1
# Copyright (C) 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
21
from bisect import bisect_right
36
from bzrlib.index import _OPTION_NODE_REFS, _OPTION_KEY_ELEMENTS, _OPTION_LEN
37
from bzrlib.transport import get_transport
40
_BTSIGNATURE = "B+Tree Graph Index 2\n"
41
_OPTION_ROW_LENGTHS = "row_lengths="
42
_LEAF_FLAG = "type=leaf\n"
43
_INTERNAL_FLAG = "type=internal\n"
44
_INTERNAL_OFFSET = "offset="
46
_RESERVED_HEADER_BYTES = 120
49
# 4K per page: 4MB - 1000 entries
50
_NODE_CACHE_SIZE = 1000
53
class _BuilderRow(object):
54
"""The stored state accumulated while writing out a row in the index.
56
:ivar spool: A temporary file used to accumulate nodes for this row
58
:ivar nodes: The count of nodes emitted so far.
62
"""Create a _BuilderRow."""
64
self.spool = None# tempfile.TemporaryFile(prefix='bzr-index-row-')
67
def finish_node(self, pad=True):
68
byte_lines, _, padding = self.writer.finish()
70
self.spool = cStringIO.StringIO()
72
self.spool.write("\x00" * _RESERVED_HEADER_BYTES)
74
# We got bigger than 1 node, switch to a temp file
75
spool = tempfile.TemporaryFile(prefix='bzr-index-row-')
76
spool.write(self.spool.getvalue())
79
if not pad and padding:
81
skipped_bytes = padding
82
self.spool.writelines(byte_lines)
83
remainder = (self.spool.tell() + skipped_bytes) % _PAGE_SIZE
85
raise AssertionError("incorrect node length: %d, %d"
86
% (self.spool.tell(), remainder))
91
class _InternalBuilderRow(_BuilderRow):
92
"""The stored state accumulated while writing out internal rows."""
94
def finish_node(self, pad=True):
96
raise AssertionError("Must pad internal nodes only.")
97
_BuilderRow.finish_node(self)
100
class _LeafBuilderRow(_BuilderRow):
101
"""The stored state accumulated while writing out a leaf rows."""
104
class BTreeBuilder(index.GraphIndexBuilder):
105
"""A Builder for B+Tree based Graph indices.
107
The resulting graph has the structure:
109
_SIGNATURE OPTIONS NODES
110
_SIGNATURE := 'B+Tree Graph Index 1' NEWLINE
111
OPTIONS := REF_LISTS KEY_ELEMENTS LENGTH
112
REF_LISTS := 'node_ref_lists=' DIGITS NEWLINE
113
KEY_ELEMENTS := 'key_elements=' DIGITS NEWLINE
114
LENGTH := 'len=' DIGITS NEWLINE
115
ROW_LENGTHS := 'row_lengths' DIGITS (COMMA DIGITS)*
116
NODES := NODE_COMPRESSED*
117
NODE_COMPRESSED:= COMPRESSED_BYTES{4096}
118
NODE_RAW := INTERNAL | LEAF
119
INTERNAL := INTERNAL_FLAG POINTERS
120
LEAF := LEAF_FLAG ROWS
121
KEY_ELEMENT := Not-whitespace-utf8
122
KEY := KEY_ELEMENT (NULL KEY_ELEMENT)*
124
ROW := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
126
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
127
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
129
VALUE := no-newline-no-null-bytes
132
def __init__(self, reference_lists=0, key_elements=1, spill_at=100000):
133
"""See GraphIndexBuilder.__init__.
135
:param spill_at: Optional parameter controlling the maximum number
136
of nodes that BTreeBuilder will hold in memory.
138
index.GraphIndexBuilder.__init__(self, reference_lists=reference_lists,
139
key_elements=key_elements)
140
self._spill_at = spill_at
141
self._backing_indices = []
142
# A map of {key: (node_refs, value)}
144
# Indicate it hasn't been built yet
145
self._nodes_by_key = None
146
self._optimize_for_size = False
148
def add_node(self, key, value, references=()):
149
"""Add a node to the index.
151
If adding the node causes the builder to reach its spill_at threshold,
152
disk spilling will be triggered.
154
:param key: The key. keys are non-empty tuples containing
155
as many whitespace-free utf8 bytestrings as the key length
156
defined for this index.
157
:param references: An iterable of iterables of keys. Each is a
158
reference to another key.
159
:param value: The value to associate with the key. It may be any
160
bytes as long as it does not contain \0 or \n.
162
# we don't care about absent_references
163
node_refs, _ = self._check_key_ref_value(key, references, value)
164
if key in self._nodes:
165
raise errors.BadIndexDuplicateKey(key, self)
166
self._nodes[key] = (node_refs, value)
168
if self._nodes_by_key is not None and self._key_length > 1:
169
self._update_nodes_by_key(key, value, node_refs)
170
if len(self._keys) < self._spill_at:
172
self._spill_mem_keys_to_disk()
174
def _spill_mem_keys_to_disk(self):
175
"""Write the in memory keys down to disk to cap memory consumption.
177
If we already have some keys written to disk, we will combine them so
178
as to preserve the sorted order. The algorithm for combining uses
179
powers of two. So on the first spill, write all mem nodes into a
180
single index. On the second spill, combine the mem nodes with the nodes
181
on disk to create a 2x sized disk index and get rid of the first index.
182
On the third spill, create a single new disk index, which will contain
183
the mem nodes, and preserve the existing 2x sized index. On the fourth,
184
combine mem with the first and second indexes, creating a new one of
185
size 4x. On the fifth create a single new one, etc.
187
if self._combine_backing_indices:
188
(new_backing_file, size,
189
backing_pos) = self._spill_mem_keys_and_combine()
191
new_backing_file, size = self._spill_mem_keys_without_combining()
192
# Note: The transport here isn't strictly needed, because we will use
193
# direct access to the new_backing._file object
194
new_backing = BTreeGraphIndex(get_transport('.'), '<temp>', size)
195
# GC will clean up the file
196
new_backing._file = new_backing_file
197
if self._combine_backing_indices:
198
if len(self._backing_indices) == backing_pos:
199
self._backing_indices.append(None)
200
self._backing_indices[backing_pos] = new_backing
201
for backing_pos in range(backing_pos):
202
self._backing_indices[backing_pos] = None
204
self._backing_indices.append(new_backing)
207
self._nodes_by_key = None
209
def _spill_mem_keys_without_combining(self):
210
return self._write_nodes(self._iter_mem_nodes(), allow_optimize=False)
212
def _spill_mem_keys_and_combine(self):
213
iterators_to_combine = [self._iter_mem_nodes()]
215
for pos, backing in enumerate(self._backing_indices):
219
iterators_to_combine.append(backing.iter_all_entries())
220
backing_pos = pos + 1
221
new_backing_file, size = \
222
self._write_nodes(self._iter_smallest(iterators_to_combine),
223
allow_optimize=False)
224
return new_backing_file, size, backing_pos
226
def add_nodes(self, nodes):
227
"""Add nodes to the index.
229
:param nodes: An iterable of (key, node_refs, value) entries to add.
231
if self.reference_lists:
232
for (key, value, node_refs) in nodes:
233
self.add_node(key, value, node_refs)
235
for (key, value) in nodes:
236
self.add_node(key, value)
238
def _iter_mem_nodes(self):
239
"""Iterate over the nodes held in memory."""
241
if self.reference_lists:
242
for key in sorted(nodes):
243
references, value = nodes[key]
244
yield self, key, value, references
246
for key in sorted(nodes):
247
references, value = nodes[key]
248
yield self, key, value
250
def _iter_smallest(self, iterators_to_combine):
251
if len(iterators_to_combine) == 1:
252
for value in iterators_to_combine[0]:
256
for iterator in iterators_to_combine:
258
current_values.append(iterator.next())
259
except StopIteration:
260
current_values.append(None)
263
# Decorate candidates with the value to allow 2.4's min to be used.
264
candidates = [(item[1][1], item) for item
265
in enumerate(current_values) if item[1] is not None]
266
if not len(candidates):
268
selected = min(candidates)
269
# undecorate back to (pos, node)
270
selected = selected[1]
271
if last == selected[1][1]:
272
raise errors.BadIndexDuplicateKey(last, self)
273
last = selected[1][1]
274
# Yield, with self as the index
275
yield (self,) + selected[1][1:]
278
current_values[pos] = iterators_to_combine[pos].next()
279
except StopIteration:
280
current_values[pos] = None
282
def _add_key(self, string_key, line, rows, allow_optimize=True):
283
"""Add a key to the current chunk.
285
:param string_key: The key to add.
286
:param line: The fully serialised key and value.
287
:param allow_optimize: If set to False, prevent setting the optimize
288
flag when writing out. This is used by the _spill_mem_keys_to_disk
291
if rows[-1].writer is None:
292
# opening a new leaf chunk;
293
for pos, internal_row in enumerate(rows[:-1]):
294
# flesh out any internal nodes that are needed to
295
# preserve the height of the tree
296
if internal_row.writer is None:
298
if internal_row.nodes == 0:
299
length -= _RESERVED_HEADER_BYTES # padded
301
optimize_for_size = self._optimize_for_size
303
optimize_for_size = False
304
internal_row.writer = chunk_writer.ChunkWriter(length, 0,
305
optimize_for_size=optimize_for_size)
306
internal_row.writer.write(_INTERNAL_FLAG)
307
internal_row.writer.write(_INTERNAL_OFFSET +
308
str(rows[pos + 1].nodes) + "\n")
311
if rows[-1].nodes == 0:
312
length -= _RESERVED_HEADER_BYTES # padded
313
rows[-1].writer = chunk_writer.ChunkWriter(length,
314
optimize_for_size=self._optimize_for_size)
315
rows[-1].writer.write(_LEAF_FLAG)
316
if rows[-1].writer.write(line):
317
# this key did not fit in the node:
318
rows[-1].finish_node()
319
key_line = string_key + "\n"
321
for row in reversed(rows[:-1]):
322
# Mark the start of the next node in the node above. If it
323
# doesn't fit then propagate upwards until we find one that
325
if row.writer.write(key_line):
328
# We've found a node that can handle the pointer.
331
# If we reached the current root without being able to mark the
332
# division point, then we need a new root:
335
if 'index' in debug.debug_flags:
336
trace.mutter('Inserting new global row.')
337
new_row = _InternalBuilderRow()
339
rows.insert(0, new_row)
340
# This will be padded, hence the -100
341
new_row.writer = chunk_writer.ChunkWriter(
342
_PAGE_SIZE - _RESERVED_HEADER_BYTES,
344
optimize_for_size=self._optimize_for_size)
345
new_row.writer.write(_INTERNAL_FLAG)
346
new_row.writer.write(_INTERNAL_OFFSET +
347
str(rows[1].nodes - 1) + "\n")
348
new_row.writer.write(key_line)
349
self._add_key(string_key, line, rows, allow_optimize=allow_optimize)
351
def _write_nodes(self, node_iterator, allow_optimize=True):
352
"""Write node_iterator out as a B+Tree.
354
:param node_iterator: An iterator of sorted nodes. Each node should
355
match the output given by iter_all_entries.
356
:param allow_optimize: If set to False, prevent setting the optimize
357
flag when writing out. This is used by the _spill_mem_keys_to_disk
359
:return: A file handle for a temporary file containing a B+Tree for
362
# The index rows - rows[0] is the root, rows[1] is the layer under it
365
# forward sorted by key. In future we may consider topological sorting,
366
# at the cost of table scans for direct lookup, or a second index for
369
# A stack with the number of nodes of each size. 0 is the root node
370
# and must always be 1 (if there are any nodes in the tree).
371
self.row_lengths = []
372
# Loop over all nodes adding them to the bottom row
373
# (rows[-1]). When we finish a chunk in a row,
374
# propagate the key that didn't fit (comes after the chunk) to the
375
# row above, transitively.
376
for node in node_iterator:
378
# First key triggers the first row
379
rows.append(_LeafBuilderRow())
381
string_key, line = _btree_serializer._flatten_node(node,
382
self.reference_lists)
383
self._add_key(string_key, line, rows, allow_optimize=allow_optimize)
384
for row in reversed(rows):
385
pad = (type(row) != _LeafBuilderRow)
386
row.finish_node(pad=pad)
387
lines = [_BTSIGNATURE]
388
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
389
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
390
lines.append(_OPTION_LEN + str(key_count) + '\n')
391
row_lengths = [row.nodes for row in rows]
392
lines.append(_OPTION_ROW_LENGTHS + ','.join(map(str, row_lengths)) + '\n')
393
if row_lengths and row_lengths[-1] > 1:
394
result = tempfile.NamedTemporaryFile(prefix='bzr-index-')
396
result = cStringIO.StringIO()
397
result.writelines(lines)
398
position = sum(map(len, lines))
400
if position > _RESERVED_HEADER_BYTES:
401
raise AssertionError("Could not fit the header in the"
402
" reserved space: %d > %d"
403
% (position, _RESERVED_HEADER_BYTES))
404
# write the rows out:
406
reserved = _RESERVED_HEADER_BYTES # reserved space for first node
409
# copy nodes to the finalised file.
410
# Special case the first node as it may be prefixed
411
node = row.spool.read(_PAGE_SIZE)
412
result.write(node[reserved:])
413
result.write("\x00" * (reserved - position))
414
position = 0 # Only the root row actually has an offset
415
copied_len = osutils.pumpfile(row.spool, result)
416
if copied_len != (row.nodes - 1) * _PAGE_SIZE:
417
if type(row) != _LeafBuilderRow:
418
raise AssertionError("Incorrect amount of data copied"
419
" expected: %d, got: %d"
420
% ((row.nodes - 1) * _PAGE_SIZE,
428
"""Finalise the index.
430
:return: A file handle for a temporary file containing the nodes added
433
return self._write_nodes(self.iter_all_entries())[0]
435
def iter_all_entries(self):
436
"""Iterate over all keys within the index
438
:return: An iterable of (index, key, value, reference_lists). There is
439
no defined order for the result iteration - it will be in the most
440
efficient order for the index (in this case dictionary hash order).
442
if 'evil' in debug.debug_flags:
443
trace.mutter_callsite(3,
444
"iter_all_entries scales with size of history.")
445
# Doing serial rather than ordered would be faster; but this shouldn't
446
# be getting called routinely anyway.
447
iterators = [self._iter_mem_nodes()]
448
for backing in self._backing_indices:
449
if backing is not None:
450
iterators.append(backing.iter_all_entries())
451
if len(iterators) == 1:
453
return self._iter_smallest(iterators)
455
def iter_entries(self, keys):
456
"""Iterate over keys within the index.
458
:param keys: An iterable providing the keys to be retrieved.
459
:return: An iterable of (index, key, value, reference_lists). There is no
460
defined order for the result iteration - it will be in the most
461
efficient order for the index (keys iteration order in this case).
464
local_keys = keys.intersection(self._keys)
465
if self.reference_lists:
466
for key in local_keys:
467
node = self._nodes[key]
468
yield self, key, node[1], node[0]
470
for key in local_keys:
471
node = self._nodes[key]
472
yield self, key, node[1]
473
# Find things that are in backing indices that have not been handled
475
if not self._backing_indices:
476
return # We won't find anything there either
477
# Remove all of the keys that we found locally
478
keys.difference_update(local_keys)
479
for backing in self._backing_indices:
484
for node in backing.iter_entries(keys):
486
yield (self,) + node[1:]
488
def iter_entries_prefix(self, keys):
489
"""Iterate over keys within the index using prefix matching.
491
Prefix matching is applied within the tuple of a key, not to within
492
the bytestring of each key element. e.g. if you have the keys ('foo',
493
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
494
only the former key is returned.
496
:param keys: An iterable providing the key prefixes to be retrieved.
497
Each key prefix takes the form of a tuple the length of a key, but
498
with the last N elements 'None' rather than a regular bytestring.
499
The first element cannot be 'None'.
500
:return: An iterable as per iter_all_entries, but restricted to the
501
keys with a matching prefix to those supplied. No additional keys
502
will be returned, and every match that is in the index will be
505
# XXX: To much duplication with the GraphIndex class; consider finding
506
# a good place to pull out the actual common logic.
510
for backing in self._backing_indices:
513
for node in backing.iter_entries_prefix(keys):
514
yield (self,) + node[1:]
515
if self._key_length == 1:
519
raise errors.BadIndexKey(key)
520
if len(key) != self._key_length:
521
raise errors.BadIndexKey(key)
523
node = self._nodes[key]
526
if self.reference_lists:
527
yield self, key, node[1], node[0]
529
yield self, key, node[1]
534
raise errors.BadIndexKey(key)
535
if len(key) != self._key_length:
536
raise errors.BadIndexKey(key)
537
# find what it refers to:
538
key_dict = self._get_nodes_by_key()
540
# find the subdict to return
542
while len(elements) and elements[0] is not None:
543
key_dict = key_dict[elements[0]]
546
# a non-existant lookup.
551
key_dict = dicts.pop(-1)
552
# can't be empty or would not exist
553
item, value = key_dict.iteritems().next()
554
if type(value) == dict:
556
dicts.extend(key_dict.itervalues())
559
for value in key_dict.itervalues():
560
yield (self, ) + value
562
yield (self, ) + key_dict
564
def _get_nodes_by_key(self):
565
if self._nodes_by_key is None:
567
if self.reference_lists:
568
for key, (references, value) in self._nodes.iteritems():
569
key_dict = nodes_by_key
570
for subkey in key[:-1]:
571
key_dict = key_dict.setdefault(subkey, {})
572
key_dict[key[-1]] = key, value, references
574
for key, (references, value) in self._nodes.iteritems():
575
key_dict = nodes_by_key
576
for subkey in key[:-1]:
577
key_dict = key_dict.setdefault(subkey, {})
578
key_dict[key[-1]] = key, value
579
self._nodes_by_key = nodes_by_key
580
return self._nodes_by_key
583
"""Return an estimate of the number of keys in this index.
585
For InMemoryGraphIndex the estimate is exact.
587
return len(self._keys) + sum(backing.key_count() for backing in
588
self._backing_indices if backing is not None)
591
"""In memory index's have no known corruption at the moment."""
594
class _LeafNode(object):
595
"""A leaf node for a serialised B+Tree index."""
597
__slots__ = ('keys', 'min_key', 'max_key')
599
def __init__(self, bytes, key_length, ref_list_length):
600
"""Parse bytes to create a leaf node object."""
601
# splitlines mangles the \r delimiters.. don't use it.
602
key_list = _btree_serializer._parse_leaf_lines(bytes,
603
key_length, ref_list_length)
605
self.min_key = key_list[0][0]
606
self.max_key = key_list[-1][0]
608
self.min_key = self.max_key = None
609
self.keys = dict(key_list)
612
class _InternalNode(object):
613
"""An internal node for a serialised B+Tree index."""
615
__slots__ = ('keys', 'offset')
617
def __init__(self, bytes):
618
"""Parse bytes to create an internal node object."""
619
# splitlines mangles the \r delimiters.. don't use it.
620
self.keys = self._parse_lines(bytes.split('\n'))
622
def _parse_lines(self, lines):
624
self.offset = int(lines[1][7:])
625
for line in lines[2:]:
628
nodes.append(tuple(map(intern, line.split('\0'))))
632
class BTreeGraphIndex(object):
633
"""Access to nodes via the standard GraphIndex interface for B+Tree's.
635
Individual nodes are held in a LRU cache. This holds the root node in
636
memory except when very large walks are done.
639
def __init__(self, transport, name, size):
640
"""Create a B+Tree index object on the index name.
642
:param transport: The transport to read data for the index from.
643
:param name: The file name of the index on transport.
644
:param size: Optional size of the index in bytes. This allows
645
compatibility with the GraphIndex API, as well as ensuring that
646
the initial read (to read the root node header) can be done
647
without over-reading even on empty indices, and on small indices
648
allows single-IO to read the entire index.
650
self._transport = transport
654
self._recommended_pages = self._compute_recommended_pages()
655
self._root_node = None
656
# Default max size is 100,000 leave values
657
self._leaf_value_cache = None # lru_cache.LRUCache(100*1000)
658
self._leaf_node_cache = lru_cache.LRUCache(_NODE_CACHE_SIZE)
659
# We could limit this, but even a 300k record btree has only 3k leaf
660
# nodes, and only 20 internal nodes. So the default of 100 nodes in an
661
# LRU would mean we always cache everything anyway, no need to pay the
663
self._internal_node_cache = fifo_cache.FIFOCache(100)
664
self._key_count = None
665
self._row_lengths = None
666
self._row_offsets = None # Start of each row, [-1] is the end
668
def __eq__(self, other):
669
"""Equal when self and other were created with the same parameters."""
671
type(self) == type(other) and
672
self._transport == other._transport and
673
self._name == other._name and
674
self._size == other._size)
676
def __ne__(self, other):
677
return not self.__eq__(other)
679
def _get_and_cache_nodes(self, nodes):
680
"""Read nodes and cache them in the lru.
682
The nodes list supplied is sorted and then read from disk, each node
683
being inserted it into the _node_cache.
685
Note: Asking for more nodes than the _node_cache can contain will
686
result in some of the results being immediately discarded, to prevent
687
this an assertion is raised if more nodes are asked for than are
690
:return: A dict of {node_pos: node}
693
start_of_leaves = None
694
for node_pos, node in self._read_nodes(sorted(nodes)):
695
if node_pos == 0: # Special case
696
self._root_node = node
698
if start_of_leaves is None:
699
start_of_leaves = self._row_offsets[-2]
700
if node_pos < start_of_leaves:
701
self._internal_node_cache.add(node_pos, node)
703
self._leaf_node_cache.add(node_pos, node)
704
found[node_pos] = node
707
def _compute_recommended_pages(self):
708
"""Convert transport's recommended_page_size into btree pages.
710
recommended_page_size is in bytes, we want to know how many _PAGE_SIZE
711
pages fit in that length.
713
recommended_read = self._transport.recommended_page_size()
714
recommended_pages = int(math.ceil(recommended_read /
716
return recommended_pages
718
def _compute_total_pages_in_index(self):
719
"""How many pages are in the index.
721
If we have read the header we will use the value stored there.
722
Otherwise it will be computed based on the length of the index.
724
if self._size is None:
725
raise AssertionError('_compute_total_pages_in_index should not be'
726
' called when self._size is None')
727
if self._root_node is not None:
728
# This is the number of pages as defined by the header
729
return self._row_offsets[-1]
730
# This is the number of pages as defined by the size of the index. They
731
# should be indentical.
732
total_pages = int(math.ceil(self._size / float(_PAGE_SIZE)))
735
def _expand_offsets(self, offsets):
736
"""Find extra pages to download.
738
The idea is that we always want to make big-enough requests (like 64kB
739
for http), so that we don't waste round trips. So given the entries
740
that we already have cached and the new pages being downloaded figure
741
out what other pages we might want to read.
743
See also doc/developers/btree_index_prefetch.txt for more details.
745
:param offsets: The offsets to be read
746
:return: A list of offsets to download
748
if 'index' in debug.debug_flags:
749
trace.mutter('expanding: %s\toffsets: %s', self._name, offsets)
751
if len(offsets) >= self._recommended_pages:
752
# Don't add more, we are already requesting more than enough
753
if 'index' in debug.debug_flags:
754
trace.mutter(' not expanding large request (%s >= %s)',
755
len(offsets), self._recommended_pages)
757
if self._size is None:
758
# Don't try anything, because we don't know where the file ends
759
if 'index' in debug.debug_flags:
760
trace.mutter(' not expanding without knowing index size')
762
total_pages = self._compute_total_pages_in_index()
763
cached_offsets = self._get_offsets_to_cached_pages()
764
# If reading recommended_pages would read the rest of the index, just
766
if total_pages - len(cached_offsets) <= self._recommended_pages:
767
# Read whatever is left
769
expanded = [x for x in xrange(total_pages)
770
if x not in cached_offsets]
772
expanded = range(total_pages)
773
if 'index' in debug.debug_flags:
774
trace.mutter(' reading all unread pages: %s', expanded)
777
if self._root_node is None:
778
# ATM on the first read of the root node of a large index, we don't
779
# bother pre-reading any other pages. This is because the
780
# likelyhood of actually reading interesting pages is very low.
781
# See doc/developers/btree_index_prefetch.txt for a discussion, and
782
# a possible implementation when we are guessing that the second
783
# layer index is small
784
final_offsets = offsets
786
tree_depth = len(self._row_lengths)
787
if len(cached_offsets) < tree_depth and len(offsets) == 1:
788
# We haven't read enough to justify expansion
789
# If we are only going to read the root node, and 1 leaf node,
790
# then it isn't worth expanding our request. Once we've read at
791
# least 2 nodes, then we are probably doing a search, and we
792
# start expanding our requests.
793
if 'index' in debug.debug_flags:
794
trace.mutter(' not expanding on first reads')
796
final_offsets = self._expand_to_neighbors(offsets, cached_offsets,
799
final_offsets = sorted(final_offsets)
800
if 'index' in debug.debug_flags:
801
trace.mutter('expanded: %s', final_offsets)
804
def _expand_to_neighbors(self, offsets, cached_offsets, total_pages):
805
"""Expand requests to neighbors until we have enough pages.
807
This is called from _expand_offsets after policy has determined that we
809
We only want to expand requests within a given layer. We cheat a little
810
bit and assume all requests will be in the same layer. This is true
811
given the current design, but if it changes this algorithm may perform
814
:param offsets: requested offsets
815
:param cached_offsets: offsets for pages we currently have cached
816
:return: A set() of offsets after expansion
818
final_offsets = set(offsets)
820
new_tips = set(final_offsets)
821
while len(final_offsets) < self._recommended_pages and new_tips:
825
first, end = self._find_layer_first_and_end(pos)
828
and previous not in cached_offsets
829
and previous not in final_offsets
830
and previous >= first):
831
next_tips.add(previous)
833
if (after < total_pages
834
and after not in cached_offsets
835
and after not in final_offsets
838
# This would keep us from going bigger than
839
# recommended_pages by only expanding the first offsets.
840
# However, if we are making a 'wide' request, it is
841
# reasonable to expand all points equally.
842
# if len(final_offsets) > recommended_pages:
844
final_offsets.update(next_tips)
848
def external_references(self, ref_list_num):
849
if self._root_node is None:
850
self._get_root_node()
851
if ref_list_num + 1 > self.node_ref_lists:
852
raise ValueError('No ref list %d, index has %d ref lists'
853
% (ref_list_num, self.node_ref_lists))
856
for node in self.iter_all_entries():
858
refs.update(node[3][ref_list_num])
861
def _find_layer_first_and_end(self, offset):
862
"""Find the start/stop nodes for the layer corresponding to offset.
864
:return: (first, end)
865
first is the first node in this layer
866
end is the first node of the next layer
869
for roffset in self._row_offsets:
876
def _get_offsets_to_cached_pages(self):
877
"""Determine what nodes we already have cached."""
878
cached_offsets = set(self._internal_node_cache.keys())
879
cached_offsets.update(self._leaf_node_cache.keys())
880
if self._root_node is not None:
881
cached_offsets.add(0)
882
return cached_offsets
884
def _get_root_node(self):
885
if self._root_node is None:
886
# We may not have a root node yet
887
self._get_internal_nodes([0])
888
return self._root_node
890
def _get_nodes(self, cache, node_indexes):
893
for idx in node_indexes:
894
if idx == 0 and self._root_node is not None:
895
found[0] = self._root_node
898
found[idx] = cache[idx]
903
needed = self._expand_offsets(needed)
904
found.update(self._get_and_cache_nodes(needed))
907
def _get_internal_nodes(self, node_indexes):
908
"""Get a node, from cache or disk.
910
After getting it, the node will be cached.
912
return self._get_nodes(self._internal_node_cache, node_indexes)
914
def _cache_leaf_values(self, nodes):
915
"""Cache directly from key => value, skipping the btree."""
916
if self._leaf_value_cache is not None:
917
for node in nodes.itervalues():
918
for key, value in node.keys.iteritems():
919
if key in self._leaf_value_cache:
920
# Don't add the rest of the keys, we've seen this node
923
self._leaf_value_cache[key] = value
925
def _get_leaf_nodes(self, node_indexes):
926
"""Get a bunch of nodes, from cache or disk."""
927
found = self._get_nodes(self._leaf_node_cache, node_indexes)
928
self._cache_leaf_values(found)
931
def iter_all_entries(self):
932
"""Iterate over all keys within the index.
934
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
935
The former tuple is used when there are no reference lists in the
936
index, making the API compatible with simple key:value index types.
937
There is no defined order for the result iteration - it will be in
938
the most efficient order for the index.
940
if 'evil' in debug.debug_flags:
941
trace.mutter_callsite(3,
942
"iter_all_entries scales with size of history.")
943
if not self.key_count():
945
if self._row_offsets[-1] == 1:
946
# There is only the root node, and we read that via key_count()
947
if self.node_ref_lists:
948
for key, (value, refs) in sorted(self._root_node.keys.items()):
949
yield (self, key, value, refs)
951
for key, (value, refs) in sorted(self._root_node.keys.items()):
952
yield (self, key, value)
954
start_of_leaves = self._row_offsets[-2]
955
end_of_leaves = self._row_offsets[-1]
956
needed_offsets = range(start_of_leaves, end_of_leaves)
957
if needed_offsets == [0]:
958
# Special case when we only have a root node, as we have already
960
nodes = [(0, self._root_node)]
962
nodes = self._read_nodes(needed_offsets)
963
# We iterate strictly in-order so that we can use this function
964
# for spilling index builds to disk.
965
if self.node_ref_lists:
966
for _, node in nodes:
967
for key, (value, refs) in sorted(node.keys.items()):
968
yield (self, key, value, refs)
970
for _, node in nodes:
971
for key, (value, refs) in sorted(node.keys.items()):
972
yield (self, key, value)
975
def _multi_bisect_right(in_keys, fixed_keys):
976
"""Find the positions where each 'in_key' would fit in fixed_keys.
978
This is equivalent to doing "bisect_right" on each in_key into
981
:param in_keys: A sorted list of keys to match with fixed_keys
982
:param fixed_keys: A sorted list of keys to match against
983
:return: A list of (integer position, [key list]) tuples.
988
# no pointers in the fixed_keys list, which means everything must
990
return [(0, in_keys)]
992
# TODO: Iterating both lists will generally take M + N steps
993
# Bisecting each key will generally take M * log2 N steps.
994
# If we had an efficient way to compare, we could pick the method
995
# based on which has the fewer number of steps.
996
# There is also the argument that bisect_right is a compiled
997
# function, so there is even more to be gained.
998
# iter_steps = len(in_keys) + len(fixed_keys)
999
# bisect_steps = len(in_keys) * math.log(len(fixed_keys), 2)
1000
if len(in_keys) == 1: # Bisect will always be faster for M = 1
1001
return [(bisect_right(fixed_keys, in_keys[0]), in_keys)]
1002
# elif bisect_steps < iter_steps:
1004
# for key in in_keys:
1005
# offsets.setdefault(bisect_right(fixed_keys, key),
1007
# return [(o, offsets[o]) for o in sorted(offsets)]
1008
in_keys_iter = iter(in_keys)
1009
fixed_keys_iter = enumerate(fixed_keys)
1010
cur_in_key = in_keys_iter.next()
1011
cur_fixed_offset, cur_fixed_key = fixed_keys_iter.next()
1013
class InputDone(Exception): pass
1014
class FixedDone(Exception): pass
1019
# TODO: Another possibility is that rather than iterating on each side,
1020
# we could use a combination of bisecting and iterating. For
1021
# example, while cur_in_key < fixed_key, bisect to find its
1022
# point, then iterate all matching keys, then bisect (restricted
1023
# to only the remainder) for the next one, etc.
1026
if cur_in_key < cur_fixed_key:
1028
cur_out = (cur_fixed_offset, cur_keys)
1029
output.append(cur_out)
1030
while cur_in_key < cur_fixed_key:
1031
cur_keys.append(cur_in_key)
1033
cur_in_key = in_keys_iter.next()
1034
except StopIteration:
1036
# At this point cur_in_key must be >= cur_fixed_key
1037
# step the cur_fixed_key until we pass the cur key, or walk off
1039
while cur_in_key >= cur_fixed_key:
1041
cur_fixed_offset, cur_fixed_key = fixed_keys_iter.next()
1042
except StopIteration:
1045
# We consumed all of the input, nothing more to do
1048
# There was some input left, but we consumed all of fixed, so we
1049
# have to add one more for the tail
1050
cur_keys = [cur_in_key]
1051
cur_keys.extend(in_keys_iter)
1052
cur_out = (len(fixed_keys), cur_keys)
1053
output.append(cur_out)
1056
def _walk_through_internal_nodes(self, keys):
1057
"""Take the given set of keys, and find the corresponding LeafNodes.
1059
:param keys: An unsorted iterable of keys to search for
1060
:return: (nodes, index_and_keys)
1061
nodes is a dict mapping {index: LeafNode}
1062
keys_at_index is a list of tuples of [(index, [keys for Leaf])]
1064
# 6 seconds spent in miss_torture using the sorted() line.
1065
# Even with out of order disk IO it seems faster not to sort it when
1066
# large queries are being made.
1067
keys_at_index = [(0, sorted(keys))]
1069
for row_pos, next_row_start in enumerate(self._row_offsets[1:-1]):
1070
node_indexes = [idx for idx, s_keys in keys_at_index]
1071
nodes = self._get_internal_nodes(node_indexes)
1073
next_nodes_and_keys = []
1074
for node_index, sub_keys in keys_at_index:
1075
node = nodes[node_index]
1076
positions = self._multi_bisect_right(sub_keys, node.keys)
1077
node_offset = next_row_start + node.offset
1078
next_nodes_and_keys.extend([(node_offset + pos, s_keys)
1079
for pos, s_keys in positions])
1080
keys_at_index = next_nodes_and_keys
1081
# We should now be at the _LeafNodes
1082
node_indexes = [idx for idx, s_keys in keys_at_index]
1084
# TODO: We may *not* want to always read all the nodes in one
1085
# big go. Consider setting a max size on this.
1086
nodes = self._get_leaf_nodes(node_indexes)
1087
return nodes, keys_at_index
1089
def iter_entries(self, keys):
1090
"""Iterate over keys within the index.
1092
:param keys: An iterable providing the keys to be retrieved.
1093
:return: An iterable as per iter_all_entries, but restricted to the
1094
keys supplied. No additional keys will be returned, and every
1095
key supplied that is in the index will be returned.
1097
# 6 seconds spent in miss_torture using the sorted() line.
1098
# Even with out of order disk IO it seems faster not to sort it when
1099
# large queries are being made.
1100
# However, now that we are doing multi-way bisecting, we need the keys
1101
# in sorted order anyway. We could change the multi-way code to not
1102
# require sorted order. (For example, it bisects for the first node,
1103
# does an in-order search until a key comes before the current point,
1104
# which it then bisects for, etc.)
1105
keys = frozenset(keys)
1109
if not self.key_count():
1113
if self._leaf_value_cache is None:
1117
value = self._leaf_value_cache.get(key, None)
1118
if value is not None:
1119
# This key is known not to be here, skip it
1121
if self.node_ref_lists:
1122
yield (self, key, value, refs)
1124
yield (self, key, value)
1126
needed_keys.append(key)
1132
nodes, nodes_and_keys = self._walk_through_internal_nodes(needed_keys)
1133
for node_index, sub_keys in nodes_and_keys:
1136
node = nodes[node_index]
1137
for next_sub_key in sub_keys:
1138
if next_sub_key in node.keys:
1139
value, refs = node.keys[next_sub_key]
1140
if self.node_ref_lists:
1141
yield (self, next_sub_key, value, refs)
1143
yield (self, next_sub_key, value)
1145
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
1146
"""Find the parent_map information for the set of keys.
1148
This populates the parent_map dict and missing_keys set based on the
1149
queried keys. It also can fill out an arbitrary number of parents that
1150
it finds while searching for the supplied keys.
1152
It is unlikely that you want to call this directly. See
1153
"CombinedGraphIndex.find_ancestry()" for a more appropriate API.
1155
:param keys: A keys whose ancestry we want to return
1156
Every key will either end up in 'parent_map' or 'missing_keys'.
1157
:param ref_list_num: This index in the ref_lists is the parents we
1159
:param parent_map: {key: parent_keys} for keys that are present in this
1160
index. This may contain more entries than were in 'keys', that are
1161
reachable ancestors of the keys requested.
1162
:param missing_keys: keys which are known to be missing in this index.
1163
This may include parents that were not directly requested, but we
1164
were able to determine that they are not present in this index.
1165
:return: search_keys parents that were found but not queried to know
1166
if they are missing or present. Callers can re-query this index for
1167
those keys, and they will be placed into parent_map or missing_keys
1169
if not self.key_count():
1170
# We use key_count() to trigger reading the root node and
1171
# determining info about this BTreeGraphIndex
1172
# If we don't have any keys, then everything is missing
1173
missing_keys.update(keys)
1175
if ref_list_num >= self.node_ref_lists:
1176
raise ValueError('No ref list %d, index has %d ref lists'
1177
% (ref_list_num, self.node_ref_lists))
1179
# The main trick we are trying to accomplish is that when we find a
1180
# key listing its parents, we expect that the parent key is also likely
1181
# to sit on the same page. Allowing us to expand parents quickly
1182
# without suffering the full stack of bisecting, etc.
1183
nodes, nodes_and_keys = self._walk_through_internal_nodes(keys)
1185
# These are parent keys which could not be immediately resolved on the
1186
# page where the child was present. Note that we may already be
1187
# searching for that key, and it may actually be present [or known
1188
# missing] on one of the other pages we are reading.
1190
# We could try searching for them in the immediate previous or next
1191
# page. If they occur "later" we could put them in a pending lookup
1192
# set, and then for each node we read thereafter we could check to
1193
# see if they are present.
1194
# However, we don't know the impact of keeping this list of things
1195
# that I'm going to search for every node I come across from here on
1197
# It doesn't handle the case when the parent key is missing on a
1198
# page that we *don't* read. So we already have to handle being
1199
# re-entrant for that.
1200
# Since most keys contain a date string, they are more likely to be
1201
# found earlier in the file than later, but we would know that right
1202
# away (key < min_key), and wouldn't keep searching it on every other
1203
# page that we read.
1204
# Mostly, it is an idea, one which should be benchmarked.
1205
parents_not_on_page = set()
1207
for node_index, sub_keys in nodes_and_keys:
1210
# sub_keys is all of the keys we are looking for that should exist
1211
# on this page, if they aren't here, then they won't be found
1212
node = nodes[node_index]
1213
node_keys = node.keys
1214
parents_to_check = set()
1215
for next_sub_key in sub_keys:
1216
if next_sub_key not in node_keys:
1217
# This one is just not present in the index at all
1218
missing_keys.add(next_sub_key)
1220
value, refs = node_keys[next_sub_key]
1221
parent_keys = refs[ref_list_num]
1222
parent_map[next_sub_key] = parent_keys
1223
parents_to_check.update(parent_keys)
1224
# Don't look for things we've already found
1225
parents_to_check = parents_to_check.difference(parent_map)
1226
# this can be used to test the benefit of having the check loop
1228
# parents_not_on_page.update(parents_to_check)
1230
while parents_to_check:
1231
next_parents_to_check = set()
1232
for key in parents_to_check:
1233
if key in node_keys:
1234
value, refs = node_keys[key]
1235
parent_keys = refs[ref_list_num]
1236
parent_map[key] = parent_keys
1237
next_parents_to_check.update(parent_keys)
1239
# This parent either is genuinely missing, or should be
1240
# found on another page. Perf test whether it is better
1241
# to check if this node should fit on this page or not.
1242
# in the 'everything-in-one-pack' scenario, this *not*
1243
# doing the check is 237ms vs 243ms.
1244
# So slightly better, but I assume the standard 'lots
1245
# of packs' is going to show a reasonable improvement
1246
# from the check, because it avoids 'going around
1247
# again' for everything that is in another index
1248
# parents_not_on_page.add(key)
1249
# Missing for some reason
1250
if key < node.min_key:
1251
# in the case of bzr.dev, 3.4k/5.3k misses are
1252
# 'earlier' misses (65%)
1253
parents_not_on_page.add(key)
1254
elif key > node.max_key:
1255
# This parent key would be present on a different
1257
parents_not_on_page.add(key)
1259
# assert key != node.min_key and key != node.max_key
1260
# If it was going to be present, it would be on
1261
# *this* page, so mark it missing.
1262
missing_keys.add(key)
1263
parents_to_check = next_parents_to_check.difference(parent_map)
1264
# Might want to do another .difference() from missing_keys
1265
# parents_not_on_page could have been found on a different page, or be
1266
# known to be missing. So cull out everything that has already been
1268
search_keys = parents_not_on_page.difference(
1269
parent_map).difference(missing_keys)
1272
def iter_entries_prefix(self, keys):
1273
"""Iterate over keys within the index using prefix matching.
1275
Prefix matching is applied within the tuple of a key, not to within
1276
the bytestring of each key element. e.g. if you have the keys ('foo',
1277
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1278
only the former key is returned.
1280
WARNING: Note that this method currently causes a full index parse
1281
unconditionally (which is reasonably appropriate as it is a means for
1282
thunking many small indices into one larger one and still supplies
1283
iter_all_entries at the thunk layer).
1285
:param keys: An iterable providing the key prefixes to be retrieved.
1286
Each key prefix takes the form of a tuple the length of a key, but
1287
with the last N elements 'None' rather than a regular bytestring.
1288
The first element cannot be 'None'.
1289
:return: An iterable as per iter_all_entries, but restricted to the
1290
keys with a matching prefix to those supplied. No additional keys
1291
will be returned, and every match that is in the index will be
1294
keys = sorted(set(keys))
1297
# Load if needed to check key lengths
1298
if self._key_count is None:
1299
self._get_root_node()
1300
# TODO: only access nodes that can satisfy the prefixes we are looking
1301
# for. For now, to meet API usage (as this function is not used by
1302
# current bzrlib) just suck the entire index and iterate in memory.
1304
if self.node_ref_lists:
1305
if self._key_length == 1:
1306
for _1, key, value, refs in self.iter_all_entries():
1307
nodes[key] = value, refs
1310
for _1, key, value, refs in self.iter_all_entries():
1311
key_value = key, value, refs
1312
# For a key of (foo, bar, baz) create
1313
# _nodes_by_key[foo][bar][baz] = key_value
1314
key_dict = nodes_by_key
1315
for subkey in key[:-1]:
1316
key_dict = key_dict.setdefault(subkey, {})
1317
key_dict[key[-1]] = key_value
1319
if self._key_length == 1:
1320
for _1, key, value in self.iter_all_entries():
1324
for _1, key, value in self.iter_all_entries():
1325
key_value = key, value
1326
# For a key of (foo, bar, baz) create
1327
# _nodes_by_key[foo][bar][baz] = key_value
1328
key_dict = nodes_by_key
1329
for subkey in key[:-1]:
1330
key_dict = key_dict.setdefault(subkey, {})
1331
key_dict[key[-1]] = key_value
1332
if self._key_length == 1:
1336
raise errors.BadIndexKey(key)
1337
if len(key) != self._key_length:
1338
raise errors.BadIndexKey(key)
1340
if self.node_ref_lists:
1341
value, node_refs = nodes[key]
1342
yield self, key, value, node_refs
1344
yield self, key, nodes[key]
1351
raise errors.BadIndexKey(key)
1352
if len(key) != self._key_length:
1353
raise errors.BadIndexKey(key)
1354
# find what it refers to:
1355
key_dict = nodes_by_key
1356
elements = list(key)
1357
# find the subdict whose contents should be returned.
1359
while len(elements) and elements[0] is not None:
1360
key_dict = key_dict[elements[0]]
1363
# a non-existant lookup.
1368
key_dict = dicts.pop(-1)
1369
# can't be empty or would not exist
1370
item, value = key_dict.iteritems().next()
1371
if type(value) == dict:
1373
dicts.extend(key_dict.itervalues())
1376
for value in key_dict.itervalues():
1377
# each value is the key:value:node refs tuple
1379
yield (self, ) + value
1381
# the last thing looked up was a terminal element
1382
yield (self, ) + key_dict
1384
def key_count(self):
1385
"""Return an estimate of the number of keys in this index.
1387
For BTreeGraphIndex the estimate is exact as it is contained in the
1390
if self._key_count is None:
1391
self._get_root_node()
1392
return self._key_count
1394
def _compute_row_offsets(self):
1395
"""Fill out the _row_offsets attribute based on _row_lengths."""
1398
for row in self._row_lengths:
1399
offsets.append(row_offset)
1401
offsets.append(row_offset)
1402
self._row_offsets = offsets
1404
def _parse_header_from_bytes(self, bytes):
1405
"""Parse the header from a region of bytes.
1407
:param bytes: The data to parse.
1408
:return: An offset, data tuple such as readv yields, for the unparsed
1409
data. (which may be of length 0).
1411
signature = bytes[0:len(self._signature())]
1412
if not signature == self._signature():
1413
raise errors.BadIndexFormatSignature(self._name, BTreeGraphIndex)
1414
lines = bytes[len(self._signature()):].splitlines()
1415
options_line = lines[0]
1416
if not options_line.startswith(_OPTION_NODE_REFS):
1417
raise errors.BadIndexOptions(self)
1419
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
1421
raise errors.BadIndexOptions(self)
1422
options_line = lines[1]
1423
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
1424
raise errors.BadIndexOptions(self)
1426
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
1428
raise errors.BadIndexOptions(self)
1429
options_line = lines[2]
1430
if not options_line.startswith(_OPTION_LEN):
1431
raise errors.BadIndexOptions(self)
1433
self._key_count = int(options_line[len(_OPTION_LEN):])
1435
raise errors.BadIndexOptions(self)
1436
options_line = lines[3]
1437
if not options_line.startswith(_OPTION_ROW_LENGTHS):
1438
raise errors.BadIndexOptions(self)
1440
self._row_lengths = map(int, [length for length in
1441
options_line[len(_OPTION_ROW_LENGTHS):].split(',')
1444
raise errors.BadIndexOptions(self)
1445
self._compute_row_offsets()
1447
# calculate the bytes we have processed
1448
header_end = (len(signature) + sum(map(len, lines[0:4])) + 4)
1449
return header_end, bytes[header_end:]
1451
def _read_nodes(self, nodes):
1452
"""Read some nodes from disk into the LRU cache.
1454
This performs a readv to get the node data into memory, and parses each
1455
node, then yields it to the caller. The nodes are requested in the
1456
supplied order. If possible doing sort() on the list before requesting
1457
a read may improve performance.
1459
:param nodes: The nodes to read. 0 - first node, 1 - second node etc.
1462
# may be the byte string of the whole file
1464
# list of (offset, length) regions of the file that should, evenually
1465
# be read in to data_ranges, either from 'bytes' or from the transport
1468
offset = index * _PAGE_SIZE
1471
# Root node - special case
1473
size = min(_PAGE_SIZE, self._size)
1475
# The only case where we don't know the size, is for very
1476
# small indexes. So we read the whole thing
1477
bytes = self._transport.get_bytes(self._name)
1478
self._size = len(bytes)
1479
# the whole thing should be parsed out of 'bytes'
1480
ranges.append((0, len(bytes)))
1483
if offset > self._size:
1484
raise AssertionError('tried to read past the end'
1485
' of the file %s > %s'
1486
% (offset, self._size))
1487
size = min(size, self._size - offset)
1488
ranges.append((offset, size))
1491
elif bytes is not None:
1492
# already have the whole file
1493
data_ranges = [(start, bytes[start:start+_PAGE_SIZE])
1494
for start in xrange(0, len(bytes), _PAGE_SIZE)]
1495
elif self._file is None:
1496
data_ranges = self._transport.readv(self._name, ranges)
1499
for offset, size in ranges:
1500
self._file.seek(offset)
1501
data_ranges.append((offset, self._file.read(size)))
1502
for offset, data in data_ranges:
1504
# extract the header
1505
offset, data = self._parse_header_from_bytes(data)
1508
bytes = zlib.decompress(data)
1509
if bytes.startswith(_LEAF_FLAG):
1510
node = _LeafNode(bytes, self._key_length, self.node_ref_lists)
1511
elif bytes.startswith(_INTERNAL_FLAG):
1512
node = _InternalNode(bytes)
1514
raise AssertionError("Unknown node type for %r" % bytes)
1515
yield offset / _PAGE_SIZE, node
1517
def _signature(self):
1518
"""The file signature for this index type."""
1522
"""Validate that everything in the index can be accessed."""
1523
# just read and parse every node.
1524
self._get_root_node()
1525
if len(self._row_lengths) > 1:
1526
start_node = self._row_offsets[1]
1528
# We shouldn't be reading anything anyway
1530
node_end = self._row_offsets[-1]
1531
for node in self._read_nodes(range(start_node, node_end)):
1536
from bzrlib import _btree_serializer_pyx as _btree_serializer
1537
except ImportError, e:
1538
osutils.failed_to_load_extension(e)
1539
from bzrlib import _btree_serializer_py as _btree_serializer
added
removed
bzrlib/btree_index.py
