1
# Copyright (C) 2007, 2008, 2009 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
43
from bzrlib.static_tuple import StaticTuple
45
_HEADER_READV = (0, 200)
46
_OPTION_KEY_ELEMENTS = "key_elements="
48
_OPTION_NODE_REFS = "node_ref_lists="
49
_SIGNATURE = "Bazaar Graph Index 1\n"
52
_whitespace_re = re.compile('[\t\n\x0b\x0c\r\x00 ]')
53
_newline_null_re = re.compile('[\n\0]')
56
def _has_key_from_parent_map(self, key):
57
"""Check if this index has one key.
59
If it's possible to check for multiple keys at once through
60
calling get_parent_map that should be faster.
62
return (key in self.get_parent_map([key]))
65
def _missing_keys_from_parent_map(self, keys):
66
return set(keys) - set(self.get_parent_map(keys))
69
class GraphIndexBuilder(object):
70
"""A builder that can build a GraphIndex.
72
The resulting graph has the structure:
74
_SIGNATURE OPTIONS NODES NEWLINE
75
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
76
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
78
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
79
KEY := Not-whitespace-utf8
81
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
82
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
83
REFERENCE := DIGITS ; digits is the byte offset in the index of the
85
VALUE := no-newline-no-null-bytes
88
def __init__(self, reference_lists=0, key_elements=1):
89
"""Create a GraphIndex builder.
91
:param reference_lists: The number of node references lists for each
93
:param key_elements: The number of bytestrings in each key.
95
self.reference_lists = reference_lists
97
# A dict of {key: (absent, ref_lists, value)}
99
self._nodes_by_key = None
100
self._key_length = key_elements
101
self._optimize_for_size = False
102
self._combine_backing_indices = True
104
def _check_key(self, key):
105
"""Raise BadIndexKey if key is not a valid key for this index."""
106
if type(key) not in (tuple, StaticTuple):
107
raise errors.BadIndexKey(key)
108
if self._key_length != len(key):
109
raise errors.BadIndexKey(key)
111
if not element or _whitespace_re.search(element) is not None:
112
raise errors.BadIndexKey(element)
114
def _external_references(self):
115
"""Return references that are not present in this index.
119
# TODO: JAM 2008-11-21 This makes an assumption about how the reference
120
# lists are used. It is currently correct for pack-0.92 through
121
# 1.9, which use the node references (3rd column) second
122
# reference list as the compression parent. Perhaps this should
123
# be moved into something higher up the stack, since it
124
# makes assumptions about how the index is used.
125
if self.reference_lists > 1:
126
for node in self.iter_all_entries():
128
refs.update(node[3][1])
131
# If reference_lists == 0 there can be no external references, and
132
# if reference_lists == 1, then there isn't a place to store the
136
def _get_nodes_by_key(self):
137
if self._nodes_by_key is None:
139
if self.reference_lists:
140
for key, (absent, references, value) in self._nodes.iteritems():
143
key_dict = nodes_by_key
144
for subkey in key[:-1]:
145
key_dict = key_dict.setdefault(subkey, {})
146
key_dict[key[-1]] = key, value, references
148
for key, (absent, references, value) in self._nodes.iteritems():
151
key_dict = nodes_by_key
152
for subkey in key[:-1]:
153
key_dict = key_dict.setdefault(subkey, {})
154
key_dict[key[-1]] = key, value
155
self._nodes_by_key = nodes_by_key
156
return self._nodes_by_key
158
def _update_nodes_by_key(self, key, value, node_refs):
159
"""Update the _nodes_by_key dict with a new key.
161
For a key of (foo, bar, baz) create
162
_nodes_by_key[foo][bar][baz] = key_value
164
if self._nodes_by_key is None:
166
key_dict = self._nodes_by_key
167
if self.reference_lists:
168
key_value = key, value, node_refs
170
key_value = key, value
171
for subkey in key[:-1]:
172
key_dict = key_dict.setdefault(subkey, {})
173
key_dict[key[-1]] = key_value
175
def _check_key_ref_value(self, key, references, value):
176
"""Check that 'key' and 'references' are all valid.
178
:param key: A key tuple. Must conform to the key interface (be a tuple,
179
be of the right length, not have any whitespace or nulls in any key
181
:param references: An iterable of reference lists. Something like
182
[[(ref, key)], [(ref, key), (other, key)]]
183
:param value: The value associate with this key. Must not contain
184
newlines or null characters.
185
:return: (node_refs, absent_references)
186
node_refs basically a packed form of 'references' where all
188
absent_references reference keys that are not in self._nodes.
189
This may contain duplicates if the same key is
190
referenced in multiple lists.
193
if _newline_null_re.search(value) is not None:
194
raise errors.BadIndexValue(value)
195
if len(references) != self.reference_lists:
196
raise errors.BadIndexValue(references)
198
absent_references = []
199
for reference_list in references:
200
for reference in reference_list:
201
# If reference *is* in self._nodes, then we know it has already
203
if reference not in self._nodes:
204
self._check_key(reference)
205
absent_references.append(reference)
207
node_refs.append(tuple(reference_list))
209
return tuple(node_refs), absent_references
211
def add_node(self, key, value, references=()):
212
"""Add a node to the index.
214
:param key: The key. keys are non-empty tuples containing
215
as many whitespace-free utf8 bytestrings as the key length
216
defined for this index.
217
:param references: An iterable of iterables of keys. Each is a
218
reference to another key.
219
:param value: The value to associate with the key. It may be any
220
bytes as long as it does not contain \0 or \n.
223
absent_references) = self._check_key_ref_value(key, references, value)
224
if key in self._nodes and self._nodes[key][0] != 'a':
225
raise errors.BadIndexDuplicateKey(key, self)
226
for reference in absent_references:
227
# There may be duplicates, but I don't think it is worth worrying
229
self._nodes[reference] = ('a', (), '')
230
self._nodes[key] = ('', node_refs, value)
232
if self._nodes_by_key is not None and self._key_length > 1:
233
self._update_nodes_by_key(key, value, node_refs)
235
def clear_cache(self):
236
"""See GraphIndex.clear_cache()
238
This is a no-op, but we need the api to conform to a generic 'Index'
244
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
245
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
246
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
247
prefix_length = sum(len(x) for x in lines)
248
# references are byte offsets. To avoid having to do nasty
249
# polynomial work to resolve offsets (references to later in the
250
# file cannot be determined until all the inbetween references have
251
# been calculated too) we pad the offsets with 0's to make them be
252
# of consistent length. Using binary offsets would break the trivial
254
# to calculate the width of zero's needed we do three passes:
255
# one to gather all the non-reference data and the number of references.
256
# one to pad all the data with reference-length and determine entry
260
# forward sorted by key. In future we may consider topological sorting,
261
# at the cost of table scans for direct lookup, or a second index for
263
nodes = sorted(self._nodes.items())
264
# if we do not prepass, we don't know how long it will be up front.
265
expected_bytes = None
266
# we only need to pre-pass if we have reference lists at all.
267
if self.reference_lists:
269
non_ref_bytes = prefix_length
271
# TODO use simple multiplication for the constants in this loop.
272
for key, (absent, references, value) in nodes:
273
# record the offset known *so far* for this key:
274
# the non reference bytes to date, and the total references to
275
# date - saves reaccumulating on the second pass
276
key_offset_info.append((key, non_ref_bytes, total_references))
277
# key is literal, value is literal, there are 3 null's, 1 NL
278
# key is variable length tuple, \x00 between elements
279
non_ref_bytes += sum(len(element) for element in key)
280
if self._key_length > 1:
281
non_ref_bytes += self._key_length - 1
282
# value is literal bytes, there are 3 null's, 1 NL.
283
non_ref_bytes += len(value) + 3 + 1
284
# one byte for absent if set.
287
elif self.reference_lists:
288
# (ref_lists -1) tabs
289
non_ref_bytes += self.reference_lists - 1
290
# (ref-1 cr's per ref_list)
291
for ref_list in references:
292
# how many references across the whole file?
293
total_references += len(ref_list)
294
# accrue reference separators
296
non_ref_bytes += len(ref_list) - 1
297
# how many digits are needed to represent the total byte count?
299
possible_total_bytes = non_ref_bytes + total_references*digits
300
while 10 ** digits < possible_total_bytes:
302
possible_total_bytes = non_ref_bytes + total_references*digits
303
expected_bytes = possible_total_bytes + 1 # terminating newline
304
# resolve key addresses.
306
for key, non_ref_bytes, total_references in key_offset_info:
307
key_addresses[key] = non_ref_bytes + total_references*digits
309
format_string = '%%0%sd' % digits
310
for key, (absent, references, value) in nodes:
311
flattened_references = []
312
for ref_list in references:
314
for reference in ref_list:
315
ref_addresses.append(format_string % key_addresses[reference])
316
flattened_references.append('\r'.join(ref_addresses))
317
string_key = '\x00'.join(key)
318
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
319
'\t'.join(flattened_references), value))
321
result = StringIO(''.join(lines))
322
if expected_bytes and len(result.getvalue()) != expected_bytes:
323
raise errors.BzrError('Failed index creation. Internal error:'
324
' mismatched output length and expected length: %d %d' %
325
(len(result.getvalue()), expected_bytes))
328
def set_optimize(self, for_size=None, combine_backing_indices=None):
329
"""Change how the builder tries to optimize the result.
331
:param for_size: Tell the builder to try and make the index as small as
333
:param combine_backing_indices: If the builder spills to disk to save
334
memory, should the on-disk indices be combined. Set to True if you
335
are going to be probing the index, but to False if you are not. (If
336
you are not querying, then the time spent combining is wasted.)
339
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
341
if for_size is not None:
342
self._optimize_for_size = for_size
343
if combine_backing_indices is not None:
344
self._combine_backing_indices = combine_backing_indices
346
def find_ancestry(self, keys, ref_list_num):
347
"""See CombinedGraphIndex.find_ancestry()"""
353
for _, key, value, ref_lists in self.iter_entries(pending):
354
parent_keys = ref_lists[ref_list_num]
355
parent_map[key] = parent_keys
356
next_pending.update([p for p in parent_keys if p not in
358
missing_keys.update(pending.difference(parent_map))
359
pending = next_pending
360
return parent_map, missing_keys
363
class GraphIndex(object):
364
"""An index for data with embedded graphs.
366
The index maps keys to a list of key reference lists, and a value.
367
Each node has the same number of key reference lists. Each key reference
368
list can be empty or an arbitrary length. The value is an opaque NULL
369
terminated string without any newlines. The storage of the index is
370
hidden in the interface: keys and key references are always tuples of
371
bytestrings, never the internal representation (e.g. dictionary offsets).
373
It is presumed that the index will not be mutated - it is static data.
375
Successive iter_all_entries calls will read the entire index each time.
376
Additionally, iter_entries calls will read the index linearly until the
377
desired keys are found. XXX: This must be fixed before the index is
378
suitable for production use. :XXX
381
def __init__(self, transport, name, size, unlimited_cache=False):
382
"""Open an index called name on transport.
384
:param transport: A bzrlib.transport.Transport.
385
:param name: A path to provide to transport API calls.
386
:param size: The size of the index in bytes. This is used for bisection
387
logic to perform partial index reads. While the size could be
388
obtained by statting the file this introduced an additional round
389
trip as well as requiring stat'able transports, both of which are
390
avoided by having it supplied. If size is None, then bisection
391
support will be disabled and accessing the index will just stream
394
self._transport = transport
396
# Becomes a dict of key:(value, reference-list-byte-locations) used by
397
# the bisection interface to store parsed but not resolved keys.
398
self._bisect_nodes = None
399
# Becomes a dict of key:(value, reference-list-keys) which are ready to
400
# be returned directly to callers.
402
# a sorted list of slice-addresses for the parsed bytes of the file.
403
# e.g. (0,1) would mean that byte 0 is parsed.
404
self._parsed_byte_map = []
405
# a sorted list of keys matching each slice address for parsed bytes
406
# e.g. (None, 'foo@bar') would mean that the first byte contained no
407
# key, and the end byte of the slice is the of the data for 'foo@bar'
408
self._parsed_key_map = []
409
self._key_count = None
410
self._keys_by_offset = None
411
self._nodes_by_key = None
413
# The number of bytes we've read so far in trying to process this file
416
def __eq__(self, other):
417
"""Equal when self and other were created with the same parameters."""
419
type(self) == type(other) and
420
self._transport == other._transport and
421
self._name == other._name and
422
self._size == other._size)
424
def __ne__(self, other):
425
return not self.__eq__(other)
428
return "%s(%r)" % (self.__class__.__name__,
429
self._transport.abspath(self._name))
431
def _buffer_all(self, stream=None):
432
"""Buffer all the index data.
434
Mutates self._nodes and self.keys_by_offset.
436
if self._nodes is not None:
437
# We already did this
439
if 'index' in debug.debug_flags:
440
mutter('Reading entire index %s', self._transport.abspath(self._name))
442
stream = self._transport.get(self._name)
443
self._read_prefix(stream)
444
self._expected_elements = 3 + self._key_length
446
# raw data keyed by offset
447
self._keys_by_offset = {}
448
# ready-to-return key:value or key:value, node_ref_lists
450
self._nodes_by_key = None
453
lines = stream.read().split('\n')
455
_, _, _, trailers = self._parse_lines(lines, pos)
456
for key, absent, references, value in self._keys_by_offset.itervalues():
459
# resolve references:
460
if self.node_ref_lists:
461
node_value = (value, self._resolve_references(references))
464
self._nodes[key] = node_value
465
# cache the keys for quick set intersections
466
self._keys = set(self._nodes)
468
# there must be one line - the empty trailer line.
469
raise errors.BadIndexData(self)
471
def clear_cache(self):
472
"""Clear out any cached/memoized values.
474
This can be called at any time, but generally it is used when we have
475
extracted some information, but don't expect to be requesting any more
479
def external_references(self, ref_list_num):
480
"""Return references that are not present in this index.
483
if ref_list_num + 1 > self.node_ref_lists:
484
raise ValueError('No ref list %d, index has %d ref lists'
485
% (ref_list_num, self.node_ref_lists))
487
for key, (value, ref_lists) in self._nodes.iteritems():
488
ref_list = ref_lists[ref_list_num]
489
refs.update(ref_list)
490
return refs - self._keys
492
def _get_nodes_by_key(self):
493
if self._nodes_by_key is None:
495
if self.node_ref_lists:
496
for key, (value, references) in self._nodes.iteritems():
497
key_dict = nodes_by_key
498
for subkey in key[:-1]:
499
key_dict = key_dict.setdefault(subkey, {})
500
key_dict[key[-1]] = key, value, references
502
for key, value in self._nodes.iteritems():
503
key_dict = nodes_by_key
504
for subkey in key[:-1]:
505
key_dict = key_dict.setdefault(subkey, {})
506
key_dict[key[-1]] = key, value
507
self._nodes_by_key = nodes_by_key
508
return self._nodes_by_key
510
def iter_all_entries(self):
511
"""Iterate over all keys within the index.
513
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
514
The former tuple is used when there are no reference lists in the
515
index, making the API compatible with simple key:value index types.
516
There is no defined order for the result iteration - it will be in
517
the most efficient order for the index.
519
if 'evil' in debug.debug_flags:
520
trace.mutter_callsite(3,
521
"iter_all_entries scales with size of history.")
522
if self._nodes is None:
524
if self.node_ref_lists:
525
for key, (value, node_ref_lists) in self._nodes.iteritems():
526
yield self, key, value, node_ref_lists
528
for key, value in self._nodes.iteritems():
529
yield self, key, value
531
def _read_prefix(self, stream):
532
signature = stream.read(len(self._signature()))
533
if not signature == self._signature():
534
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
535
options_line = stream.readline()
536
if not options_line.startswith(_OPTION_NODE_REFS):
537
raise errors.BadIndexOptions(self)
539
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
541
raise errors.BadIndexOptions(self)
542
options_line = stream.readline()
543
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
544
raise errors.BadIndexOptions(self)
546
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
548
raise errors.BadIndexOptions(self)
549
options_line = stream.readline()
550
if not options_line.startswith(_OPTION_LEN):
551
raise errors.BadIndexOptions(self)
553
self._key_count = int(options_line[len(_OPTION_LEN):-1])
555
raise errors.BadIndexOptions(self)
557
def _resolve_references(self, references):
558
"""Return the resolved key references for references.
560
References are resolved by looking up the location of the key in the
561
_keys_by_offset map and substituting the key name, preserving ordering.
563
:param references: An iterable of iterables of key locations. e.g.
565
:return: A tuple of tuples of keys.
568
for ref_list in references:
569
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
570
return tuple(node_refs)
572
def _find_index(self, range_map, key):
573
"""Helper for the _parsed_*_index calls.
575
Given a range map - [(start, end), ...], finds the index of the range
576
in the map for key if it is in the map, and if it is not there, the
577
immediately preceeding range in the map.
579
result = bisect_right(range_map, key) - 1
580
if result + 1 < len(range_map):
581
# check the border condition, it may be in result + 1
582
if range_map[result + 1][0] == key[0]:
586
def _parsed_byte_index(self, offset):
587
"""Return the index of the entry immediately before offset.
589
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
590
there is one unparsed byte (the 11th, addressed as[10]). then:
591
asking for 0 will return 0
592
asking for 10 will return 0
593
asking for 11 will return 1
594
asking for 12 will return 1
597
return self._find_index(self._parsed_byte_map, key)
599
def _parsed_key_index(self, key):
600
"""Return the index of the entry immediately before key.
602
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
603
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
604
have been parsed, then:
605
asking for '' will return 0
606
asking for 'a' will return 0
607
asking for 'b' will return 1
608
asking for 'e' will return 1
610
search_key = (key, None)
611
return self._find_index(self._parsed_key_map, search_key)
613
def _is_parsed(self, offset):
614
"""Returns True if offset has been parsed."""
615
index = self._parsed_byte_index(offset)
616
if index == len(self._parsed_byte_map):
617
return offset < self._parsed_byte_map[index - 1][1]
618
start, end = self._parsed_byte_map[index]
619
return offset >= start and offset < end
621
def _iter_entries_from_total_buffer(self, keys):
622
"""Iterate over keys when the entire index is parsed."""
623
keys = keys.intersection(self._keys)
624
if self.node_ref_lists:
626
value, node_refs = self._nodes[key]
627
yield self, key, value, node_refs
630
yield self, key, self._nodes[key]
632
def iter_entries(self, keys):
633
"""Iterate over keys within the index.
635
:param keys: An iterable providing the keys to be retrieved.
636
:return: An iterable as per iter_all_entries, but restricted to the
637
keys supplied. No additional keys will be returned, and every
638
key supplied that is in the index will be returned.
643
if self._size is None and self._nodes is None:
646
# We fit about 20 keys per minimum-read (4K), so if we are looking for
647
# more than 1/20th of the index its likely (assuming homogenous key
648
# spread) that we'll read the entire index. If we're going to do that,
649
# buffer the whole thing. A better analysis might take key spread into
650
# account - but B+Tree indices are better anyway.
651
# We could look at all data read, and use a threshold there, which will
652
# trigger on ancestry walks, but that is not yet fully mapped out.
653
if self._nodes is None and len(keys) * 20 > self.key_count():
655
if self._nodes is not None:
656
return self._iter_entries_from_total_buffer(keys)
658
return (result[1] for result in bisect_multi_bytes(
659
self._lookup_keys_via_location, self._size, keys))
661
def iter_entries_prefix(self, keys):
662
"""Iterate over keys within the index using prefix matching.
664
Prefix matching is applied within the tuple of a key, not to within
665
the bytestring of each key element. e.g. if you have the keys ('foo',
666
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
667
only the former key is returned.
669
WARNING: Note that this method currently causes a full index parse
670
unconditionally (which is reasonably appropriate as it is a means for
671
thunking many small indices into one larger one and still supplies
672
iter_all_entries at the thunk layer).
674
:param keys: An iterable providing the key prefixes to be retrieved.
675
Each key prefix takes the form of a tuple the length of a key, but
676
with the last N elements 'None' rather than a regular bytestring.
677
The first element cannot be 'None'.
678
:return: An iterable as per iter_all_entries, but restricted to the
679
keys with a matching prefix to those supplied. No additional keys
680
will be returned, and every match that is in the index will be
686
# load data - also finds key lengths
687
if self._nodes is None:
689
if self._key_length == 1:
693
raise errors.BadIndexKey(key)
694
if len(key) != self._key_length:
695
raise errors.BadIndexKey(key)
696
if self.node_ref_lists:
697
value, node_refs = self._nodes[key]
698
yield self, key, value, node_refs
700
yield self, key, self._nodes[key]
702
nodes_by_key = self._get_nodes_by_key()
706
raise errors.BadIndexKey(key)
707
if len(key) != self._key_length:
708
raise errors.BadIndexKey(key)
709
# find what it refers to:
710
key_dict = nodes_by_key
712
# find the subdict whose contents should be returned.
714
while len(elements) and elements[0] is not None:
715
key_dict = key_dict[elements[0]]
718
# a non-existant lookup.
723
key_dict = dicts.pop(-1)
724
# can't be empty or would not exist
725
item, value = key_dict.iteritems().next()
726
if type(value) == dict:
728
dicts.extend(key_dict.itervalues())
731
for value in key_dict.itervalues():
732
# each value is the key:value:node refs tuple
734
yield (self, ) + value
736
# the last thing looked up was a terminal element
737
yield (self, ) + key_dict
739
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
740
"""See BTreeIndex._find_ancestors."""
741
# The api can be implemented as a trivial overlay on top of
742
# iter_entries, it is not an efficient implementation, but it at least
746
for index, key, value, refs in self.iter_entries(keys):
747
parent_keys = refs[ref_list_num]
749
parent_map[key] = parent_keys
750
search_keys.update(parent_keys)
751
# Figure out what, if anything, was missing
752
missing_keys.update(set(keys).difference(found_keys))
753
search_keys = search_keys.difference(parent_map)
757
"""Return an estimate of the number of keys in this index.
759
For GraphIndex the estimate is exact.
761
if self._key_count is None:
762
self._read_and_parse([_HEADER_READV])
763
return self._key_count
765
def _lookup_keys_via_location(self, location_keys):
766
"""Public interface for implementing bisection.
768
If _buffer_all has been called, then all the data for the index is in
769
memory, and this method should not be called, as it uses a separate
770
cache because it cannot pre-resolve all indices, which buffer_all does
773
:param location_keys: A list of location(byte offset), key tuples.
774
:return: A list of (location_key, result) tuples as expected by
775
bzrlib.bisect_multi.bisect_multi_bytes.
777
# Possible improvements:
778
# - only bisect lookup each key once
779
# - sort the keys first, and use that to reduce the bisection window
781
# this progresses in three parts:
784
# attempt to answer the question from the now in memory data.
785
# build the readv request
786
# for each location, ask for 800 bytes - much more than rows we've seen
789
for location, key in location_keys:
790
# can we answer from cache?
791
if self._bisect_nodes and key in self._bisect_nodes:
792
# We have the key parsed.
794
index = self._parsed_key_index(key)
795
if (len(self._parsed_key_map) and
796
self._parsed_key_map[index][0] <= key and
797
(self._parsed_key_map[index][1] >= key or
798
# end of the file has been parsed
799
self._parsed_byte_map[index][1] == self._size)):
800
# the key has been parsed, so no lookup is needed even if its
803
# - if we have examined this part of the file already - yes
804
index = self._parsed_byte_index(location)
805
if (len(self._parsed_byte_map) and
806
self._parsed_byte_map[index][0] <= location and
807
self._parsed_byte_map[index][1] > location):
808
# the byte region has been parsed, so no read is needed.
811
if location + length > self._size:
812
length = self._size - location
813
# todo, trim out parsed locations.
815
readv_ranges.append((location, length))
816
# read the header if needed
817
if self._bisect_nodes is None:
818
readv_ranges.append(_HEADER_READV)
819
self._read_and_parse(readv_ranges)
821
if self._nodes is not None:
822
# _read_and_parse triggered a _buffer_all because we requested the
824
for location, key in location_keys:
825
if key not in self._nodes: # not present
826
result.append(((location, key), False))
827
elif self.node_ref_lists:
828
value, refs = self._nodes[key]
829
result.append(((location, key),
830
(self, key, value, refs)))
832
result.append(((location, key),
833
(self, key, self._nodes[key])))
836
# - figure out <, >, missing, present
837
# - result present references so we can return them.
838
# keys that we cannot answer until we resolve references
839
pending_references = []
840
pending_locations = set()
841
for location, key in location_keys:
842
# can we answer from cache?
843
if key in self._bisect_nodes:
844
# the key has been parsed, so no lookup is needed
845
if self.node_ref_lists:
846
# the references may not have been all parsed.
847
value, refs = self._bisect_nodes[key]
848
wanted_locations = []
849
for ref_list in refs:
851
if ref not in self._keys_by_offset:
852
wanted_locations.append(ref)
854
pending_locations.update(wanted_locations)
855
pending_references.append((location, key))
857
result.append(((location, key), (self, key,
858
value, self._resolve_references(refs))))
860
result.append(((location, key),
861
(self, key, self._bisect_nodes[key])))
864
# has the region the key should be in, been parsed?
865
index = self._parsed_key_index(key)
866
if (self._parsed_key_map[index][0] <= key and
867
(self._parsed_key_map[index][1] >= key or
868
# end of the file has been parsed
869
self._parsed_byte_map[index][1] == self._size)):
870
result.append(((location, key), False))
872
# no, is the key above or below the probed location:
873
# get the range of the probed & parsed location
874
index = self._parsed_byte_index(location)
875
# if the key is below the start of the range, its below
876
if key < self._parsed_key_map[index][0]:
880
result.append(((location, key), direction))
882
# lookup data to resolve references
883
for location in pending_locations:
885
if location + length > self._size:
886
length = self._size - location
887
# TODO: trim out parsed locations (e.g. if the 800 is into the
888
# parsed region trim it, and dont use the adjust_for_latency
891
readv_ranges.append((location, length))
892
self._read_and_parse(readv_ranges)
893
if self._nodes is not None:
894
# The _read_and_parse triggered a _buffer_all, grab the data and
896
for location, key in pending_references:
897
value, refs = self._nodes[key]
898
result.append(((location, key), (self, key, value, refs)))
900
for location, key in pending_references:
901
# answer key references we had to look-up-late.
902
value, refs = self._bisect_nodes[key]
903
result.append(((location, key), (self, key,
904
value, self._resolve_references(refs))))
907
def _parse_header_from_bytes(self, bytes):
908
"""Parse the header from a region of bytes.
910
:param bytes: The data to parse.
911
:return: An offset, data tuple such as readv yields, for the unparsed
912
data. (which may length 0).
914
signature = bytes[0:len(self._signature())]
915
if not signature == self._signature():
916
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
917
lines = bytes[len(self._signature()):].splitlines()
918
options_line = lines[0]
919
if not options_line.startswith(_OPTION_NODE_REFS):
920
raise errors.BadIndexOptions(self)
922
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
924
raise errors.BadIndexOptions(self)
925
options_line = lines[1]
926
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
927
raise errors.BadIndexOptions(self)
929
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
931
raise errors.BadIndexOptions(self)
932
options_line = lines[2]
933
if not options_line.startswith(_OPTION_LEN):
934
raise errors.BadIndexOptions(self)
936
self._key_count = int(options_line[len(_OPTION_LEN):])
938
raise errors.BadIndexOptions(self)
939
# calculate the bytes we have processed
940
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
942
self._parsed_bytes(0, None, header_end, None)
943
# setup parsing state
944
self._expected_elements = 3 + self._key_length
945
# raw data keyed by offset
946
self._keys_by_offset = {}
947
# keys with the value and node references
948
self._bisect_nodes = {}
949
return header_end, bytes[header_end:]
951
def _parse_region(self, offset, data):
952
"""Parse node data returned from a readv operation.
954
:param offset: The byte offset the data starts at.
955
:param data: The data to parse.
959
end = offset + len(data)
962
# Trivial test - if the current index's end is within the
963
# low-matching parsed range, we're done.
964
index = self._parsed_byte_index(high_parsed)
965
if end < self._parsed_byte_map[index][1]:
967
# print "[%d:%d]" % (offset, end), \
968
# self._parsed_byte_map[index:index + 2]
969
high_parsed, last_segment = self._parse_segment(
970
offset, data, end, index)
974
def _parse_segment(self, offset, data, end, index):
975
"""Parse one segment of data.
977
:param offset: Where 'data' begins in the file.
978
:param data: Some data to parse a segment of.
979
:param end: Where data ends
980
:param index: The current index into the parsed bytes map.
981
:return: True if the parsed segment is the last possible one in the
983
:return: high_parsed_byte, last_segment.
984
high_parsed_byte is the location of the highest parsed byte in this
985
segment, last_segment is True if the parsed segment is the last
986
possible one in the data block.
988
# default is to use all data
990
# accomodate overlap with data before this.
991
if offset < self._parsed_byte_map[index][1]:
992
# overlaps the lower parsed region
993
# skip the parsed data
994
trim_start = self._parsed_byte_map[index][1] - offset
995
# don't trim the start for \n
996
start_adjacent = True
997
elif offset == self._parsed_byte_map[index][1]:
998
# abuts the lower parsed region
1001
# do not trim anything
1002
start_adjacent = True
1004
# does not overlap the lower parsed region
1007
# but trim the leading \n
1008
start_adjacent = False
1009
if end == self._size:
1010
# lines up to the end of all data:
1013
# do not strip to the last \n
1016
elif index + 1 == len(self._parsed_byte_map):
1017
# at the end of the parsed data
1020
# but strip to the last \n
1021
end_adjacent = False
1023
elif end == self._parsed_byte_map[index + 1][0]:
1024
# buts up against the next parsed region
1027
# do not strip to the last \n
1030
elif end > self._parsed_byte_map[index + 1][0]:
1031
# overlaps into the next parsed region
1032
# only consider the unparsed data
1033
trim_end = self._parsed_byte_map[index + 1][0] - offset
1034
# do not strip to the last \n as we know its an entire record
1036
last_segment = end < self._parsed_byte_map[index + 1][1]
1038
# does not overlap into the next region
1041
# but strip to the last \n
1042
end_adjacent = False
1044
# now find bytes to discard if needed
1045
if not start_adjacent:
1046
# work around python bug in rfind
1047
if trim_start is None:
1048
trim_start = data.find('\n') + 1
1050
trim_start = data.find('\n', trim_start) + 1
1051
if not (trim_start != 0):
1052
raise AssertionError('no \n was present')
1053
# print 'removing start', offset, trim_start, repr(data[:trim_start])
1054
if not end_adjacent:
1055
# work around python bug in rfind
1056
if trim_end is None:
1057
trim_end = data.rfind('\n') + 1
1059
trim_end = data.rfind('\n', None, trim_end) + 1
1060
if not (trim_end != 0):
1061
raise AssertionError('no \n was present')
1062
# print 'removing end', offset, trim_end, repr(data[trim_end:])
1063
# adjust offset and data to the parseable data.
1064
trimmed_data = data[trim_start:trim_end]
1065
if not (trimmed_data):
1066
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
1067
% (trim_start, trim_end, offset, offset + len(data)))
1069
offset += trim_start
1070
# print "parsing", repr(trimmed_data)
1071
# splitlines mangles the \r delimiters.. don't use it.
1072
lines = trimmed_data.split('\n')
1075
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
1076
for key, value in nodes:
1077
self._bisect_nodes[key] = value
1078
self._parsed_bytes(offset, first_key,
1079
offset + len(trimmed_data), last_key)
1080
return offset + len(trimmed_data), last_segment
1082
def _parse_lines(self, lines, pos):
1089
# must be at the end
1091
if not (self._size == pos + 1):
1092
raise AssertionError("%s %s" % (self._size, pos))
1095
elements = line.split('\0')
1096
if len(elements) != self._expected_elements:
1097
raise errors.BadIndexData(self)
1098
# keys are tuples. Each element is a string that may occur many
1099
# times, so we intern them to save space. AB, RC, 200807
1100
key = tuple([intern(element) for element in elements[:self._key_length]])
1101
if first_key is None:
1103
absent, references, value = elements[-3:]
1105
for ref_string in references.split('\t'):
1106
ref_lists.append(tuple([
1107
int(ref) for ref in ref_string.split('\r') if ref
1109
ref_lists = tuple(ref_lists)
1110
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1111
pos += len(line) + 1 # +1 for the \n
1114
if self.node_ref_lists:
1115
node_value = (value, ref_lists)
1118
nodes.append((key, node_value))
1119
# print "parsed ", key
1120
return first_key, key, nodes, trailers
1122
def _parsed_bytes(self, start, start_key, end, end_key):
1123
"""Mark the bytes from start to end as parsed.
1125
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1128
:param start: The start of the parsed region.
1129
:param end: The end of the parsed region.
1131
index = self._parsed_byte_index(start)
1132
new_value = (start, end)
1133
new_key = (start_key, end_key)
1135
# first range parsed is always the beginning.
1136
self._parsed_byte_map.insert(index, new_value)
1137
self._parsed_key_map.insert(index, new_key)
1141
# extend lower region
1142
# extend higher region
1143
# combine two regions
1144
if (index + 1 < len(self._parsed_byte_map) and
1145
self._parsed_byte_map[index][1] == start and
1146
self._parsed_byte_map[index + 1][0] == end):
1147
# combine two regions
1148
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1149
self._parsed_byte_map[index + 1][1])
1150
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1151
self._parsed_key_map[index + 1][1])
1152
del self._parsed_byte_map[index + 1]
1153
del self._parsed_key_map[index + 1]
1154
elif self._parsed_byte_map[index][1] == start:
1155
# extend the lower entry
1156
self._parsed_byte_map[index] = (
1157
self._parsed_byte_map[index][0], end)
1158
self._parsed_key_map[index] = (
1159
self._parsed_key_map[index][0], end_key)
1160
elif (index + 1 < len(self._parsed_byte_map) and
1161
self._parsed_byte_map[index + 1][0] == end):
1162
# extend the higher entry
1163
self._parsed_byte_map[index + 1] = (
1164
start, self._parsed_byte_map[index + 1][1])
1165
self._parsed_key_map[index + 1] = (
1166
start_key, self._parsed_key_map[index + 1][1])
1169
self._parsed_byte_map.insert(index + 1, new_value)
1170
self._parsed_key_map.insert(index + 1, new_key)
1172
def _read_and_parse(self, readv_ranges):
1173
"""Read the ranges and parse the resulting data.
1175
:param readv_ranges: A prepared readv range list.
1177
if not readv_ranges:
1179
if self._nodes is None and self._bytes_read * 2 >= self._size:
1180
# We've already read more than 50% of the file and we are about to
1181
# request more data, just _buffer_all() and be done
1185
readv_data = self._transport.readv(self._name, readv_ranges, True,
1188
for offset, data in readv_data:
1189
self._bytes_read += len(data)
1190
if offset == 0 and len(data) == self._size:
1191
# We read the whole range, most likely because the
1192
# Transport upcast our readv ranges into one long request
1193
# for enough total data to grab the whole index.
1194
self._buffer_all(StringIO(data))
1196
if self._bisect_nodes is None:
1197
# this must be the start
1198
if not (offset == 0):
1199
raise AssertionError()
1200
offset, data = self._parse_header_from_bytes(data)
1201
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1202
self._parse_region(offset, data)
1204
def _signature(self):
1205
"""The file signature for this index type."""
1209
"""Validate that everything in the index can be accessed."""
1210
# iter_all validates completely at the moment, so just do that.
1211
for node in self.iter_all_entries():
1215
class CombinedGraphIndex(object):
1216
"""A GraphIndex made up from smaller GraphIndices.
1218
The backing indices must implement GraphIndex, and are presumed to be
1221
Queries against the combined index will be made against the first index,
1222
and then the second and so on. The order of index's can thus influence
1223
performance significantly. For example, if one index is on local disk and a
1224
second on a remote server, the local disk index should be before the other
1228
def __init__(self, indices, reload_func=None):
1229
"""Create a CombinedGraphIndex backed by indices.
1231
:param indices: An ordered list of indices to query for data.
1232
:param reload_func: A function to call if we find we are missing an
1233
index. Should have the form reload_func() => True/False to indicate
1234
if reloading actually changed anything.
1236
self._indices = indices
1237
self._reload_func = reload_func
1241
self.__class__.__name__,
1242
', '.join(map(repr, self._indices)))
1244
def clear_cache(self):
1245
"""See GraphIndex.clear_cache()"""
1246
for index in self._indices:
1249
def get_parent_map(self, keys):
1250
"""See graph.StackedParentsProvider.get_parent_map"""
1251
search_keys = set(keys)
1252
if NULL_REVISION in search_keys:
1253
search_keys.discard(NULL_REVISION)
1254
found_parents = {NULL_REVISION:[]}
1257
for index, key, value, refs in self.iter_entries(search_keys):
1260
parents = (NULL_REVISION,)
1261
found_parents[key] = parents
1262
return found_parents
1264
has_key = _has_key_from_parent_map
1266
def insert_index(self, pos, index):
1267
"""Insert a new index in the list of indices to query.
1269
:param pos: The position to insert the index.
1270
:param index: The index to insert.
1272
self._indices.insert(pos, index)
1274
def iter_all_entries(self):
1275
"""Iterate over all keys within the index
1277
Duplicate keys across child indices are presumed to have the same
1278
value and are only reported once.
1280
:return: An iterable of (index, key, reference_lists, value).
1281
There is no defined order for the result iteration - it will be in
1282
the most efficient order for the index.
1287
for index in self._indices:
1288
for node in index.iter_all_entries():
1289
if node[1] not in seen_keys:
1291
seen_keys.add(node[1])
1293
except errors.NoSuchFile:
1294
self._reload_or_raise()
1296
def iter_entries(self, keys):
1297
"""Iterate over keys within the index.
1299
Duplicate keys across child indices are presumed to have the same
1300
value and are only reported once.
1302
:param keys: An iterable providing the keys to be retrieved.
1303
:return: An iterable of (index, key, reference_lists, value). There is no
1304
defined order for the result iteration - it will be in the most
1305
efficient order for the index.
1310
for index in self._indices:
1313
for node in index.iter_entries(keys):
1314
keys.remove(node[1])
1317
except errors.NoSuchFile:
1318
self._reload_or_raise()
1320
def iter_entries_prefix(self, keys):
1321
"""Iterate over keys within the index using prefix matching.
1323
Duplicate keys across child indices are presumed to have the same
1324
value and are only reported once.
1326
Prefix matching is applied within the tuple of a key, not to within
1327
the bytestring of each key element. e.g. if you have the keys ('foo',
1328
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1329
only the former key is returned.
1331
:param keys: An iterable providing the key prefixes to be retrieved.
1332
Each key prefix takes the form of a tuple the length of a key, but
1333
with the last N elements 'None' rather than a regular bytestring.
1334
The first element cannot be 'None'.
1335
:return: An iterable as per iter_all_entries, but restricted to the
1336
keys with a matching prefix to those supplied. No additional keys
1337
will be returned, and every match that is in the index will be
1346
for index in self._indices:
1347
for node in index.iter_entries_prefix(keys):
1348
if node[1] in seen_keys:
1350
seen_keys.add(node[1])
1353
except errors.NoSuchFile:
1354
self._reload_or_raise()
1356
def find_ancestry(self, keys, ref_list_num):
1357
"""Find the complete ancestry for the given set of keys.
1359
Note that this is a whole-ancestry request, so it should be used
1362
:param keys: An iterable of keys to look for
1363
:param ref_list_num: The reference list which references the parents
1365
:return: (parent_map, missing_keys)
1367
missing_keys = set()
1369
keys_to_lookup = set(keys)
1371
while keys_to_lookup:
1372
# keys that *all* indexes claim are missing, stop searching them
1374
all_index_missing = None
1375
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1376
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1378
# len(missing_keys))
1379
for index_idx, index in enumerate(self._indices):
1380
# TODO: we should probably be doing something with
1381
# 'missing_keys' since we've already determined that
1382
# those revisions have not been found anywhere
1383
index_missing_keys = set()
1384
# Find all of the ancestry we can from this index
1385
# keep looking until the search_keys set is empty, which means
1386
# things we didn't find should be in index_missing_keys
1387
search_keys = keys_to_lookup
1389
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1390
# index_idx, len(search_keys),
1391
# len(parent_map), len(index_missing_keys))
1394
# TODO: ref_list_num should really be a parameter, since
1395
# CombinedGraphIndex does not know what the ref lists
1397
search_keys = index._find_ancestors(search_keys,
1398
ref_list_num, parent_map, index_missing_keys)
1399
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1400
# sub_generation, len(search_keys),
1401
# len(parent_map), len(index_missing_keys))
1402
# Now set whatever was missing to be searched in the next index
1403
keys_to_lookup = index_missing_keys
1404
if all_index_missing is None:
1405
all_index_missing = set(index_missing_keys)
1407
all_index_missing.intersection_update(index_missing_keys)
1408
if not keys_to_lookup:
1410
if all_index_missing is None:
1411
# There were no indexes, so all search keys are 'missing'
1412
missing_keys.update(keys_to_lookup)
1413
keys_to_lookup = None
1415
missing_keys.update(all_index_missing)
1416
keys_to_lookup.difference_update(all_index_missing)
1417
return parent_map, missing_keys
1419
def key_count(self):
1420
"""Return an estimate of the number of keys in this index.
1422
For CombinedGraphIndex this is approximated by the sum of the keys of
1423
the child indices. As child indices may have duplicate keys this can
1424
have a maximum error of the number of child indices * largest number of
1429
return sum((index.key_count() for index in self._indices), 0)
1430
except errors.NoSuchFile:
1431
self._reload_or_raise()
1433
missing_keys = _missing_keys_from_parent_map
1435
def _reload_or_raise(self):
1436
"""We just got a NoSuchFile exception.
1438
Try to reload the indices, if it fails, just raise the current
1441
if self._reload_func is None:
1443
exc_type, exc_value, exc_traceback = sys.exc_info()
1444
trace.mutter('Trying to reload after getting exception: %s',
1446
if not self._reload_func():
1447
# We tried to reload, but nothing changed, so we fail anyway
1448
trace.mutter('_reload_func indicated nothing has changed.'
1449
' Raising original exception.')
1450
raise exc_type, exc_value, exc_traceback
1453
"""Validate that everything in the index can be accessed."""
1456
for index in self._indices:
1459
except errors.NoSuchFile:
1460
self._reload_or_raise()
1463
class InMemoryGraphIndex(GraphIndexBuilder):
1464
"""A GraphIndex which operates entirely out of memory and is mutable.
1466
This is designed to allow the accumulation of GraphIndex entries during a
1467
single write operation, where the accumulated entries need to be immediately
1468
available - for example via a CombinedGraphIndex.
1471
def add_nodes(self, nodes):
1472
"""Add nodes to the index.
1474
:param nodes: An iterable of (key, node_refs, value) entries to add.
1476
if self.reference_lists:
1477
for (key, value, node_refs) in nodes:
1478
self.add_node(key, value, node_refs)
1480
for (key, value) in nodes:
1481
self.add_node(key, value)
1483
def iter_all_entries(self):
1484
"""Iterate over all keys within the index
1486
:return: An iterable of (index, key, reference_lists, value). There is no
1487
defined order for the result iteration - it will be in the most
1488
efficient order for the index (in this case dictionary hash order).
1490
if 'evil' in debug.debug_flags:
1491
trace.mutter_callsite(3,
1492
"iter_all_entries scales with size of history.")
1493
if self.reference_lists:
1494
for key, (absent, references, value) in self._nodes.iteritems():
1496
yield self, key, value, references
1498
for key, (absent, references, value) in self._nodes.iteritems():
1500
yield self, key, value
1502
def iter_entries(self, keys):
1503
"""Iterate over keys within the index.
1505
:param keys: An iterable providing the keys to be retrieved.
1506
:return: An iterable of (index, key, value, reference_lists). There is no
1507
defined order for the result iteration - it will be in the most
1508
efficient order for the index (keys iteration order in this case).
1511
if self.reference_lists:
1512
for key in keys.intersection(self._keys):
1513
node = self._nodes[key]
1515
yield self, key, node[2], node[1]
1517
for key in keys.intersection(self._keys):
1518
node = self._nodes[key]
1520
yield self, key, node[2]
1522
def iter_entries_prefix(self, keys):
1523
"""Iterate over keys within the index using prefix matching.
1525
Prefix matching is applied within the tuple of a key, not to within
1526
the bytestring of each key element. e.g. if you have the keys ('foo',
1527
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1528
only the former key is returned.
1530
:param keys: An iterable providing the key prefixes to be retrieved.
1531
Each key prefix takes the form of a tuple the length of a key, but
1532
with the last N elements 'None' rather than a regular bytestring.
1533
The first element cannot be 'None'.
1534
:return: An iterable as per iter_all_entries, but restricted to the
1535
keys with a matching prefix to those supplied. No additional keys
1536
will be returned, and every match that is in the index will be
1539
# XXX: To much duplication with the GraphIndex class; consider finding
1540
# a good place to pull out the actual common logic.
1544
if self._key_length == 1:
1548
raise errors.BadIndexKey(key)
1549
if len(key) != self._key_length:
1550
raise errors.BadIndexKey(key)
1551
node = self._nodes[key]
1554
if self.reference_lists:
1555
yield self, key, node[2], node[1]
1557
yield self, key, node[2]
1559
nodes_by_key = self._get_nodes_by_key()
1563
raise errors.BadIndexKey(key)
1564
if len(key) != self._key_length:
1565
raise errors.BadIndexKey(key)
1566
# find what it refers to:
1567
key_dict = nodes_by_key
1568
elements = list(key)
1569
# find the subdict to return
1571
while len(elements) and elements[0] is not None:
1572
key_dict = key_dict[elements[0]]
1575
# a non-existant lookup.
1580
key_dict = dicts.pop(-1)
1581
# can't be empty or would not exist
1582
item, value = key_dict.iteritems().next()
1583
if type(value) == dict:
1585
dicts.extend(key_dict.itervalues())
1588
for value in key_dict.itervalues():
1589
yield (self, ) + value
1591
yield (self, ) + key_dict
1593
def key_count(self):
1594
"""Return an estimate of the number of keys in this index.
1596
For InMemoryGraphIndex the estimate is exact.
1598
return len(self._keys)
1601
"""In memory index's have no known corruption at the moment."""
1604
class GraphIndexPrefixAdapter(object):
1605
"""An adapter between GraphIndex with different key lengths.
1607
Queries against this will emit queries against the adapted Graph with the
1608
prefix added, queries for all items use iter_entries_prefix. The returned
1609
nodes will have their keys and node references adjusted to remove the
1610
prefix. Finally, an add_nodes_callback can be supplied - when called the
1611
nodes and references being added will have prefix prepended.
1614
def __init__(self, adapted, prefix, missing_key_length,
1615
add_nodes_callback=None):
1616
"""Construct an adapter against adapted with prefix."""
1617
self.adapted = adapted
1618
self.prefix_key = prefix + (None,)*missing_key_length
1619
self.prefix = prefix
1620
self.prefix_len = len(prefix)
1621
self.add_nodes_callback = add_nodes_callback
1623
def add_nodes(self, nodes):
1624
"""Add nodes to the index.
1626
:param nodes: An iterable of (key, node_refs, value) entries to add.
1628
# save nodes in case its an iterator
1629
nodes = tuple(nodes)
1630
translated_nodes = []
1632
# Add prefix_key to each reference node_refs is a tuple of tuples,
1633
# so split it apart, and add prefix_key to the internal reference
1634
for (key, value, node_refs) in nodes:
1635
adjusted_references = (
1636
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1637
for ref_list in node_refs))
1638
translated_nodes.append((self.prefix + key, value,
1639
adjusted_references))
1641
# XXX: TODO add an explicit interface for getting the reference list
1642
# status, to handle this bit of user-friendliness in the API more
1644
for (key, value) in nodes:
1645
translated_nodes.append((self.prefix + key, value))
1646
self.add_nodes_callback(translated_nodes)
1648
def add_node(self, key, value, references=()):
1649
"""Add a node to the index.
1651
:param key: The key. keys are non-empty tuples containing
1652
as many whitespace-free utf8 bytestrings as the key length
1653
defined for this index.
1654
:param references: An iterable of iterables of keys. Each is a
1655
reference to another key.
1656
:param value: The value to associate with the key. It may be any
1657
bytes as long as it does not contain \0 or \n.
1659
self.add_nodes(((key, value, references), ))
1661
def _strip_prefix(self, an_iter):
1662
"""Strip prefix data from nodes and return it."""
1663
for node in an_iter:
1665
if node[1][:self.prefix_len] != self.prefix:
1666
raise errors.BadIndexData(self)
1667
for ref_list in node[3]:
1668
for ref_node in ref_list:
1669
if ref_node[:self.prefix_len] != self.prefix:
1670
raise errors.BadIndexData(self)
1671
yield node[0], node[1][self.prefix_len:], node[2], (
1672
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1673
for ref_list in node[3]))
1675
def iter_all_entries(self):
1676
"""Iterate over all keys within the index
1678
iter_all_entries is implemented against the adapted index using
1679
iter_entries_prefix.
1681
:return: An iterable of (index, key, reference_lists, value). There is no
1682
defined order for the result iteration - it will be in the most
1683
efficient order for the index (in this case dictionary hash order).
1685
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1687
def iter_entries(self, keys):
1688
"""Iterate over keys within the index.
1690
:param keys: An iterable providing the keys to be retrieved.
1691
:return: An iterable of (index, key, value, reference_lists). There is no
1692
defined order for the result iteration - it will be in the most
1693
efficient order for the index (keys iteration order in this case).
1695
return self._strip_prefix(self.adapted.iter_entries(
1696
self.prefix + key for key in keys))
1698
def iter_entries_prefix(self, keys):
1699
"""Iterate over keys within the index using prefix matching.
1701
Prefix matching is applied within the tuple of a key, not to within
1702
the bytestring of each key element. e.g. if you have the keys ('foo',
1703
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1704
only the former key is returned.
1706
:param keys: An iterable providing the key prefixes to be retrieved.
1707
Each key prefix takes the form of a tuple the length of a key, but
1708
with the last N elements 'None' rather than a regular bytestring.
1709
The first element cannot be 'None'.
1710
:return: An iterable as per iter_all_entries, but restricted to the
1711
keys with a matching prefix to those supplied. No additional keys
1712
will be returned, and every match that is in the index will be
1715
return self._strip_prefix(self.adapted.iter_entries_prefix(
1716
self.prefix + key for key in keys))
1718
def key_count(self):
1719
"""Return an estimate of the number of keys in this index.
1721
For GraphIndexPrefixAdapter this is relatively expensive - key
1722
iteration with the prefix is done.
1724
return len(list(self.iter_all_entries()))
1727
"""Call the adapted's validate."""
1728
self.adapted.validate()
added
removed
bzrlib/index.py
