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
83
# A dict of {key: (absent, ref_lists, value)}
85
self._nodes_by_key = None
86
self._key_length = key_elements
87
self._optimize_for_size = False
89
def _check_key(self, key):
90
"""Raise BadIndexKey if key is not a valid key for this index."""
91
if type(key) != tuple:
92
raise errors.BadIndexKey(key)
93
if self._key_length != len(key):
94
raise errors.BadIndexKey(key)
96
if not element or _whitespace_re.search(element) is not None:
97
raise errors.BadIndexKey(element)
99
def _get_nodes_by_key(self):
100
if self._nodes_by_key is None:
102
if self.reference_lists:
103
for key, (absent, references, value) in self._nodes.iteritems():
106
key_dict = nodes_by_key
107
for subkey in key[:-1]:
108
key_dict = key_dict.setdefault(subkey, {})
109
key_dict[key[-1]] = key, value, references
111
for key, (absent, references, value) in self._nodes.iteritems():
114
key_dict = nodes_by_key
115
for subkey in key[:-1]:
116
key_dict = key_dict.setdefault(subkey, {})
117
key_dict[key[-1]] = key, value
118
self._nodes_by_key = nodes_by_key
119
return self._nodes_by_key
121
def _update_nodes_by_key(self, key, value, node_refs):
122
"""Update the _nodes_by_key dict with a new key.
124
For a key of (foo, bar, baz) create
125
_nodes_by_key[foo][bar][baz] = key_value
127
if self._nodes_by_key is None:
129
key_dict = self._nodes_by_key
130
if self.reference_lists:
131
key_value = key, value, node_refs
133
key_value = key, value
134
for subkey in key[:-1]:
135
key_dict = key_dict.setdefault(subkey, {})
136
key_dict[key[-1]] = key_value
138
def _check_key_ref_value(self, key, references, value):
139
"""Check that 'key' and 'references' are all valid.
141
:param key: A key tuple. Must conform to the key interface (be a tuple,
142
be of the right length, not have any whitespace or nulls in any key
144
:param references: An iterable of reference lists. Something like
145
[[(ref, key)], [(ref, key), (other, key)]]
146
:param value: The value associate with this key. Must not contain
147
newlines or null characters.
148
:return: (node_refs, absent_references)
149
node_refs basically a packed form of 'references' where all
151
absent_references reference keys that are not in self._nodes.
152
This may contain duplicates if the same key is
153
referenced in multiple lists.
156
if _newline_null_re.search(value) is not None:
157
raise errors.BadIndexValue(value)
158
if len(references) != self.reference_lists:
159
raise errors.BadIndexValue(references)
161
absent_references = []
162
for reference_list in references:
163
for reference in reference_list:
164
# If reference *is* in self._nodes, then we know it has already
166
if reference not in self._nodes:
167
self._check_key(reference)
168
absent_references.append(reference)
169
node_refs.append(tuple(reference_list))
170
return tuple(node_refs), absent_references
172
def add_node(self, key, value, references=()):
173
"""Add a node to the index.
175
:param key: The key. keys are non-empty tuples containing
176
as many whitespace-free utf8 bytestrings as the key length
177
defined for this index.
178
:param references: An iterable of iterables of keys. Each is a
179
reference to another key.
180
:param value: The value to associate with the key. It may be any
181
bytes as long as it does not contain \0 or \n.
184
absent_references) = self._check_key_ref_value(key, references, value)
185
if key in self._nodes and self._nodes[key][0] != 'a':
186
raise errors.BadIndexDuplicateKey(key, self)
187
for reference in absent_references:
188
# There may be duplicates, but I don't think it is worth worrying
190
self._nodes[reference] = ('a', (), '')
191
self._nodes[key] = ('', node_refs, value)
193
if self._nodes_by_key is not None and self._key_length > 1:
194
self._update_nodes_by_key(key, value, node_refs)
198
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
199
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
200
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
201
prefix_length = sum(len(x) for x in lines)
202
# references are byte offsets. To avoid having to do nasty
203
# polynomial work to resolve offsets (references to later in the
204
# file cannot be determined until all the inbetween references have
205
# been calculated too) we pad the offsets with 0's to make them be
206
# of consistent length. Using binary offsets would break the trivial
208
# to calculate the width of zero's needed we do three passes:
209
# one to gather all the non-reference data and the number of references.
210
# one to pad all the data with reference-length and determine entry
214
# forward sorted by key. In future we may consider topological sorting,
215
# at the cost of table scans for direct lookup, or a second index for
217
nodes = sorted(self._nodes.items())
218
# if we do not prepass, we don't know how long it will be up front.
219
expected_bytes = None
220
# we only need to pre-pass if we have reference lists at all.
221
if self.reference_lists:
223
non_ref_bytes = prefix_length
225
# TODO use simple multiplication for the constants in this loop.
226
for key, (absent, references, value) in nodes:
227
# record the offset known *so far* for this key:
228
# the non reference bytes to date, and the total references to
229
# date - saves reaccumulating on the second pass
230
key_offset_info.append((key, non_ref_bytes, total_references))
231
# key is literal, value is literal, there are 3 null's, 1 NL
232
# key is variable length tuple, \x00 between elements
233
non_ref_bytes += sum(len(element) for element in key)
234
if self._key_length > 1:
235
non_ref_bytes += self._key_length - 1
236
# value is literal bytes, there are 3 null's, 1 NL.
237
non_ref_bytes += len(value) + 3 + 1
238
# one byte for absent if set.
241
elif self.reference_lists:
242
# (ref_lists -1) tabs
243
non_ref_bytes += self.reference_lists - 1
244
# (ref-1 cr's per ref_list)
245
for ref_list in references:
246
# how many references across the whole file?
247
total_references += len(ref_list)
248
# accrue reference separators
250
non_ref_bytes += len(ref_list) - 1
251
# how many digits are needed to represent the total byte count?
253
possible_total_bytes = non_ref_bytes + total_references*digits
254
while 10 ** digits < possible_total_bytes:
256
possible_total_bytes = non_ref_bytes + total_references*digits
257
expected_bytes = possible_total_bytes + 1 # terminating newline
258
# resolve key addresses.
260
for key, non_ref_bytes, total_references in key_offset_info:
261
key_addresses[key] = non_ref_bytes + total_references*digits
263
format_string = '%%0%sd' % digits
264
for key, (absent, references, value) in nodes:
265
flattened_references = []
266
for ref_list in references:
268
for reference in ref_list:
269
ref_addresses.append(format_string % key_addresses[reference])
270
flattened_references.append('\r'.join(ref_addresses))
271
string_key = '\x00'.join(key)
272
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
273
'\t'.join(flattened_references), value))
275
result = StringIO(''.join(lines))
276
if expected_bytes and len(result.getvalue()) != expected_bytes:
277
raise errors.BzrError('Failed index creation. Internal error:'
278
' mismatched output length and expected length: %d %d' %
279
(len(result.getvalue()), expected_bytes))
282
def set_optimize(self, for_size=True):
283
"""Change how the builder tries to optimize the result.
285
:param for_size: Tell the builder to try and make the index as small as
289
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
291
self._optimize_for_size = for_size
294
class GraphIndex(object):
295
"""An index for data with embedded graphs.
297
The index maps keys to a list of key reference lists, and a value.
298
Each node has the same number of key reference lists. Each key reference
299
list can be empty or an arbitrary length. The value is an opaque NULL
300
terminated string without any newlines. The storage of the index is
301
hidden in the interface: keys and key references are always tuples of
302
bytestrings, never the internal representation (e.g. dictionary offsets).
304
It is presumed that the index will not be mutated - it is static data.
306
Successive iter_all_entries calls will read the entire index each time.
307
Additionally, iter_entries calls will read the index linearly until the
308
desired keys are found. XXX: This must be fixed before the index is
309
suitable for production use. :XXX
312
def __init__(self, transport, name, size):
313
"""Open an index called name on transport.
315
:param transport: A bzrlib.transport.Transport.
316
:param name: A path to provide to transport API calls.
317
:param size: The size of the index in bytes. This is used for bisection
318
logic to perform partial index reads. While the size could be
319
obtained by statting the file this introduced an additional round
320
trip as well as requiring stat'able transports, both of which are
321
avoided by having it supplied. If size is None, then bisection
322
support will be disabled and accessing the index will just stream
325
self._transport = transport
327
# Becomes a dict of key:(value, reference-list-byte-locations) used by
328
# the bisection interface to store parsed but not resolved keys.
329
self._bisect_nodes = None
330
# Becomes a dict of key:(value, reference-list-keys) which are ready to
331
# be returned directly to callers.
333
# a sorted list of slice-addresses for the parsed bytes of the file.
334
# e.g. (0,1) would mean that byte 0 is parsed.
335
self._parsed_byte_map = []
336
# a sorted list of keys matching each slice address for parsed bytes
337
# e.g. (None, 'foo@bar') would mean that the first byte contained no
338
# key, and the end byte of the slice is the of the data for 'foo@bar'
339
self._parsed_key_map = []
340
self._key_count = None
341
self._keys_by_offset = None
342
self._nodes_by_key = None
344
# The number of bytes we've read so far in trying to process this file
347
def __eq__(self, other):
348
"""Equal when self and other were created with the same parameters."""
350
type(self) == type(other) and
351
self._transport == other._transport and
352
self._name == other._name and
353
self._size == other._size)
355
def __ne__(self, other):
356
return not self.__eq__(other)
359
return "%s(%r)" % (self.__class__.__name__,
360
self._transport.abspath(self._name))
362
def _buffer_all(self, stream=None):
363
"""Buffer all the index data.
365
Mutates self._nodes and self.keys_by_offset.
367
if self._nodes is not None:
368
# We already did this
370
if 'index' in debug.debug_flags:
371
mutter('Reading entire index %s', self._transport.abspath(self._name))
373
stream = self._transport.get(self._name)
374
self._read_prefix(stream)
375
self._expected_elements = 3 + self._key_length
377
# raw data keyed by offset
378
self._keys_by_offset = {}
379
# ready-to-return key:value or key:value, node_ref_lists
381
self._nodes_by_key = None
384
lines = stream.read().split('\n')
386
_, _, _, trailers = self._parse_lines(lines, pos)
387
for key, absent, references, value in self._keys_by_offset.itervalues():
390
# resolve references:
391
if self.node_ref_lists:
392
node_value = (value, self._resolve_references(references))
395
self._nodes[key] = node_value
396
# cache the keys for quick set intersections
397
self._keys = set(self._nodes)
399
# there must be one line - the empty trailer line.
400
raise errors.BadIndexData(self)
402
def _get_nodes_by_key(self):
403
if self._nodes_by_key is None:
405
if self.node_ref_lists:
406
for key, (value, references) in self._nodes.iteritems():
407
key_dict = nodes_by_key
408
for subkey in key[:-1]:
409
key_dict = key_dict.setdefault(subkey, {})
410
key_dict[key[-1]] = key, value, references
412
for key, value in self._nodes.iteritems():
413
key_dict = nodes_by_key
414
for subkey in key[:-1]:
415
key_dict = key_dict.setdefault(subkey, {})
416
key_dict[key[-1]] = key, value
417
self._nodes_by_key = nodes_by_key
418
return self._nodes_by_key
420
def iter_all_entries(self):
421
"""Iterate over all keys within the index.
423
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
424
The former tuple is used when there are no reference lists in the
425
index, making the API compatible with simple key:value index types.
426
There is no defined order for the result iteration - it will be in
427
the most efficient order for the index.
429
if 'evil' in debug.debug_flags:
430
trace.mutter_callsite(3,
431
"iter_all_entries scales with size of history.")
432
if self._nodes is None:
434
if self.node_ref_lists:
435
for key, (value, node_ref_lists) in self._nodes.iteritems():
436
yield self, key, value, node_ref_lists
438
for key, value in self._nodes.iteritems():
439
yield self, key, value
441
def _read_prefix(self, stream):
442
signature = stream.read(len(self._signature()))
443
if not signature == self._signature():
444
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
445
options_line = stream.readline()
446
if not options_line.startswith(_OPTION_NODE_REFS):
447
raise errors.BadIndexOptions(self)
449
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
451
raise errors.BadIndexOptions(self)
452
options_line = stream.readline()
453
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
454
raise errors.BadIndexOptions(self)
456
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
458
raise errors.BadIndexOptions(self)
459
options_line = stream.readline()
460
if not options_line.startswith(_OPTION_LEN):
461
raise errors.BadIndexOptions(self)
463
self._key_count = int(options_line[len(_OPTION_LEN):-1])
465
raise errors.BadIndexOptions(self)
467
def _resolve_references(self, references):
468
"""Return the resolved key references for references.
470
References are resolved by looking up the location of the key in the
471
_keys_by_offset map and substituting the key name, preserving ordering.
473
:param references: An iterable of iterables of key locations. e.g.
475
:return: A tuple of tuples of keys.
478
for ref_list in references:
479
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
480
return tuple(node_refs)
482
def _find_index(self, range_map, key):
483
"""Helper for the _parsed_*_index calls.
485
Given a range map - [(start, end), ...], finds the index of the range
486
in the map for key if it is in the map, and if it is not there, the
487
immediately preceeding range in the map.
489
result = bisect_right(range_map, key) - 1
490
if result + 1 < len(range_map):
491
# check the border condition, it may be in result + 1
492
if range_map[result + 1][0] == key[0]:
496
def _parsed_byte_index(self, offset):
497
"""Return the index of the entry immediately before offset.
499
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
500
there is one unparsed byte (the 11th, addressed as[10]). then:
501
asking for 0 will return 0
502
asking for 10 will return 0
503
asking for 11 will return 1
504
asking for 12 will return 1
507
return self._find_index(self._parsed_byte_map, key)
509
def _parsed_key_index(self, key):
510
"""Return the index of the entry immediately before key.
512
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
513
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
514
have been parsed, then:
515
asking for '' will return 0
516
asking for 'a' will return 0
517
asking for 'b' will return 1
518
asking for 'e' will return 1
520
search_key = (key, None)
521
return self._find_index(self._parsed_key_map, search_key)
523
def _is_parsed(self, offset):
524
"""Returns True if offset has been parsed."""
525
index = self._parsed_byte_index(offset)
526
if index == len(self._parsed_byte_map):
527
return offset < self._parsed_byte_map[index - 1][1]
528
start, end = self._parsed_byte_map[index]
529
return offset >= start and offset < end
531
def _iter_entries_from_total_buffer(self, keys):
532
"""Iterate over keys when the entire index is parsed."""
533
keys = keys.intersection(self._keys)
534
if self.node_ref_lists:
536
value, node_refs = self._nodes[key]
537
yield self, key, value, node_refs
540
yield self, key, self._nodes[key]
542
def iter_entries(self, keys):
543
"""Iterate over keys within the index.
545
:param keys: An iterable providing the keys to be retrieved.
546
:return: An iterable as per iter_all_entries, but restricted to the
547
keys supplied. No additional keys will be returned, and every
548
key supplied that is in the index will be returned.
553
if self._size is None and self._nodes is None:
556
# We fit about 20 keys per minimum-read (4K), so if we are looking for
557
# more than 1/20th of the index its likely (assuming homogenous key
558
# spread) that we'll read the entire index. If we're going to do that,
559
# buffer the whole thing. A better analysis might take key spread into
560
# account - but B+Tree indices are better anyway.
561
# We could look at all data read, and use a threshold there, which will
562
# trigger on ancestry walks, but that is not yet fully mapped out.
563
if self._nodes is None and len(keys) * 20 > self.key_count():
565
if self._nodes is not None:
566
return self._iter_entries_from_total_buffer(keys)
568
return (result[1] for result in bisect_multi_bytes(
569
self._lookup_keys_via_location, self._size, keys))
571
def iter_entries_prefix(self, keys):
572
"""Iterate over keys within the index using prefix matching.
574
Prefix matching is applied within the tuple of a key, not to within
575
the bytestring of each key element. e.g. if you have the keys ('foo',
576
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
577
only the former key is returned.
579
WARNING: Note that this method currently causes a full index parse
580
unconditionally (which is reasonably appropriate as it is a means for
581
thunking many small indices into one larger one and still supplies
582
iter_all_entries at the thunk layer).
584
:param keys: An iterable providing the key prefixes to be retrieved.
585
Each key prefix takes the form of a tuple the length of a key, but
586
with the last N elements 'None' rather than a regular bytestring.
587
The first element cannot be 'None'.
588
:return: An iterable as per iter_all_entries, but restricted to the
589
keys with a matching prefix to those supplied. No additional keys
590
will be returned, and every match that is in the index will be
596
# load data - also finds key lengths
597
if self._nodes is None:
599
if self._key_length == 1:
603
raise errors.BadIndexKey(key)
604
if len(key) != self._key_length:
605
raise errors.BadIndexKey(key)
606
if self.node_ref_lists:
607
value, node_refs = self._nodes[key]
608
yield self, key, value, node_refs
610
yield self, key, self._nodes[key]
612
nodes_by_key = self._get_nodes_by_key()
616
raise errors.BadIndexKey(key)
617
if len(key) != self._key_length:
618
raise errors.BadIndexKey(key)
619
# find what it refers to:
620
key_dict = nodes_by_key
622
# find the subdict whose contents should be returned.
624
while len(elements) and elements[0] is not None:
625
key_dict = key_dict[elements[0]]
628
# a non-existant lookup.
633
key_dict = dicts.pop(-1)
634
# can't be empty or would not exist
635
item, value = key_dict.iteritems().next()
636
if type(value) == dict:
638
dicts.extend(key_dict.itervalues())
641
for value in key_dict.itervalues():
642
# each value is the key:value:node refs tuple
644
yield (self, ) + value
646
# the last thing looked up was a terminal element
647
yield (self, ) + key_dict
650
"""Return an estimate of the number of keys in this index.
652
For GraphIndex the estimate is exact.
654
if self._key_count is None:
655
self._read_and_parse([_HEADER_READV])
656
return self._key_count
658
def _lookup_keys_via_location(self, location_keys):
659
"""Public interface for implementing bisection.
661
If _buffer_all has been called, then all the data for the index is in
662
memory, and this method should not be called, as it uses a separate
663
cache because it cannot pre-resolve all indices, which buffer_all does
666
:param location_keys: A list of location(byte offset), key tuples.
667
:return: A list of (location_key, result) tuples as expected by
668
bzrlib.bisect_multi.bisect_multi_bytes.
670
# Possible improvements:
671
# - only bisect lookup each key once
672
# - sort the keys first, and use that to reduce the bisection window
674
# this progresses in three parts:
677
# attempt to answer the question from the now in memory data.
678
# build the readv request
679
# for each location, ask for 800 bytes - much more than rows we've seen
682
for location, key in location_keys:
683
# can we answer from cache?
684
if self._bisect_nodes and key in self._bisect_nodes:
685
# We have the key parsed.
687
index = self._parsed_key_index(key)
688
if (len(self._parsed_key_map) and
689
self._parsed_key_map[index][0] <= key and
690
(self._parsed_key_map[index][1] >= key or
691
# end of the file has been parsed
692
self._parsed_byte_map[index][1] == self._size)):
693
# the key has been parsed, so no lookup is needed even if its
696
# - if we have examined this part of the file already - yes
697
index = self._parsed_byte_index(location)
698
if (len(self._parsed_byte_map) and
699
self._parsed_byte_map[index][0] <= location and
700
self._parsed_byte_map[index][1] > location):
701
# the byte region has been parsed, so no read is needed.
704
if location + length > self._size:
705
length = self._size - location
706
# todo, trim out parsed locations.
708
readv_ranges.append((location, length))
709
# read the header if needed
710
if self._bisect_nodes is None:
711
readv_ranges.append(_HEADER_READV)
712
self._read_and_parse(readv_ranges)
714
if self._nodes is not None:
715
# _read_and_parse triggered a _buffer_all because we requested the
717
for location, key in location_keys:
718
if key not in self._nodes: # not present
719
result.append(((location, key), False))
720
elif self.node_ref_lists:
721
value, refs = self._nodes[key]
722
result.append(((location, key),
723
(self, key, value, refs)))
725
result.append(((location, key),
726
(self, key, self._nodes[key])))
729
# - figure out <, >, missing, present
730
# - result present references so we can return them.
731
# keys that we cannot answer until we resolve references
732
pending_references = []
733
pending_locations = set()
734
for location, key in location_keys:
735
# can we answer from cache?
736
if key in self._bisect_nodes:
737
# the key has been parsed, so no lookup is needed
738
if self.node_ref_lists:
739
# the references may not have been all parsed.
740
value, refs = self._bisect_nodes[key]
741
wanted_locations = []
742
for ref_list in refs:
744
if ref not in self._keys_by_offset:
745
wanted_locations.append(ref)
747
pending_locations.update(wanted_locations)
748
pending_references.append((location, key))
750
result.append(((location, key), (self, key,
751
value, self._resolve_references(refs))))
753
result.append(((location, key),
754
(self, key, self._bisect_nodes[key])))
757
# has the region the key should be in, been parsed?
758
index = self._parsed_key_index(key)
759
if (self._parsed_key_map[index][0] <= key and
760
(self._parsed_key_map[index][1] >= key or
761
# end of the file has been parsed
762
self._parsed_byte_map[index][1] == self._size)):
763
result.append(((location, key), False))
765
# no, is the key above or below the probed location:
766
# get the range of the probed & parsed location
767
index = self._parsed_byte_index(location)
768
# if the key is below the start of the range, its below
769
if key < self._parsed_key_map[index][0]:
773
result.append(((location, key), direction))
775
# lookup data to resolve references
776
for location in pending_locations:
778
if location + length > self._size:
779
length = self._size - location
780
# TODO: trim out parsed locations (e.g. if the 800 is into the
781
# parsed region trim it, and dont use the adjust_for_latency
784
readv_ranges.append((location, length))
785
self._read_and_parse(readv_ranges)
786
if self._nodes is not None:
787
# The _read_and_parse triggered a _buffer_all, grab the data and
789
for location, key in pending_references:
790
value, refs = self._nodes[key]
791
result.append(((location, key), (self, key, value, refs)))
793
for location, key in pending_references:
794
# answer key references we had to look-up-late.
795
value, refs = self._bisect_nodes[key]
796
result.append(((location, key), (self, key,
797
value, self._resolve_references(refs))))
800
def _parse_header_from_bytes(self, bytes):
801
"""Parse the header from a region of bytes.
803
:param bytes: The data to parse.
804
:return: An offset, data tuple such as readv yields, for the unparsed
805
data. (which may length 0).
807
signature = bytes[0:len(self._signature())]
808
if not signature == self._signature():
809
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
810
lines = bytes[len(self._signature()):].splitlines()
811
options_line = lines[0]
812
if not options_line.startswith(_OPTION_NODE_REFS):
813
raise errors.BadIndexOptions(self)
815
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
817
raise errors.BadIndexOptions(self)
818
options_line = lines[1]
819
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
820
raise errors.BadIndexOptions(self)
822
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
824
raise errors.BadIndexOptions(self)
825
options_line = lines[2]
826
if not options_line.startswith(_OPTION_LEN):
827
raise errors.BadIndexOptions(self)
829
self._key_count = int(options_line[len(_OPTION_LEN):])
831
raise errors.BadIndexOptions(self)
832
# calculate the bytes we have processed
833
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
835
self._parsed_bytes(0, None, header_end, None)
836
# setup parsing state
837
self._expected_elements = 3 + self._key_length
838
# raw data keyed by offset
839
self._keys_by_offset = {}
840
# keys with the value and node references
841
self._bisect_nodes = {}
842
return header_end, bytes[header_end:]
844
def _parse_region(self, offset, data):
845
"""Parse node data returned from a readv operation.
847
:param offset: The byte offset the data starts at.
848
:param data: The data to parse.
852
end = offset + len(data)
855
# Trivial test - if the current index's end is within the
856
# low-matching parsed range, we're done.
857
index = self._parsed_byte_index(high_parsed)
858
if end < self._parsed_byte_map[index][1]:
860
# print "[%d:%d]" % (offset, end), \
861
# self._parsed_byte_map[index:index + 2]
862
high_parsed, last_segment = self._parse_segment(
863
offset, data, end, index)
867
def _parse_segment(self, offset, data, end, index):
868
"""Parse one segment of data.
870
:param offset: Where 'data' begins in the file.
871
:param data: Some data to parse a segment of.
872
:param end: Where data ends
873
:param index: The current index into the parsed bytes map.
874
:return: True if the parsed segment is the last possible one in the
876
:return: high_parsed_byte, last_segment.
877
high_parsed_byte is the location of the highest parsed byte in this
878
segment, last_segment is True if the parsed segment is the last
879
possible one in the data block.
881
# default is to use all data
883
# accomodate overlap with data before this.
884
if offset < self._parsed_byte_map[index][1]:
885
# overlaps the lower parsed region
886
# skip the parsed data
887
trim_start = self._parsed_byte_map[index][1] - offset
888
# don't trim the start for \n
889
start_adjacent = True
890
elif offset == self._parsed_byte_map[index][1]:
891
# abuts the lower parsed region
894
# do not trim anything
895
start_adjacent = True
897
# does not overlap the lower parsed region
900
# but trim the leading \n
901
start_adjacent = False
902
if end == self._size:
903
# lines up to the end of all data:
906
# do not strip to the last \n
909
elif index + 1 == len(self._parsed_byte_map):
910
# at the end of the parsed data
913
# but strip to the last \n
916
elif end == self._parsed_byte_map[index + 1][0]:
917
# buts up against the next parsed region
920
# do not strip to the last \n
923
elif end > self._parsed_byte_map[index + 1][0]:
924
# overlaps into the next parsed region
925
# only consider the unparsed data
926
trim_end = self._parsed_byte_map[index + 1][0] - offset
927
# do not strip to the last \n as we know its an entire record
929
last_segment = end < self._parsed_byte_map[index + 1][1]
931
# does not overlap into the next region
934
# but strip to the last \n
937
# now find bytes to discard if needed
938
if not start_adjacent:
939
# work around python bug in rfind
940
if trim_start is None:
941
trim_start = data.find('\n') + 1
943
trim_start = data.find('\n', trim_start) + 1
944
if not (trim_start != 0):
945
raise AssertionError('no \n was present')
946
# print 'removing start', offset, trim_start, repr(data[:trim_start])
948
# work around python bug in rfind
950
trim_end = data.rfind('\n') + 1
952
trim_end = data.rfind('\n', None, trim_end) + 1
953
if not (trim_end != 0):
954
raise AssertionError('no \n was present')
955
# print 'removing end', offset, trim_end, repr(data[trim_end:])
956
# adjust offset and data to the parseable data.
957
trimmed_data = data[trim_start:trim_end]
958
if not (trimmed_data):
959
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
960
% (trim_start, trim_end, offset, offset + len(data)))
963
# print "parsing", repr(trimmed_data)
964
# splitlines mangles the \r delimiters.. don't use it.
965
lines = trimmed_data.split('\n')
968
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
969
for key, value in nodes:
970
self._bisect_nodes[key] = value
971
self._parsed_bytes(offset, first_key,
972
offset + len(trimmed_data), last_key)
973
return offset + len(trimmed_data), last_segment
975
def _parse_lines(self, lines, pos):
984
if not (self._size == pos + 1):
985
raise AssertionError("%s %s" % (self._size, pos))
988
elements = line.split('\0')
989
if len(elements) != self._expected_elements:
990
raise errors.BadIndexData(self)
991
# keys are tuples. Each element is a string that may occur many
992
# times, so we intern them to save space. AB, RC, 200807
993
key = tuple([intern(element) for element in elements[:self._key_length]])
994
if first_key is None:
996
absent, references, value = elements[-3:]
998
for ref_string in references.split('\t'):
999
ref_lists.append(tuple([
1000
int(ref) for ref in ref_string.split('\r') if ref
1002
ref_lists = tuple(ref_lists)
1003
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1004
pos += len(line) + 1 # +1 for the \n
1007
if self.node_ref_lists:
1008
node_value = (value, ref_lists)
1011
nodes.append((key, node_value))
1012
# print "parsed ", key
1013
return first_key, key, nodes, trailers
1015
def _parsed_bytes(self, start, start_key, end, end_key):
1016
"""Mark the bytes from start to end as parsed.
1018
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1021
:param start: The start of the parsed region.
1022
:param end: The end of the parsed region.
1024
index = self._parsed_byte_index(start)
1025
new_value = (start, end)
1026
new_key = (start_key, end_key)
1028
# first range parsed is always the beginning.
1029
self._parsed_byte_map.insert(index, new_value)
1030
self._parsed_key_map.insert(index, new_key)
1034
# extend lower region
1035
# extend higher region
1036
# combine two regions
1037
if (index + 1 < len(self._parsed_byte_map) and
1038
self._parsed_byte_map[index][1] == start and
1039
self._parsed_byte_map[index + 1][0] == end):
1040
# combine two regions
1041
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1042
self._parsed_byte_map[index + 1][1])
1043
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1044
self._parsed_key_map[index + 1][1])
1045
del self._parsed_byte_map[index + 1]
1046
del self._parsed_key_map[index + 1]
1047
elif self._parsed_byte_map[index][1] == start:
1048
# extend the lower entry
1049
self._parsed_byte_map[index] = (
1050
self._parsed_byte_map[index][0], end)
1051
self._parsed_key_map[index] = (
1052
self._parsed_key_map[index][0], end_key)
1053
elif (index + 1 < len(self._parsed_byte_map) and
1054
self._parsed_byte_map[index + 1][0] == end):
1055
# extend the higher entry
1056
self._parsed_byte_map[index + 1] = (
1057
start, self._parsed_byte_map[index + 1][1])
1058
self._parsed_key_map[index + 1] = (
1059
start_key, self._parsed_key_map[index + 1][1])
1062
self._parsed_byte_map.insert(index + 1, new_value)
1063
self._parsed_key_map.insert(index + 1, new_key)
1065
def _read_and_parse(self, readv_ranges):
1066
"""Read the the ranges and parse the resulting data.
1068
:param readv_ranges: A prepared readv range list.
1070
if not readv_ranges:
1072
if self._nodes is None and self._bytes_read * 2 >= self._size:
1073
# We've already read more than 50% of the file and we are about to
1074
# request more data, just _buffer_all() and be done
1078
readv_data = self._transport.readv(self._name, readv_ranges, True,
1081
for offset, data in readv_data:
1082
self._bytes_read += len(data)
1083
if offset == 0 and len(data) == self._size:
1084
# We read the whole range, most likely because the
1085
# Transport upcast our readv ranges into one long request
1086
# for enough total data to grab the whole index.
1087
self._buffer_all(StringIO(data))
1089
if self._bisect_nodes is None:
1090
# this must be the start
1091
if not (offset == 0):
1092
raise AssertionError()
1093
offset, data = self._parse_header_from_bytes(data)
1094
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1095
self._parse_region(offset, data)
1097
def _signature(self):
1098
"""The file signature for this index type."""
1102
"""Validate that everything in the index can be accessed."""
1103
# iter_all validates completely at the moment, so just do that.
1104
for node in self.iter_all_entries():
1108
class CombinedGraphIndex(object):
1109
"""A GraphIndex made up from smaller GraphIndices.
1111
The backing indices must implement GraphIndex, and are presumed to be
1114
Queries against the combined index will be made against the first index,
1115
and then the second and so on. The order of index's can thus influence
1116
performance significantly. For example, if one index is on local disk and a
1117
second on a remote server, the local disk index should be before the other
1121
def __init__(self, indices):
1122
"""Create a CombinedGraphIndex backed by indices.
1124
:param indices: An ordered list of indices to query for data.
1126
self._indices = indices
1130
self.__class__.__name__,
1131
', '.join(map(repr, self._indices)))
1133
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1134
def get_parents(self, revision_ids):
1135
"""See graph._StackedParentsProvider.get_parents.
1137
This implementation thunks the graph.Graph.get_parents api across to
1140
:param revision_ids: An iterable of graph keys for this graph.
1141
:return: A list of parent details for each key in revision_ids.
1142
Each parent details will be one of:
1143
* None when the key was missing
1144
* (NULL_REVISION,) when the key has no parents.
1145
* (parent_key, parent_key...) otherwise.
1147
parent_map = self.get_parent_map(revision_ids)
1148
return [parent_map.get(r, None) for r in revision_ids]
1150
def get_parent_map(self, keys):
1151
"""See graph._StackedParentsProvider.get_parent_map"""
1152
search_keys = set(keys)
1153
if NULL_REVISION in search_keys:
1154
search_keys.discard(NULL_REVISION)
1155
found_parents = {NULL_REVISION:[]}
1158
for index, key, value, refs in self.iter_entries(search_keys):
1161
parents = (NULL_REVISION,)
1162
found_parents[key] = parents
1163
return found_parents
1165
def insert_index(self, pos, index):
1166
"""Insert a new index in the list of indices to query.
1168
:param pos: The position to insert the index.
1169
:param index: The index to insert.
1171
self._indices.insert(pos, index)
1173
def iter_all_entries(self):
1174
"""Iterate over all keys within the index
1176
Duplicate keys across child indices are presumed to have the same
1177
value and are only reported once.
1179
:return: An iterable of (index, key, reference_lists, value).
1180
There is no defined order for the result iteration - it will be in
1181
the most efficient order for the index.
1184
for index in self._indices:
1185
for node in index.iter_all_entries():
1186
if node[1] not in seen_keys:
1188
seen_keys.add(node[1])
1190
def iter_entries(self, keys):
1191
"""Iterate over keys within the index.
1193
Duplicate keys across child indices are presumed to have the same
1194
value and are only reported once.
1196
:param keys: An iterable providing the keys to be retrieved.
1197
:return: An iterable of (index, key, reference_lists, value). There is no
1198
defined order for the result iteration - it will be in the most
1199
efficient order for the index.
1202
for index in self._indices:
1205
for node in index.iter_entries(keys):
1206
keys.remove(node[1])
1209
def iter_entries_prefix(self, keys):
1210
"""Iterate over keys within the index using prefix matching.
1212
Duplicate keys across child indices are presumed to have the same
1213
value and are only reported once.
1215
Prefix matching is applied within the tuple of a key, not to within
1216
the bytestring of each key element. e.g. if you have the keys ('foo',
1217
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1218
only the former key is returned.
1220
:param keys: An iterable providing the key prefixes to be retrieved.
1221
Each key prefix takes the form of a tuple the length of a key, but
1222
with the last N elements 'None' rather than a regular bytestring.
1223
The first element cannot be 'None'.
1224
:return: An iterable as per iter_all_entries, but restricted to the
1225
keys with a matching prefix to those supplied. No additional keys
1226
will be returned, and every match that is in the index will be
1233
for index in self._indices:
1234
for node in index.iter_entries_prefix(keys):
1235
if node[1] in seen_keys:
1237
seen_keys.add(node[1])
1240
def key_count(self):
1241
"""Return an estimate of the number of keys in this index.
1243
For CombinedGraphIndex this is approximated by the sum of the keys of
1244
the child indices. As child indices may have duplicate keys this can
1245
have a maximum error of the number of child indices * largest number of
1248
return sum((index.key_count() for index in self._indices), 0)
1251
"""Validate that everything in the index can be accessed."""
1252
for index in self._indices:
1256
class InMemoryGraphIndex(GraphIndexBuilder):
1257
"""A GraphIndex which operates entirely out of memory and is mutable.
1259
This is designed to allow the accumulation of GraphIndex entries during a
1260
single write operation, where the accumulated entries need to be immediately
1261
available - for example via a CombinedGraphIndex.
1264
def add_nodes(self, nodes):
1265
"""Add nodes to the index.
1267
:param nodes: An iterable of (key, node_refs, value) entries to add.
1269
if self.reference_lists:
1270
for (key, value, node_refs) in nodes:
1271
self.add_node(key, value, node_refs)
1273
for (key, value) in nodes:
1274
self.add_node(key, value)
1276
def iter_all_entries(self):
1277
"""Iterate over all keys within the index
1279
:return: An iterable of (index, key, reference_lists, value). There is no
1280
defined order for the result iteration - it will be in the most
1281
efficient order for the index (in this case dictionary hash order).
1283
if 'evil' in debug.debug_flags:
1284
trace.mutter_callsite(3,
1285
"iter_all_entries scales with size of history.")
1286
if self.reference_lists:
1287
for key, (absent, references, value) in self._nodes.iteritems():
1289
yield self, key, value, references
1291
for key, (absent, references, value) in self._nodes.iteritems():
1293
yield self, key, value
1295
def iter_entries(self, keys):
1296
"""Iterate over keys within the index.
1298
:param keys: An iterable providing the keys to be retrieved.
1299
:return: An iterable of (index, key, value, reference_lists). There is no
1300
defined order for the result iteration - it will be in the most
1301
efficient order for the index (keys iteration order in this case).
1304
if self.reference_lists:
1305
for key in keys.intersection(self._keys):
1306
node = self._nodes[key]
1308
yield self, key, node[2], node[1]
1310
for key in keys.intersection(self._keys):
1311
node = self._nodes[key]
1313
yield self, key, node[2]
1315
def iter_entries_prefix(self, keys):
1316
"""Iterate over keys within the index using prefix matching.
1318
Prefix matching is applied within the tuple of a key, not to within
1319
the bytestring of each key element. e.g. if you have the keys ('foo',
1320
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1321
only the former key is returned.
1323
:param keys: An iterable providing the key prefixes to be retrieved.
1324
Each key prefix takes the form of a tuple the length of a key, but
1325
with the last N elements 'None' rather than a regular bytestring.
1326
The first element cannot be 'None'.
1327
:return: An iterable as per iter_all_entries, but restricted to the
1328
keys with a matching prefix to those supplied. No additional keys
1329
will be returned, and every match that is in the index will be
1332
# XXX: To much duplication with the GraphIndex class; consider finding
1333
# a good place to pull out the actual common logic.
1337
if self._key_length == 1:
1341
raise errors.BadIndexKey(key)
1342
if len(key) != self._key_length:
1343
raise errors.BadIndexKey(key)
1344
node = self._nodes[key]
1347
if self.reference_lists:
1348
yield self, key, node[2], node[1]
1350
yield self, key, node[2]
1352
nodes_by_key = self._get_nodes_by_key()
1356
raise errors.BadIndexKey(key)
1357
if len(key) != self._key_length:
1358
raise errors.BadIndexKey(key)
1359
# find what it refers to:
1360
key_dict = nodes_by_key
1361
elements = list(key)
1362
# find the subdict to return
1364
while len(elements) and elements[0] is not None:
1365
key_dict = key_dict[elements[0]]
1368
# a non-existant lookup.
1373
key_dict = dicts.pop(-1)
1374
# can't be empty or would not exist
1375
item, value = key_dict.iteritems().next()
1376
if type(value) == dict:
1378
dicts.extend(key_dict.itervalues())
1381
for value in key_dict.itervalues():
1382
yield (self, ) + value
1384
yield (self, ) + key_dict
1386
def key_count(self):
1387
"""Return an estimate of the number of keys in this index.
1389
For InMemoryGraphIndex the estimate is exact.
1391
return len(self._keys)
1394
"""In memory index's have no known corruption at the moment."""
1397
class GraphIndexPrefixAdapter(object):
1398
"""An adapter between GraphIndex with different key lengths.
1400
Queries against this will emit queries against the adapted Graph with the
1401
prefix added, queries for all items use iter_entries_prefix. The returned
1402
nodes will have their keys and node references adjusted to remove the
1403
prefix. Finally, an add_nodes_callback can be supplied - when called the
1404
nodes and references being added will have prefix prepended.
1407
def __init__(self, adapted, prefix, missing_key_length,
1408
add_nodes_callback=None):
1409
"""Construct an adapter against adapted with prefix."""
1410
self.adapted = adapted
1411
self.prefix_key = prefix + (None,)*missing_key_length
1412
self.prefix = prefix
1413
self.prefix_len = len(prefix)
1414
self.add_nodes_callback = add_nodes_callback
1416
def add_nodes(self, nodes):
1417
"""Add nodes to the index.
1419
:param nodes: An iterable of (key, node_refs, value) entries to add.
1421
# save nodes in case its an iterator
1422
nodes = tuple(nodes)
1423
translated_nodes = []
1425
# Add prefix_key to each reference node_refs is a tuple of tuples,
1426
# so split it apart, and add prefix_key to the internal reference
1427
for (key, value, node_refs) in nodes:
1428
adjusted_references = (
1429
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1430
for ref_list in node_refs))
1431
translated_nodes.append((self.prefix + key, value,
1432
adjusted_references))
1434
# XXX: TODO add an explicit interface for getting the reference list
1435
# status, to handle this bit of user-friendliness in the API more
1437
for (key, value) in nodes:
1438
translated_nodes.append((self.prefix + key, value))
1439
self.add_nodes_callback(translated_nodes)
1441
def add_node(self, key, value, references=()):
1442
"""Add a node to the index.
1444
:param key: The key. keys are non-empty tuples containing
1445
as many whitespace-free utf8 bytestrings as the key length
1446
defined for this index.
1447
:param references: An iterable of iterables of keys. Each is a
1448
reference to another key.
1449
:param value: The value to associate with the key. It may be any
1450
bytes as long as it does not contain \0 or \n.
1452
self.add_nodes(((key, value, references), ))
1454
def _strip_prefix(self, an_iter):
1455
"""Strip prefix data from nodes and return it."""
1456
for node in an_iter:
1458
if node[1][:self.prefix_len] != self.prefix:
1459
raise errors.BadIndexData(self)
1460
for ref_list in node[3]:
1461
for ref_node in ref_list:
1462
if ref_node[:self.prefix_len] != self.prefix:
1463
raise errors.BadIndexData(self)
1464
yield node[0], node[1][self.prefix_len:], node[2], (
1465
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1466
for ref_list in node[3]))
1468
def iter_all_entries(self):
1469
"""Iterate over all keys within the index
1471
iter_all_entries is implemented against the adapted index using
1472
iter_entries_prefix.
1474
:return: An iterable of (index, key, reference_lists, value). There is no
1475
defined order for the result iteration - it will be in the most
1476
efficient order for the index (in this case dictionary hash order).
1478
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1480
def iter_entries(self, keys):
1481
"""Iterate over keys within the index.
1483
:param keys: An iterable providing the keys to be retrieved.
1484
:return: An iterable of (index, key, value, reference_lists). There is no
1485
defined order for the result iteration - it will be in the most
1486
efficient order for the index (keys iteration order in this case).
1488
return self._strip_prefix(self.adapted.iter_entries(
1489
self.prefix + key for key in keys))
1491
def iter_entries_prefix(self, keys):
1492
"""Iterate over keys within the index using prefix matching.
1494
Prefix matching is applied within the tuple of a key, not to within
1495
the bytestring of each key element. e.g. if you have the keys ('foo',
1496
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1497
only the former key is returned.
1499
:param keys: An iterable providing the key prefixes to be retrieved.
1500
Each key prefix takes the form of a tuple the length of a key, but
1501
with the last N elements 'None' rather than a regular bytestring.
1502
The first element cannot be 'None'.
1503
:return: An iterable as per iter_all_entries, but restricted to the
1504
keys with a matching prefix to those supplied. No additional keys
1505
will be returned, and every match that is in the index will be
1508
return self._strip_prefix(self.adapted.iter_entries_prefix(
1509
self.prefix + key for key in keys))
1511
def key_count(self):
1512
"""Return an estimate of the number of keys in this index.
1514
For GraphIndexPrefixAdapter this is relatively expensive - key
1515
iteration with the prefix is done.
1517
return len(list(self.iter_all_entries()))
1520
"""Call the adapted's validate."""
1521
self.adapted.validate()
added
removed
bzrlib/index.py
