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))
222
return StringIO(''.join(lines))
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)
287
def _buffer_all(self):
288
"""Buffer all the index data.
290
Mutates self._nodes and self.keys_by_offset.
292
if 'index' in debug.debug_flags:
293
mutter('Reading entire index %s', self._transport.abspath(self._name))
294
stream = self._transport.get(self._name)
295
self._read_prefix(stream)
296
self._expected_elements = 3 + self._key_length
298
# raw data keyed by offset
299
self._keys_by_offset = {}
300
# ready-to-return key:value or key:value, node_ref_lists
302
self._nodes_by_key = {}
305
lines = stream.read().split('\n')
307
_, _, _, trailers = self._parse_lines(lines, pos)
308
for key, absent, references, value in self._keys_by_offset.itervalues():
311
# resolve references:
312
if self.node_ref_lists:
313
node_value = (value, self._resolve_references(references))
316
self._nodes[key] = node_value
317
if self._key_length > 1:
318
subkey = list(reversed(key[:-1]))
319
key_dict = self._nodes_by_key
320
if self.node_ref_lists:
321
key_value = key, node_value[0], node_value[1]
323
key_value = key, node_value
324
# possibly should do this on-demand, but it seems likely it is
326
# For a key of (foo, bar, baz) create
327
# _nodes_by_key[foo][bar][baz] = key_value
328
for subkey in key[:-1]:
329
key_dict = key_dict.setdefault(subkey, {})
330
key_dict[key[-1]] = key_value
331
# cache the keys for quick set intersections
332
self._keys = set(self._nodes)
334
# there must be one line - the empty trailer line.
335
raise errors.BadIndexData(self)
337
def iter_all_entries(self):
338
"""Iterate over all keys within the index.
340
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
341
The former tuple is used when there are no reference lists in the
342
index, making the API compatible with simple key:value index types.
343
There is no defined order for the result iteration - it will be in
344
the most efficient order for the index.
346
if 'evil' in debug.debug_flags:
347
trace.mutter_callsite(3,
348
"iter_all_entries scales with size of history.")
349
if self._nodes is None:
351
if self.node_ref_lists:
352
for key, (value, node_ref_lists) in self._nodes.iteritems():
353
yield self, key, value, node_ref_lists
355
for key, value in self._nodes.iteritems():
356
yield self, key, value
358
def _read_prefix(self, stream):
359
signature = stream.read(len(self._signature()))
360
if not signature == self._signature():
361
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
362
options_line = stream.readline()
363
if not options_line.startswith(_OPTION_NODE_REFS):
364
raise errors.BadIndexOptions(self)
366
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
368
raise errors.BadIndexOptions(self)
369
options_line = stream.readline()
370
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
371
raise errors.BadIndexOptions(self)
373
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
375
raise errors.BadIndexOptions(self)
376
options_line = stream.readline()
377
if not options_line.startswith(_OPTION_LEN):
378
raise errors.BadIndexOptions(self)
380
self._key_count = int(options_line[len(_OPTION_LEN):-1])
382
raise errors.BadIndexOptions(self)
384
def _resolve_references(self, references):
385
"""Return the resolved key references for references.
387
References are resolved by looking up the location of the key in the
388
_keys_by_offset map and substituting the key name, preserving ordering.
390
:param references: An iterable of iterables of key locations. e.g.
392
:return: A tuple of tuples of keys.
395
for ref_list in references:
396
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
397
return tuple(node_refs)
399
def _find_index(self, range_map, key):
400
"""Helper for the _parsed_*_index calls.
402
Given a range map - [(start, end), ...], finds the index of the range
403
in the map for key if it is in the map, and if it is not there, the
404
immediately preceeding range in the map.
406
result = bisect_right(range_map, key) - 1
407
if result + 1 < len(range_map):
408
# check the border condition, it may be in result + 1
409
if range_map[result + 1][0] == key[0]:
413
def _parsed_byte_index(self, offset):
414
"""Return the index of the entry immediately before offset.
416
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
417
there is one unparsed byte (the 11th, addressed as[10]). then:
418
asking for 0 will return 0
419
asking for 10 will return 0
420
asking for 11 will return 1
421
asking for 12 will return 1
424
return self._find_index(self._parsed_byte_map, key)
426
def _parsed_key_index(self, key):
427
"""Return the index of the entry immediately before key.
429
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
430
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
431
have been parsed, then:
432
asking for '' will return 0
433
asking for 'a' will return 0
434
asking for 'b' will return 1
435
asking for 'e' will return 1
437
search_key = (key, None)
438
return self._find_index(self._parsed_key_map, search_key)
440
def _is_parsed(self, offset):
441
"""Returns True if offset has been parsed."""
442
index = self._parsed_byte_index(offset)
443
if index == len(self._parsed_byte_map):
444
return offset < self._parsed_byte_map[index - 1][1]
445
start, end = self._parsed_byte_map[index]
446
return offset >= start and offset < end
448
def _iter_entries_from_total_buffer(self, keys):
449
"""Iterate over keys when the entire index is parsed."""
450
keys = keys.intersection(self._keys)
451
if self.node_ref_lists:
453
value, node_refs = self._nodes[key]
454
yield self, key, value, node_refs
457
yield self, key, self._nodes[key]
459
def iter_entries(self, keys):
460
"""Iterate over keys within the index.
462
:param keys: An iterable providing the keys to be retrieved.
463
:return: An iterable as per iter_all_entries, but restricted to the
464
keys supplied. No additional keys will be returned, and every
465
key supplied that is in the index will be returned.
467
# PERFORMANCE TODO: parse and bisect all remaining data at some
468
# threshold of total-index processing/get calling layers that expect to
469
# read the entire index to use the iter_all_entries method instead.
473
if self._size is None and self._nodes is None:
475
if self._nodes is not None:
476
return self._iter_entries_from_total_buffer(keys)
478
return (result[1] for result in bisect_multi_bytes(
479
self._lookup_keys_via_location, self._size, keys))
481
def iter_entries_prefix(self, keys):
482
"""Iterate over keys within the index using prefix matching.
484
Prefix matching is applied within the tuple of a key, not to within
485
the bytestring of each key element. e.g. if you have the keys ('foo',
486
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
487
only the former key is returned.
489
WARNING: Note that this method currently causes a full index parse
490
unconditionally (which is reasonably appropriate as it is a means for
491
thunking many small indices into one larger one and still supplies
492
iter_all_entries at the thunk layer).
494
:param keys: An iterable providing the key prefixes to be retrieved.
495
Each key prefix takes the form of a tuple the length of a key, but
496
with the last N elements 'None' rather than a regular bytestring.
497
The first element cannot be 'None'.
498
:return: An iterable as per iter_all_entries, but restricted to the
499
keys with a matching prefix to those supplied. No additional keys
500
will be returned, and every match that is in the index will be
506
# load data - also finds key lengths
507
if self._nodes is None:
509
if self._key_length == 1:
513
raise errors.BadIndexKey(key)
514
if len(key) != self._key_length:
515
raise errors.BadIndexKey(key)
516
if self.node_ref_lists:
517
value, node_refs = self._nodes[key]
518
yield self, key, value, node_refs
520
yield self, key, self._nodes[key]
525
raise errors.BadIndexKey(key)
526
if len(key) != self._key_length:
527
raise errors.BadIndexKey(key)
528
# find what it refers to:
529
key_dict = self._nodes_by_key
531
# find the subdict whose contents should be returned.
533
while len(elements) and elements[0] is not None:
534
key_dict = key_dict[elements[0]]
537
# a non-existant lookup.
542
key_dict = dicts.pop(-1)
543
# can't be empty or would not exist
544
item, value = key_dict.iteritems().next()
545
if type(value) == dict:
547
dicts.extend(key_dict.itervalues())
550
for value in key_dict.itervalues():
551
# each value is the key:value:node refs tuple
553
yield (self, ) + value
555
# the last thing looked up was a terminal element
556
yield (self, ) + key_dict
559
"""Return an estimate of the number of keys in this index.
561
For GraphIndex the estimate is exact.
563
if self._key_count is None:
564
self._read_and_parse([_HEADER_READV])
565
return self._key_count
567
def _lookup_keys_via_location(self, location_keys):
568
"""Public interface for implementing bisection.
570
If _buffer_all has been called, then all the data for the index is in
571
memory, and this method should not be called, as it uses a separate
572
cache because it cannot pre-resolve all indices, which buffer_all does
575
:param location_keys: A list of location(byte offset), key tuples.
576
:return: A list of (location_key, result) tuples as expected by
577
bzrlib.bisect_multi.bisect_multi_bytes.
579
# Possible improvements:
580
# - only bisect lookup each key once
581
# - sort the keys first, and use that to reduce the bisection window
583
# this progresses in three parts:
586
# attempt to answer the question from the now in memory data.
587
# build the readv request
588
# for each location, ask for 800 bytes - much more than rows we've seen
591
for location, key in location_keys:
592
# can we answer from cache?
593
if self._bisect_nodes and key in self._bisect_nodes:
594
# We have the key parsed.
596
index = self._parsed_key_index(key)
597
if (len(self._parsed_key_map) and
598
self._parsed_key_map[index][0] <= key and
599
(self._parsed_key_map[index][1] >= key or
600
# end of the file has been parsed
601
self._parsed_byte_map[index][1] == self._size)):
602
# the key has been parsed, so no lookup is needed even if its
605
# - if we have examined this part of the file already - yes
606
index = self._parsed_byte_index(location)
607
if (len(self._parsed_byte_map) and
608
self._parsed_byte_map[index][0] <= location and
609
self._parsed_byte_map[index][1] > location):
610
# the byte region has been parsed, so no read is needed.
613
if location + length > self._size:
614
length = self._size - location
615
# todo, trim out parsed locations.
617
readv_ranges.append((location, length))
618
# read the header if needed
619
if self._bisect_nodes is None:
620
readv_ranges.append(_HEADER_READV)
621
self._read_and_parse(readv_ranges)
623
# - figure out <, >, missing, present
624
# - result present references so we can return them.
626
# keys that we cannot answer until we resolve references
627
pending_references = []
628
pending_locations = set()
629
for location, key in location_keys:
630
# can we answer from cache?
631
if key in self._bisect_nodes:
632
# the key has been parsed, so no lookup is needed
633
if self.node_ref_lists:
634
# the references may not have been all parsed.
635
value, refs = self._bisect_nodes[key]
636
wanted_locations = []
637
for ref_list in refs:
639
if ref not in self._keys_by_offset:
640
wanted_locations.append(ref)
642
pending_locations.update(wanted_locations)
643
pending_references.append((location, key))
645
result.append(((location, key), (self, key,
646
value, self._resolve_references(refs))))
648
result.append(((location, key),
649
(self, key, self._bisect_nodes[key])))
652
# has the region the key should be in, been parsed?
653
index = self._parsed_key_index(key)
654
if (self._parsed_key_map[index][0] <= key and
655
(self._parsed_key_map[index][1] >= key or
656
# end of the file has been parsed
657
self._parsed_byte_map[index][1] == self._size)):
658
result.append(((location, key), False))
660
# no, is the key above or below the probed location:
661
# get the range of the probed & parsed location
662
index = self._parsed_byte_index(location)
663
# if the key is below the start of the range, its below
664
if key < self._parsed_key_map[index][0]:
668
result.append(((location, key), direction))
670
# lookup data to resolve references
671
for location in pending_locations:
673
if location + length > self._size:
674
length = self._size - location
675
# TODO: trim out parsed locations (e.g. if the 800 is into the
676
# parsed region trim it, and dont use the adjust_for_latency
679
readv_ranges.append((location, length))
680
self._read_and_parse(readv_ranges)
681
for location, key in pending_references:
682
# answer key references we had to look-up-late.
683
index = self._parsed_key_index(key)
684
value, refs = self._bisect_nodes[key]
685
result.append(((location, key), (self, key,
686
value, self._resolve_references(refs))))
689
def _parse_header_from_bytes(self, bytes):
690
"""Parse the header from a region of bytes.
692
:param bytes: The data to parse.
693
:return: An offset, data tuple such as readv yields, for the unparsed
694
data. (which may length 0).
696
signature = bytes[0:len(self._signature())]
697
if not signature == self._signature():
698
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
699
lines = bytes[len(self._signature()):].splitlines()
700
options_line = lines[0]
701
if not options_line.startswith(_OPTION_NODE_REFS):
702
raise errors.BadIndexOptions(self)
704
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
706
raise errors.BadIndexOptions(self)
707
options_line = lines[1]
708
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
709
raise errors.BadIndexOptions(self)
711
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
713
raise errors.BadIndexOptions(self)
714
options_line = lines[2]
715
if not options_line.startswith(_OPTION_LEN):
716
raise errors.BadIndexOptions(self)
718
self._key_count = int(options_line[len(_OPTION_LEN):])
720
raise errors.BadIndexOptions(self)
721
# calculate the bytes we have processed
722
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
724
self._parsed_bytes(0, None, header_end, None)
725
# setup parsing state
726
self._expected_elements = 3 + self._key_length
727
# raw data keyed by offset
728
self._keys_by_offset = {}
729
# keys with the value and node references
730
self._bisect_nodes = {}
731
return header_end, bytes[header_end:]
733
def _parse_region(self, offset, data):
734
"""Parse node data returned from a readv operation.
736
:param offset: The byte offset the data starts at.
737
:param data: The data to parse.
741
end = offset + len(data)
744
# Trivial test - if the current index's end is within the
745
# low-matching parsed range, we're done.
746
index = self._parsed_byte_index(high_parsed)
747
if end < self._parsed_byte_map[index][1]:
749
# print "[%d:%d]" % (offset, end), \
750
# self._parsed_byte_map[index:index + 2]
751
high_parsed, last_segment = self._parse_segment(
752
offset, data, end, index)
756
def _parse_segment(self, offset, data, end, index):
757
"""Parse one segment of data.
759
:param offset: Where 'data' begins in the file.
760
:param data: Some data to parse a segment of.
761
:param end: Where data ends
762
:param index: The current index into the parsed bytes map.
763
:return: True if the parsed segment is the last possible one in the
765
:return: high_parsed_byte, last_segment.
766
high_parsed_byte is the location of the highest parsed byte in this
767
segment, last_segment is True if the parsed segment is the last
768
possible one in the data block.
770
# default is to use all data
772
# accomodate overlap with data before this.
773
if offset < self._parsed_byte_map[index][1]:
774
# overlaps the lower parsed region
775
# skip the parsed data
776
trim_start = self._parsed_byte_map[index][1] - offset
777
# don't trim the start for \n
778
start_adjacent = True
779
elif offset == self._parsed_byte_map[index][1]:
780
# abuts the lower parsed region
783
# do not trim anything
784
start_adjacent = True
786
# does not overlap the lower parsed region
789
# but trim the leading \n
790
start_adjacent = False
791
if end == self._size:
792
# lines up to the end of all data:
795
# do not strip to the last \n
798
elif index + 1 == len(self._parsed_byte_map):
799
# at the end of the parsed data
802
# but strip to the last \n
805
elif end == self._parsed_byte_map[index + 1][0]:
806
# buts up against the next parsed region
809
# do not strip to the last \n
812
elif end > self._parsed_byte_map[index + 1][0]:
813
# overlaps into the next parsed region
814
# only consider the unparsed data
815
trim_end = self._parsed_byte_map[index + 1][0] - offset
816
# do not strip to the last \n as we know its an entire record
818
last_segment = end < self._parsed_byte_map[index + 1][1]
820
# does not overlap into the next region
823
# but strip to the last \n
826
# now find bytes to discard if needed
827
if not start_adjacent:
828
# work around python bug in rfind
829
if trim_start is None:
830
trim_start = data.find('\n') + 1
832
trim_start = data.find('\n', trim_start) + 1
833
assert trim_start != 0, 'no \n was present'
834
# print 'removing start', offset, trim_start, repr(data[:trim_start])
836
# work around python bug in rfind
838
trim_end = data.rfind('\n') + 1
840
trim_end = data.rfind('\n', None, trim_end) + 1
841
assert trim_end != 0, 'no \n was present'
842
# print 'removing end', offset, trim_end, repr(data[trim_end:])
843
# adjust offset and data to the parseable data.
844
trimmed_data = data[trim_start:trim_end]
845
assert trimmed_data, 'read unneeded data [%d:%d] from [%d:%d]' % (
846
trim_start, trim_end, offset, offset + len(data))
849
# print "parsing", repr(trimmed_data)
850
# splitlines mangles the \r delimiters.. don't use it.
851
lines = trimmed_data.split('\n')
854
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
855
for key, value in nodes:
856
self._bisect_nodes[key] = value
857
self._parsed_bytes(offset, first_key,
858
offset + len(trimmed_data), last_key)
859
return offset + len(trimmed_data), last_segment
861
def _parse_lines(self, lines, pos):
870
assert self._size == pos + 1, "%s %s" % (self._size, pos)
873
elements = line.split('\0')
874
if len(elements) != self._expected_elements:
875
raise errors.BadIndexData(self)
877
key = tuple(elements[:self._key_length])
878
if first_key is None:
880
absent, references, value = elements[-3:]
882
for ref_string in references.split('\t'):
883
ref_lists.append(tuple([
884
int(ref) for ref in ref_string.split('\r') if ref
886
ref_lists = tuple(ref_lists)
887
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
888
pos += len(line) + 1 # +1 for the \n
891
if self.node_ref_lists:
892
node_value = (value, ref_lists)
895
nodes.append((key, node_value))
896
# print "parsed ", key
897
return first_key, key, nodes, trailers
899
def _parsed_bytes(self, start, start_key, end, end_key):
900
"""Mark the bytes from start to end as parsed.
902
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
905
:param start: The start of the parsed region.
906
:param end: The end of the parsed region.
908
index = self._parsed_byte_index(start)
909
new_value = (start, end)
910
new_key = (start_key, end_key)
912
# first range parsed is always the beginning.
913
self._parsed_byte_map.insert(index, new_value)
914
self._parsed_key_map.insert(index, new_key)
918
# extend lower region
919
# extend higher region
920
# combine two regions
921
if (index + 1 < len(self._parsed_byte_map) and
922
self._parsed_byte_map[index][1] == start and
923
self._parsed_byte_map[index + 1][0] == end):
924
# combine two regions
925
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
926
self._parsed_byte_map[index + 1][1])
927
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
928
self._parsed_key_map[index + 1][1])
929
del self._parsed_byte_map[index + 1]
930
del self._parsed_key_map[index + 1]
931
elif self._parsed_byte_map[index][1] == start:
932
# extend the lower entry
933
self._parsed_byte_map[index] = (
934
self._parsed_byte_map[index][0], end)
935
self._parsed_key_map[index] = (
936
self._parsed_key_map[index][0], end_key)
937
elif (index + 1 < len(self._parsed_byte_map) and
938
self._parsed_byte_map[index + 1][0] == end):
939
# extend the higher entry
940
self._parsed_byte_map[index + 1] = (
941
start, self._parsed_byte_map[index + 1][1])
942
self._parsed_key_map[index + 1] = (
943
start_key, self._parsed_key_map[index + 1][1])
946
self._parsed_byte_map.insert(index + 1, new_value)
947
self._parsed_key_map.insert(index + 1, new_key)
949
def _read_and_parse(self, readv_ranges):
950
"""Read the the ranges and parse the resulting data.
952
:param readv_ranges: A prepared readv range list.
955
readv_data = self._transport.readv(self._name, readv_ranges, True,
958
for offset, data in readv_data:
959
if self._bisect_nodes is None:
960
# this must be the start
962
offset, data = self._parse_header_from_bytes(data)
963
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
964
self._parse_region(offset, data)
966
def _signature(self):
967
"""The file signature for this index type."""
971
"""Validate that everything in the index can be accessed."""
972
# iter_all validates completely at the moment, so just do that.
973
for node in self.iter_all_entries():
977
class CombinedGraphIndex(object):
978
"""A GraphIndex made up from smaller GraphIndices.
980
The backing indices must implement GraphIndex, and are presumed to be
983
Queries against the combined index will be made against the first index,
984
and then the second and so on. The order of index's can thus influence
985
performance significantly. For example, if one index is on local disk and a
986
second on a remote server, the local disk index should be before the other
990
def __init__(self, indices):
991
"""Create a CombinedGraphIndex backed by indices.
993
:param indices: An ordered list of indices to query for data.
995
self._indices = indices
999
self.__class__.__name__,
1000
', '.join(map(repr, self._indices)))
1002
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1003
def get_parents(self, revision_ids):
1004
"""See graph._StackedParentsProvider.get_parents.
1006
This implementation thunks the graph.Graph.get_parents api across to
1009
:param revision_ids: An iterable of graph keys for this graph.
1010
:return: A list of parent details for each key in revision_ids.
1011
Each parent details will be one of:
1012
* None when the key was missing
1013
* (NULL_REVISION,) when the key has no parents.
1014
* (parent_key, parent_key...) otherwise.
1016
parent_map = self.get_parent_map(revision_ids)
1017
return [parent_map.get(r, None) for r in revision_ids]
1019
def get_parent_map(self, keys):
1020
"""See graph._StackedParentsProvider.get_parent_map"""
1021
search_keys = set(keys)
1022
if NULL_REVISION in search_keys:
1023
search_keys.discard(NULL_REVISION)
1024
found_parents = {NULL_REVISION:[]}
1027
for index, key, value, refs in self.iter_entries(search_keys):
1030
parents = (NULL_REVISION,)
1031
found_parents[key] = parents
1032
return found_parents
1034
def insert_index(self, pos, index):
1035
"""Insert a new index in the list of indices to query.
1037
:param pos: The position to insert the index.
1038
:param index: The index to insert.
1040
self._indices.insert(pos, index)
1042
def iter_all_entries(self):
1043
"""Iterate over all keys within the index
1045
Duplicate keys across child indices are presumed to have the same
1046
value and are only reported once.
1048
:return: An iterable of (index, key, reference_lists, value).
1049
There is no defined order for the result iteration - it will be in
1050
the most efficient order for the index.
1053
for index in self._indices:
1054
for node in index.iter_all_entries():
1055
if node[1] not in seen_keys:
1057
seen_keys.add(node[1])
1059
def iter_entries(self, keys):
1060
"""Iterate over keys within the index.
1062
Duplicate keys across child indices are presumed to have the same
1063
value and are only reported once.
1065
:param keys: An iterable providing the keys to be retrieved.
1066
:return: An iterable of (index, key, reference_lists, value). There is no
1067
defined order for the result iteration - it will be in the most
1068
efficient order for the index.
1071
for index in self._indices:
1074
for node in index.iter_entries(keys):
1075
keys.remove(node[1])
1078
def iter_entries_prefix(self, keys):
1079
"""Iterate over keys within the index using prefix matching.
1081
Duplicate keys across child indices are presumed to have the same
1082
value and are only reported once.
1084
Prefix matching is applied within the tuple of a key, not to within
1085
the bytestring of each key element. e.g. if you have the keys ('foo',
1086
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1087
only the former key is returned.
1089
:param keys: An iterable providing the key prefixes to be retrieved.
1090
Each key prefix takes the form of a tuple the length of a key, but
1091
with the last N elements 'None' rather than a regular bytestring.
1092
The first element cannot be 'None'.
1093
:return: An iterable as per iter_all_entries, but restricted to the
1094
keys with a matching prefix to those supplied. No additional keys
1095
will be returned, and every match that is in the index will be
1102
for index in self._indices:
1103
for node in index.iter_entries_prefix(keys):
1104
if node[1] in seen_keys:
1106
seen_keys.add(node[1])
1109
def key_count(self):
1110
"""Return an estimate of the number of keys in this index.
1112
For CombinedGraphIndex this is approximated by the sum of the keys of
1113
the child indices. As child indices may have duplicate keys this can
1114
have a maximum error of the number of child indices * largest number of
1117
return sum((index.key_count() for index in self._indices), 0)
1120
"""Validate that everything in the index can be accessed."""
1121
for index in self._indices:
1125
class InMemoryGraphIndex(GraphIndexBuilder):
1126
"""A GraphIndex which operates entirely out of memory and is mutable.
1128
This is designed to allow the accumulation of GraphIndex entries during a
1129
single write operation, where the accumulated entries need to be immediately
1130
available - for example via a CombinedGraphIndex.
1133
def add_nodes(self, nodes):
1134
"""Add nodes to the index.
1136
:param nodes: An iterable of (key, node_refs, value) entries to add.
1138
if self.reference_lists:
1139
for (key, value, node_refs) in nodes:
1140
self.add_node(key, value, node_refs)
1142
for (key, value) in nodes:
1143
self.add_node(key, value)
1145
def iter_all_entries(self):
1146
"""Iterate over all keys within the index
1148
:return: An iterable of (index, key, reference_lists, value). There is no
1149
defined order for the result iteration - it will be in the most
1150
efficient order for the index (in this case dictionary hash order).
1152
if 'evil' in debug.debug_flags:
1153
trace.mutter_callsite(3,
1154
"iter_all_entries scales with size of history.")
1155
if self.reference_lists:
1156
for key, (absent, references, value) in self._nodes.iteritems():
1158
yield self, key, value, references
1160
for key, (absent, references, value) in self._nodes.iteritems():
1162
yield self, key, value
1164
def iter_entries(self, keys):
1165
"""Iterate over keys within the index.
1167
:param keys: An iterable providing the keys to be retrieved.
1168
:return: An iterable of (index, key, value, reference_lists). There is no
1169
defined order for the result iteration - it will be in the most
1170
efficient order for the index (keys iteration order in this case).
1173
if self.reference_lists:
1174
for key in keys.intersection(self._keys):
1175
node = self._nodes[key]
1177
yield self, key, node[2], node[1]
1179
for key in keys.intersection(self._keys):
1180
node = self._nodes[key]
1182
yield self, key, node[2]
1184
def iter_entries_prefix(self, keys):
1185
"""Iterate over keys within the index using prefix matching.
1187
Prefix matching is applied within the tuple of a key, not to within
1188
the bytestring of each key element. e.g. if you have the keys ('foo',
1189
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1190
only the former key is returned.
1192
:param keys: An iterable providing the key prefixes to be retrieved.
1193
Each key prefix takes the form of a tuple the length of a key, but
1194
with the last N elements 'None' rather than a regular bytestring.
1195
The first element cannot be 'None'.
1196
:return: An iterable as per iter_all_entries, but restricted to the
1197
keys with a matching prefix to those supplied. No additional keys
1198
will be returned, and every match that is in the index will be
1201
# XXX: To much duplication with the GraphIndex class; consider finding
1202
# a good place to pull out the actual common logic.
1206
if self._key_length == 1:
1210
raise errors.BadIndexKey(key)
1211
if len(key) != self._key_length:
1212
raise errors.BadIndexKey(key)
1213
node = self._nodes[key]
1216
if self.reference_lists:
1217
yield self, key, node[2], node[1]
1219
yield self, key, node[2]
1224
raise errors.BadIndexKey(key)
1225
if len(key) != self._key_length:
1226
raise errors.BadIndexKey(key)
1227
# find what it refers to:
1228
key_dict = self._nodes_by_key
1229
elements = list(key)
1230
# find the subdict to return
1232
while len(elements) and elements[0] is not None:
1233
key_dict = key_dict[elements[0]]
1236
# a non-existant lookup.
1241
key_dict = dicts.pop(-1)
1242
# can't be empty or would not exist
1243
item, value = key_dict.iteritems().next()
1244
if type(value) == dict:
1246
dicts.extend(key_dict.itervalues())
1249
for value in key_dict.itervalues():
1250
yield (self, ) + value
1252
yield (self, ) + key_dict
1254
def key_count(self):
1255
"""Return an estimate of the number of keys in this index.
1257
For InMemoryGraphIndex the estimate is exact.
1259
return len(self._keys)
1262
"""In memory index's have no known corruption at the moment."""
1265
class GraphIndexPrefixAdapter(object):
1266
"""An adapter between GraphIndex with different key lengths.
1268
Queries against this will emit queries against the adapted Graph with the
1269
prefix added, queries for all items use iter_entries_prefix. The returned
1270
nodes will have their keys and node references adjusted to remove the
1271
prefix. Finally, an add_nodes_callback can be supplied - when called the
1272
nodes and references being added will have prefix prepended.
1275
def __init__(self, adapted, prefix, missing_key_length,
1276
add_nodes_callback=None):
1277
"""Construct an adapter against adapted with prefix."""
1278
self.adapted = adapted
1279
self.prefix_key = prefix + (None,)*missing_key_length
1280
self.prefix = prefix
1281
self.prefix_len = len(prefix)
1282
self.add_nodes_callback = add_nodes_callback
1284
def add_nodes(self, nodes):
1285
"""Add nodes to the index.
1287
:param nodes: An iterable of (key, node_refs, value) entries to add.
1289
# save nodes in case its an iterator
1290
nodes = tuple(nodes)
1291
translated_nodes = []
1293
# Add prefix_key to each reference node_refs is a tuple of tuples,
1294
# so split it apart, and add prefix_key to the internal reference
1295
for (key, value, node_refs) in nodes:
1296
adjusted_references = (
1297
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1298
for ref_list in node_refs))
1299
translated_nodes.append((self.prefix + key, value,
1300
adjusted_references))
1302
# XXX: TODO add an explicit interface for getting the reference list
1303
# status, to handle this bit of user-friendliness in the API more
1305
for (key, value) in nodes:
1306
translated_nodes.append((self.prefix + key, value))
1307
self.add_nodes_callback(translated_nodes)
1309
def add_node(self, key, value, references=()):
1310
"""Add a node to the index.
1312
:param key: The key. keys are non-empty tuples containing
1313
as many whitespace-free utf8 bytestrings as the key length
1314
defined for this index.
1315
:param references: An iterable of iterables of keys. Each is a
1316
reference to another key.
1317
:param value: The value to associate with the key. It may be any
1318
bytes as long as it does not contain \0 or \n.
1320
self.add_nodes(((key, value, references), ))
1322
def _strip_prefix(self, an_iter):
1323
"""Strip prefix data from nodes and return it."""
1324
for node in an_iter:
1326
if node[1][:self.prefix_len] != self.prefix:
1327
raise errors.BadIndexData(self)
1328
for ref_list in node[3]:
1329
for ref_node in ref_list:
1330
if ref_node[:self.prefix_len] != self.prefix:
1331
raise errors.BadIndexData(self)
1332
yield node[0], node[1][self.prefix_len:], node[2], (
1333
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1334
for ref_list in node[3]))
1336
def iter_all_entries(self):
1337
"""Iterate over all keys within the index
1339
iter_all_entries is implemented against the adapted index using
1340
iter_entries_prefix.
1342
:return: An iterable of (index, key, reference_lists, value). There is no
1343
defined order for the result iteration - it will be in the most
1344
efficient order for the index (in this case dictionary hash order).
1346
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1348
def iter_entries(self, keys):
1349
"""Iterate over keys within the index.
1351
:param keys: An iterable providing the keys to be retrieved.
1352
:return: An iterable of (index, key, value, reference_lists). There is no
1353
defined order for the result iteration - it will be in the most
1354
efficient order for the index (keys iteration order in this case).
1356
return self._strip_prefix(self.adapted.iter_entries(
1357
self.prefix + key for key in keys))
1359
def iter_entries_prefix(self, keys):
1360
"""Iterate over keys within the index using prefix matching.
1362
Prefix matching is applied within the tuple of a key, not to within
1363
the bytestring of each key element. e.g. if you have the keys ('foo',
1364
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1365
only the former key is returned.
1367
:param keys: An iterable providing the key prefixes to be retrieved.
1368
Each key prefix takes the form of a tuple the length of a key, but
1369
with the last N elements 'None' rather than a regular bytestring.
1370
The first element cannot be 'None'.
1371
:return: An iterable as per iter_all_entries, but restricted to the
1372
keys with a matching prefix to those supplied. No additional keys
1373
will be returned, and every match that is in the index will be
1376
return self._strip_prefix(self.adapted.iter_entries_prefix(
1377
self.prefix + key for key in keys))
1379
def key_count(self):
1380
"""Return an estimate of the number of keys in this index.
1382
For GraphIndexPrefixAdapter this is relatively expensive - key
1383
iteration with the prefix is done.
1385
return len(list(self.iter_all_entries()))
1388
"""Call the adapted's validate."""
1389
self.adapted.validate()
added
removed
bzrlib/index.py
