112
88
if not element or _whitespace_re.search(element) is not None:
113
89
raise errors.BadIndexKey(element)
115
def _external_references(self):
116
"""Return references that are not present in this index.
120
# TODO: JAM 2008-11-21 This makes an assumption about how the reference
121
# lists are used. It is currently correct for pack-0.92 through
122
# 1.9, which use the node references (3rd column) second
123
# reference list as the compression parent. Perhaps this should
124
# be moved into something higher up the stack, since it
125
# makes assumptions about how the index is used.
126
if self.reference_lists > 1:
127
for node in self.iter_all_entries():
129
refs.update(node[3][1])
132
# If reference_lists == 0 there can be no external references, and
133
# if reference_lists == 1, then there isn't a place to store the
137
def _get_nodes_by_key(self):
138
if self._nodes_by_key is None:
140
if self.reference_lists:
141
for key, (absent, references, value) in self._nodes.iteritems():
144
key_dict = nodes_by_key
145
for subkey in key[:-1]:
146
key_dict = key_dict.setdefault(subkey, {})
147
key_dict[key[-1]] = key, value, references
149
for key, (absent, references, value) in self._nodes.iteritems():
152
key_dict = nodes_by_key
153
for subkey in key[:-1]:
154
key_dict = key_dict.setdefault(subkey, {})
155
key_dict[key[-1]] = key, value
156
self._nodes_by_key = nodes_by_key
157
return self._nodes_by_key
159
def _update_nodes_by_key(self, key, value, node_refs):
160
"""Update the _nodes_by_key dict with a new key.
162
For a key of (foo, bar, baz) create
163
_nodes_by_key[foo][bar][baz] = key_value
165
if self._nodes_by_key is None:
167
key_dict = self._nodes_by_key
168
if self.reference_lists:
169
key_value = StaticTuple(key, value, node_refs)
171
key_value = StaticTuple(key, value)
172
for subkey in key[:-1]:
173
key_dict = key_dict.setdefault(subkey, {})
174
key_dict[key[-1]] = key_value
176
def _check_key_ref_value(self, key, references, value):
177
"""Check that 'key' and 'references' are all valid.
179
:param key: A key tuple. Must conform to the key interface (be a tuple,
180
be of the right length, not have any whitespace or nulls in any key
182
:param references: An iterable of reference lists. Something like
183
[[(ref, key)], [(ref, key), (other, key)]]
184
:param value: The value associate with this key. Must not contain
185
newlines or null characters.
186
:return: (node_refs, absent_references)
187
node_refs basically a packed form of 'references' where all
189
absent_references reference keys that are not in self._nodes.
190
This may contain duplicates if the same key is
191
referenced in multiple lists.
193
as_st = StaticTuple.from_sequence
91
def add_node(self, key, value, references=()):
92
"""Add a node to the index.
94
:param key: The key. keys are non-empty tuples containing
95
as many whitespace-free utf8 bytestrings as the key length
96
defined for this index.
97
:param references: An iterable of iterables of keys. Each is a
98
reference to another key.
99
:param value: The value to associate with the key. It may be any
100
bytes as long as it does not contain \0 or \n.
194
102
self._check_key(key)
195
103
if _newline_null_re.search(value) is not None:
196
104
raise errors.BadIndexValue(value)
197
105
if len(references) != self.reference_lists:
198
106
raise errors.BadIndexValue(references)
200
absent_references = []
201
108
for reference_list in references:
202
109
for reference in reference_list:
203
# If reference *is* in self._nodes, then we know it has already
110
self._check_key(reference)
205
111
if reference not in self._nodes:
206
self._check_key(reference)
207
absent_references.append(reference)
208
reference_list = as_st([as_st(ref).intern()
209
for ref in reference_list])
210
node_refs.append(reference_list)
211
return as_st(node_refs), absent_references
213
def add_node(self, key, value, references=()):
214
"""Add a node to the index.
216
:param key: The key. keys are non-empty tuples containing
217
as many whitespace-free utf8 bytestrings as the key length
218
defined for this index.
219
:param references: An iterable of iterables of keys. Each is a
220
reference to another key.
221
:param value: The value to associate with the key. It may be any
222
bytes as long as it does not contain \0 or \n.
225
absent_references) = self._check_key_ref_value(key, references, value)
226
if key in self._nodes and self._nodes[key][0] != 'a':
112
self._nodes[reference] = ('a', (), '')
113
node_refs.append(tuple(reference_list))
114
if key in self._nodes and self._nodes[key][0] == '':
227
115
raise errors.BadIndexDuplicateKey(key, self)
228
for reference in absent_references:
229
# There may be duplicates, but I don't think it is worth worrying
231
self._nodes[reference] = ('a', (), '')
232
self._absent_keys.update(absent_references)
233
self._absent_keys.discard(key)
234
self._nodes[key] = ('', node_refs, value)
235
if self._nodes_by_key is not None and self._key_length > 1:
236
self._update_nodes_by_key(key, value, node_refs)
238
def clear_cache(self):
239
"""See GraphIndex.clear_cache()
241
This is a no-op, but we need the api to conform to a generic 'Index'
116
self._nodes[key] = ('', tuple(node_refs), value)
118
if self._key_length > 1:
119
key_dict = self._nodes_by_key
120
if self.reference_lists:
121
key_value = key, value, tuple(node_refs)
123
key_value = key, value
124
# possibly should do this on-demand, but it seems likely it is
126
# For a key of (foo, bar, baz) create
127
# _nodes_by_key[foo][bar][baz] = key_value
128
for subkey in key[:-1]:
129
key_dict = key_dict.setdefault(subkey, {})
130
key_dict[key[-1]] = key_value
245
132
def finish(self):
246
133
lines = [_SIGNATURE]
247
134
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
248
135
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
249
key_count = len(self._nodes) - len(self._absent_keys)
250
lines.append(_OPTION_LEN + str(key_count) + '\n')
136
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
251
137
prefix_length = sum(len(x) for x in lines)
252
138
# references are byte offsets. To avoid having to do nasty
253
# polynomial work to resolve offsets (references to later in the
139
# polynomial work to resolve offsets (references to later in the
254
140
# file cannot be determined until all the inbetween references have
255
141
# been calculated too) we pad the offsets with 0's to make them be
256
142
# of consistent length. Using binary offsets would break the trivial
327
213
raise errors.BzrError('Failed index creation. Internal error:'
328
214
' mismatched output length and expected length: %d %d' %
329
215
(len(result.getvalue()), expected_bytes))
332
def set_optimize(self, for_size=None, combine_backing_indices=None):
333
"""Change how the builder tries to optimize the result.
335
:param for_size: Tell the builder to try and make the index as small as
337
:param combine_backing_indices: If the builder spills to disk to save
338
memory, should the on-disk indices be combined. Set to True if you
339
are going to be probing the index, but to False if you are not. (If
340
you are not querying, then the time spent combining is wasted.)
343
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
345
if for_size is not None:
346
self._optimize_for_size = for_size
347
if combine_backing_indices is not None:
348
self._combine_backing_indices = combine_backing_indices
350
def find_ancestry(self, keys, ref_list_num):
351
"""See CombinedGraphIndex.find_ancestry()"""
357
for _, key, value, ref_lists in self.iter_entries(pending):
358
parent_keys = ref_lists[ref_list_num]
359
parent_map[key] = parent_keys
360
next_pending.update([p for p in parent_keys if p not in
362
missing_keys.update(pending.difference(parent_map))
363
pending = next_pending
364
return parent_map, missing_keys
216
return StringIO(''.join(lines))
367
219
class GraphIndex(object):
368
220
"""An index for data with embedded graphs.
370
222
The index maps keys to a list of key reference lists, and a value.
371
223
Each node has the same number of key reference lists. Each key reference
372
224
list can be empty or an arbitrary length. The value is an opaque NULL
373
terminated string without any newlines. The storage of the index is
225
terminated string without any newlines. The storage of the index is
374
226
hidden in the interface: keys and key references are always tuples of
375
227
bytestrings, never the internal representation (e.g. dictionary offsets).
475
309
node_value = value
476
310
self._nodes[key] = node_value
311
if self._key_length > 1:
312
subkey = list(reversed(key[:-1]))
313
key_dict = self._nodes_by_key
314
if self.node_ref_lists:
315
key_value = key, node_value[0], node_value[1]
317
key_value = key, node_value
318
# possibly should do this on-demand, but it seems likely it is
320
# For a key of (foo, bar, baz) create
321
# _nodes_by_key[foo][bar][baz] = key_value
322
for subkey in key[:-1]:
323
key_dict = key_dict.setdefault(subkey, {})
324
key_dict[key[-1]] = key_value
477
325
# cache the keys for quick set intersections
326
self._keys = set(self._nodes)
478
327
if trailers != 1:
479
328
# there must be one line - the empty trailer line.
480
329
raise errors.BadIndexData(self)
482
def clear_cache(self):
483
"""Clear out any cached/memoized values.
485
This can be called at any time, but generally it is used when we have
486
extracted some information, but don't expect to be requesting any more
490
def external_references(self, ref_list_num):
491
"""Return references that are not present in this index.
494
if ref_list_num + 1 > self.node_ref_lists:
495
raise ValueError('No ref list %d, index has %d ref lists'
496
% (ref_list_num, self.node_ref_lists))
499
for key, (value, ref_lists) in nodes.iteritems():
500
ref_list = ref_lists[ref_list_num]
501
refs.update([ref for ref in ref_list if ref not in nodes])
504
def _get_nodes_by_key(self):
505
if self._nodes_by_key is None:
507
if self.node_ref_lists:
508
for key, (value, references) in self._nodes.iteritems():
509
key_dict = nodes_by_key
510
for subkey in key[:-1]:
511
key_dict = key_dict.setdefault(subkey, {})
512
key_dict[key[-1]] = key, value, references
514
for key, value in self._nodes.iteritems():
515
key_dict = nodes_by_key
516
for subkey in key[:-1]:
517
key_dict = key_dict.setdefault(subkey, {})
518
key_dict[key[-1]] = key, value
519
self._nodes_by_key = nodes_by_key
520
return self._nodes_by_key
522
331
def iter_all_entries(self):
523
332
"""Iterate over all keys within the index.
830
612
readv_ranges.append((location, length))
831
613
# read the header if needed
832
614
if self._bisect_nodes is None:
833
readv_ranges.append(_HEADER_READV)
615
readv_ranges.append((0, 200))
834
616
self._read_and_parse(readv_ranges)
836
if self._nodes is not None:
837
# _read_and_parse triggered a _buffer_all because we requested the
839
for location, key in location_keys:
840
if key not in self._nodes: # not present
841
result.append(((location, key), False))
842
elif self.node_ref_lists:
843
value, refs = self._nodes[key]
844
result.append(((location, key),
845
(self, key, value, refs)))
847
result.append(((location, key),
848
(self, key, self._nodes[key])))
850
617
# generate results:
851
618
# - figure out <, >, missing, present
852
619
# - result present references so we can return them.
853
621
# keys that we cannot answer until we resolve references
854
622
pending_references = []
855
623
pending_locations = set()
1185
942
self._parsed_key_map.insert(index + 1, new_key)
1187
944
def _read_and_parse(self, readv_ranges):
1188
"""Read the ranges and parse the resulting data.
945
"""Read the the ranges and parse the resulting data.
1190
947
:param readv_ranges: A prepared readv range list.
1192
if not readv_ranges:
1194
if self._nodes is None and self._bytes_read * 2 >= self._size:
1195
# We've already read more than 50% of the file and we are about to
1196
# request more data, just _buffer_all() and be done
1200
base_offset = self._base_offset
1201
if base_offset != 0:
1202
# Rewrite the ranges for the offset
1203
readv_ranges = [(start+base_offset, size)
1204
for start, size in readv_ranges]
1205
readv_data = self._transport.readv(self._name, readv_ranges, True,
1206
self._size + self._base_offset)
1208
for offset, data in readv_data:
1209
offset -= base_offset
1210
self._bytes_read += len(data)
1212
# transport.readv() expanded to extra data which isn't part of
1214
data = data[-offset:]
1216
if offset == 0 and len(data) == self._size:
1217
# We read the whole range, most likely because the
1218
# Transport upcast our readv ranges into one long request
1219
# for enough total data to grab the whole index.
1220
self._buffer_all(StringIO(data))
1222
if self._bisect_nodes is None:
1223
# this must be the start
1224
if not (offset == 0):
1225
raise AssertionError()
1226
offset, data = self._parse_header_from_bytes(data)
1227
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1228
self._parse_region(offset, data)
950
readv_data = self._transport.readv(self._name, readv_ranges, True,
953
for offset, data in readv_data:
954
if self._bisect_nodes is None:
955
# this must be the start
957
offset, data = self._parse_header_from_bytes(data)
958
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
959
self._parse_region(offset, data)
1230
961
def _signature(self):
1231
962
"""The file signature for this index type."""
1241
972
class CombinedGraphIndex(object):
1242
973
"""A GraphIndex made up from smaller GraphIndices.
1244
975
The backing indices must implement GraphIndex, and are presumed to be
1247
978
Queries against the combined index will be made against the first index,
1248
and then the second and so on. The order of indices can thus influence
979
and then the second and so on. The order of index's can thus influence
1249
980
performance significantly. For example, if one index is on local disk and a
1250
981
second on a remote server, the local disk index should be before the other
1251
982
in the index list.
1253
Also, queries tend to need results from the same indices as previous
1254
queries. So the indices will be reordered after every query to put the
1255
indices that had the result(s) of that query first (while otherwise
1256
preserving the relative ordering).
1259
def __init__(self, indices, reload_func=None):
985
def __init__(self, indices):
1260
986
"""Create a CombinedGraphIndex backed by indices.
1262
988
:param indices: An ordered list of indices to query for data.
1263
:param reload_func: A function to call if we find we are missing an
1264
index. Should have the form reload_func() => True/False to indicate
1265
if reloading actually changed anything.
1267
990
self._indices = indices
1268
self._reload_func = reload_func
1269
# Sibling indices are other CombinedGraphIndex that we should call
1270
# _move_to_front_by_name on when we auto-reorder ourself.
1271
self._sibling_indices = []
1272
# A list of names that corresponds to the instances in self._indices,
1273
# so _index_names[0] is always the name for _indices[0], etc. Sibling
1274
# indices must all use the same set of names as each other.
1275
self._index_names = [None] * len(self._indices)
1277
992
def __repr__(self):
1278
993
return "%s(%s)" % (
1279
994
self.__class__.__name__,
1280
995
', '.join(map(repr, self._indices)))
1282
def clear_cache(self):
1283
"""See GraphIndex.clear_cache()"""
1284
for index in self._indices:
1287
def get_parent_map(self, keys):
1288
"""See graph.StackedParentsProvider.get_parent_map"""
1289
search_keys = set(keys)
1290
if NULL_REVISION in search_keys:
1291
search_keys.discard(NULL_REVISION)
1292
found_parents = {NULL_REVISION:[]}
1295
for index, key, value, refs in self.iter_entries(search_keys):
1298
parents = (NULL_REVISION,)
1299
found_parents[key] = parents
1300
return found_parents
1302
has_key = _has_key_from_parent_map
1304
def insert_index(self, pos, index, name=None):
997
def insert_index(self, pos, index):
1305
998
"""Insert a new index in the list of indices to query.
1307
1000
:param pos: The position to insert the index.
1308
1001
:param index: The index to insert.
1309
:param name: a name for this index, e.g. a pack name. These names can
1310
be used to reflect index reorderings to related CombinedGraphIndex
1311
instances that use the same names. (see set_sibling_indices)
1313
1003
self._indices.insert(pos, index)
1314
self._index_names.insert(pos, name)
1316
1005
def iter_all_entries(self):
1317
1006
"""Iterate over all keys within the index
1391
1064
seen_keys = set()
1395
for index in self._indices:
1397
for node in index.iter_entries_prefix(keys):
1398
if node[1] in seen_keys:
1400
seen_keys.add(node[1])
1404
hit_indices.append(index)
1406
except errors.NoSuchFile:
1407
self._reload_or_raise()
1408
self._move_to_front(hit_indices)
1410
def _move_to_front(self, hit_indices):
1411
"""Rearrange self._indices so that hit_indices are first.
1413
Order is maintained as much as possible, e.g. the first unhit index
1414
will be the first index in _indices after the hit_indices, and the
1415
hit_indices will be present in exactly the order they are passed to
1418
_move_to_front propagates to all objects in self._sibling_indices by
1419
calling _move_to_front_by_name.
1421
if self._indices[:len(hit_indices)] == hit_indices:
1422
# The 'hit_indices' are already at the front (and in the same
1423
# order), no need to re-order
1425
hit_names = self._move_to_front_by_index(hit_indices)
1426
for sibling_idx in self._sibling_indices:
1427
sibling_idx._move_to_front_by_name(hit_names)
1429
def _move_to_front_by_index(self, hit_indices):
1430
"""Core logic for _move_to_front.
1432
Returns a list of names corresponding to the hit_indices param.
1434
indices_info = zip(self._index_names, self._indices)
1435
if 'index' in debug.debug_flags:
1436
mutter('CombinedGraphIndex reordering: currently %r, promoting %r',
1437
indices_info, hit_indices)
1440
new_hit_indices = []
1443
for offset, (name, idx) in enumerate(indices_info):
1444
if idx in hit_indices:
1445
hit_names.append(name)
1446
new_hit_indices.append(idx)
1447
if len(new_hit_indices) == len(hit_indices):
1448
# We've found all of the hit entries, everything else is
1450
unhit_names.extend(self._index_names[offset+1:])
1451
unhit_indices.extend(self._indices[offset+1:])
1454
unhit_names.append(name)
1455
unhit_indices.append(idx)
1457
self._indices = new_hit_indices + unhit_indices
1458
self._index_names = hit_names + unhit_names
1459
if 'index' in debug.debug_flags:
1460
mutter('CombinedGraphIndex reordered: %r', self._indices)
1463
def _move_to_front_by_name(self, hit_names):
1464
"""Moves indices named by 'hit_names' to front of the search order, as
1465
described in _move_to_front.
1467
# Translate names to index instances, and then call
1468
# _move_to_front_by_index.
1469
indices_info = zip(self._index_names, self._indices)
1471
for name, idx in indices_info:
1472
if name in hit_names:
1473
hit_indices.append(idx)
1474
self._move_to_front_by_index(hit_indices)
1476
def find_ancestry(self, keys, ref_list_num):
1477
"""Find the complete ancestry for the given set of keys.
1479
Note that this is a whole-ancestry request, so it should be used
1482
:param keys: An iterable of keys to look for
1483
:param ref_list_num: The reference list which references the parents
1485
:return: (parent_map, missing_keys)
1487
# XXX: make this call _move_to_front?
1488
missing_keys = set()
1490
keys_to_lookup = set(keys)
1492
while keys_to_lookup:
1493
# keys that *all* indexes claim are missing, stop searching them
1495
all_index_missing = None
1496
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1497
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1499
# len(missing_keys))
1500
for index_idx, index in enumerate(self._indices):
1501
# TODO: we should probably be doing something with
1502
# 'missing_keys' since we've already determined that
1503
# those revisions have not been found anywhere
1504
index_missing_keys = set()
1505
# Find all of the ancestry we can from this index
1506
# keep looking until the search_keys set is empty, which means
1507
# things we didn't find should be in index_missing_keys
1508
search_keys = keys_to_lookup
1510
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1511
# index_idx, len(search_keys),
1512
# len(parent_map), len(index_missing_keys))
1515
# TODO: ref_list_num should really be a parameter, since
1516
# CombinedGraphIndex does not know what the ref lists
1518
search_keys = index._find_ancestors(search_keys,
1519
ref_list_num, parent_map, index_missing_keys)
1520
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1521
# sub_generation, len(search_keys),
1522
# len(parent_map), len(index_missing_keys))
1523
# Now set whatever was missing to be searched in the next index
1524
keys_to_lookup = index_missing_keys
1525
if all_index_missing is None:
1526
all_index_missing = set(index_missing_keys)
1528
all_index_missing.intersection_update(index_missing_keys)
1529
if not keys_to_lookup:
1531
if all_index_missing is None:
1532
# There were no indexes, so all search keys are 'missing'
1533
missing_keys.update(keys_to_lookup)
1534
keys_to_lookup = None
1536
missing_keys.update(all_index_missing)
1537
keys_to_lookup.difference_update(all_index_missing)
1538
return parent_map, missing_keys
1065
for index in self._indices:
1066
for node in index.iter_entries_prefix(keys):
1067
if node[1] in seen_keys:
1069
seen_keys.add(node[1])
1540
1072
def key_count(self):
1541
1073
"""Return an estimate of the number of keys in this index.
1543
1075
For CombinedGraphIndex this is approximated by the sum of the keys of
1544
1076
the child indices. As child indices may have duplicate keys this can
1545
1077
have a maximum error of the number of child indices * largest number of
1546
1078
keys in any index.
1550
return sum((index.key_count() for index in self._indices), 0)
1551
except errors.NoSuchFile:
1552
self._reload_or_raise()
1554
missing_keys = _missing_keys_from_parent_map
1556
def _reload_or_raise(self):
1557
"""We just got a NoSuchFile exception.
1559
Try to reload the indices, if it fails, just raise the current
1562
if self._reload_func is None:
1564
exc_type, exc_value, exc_traceback = sys.exc_info()
1565
trace.mutter('Trying to reload after getting exception: %s',
1567
if not self._reload_func():
1568
# We tried to reload, but nothing changed, so we fail anyway
1569
trace.mutter('_reload_func indicated nothing has changed.'
1570
' Raising original exception.')
1571
raise exc_type, exc_value, exc_traceback
1573
def set_sibling_indices(self, sibling_combined_graph_indices):
1574
"""Set the CombinedGraphIndex objects to reorder after reordering self.
1576
self._sibling_indices = sibling_combined_graph_indices
1080
return sum((index.key_count() for index in self._indices), 0)
1578
1082
def validate(self):
1579
1083
"""Validate that everything in the index can be accessed."""
1582
for index in self._indices:
1585
except errors.NoSuchFile:
1586
self._reload_or_raise()
1084
for index in self._indices:
1589
1088
class InMemoryGraphIndex(GraphIndexBuilder):
added
removed
bzrlib/index.py
