40
from __future__ import absolute_import
43
from bzrlib import lazy_import
44
lazy_import.lazy_import(globals(), """
57
from bzrlib.static_tuple import StaticTuple
53
from ..sixish import (
57
from ..sixish import PY3
58
from ..static_tuple import StaticTuple
60
61
# If each line is 50 bytes, and you have 255 internal pages, with 255-way fan
61
62
# out, it takes 3.1MB to cache the layer.
62
_PAGE_CACHE_SIZE = 4*1024*1024
63
_PAGE_CACHE_SIZE = 4 * 1024 * 1024
63
64
# Per thread caches for 2 reasons:
64
65
# - in the server we may be serving very different content, so we get less
90
92
_INTERESTING_NEW_SIZE = 50
91
93
# If a ChildNode shrinks by more than this amount, we check for a remap
92
94
_INTERESTING_SHRINKAGE_LIMIT = 20
93
# If we delete more than this many nodes applying a delta, we check for a remap
94
_INTERESTING_DELETES_LIMIT = 5
97
97
def _search_key_plain(key):
98
98
"""Map the key tuple into a search string that just uses the key bytes."""
99
return '\x00'.join(key)
99
return b'\x00'.join(key)
102
102
search_key_registry = registry.Registry()
103
search_key_registry.register('plain', _search_key_plain)
103
search_key_registry.register(b'plain', _search_key_plain)
106
106
class CHKMap(object):
135
135
into the map; if old_key is not None, then the old mapping
136
136
of old_key is removed.
139
139
# Check preconditions first.
140
140
as_st = StaticTuple.from_sequence
141
new_items = set([as_st(key) for (old, key, value) in delta
142
if key is not None and old is None])
141
new_items = {as_st(key) for (old, key, value) in delta
142
if key is not None and old is None}
143
143
existing_new = list(self.iteritems(key_filter=new_items))
145
145
raise errors.InconsistentDeltaDelta(delta,
146
"New items are already in the map %r." % existing_new)
146
"New items are already in the map %r." % existing_new)
147
147
# Now apply changes.
148
148
for old, new, value in delta:
149
149
if old is not None and old != new:
150
150
self.unmap(old, check_remap=False)
152
152
for old, new, value in delta:
153
153
if new is not None:
154
154
self.map(new, value)
155
if delete_count > _INTERESTING_DELETES_LIMIT:
156
trace.mutter("checking remap as %d deletions", delete_count)
157
156
self._check_remap()
158
157
return self._save()
160
159
def _ensure_root(self):
161
160
"""Ensure that the root node is an object not a key."""
162
if type(self._root_node) is StaticTuple:
161
if isinstance(self._root_node, StaticTuple):
163
162
# Demand-load the root
164
163
self._root_node = self._get_node(self._root_node)
173
172
:param node: A tuple key or node object.
174
173
:return: A node object.
176
if type(node) is StaticTuple:
175
if isinstance(node, StaticTuple):
177
176
bytes = self._read_bytes(node)
178
177
return _deserialise(bytes, node,
179
search_key_func=self._search_key_func)
178
search_key_func=self._search_key_func)
185
184
return _get_cache()[key]
187
186
stream = self._store.get_record_stream([key], 'unordered', True)
188
bytes = stream.next().get_bytes_as('fulltext')
187
bytes = next(stream).get_bytes_as('fulltext')
189
188
_get_cache()[key] = bytes
192
def _dump_tree(self, include_keys=False):
191
def _dump_tree(self, include_keys=False, encoding='utf-8'):
193
192
"""Return the tree in a string representation."""
194
193
self._ensure_root()
195
res = self._dump_tree_node(self._root_node, prefix='', indent='',
196
include_keys=include_keys)
197
res.append('') # Give a trailing '\n'
195
def decode(x): return x.decode(encoding)
197
def decode(x): return x
198
res = self._dump_tree_node(self._root_node, prefix=b'', indent='',
199
decode=decode, include_keys=include_keys)
200
res.append('') # Give a trailing '\n'
198
201
return '\n'.join(res)
200
def _dump_tree_node(self, node, prefix, indent, include_keys=True):
203
def _dump_tree_node(self, node, prefix, indent, decode, include_keys=True):
201
204
"""For this node and all children, generate a string representation."""
203
206
if not include_keys:
206
209
node_key = node.key()
207
210
if node_key is not None:
208
key_str = ' %s' % (node_key[0],)
211
key_str = ' %s' % (decode(node_key[0]),)
210
213
key_str = ' None'
211
result.append('%s%r %s%s' % (indent, prefix, node.__class__.__name__,
214
result.append('%s%r %s%s' % (indent, decode(prefix), node.__class__.__name__,
213
if type(node) is InternalNode:
216
if isinstance(node, InternalNode):
214
217
# Trigger all child nodes to get loaded
215
218
list(node._iter_nodes(self._store))
216
for prefix, sub in sorted(node._items.iteritems()):
219
for prefix, sub in sorted(viewitems(node._items)):
217
220
result.extend(self._dump_tree_node(sub, prefix, indent + ' ',
218
include_keys=include_keys))
221
decode=decode, include_keys=include_keys))
220
for key, value in sorted(node._items.iteritems()):
223
for key, value in sorted(viewitems(node._items)):
221
224
# Don't use prefix nor indent here to line up when used in
222
225
# tests in conjunction with assertEqualDiff
223
result.append(' %r %r' % (tuple(key), value))
226
result.append(' %r %r' % (
227
tuple([decode(ke) for ke in key]), decode(value)))
227
231
def from_dict(klass, store, initial_value, maximum_size=0, key_width=1,
228
search_key_func=None):
232
search_key_func=None):
229
233
"""Create a CHKMap in store with initial_value as the content.
231
235
:param store: The store to record initial_value in, a VersionedFiles
242
246
:return: The root chk of the resulting CHKMap.
244
248
root_key = klass._create_directly(store, initial_value,
245
maximum_size=maximum_size, key_width=key_width,
246
search_key_func=search_key_func)
247
if type(root_key) is not StaticTuple:
249
maximum_size=maximum_size, key_width=key_width,
250
search_key_func=search_key_func)
251
if not isinstance(root_key, StaticTuple):
248
252
raise AssertionError('we got a %s instead of a StaticTuple'
249
253
% (type(root_key),))
256
260
result._root_node.set_maximum_size(maximum_size)
257
261
result._root_node._key_width = key_width
259
for key, value in initial_value.items():
263
for key, value in viewitems(initial_value):
260
264
delta.append((None, key, value))
261
265
root_key = result.apply_delta(delta)
268
272
node.set_maximum_size(maximum_size)
269
273
node._key_width = key_width
270
274
as_st = StaticTuple.from_sequence
271
node._items = dict([(as_st(key), val) for key, val
272
in initial_value.iteritems()])
273
node._raw_size = sum([node._key_value_len(key, value)
274
for key,value in node._items.iteritems()])
275
node._items = dict((as_st(key), val)
276
for key, val in viewitems(initial_value))
277
node._raw_size = sum(node._key_value_len(key, value)
278
for key, value in viewitems(node._items))
275
279
node._len = len(node._items)
276
280
node._compute_search_prefix()
277
281
node._compute_serialised_prefix()
280
and node._current_size() > maximum_size):
282
if (node._len > 1 and
284
node._current_size() > maximum_size):
281
285
prefix, node_details = node._split(store)
282
286
if len(node_details) == 1:
283
287
raise AssertionError('Failed to split using node._split')
329
333
# key_path (a list of tuples, tail-sharing down the tree.)
330
334
self_pending = []
331
335
basis_pending = []
332
337
def process_node(node, path, a_map, pending):
333
338
# take a node and expand it
334
339
node = a_map._get_node(node)
335
if type(node) == LeafNode:
340
if isinstance(node, LeafNode):
336
341
path = (node._key, path)
337
for key, value in node._items.items():
342
for key, value in viewitems(node._items):
338
343
# For a LeafNode, the key is a serialized_key, rather than
339
344
# a search_key, but the heap is using search_keys
340
345
search_key = node._search_key_func(key)
343
348
# type(node) == InternalNode
344
349
path = (node._key, path)
345
for prefix, child in node._items.items():
350
for prefix, child in viewitems(node._items):
346
351
heapq.heappush(pending, (prefix, None, child, path))
347
353
def process_common_internal_nodes(self_node, basis_node):
348
self_items = set(self_node._items.items())
349
basis_items = set(basis_node._items.items())
354
self_items = set(viewitems(self_node._items))
355
basis_items = set(viewitems(basis_node._items))
350
356
path = (self_node._key, None)
351
357
for prefix, child in self_items - basis_items:
352
358
heapq.heappush(self_pending, (prefix, None, child, path))
353
359
path = (basis_node._key, None)
354
360
for prefix, child in basis_items - self_items:
355
361
heapq.heappush(basis_pending, (prefix, None, child, path))
356
363
def process_common_leaf_nodes(self_node, basis_node):
357
self_items = set(self_node._items.items())
358
basis_items = set(basis_node._items.items())
364
self_items = set(viewitems(self_node._items))
365
basis_items = set(viewitems(basis_node._items))
359
366
path = (self_node._key, None)
360
367
for key, value in self_items - basis_items:
361
368
prefix = self._search_key_func(key)
364
371
for key, value in basis_items - self_items:
365
372
prefix = basis._search_key_func(key)
366
373
heapq.heappush(basis_pending, (prefix, key, value, path))
367
375
def process_common_prefix_nodes(self_node, self_path,
368
376
basis_node, basis_path):
369
377
# Would it be more efficient if we could request both at the same
371
379
self_node = self._get_node(self_node)
372
380
basis_node = basis._get_node(basis_node)
373
if (type(self_node) == InternalNode
374
and type(basis_node) == InternalNode):
381
if (isinstance(self_node, InternalNode) and
382
isinstance(basis_node, InternalNode)):
375
383
# Matching internal nodes
376
384
process_common_internal_nodes(self_node, basis_node)
377
elif (type(self_node) == LeafNode
378
and type(basis_node) == LeafNode):
385
elif (isinstance(self_node, LeafNode) and
386
isinstance(basis_node, LeafNode)):
379
387
process_common_leaf_nodes(self_node, basis_node)
381
389
process_node(self_node, self_path, self, self_pending)
384
392
self_seen = set()
385
393
basis_seen = set()
386
394
excluded_keys = set()
387
396
def check_excluded(key_path):
388
397
# Note that this is N^2, it depends on us trimming trees
389
398
# aggressively to not become slow.
390
399
# A better implementation would probably have a reverse map
391
400
# back to the children of a node, and jump straight to it when
392
401
# a common node is detected, the proceed to remove the already
393
# pending children. bzrlib.graph has a searcher module with a
402
# pending children. breezy.graph has a searcher module with a
394
403
# similar problem.
395
404
while key_path is not None:
396
405
key, key_path = key_path
473
482
basis_details = heapq.heappop(basis_pending)
474
483
if self_details[2] != basis_details[2]:
475
484
yield (self_details[1],
476
basis_details[2], self_details[2])
485
basis_details[2], self_details[2])
478
487
# At least one side wasn't a simple value
479
if (self._node_key(self_pending[0][2]) ==
480
self._node_key(basis_pending[0][2])):
488
if (self._node_key(self_pending[0][2])
489
== self._node_key(basis_pending[0][2])):
481
490
# Identical pointers, skip (and don't bother adding to
482
491
# excluded, it won't turn up again.
483
492
heapq.heappop(self_pending)
543
552
self._root_node = node_details[0][1]
545
554
self._root_node = InternalNode(prefix,
546
search_key_func=self._search_key_func)
555
search_key_func=self._search_key_func)
547
556
self._root_node.set_maximum_size(node_details[0][1].maximum_size)
548
557
self._root_node._key_width = node_details[0][1]._key_width
549
558
for split, node in node_details:
552
561
def _node_key(self, node):
553
562
"""Get the key for a node whether it's a tuple or node."""
554
if type(node) is tuple:
563
if isinstance(node, tuple):
555
564
node = StaticTuple.from_sequence(node)
556
if type(node) is StaticTuple:
565
if isinstance(node, StaticTuple):
562
571
"""remove key from the map."""
563
572
key = StaticTuple.from_sequence(key)
564
573
self._ensure_root()
565
if type(self._root_node) is InternalNode:
574
if isinstance(self._root_node, InternalNode):
566
575
unmapped = self._root_node.unmap(self._store, key,
567
check_remap=check_remap)
576
check_remap=check_remap)
569
578
unmapped = self._root_node.unmap(self._store, key)
570
579
self._root_node = unmapped
572
581
def _check_remap(self):
573
582
"""Check if nodes can be collapsed."""
574
583
self._ensure_root()
575
if type(self._root_node) is InternalNode:
576
self._root_node._check_remap(self._store)
584
if isinstance(self._root_node, InternalNode):
585
self._root_node = self._root_node._check_remap(self._store)
579
588
"""Save the map completely.
581
590
:return: The key of the root node.
583
if type(self._root_node) is StaticTuple:
592
if isinstance(self._root_node, StaticTuple):
585
594
return self._root_node
586
595
keys = list(self._root_node.serialise(self._store))
597
606
__slots__ = ('_key', '_len', '_maximum_size', '_key_width',
598
607
'_raw_size', '_items', '_search_prefix', '_search_key_func'
601
610
def __init__(self, key_width=1):
602
611
"""Create a node.
711
721
'%s(key:%s len:%s size:%s max:%s prefix:%s keywidth:%s items:%s)' \
712
722
% (self.__class__.__name__, self._key, self._len, self._raw_size,
713
self._maximum_size, self._search_prefix, self._key_width, items_str)
723
self._maximum_size, self._search_prefix, self._key_width, items_str)
715
725
def _current_size(self):
716
726
"""Answer the current serialised size of this node.
728
738
prefix_len = len(self._common_serialised_prefix)
729
739
bytes_for_items = (self._raw_size - (prefix_len * self._len))
730
return (9 # 'chkleaf:\n'
731
+ len(str(self._maximum_size)) + 1
732
+ len(str(self._key_width)) + 1
733
+ len(str(self._len)) + 1
740
return (9 + # 'chkleaf:\n' +
741
len(str(self._maximum_size)) + 1 +
742
len(str(self._key_width)) + 1 +
743
len(str(self._len)) + 1 +
738
748
def deserialise(klass, bytes, key, search_key_func=None):
769
779
# Short items, we need to match based on a prefix
770
length_filter = filters.setdefault(len(key), set())
771
length_filter.add(key)
780
filters.setdefault(len(key), set()).add(key)
773
filters = filters.items()
774
for item in self._items.iteritems():
775
for length, length_filter in filters:
782
filters_itemview = viewitems(filters)
783
for item in viewitems(self._items):
784
for length, length_filter in filters_itemview:
776
785
if item[0][:length] in length_filter:
780
for item in self._items.iteritems():
789
for item in viewitems(self._items):
783
792
def _key_value_len(self, key, value):
784
793
# TODO: Should probably be done without actually joining the key, but
785
794
# then that can be done via the C extension
786
return (len(self._serialise_key(key)) + 1
787
+ len(str(value.count('\n'))) + 1
795
return (len(self._serialise_key(key)) + 1 +
796
len(b'%d' % value.count(b'\n')) + 1 +
790
799
def _search_key(self, key):
791
800
return self._search_key_func(key)
816
825
self._search_prefix = self.common_prefix(
817
826
self._search_prefix, search_key)
819
and self._maximum_size
820
and self._current_size() > self._maximum_size):
827
if (self._len > 1 and
828
self._maximum_size and
829
self._current_size() > self._maximum_size):
821
830
# Check to see if all of the search_keys for this node are
822
831
# identical. We allow the node to grow under that circumstance
823
832
# (we could track this as common state, but it is infrequent)
824
if (search_key != self._search_prefix
825
or not self._are_search_keys_identical()):
833
if (search_key != self._search_prefix or
834
not self._are_search_keys_identical()):
839
848
common_prefix = self._search_prefix
840
849
split_at = len(common_prefix) + 1
842
for key, value in self._items.iteritems():
851
for key, value in viewitems(self._items):
843
852
search_key = self._search_key(key)
844
853
prefix = search_key[:split_at]
845
854
# TODO: Generally only 1 key can be exactly the right length,
851
860
# may get a '\00' node anywhere, but won't have keys of
852
861
# different lengths.
853
862
if len(prefix) < split_at:
854
prefix += '\x00'*(split_at - len(prefix))
863
prefix += b'\x00' * (split_at - len(prefix))
855
864
if prefix not in result:
856
865
node = LeafNode(search_key_func=self._search_key_func)
857
866
node.set_maximum_size(self._maximum_size)
867
876
result.pop(prefix)
868
877
new_node = InternalNode(sub_prefix,
869
search_key_func=self._search_key_func)
878
search_key_func=self._search_key_func)
870
879
new_node.set_maximum_size(self._maximum_size)
871
880
new_node._key_width = self._key_width
872
881
for split, node in node_details:
873
882
new_node.add_node(split, node)
874
883
result[prefix] = new_node
875
return common_prefix, result.items()
884
return common_prefix, list(viewitems(result))
877
886
def map(self, store, key, value):
878
887
"""Map key to value."""
886
895
if self._search_prefix is _unknown:
887
896
raise AssertionError('%r must be known' % self._search_prefix)
888
return self._search_prefix, [("", self)]
897
return self._search_prefix, [(b"", self)]
890
_serialise_key = '\x00'.join
899
_serialise_key = b'\x00'.join
892
901
def serialise(self, store):
893
902
"""Serialise the LeafNode to store.
895
904
:param store: A VersionedFiles honouring the CHK extensions.
896
905
:return: An iterable of the keys inserted by this operation.
898
lines = ["chkleaf:\n"]
899
lines.append("%d\n" % self._maximum_size)
900
lines.append("%d\n" % self._key_width)
901
lines.append("%d\n" % self._len)
907
lines = [b"chkleaf:\n"]
908
lines.append(b"%d\n" % self._maximum_size)
909
lines.append(b"%d\n" % self._key_width)
910
lines.append(b"%d\n" % self._len)
902
911
if self._common_serialised_prefix is None:
904
913
if len(self._items) != 0:
905
914
raise AssertionError('If _common_serialised_prefix is None'
906
' we should have no items')
915
' we should have no items')
908
lines.append('%s\n' % (self._common_serialised_prefix,))
917
lines.append(b'%s\n' % (self._common_serialised_prefix,))
909
918
prefix_len = len(self._common_serialised_prefix)
910
for key, value in sorted(self._items.items()):
919
for key, value in sorted(viewitems(self._items)):
911
920
# Always add a final newline
912
value_lines = osutils.chunks_to_lines([value + '\n'])
913
serialized = "%s\x00%s\n" % (self._serialise_key(key),
921
value_lines = osutils.chunks_to_lines([value + b'\n'])
922
serialized = b"%s\x00%d\n" % (self._serialise_key(key),
915
924
if not serialized.startswith(self._common_serialised_prefix):
916
925
raise AssertionError('We thought the common prefix was %r'
917
' but entry %r does not have it in common'
918
% (self._common_serialised_prefix, serialized))
926
' but entry %r does not have it in common'
927
% (self._common_serialised_prefix, serialized))
919
928
lines.append(serialized[prefix_len:])
920
929
lines.extend(value_lines)
921
930
sha1, _, _ = store.add_lines((None,), (), lines)
922
self._key = StaticTuple("sha1:" + sha1,).intern()
923
bytes = ''.join(lines)
924
if len(bytes) != self._current_size():
931
self._key = StaticTuple(b"sha1:" + sha1,).intern()
932
data = b''.join(lines)
933
if len(data) != self._current_size():
925
934
raise AssertionError('Invalid _current_size')
926
_get_cache().add(self._key, bytes)
935
_get_cache()[self._key] = data
927
936
return [self._key]
996
1005
__slots__ = ('_node_width',)
998
def __init__(self, prefix='', search_key_func=None):
1007
def __init__(self, prefix=b'', search_key_func=None):
999
1008
Node.__init__(self)
1000
1009
# The size of an internalnode with default values and no children.
1001
1010
# How many octets key prefixes within this node are.
1016
1025
raise AssertionError("_search_prefix should not be None")
1017
1026
if not prefix.startswith(self._search_prefix):
1018
1027
raise AssertionError("prefixes mismatch: %s must start with %s"
1019
% (prefix,self._search_prefix))
1028
% (prefix, self._search_prefix))
1020
1029
if len(prefix) != len(self._search_prefix) + 1:
1021
1030
raise AssertionError("prefix wrong length: len(%s) is not %d" %
1022
(prefix, len(self._search_prefix) + 1))
1031
(prefix, len(self._search_prefix) + 1))
1023
1032
self._len += len(node)
1024
1033
if not len(self._items):
1025
1034
self._node_width = len(prefix)
1026
1035
if self._node_width != len(self._search_prefix) + 1:
1027
1036
raise AssertionError("node width mismatch: %d is not %d" %
1028
(self._node_width, len(self._search_prefix) + 1))
1037
(self._node_width, len(self._search_prefix) + 1))
1029
1038
self._items[prefix] = node
1030
1039
self._key = None
1032
1041
def _current_size(self):
1033
1042
"""Answer the current serialised size of this node."""
1034
return (self._raw_size + len(str(self._len)) + len(str(self._key_width)) +
1035
len(str(self._maximum_size)))
1043
return (self._raw_size + len(str(self._len)) + len(str(self._key_width))
1044
+ len(str(self._maximum_size)))
1038
1047
def deserialise(klass, bytes, key, search_key_func=None):
1072
1081
# yielding all nodes, yield whatever we have, and queue up a read
1073
1082
# for whatever we are missing
1074
1083
shortcut = True
1075
for prefix, node in self._items.iteritems():
1084
for prefix, node in viewitems(self._items):
1076
1085
if node.__class__ is StaticTuple:
1077
1086
keys[node] = (prefix, None)
1123
1132
for key in key_filter:
1124
1133
search_prefix = self._search_prefix_filter(key)
1125
1134
length_filter = length_filters.setdefault(
1126
len(search_prefix), set())
1135
len(search_prefix), set())
1127
1136
length_filter.add(search_prefix)
1128
1137
prefix_to_keys.setdefault(search_prefix, []).append(key)
1130
if (self._node_width in length_filters
1131
and len(length_filters) == 1):
1139
if (self._node_width in length_filters and
1140
len(length_filters) == 1):
1132
1141
# all of the search prefixes match exactly _node_width. This
1133
1142
# means that everything is an exact match, and we can do a
1134
1143
# lookup into self._items, rather than iterating over the items
1149
1158
# The slow way. We walk every item in self._items, and check to
1150
1159
# see if there are any matches
1151
length_filters = length_filters.items()
1152
for prefix, node in self._items.iteritems():
1160
length_filters_itemview = viewitems(length_filters)
1161
for prefix, node in viewitems(self._items):
1153
1162
node_key_filter = []
1154
for length, length_filter in length_filters:
1163
for length, length_filter in length_filters_itemview:
1155
1164
sub_prefix = prefix[:length]
1156
1165
if sub_prefix in length_filter:
1157
1166
node_key_filter.extend(prefix_to_keys[sub_prefix])
1158
if node_key_filter: # this key matched something, yield it
1167
if node_key_filter: # this key matched something, yield it
1159
1168
if node.__class__ is StaticTuple:
1160
1169
keys[node] = (prefix, node_key_filter)
1172
1181
node = _deserialise(bytes, key,
1173
search_key_func=self._search_key_func)
1182
search_key_func=self._search_key_func)
1174
1183
prefix, node_key_filter = keys[key]
1175
1184
self._items[prefix] = node
1176
1185
found_keys.add(key)
1192
1201
for record in stream:
1193
1202
bytes = record.get_bytes_as('fulltext')
1194
1203
node = _deserialise(bytes, record.key,
1195
search_key_func=self._search_key_func)
1204
search_key_func=self._search_key_func)
1196
1205
prefix, node_key_filter = keys[record.key]
1197
1206
node_and_filters.append((node, node_key_filter))
1198
1207
self._items[prefix] = node
1199
_get_cache().add(record.key, bytes)
1208
_get_cache()[record.key] = bytes
1200
1209
for info in node_and_filters:
1207
1216
search_key = self._search_key(key)
1208
1217
if self._node_width != len(self._search_prefix) + 1:
1209
1218
raise AssertionError("node width mismatch: %d is not %d" %
1210
(self._node_width, len(self._search_prefix) + 1))
1219
(self._node_width, len(self._search_prefix) + 1))
1211
1220
if not search_key.startswith(self._search_prefix):
1212
1221
# This key doesn't fit in this index, so we need to split at the
1213
1222
# point where it would fit, insert self into that internal node,
1215
1224
new_prefix = self.common_prefix(self._search_prefix,
1217
1226
new_parent = InternalNode(new_prefix,
1218
search_key_func=self._search_key_func)
1227
search_key_func=self._search_key_func)
1219
1228
new_parent.set_maximum_size(self._maximum_size)
1220
1229
new_parent._key_width = self._key_width
1221
new_parent.add_node(self._search_prefix[:len(new_prefix)+1],
1230
new_parent.add_node(self._search_prefix[:len(new_prefix) + 1],
1223
1232
return new_parent.map(store, key, value)
1224
children = [node for node, _
1225
in self._iter_nodes(store, key_filter=[key])]
1233
children = [node for node, _ in self._iter_nodes(
1234
store, key_filter=[key])]
1227
1236
child = children[0]
1229
1238
# new child needed:
1230
1239
child = self._new_child(search_key, LeafNode)
1231
1240
old_len = len(child)
1232
if type(child) is LeafNode:
1241
if isinstance(child, LeafNode):
1233
1242
old_size = child._current_size()
1235
1244
old_size = None
1257
1266
# amount is over a configurable limit.
1258
1267
new_size = child._current_size()
1259
1268
shrinkage = old_size - new_size
1260
if (shrinkage > 0 and new_size < _INTERESTING_NEW_SIZE
1261
or shrinkage > _INTERESTING_SHRINKAGE_LIMIT):
1269
if (shrinkage > 0 and new_size < _INTERESTING_NEW_SIZE or
1270
shrinkage > _INTERESTING_SHRINKAGE_LIMIT):
1263
1272
"checking remap as size shrunk by %d to be %d",
1264
1273
shrinkage, new_size)
1265
1274
new_node = self._check_remap(store)
1266
1275
if new_node._search_prefix is None:
1267
1276
raise AssertionError("_search_prefix should not be None")
1268
return new_node._search_prefix, [('', new_node)]
1277
return new_node._search_prefix, [(b'', new_node)]
1269
1278
# child has overflown - create a new intermediate node.
1270
1279
# XXX: This is where we might want to try and expand our depth
1271
1280
# to refer to more bytes of every child (which would give us
1276
1285
child.add_node(split, node)
1277
1286
self._len = self._len - old_len + len(child)
1278
1287
self._key = None
1279
return self._search_prefix, [("", self)]
1288
return self._search_prefix, [(b"", self)]
1281
1290
def _new_child(self, search_key, klass):
1282
1291
"""Create a new child node of type klass."""
1293
1302
:param store: A VersionedFiles honouring the CHK extensions.
1294
1303
:return: An iterable of the keys inserted by this operation.
1296
for node in self._items.itervalues():
1297
if type(node) is StaticTuple:
1305
for node in viewvalues(self._items):
1306
if isinstance(node, StaticTuple):
1298
1307
# Never deserialised.
1300
1309
if node._key is not None:
1303
1312
for key in node.serialise(store):
1305
lines = ["chknode:\n"]
1306
lines.append("%d\n" % self._maximum_size)
1307
lines.append("%d\n" % self._key_width)
1308
lines.append("%d\n" % self._len)
1314
lines = [b"chknode:\n"]
1315
lines.append(b"%d\n" % self._maximum_size)
1316
lines.append(b"%d\n" % self._key_width)
1317
lines.append(b"%d\n" % self._len)
1309
1318
if self._search_prefix is None:
1310
1319
raise AssertionError("_search_prefix should not be None")
1311
lines.append('%s\n' % (self._search_prefix,))
1320
lines.append(b'%s\n' % (self._search_prefix,))
1312
1321
prefix_len = len(self._search_prefix)
1313
for prefix, node in sorted(self._items.items()):
1314
if type(node) is StaticTuple:
1322
for prefix, node in sorted(viewitems(self._items)):
1323
if isinstance(node, StaticTuple):
1317
1326
key = node._key[0]
1318
serialised = "%s\x00%s\n" % (prefix, key)
1327
serialised = b"%s\x00%s\n" % (prefix, key)
1319
1328
if not serialised.startswith(self._search_prefix):
1320
1329
raise AssertionError("prefixes mismatch: %s must start with %s"
1321
% (serialised, self._search_prefix))
1330
% (serialised, self._search_prefix))
1322
1331
lines.append(serialised[prefix_len:])
1323
1332
sha1, _, _ = store.add_lines((None,), (), lines)
1324
self._key = StaticTuple("sha1:" + sha1,).intern()
1325
_get_cache().add(self._key, ''.join(lines))
1333
self._key = StaticTuple(b"sha1:" + sha1,).intern()
1334
_get_cache()[self._key] = b''.join(lines)
1326
1335
yield self._key
1328
1337
def _search_key(self, key):
1329
1338
"""Return the serialised key for key in this node."""
1330
1339
# search keys are fixed width. All will be self._node_width wide, so we
1331
1340
# pad as necessary.
1332
return (self._search_key_func(key) + '\x00'*self._node_width)[:self._node_width]
1341
return (self._search_key_func(key) + b'\x00' * self._node_width)[:self._node_width]
1334
1343
def _search_prefix_filter(self, key):
1335
1344
"""Serialise key for use as a prefix filter in iteritems."""
1343
1352
prefix for reaching node.
1345
1354
if offset >= self._node_width:
1346
for node in self._items.values():
1355
for node in valueview(self._items):
1347
1356
for result in node._split(offset):
1350
for key, node in self._items.items():
1353
1359
def refs(self):
1354
1360
"""Return the references to other CHK's held by this node."""
1355
1361
if self._key is None:
1356
1362
raise AssertionError("unserialised nodes have no refs.")
1358
for value in self._items.itervalues():
1359
if type(value) is StaticTuple:
1364
for value in viewvalues(self._items):
1365
if isinstance(value, StaticTuple):
1360
1366
refs.append(value)
1362
1368
refs.append(value.key())
1372
1378
return self._search_prefix
1374
1380
def unmap(self, store, key, check_remap=True):
1375
"""Remove key from this node and it's children."""
1381
"""Remove key from this node and its children."""
1376
1382
if not len(self._items):
1377
1383
raise AssertionError("can't unmap in an empty InternalNode.")
1378
1384
children = [node for node, _
1379
in self._iter_nodes(store, key_filter=[key])]
1385
in self._iter_nodes(store, key_filter=[key])]
1381
1387
child = children[0]
1394
1400
self._items[search_key] = unmapped
1395
1401
if len(self._items) == 1:
1396
1402
# this node is no longer needed:
1397
return self._items.values()[0]
1398
if type(unmapped) is InternalNode:
1403
return list(viewvalues(self._items))[0]
1404
if isinstance(unmapped, InternalNode):
1400
1406
if check_remap:
1401
1407
return self._check_remap(store)
1441
1447
# c) With 255-way fan out, we don't want to read all 255 and destroy
1442
1448
# the page cache, just to determine that we really don't need it.
1443
1449
for node, _ in self._iter_nodes(store, batch_size=16):
1444
if type(node) is InternalNode:
1450
if isinstance(node, InternalNode):
1445
1451
# Without looking at any leaf nodes, we are sure
1447
for key, value in node._items.iteritems():
1453
for key, value in viewitems(node._items):
1448
1454
if new_leaf._map_no_split(key, value):
1450
1456
trace.mutter("remap generated a new LeafNode")
1451
1457
return new_leaf
1454
def _deserialise(bytes, key, search_key_func):
1460
def _deserialise(data, key, search_key_func):
1455
1461
"""Helper for repositorydetails - convert bytes to a node."""
1456
if bytes.startswith("chkleaf:\n"):
1457
node = LeafNode.deserialise(bytes, key, search_key_func=search_key_func)
1458
elif bytes.startswith("chknode:\n"):
1459
node = InternalNode.deserialise(bytes, key,
1460
search_key_func=search_key_func)
1462
if data.startswith(b"chkleaf:\n"):
1463
node = LeafNode.deserialise(data, key, search_key_func=search_key_func)
1464
elif data.startswith(b"chknode:\n"):
1465
node = InternalNode.deserialise(data, key,
1466
search_key_func=search_key_func)
1462
1468
raise AssertionError("Unknown node type.")
1526
1532
bytes = record.get_bytes_as('fulltext')
1527
1533
node = _deserialise(bytes, record.key,
1528
1534
search_key_func=self._search_key_func)
1529
if type(node) is InternalNode:
1535
if isinstance(node, InternalNode):
1530
1536
# Note we don't have to do node.refs() because we know that
1531
1537
# there are no children that have been pushed into this node
1532
1538
# Note: Using as_st() here seemed to save 1.2MB, which would
1533
1539
# indicate that we keep 100k prefix_refs around while
1534
1540
# processing. They *should* be shorter lived than that...
1535
1541
# It does cost us ~10s of processing time
1536
#prefix_refs = [as_st(item) for item in node._items.iteritems()]
1537
prefix_refs = node._items.items()
1542
prefix_refs = list(viewitems(node._items))
1540
1545
prefix_refs = []
1541
1546
# Note: We don't use a StaticTuple here. Profiling showed a
1542
1547
# minor memory improvement (0.8MB out of 335MB peak 0.2%)
1543
1548
# But a significant slowdown (15s / 145s, or 10%)
1544
items = node._items.items()
1549
items = list(viewitems(node._items))
1545
1550
yield record, node, prefix_refs, items
1547
1552
def _read_old_roots(self):
1551
1556
self._read_nodes_from_store(self._old_root_keys):
1552
1557
# Uninteresting node
1553
1558
prefix_refs = [p_r for p_r in prefix_refs
1554
if p_r[1] not in all_old_chks]
1559
if p_r[1] not in all_old_chks]
1555
1560
new_refs = [p_r[1] for p_r in prefix_refs]
1556
1561
all_old_chks.update(new_refs)
1557
1562
# TODO: This might be a good time to turn items into StaticTuple
1571
1576
# handled the interesting ones
1572
1577
for prefix, ref in old_chks_to_enqueue:
1573
1578
not_interesting = True
1574
for i in xrange(len(prefix), 0, -1):
1579
for i in range(len(prefix), 0, -1):
1575
1580
if prefix[:i] in new_prefixes:
1576
1581
not_interesting = False
1607
1612
# At this level, we now know all the uninteresting references
1608
1613
# So we filter and queue up whatever is remaining
1609
1614
prefix_refs = [p_r for p_r in prefix_refs
1610
if p_r[1] not in self._all_old_chks
1611
and p_r[1] not in processed_new_refs]
1615
if p_r[1] not in self._all_old_chks and
1616
p_r[1] not in processed_new_refs]
1612
1617
refs = [p_r[1] for p_r in prefix_refs]
1613
1618
new_prefixes.update([p_r[0] for p_r in prefix_refs])
1614
1619
self._new_queue.extend(refs)
1620
1625
# self._new_item_queue will hold the contents of multiple
1621
1626
# records for an extended lifetime
1622
1627
new_items = [item for item in items
1623
if item not in self._all_old_items]
1628
if item not in self._all_old_items]
1624
1629
self._new_item_queue.extend(new_items)
1625
1630
new_prefixes.update([self._search_key_func(item[0])
1626
1631
for item in new_items])
1631
1636
# 'ab', then we also need to include 'a'.) So expand the
1632
1637
# new_prefixes to include all shorter prefixes
1633
1638
for prefix in list(new_prefixes):
1634
new_prefixes.update([prefix[:i] for i in xrange(1, len(prefix))])
1639
new_prefixes.update([prefix[:i] for i in range(1, len(prefix))])
1635
1640
self._enqueue_old(new_prefixes, old_chks_to_enqueue)
1637
1642
def _flush_new_queue(self):
1688
1693
for record, _, prefix_refs, items in self._read_nodes_from_store(refs):
1689
1694
# TODO: Use StaticTuple here?
1690
1695
self._all_old_items.update(items)
1691
refs = [r for _,r in prefix_refs if r not in all_old_chks]
1696
refs = [r for _, r in prefix_refs if r not in all_old_chks]
1692
1697
self._old_queue.extend(refs)
1693
1698
all_old_chks.update(refs)
1729
from bzrlib._chk_map_pyx import (
1734
from ._chk_map_pyx import (
1730
1736
_search_key_16,
1731
1737
_search_key_255,
1732
1738
_deserialise_leaf_node,
1733
1739
_deserialise_internal_node,
1735
except ImportError, e:
1741
except ImportError as e:
1736
1742
osutils.failed_to_load_extension(e)
1737
from bzrlib._chk_map_py import (
1743
from ._chk_map_py import (
1738
1745
_search_key_16,
1739
1746
_search_key_255,
1740
1747
_deserialise_leaf_node,
1741
1748
_deserialise_internal_node,
1743
search_key_registry.register('hash-16-way', _search_key_16)
1744
search_key_registry.register('hash-255-way', _search_key_255)
1750
search_key_registry.register(b'hash-16-way', _search_key_16)
1751
search_key_registry.register(b'hash-255-way', _search_key_255)
1747
1754
def _check_key(key):
1750
1757
This generally shouldn't be used in production code, but it can be helpful
1751
1758
to debug problems.
1753
if type(key) is not StaticTuple:
1760
if not isinstance(key, StaticTuple):
1754
1761
raise TypeError('key %r is not StaticTuple but %s' % (key, type(key)))
1755
1762
if len(key) != 1:
1756
raise ValueError('key %r should have length 1, not %d' % (key, len(key),))
1757
if type(key[0]) is not str:
1763
raise ValueError('key %r should have length 1, not %d' %
1765
if not isinstance(key[0], str):
1758
1766
raise TypeError('key %r should hold a str, not %r'
1759
1767
% (key, type(key[0])))
1760
1768
if not key[0].startswith('sha1:'):
1761
1769
raise ValueError('key %r should point to a sha1:' % (key,))
added
removed
breezy/bzr/chk_map.py
