1
# Copyright (C) 2007, 2008 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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
"""Indexing facilities."""
23
'GraphIndexPrefixAdapter',
27
from bisect import bisect_right
28
from cStringIO import StringIO
32
from bzrlib.lazy_import import lazy_import
33
lazy_import(globals(), """
34
from bzrlib import trace
35
from bzrlib.bisect_multi import bisect_multi_bytes
36
from bzrlib.revision import NULL_REVISION
37
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
def _has_key_from_parent_map(self, key):
56
"""Check if this index has one key.
58
If it's possible to check for multiple keys at once through
59
calling get_parent_map that should be faster.
61
return (key in self.get_parent_map([key]))
64
def _missing_keys_from_parent_map(self, keys):
65
return set(keys) - set(self.get_parent_map(keys))
68
class GraphIndexBuilder(object):
69
"""A builder that can build a GraphIndex.
71
The resulting graph has the structure:
73
_SIGNATURE OPTIONS NODES NEWLINE
74
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
75
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
77
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
78
KEY := Not-whitespace-utf8
80
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
81
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
82
REFERENCE := DIGITS ; digits is the byte offset in the index of the
84
VALUE := no-newline-no-null-bytes
87
def __init__(self, reference_lists=0, key_elements=1):
88
"""Create a GraphIndex builder.
90
:param reference_lists: The number of node references lists for each
92
:param key_elements: The number of bytestrings in each key.
94
self.reference_lists = reference_lists
96
# A dict of {key: (absent, ref_lists, value)}
98
self._nodes_by_key = None
99
self._key_length = key_elements
100
self._optimize_for_size = False
101
self._combine_backing_indices = True
103
def _check_key(self, key):
104
"""Raise BadIndexKey if key is not a valid key for this index."""
105
if type(key) != tuple:
106
raise errors.BadIndexKey(key)
107
if self._key_length != len(key):
108
raise errors.BadIndexKey(key)
110
if not element or _whitespace_re.search(element) is not None:
111
raise errors.BadIndexKey(element)
113
def _external_references(self):
114
"""Return references that are not present in this index.
118
# TODO: JAM 2008-11-21 This makes an assumption about how the reference
119
# lists are used. It is currently correct for pack-0.92 through
120
# 1.9, which use the node references (3rd column) second
121
# reference list as the compression parent. Perhaps this should
122
# be moved into something higher up the stack, since it
123
# makes assumptions about how the index is used.
124
if self.reference_lists > 1:
125
for node in self.iter_all_entries():
127
refs.update(node[3][1])
130
# If reference_lists == 0 there can be no external references, and
131
# if reference_lists == 1, then there isn't a place to store the
135
def _get_nodes_by_key(self):
136
if self._nodes_by_key is None:
138
if self.reference_lists:
139
for key, (absent, references, value) in self._nodes.iteritems():
142
key_dict = nodes_by_key
143
for subkey in key[:-1]:
144
key_dict = key_dict.setdefault(subkey, {})
145
key_dict[key[-1]] = key, value, references
147
for key, (absent, references, value) in self._nodes.iteritems():
150
key_dict = nodes_by_key
151
for subkey in key[:-1]:
152
key_dict = key_dict.setdefault(subkey, {})
153
key_dict[key[-1]] = key, value
154
self._nodes_by_key = nodes_by_key
155
return self._nodes_by_key
157
def _update_nodes_by_key(self, key, value, node_refs):
158
"""Update the _nodes_by_key dict with a new key.
160
For a key of (foo, bar, baz) create
161
_nodes_by_key[foo][bar][baz] = key_value
163
if self._nodes_by_key is None:
165
key_dict = self._nodes_by_key
166
if self.reference_lists:
167
key_value = key, value, node_refs
169
key_value = key, value
170
for subkey in key[:-1]:
171
key_dict = key_dict.setdefault(subkey, {})
172
key_dict[key[-1]] = key_value
174
def _check_key_ref_value(self, key, references, value):
175
"""Check that 'key' and 'references' are all valid.
177
:param key: A key tuple. Must conform to the key interface (be a tuple,
178
be of the right length, not have any whitespace or nulls in any key
180
:param references: An iterable of reference lists. Something like
181
[[(ref, key)], [(ref, key), (other, key)]]
182
:param value: The value associate with this key. Must not contain
183
newlines or null characters.
184
:return: (node_refs, absent_references)
185
node_refs basically a packed form of 'references' where all
187
absent_references reference keys that are not in self._nodes.
188
This may contain duplicates if the same key is
189
referenced in multiple lists.
192
if _newline_null_re.search(value) is not None:
193
raise errors.BadIndexValue(value)
194
if len(references) != self.reference_lists:
195
raise errors.BadIndexValue(references)
197
absent_references = []
198
for reference_list in references:
199
for reference in reference_list:
200
# If reference *is* in self._nodes, then we know it has already
202
if reference not in self._nodes:
203
self._check_key(reference)
204
absent_references.append(reference)
205
node_refs.append(tuple(reference_list))
206
return tuple(node_refs), absent_references
208
def add_node(self, key, value, references=()):
209
"""Add a node to the index.
211
:param key: The key. keys are non-empty tuples containing
212
as many whitespace-free utf8 bytestrings as the key length
213
defined for this index.
214
:param references: An iterable of iterables of keys. Each is a
215
reference to another key.
216
:param value: The value to associate with the key. It may be any
217
bytes as long as it does not contain \0 or \n.
220
absent_references) = self._check_key_ref_value(key, references, value)
221
if key in self._nodes and self._nodes[key][0] != 'a':
222
raise errors.BadIndexDuplicateKey(key, self)
223
for reference in absent_references:
224
# There may be duplicates, but I don't think it is worth worrying
226
self._nodes[reference] = ('a', (), '')
227
self._nodes[key] = ('', node_refs, value)
229
if self._nodes_by_key is not None and self._key_length > 1:
230
self._update_nodes_by_key(key, value, node_refs)
234
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
235
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
236
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
237
prefix_length = sum(len(x) for x in lines)
238
# references are byte offsets. To avoid having to do nasty
239
# polynomial work to resolve offsets (references to later in the
240
# file cannot be determined until all the inbetween references have
241
# been calculated too) we pad the offsets with 0's to make them be
242
# of consistent length. Using binary offsets would break the trivial
244
# to calculate the width of zero's needed we do three passes:
245
# one to gather all the non-reference data and the number of references.
246
# one to pad all the data with reference-length and determine entry
250
# forward sorted by key. In future we may consider topological sorting,
251
# at the cost of table scans for direct lookup, or a second index for
253
nodes = sorted(self._nodes.items())
254
# if we do not prepass, we don't know how long it will be up front.
255
expected_bytes = None
256
# we only need to pre-pass if we have reference lists at all.
257
if self.reference_lists:
259
non_ref_bytes = prefix_length
261
# TODO use simple multiplication for the constants in this loop.
262
for key, (absent, references, value) in nodes:
263
# record the offset known *so far* for this key:
264
# the non reference bytes to date, and the total references to
265
# date - saves reaccumulating on the second pass
266
key_offset_info.append((key, non_ref_bytes, total_references))
267
# key is literal, value is literal, there are 3 null's, 1 NL
268
# key is variable length tuple, \x00 between elements
269
non_ref_bytes += sum(len(element) for element in key)
270
if self._key_length > 1:
271
non_ref_bytes += self._key_length - 1
272
# value is literal bytes, there are 3 null's, 1 NL.
273
non_ref_bytes += len(value) + 3 + 1
274
# one byte for absent if set.
277
elif self.reference_lists:
278
# (ref_lists -1) tabs
279
non_ref_bytes += self.reference_lists - 1
280
# (ref-1 cr's per ref_list)
281
for ref_list in references:
282
# how many references across the whole file?
283
total_references += len(ref_list)
284
# accrue reference separators
286
non_ref_bytes += len(ref_list) - 1
287
# how many digits are needed to represent the total byte count?
289
possible_total_bytes = non_ref_bytes + total_references*digits
290
while 10 ** digits < possible_total_bytes:
292
possible_total_bytes = non_ref_bytes + total_references*digits
293
expected_bytes = possible_total_bytes + 1 # terminating newline
294
# resolve key addresses.
296
for key, non_ref_bytes, total_references in key_offset_info:
297
key_addresses[key] = non_ref_bytes + total_references*digits
299
format_string = '%%0%sd' % digits
300
for key, (absent, references, value) in nodes:
301
flattened_references = []
302
for ref_list in references:
304
for reference in ref_list:
305
ref_addresses.append(format_string % key_addresses[reference])
306
flattened_references.append('\r'.join(ref_addresses))
307
string_key = '\x00'.join(key)
308
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
309
'\t'.join(flattened_references), value))
311
result = StringIO(''.join(lines))
312
if expected_bytes and len(result.getvalue()) != expected_bytes:
313
raise errors.BzrError('Failed index creation. Internal error:'
314
' mismatched output length and expected length: %d %d' %
315
(len(result.getvalue()), expected_bytes))
318
def set_optimize(self, for_size=None, combine_backing_indices=None):
319
"""Change how the builder tries to optimize the result.
321
:param for_size: Tell the builder to try and make the index as small as
323
:param combine_backing_indices: If the builder spills to disk to save
324
memory, should the on-disk indices be combined. Set to True if you
325
are going to be probing the index, but to False if you are not. (If
326
you are not querying, then the time spent combining is wasted.)
329
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
331
if for_size is not None:
332
self._optimize_for_size = for_size
333
if combine_backing_indices is not None:
334
self._combine_backing_indices = combine_backing_indices
337
class GraphIndex(object):
338
"""An index for data with embedded graphs.
340
The index maps keys to a list of key reference lists, and a value.
341
Each node has the same number of key reference lists. Each key reference
342
list can be empty or an arbitrary length. The value is an opaque NULL
343
terminated string without any newlines. The storage of the index is
344
hidden in the interface: keys and key references are always tuples of
345
bytestrings, never the internal representation (e.g. dictionary offsets).
347
It is presumed that the index will not be mutated - it is static data.
349
Successive iter_all_entries calls will read the entire index each time.
350
Additionally, iter_entries calls will read the index linearly until the
351
desired keys are found. XXX: This must be fixed before the index is
352
suitable for production use. :XXX
355
def __init__(self, transport, name, size):
356
"""Open an index called name on transport.
358
:param transport: A bzrlib.transport.Transport.
359
:param name: A path to provide to transport API calls.
360
:param size: The size of the index in bytes. This is used for bisection
361
logic to perform partial index reads. While the size could be
362
obtained by statting the file this introduced an additional round
363
trip as well as requiring stat'able transports, both of which are
364
avoided by having it supplied. If size is None, then bisection
365
support will be disabled and accessing the index will just stream
368
self._transport = transport
370
# Becomes a dict of key:(value, reference-list-byte-locations) used by
371
# the bisection interface to store parsed but not resolved keys.
372
self._bisect_nodes = None
373
# Becomes a dict of key:(value, reference-list-keys) which are ready to
374
# be returned directly to callers.
376
# a sorted list of slice-addresses for the parsed bytes of the file.
377
# e.g. (0,1) would mean that byte 0 is parsed.
378
self._parsed_byte_map = []
379
# a sorted list of keys matching each slice address for parsed bytes
380
# e.g. (None, 'foo@bar') would mean that the first byte contained no
381
# key, and the end byte of the slice is the of the data for 'foo@bar'
382
self._parsed_key_map = []
383
self._key_count = None
384
self._keys_by_offset = None
385
self._nodes_by_key = None
387
# The number of bytes we've read so far in trying to process this file
390
def __eq__(self, other):
391
"""Equal when self and other were created with the same parameters."""
393
type(self) == type(other) and
394
self._transport == other._transport and
395
self._name == other._name and
396
self._size == other._size)
398
def __ne__(self, other):
399
return not self.__eq__(other)
402
return "%s(%r)" % (self.__class__.__name__,
403
self._transport.abspath(self._name))
405
def _buffer_all(self, stream=None):
406
"""Buffer all the index data.
408
Mutates self._nodes and self.keys_by_offset.
410
if self._nodes is not None:
411
# We already did this
413
if 'index' in debug.debug_flags:
414
mutter('Reading entire index %s', self._transport.abspath(self._name))
416
stream = self._transport.get(self._name)
417
self._read_prefix(stream)
418
self._expected_elements = 3 + self._key_length
420
# raw data keyed by offset
421
self._keys_by_offset = {}
422
# ready-to-return key:value or key:value, node_ref_lists
424
self._nodes_by_key = None
427
lines = stream.read().split('\n')
429
_, _, _, trailers = self._parse_lines(lines, pos)
430
for key, absent, references, value in self._keys_by_offset.itervalues():
433
# resolve references:
434
if self.node_ref_lists:
435
node_value = (value, self._resolve_references(references))
438
self._nodes[key] = node_value
439
# cache the keys for quick set intersections
440
self._keys = set(self._nodes)
442
# there must be one line - the empty trailer line.
443
raise errors.BadIndexData(self)
445
def external_references(self, ref_list_num):
446
"""Return references that are not present in this index.
449
if ref_list_num + 1 > self.node_ref_lists:
450
raise ValueError('No ref list %d, index has %d ref lists'
451
% (ref_list_num, self.node_ref_lists))
453
for key, (value, ref_lists) in self._nodes.iteritems():
454
ref_list = ref_lists[ref_list_num]
455
refs.update(ref_list)
456
return refs - self._keys
458
def _get_nodes_by_key(self):
459
if self._nodes_by_key is None:
461
if self.node_ref_lists:
462
for key, (value, references) in self._nodes.iteritems():
463
key_dict = nodes_by_key
464
for subkey in key[:-1]:
465
key_dict = key_dict.setdefault(subkey, {})
466
key_dict[key[-1]] = key, value, references
468
for key, value in self._nodes.iteritems():
469
key_dict = nodes_by_key
470
for subkey in key[:-1]:
471
key_dict = key_dict.setdefault(subkey, {})
472
key_dict[key[-1]] = key, value
473
self._nodes_by_key = nodes_by_key
474
return self._nodes_by_key
476
def iter_all_entries(self):
477
"""Iterate over all keys within the index.
479
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
480
The former tuple is used when there are no reference lists in the
481
index, making the API compatible with simple key:value index types.
482
There is no defined order for the result iteration - it will be in
483
the most efficient order for the index.
485
if 'evil' in debug.debug_flags:
486
trace.mutter_callsite(3,
487
"iter_all_entries scales with size of history.")
488
if self._nodes is None:
490
if self.node_ref_lists:
491
for key, (value, node_ref_lists) in self._nodes.iteritems():
492
yield self, key, value, node_ref_lists
494
for key, value in self._nodes.iteritems():
495
yield self, key, value
497
def _read_prefix(self, stream):
498
signature = stream.read(len(self._signature()))
499
if not signature == self._signature():
500
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
501
options_line = stream.readline()
502
if not options_line.startswith(_OPTION_NODE_REFS):
503
raise errors.BadIndexOptions(self)
505
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
507
raise errors.BadIndexOptions(self)
508
options_line = stream.readline()
509
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
510
raise errors.BadIndexOptions(self)
512
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
514
raise errors.BadIndexOptions(self)
515
options_line = stream.readline()
516
if not options_line.startswith(_OPTION_LEN):
517
raise errors.BadIndexOptions(self)
519
self._key_count = int(options_line[len(_OPTION_LEN):-1])
521
raise errors.BadIndexOptions(self)
523
def _resolve_references(self, references):
524
"""Return the resolved key references for references.
526
References are resolved by looking up the location of the key in the
527
_keys_by_offset map and substituting the key name, preserving ordering.
529
:param references: An iterable of iterables of key locations. e.g.
531
:return: A tuple of tuples of keys.
534
for ref_list in references:
535
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
536
return tuple(node_refs)
538
def _find_index(self, range_map, key):
539
"""Helper for the _parsed_*_index calls.
541
Given a range map - [(start, end), ...], finds the index of the range
542
in the map for key if it is in the map, and if it is not there, the
543
immediately preceeding range in the map.
545
result = bisect_right(range_map, key) - 1
546
if result + 1 < len(range_map):
547
# check the border condition, it may be in result + 1
548
if range_map[result + 1][0] == key[0]:
552
def _parsed_byte_index(self, offset):
553
"""Return the index of the entry immediately before offset.
555
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
556
there is one unparsed byte (the 11th, addressed as[10]). then:
557
asking for 0 will return 0
558
asking for 10 will return 0
559
asking for 11 will return 1
560
asking for 12 will return 1
563
return self._find_index(self._parsed_byte_map, key)
565
def _parsed_key_index(self, key):
566
"""Return the index of the entry immediately before key.
568
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
569
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
570
have been parsed, then:
571
asking for '' will return 0
572
asking for 'a' will return 0
573
asking for 'b' will return 1
574
asking for 'e' will return 1
576
search_key = (key, None)
577
return self._find_index(self._parsed_key_map, search_key)
579
def _is_parsed(self, offset):
580
"""Returns True if offset has been parsed."""
581
index = self._parsed_byte_index(offset)
582
if index == len(self._parsed_byte_map):
583
return offset < self._parsed_byte_map[index - 1][1]
584
start, end = self._parsed_byte_map[index]
585
return offset >= start and offset < end
587
def _iter_entries_from_total_buffer(self, keys):
588
"""Iterate over keys when the entire index is parsed."""
589
keys = keys.intersection(self._keys)
590
if self.node_ref_lists:
592
value, node_refs = self._nodes[key]
593
yield self, key, value, node_refs
596
yield self, key, self._nodes[key]
598
def iter_entries(self, keys):
599
"""Iterate over keys within the index.
601
:param keys: An iterable providing the keys to be retrieved.
602
:return: An iterable as per iter_all_entries, but restricted to the
603
keys supplied. No additional keys will be returned, and every
604
key supplied that is in the index will be returned.
609
if self._size is None and self._nodes is None:
612
# We fit about 20 keys per minimum-read (4K), so if we are looking for
613
# more than 1/20th of the index its likely (assuming homogenous key
614
# spread) that we'll read the entire index. If we're going to do that,
615
# buffer the whole thing. A better analysis might take key spread into
616
# account - but B+Tree indices are better anyway.
617
# We could look at all data read, and use a threshold there, which will
618
# trigger on ancestry walks, but that is not yet fully mapped out.
619
if self._nodes is None and len(keys) * 20 > self.key_count():
621
if self._nodes is not None:
622
return self._iter_entries_from_total_buffer(keys)
624
return (result[1] for result in bisect_multi_bytes(
625
self._lookup_keys_via_location, self._size, keys))
627
def iter_entries_prefix(self, keys):
628
"""Iterate over keys within the index using prefix matching.
630
Prefix matching is applied within the tuple of a key, not to within
631
the bytestring of each key element. e.g. if you have the keys ('foo',
632
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
633
only the former key is returned.
635
WARNING: Note that this method currently causes a full index parse
636
unconditionally (which is reasonably appropriate as it is a means for
637
thunking many small indices into one larger one and still supplies
638
iter_all_entries at the thunk layer).
640
:param keys: An iterable providing the key prefixes to be retrieved.
641
Each key prefix takes the form of a tuple the length of a key, but
642
with the last N elements 'None' rather than a regular bytestring.
643
The first element cannot be 'None'.
644
:return: An iterable as per iter_all_entries, but restricted to the
645
keys with a matching prefix to those supplied. No additional keys
646
will be returned, and every match that is in the index will be
652
# load data - also finds key lengths
653
if self._nodes is None:
655
if self._key_length == 1:
659
raise errors.BadIndexKey(key)
660
if len(key) != self._key_length:
661
raise errors.BadIndexKey(key)
662
if self.node_ref_lists:
663
value, node_refs = self._nodes[key]
664
yield self, key, value, node_refs
666
yield self, key, self._nodes[key]
668
nodes_by_key = self._get_nodes_by_key()
672
raise errors.BadIndexKey(key)
673
if len(key) != self._key_length:
674
raise errors.BadIndexKey(key)
675
# find what it refers to:
676
key_dict = nodes_by_key
678
# find the subdict whose contents should be returned.
680
while len(elements) and elements[0] is not None:
681
key_dict = key_dict[elements[0]]
684
# a non-existant lookup.
689
key_dict = dicts.pop(-1)
690
# can't be empty or would not exist
691
item, value = key_dict.iteritems().next()
692
if type(value) == dict:
694
dicts.extend(key_dict.itervalues())
697
for value in key_dict.itervalues():
698
# each value is the key:value:node refs tuple
700
yield (self, ) + value
702
# the last thing looked up was a terminal element
703
yield (self, ) + key_dict
705
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
706
"""See BTreeIndex._find_ancestors."""
707
# The api can be implemented as a trivial overlay on top of
708
# iter_entries, it is not an efficient implementation, but it at least
712
for index, key, value, refs in self.iter_entries(keys):
713
parent_keys = refs[ref_list_num]
715
parent_map[key] = parent_keys
716
search_keys.update(parent_keys)
717
# Figure out what, if anything, was missing
718
missing_keys.update(set(keys).difference(found_keys))
719
search_keys = search_keys.difference(parent_map)
723
"""Return an estimate of the number of keys in this index.
725
For GraphIndex the estimate is exact.
727
if self._key_count is None:
728
self._read_and_parse([_HEADER_READV])
729
return self._key_count
731
def _lookup_keys_via_location(self, location_keys):
732
"""Public interface for implementing bisection.
734
If _buffer_all has been called, then all the data for the index is in
735
memory, and this method should not be called, as it uses a separate
736
cache because it cannot pre-resolve all indices, which buffer_all does
739
:param location_keys: A list of location(byte offset), key tuples.
740
:return: A list of (location_key, result) tuples as expected by
741
bzrlib.bisect_multi.bisect_multi_bytes.
743
# Possible improvements:
744
# - only bisect lookup each key once
745
# - sort the keys first, and use that to reduce the bisection window
747
# this progresses in three parts:
750
# attempt to answer the question from the now in memory data.
751
# build the readv request
752
# for each location, ask for 800 bytes - much more than rows we've seen
755
for location, key in location_keys:
756
# can we answer from cache?
757
if self._bisect_nodes and key in self._bisect_nodes:
758
# We have the key parsed.
760
index = self._parsed_key_index(key)
761
if (len(self._parsed_key_map) and
762
self._parsed_key_map[index][0] <= key and
763
(self._parsed_key_map[index][1] >= key or
764
# end of the file has been parsed
765
self._parsed_byte_map[index][1] == self._size)):
766
# the key has been parsed, so no lookup is needed even if its
769
# - if we have examined this part of the file already - yes
770
index = self._parsed_byte_index(location)
771
if (len(self._parsed_byte_map) and
772
self._parsed_byte_map[index][0] <= location and
773
self._parsed_byte_map[index][1] > location):
774
# the byte region has been parsed, so no read is needed.
777
if location + length > self._size:
778
length = self._size - location
779
# todo, trim out parsed locations.
781
readv_ranges.append((location, length))
782
# read the header if needed
783
if self._bisect_nodes is None:
784
readv_ranges.append(_HEADER_READV)
785
self._read_and_parse(readv_ranges)
787
if self._nodes is not None:
788
# _read_and_parse triggered a _buffer_all because we requested the
790
for location, key in location_keys:
791
if key not in self._nodes: # not present
792
result.append(((location, key), False))
793
elif self.node_ref_lists:
794
value, refs = self._nodes[key]
795
result.append(((location, key),
796
(self, key, value, refs)))
798
result.append(((location, key),
799
(self, key, self._nodes[key])))
802
# - figure out <, >, missing, present
803
# - result present references so we can return them.
804
# keys that we cannot answer until we resolve references
805
pending_references = []
806
pending_locations = set()
807
for location, key in location_keys:
808
# can we answer from cache?
809
if key in self._bisect_nodes:
810
# the key has been parsed, so no lookup is needed
811
if self.node_ref_lists:
812
# the references may not have been all parsed.
813
value, refs = self._bisect_nodes[key]
814
wanted_locations = []
815
for ref_list in refs:
817
if ref not in self._keys_by_offset:
818
wanted_locations.append(ref)
820
pending_locations.update(wanted_locations)
821
pending_references.append((location, key))
823
result.append(((location, key), (self, key,
824
value, self._resolve_references(refs))))
826
result.append(((location, key),
827
(self, key, self._bisect_nodes[key])))
830
# has the region the key should be in, been parsed?
831
index = self._parsed_key_index(key)
832
if (self._parsed_key_map[index][0] <= key and
833
(self._parsed_key_map[index][1] >= key or
834
# end of the file has been parsed
835
self._parsed_byte_map[index][1] == self._size)):
836
result.append(((location, key), False))
838
# no, is the key above or below the probed location:
839
# get the range of the probed & parsed location
840
index = self._parsed_byte_index(location)
841
# if the key is below the start of the range, its below
842
if key < self._parsed_key_map[index][0]:
846
result.append(((location, key), direction))
848
# lookup data to resolve references
849
for location in pending_locations:
851
if location + length > self._size:
852
length = self._size - location
853
# TODO: trim out parsed locations (e.g. if the 800 is into the
854
# parsed region trim it, and dont use the adjust_for_latency
857
readv_ranges.append((location, length))
858
self._read_and_parse(readv_ranges)
859
if self._nodes is not None:
860
# The _read_and_parse triggered a _buffer_all, grab the data and
862
for location, key in pending_references:
863
value, refs = self._nodes[key]
864
result.append(((location, key), (self, key, value, refs)))
866
for location, key in pending_references:
867
# answer key references we had to look-up-late.
868
value, refs = self._bisect_nodes[key]
869
result.append(((location, key), (self, key,
870
value, self._resolve_references(refs))))
873
def _parse_header_from_bytes(self, bytes):
874
"""Parse the header from a region of bytes.
876
:param bytes: The data to parse.
877
:return: An offset, data tuple such as readv yields, for the unparsed
878
data. (which may length 0).
880
signature = bytes[0:len(self._signature())]
881
if not signature == self._signature():
882
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
883
lines = bytes[len(self._signature()):].splitlines()
884
options_line = lines[0]
885
if not options_line.startswith(_OPTION_NODE_REFS):
886
raise errors.BadIndexOptions(self)
888
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
890
raise errors.BadIndexOptions(self)
891
options_line = lines[1]
892
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
893
raise errors.BadIndexOptions(self)
895
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
897
raise errors.BadIndexOptions(self)
898
options_line = lines[2]
899
if not options_line.startswith(_OPTION_LEN):
900
raise errors.BadIndexOptions(self)
902
self._key_count = int(options_line[len(_OPTION_LEN):])
904
raise errors.BadIndexOptions(self)
905
# calculate the bytes we have processed
906
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
908
self._parsed_bytes(0, None, header_end, None)
909
# setup parsing state
910
self._expected_elements = 3 + self._key_length
911
# raw data keyed by offset
912
self._keys_by_offset = {}
913
# keys with the value and node references
914
self._bisect_nodes = {}
915
return header_end, bytes[header_end:]
917
def _parse_region(self, offset, data):
918
"""Parse node data returned from a readv operation.
920
:param offset: The byte offset the data starts at.
921
:param data: The data to parse.
925
end = offset + len(data)
928
# Trivial test - if the current index's end is within the
929
# low-matching parsed range, we're done.
930
index = self._parsed_byte_index(high_parsed)
931
if end < self._parsed_byte_map[index][1]:
933
# print "[%d:%d]" % (offset, end), \
934
# self._parsed_byte_map[index:index + 2]
935
high_parsed, last_segment = self._parse_segment(
936
offset, data, end, index)
940
def _parse_segment(self, offset, data, end, index):
941
"""Parse one segment of data.
943
:param offset: Where 'data' begins in the file.
944
:param data: Some data to parse a segment of.
945
:param end: Where data ends
946
:param index: The current index into the parsed bytes map.
947
:return: True if the parsed segment is the last possible one in the
949
:return: high_parsed_byte, last_segment.
950
high_parsed_byte is the location of the highest parsed byte in this
951
segment, last_segment is True if the parsed segment is the last
952
possible one in the data block.
954
# default is to use all data
956
# accomodate overlap with data before this.
957
if offset < self._parsed_byte_map[index][1]:
958
# overlaps the lower parsed region
959
# skip the parsed data
960
trim_start = self._parsed_byte_map[index][1] - offset
961
# don't trim the start for \n
962
start_adjacent = True
963
elif offset == self._parsed_byte_map[index][1]:
964
# abuts the lower parsed region
967
# do not trim anything
968
start_adjacent = True
970
# does not overlap the lower parsed region
973
# but trim the leading \n
974
start_adjacent = False
975
if end == self._size:
976
# lines up to the end of all data:
979
# do not strip to the last \n
982
elif index + 1 == len(self._parsed_byte_map):
983
# at the end of the parsed data
986
# but strip to the last \n
989
elif end == self._parsed_byte_map[index + 1][0]:
990
# buts up against the next parsed region
993
# do not strip to the last \n
996
elif end > self._parsed_byte_map[index + 1][0]:
997
# overlaps into the next parsed region
998
# only consider the unparsed data
999
trim_end = self._parsed_byte_map[index + 1][0] - offset
1000
# do not strip to the last \n as we know its an entire record
1002
last_segment = end < self._parsed_byte_map[index + 1][1]
1004
# does not overlap into the next region
1007
# but strip to the last \n
1008
end_adjacent = False
1010
# now find bytes to discard if needed
1011
if not start_adjacent:
1012
# work around python bug in rfind
1013
if trim_start is None:
1014
trim_start = data.find('\n') + 1
1016
trim_start = data.find('\n', trim_start) + 1
1017
if not (trim_start != 0):
1018
raise AssertionError('no \n was present')
1019
# print 'removing start', offset, trim_start, repr(data[:trim_start])
1020
if not end_adjacent:
1021
# work around python bug in rfind
1022
if trim_end is None:
1023
trim_end = data.rfind('\n') + 1
1025
trim_end = data.rfind('\n', None, trim_end) + 1
1026
if not (trim_end != 0):
1027
raise AssertionError('no \n was present')
1028
# print 'removing end', offset, trim_end, repr(data[trim_end:])
1029
# adjust offset and data to the parseable data.
1030
trimmed_data = data[trim_start:trim_end]
1031
if not (trimmed_data):
1032
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
1033
% (trim_start, trim_end, offset, offset + len(data)))
1035
offset += trim_start
1036
# print "parsing", repr(trimmed_data)
1037
# splitlines mangles the \r delimiters.. don't use it.
1038
lines = trimmed_data.split('\n')
1041
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1042
for key, value in nodes:
1043
self._bisect_nodes[key] = value
1044
self._parsed_bytes(offset, first_key,
1045
offset + len(trimmed_data), last_key)
1046
return offset + len(trimmed_data), last_segment
1048
def _parse_lines(self, lines, pos):
1055
# must be at the end
1057
if not (self._size == pos + 1):
1058
raise AssertionError("%s %s" % (self._size, pos))
1061
elements = line.split('\0')
1062
if len(elements) != self._expected_elements:
1063
raise errors.BadIndexData(self)
1064
# keys are tuples. Each element is a string that may occur many
1065
# times, so we intern them to save space. AB, RC, 200807
1066
key = tuple([intern(element) for element in elements[:self._key_length]])
1067
if first_key is None:
1069
absent, references, value = elements[-3:]
1071
for ref_string in references.split('\t'):
1072
ref_lists.append(tuple([
1073
int(ref) for ref in ref_string.split('\r') if ref
1075
ref_lists = tuple(ref_lists)
1076
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1077
pos += len(line) + 1 # +1 for the \n
1080
if self.node_ref_lists:
1081
node_value = (value, ref_lists)
1084
nodes.append((key, node_value))
1085
# print "parsed ", key
1086
return first_key, key, nodes, trailers
1088
def _parsed_bytes(self, start, start_key, end, end_key):
1089
"""Mark the bytes from start to end as parsed.
1091
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1094
:param start: The start of the parsed region.
1095
:param end: The end of the parsed region.
1097
index = self._parsed_byte_index(start)
1098
new_value = (start, end)
1099
new_key = (start_key, end_key)
1101
# first range parsed is always the beginning.
1102
self._parsed_byte_map.insert(index, new_value)
1103
self._parsed_key_map.insert(index, new_key)
1107
# extend lower region
1108
# extend higher region
1109
# combine two regions
1110
if (index + 1 < len(self._parsed_byte_map) and
1111
self._parsed_byte_map[index][1] == start and
1112
self._parsed_byte_map[index + 1][0] == end):
1113
# combine two regions
1114
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1115
self._parsed_byte_map[index + 1][1])
1116
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1117
self._parsed_key_map[index + 1][1])
1118
del self._parsed_byte_map[index + 1]
1119
del self._parsed_key_map[index + 1]
1120
elif self._parsed_byte_map[index][1] == start:
1121
# extend the lower entry
1122
self._parsed_byte_map[index] = (
1123
self._parsed_byte_map[index][0], end)
1124
self._parsed_key_map[index] = (
1125
self._parsed_key_map[index][0], end_key)
1126
elif (index + 1 < len(self._parsed_byte_map) and
1127
self._parsed_byte_map[index + 1][0] == end):
1128
# extend the higher entry
1129
self._parsed_byte_map[index + 1] = (
1130
start, self._parsed_byte_map[index + 1][1])
1131
self._parsed_key_map[index + 1] = (
1132
start_key, self._parsed_key_map[index + 1][1])
1135
self._parsed_byte_map.insert(index + 1, new_value)
1136
self._parsed_key_map.insert(index + 1, new_key)
1138
def _read_and_parse(self, readv_ranges):
1139
"""Read the the ranges and parse the resulting data.
1141
:param readv_ranges: A prepared readv range list.
1143
if not readv_ranges:
1145
if self._nodes is None and self._bytes_read * 2 >= self._size:
1146
# We've already read more than 50% of the file and we are about to
1147
# request more data, just _buffer_all() and be done
1151
readv_data = self._transport.readv(self._name, readv_ranges, True,
1154
for offset, data in readv_data:
1155
self._bytes_read += len(data)
1156
if offset == 0 and len(data) == self._size:
1157
# We read the whole range, most likely because the
1158
# Transport upcast our readv ranges into one long request
1159
# for enough total data to grab the whole index.
1160
self._buffer_all(StringIO(data))
1162
if self._bisect_nodes is None:
1163
# this must be the start
1164
if not (offset == 0):
1165
raise AssertionError()
1166
offset, data = self._parse_header_from_bytes(data)
1167
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1168
self._parse_region(offset, data)
1170
def _signature(self):
1171
"""The file signature for this index type."""
1175
"""Validate that everything in the index can be accessed."""
1176
# iter_all validates completely at the moment, so just do that.
1177
for node in self.iter_all_entries():
1181
class CombinedGraphIndex(object):
1182
"""A GraphIndex made up from smaller GraphIndices.
1184
The backing indices must implement GraphIndex, and are presumed to be
1187
Queries against the combined index will be made against the first index,
1188
and then the second and so on. The order of index's can thus influence
1189
performance significantly. For example, if one index is on local disk and a
1190
second on a remote server, the local disk index should be before the other
1194
def __init__(self, indices, reload_func=None):
1195
"""Create a CombinedGraphIndex backed by indices.
1197
:param indices: An ordered list of indices to query for data.
1198
:param reload_func: A function to call if we find we are missing an
1199
index. Should have the form reload_func() => True/False to indicate
1200
if reloading actually changed anything.
1202
self._indices = indices
1203
self._reload_func = reload_func
1207
self.__class__.__name__,
1208
', '.join(map(repr, self._indices)))
1210
def get_parent_map(self, keys):
1211
"""See graph.StackedParentsProvider.get_parent_map"""
1212
search_keys = set(keys)
1213
if NULL_REVISION in search_keys:
1214
search_keys.discard(NULL_REVISION)
1215
found_parents = {NULL_REVISION:[]}
1218
for index, key, value, refs in self.iter_entries(search_keys):
1221
parents = (NULL_REVISION,)
1222
found_parents[key] = parents
1223
return found_parents
1225
has_key = _has_key_from_parent_map
1227
def insert_index(self, pos, index):
1228
"""Insert a new index in the list of indices to query.
1230
:param pos: The position to insert the index.
1231
:param index: The index to insert.
1233
self._indices.insert(pos, index)
1235
def iter_all_entries(self):
1236
"""Iterate over all keys within the index
1238
Duplicate keys across child indices are presumed to have the same
1239
value and are only reported once.
1241
:return: An iterable of (index, key, reference_lists, value).
1242
There is no defined order for the result iteration - it will be in
1243
the most efficient order for the index.
1248
for index in self._indices:
1249
for node in index.iter_all_entries():
1250
if node[1] not in seen_keys:
1252
seen_keys.add(node[1])
1254
except errors.NoSuchFile:
1255
self._reload_or_raise()
1257
def iter_entries(self, keys):
1258
"""Iterate over keys within the index.
1260
Duplicate keys across child indices are presumed to have the same
1261
value and are only reported once.
1263
:param keys: An iterable providing the keys to be retrieved.
1264
:return: An iterable of (index, key, reference_lists, value). There is no
1265
defined order for the result iteration - it will be in the most
1266
efficient order for the index.
1271
for index in self._indices:
1274
for node in index.iter_entries(keys):
1275
keys.remove(node[1])
1278
except errors.NoSuchFile:
1279
self._reload_or_raise()
1281
def iter_entries_prefix(self, keys):
1282
"""Iterate over keys within the index using prefix matching.
1284
Duplicate keys across child indices are presumed to have the same
1285
value and are only reported once.
1287
Prefix matching is applied within the tuple of a key, not to within
1288
the bytestring of each key element. e.g. if you have the keys ('foo',
1289
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1290
only the former key is returned.
1292
:param keys: An iterable providing the key prefixes to be retrieved.
1293
Each key prefix takes the form of a tuple the length of a key, but
1294
with the last N elements 'None' rather than a regular bytestring.
1295
The first element cannot be 'None'.
1296
:return: An iterable as per iter_all_entries, but restricted to the
1297
keys with a matching prefix to those supplied. No additional keys
1298
will be returned, and every match that is in the index will be
1307
for index in self._indices:
1308
for node in index.iter_entries_prefix(keys):
1309
if node[1] in seen_keys:
1311
seen_keys.add(node[1])
1314
except errors.NoSuchFile:
1315
self._reload_or_raise()
1317
def find_ancestry(self, keys, ref_list_num):
1318
"""Find the complete ancestry for the given set of keys.
1320
Note that this is a whole-ancestry request, so it should be used
1323
:param keys: An iterable of keys to look for
1324
:param ref_list_num: The reference list which references the parents
1326
:return: (parent_map, missing_keys)
1328
missing_keys = set()
1330
keys_to_lookup = set(keys)
1332
while keys_to_lookup:
1333
# keys that *all* indexes claim are missing, stop searching them
1335
all_index_missing = None
1336
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1337
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1339
# len(missing_keys))
1340
for index_idx, index in enumerate(self._indices):
1341
# TODO: we should probably be doing something with
1342
# 'missing_keys' since we've already determined that
1343
# those revisions have not been found anywhere
1344
index_missing_keys = set()
1345
# Find all of the ancestry we can from this index
1346
# keep looking until the search_keys set is empty, which means
1347
# things we didn't find should be in index_missing_keys
1348
search_keys = keys_to_lookup
1350
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1351
# index_idx, len(search_keys),
1352
# len(parent_map), len(index_missing_keys))
1355
# TODO: ref_list_num should really be a parameter, since
1356
# CombinedGraphIndex does not know what the ref lists
1358
search_keys = index._find_ancestors(search_keys,
1359
ref_list_num, parent_map, index_missing_keys)
1360
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1361
# sub_generation, len(search_keys),
1362
# len(parent_map), len(index_missing_keys))
1363
# Now set whatever was missing to be searched in the next index
1364
keys_to_lookup = index_missing_keys
1365
if all_index_missing is None:
1366
all_index_missing = set(index_missing_keys)
1368
all_index_missing.intersection_update(index_missing_keys)
1369
if not keys_to_lookup:
1371
if all_index_missing is None:
1372
# There were no indexes, so all search keys are 'missing'
1373
missing_keys.update(keys_to_lookup)
1374
keys_to_lookup = None
1376
missing_keys.update(all_index_missing)
1377
keys_to_lookup.difference_update(all_index_missing)
1378
return parent_map, missing_keys
1380
def key_count(self):
1381
"""Return an estimate of the number of keys in this index.
1383
For CombinedGraphIndex this is approximated by the sum of the keys of
1384
the child indices. As child indices may have duplicate keys this can
1385
have a maximum error of the number of child indices * largest number of
1390
return sum((index.key_count() for index in self._indices), 0)
1391
except errors.NoSuchFile:
1392
self._reload_or_raise()
1394
missing_keys = _missing_keys_from_parent_map
1396
def _reload_or_raise(self):
1397
"""We just got a NoSuchFile exception.
1399
Try to reload the indices, if it fails, just raise the current
1402
if self._reload_func is None:
1404
exc_type, exc_value, exc_traceback = sys.exc_info()
1405
trace.mutter('Trying to reload after getting exception: %s',
1407
if not self._reload_func():
1408
# We tried to reload, but nothing changed, so we fail anyway
1409
trace.mutter('_reload_func indicated nothing has changed.'
1410
' Raising original exception.')
1411
raise exc_type, exc_value, exc_traceback
1414
"""Validate that everything in the index can be accessed."""
1417
for index in self._indices:
1420
except errors.NoSuchFile:
1421
self._reload_or_raise()
1424
class InMemoryGraphIndex(GraphIndexBuilder):
1425
"""A GraphIndex which operates entirely out of memory and is mutable.
1427
This is designed to allow the accumulation of GraphIndex entries during a
1428
single write operation, where the accumulated entries need to be immediately
1429
available - for example via a CombinedGraphIndex.
1432
def add_nodes(self, nodes):
1433
"""Add nodes to the index.
1435
:param nodes: An iterable of (key, node_refs, value) entries to add.
1437
if self.reference_lists:
1438
for (key, value, node_refs) in nodes:
1439
self.add_node(key, value, node_refs)
1441
for (key, value) in nodes:
1442
self.add_node(key, value)
1444
def iter_all_entries(self):
1445
"""Iterate over all keys within the index
1447
:return: An iterable of (index, key, reference_lists, value). There is no
1448
defined order for the result iteration - it will be in the most
1449
efficient order for the index (in this case dictionary hash order).
1451
if 'evil' in debug.debug_flags:
1452
trace.mutter_callsite(3,
1453
"iter_all_entries scales with size of history.")
1454
if self.reference_lists:
1455
for key, (absent, references, value) in self._nodes.iteritems():
1457
yield self, key, value, references
1459
for key, (absent, references, value) in self._nodes.iteritems():
1461
yield self, key, value
1463
def iter_entries(self, keys):
1464
"""Iterate over keys within the index.
1466
:param keys: An iterable providing the keys to be retrieved.
1467
:return: An iterable of (index, key, value, reference_lists). There is no
1468
defined order for the result iteration - it will be in the most
1469
efficient order for the index (keys iteration order in this case).
1472
if self.reference_lists:
1473
for key in keys.intersection(self._keys):
1474
node = self._nodes[key]
1476
yield self, key, node[2], node[1]
1478
for key in keys.intersection(self._keys):
1479
node = self._nodes[key]
1481
yield self, key, node[2]
1483
def iter_entries_prefix(self, keys):
1484
"""Iterate over keys within the index using prefix matching.
1486
Prefix matching is applied within the tuple of a key, not to within
1487
the bytestring of each key element. e.g. if you have the keys ('foo',
1488
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1489
only the former key is returned.
1491
:param keys: An iterable providing the key prefixes to be retrieved.
1492
Each key prefix takes the form of a tuple the length of a key, but
1493
with the last N elements 'None' rather than a regular bytestring.
1494
The first element cannot be 'None'.
1495
:return: An iterable as per iter_all_entries, but restricted to the
1496
keys with a matching prefix to those supplied. No additional keys
1497
will be returned, and every match that is in the index will be
1500
# XXX: To much duplication with the GraphIndex class; consider finding
1501
# a good place to pull out the actual common logic.
1505
if self._key_length == 1:
1509
raise errors.BadIndexKey(key)
1510
if len(key) != self._key_length:
1511
raise errors.BadIndexKey(key)
1512
node = self._nodes[key]
1515
if self.reference_lists:
1516
yield self, key, node[2], node[1]
1518
yield self, key, node[2]
1520
nodes_by_key = self._get_nodes_by_key()
1524
raise errors.BadIndexKey(key)
1525
if len(key) != self._key_length:
1526
raise errors.BadIndexKey(key)
1527
# find what it refers to:
1528
key_dict = nodes_by_key
1529
elements = list(key)
1530
# find the subdict to return
1532
while len(elements) and elements[0] is not None:
1533
key_dict = key_dict[elements[0]]
1536
# a non-existant lookup.
1541
key_dict = dicts.pop(-1)
1542
# can't be empty or would not exist
1543
item, value = key_dict.iteritems().next()
1544
if type(value) == dict:
1546
dicts.extend(key_dict.itervalues())
1549
for value in key_dict.itervalues():
1550
yield (self, ) + value
1552
yield (self, ) + key_dict
1554
def key_count(self):
1555
"""Return an estimate of the number of keys in this index.
1557
For InMemoryGraphIndex the estimate is exact.
1559
return len(self._keys)
1562
"""In memory index's have no known corruption at the moment."""
1565
class GraphIndexPrefixAdapter(object):
1566
"""An adapter between GraphIndex with different key lengths.
1568
Queries against this will emit queries against the adapted Graph with the
1569
prefix added, queries for all items use iter_entries_prefix. The returned
1570
nodes will have their keys and node references adjusted to remove the
1571
prefix. Finally, an add_nodes_callback can be supplied - when called the
1572
nodes and references being added will have prefix prepended.
1575
def __init__(self, adapted, prefix, missing_key_length,
1576
add_nodes_callback=None):
1577
"""Construct an adapter against adapted with prefix."""
1578
self.adapted = adapted
1579
self.prefix_key = prefix + (None,)*missing_key_length
1580
self.prefix = prefix
1581
self.prefix_len = len(prefix)
1582
self.add_nodes_callback = add_nodes_callback
1584
def add_nodes(self, nodes):
1585
"""Add nodes to the index.
1587
:param nodes: An iterable of (key, node_refs, value) entries to add.
1589
# save nodes in case its an iterator
1590
nodes = tuple(nodes)
1591
translated_nodes = []
1593
# Add prefix_key to each reference node_refs is a tuple of tuples,
1594
# so split it apart, and add prefix_key to the internal reference
1595
for (key, value, node_refs) in nodes:
1596
adjusted_references = (
1597
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1598
for ref_list in node_refs))
1599
translated_nodes.append((self.prefix + key, value,
1600
adjusted_references))
1602
# XXX: TODO add an explicit interface for getting the reference list
1603
# status, to handle this bit of user-friendliness in the API more
1605
for (key, value) in nodes:
1606
translated_nodes.append((self.prefix + key, value))
1607
self.add_nodes_callback(translated_nodes)
1609
def add_node(self, key, value, references=()):
1610
"""Add a node to the index.
1612
:param key: The key. keys are non-empty tuples containing
1613
as many whitespace-free utf8 bytestrings as the key length
1614
defined for this index.
1615
:param references: An iterable of iterables of keys. Each is a
1616
reference to another key.
1617
:param value: The value to associate with the key. It may be any
1618
bytes as long as it does not contain \0 or \n.
1620
self.add_nodes(((key, value, references), ))
1622
def _strip_prefix(self, an_iter):
1623
"""Strip prefix data from nodes and return it."""
1624
for node in an_iter:
1626
if node[1][:self.prefix_len] != self.prefix:
1627
raise errors.BadIndexData(self)
1628
for ref_list in node[3]:
1629
for ref_node in ref_list:
1630
if ref_node[:self.prefix_len] != self.prefix:
1631
raise errors.BadIndexData(self)
1632
yield node[0], node[1][self.prefix_len:], node[2], (
1633
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1634
for ref_list in node[3]))
1636
def iter_all_entries(self):
1637
"""Iterate over all keys within the index
1639
iter_all_entries is implemented against the adapted index using
1640
iter_entries_prefix.
1642
:return: An iterable of (index, key, reference_lists, value). There is no
1643
defined order for the result iteration - it will be in the most
1644
efficient order for the index (in this case dictionary hash order).
1646
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1648
def iter_entries(self, keys):
1649
"""Iterate over keys within the index.
1651
:param keys: An iterable providing the keys to be retrieved.
1652
:return: An iterable of (index, key, value, reference_lists). There is no
1653
defined order for the result iteration - it will be in the most
1654
efficient order for the index (keys iteration order in this case).
1656
return self._strip_prefix(self.adapted.iter_entries(
1657
self.prefix + key for key in keys))
1659
def iter_entries_prefix(self, keys):
1660
"""Iterate over keys within the index using prefix matching.
1662
Prefix matching is applied within the tuple of a key, not to within
1663
the bytestring of each key element. e.g. if you have the keys ('foo',
1664
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1665
only the former key is returned.
1667
:param keys: An iterable providing the key prefixes to be retrieved.
1668
Each key prefix takes the form of a tuple the length of a key, but
1669
with the last N elements 'None' rather than a regular bytestring.
1670
The first element cannot be 'None'.
1671
:return: An iterable as per iter_all_entries, but restricted to the
1672
keys with a matching prefix to those supplied. No additional keys
1673
will be returned, and every match that is in the index will be
1676
return self._strip_prefix(self.adapted.iter_entries_prefix(
1677
self.prefix + key for key in keys))
1679
def key_count(self):
1680
"""Return an estimate of the number of keys in this index.
1682
For GraphIndexPrefixAdapter this is relatively expensive - key
1683
iteration with the prefix is done.
1685
return len(list(self.iter_all_entries()))
1688
"""Call the adapted's validate."""
1689
self.adapted.validate()
added
removed
bzrlib/index.py
