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.
471
# PERFORMANCE TODO: parse and bisect all remaining data at some
472
# threshold of total-index processing/get calling layers that expect to
473
# read the entire index to use the iter_all_entries method instead.
477
if self._size is None and self._nodes is None:
479
if self._nodes is not None:
480
return self._iter_entries_from_total_buffer(keys)
482
return (result[1] for result in bisect_multi_bytes(
483
self._lookup_keys_via_location, self._size, keys))
485
def iter_entries_prefix(self, keys):
486
"""Iterate over keys within the index using prefix matching.
488
Prefix matching is applied within the tuple of a key, not to within
489
the bytestring of each key element. e.g. if you have the keys ('foo',
490
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
491
only the former key is returned.
493
WARNING: Note that this method currently causes a full index parse
494
unconditionally (which is reasonably appropriate as it is a means for
495
thunking many small indices into one larger one and still supplies
496
iter_all_entries at the thunk layer).
498
:param keys: An iterable providing the key prefixes to be retrieved.
499
Each key prefix takes the form of a tuple the length of a key, but
500
with the last N elements 'None' rather than a regular bytestring.
501
The first element cannot be 'None'.
502
:return: An iterable as per iter_all_entries, but restricted to the
503
keys with a matching prefix to those supplied. No additional keys
504
will be returned, and every match that is in the index will be
510
# load data - also finds key lengths
511
if self._nodes is None:
513
if self._key_length == 1:
517
raise errors.BadIndexKey(key)
518
if len(key) != self._key_length:
519
raise errors.BadIndexKey(key)
520
if self.node_ref_lists:
521
value, node_refs = self._nodes[key]
522
yield self, key, value, node_refs
524
yield self, key, self._nodes[key]
529
raise errors.BadIndexKey(key)
530
if len(key) != self._key_length:
531
raise errors.BadIndexKey(key)
532
# find what it refers to:
533
key_dict = self._nodes_by_key
535
# find the subdict whose contents should be returned.
537
while len(elements) and elements[0] is not None:
538
key_dict = key_dict[elements[0]]
541
# a non-existant lookup.
546
key_dict = dicts.pop(-1)
547
# can't be empty or would not exist
548
item, value = key_dict.iteritems().next()
549
if type(value) == dict:
551
dicts.extend(key_dict.itervalues())
554
for value in key_dict.itervalues():
555
# each value is the key:value:node refs tuple
557
yield (self, ) + value
559
# the last thing looked up was a terminal element
560
yield (self, ) + key_dict
563
"""Return an estimate of the number of keys in this index.
565
For GraphIndex the estimate is exact.
567
if self._key_count is None:
568
self._read_and_parse([_HEADER_READV])
569
return self._key_count
571
def _lookup_keys_via_location(self, location_keys):
572
"""Public interface for implementing bisection.
574
If _buffer_all has been called, then all the data for the index is in
575
memory, and this method should not be called, as it uses a separate
576
cache because it cannot pre-resolve all indices, which buffer_all does
579
:param location_keys: A list of location(byte offset), key tuples.
580
:return: A list of (location_key, result) tuples as expected by
581
bzrlib.bisect_multi.bisect_multi_bytes.
583
# Possible improvements:
584
# - only bisect lookup each key once
585
# - sort the keys first, and use that to reduce the bisection window
587
# this progresses in three parts:
590
# attempt to answer the question from the now in memory data.
591
# build the readv request
592
# for each location, ask for 800 bytes - much more than rows we've seen
595
for location, key in location_keys:
596
# can we answer from cache?
597
if self._bisect_nodes and key in self._bisect_nodes:
598
# We have the key parsed.
600
index = self._parsed_key_index(key)
601
if (len(self._parsed_key_map) and
602
self._parsed_key_map[index][0] <= key and
603
(self._parsed_key_map[index][1] >= key or
604
# end of the file has been parsed
605
self._parsed_byte_map[index][1] == self._size)):
606
# the key has been parsed, so no lookup is needed even if its
609
# - if we have examined this part of the file already - yes
610
index = self._parsed_byte_index(location)
611
if (len(self._parsed_byte_map) and
612
self._parsed_byte_map[index][0] <= location and
613
self._parsed_byte_map[index][1] > location):
614
# the byte region has been parsed, so no read is needed.
617
if location + length > self._size:
618
length = self._size - location
619
# todo, trim out parsed locations.
621
readv_ranges.append((location, length))
622
# read the header if needed
623
if self._bisect_nodes is None:
624
readv_ranges.append(_HEADER_READV)
625
self._read_and_parse(readv_ranges)
627
# - figure out <, >, missing, present
628
# - result present references so we can return them.
630
# keys that we cannot answer until we resolve references
631
pending_references = []
632
pending_locations = set()
633
for location, key in location_keys:
634
# can we answer from cache?
635
if key in self._bisect_nodes:
636
# the key has been parsed, so no lookup is needed
637
if self.node_ref_lists:
638
# the references may not have been all parsed.
639
value, refs = self._bisect_nodes[key]
640
wanted_locations = []
641
for ref_list in refs:
643
if ref not in self._keys_by_offset:
644
wanted_locations.append(ref)
646
pending_locations.update(wanted_locations)
647
pending_references.append((location, key))
649
result.append(((location, key), (self, key,
650
value, self._resolve_references(refs))))
652
result.append(((location, key),
653
(self, key, self._bisect_nodes[key])))
656
# has the region the key should be in, been parsed?
657
index = self._parsed_key_index(key)
658
if (self._parsed_key_map[index][0] <= key and
659
(self._parsed_key_map[index][1] >= key or
660
# end of the file has been parsed
661
self._parsed_byte_map[index][1] == self._size)):
662
result.append(((location, key), False))
664
# no, is the key above or below the probed location:
665
# get the range of the probed & parsed location
666
index = self._parsed_byte_index(location)
667
# if the key is below the start of the range, its below
668
if key < self._parsed_key_map[index][0]:
672
result.append(((location, key), direction))
674
# lookup data to resolve references
675
for location in pending_locations:
677
if location + length > self._size:
678
length = self._size - location
679
# TODO: trim out parsed locations (e.g. if the 800 is into the
680
# parsed region trim it, and dont use the adjust_for_latency
683
readv_ranges.append((location, length))
684
self._read_and_parse(readv_ranges)
685
for location, key in pending_references:
686
# answer key references we had to look-up-late.
687
index = self._parsed_key_index(key)
688
value, refs = self._bisect_nodes[key]
689
result.append(((location, key), (self, key,
690
value, self._resolve_references(refs))))
693
def _parse_header_from_bytes(self, bytes):
694
"""Parse the header from a region of bytes.
696
:param bytes: The data to parse.
697
:return: An offset, data tuple such as readv yields, for the unparsed
698
data. (which may length 0).
700
signature = bytes[0:len(self._signature())]
701
if not signature == self._signature():
702
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
703
lines = bytes[len(self._signature()):].splitlines()
704
options_line = lines[0]
705
if not options_line.startswith(_OPTION_NODE_REFS):
706
raise errors.BadIndexOptions(self)
708
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
710
raise errors.BadIndexOptions(self)
711
options_line = lines[1]
712
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
713
raise errors.BadIndexOptions(self)
715
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
717
raise errors.BadIndexOptions(self)
718
options_line = lines[2]
719
if not options_line.startswith(_OPTION_LEN):
720
raise errors.BadIndexOptions(self)
722
self._key_count = int(options_line[len(_OPTION_LEN):])
724
raise errors.BadIndexOptions(self)
725
# calculate the bytes we have processed
726
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
728
self._parsed_bytes(0, None, header_end, None)
729
# setup parsing state
730
self._expected_elements = 3 + self._key_length
731
# raw data keyed by offset
732
self._keys_by_offset = {}
733
# keys with the value and node references
734
self._bisect_nodes = {}
735
return header_end, bytes[header_end:]
737
def _parse_region(self, offset, data):
738
"""Parse node data returned from a readv operation.
740
:param offset: The byte offset the data starts at.
741
:param data: The data to parse.
745
end = offset + len(data)
748
# Trivial test - if the current index's end is within the
749
# low-matching parsed range, we're done.
750
index = self._parsed_byte_index(high_parsed)
751
if end < self._parsed_byte_map[index][1]:
753
# print "[%d:%d]" % (offset, end), \
754
# self._parsed_byte_map[index:index + 2]
755
high_parsed, last_segment = self._parse_segment(
756
offset, data, end, index)
760
def _parse_segment(self, offset, data, end, index):
761
"""Parse one segment of data.
763
:param offset: Where 'data' begins in the file.
764
:param data: Some data to parse a segment of.
765
:param end: Where data ends
766
:param index: The current index into the parsed bytes map.
767
:return: True if the parsed segment is the last possible one in the
769
:return: high_parsed_byte, last_segment.
770
high_parsed_byte is the location of the highest parsed byte in this
771
segment, last_segment is True if the parsed segment is the last
772
possible one in the data block.
774
# default is to use all data
776
# accomodate overlap with data before this.
777
if offset < self._parsed_byte_map[index][1]:
778
# overlaps the lower parsed region
779
# skip the parsed data
780
trim_start = self._parsed_byte_map[index][1] - offset
781
# don't trim the start for \n
782
start_adjacent = True
783
elif offset == self._parsed_byte_map[index][1]:
784
# abuts the lower parsed region
787
# do not trim anything
788
start_adjacent = True
790
# does not overlap the lower parsed region
793
# but trim the leading \n
794
start_adjacent = False
795
if end == self._size:
796
# lines up to the end of all data:
799
# do not strip to the last \n
802
elif index + 1 == len(self._parsed_byte_map):
803
# at the end of the parsed data
806
# but strip to the last \n
809
elif end == self._parsed_byte_map[index + 1][0]:
810
# buts up against the next parsed region
813
# do not strip to the last \n
816
elif end > self._parsed_byte_map[index + 1][0]:
817
# overlaps into the next parsed region
818
# only consider the unparsed data
819
trim_end = self._parsed_byte_map[index + 1][0] - offset
820
# do not strip to the last \n as we know its an entire record
822
last_segment = end < self._parsed_byte_map[index + 1][1]
824
# does not overlap into the next region
827
# but strip to the last \n
830
# now find bytes to discard if needed
831
if not start_adjacent:
832
# work around python bug in rfind
833
if trim_start is None:
834
trim_start = data.find('\n') + 1
836
trim_start = data.find('\n', trim_start) + 1
837
if not (trim_start != 0):
838
raise AssertionError('no \n was present')
839
# print 'removing start', offset, trim_start, repr(data[:trim_start])
841
# work around python bug in rfind
843
trim_end = data.rfind('\n') + 1
845
trim_end = data.rfind('\n', None, trim_end) + 1
846
if not (trim_end != 0):
847
raise AssertionError('no \n was present')
848
# print 'removing end', offset, trim_end, repr(data[trim_end:])
849
# adjust offset and data to the parseable data.
850
trimmed_data = data[trim_start:trim_end]
851
if not (trimmed_data):
852
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
853
% (trim_start, trim_end, offset, offset + len(data)))
856
# print "parsing", repr(trimmed_data)
857
# splitlines mangles the \r delimiters.. don't use it.
858
lines = trimmed_data.split('\n')
861
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
862
for key, value in nodes:
863
self._bisect_nodes[key] = value
864
self._parsed_bytes(offset, first_key,
865
offset + len(trimmed_data), last_key)
866
return offset + len(trimmed_data), last_segment
868
def _parse_lines(self, lines, pos):
877
if not (self._size == pos + 1):
878
raise AssertionError("%s %s" % (self._size, pos))
881
elements = line.split('\0')
882
if len(elements) != self._expected_elements:
883
raise errors.BadIndexData(self)
884
# keys are tuples. Each element is a string that may occur many
885
# times, so we intern them to save space. AB, RC, 200807
886
key = tuple(intern(element) for element in elements[:self._key_length])
887
if first_key is None:
889
absent, references, value = elements[-3:]
891
for ref_string in references.split('\t'):
892
ref_lists.append(tuple([
893
int(ref) for ref in ref_string.split('\r') if ref
895
ref_lists = tuple(ref_lists)
896
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
897
pos += len(line) + 1 # +1 for the \n
900
if self.node_ref_lists:
901
node_value = (value, ref_lists)
904
nodes.append((key, node_value))
905
# print "parsed ", key
906
return first_key, key, nodes, trailers
908
def _parsed_bytes(self, start, start_key, end, end_key):
909
"""Mark the bytes from start to end as parsed.
911
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
914
:param start: The start of the parsed region.
915
:param end: The end of the parsed region.
917
index = self._parsed_byte_index(start)
918
new_value = (start, end)
919
new_key = (start_key, end_key)
921
# first range parsed is always the beginning.
922
self._parsed_byte_map.insert(index, new_value)
923
self._parsed_key_map.insert(index, new_key)
927
# extend lower region
928
# extend higher region
929
# combine two regions
930
if (index + 1 < len(self._parsed_byte_map) and
931
self._parsed_byte_map[index][1] == start and
932
self._parsed_byte_map[index + 1][0] == end):
933
# combine two regions
934
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
935
self._parsed_byte_map[index + 1][1])
936
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
937
self._parsed_key_map[index + 1][1])
938
del self._parsed_byte_map[index + 1]
939
del self._parsed_key_map[index + 1]
940
elif self._parsed_byte_map[index][1] == start:
941
# extend the lower entry
942
self._parsed_byte_map[index] = (
943
self._parsed_byte_map[index][0], end)
944
self._parsed_key_map[index] = (
945
self._parsed_key_map[index][0], end_key)
946
elif (index + 1 < len(self._parsed_byte_map) and
947
self._parsed_byte_map[index + 1][0] == end):
948
# extend the higher entry
949
self._parsed_byte_map[index + 1] = (
950
start, self._parsed_byte_map[index + 1][1])
951
self._parsed_key_map[index + 1] = (
952
start_key, self._parsed_key_map[index + 1][1])
955
self._parsed_byte_map.insert(index + 1, new_value)
956
self._parsed_key_map.insert(index + 1, new_key)
958
def _read_and_parse(self, readv_ranges):
959
"""Read the the ranges and parse the resulting data.
961
:param readv_ranges: A prepared readv range list.
964
readv_data = self._transport.readv(self._name, readv_ranges, True,
967
for offset, data in readv_data:
968
if self._bisect_nodes is None:
969
# this must be the start
970
if not (offset == 0):
971
raise AssertionError()
972
offset, data = self._parse_header_from_bytes(data)
973
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
974
self._parse_region(offset, data)
976
def _signature(self):
977
"""The file signature for this index type."""
981
"""Validate that everything in the index can be accessed."""
982
# iter_all validates completely at the moment, so just do that.
983
for node in self.iter_all_entries():
987
class CombinedGraphIndex(object):
988
"""A GraphIndex made up from smaller GraphIndices.
990
The backing indices must implement GraphIndex, and are presumed to be
993
Queries against the combined index will be made against the first index,
994
and then the second and so on. The order of index's can thus influence
995
performance significantly. For example, if one index is on local disk and a
996
second on a remote server, the local disk index should be before the other
1000
def __init__(self, indices):
1001
"""Create a CombinedGraphIndex backed by indices.
1003
:param indices: An ordered list of indices to query for data.
1005
self._indices = indices
1009
self.__class__.__name__,
1010
', '.join(map(repr, self._indices)))
1012
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1013
def get_parents(self, revision_ids):
1014
"""See graph._StackedParentsProvider.get_parents.
1016
This implementation thunks the graph.Graph.get_parents api across to
1019
:param revision_ids: An iterable of graph keys for this graph.
1020
:return: A list of parent details for each key in revision_ids.
1021
Each parent details will be one of:
1022
* None when the key was missing
1023
* (NULL_REVISION,) when the key has no parents.
1024
* (parent_key, parent_key...) otherwise.
1026
parent_map = self.get_parent_map(revision_ids)
1027
return [parent_map.get(r, None) for r in revision_ids]
1029
def get_parent_map(self, keys):
1030
"""See graph._StackedParentsProvider.get_parent_map"""
1031
search_keys = set(keys)
1032
if NULL_REVISION in search_keys:
1033
search_keys.discard(NULL_REVISION)
1034
found_parents = {NULL_REVISION:[]}
1037
for index, key, value, refs in self.iter_entries(search_keys):
1040
parents = (NULL_REVISION,)
1041
found_parents[key] = parents
1042
return found_parents
1044
def insert_index(self, pos, index):
1045
"""Insert a new index in the list of indices to query.
1047
:param pos: The position to insert the index.
1048
:param index: The index to insert.
1050
self._indices.insert(pos, index)
1052
def iter_all_entries(self):
1053
"""Iterate over all keys within the index
1055
Duplicate keys across child indices are presumed to have the same
1056
value and are only reported once.
1058
:return: An iterable of (index, key, reference_lists, value).
1059
There is no defined order for the result iteration - it will be in
1060
the most efficient order for the index.
1063
for index in self._indices:
1064
for node in index.iter_all_entries():
1065
if node[1] not in seen_keys:
1067
seen_keys.add(node[1])
1069
def iter_entries(self, keys):
1070
"""Iterate over keys within the index.
1072
Duplicate keys across child indices are presumed to have the same
1073
value and are only reported once.
1075
:param keys: An iterable providing the keys to be retrieved.
1076
:return: An iterable of (index, key, reference_lists, value). There is no
1077
defined order for the result iteration - it will be in the most
1078
efficient order for the index.
1081
for index in self._indices:
1084
for node in index.iter_entries(keys):
1085
keys.remove(node[1])
1088
def iter_entries_prefix(self, keys):
1089
"""Iterate over keys within the index using prefix matching.
1091
Duplicate keys across child indices are presumed to have the same
1092
value and are only reported once.
1094
Prefix matching is applied within the tuple of a key, not to within
1095
the bytestring of each key element. e.g. if you have the keys ('foo',
1096
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1097
only the former key is returned.
1099
:param keys: An iterable providing the key prefixes to be retrieved.
1100
Each key prefix takes the form of a tuple the length of a key, but
1101
with the last N elements 'None' rather than a regular bytestring.
1102
The first element cannot be 'None'.
1103
:return: An iterable as per iter_all_entries, but restricted to the
1104
keys with a matching prefix to those supplied. No additional keys
1105
will be returned, and every match that is in the index will be
1112
for index in self._indices:
1113
for node in index.iter_entries_prefix(keys):
1114
if node[1] in seen_keys:
1116
seen_keys.add(node[1])
1119
def key_count(self):
1120
"""Return an estimate of the number of keys in this index.
1122
For CombinedGraphIndex this is approximated by the sum of the keys of
1123
the child indices. As child indices may have duplicate keys this can
1124
have a maximum error of the number of child indices * largest number of
1127
return sum((index.key_count() for index in self._indices), 0)
1130
"""Validate that everything in the index can be accessed."""
1131
for index in self._indices:
1135
class InMemoryGraphIndex(GraphIndexBuilder):
1136
"""A GraphIndex which operates entirely out of memory and is mutable.
1138
This is designed to allow the accumulation of GraphIndex entries during a
1139
single write operation, where the accumulated entries need to be immediately
1140
available - for example via a CombinedGraphIndex.
1143
def add_nodes(self, nodes):
1144
"""Add nodes to the index.
1146
:param nodes: An iterable of (key, node_refs, value) entries to add.
1148
if self.reference_lists:
1149
for (key, value, node_refs) in nodes:
1150
self.add_node(key, value, node_refs)
1152
for (key, value) in nodes:
1153
self.add_node(key, value)
1155
def iter_all_entries(self):
1156
"""Iterate over all keys within the index
1158
:return: An iterable of (index, key, reference_lists, value). There is no
1159
defined order for the result iteration - it will be in the most
1160
efficient order for the index (in this case dictionary hash order).
1162
if 'evil' in debug.debug_flags:
1163
trace.mutter_callsite(3,
1164
"iter_all_entries scales with size of history.")
1165
if self.reference_lists:
1166
for key, (absent, references, value) in self._nodes.iteritems():
1168
yield self, key, value, references
1170
for key, (absent, references, value) in self._nodes.iteritems():
1172
yield self, key, value
1174
def iter_entries(self, keys):
1175
"""Iterate over keys within the index.
1177
:param keys: An iterable providing the keys to be retrieved.
1178
:return: An iterable of (index, key, value, reference_lists). There is no
1179
defined order for the result iteration - it will be in the most
1180
efficient order for the index (keys iteration order in this case).
1183
if self.reference_lists:
1184
for key in keys.intersection(self._keys):
1185
node = self._nodes[key]
1187
yield self, key, node[2], node[1]
1189
for key in keys.intersection(self._keys):
1190
node = self._nodes[key]
1192
yield self, key, node[2]
1194
def iter_entries_prefix(self, keys):
1195
"""Iterate over keys within the index using prefix matching.
1197
Prefix matching is applied within the tuple of a key, not to within
1198
the bytestring of each key element. e.g. if you have the keys ('foo',
1199
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1200
only the former key is returned.
1202
:param keys: An iterable providing the key prefixes to be retrieved.
1203
Each key prefix takes the form of a tuple the length of a key, but
1204
with the last N elements 'None' rather than a regular bytestring.
1205
The first element cannot be 'None'.
1206
:return: An iterable as per iter_all_entries, but restricted to the
1207
keys with a matching prefix to those supplied. No additional keys
1208
will be returned, and every match that is in the index will be
1211
# XXX: To much duplication with the GraphIndex class; consider finding
1212
# a good place to pull out the actual common logic.
1216
if self._key_length == 1:
1220
raise errors.BadIndexKey(key)
1221
if len(key) != self._key_length:
1222
raise errors.BadIndexKey(key)
1223
node = self._nodes[key]
1226
if self.reference_lists:
1227
yield self, key, node[2], node[1]
1229
yield self, key, node[2]
1234
raise errors.BadIndexKey(key)
1235
if len(key) != self._key_length:
1236
raise errors.BadIndexKey(key)
1237
# find what it refers to:
1238
key_dict = self._nodes_by_key
1239
elements = list(key)
1240
# find the subdict to return
1242
while len(elements) and elements[0] is not None:
1243
key_dict = key_dict[elements[0]]
1246
# a non-existant lookup.
1251
key_dict = dicts.pop(-1)
1252
# can't be empty or would not exist
1253
item, value = key_dict.iteritems().next()
1254
if type(value) == dict:
1256
dicts.extend(key_dict.itervalues())
1259
for value in key_dict.itervalues():
1260
yield (self, ) + value
1262
yield (self, ) + key_dict
1264
def key_count(self):
1265
"""Return an estimate of the number of keys in this index.
1267
For InMemoryGraphIndex the estimate is exact.
1269
return len(self._keys)
1272
"""In memory index's have no known corruption at the moment."""
1275
class GraphIndexPrefixAdapter(object):
1276
"""An adapter between GraphIndex with different key lengths.
1278
Queries against this will emit queries against the adapted Graph with the
1279
prefix added, queries for all items use iter_entries_prefix. The returned
1280
nodes will have their keys and node references adjusted to remove the
1281
prefix. Finally, an add_nodes_callback can be supplied - when called the
1282
nodes and references being added will have prefix prepended.
1285
def __init__(self, adapted, prefix, missing_key_length,
1286
add_nodes_callback=None):
1287
"""Construct an adapter against adapted with prefix."""
1288
self.adapted = adapted
1289
self.prefix_key = prefix + (None,)*missing_key_length
1290
self.prefix = prefix
1291
self.prefix_len = len(prefix)
1292
self.add_nodes_callback = add_nodes_callback
1294
def add_nodes(self, nodes):
1295
"""Add nodes to the index.
1297
:param nodes: An iterable of (key, node_refs, value) entries to add.
1299
# save nodes in case its an iterator
1300
nodes = tuple(nodes)
1301
translated_nodes = []
1303
# Add prefix_key to each reference node_refs is a tuple of tuples,
1304
# so split it apart, and add prefix_key to the internal reference
1305
for (key, value, node_refs) in nodes:
1306
adjusted_references = (
1307
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1308
for ref_list in node_refs))
1309
translated_nodes.append((self.prefix + key, value,
1310
adjusted_references))
1312
# XXX: TODO add an explicit interface for getting the reference list
1313
# status, to handle this bit of user-friendliness in the API more
1315
for (key, value) in nodes:
1316
translated_nodes.append((self.prefix + key, value))
1317
self.add_nodes_callback(translated_nodes)
1319
def add_node(self, key, value, references=()):
1320
"""Add a node to the index.
1322
:param key: The key. keys are non-empty tuples containing
1323
as many whitespace-free utf8 bytestrings as the key length
1324
defined for this index.
1325
:param references: An iterable of iterables of keys. Each is a
1326
reference to another key.
1327
:param value: The value to associate with the key. It may be any
1328
bytes as long as it does not contain \0 or \n.
1330
self.add_nodes(((key, value, references), ))
1332
def _strip_prefix(self, an_iter):
1333
"""Strip prefix data from nodes and return it."""
1334
for node in an_iter:
1336
if node[1][:self.prefix_len] != self.prefix:
1337
raise errors.BadIndexData(self)
1338
for ref_list in node[3]:
1339
for ref_node in ref_list:
1340
if ref_node[:self.prefix_len] != self.prefix:
1341
raise errors.BadIndexData(self)
1342
yield node[0], node[1][self.prefix_len:], node[2], (
1343
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1344
for ref_list in node[3]))
1346
def iter_all_entries(self):
1347
"""Iterate over all keys within the index
1349
iter_all_entries is implemented against the adapted index using
1350
iter_entries_prefix.
1352
:return: An iterable of (index, key, reference_lists, value). There is no
1353
defined order for the result iteration - it will be in the most
1354
efficient order for the index (in this case dictionary hash order).
1356
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1358
def iter_entries(self, keys):
1359
"""Iterate over keys within the index.
1361
:param keys: An iterable providing the keys to be retrieved.
1362
:return: An iterable of (index, key, value, reference_lists). There is no
1363
defined order for the result iteration - it will be in the most
1364
efficient order for the index (keys iteration order in this case).
1366
return self._strip_prefix(self.adapted.iter_entries(
1367
self.prefix + key for key in keys))
1369
def iter_entries_prefix(self, keys):
1370
"""Iterate over keys within the index using prefix matching.
1372
Prefix matching is applied within the tuple of a key, not to within
1373
the bytestring of each key element. e.g. if you have the keys ('foo',
1374
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1375
only the former key is returned.
1377
:param keys: An iterable providing the key prefixes to be retrieved.
1378
Each key prefix takes the form of a tuple the length of a key, but
1379
with the last N elements 'None' rather than a regular bytestring.
1380
The first element cannot be 'None'.
1381
:return: An iterable as per iter_all_entries, but restricted to the
1382
keys with a matching prefix to those supplied. No additional keys
1383
will be returned, and every match that is in the index will be
1386
return self._strip_prefix(self.adapted.iter_entries_prefix(
1387
self.prefix + key for key in keys))
1389
def key_count(self):
1390
"""Return an estimate of the number of keys in this index.
1392
For GraphIndexPrefixAdapter this is relatively expensive - key
1393
iteration with the prefix is done.
1395
return len(list(self.iter_all_entries()))
1398
"""Call the adapted's validate."""
1399
self.adapted.validate()
added
removed
bzrlib/index.py
