:param keys: An iterable providing the key prefixes to be retrieved.
1384
Each key prefix takes the form of a tuple the length of a key, but
1385
with the last N elements 'None' rather than a regular bytestring.
1386
The first element cannot be 'None'.
1387
:return: An iterable as per iter_all_entries, but restricted to the
1388
keys with a matching prefix to those supplied. No additional keys
1389
will be returned, and every match that is in the index will be
1392
return self._strip_prefix(self.adapted.iter_entries_prefix(
1393
self.prefix + key for key in keys))
1395
def key_count(self):
1396
"""Return an estimate of the number of keys in this index.
1398
For GraphIndexPrefixAdapter this is relatively expensive - key
1399
iteration with the prefix is done.
1401
return len(list(self.iter_all_entries()))
1404
"""Call the adapted's validate."""
1405
self.adapted.validate()
added
removed
bzrlib/index.py
