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)
167
self._nodes[key] = (node_refs, value)
169
if self._nodes_by_key is not None and self._key_length > 1:
170
self._update_nodes_by_key(key, value, node_refs)
171
if len(self._keys) < self._spill_at:
173
self._spill_mem_keys_to_disk()
175
def _spill_mem_keys_to_disk(self):
176
"""Write the in memory keys down to disk to cap memory consumption.
178
If we already have some keys written to disk, we will combine them so
179
as to preserve the sorted order. The algorithm for combining uses
180
powers of two. So on the first spill, write all mem nodes into a
181
single index. On the second spill, combine the mem nodes with the nodes
182
on disk to create a 2x sized disk index and get rid of the first index.
183
On the third spill, create a single new disk index, which will contain
184
the mem nodes, and preserve the existing 2x sized index. On the fourth,
185
combine mem with the first and second indexes, creating a new one of
186
size 4x. On the fifth create a single new one, etc.
188
if self._combine_backing_indices:
189
(new_backing_file, size,
190
backing_pos) = self._spill_mem_keys_and_combine()
192
new_backing_file, size = self._spill_mem_keys_without_combining()
193
# Note: The transport here isn't strictly needed, because we will use
194
# direct access to the new_backing._file object
195
new_backing = BTreeGraphIndex(get_transport('.'), '<temp>', size)
196
# GC will clean up the file
197
new_backing._file = new_backing_file
198
if self._combine_backing_indices:
199
if len(self._backing_indices) == backing_pos:
200
self._backing_indices.append(None)
201
self._backing_indices[backing_pos] = new_backing
202
for backing_pos in range(backing_pos):
203
self._backing_indices[backing_pos] = None
205
self._backing_indices.append(new_backing)
208
self._nodes_by_key = None
210
def _spill_mem_keys_without_combining(self):
211
return self._write_nodes(self._iter_mem_nodes(), allow_optimize=False)
213
def _spill_mem_keys_and_combine(self):
214
iterators_to_combine = [self._iter_mem_nodes()]
216
for pos, backing in enumerate(self._backing_indices):
220
iterators_to_combine.append(backing.iter_all_entries())
221
backing_pos = pos + 1
222
new_backing_file, size = \
223
self._write_nodes(self._iter_smallest(iterators_to_combine),
224
allow_optimize=False)
225
return new_backing_file, size, backing_pos
227
def add_nodes(self, nodes):
228
"""Add nodes to the index.
230
:param nodes: An iterable of (key, node_refs, value) entries to add.
232
if self.reference_lists:
233
for (key, value, node_refs) in nodes:
234
self.add_node(key, value, node_refs)
236
for (key, value) in nodes:
237
self.add_node(key, value)
239
def _iter_mem_nodes(self):
240
"""Iterate over the nodes held in memory."""
242
if self.reference_lists:
243
for key in sorted(nodes):
244
references, value = nodes[key]
245
yield self, key, value, references
247
for key in sorted(nodes):
248
references, value = nodes[key]
249
yield self, key, value
251
def _iter_smallest(self, iterators_to_combine):
252
if len(iterators_to_combine) == 1:
253
for value in iterators_to_combine[0]:
257
for iterator in iterators_to_combine:
259
current_values.append(iterator.next())
260
except StopIteration:
261
current_values.append(None)
264
# Decorate candidates with the value to allow 2.4's min to be used.
265
candidates = [(item[1][1], item) for item
266
in enumerate(current_values) if item[1] is not None]
267
if not len(candidates):
269
selected = min(candidates)
270
# undecorate back to (pos, node)
271
selected = selected[1]
272
if last == selected[1][1]:
273
raise errors.BadIndexDuplicateKey(last, self)
274
last = selected[1][1]
275
# Yield, with self as the index
276
yield (self,) + selected[1][1:]
279
current_values[pos] = iterators_to_combine[pos].next()
280
except StopIteration:
281
current_values[pos] = None
283
def _add_key(self, string_key, line, rows, allow_optimize=True):
284
"""Add a key to the current chunk.
286
:param string_key: The key to add.
287
:param line: The fully serialised key and value.
288
:param allow_optimize: If set to False, prevent setting the optimize
289
flag when writing out. This is used by the _spill_mem_keys_to_disk
292
if rows[-1].writer is None:
293
# opening a new leaf chunk;
294
for pos, internal_row in enumerate(rows[:-1]):
295
# flesh out any internal nodes that are needed to
296
# preserve the height of the tree
297
if internal_row.writer is None:
299
if internal_row.nodes == 0:
300
length -= _RESERVED_HEADER_BYTES # padded
302
optimize_for_size = self._optimize_for_size
304
optimize_for_size = False
305
internal_row.writer = chunk_writer.ChunkWriter(length, 0,
306
optimize_for_size=optimize_for_size)
307
internal_row.writer.write(_INTERNAL_FLAG)
308
internal_row.writer.write(_INTERNAL_OFFSET +
309
str(rows[pos + 1].nodes) + "\n")
312
if rows[-1].nodes == 0:
313
length -= _RESERVED_HEADER_BYTES # padded
314
rows[-1].writer = chunk_writer.ChunkWriter(length,
315
optimize_for_size=self._optimize_for_size)
316
rows[-1].writer.write(_LEAF_FLAG)
317
if rows[-1].writer.write(line):
318
# this key did not fit in the node:
319
rows[-1].finish_node()
320
key_line = string_key + "\n"
322
for row in reversed(rows[:-1]):
323
# Mark the start of the next node in the node above. If it
324
# doesn't fit then propagate upwards until we find one that
326
if row.writer.write(key_line):
329
# We've found a node that can handle the pointer.
332
# If we reached the current root without being able to mark the
333
# division point, then we need a new root:
336
if 'index' in debug.debug_flags:
337
trace.mutter('Inserting new global row.')
338
new_row = _InternalBuilderRow()
340
rows.insert(0, new_row)
341
# This will be padded, hence the -100
342
new_row.writer = chunk_writer.ChunkWriter(
343
_PAGE_SIZE - _RESERVED_HEADER_BYTES,
345
optimize_for_size=self._optimize_for_size)
346
new_row.writer.write(_INTERNAL_FLAG)
347
new_row.writer.write(_INTERNAL_OFFSET +
348
str(rows[1].nodes - 1) + "\n")
349
new_row.writer.write(key_line)
350
self._add_key(string_key, line, rows, allow_optimize=allow_optimize)
352
def _write_nodes(self, node_iterator, allow_optimize=True):
353
"""Write node_iterator out as a B+Tree.
355
:param node_iterator: An iterator of sorted nodes. Each node should
356
match the output given by iter_all_entries.
357
:param allow_optimize: If set to False, prevent setting the optimize
358
flag when writing out. This is used by the _spill_mem_keys_to_disk
360
:return: A file handle for a temporary file containing a B+Tree for
363
# The index rows - rows[0] is the root, rows[1] is the layer under it
366
# forward sorted by key. In future we may consider topological sorting,
367
# at the cost of table scans for direct lookup, or a second index for
370
# A stack with the number of nodes of each size. 0 is the root node
371
# and must always be 1 (if there are any nodes in the tree).
372
self.row_lengths = []
373
# Loop over all nodes adding them to the bottom row
374
# (rows[-1]). When we finish a chunk in a row,
375
# propagate the key that didn't fit (comes after the chunk) to the
376
# row above, transitively.
377
for node in node_iterator:
379
# First key triggers the first row
380
rows.append(_LeafBuilderRow())
382
string_key, line = _btree_serializer._flatten_node(node,
383
self.reference_lists)
384
self._add_key(string_key, line, rows, allow_optimize=allow_optimize)
385
for row in reversed(rows):
386
pad = (type(row) != _LeafBuilderRow)
387
row.finish_node(pad=pad)
388
lines = [_BTSIGNATURE]
389
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
390
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
391
lines.append(_OPTION_LEN + str(key_count) + '\n')
392
row_lengths = [row.nodes for row in rows]
393
lines.append(_OPTION_ROW_LENGTHS + ','.join(map(str, row_lengths)) + '\n')
394
if row_lengths and row_lengths[-1] > 1:
395
result = tempfile.NamedTemporaryFile(prefix='bzr-index-')
397
result = cStringIO.StringIO()
398
result.writelines(lines)
399
position = sum(map(len, lines))
401
if position > _RESERVED_HEADER_BYTES:
402
raise AssertionError("Could not fit the header in the"
403
" reserved space: %d > %d"
404
% (position, _RESERVED_HEADER_BYTES))
405
# write the rows out:
407
reserved = _RESERVED_HEADER_BYTES # reserved space for first node
410
# copy nodes to the finalised file.
411
# Special case the first node as it may be prefixed
412
node = row.spool.read(_PAGE_SIZE)
413
result.write(node[reserved:])
414
if len(node) == _PAGE_SIZE:
415
result.write("\x00" * (reserved - position))
416
position = 0 # Only the root row actually has an offset
417
copied_len = osutils.pumpfile(row.spool, result)
418
if copied_len != (row.nodes - 1) * _PAGE_SIZE:
419
if type(row) != _LeafBuilderRow:
420
raise AssertionError("Incorrect amount of data copied"
421
" expected: %d, got: %d"
422
% ((row.nodes - 1) * _PAGE_SIZE,
430
"""Finalise the index.
432
:return: A file handle for a temporary file containing the nodes added
435
return self._write_nodes(self.iter_all_entries())[0]
437
def iter_all_entries(self):
438
"""Iterate over all keys within the index
440
:return: An iterable of (index, key, value, reference_lists). There is
441
no defined order for the result iteration - it will be in the most
442
efficient order for the index (in this case dictionary hash order).
444
if 'evil' in debug.debug_flags:
445
trace.mutter_callsite(3,
446
"iter_all_entries scales with size of history.")
447
# Doing serial rather than ordered would be faster; but this shouldn't
448
# be getting called routinely anyway.
449
iterators = [self._iter_mem_nodes()]
450
for backing in self._backing_indices:
451
if backing is not None:
452
iterators.append(backing.iter_all_entries())
453
if len(iterators) == 1:
455
return self._iter_smallest(iterators)
457
def iter_entries(self, keys):
458
"""Iterate over keys within the index.
460
:param keys: An iterable providing the keys to be retrieved.
461
:return: An iterable of (index, key, value, reference_lists). There is no
462
defined order for the result iteration - it will be in the most
463
efficient order for the index (keys iteration order in this case).
466
local_keys = keys.intersection(self._keys)
467
if self.reference_lists:
468
for key in local_keys:
469
node = self._nodes[key]
470
yield self, key, node[1], node[0]
472
for key in local_keys:
473
node = self._nodes[key]
474
yield self, key, node[1]
475
# Find things that are in backing indices that have not been handled
477
if not self._backing_indices:
478
return # We won't find anything there either
479
# Remove all of the keys that we found locally
480
keys.difference_update(local_keys)
481
for backing in self._backing_indices:
486
for node in backing.iter_entries(keys):
488
yield (self,) + node[1:]
490
def iter_entries_prefix(self, keys):
491
"""Iterate over keys within the index using prefix matching.
493
Prefix matching is applied within the tuple of a key, not to within
494
the bytestring of each key element. e.g. if you have the keys ('foo',
495
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
496
only the former key is returned.
498
:param keys: An iterable providing the key prefixes to be retrieved.
499
Each key prefix takes the form of a tuple the length of a key, but
500
with the last N elements 'None' rather than a regular bytestring.
501
The first element cannot be 'None'.
502
:return: An iterable as per iter_all_entries, but restricted to the
503
keys with a matching prefix to those supplied. No additional keys
504
will be returned, and every match that is in the index will be
507
# XXX: To much duplication with the GraphIndex class; consider finding
508
# a good place to pull out the actual common logic.
512
for backing in self._backing_indices:
515
for node in backing.iter_entries_prefix(keys):
516
yield (self,) + node[1:]
517
if self._key_length == 1:
521
raise errors.BadIndexKey(key)
522
if len(key) != self._key_length:
523
raise errors.BadIndexKey(key)
525
node = self._nodes[key]
528
if self.reference_lists:
529
yield self, key, node[1], node[0]
531
yield self, key, node[1]
536
raise errors.BadIndexKey(key)
537
if len(key) != self._key_length:
538
raise errors.BadIndexKey(key)
539
# find what it refers to:
540
key_dict = self._get_nodes_by_key()
542
# find the subdict to return
544
while len(elements) and elements[0] is not None:
545
key_dict = key_dict[elements[0]]
548
# a non-existant lookup.
553
key_dict = dicts.pop(-1)
554
# can't be empty or would not exist
555
item, value = key_dict.iteritems().next()
556
if type(value) == dict:
558
dicts.extend(key_dict.itervalues())
561
for value in key_dict.itervalues():
562
yield (self, ) + value
564
yield (self, ) + key_dict
566
def _get_nodes_by_key(self):
567
if self._nodes_by_key is None:
569
if self.reference_lists:
570
for key, (references, value) in self._nodes.iteritems():
571
key_dict = nodes_by_key
572
for subkey in key[:-1]:
573
key_dict = key_dict.setdefault(subkey, {})
574
key_dict[key[-1]] = key, value, references
576
for key, (references, value) in self._nodes.iteritems():
577
key_dict = nodes_by_key
578
for subkey in key[:-1]:
579
key_dict = key_dict.setdefault(subkey, {})
580
key_dict[key[-1]] = key, value
581
self._nodes_by_key = nodes_by_key
582
return self._nodes_by_key
585
"""Return an estimate of the number of keys in this index.
587
For InMemoryGraphIndex the estimate is exact.
589
return len(self._keys) + sum(backing.key_count() for backing in
590
self._backing_indices if backing is not None)
593
"""In memory index's have no known corruption at the moment."""
596
class _LeafNode(object):
597
"""A leaf node for a serialised B+Tree index."""
599
__slots__ = ('keys', 'min_key', 'max_key')
601
def __init__(self, bytes, key_length, ref_list_length):
602
"""Parse bytes to create a leaf node object."""
603
# splitlines mangles the \r delimiters.. don't use it.
604
key_list = _btree_serializer._parse_leaf_lines(bytes,
605
key_length, ref_list_length)
607
self.min_key = key_list[0][0]
608
self.max_key = key_list[-1][0]
610
self.min_key = self.max_key = None
611
self.keys = dict(key_list)
614
class _InternalNode(object):
615
"""An internal node for a serialised B+Tree index."""
617
__slots__ = ('keys', 'offset')
619
def __init__(self, bytes):
620
"""Parse bytes to create an internal node object."""
621
# splitlines mangles the \r delimiters.. don't use it.
622
self.keys = self._parse_lines(bytes.split('\n'))
624
def _parse_lines(self, lines):
626
self.offset = int(lines[1][7:])
627
for line in lines[2:]:
630
# TODO: Switch to StaticTuple here.
631
nodes.append(tuple(map(intern, line.split('\0'))))
635
class BTreeGraphIndex(object):
636
"""Access to nodes via the standard GraphIndex interface for B+Tree's.
638
Individual nodes are held in a LRU cache. This holds the root node in
639
memory except when very large walks are done.
642
def __init__(self, transport, name, size, unlimited_cache=False):
643
"""Create a B+Tree index object on the index name.
645
:param transport: The transport to read data for the index from.
646
:param name: The file name of the index on transport.
647
:param size: Optional size of the index in bytes. This allows
648
compatibility with the GraphIndex API, as well as ensuring that
649
the initial read (to read the root node header) can be done
650
without over-reading even on empty indices, and on small indices
651
allows single-IO to read the entire index.
652
:param unlimited_cache: If set to True, then instead of using an
653
LRUCache with size _NODE_CACHE_SIZE, we will use a dict and always
654
cache all leaf nodes.
656
self._transport = transport
660
self._recommended_pages = self._compute_recommended_pages()
661
self._root_node = None
662
# Default max size is 100,000 leave values
663
self._leaf_value_cache = None # lru_cache.LRUCache(100*1000)
665
self._leaf_node_cache = {}
666
self._internal_node_cache = {}
668
self._leaf_node_cache = lru_cache.LRUCache(_NODE_CACHE_SIZE)
669
# We use a FIFO here just to prevent possible blowout. However, a
670
# 300k record btree has only 3k leaf nodes, and only 20 internal
671
# nodes. A value of 100 scales to ~100*100*100 = 1M records.
672
self._internal_node_cache = fifo_cache.FIFOCache(100)
673
self._key_count = None
674
self._row_lengths = None
675
self._row_offsets = None # Start of each row, [-1] is the end
677
def __eq__(self, other):
678
"""Equal when self and other were created with the same parameters."""
680
type(self) == type(other) and
681
self._transport == other._transport and
682
self._name == other._name and
683
self._size == other._size)
685
def __ne__(self, other):
686
return not self.__eq__(other)
688
def _get_and_cache_nodes(self, nodes):
689
"""Read nodes and cache them in the lru.
691
The nodes list supplied is sorted and then read from disk, each node
692
being inserted it into the _node_cache.
694
Note: Asking for more nodes than the _node_cache can contain will
695
result in some of the results being immediately discarded, to prevent
696
this an assertion is raised if more nodes are asked for than are
699
:return: A dict of {node_pos: node}
702
start_of_leaves = None
703
for node_pos, node in self._read_nodes(sorted(nodes)):
704
if node_pos == 0: # Special case
705
self._root_node = node
707
if start_of_leaves is None:
708
start_of_leaves = self._row_offsets[-2]
709
if node_pos < start_of_leaves:
710
self._internal_node_cache[node_pos] = node
712
self._leaf_node_cache[node_pos] = node
713
found[node_pos] = node
716
def _compute_recommended_pages(self):
717
"""Convert transport's recommended_page_size into btree pages.
719
recommended_page_size is in bytes, we want to know how many _PAGE_SIZE
720
pages fit in that length.
722
recommended_read = self._transport.recommended_page_size()
723
recommended_pages = int(math.ceil(recommended_read /
725
return recommended_pages
727
def _compute_total_pages_in_index(self):
728
"""How many pages are in the index.
730
If we have read the header we will use the value stored there.
731
Otherwise it will be computed based on the length of the index.
733
if self._size is None:
734
raise AssertionError('_compute_total_pages_in_index should not be'
735
' called when self._size is None')
736
if self._root_node is not None:
737
# This is the number of pages as defined by the header
738
return self._row_offsets[-1]
739
# This is the number of pages as defined by the size of the index. They
740
# should be indentical.
741
total_pages = int(math.ceil(self._size / float(_PAGE_SIZE)))
744
def _expand_offsets(self, offsets):
745
"""Find extra pages to download.
747
The idea is that we always want to make big-enough requests (like 64kB
748
for http), so that we don't waste round trips. So given the entries
749
that we already have cached and the new pages being downloaded figure
750
out what other pages we might want to read.
752
See also doc/developers/btree_index_prefetch.txt for more details.
754
:param offsets: The offsets to be read
755
:return: A list of offsets to download
757
if 'index' in debug.debug_flags:
758
trace.mutter('expanding: %s\toffsets: %s', self._name, offsets)
760
if len(offsets) >= self._recommended_pages:
761
# Don't add more, we are already requesting more than enough
762
if 'index' in debug.debug_flags:
763
trace.mutter(' not expanding large request (%s >= %s)',
764
len(offsets), self._recommended_pages)
766
if self._size is None:
767
# Don't try anything, because we don't know where the file ends
768
if 'index' in debug.debug_flags:
769
trace.mutter(' not expanding without knowing index size')
771
total_pages = self._compute_total_pages_in_index()
772
cached_offsets = self._get_offsets_to_cached_pages()
773
# If reading recommended_pages would read the rest of the index, just
775
if total_pages - len(cached_offsets) <= self._recommended_pages:
776
# Read whatever is left
778
expanded = [x for x in xrange(total_pages)
779
if x not in cached_offsets]
781
expanded = range(total_pages)
782
if 'index' in debug.debug_flags:
783
trace.mutter(' reading all unread pages: %s', expanded)
786
if self._root_node is None:
787
# ATM on the first read of the root node of a large index, we don't
788
# bother pre-reading any other pages. This is because the
789
# likelyhood of actually reading interesting pages is very low.
790
# See doc/developers/btree_index_prefetch.txt for a discussion, and
791
# a possible implementation when we are guessing that the second
792
# layer index is small
793
final_offsets = offsets
795
tree_depth = len(self._row_lengths)
796
if len(cached_offsets) < tree_depth and len(offsets) == 1:
797
# We haven't read enough to justify expansion
798
# If we are only going to read the root node, and 1 leaf node,
799
# then it isn't worth expanding our request. Once we've read at
800
# least 2 nodes, then we are probably doing a search, and we
801
# start expanding our requests.
802
if 'index' in debug.debug_flags:
803
trace.mutter(' not expanding on first reads')
805
final_offsets = self._expand_to_neighbors(offsets, cached_offsets,
808
final_offsets = sorted(final_offsets)
809
if 'index' in debug.debug_flags:
810
trace.mutter('expanded: %s', final_offsets)
813
def _expand_to_neighbors(self, offsets, cached_offsets, total_pages):
814
"""Expand requests to neighbors until we have enough pages.
816
This is called from _expand_offsets after policy has determined that we
818
We only want to expand requests within a given layer. We cheat a little
819
bit and assume all requests will be in the same layer. This is true
820
given the current design, but if it changes this algorithm may perform
823
:param offsets: requested offsets
824
:param cached_offsets: offsets for pages we currently have cached
825
:return: A set() of offsets after expansion
827
final_offsets = set(offsets)
829
new_tips = set(final_offsets)
830
while len(final_offsets) < self._recommended_pages and new_tips:
834
first, end = self._find_layer_first_and_end(pos)
837
and previous not in cached_offsets
838
and previous not in final_offsets
839
and previous >= first):
840
next_tips.add(previous)
842
if (after < total_pages
843
and after not in cached_offsets
844
and after not in final_offsets
847
# This would keep us from going bigger than
848
# recommended_pages by only expanding the first offsets.
849
# However, if we are making a 'wide' request, it is
850
# reasonable to expand all points equally.
851
# if len(final_offsets) > recommended_pages:
853
final_offsets.update(next_tips)
857
def clear_cache(self):
858
"""Clear out any cached/memoized values.
860
This can be called at any time, but generally it is used when we have
861
extracted some information, but don't expect to be requesting any more
864
# Note that we don't touch self._root_node or self._internal_node_cache
865
# We don't expect either of those to be big, and it can save
866
# round-trips in the future. We may re-evaluate this if InternalNode
867
# memory starts to be an issue.
868
self._leaf_node_cache.clear()
870
def external_references(self, ref_list_num):
871
if self._root_node is None:
872
self._get_root_node()
873
if ref_list_num + 1 > self.node_ref_lists:
874
raise ValueError('No ref list %d, index has %d ref lists'
875
% (ref_list_num, self.node_ref_lists))
878
for node in self.iter_all_entries():
880
refs.update(node[3][ref_list_num])
883
def _find_layer_first_and_end(self, offset):
884
"""Find the start/stop nodes for the layer corresponding to offset.
886
:return: (first, end)
887
first is the first node in this layer
888
end is the first node of the next layer
891
for roffset in self._row_offsets:
898
def _get_offsets_to_cached_pages(self):
899
"""Determine what nodes we already have cached."""
900
cached_offsets = set(self._internal_node_cache.keys())
901
cached_offsets.update(self._leaf_node_cache.keys())
902
if self._root_node is not None:
903
cached_offsets.add(0)
904
return cached_offsets
906
def _get_root_node(self):
907
if self._root_node is None:
908
# We may not have a root node yet
909
self._get_internal_nodes([0])
910
return self._root_node
912
def _get_nodes(self, cache, node_indexes):
915
for idx in node_indexes:
916
if idx == 0 and self._root_node is not None:
917
found[0] = self._root_node
920
found[idx] = cache[idx]
925
needed = self._expand_offsets(needed)
926
found.update(self._get_and_cache_nodes(needed))
929
def _get_internal_nodes(self, node_indexes):
930
"""Get a node, from cache or disk.
932
After getting it, the node will be cached.
934
return self._get_nodes(self._internal_node_cache, node_indexes)
936
def _cache_leaf_values(self, nodes):
937
"""Cache directly from key => value, skipping the btree."""
938
if self._leaf_value_cache is not None:
939
for node in nodes.itervalues():
940
for key, value in node.keys.iteritems():
941
if key in self._leaf_value_cache:
942
# Don't add the rest of the keys, we've seen this node
945
self._leaf_value_cache[key] = value
947
def _get_leaf_nodes(self, node_indexes):
948
"""Get a bunch of nodes, from cache or disk."""
949
found = self._get_nodes(self._leaf_node_cache, node_indexes)
950
self._cache_leaf_values(found)
953
def iter_all_entries(self):
954
"""Iterate over all keys within the index.
956
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
957
The former tuple is used when there are no reference lists in the
958
index, making the API compatible with simple key:value index types.
959
There is no defined order for the result iteration - it will be in
960
the most efficient order for the index.
962
if 'evil' in debug.debug_flags:
963
trace.mutter_callsite(3,
964
"iter_all_entries scales with size of history.")
965
if not self.key_count():
967
if self._row_offsets[-1] == 1:
968
# There is only the root node, and we read that via key_count()
969
if self.node_ref_lists:
970
for key, (value, refs) in sorted(self._root_node.keys.items()):
971
yield (self, key, value, refs)
973
for key, (value, refs) in sorted(self._root_node.keys.items()):
974
yield (self, key, value)
976
start_of_leaves = self._row_offsets[-2]
977
end_of_leaves = self._row_offsets[-1]
978
needed_offsets = range(start_of_leaves, end_of_leaves)
979
if needed_offsets == [0]:
980
# Special case when we only have a root node, as we have already
982
nodes = [(0, self._root_node)]
984
nodes = self._read_nodes(needed_offsets)
985
# We iterate strictly in-order so that we can use this function
986
# for spilling index builds to disk.
987
if self.node_ref_lists:
988
for _, node in nodes:
989
for key, (value, refs) in sorted(node.keys.items()):
990
yield (self, key, value, refs)
992
for _, node in nodes:
993
for key, (value, refs) in sorted(node.keys.items()):
994
yield (self, key, value)
997
def _multi_bisect_right(in_keys, fixed_keys):
998
"""Find the positions where each 'in_key' would fit in fixed_keys.
1000
This is equivalent to doing "bisect_right" on each in_key into
1003
:param in_keys: A sorted list of keys to match with fixed_keys
1004
:param fixed_keys: A sorted list of keys to match against
1005
:return: A list of (integer position, [key list]) tuples.
1010
# no pointers in the fixed_keys list, which means everything must
1012
return [(0, in_keys)]
1014
# TODO: Iterating both lists will generally take M + N steps
1015
# Bisecting each key will generally take M * log2 N steps.
1016
# If we had an efficient way to compare, we could pick the method
1017
# based on which has the fewer number of steps.
1018
# There is also the argument that bisect_right is a compiled
1019
# function, so there is even more to be gained.
1020
# iter_steps = len(in_keys) + len(fixed_keys)
1021
# bisect_steps = len(in_keys) * math.log(len(fixed_keys), 2)
1022
if len(in_keys) == 1: # Bisect will always be faster for M = 1
1023
return [(bisect_right(fixed_keys, in_keys[0]), in_keys)]
1024
# elif bisect_steps < iter_steps:
1026
# for key in in_keys:
1027
# offsets.setdefault(bisect_right(fixed_keys, key),
1029
# return [(o, offsets[o]) for o in sorted(offsets)]
1030
in_keys_iter = iter(in_keys)
1031
fixed_keys_iter = enumerate(fixed_keys)
1032
cur_in_key = in_keys_iter.next()
1033
cur_fixed_offset, cur_fixed_key = fixed_keys_iter.next()
1035
class InputDone(Exception): pass
1036
class FixedDone(Exception): pass
1041
# TODO: Another possibility is that rather than iterating on each side,
1042
# we could use a combination of bisecting and iterating. For
1043
# example, while cur_in_key < fixed_key, bisect to find its
1044
# point, then iterate all matching keys, then bisect (restricted
1045
# to only the remainder) for the next one, etc.
1048
if cur_in_key < cur_fixed_key:
1050
cur_out = (cur_fixed_offset, cur_keys)
1051
output.append(cur_out)
1052
while cur_in_key < cur_fixed_key:
1053
cur_keys.append(cur_in_key)
1055
cur_in_key = in_keys_iter.next()
1056
except StopIteration:
1058
# At this point cur_in_key must be >= cur_fixed_key
1059
# step the cur_fixed_key until we pass the cur key, or walk off
1061
while cur_in_key >= cur_fixed_key:
1063
cur_fixed_offset, cur_fixed_key = fixed_keys_iter.next()
1064
except StopIteration:
1067
# We consumed all of the input, nothing more to do
1070
# There was some input left, but we consumed all of fixed, so we
1071
# have to add one more for the tail
1072
cur_keys = [cur_in_key]
1073
cur_keys.extend(in_keys_iter)
1074
cur_out = (len(fixed_keys), cur_keys)
1075
output.append(cur_out)
1078
def _walk_through_internal_nodes(self, keys):
1079
"""Take the given set of keys, and find the corresponding LeafNodes.
1081
:param keys: An unsorted iterable of keys to search for
1082
:return: (nodes, index_and_keys)
1083
nodes is a dict mapping {index: LeafNode}
1084
keys_at_index is a list of tuples of [(index, [keys for Leaf])]
1086
# 6 seconds spent in miss_torture using the sorted() line.
1087
# Even with out of order disk IO it seems faster not to sort it when
1088
# large queries are being made.
1089
keys_at_index = [(0, sorted(keys))]
1091
for row_pos, next_row_start in enumerate(self._row_offsets[1:-1]):
1092
node_indexes = [idx for idx, s_keys in keys_at_index]
1093
nodes = self._get_internal_nodes(node_indexes)
1095
next_nodes_and_keys = []
1096
for node_index, sub_keys in keys_at_index:
1097
node = nodes[node_index]
1098
positions = self._multi_bisect_right(sub_keys, node.keys)
1099
node_offset = next_row_start + node.offset
1100
next_nodes_and_keys.extend([(node_offset + pos, s_keys)
1101
for pos, s_keys in positions])
1102
keys_at_index = next_nodes_and_keys
1103
# We should now be at the _LeafNodes
1104
node_indexes = [idx for idx, s_keys in keys_at_index]
1106
# TODO: We may *not* want to always read all the nodes in one
1107
# big go. Consider setting a max size on this.
1108
nodes = self._get_leaf_nodes(node_indexes)
1109
return nodes, keys_at_index
1111
def iter_entries(self, keys):
1112
"""Iterate over keys within the index.
1114
:param keys: An iterable providing the keys to be retrieved.
1115
:return: An iterable as per iter_all_entries, but restricted to the
1116
keys supplied. No additional keys will be returned, and every
1117
key supplied that is in the index will be returned.
1119
# 6 seconds spent in miss_torture using the sorted() line.
1120
# Even with out of order disk IO it seems faster not to sort it when
1121
# large queries are being made.
1122
# However, now that we are doing multi-way bisecting, we need the keys
1123
# in sorted order anyway. We could change the multi-way code to not
1124
# require sorted order. (For example, it bisects for the first node,
1125
# does an in-order search until a key comes before the current point,
1126
# which it then bisects for, etc.)
1127
keys = frozenset(keys)
1131
if not self.key_count():
1135
if self._leaf_value_cache is None:
1139
value = self._leaf_value_cache.get(key, None)
1140
if value is not None:
1141
# This key is known not to be here, skip it
1143
if self.node_ref_lists:
1144
yield (self, key, value, refs)
1146
yield (self, key, value)
1148
needed_keys.append(key)
1154
nodes, nodes_and_keys = self._walk_through_internal_nodes(needed_keys)
1155
for node_index, sub_keys in nodes_and_keys:
1158
node = nodes[node_index]
1159
for next_sub_key in sub_keys:
1160
if next_sub_key in node.keys:
1161
value, refs = node.keys[next_sub_key]
1162
if self.node_ref_lists:
1163
yield (self, next_sub_key, value, refs)
1165
yield (self, next_sub_key, value)
1167
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
1168
"""Find the parent_map information for the set of keys.
1170
This populates the parent_map dict and missing_keys set based on the
1171
queried keys. It also can fill out an arbitrary number of parents that
1172
it finds while searching for the supplied keys.
1174
It is unlikely that you want to call this directly. See
1175
"CombinedGraphIndex.find_ancestry()" for a more appropriate API.
1177
:param keys: A keys whose ancestry we want to return
1178
Every key will either end up in 'parent_map' or 'missing_keys'.
1179
:param ref_list_num: This index in the ref_lists is the parents we
1181
:param parent_map: {key: parent_keys} for keys that are present in this
1182
index. This may contain more entries than were in 'keys', that are
1183
reachable ancestors of the keys requested.
1184
:param missing_keys: keys which are known to be missing in this index.
1185
This may include parents that were not directly requested, but we
1186
were able to determine that they are not present in this index.
1187
:return: search_keys parents that were found but not queried to know
1188
if they are missing or present. Callers can re-query this index for
1189
those keys, and they will be placed into parent_map or missing_keys
1191
if not self.key_count():
1192
# We use key_count() to trigger reading the root node and
1193
# determining info about this BTreeGraphIndex
1194
# If we don't have any keys, then everything is missing
1195
missing_keys.update(keys)
1197
if ref_list_num >= self.node_ref_lists:
1198
raise ValueError('No ref list %d, index has %d ref lists'
1199
% (ref_list_num, self.node_ref_lists))
1201
# The main trick we are trying to accomplish is that when we find a
1202
# key listing its parents, we expect that the parent key is also likely
1203
# to sit on the same page. Allowing us to expand parents quickly
1204
# without suffering the full stack of bisecting, etc.
1205
nodes, nodes_and_keys = self._walk_through_internal_nodes(keys)
1207
# These are parent keys which could not be immediately resolved on the
1208
# page where the child was present. Note that we may already be
1209
# searching for that key, and it may actually be present [or known
1210
# missing] on one of the other pages we are reading.
1212
# We could try searching for them in the immediate previous or next
1213
# page. If they occur "later" we could put them in a pending lookup
1214
# set, and then for each node we read thereafter we could check to
1215
# see if they are present.
1216
# However, we don't know the impact of keeping this list of things
1217
# that I'm going to search for every node I come across from here on
1219
# It doesn't handle the case when the parent key is missing on a
1220
# page that we *don't* read. So we already have to handle being
1221
# re-entrant for that.
1222
# Since most keys contain a date string, they are more likely to be
1223
# found earlier in the file than later, but we would know that right
1224
# away (key < min_key), and wouldn't keep searching it on every other
1225
# page that we read.
1226
# Mostly, it is an idea, one which should be benchmarked.
1227
parents_not_on_page = set()
1229
for node_index, sub_keys in nodes_and_keys:
1232
# sub_keys is all of the keys we are looking for that should exist
1233
# on this page, if they aren't here, then they won't be found
1234
node = nodes[node_index]
1235
node_keys = node.keys
1236
parents_to_check = set()
1237
for next_sub_key in sub_keys:
1238
if next_sub_key not in node_keys:
1239
# This one is just not present in the index at all
1240
missing_keys.add(next_sub_key)
1242
value, refs = node_keys[next_sub_key]
1243
parent_keys = refs[ref_list_num]
1244
parent_map[next_sub_key] = parent_keys
1245
parents_to_check.update(parent_keys)
1246
# Don't look for things we've already found
1247
parents_to_check = parents_to_check.difference(parent_map)
1248
# this can be used to test the benefit of having the check loop
1250
# parents_not_on_page.update(parents_to_check)
1252
while parents_to_check:
1253
next_parents_to_check = set()
1254
for key in parents_to_check:
1255
if key in node_keys:
1256
value, refs = node_keys[key]
1257
parent_keys = refs[ref_list_num]
1258
parent_map[key] = parent_keys
1259
next_parents_to_check.update(parent_keys)
1261
# This parent either is genuinely missing, or should be
1262
# found on another page. Perf test whether it is better
1263
# to check if this node should fit on this page or not.
1264
# in the 'everything-in-one-pack' scenario, this *not*
1265
# doing the check is 237ms vs 243ms.
1266
# So slightly better, but I assume the standard 'lots
1267
# of packs' is going to show a reasonable improvement
1268
# from the check, because it avoids 'going around
1269
# again' for everything that is in another index
1270
# parents_not_on_page.add(key)
1271
# Missing for some reason
1272
if key < node.min_key:
1273
# in the case of bzr.dev, 3.4k/5.3k misses are
1274
# 'earlier' misses (65%)
1275
parents_not_on_page.add(key)
1276
elif key > node.max_key:
1277
# This parent key would be present on a different
1279
parents_not_on_page.add(key)
1281
# assert key != node.min_key and key != node.max_key
1282
# If it was going to be present, it would be on
1283
# *this* page, so mark it missing.
1284
missing_keys.add(key)
1285
parents_to_check = next_parents_to_check.difference(parent_map)
1286
# Might want to do another .difference() from missing_keys
1287
# parents_not_on_page could have been found on a different page, or be
1288
# known to be missing. So cull out everything that has already been
1290
search_keys = parents_not_on_page.difference(
1291
parent_map).difference(missing_keys)
1294
def iter_entries_prefix(self, keys):
1295
"""Iterate over keys within the index using prefix matching.
1297
Prefix matching is applied within the tuple of a key, not to within
1298
the bytestring of each key element. e.g. if you have the keys ('foo',
1299
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1300
only the former key is returned.
1302
WARNING: Note that this method currently causes a full index parse
1303
unconditionally (which is reasonably appropriate as it is a means for
1304
thunking many small indices into one larger one and still supplies
1305
iter_all_entries at the thunk layer).
1307
:param keys: An iterable providing the key prefixes to be retrieved.
1308
Each key prefix takes the form of a tuple the length of a key, but
1309
with the last N elements 'None' rather than a regular bytestring.
1310
The first element cannot be 'None'.
1311
:return: An iterable as per iter_all_entries, but restricted to the
1312
keys with a matching prefix to those supplied. No additional keys
1313
will be returned, and every match that is in the index will be
1316
keys = sorted(set(keys))
1319
# Load if needed to check key lengths
1320
if self._key_count is None:
1321
self._get_root_node()
1322
# TODO: only access nodes that can satisfy the prefixes we are looking
1323
# for. For now, to meet API usage (as this function is not used by
1324
# current bzrlib) just suck the entire index and iterate in memory.
1326
if self.node_ref_lists:
1327
if self._key_length == 1:
1328
for _1, key, value, refs in self.iter_all_entries():
1329
nodes[key] = value, refs
1332
for _1, key, value, refs in self.iter_all_entries():
1333
key_value = key, value, refs
1334
# For a key of (foo, bar, baz) create
1335
# _nodes_by_key[foo][bar][baz] = key_value
1336
key_dict = nodes_by_key
1337
for subkey in key[:-1]:
1338
key_dict = key_dict.setdefault(subkey, {})
1339
key_dict[key[-1]] = key_value
1341
if self._key_length == 1:
1342
for _1, key, value in self.iter_all_entries():
1346
for _1, key, value in self.iter_all_entries():
1347
key_value = key, value
1348
# For a key of (foo, bar, baz) create
1349
# _nodes_by_key[foo][bar][baz] = key_value
1350
key_dict = nodes_by_key
1351
for subkey in key[:-1]:
1352
key_dict = key_dict.setdefault(subkey, {})
1353
key_dict[key[-1]] = key_value
1354
if self._key_length == 1:
1358
raise errors.BadIndexKey(key)
1359
if len(key) != self._key_length:
1360
raise errors.BadIndexKey(key)
1362
if self.node_ref_lists:
1363
value, node_refs = nodes[key]
1364
yield self, key, value, node_refs
1366
yield self, key, nodes[key]
1373
raise errors.BadIndexKey(key)
1374
if len(key) != self._key_length:
1375
raise errors.BadIndexKey(key)
1376
# find what it refers to:
1377
key_dict = nodes_by_key
1378
elements = list(key)
1379
# find the subdict whose contents should be returned.
1381
while len(elements) and elements[0] is not None:
1382
key_dict = key_dict[elements[0]]
1385
# a non-existant lookup.
1390
key_dict = dicts.pop(-1)
1391
# can't be empty or would not exist
1392
item, value = key_dict.iteritems().next()
1393
if type(value) == dict:
1395
dicts.extend(key_dict.itervalues())
1398
for value in key_dict.itervalues():
1399
# each value is the key:value:node refs tuple
1401
yield (self, ) + value
1403
# the last thing looked up was a terminal element
1404
yield (self, ) + key_dict
1406
def key_count(self):
1407
"""Return an estimate of the number of keys in this index.
1409
For BTreeGraphIndex the estimate is exact as it is contained in the
1412
if self._key_count is None:
1413
self._get_root_node()
1414
return self._key_count
1416
def _compute_row_offsets(self):
1417
"""Fill out the _row_offsets attribute based on _row_lengths."""
1420
for row in self._row_lengths:
1421
offsets.append(row_offset)
1423
offsets.append(row_offset)
1424
self._row_offsets = offsets
1426
def _parse_header_from_bytes(self, bytes):
1427
"""Parse the header from a region of bytes.
1429
:param bytes: The data to parse.
1430
:return: An offset, data tuple such as readv yields, for the unparsed
1431
data. (which may be of length 0).
1433
signature = bytes[0:len(self._signature())]
1434
if not signature == self._signature():
1435
raise errors.BadIndexFormatSignature(self._name, BTreeGraphIndex)
1436
lines = bytes[len(self._signature()):].splitlines()
1437
options_line = lines[0]
1438
if not options_line.startswith(_OPTION_NODE_REFS):
1439
raise errors.BadIndexOptions(self)
1441
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
1443
raise errors.BadIndexOptions(self)
1444
options_line = lines[1]
1445
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
1446
raise errors.BadIndexOptions(self)
1448
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
1450
raise errors.BadIndexOptions(self)
1451
options_line = lines[2]
1452
if not options_line.startswith(_OPTION_LEN):
1453
raise errors.BadIndexOptions(self)
1455
self._key_count = int(options_line[len(_OPTION_LEN):])
1457
raise errors.BadIndexOptions(self)
1458
options_line = lines[3]
1459
if not options_line.startswith(_OPTION_ROW_LENGTHS):
1460
raise errors.BadIndexOptions(self)
1462
self._row_lengths = map(int, [length for length in
1463
options_line[len(_OPTION_ROW_LENGTHS):].split(',')
1466
raise errors.BadIndexOptions(self)
1467
self._compute_row_offsets()
1469
# calculate the bytes we have processed
1470
header_end = (len(signature) + sum(map(len, lines[0:4])) + 4)
1471
return header_end, bytes[header_end:]
1473
def _read_nodes(self, nodes):
1474
"""Read some nodes from disk into the LRU cache.
1476
This performs a readv to get the node data into memory, and parses each
1477
node, then yields it to the caller. The nodes are requested in the
1478
supplied order. If possible doing sort() on the list before requesting
1479
a read may improve performance.
1481
:param nodes: The nodes to read. 0 - first node, 1 - second node etc.
1484
# may be the byte string of the whole file
1486
# list of (offset, length) regions of the file that should, evenually
1487
# be read in to data_ranges, either from 'bytes' or from the transport
1490
offset = index * _PAGE_SIZE
1493
# Root node - special case
1495
size = min(_PAGE_SIZE, self._size)
1497
# The only case where we don't know the size, is for very
1498
# small indexes. So we read the whole thing
1499
bytes = self._transport.get_bytes(self._name)
1500
self._size = len(bytes)
1501
# the whole thing should be parsed out of 'bytes'
1502
ranges.append((0, len(bytes)))
1505
if offset > self._size:
1506
raise AssertionError('tried to read past the end'
1507
' of the file %s > %s'
1508
% (offset, self._size))
1509
size = min(size, self._size - offset)
1510
ranges.append((offset, size))
1513
elif bytes is not None:
1514
# already have the whole file
1515
data_ranges = [(start, bytes[start:start+_PAGE_SIZE])
1516
for start in xrange(0, len(bytes), _PAGE_SIZE)]
1517
elif self._file is None:
1518
data_ranges = self._transport.readv(self._name, ranges)
1521
for offset, size in ranges:
1522
self._file.seek(offset)
1523
data_ranges.append((offset, self._file.read(size)))
1524
for offset, data in data_ranges:
1526
# extract the header
1527
offset, data = self._parse_header_from_bytes(data)
1530
bytes = zlib.decompress(data)
1531
if bytes.startswith(_LEAF_FLAG):
1532
node = _LeafNode(bytes, self._key_length, self.node_ref_lists)
1533
elif bytes.startswith(_INTERNAL_FLAG):
1534
node = _InternalNode(bytes)
1536
raise AssertionError("Unknown node type for %r" % bytes)
1537
yield offset / _PAGE_SIZE, node
1539
def _signature(self):
1540
"""The file signature for this index type."""
1544
"""Validate that everything in the index can be accessed."""
1545
# just read and parse every node.
1546
self._get_root_node()
1547
if len(self._row_lengths) > 1:
1548
start_node = self._row_offsets[1]
1550
# We shouldn't be reading anything anyway
1552
node_end = self._row_offsets[-1]
1553
for node in self._read_nodes(range(start_node, node_end)):
1558
from bzrlib import _btree_serializer_pyx as _btree_serializer
1559
except ImportError, e:
1560
osutils.failed_to_load_extension(e)
1561
from bzrlib import _btree_serializer_py as _btree_serializer
added
removed
bzrlib/btree_index.py
