112
94
if not element or _whitespace_re.search(element) is not None:
113
95
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
97
def add_node(self, key, value, references=()):
98
"""Add a node to the index.
100
:param key: The key. keys are non-empty tuples containing
101
as many whitespace-free utf8 bytestrings as the key length
102
defined for this index.
103
:param references: An iterable of iterables of keys. Each is a
104
reference to another key.
105
:param value: The value to associate with the key. It may be any
106
bytes as long as it does not contain \0 or \n.
194
108
self._check_key(key)
195
109
if _newline_null_re.search(value) is not None:
196
110
raise errors.BadIndexValue(value)
197
111
if len(references) != self.reference_lists:
198
112
raise errors.BadIndexValue(references)
200
absent_references = []
201
114
for reference_list in references:
202
115
for reference in reference_list:
203
# If reference *is* in self._nodes, then we know it has already
116
self._check_key(reference)
205
117
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':
118
self._nodes[reference] = ('a', (), '')
119
node_refs.append(tuple(reference_list))
120
if key in self._nodes and self._nodes[key][0] == '':
227
121
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'
122
self._nodes[key] = ('', tuple(node_refs), value)
124
if self._key_length > 1:
125
key_dict = self._nodes_by_key
126
if self.reference_lists:
127
key_value = key, value, tuple(node_refs)
129
key_value = key, value
130
# possibly should do this on-demand, but it seems likely it is
132
# For a key of (foo, bar, baz) create
133
# _nodes_by_key[foo][bar][baz] = key_value
134
for subkey in key[:-1]:
135
key_dict = key_dict.setdefault(subkey, {})
136
key_dict[key[-1]] = key_value
245
138
def finish(self):
246
139
lines = [_SIGNATURE]
247
140
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
248
141
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')
142
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
251
143
prefix_length = sum(len(x) for x in lines)
252
144
# references are byte offsets. To avoid having to do nasty
253
# polynomial work to resolve offsets (references to later in the
145
# polynomial work to resolve offsets (references to later in the
254
146
# file cannot be determined until all the inbetween references have
255
147
# been calculated too) we pad the offsets with 0's to make them be
256
148
# of consistent length. Using binary offsets would break the trivial
327
219
raise errors.BzrError('Failed index creation. Internal error:'
328
220
' mismatched output length and expected length: %d %d' %
329
221
(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
222
return StringIO(''.join(lines))
367
225
class GraphIndex(object):
368
226
"""An index for data with embedded graphs.
370
228
The index maps keys to a list of key reference lists, and a value.
371
229
Each node has the same number of key reference lists. Each key reference
372
230
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
231
terminated string without any newlines. The storage of the index is
374
232
hidden in the interface: keys and key references are always tuples of
375
233
bytestrings, never the internal representation (e.g. dictionary offsets).
476
315
node_value = value
477
316
self._nodes[key] = node_value
317
if self._key_length > 1:
318
subkey = list(reversed(key[:-1]))
319
key_dict = self._nodes_by_key
320
if self.node_ref_lists:
321
key_value = key, node_value[0], node_value[1]
323
key_value = key, node_value
324
# possibly should do this on-demand, but it seems likely it is
326
# For a key of (foo, bar, baz) create
327
# _nodes_by_key[foo][bar][baz] = key_value
328
for subkey in key[:-1]:
329
key_dict = key_dict.setdefault(subkey, {})
330
key_dict[key[-1]] = key_value
478
331
# cache the keys for quick set intersections
332
self._keys = set(self._nodes)
479
333
if trailers != 1:
480
334
# there must be one line - the empty trailer line.
481
335
raise errors.BadIndexData(self)
483
def clear_cache(self):
484
"""Clear out any cached/memoized values.
486
This can be called at any time, but generally it is used when we have
487
extracted some information, but don't expect to be requesting any more
491
def external_references(self, ref_list_num):
492
"""Return references that are not present in this index.
495
if ref_list_num + 1 > self.node_ref_lists:
496
raise ValueError('No ref list %d, index has %d ref lists'
497
% (ref_list_num, self.node_ref_lists))
500
for key, (value, ref_lists) in nodes.iteritems():
501
ref_list = ref_lists[ref_list_num]
502
refs.update([ref for ref in ref_list if ref not in nodes])
505
def _get_nodes_by_key(self):
506
if self._nodes_by_key is None:
508
if self.node_ref_lists:
509
for key, (value, references) in self._nodes.iteritems():
510
key_dict = nodes_by_key
511
for subkey in key[:-1]:
512
key_dict = key_dict.setdefault(subkey, {})
513
key_dict[key[-1]] = key, value, references
515
for key, value in self._nodes.iteritems():
516
key_dict = nodes_by_key
517
for subkey in key[:-1]:
518
key_dict = key_dict.setdefault(subkey, {})
519
key_dict[key[-1]] = key, value
520
self._nodes_by_key = nodes_by_key
521
return self._nodes_by_key
523
337
def iter_all_entries(self):
524
338
"""Iterate over all keys within the index.
1186
947
self._parsed_key_map.insert(index + 1, new_key)
1188
949
def _read_and_parse(self, readv_ranges):
1189
"""Read the ranges and parse the resulting data.
950
"""Read the the ranges and parse the resulting data.
1191
952
:param readv_ranges: A prepared readv range list.
1193
if not readv_ranges:
1195
if self._nodes is None and self._bytes_read * 2 >= self._size:
1196
# We've already read more than 50% of the file and we are about to
1197
# request more data, just _buffer_all() and be done
1201
base_offset = self._base_offset
1202
if base_offset != 0:
1203
# Rewrite the ranges for the offset
1204
readv_ranges = [(start+base_offset, size)
1205
for start, size in readv_ranges]
1206
readv_data = self._transport.readv(self._name, readv_ranges, True,
1207
self._size + self._base_offset)
1209
for offset, data in readv_data:
1210
offset -= base_offset
1211
self._bytes_read += len(data)
1213
# transport.readv() expanded to extra data which isn't part of
1215
data = data[-offset:]
1217
if offset == 0 and len(data) == self._size:
1218
# We read the whole range, most likely because the
1219
# Transport upcast our readv ranges into one long request
1220
# for enough total data to grab the whole index.
1221
self._buffer_all(StringIO(data))
1223
if self._bisect_nodes is None:
1224
# this must be the start
1225
if not (offset == 0):
1226
raise AssertionError()
1227
offset, data = self._parse_header_from_bytes(data)
1228
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1229
self._parse_region(offset, data)
955
readv_data = self._transport.readv(self._name, readv_ranges, True,
958
for offset, data in readv_data:
959
if self._bisect_nodes is None:
960
# this must be the start
962
offset, data = self._parse_header_from_bytes(data)
963
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
964
self._parse_region(offset, data)
1231
966
def _signature(self):
1232
967
"""The file signature for this index type."""
1242
977
class CombinedGraphIndex(object):
1243
978
"""A GraphIndex made up from smaller GraphIndices.
1245
980
The backing indices must implement GraphIndex, and are presumed to be
1248
983
Queries against the combined index will be made against the first index,
1249
and then the second and so on. The order of indices can thus influence
984
and then the second and so on. The order of index's can thus influence
1250
985
performance significantly. For example, if one index is on local disk and a
1251
986
second on a remote server, the local disk index should be before the other
1252
987
in the index list.
1254
Also, queries tend to need results from the same indices as previous
1255
queries. So the indices will be reordered after every query to put the
1256
indices that had the result(s) of that query first (while otherwise
1257
preserving the relative ordering).
1260
def __init__(self, indices, reload_func=None):
990
def __init__(self, indices):
1261
991
"""Create a CombinedGraphIndex backed by indices.
1263
993
:param indices: An ordered list of indices to query for data.
1264
:param reload_func: A function to call if we find we are missing an
1265
index. Should have the form reload_func() => True/False to indicate
1266
if reloading actually changed anything.
1268
995
self._indices = indices
1269
self._reload_func = reload_func
1270
# Sibling indices are other CombinedGraphIndex that we should call
1271
# _move_to_front_by_name on when we auto-reorder ourself.
1272
self._sibling_indices = []
1273
# A list of names that corresponds to the instances in self._indices,
1274
# so _index_names[0] is always the name for _indices[0], etc. Sibling
1275
# indices must all use the same set of names as each other.
1276
self._index_names = [None] * len(self._indices)
1278
997
def __repr__(self):
1279
998
return "%s(%s)" % (
1280
999
self.__class__.__name__,
1281
1000
', '.join(map(repr, self._indices)))
1283
def clear_cache(self):
1284
"""See GraphIndex.clear_cache()"""
1285
for index in self._indices:
1002
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1003
def get_parents(self, revision_ids):
1004
"""See graph._StackedParentsProvider.get_parents.
1006
This implementation thunks the graph.Graph.get_parents api across to
1009
:param revision_ids: An iterable of graph keys for this graph.
1010
:return: A list of parent details for each key in revision_ids.
1011
Each parent details will be one of:
1012
* None when the key was missing
1013
* (NULL_REVISION,) when the key has no parents.
1014
* (parent_key, parent_key...) otherwise.
1016
parent_map = self.get_parent_map(revision_ids)
1017
return [parent_map.get(r, None) for r in revision_ids]
1288
1019
def get_parent_map(self, keys):
1289
"""See graph.StackedParentsProvider.get_parent_map"""
1020
"""See graph._StackedParentsProvider.get_parent_map"""
1290
1021
search_keys = set(keys)
1291
1022
if NULL_REVISION in search_keys:
1292
1023
search_keys.discard(NULL_REVISION)
1392
1101
seen_keys = set()
1396
for index in self._indices:
1398
for node in index.iter_entries_prefix(keys):
1399
if node[1] in seen_keys:
1401
seen_keys.add(node[1])
1405
hit_indices.append(index)
1407
except errors.NoSuchFile:
1408
self._reload_or_raise()
1409
self._move_to_front(hit_indices)
1411
def _move_to_front(self, hit_indices):
1412
"""Rearrange self._indices so that hit_indices are first.
1414
Order is maintained as much as possible, e.g. the first unhit index
1415
will be the first index in _indices after the hit_indices, and the
1416
hit_indices will be present in exactly the order they are passed to
1419
_move_to_front propagates to all objects in self._sibling_indices by
1420
calling _move_to_front_by_name.
1422
if self._indices[:len(hit_indices)] == hit_indices:
1423
# The 'hit_indices' are already at the front (and in the same
1424
# order), no need to re-order
1426
hit_names = self._move_to_front_by_index(hit_indices)
1427
for sibling_idx in self._sibling_indices:
1428
sibling_idx._move_to_front_by_name(hit_names)
1430
def _move_to_front_by_index(self, hit_indices):
1431
"""Core logic for _move_to_front.
1433
Returns a list of names corresponding to the hit_indices param.
1435
indices_info = zip(self._index_names, self._indices)
1436
if 'index' in debug.debug_flags:
1437
mutter('CombinedGraphIndex reordering: currently %r, promoting %r',
1438
indices_info, hit_indices)
1441
new_hit_indices = []
1444
for offset, (name, idx) in enumerate(indices_info):
1445
if idx in hit_indices:
1446
hit_names.append(name)
1447
new_hit_indices.append(idx)
1448
if len(new_hit_indices) == len(hit_indices):
1449
# We've found all of the hit entries, everything else is
1451
unhit_names.extend(self._index_names[offset+1:])
1452
unhit_indices.extend(self._indices[offset+1:])
1455
unhit_names.append(name)
1456
unhit_indices.append(idx)
1458
self._indices = new_hit_indices + unhit_indices
1459
self._index_names = hit_names + unhit_names
1460
if 'index' in debug.debug_flags:
1461
mutter('CombinedGraphIndex reordered: %r', self._indices)
1464
def _move_to_front_by_name(self, hit_names):
1465
"""Moves indices named by 'hit_names' to front of the search order, as
1466
described in _move_to_front.
1468
# Translate names to index instances, and then call
1469
# _move_to_front_by_index.
1470
indices_info = zip(self._index_names, self._indices)
1472
for name, idx in indices_info:
1473
if name in hit_names:
1474
hit_indices.append(idx)
1475
self._move_to_front_by_index(hit_indices)
1477
def find_ancestry(self, keys, ref_list_num):
1478
"""Find the complete ancestry for the given set of keys.
1480
Note that this is a whole-ancestry request, so it should be used
1483
:param keys: An iterable of keys to look for
1484
:param ref_list_num: The reference list which references the parents
1486
:return: (parent_map, missing_keys)
1488
# XXX: make this call _move_to_front?
1489
missing_keys = set()
1491
keys_to_lookup = set(keys)
1493
while keys_to_lookup:
1494
# keys that *all* indexes claim are missing, stop searching them
1496
all_index_missing = None
1497
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1498
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1500
# len(missing_keys))
1501
for index_idx, index in enumerate(self._indices):
1502
# TODO: we should probably be doing something with
1503
# 'missing_keys' since we've already determined that
1504
# those revisions have not been found anywhere
1505
index_missing_keys = set()
1506
# Find all of the ancestry we can from this index
1507
# keep looking until the search_keys set is empty, which means
1508
# things we didn't find should be in index_missing_keys
1509
search_keys = keys_to_lookup
1511
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1512
# index_idx, len(search_keys),
1513
# len(parent_map), len(index_missing_keys))
1516
# TODO: ref_list_num should really be a parameter, since
1517
# CombinedGraphIndex does not know what the ref lists
1519
search_keys = index._find_ancestors(search_keys,
1520
ref_list_num, parent_map, index_missing_keys)
1521
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1522
# sub_generation, len(search_keys),
1523
# len(parent_map), len(index_missing_keys))
1524
# Now set whatever was missing to be searched in the next index
1525
keys_to_lookup = index_missing_keys
1526
if all_index_missing is None:
1527
all_index_missing = set(index_missing_keys)
1529
all_index_missing.intersection_update(index_missing_keys)
1530
if not keys_to_lookup:
1532
if all_index_missing is None:
1533
# There were no indexes, so all search keys are 'missing'
1534
missing_keys.update(keys_to_lookup)
1535
keys_to_lookup = None
1537
missing_keys.update(all_index_missing)
1538
keys_to_lookup.difference_update(all_index_missing)
1539
return parent_map, missing_keys
1102
for index in self._indices:
1103
for node in index.iter_entries_prefix(keys):
1104
if node[1] in seen_keys:
1106
seen_keys.add(node[1])
1541
1109
def key_count(self):
1542
1110
"""Return an estimate of the number of keys in this index.
1544
1112
For CombinedGraphIndex this is approximated by the sum of the keys of
1545
1113
the child indices. As child indices may have duplicate keys this can
1546
1114
have a maximum error of the number of child indices * largest number of
1547
1115
keys in any index.
1551
return sum((index.key_count() for index in self._indices), 0)
1552
except errors.NoSuchFile:
1553
self._reload_or_raise()
1555
missing_keys = _missing_keys_from_parent_map
1557
def _reload_or_raise(self):
1558
"""We just got a NoSuchFile exception.
1560
Try to reload the indices, if it fails, just raise the current
1563
if self._reload_func is None:
1565
exc_type, exc_value, exc_traceback = sys.exc_info()
1566
trace.mutter('Trying to reload after getting exception: %s',
1568
if not self._reload_func():
1569
# We tried to reload, but nothing changed, so we fail anyway
1570
trace.mutter('_reload_func indicated nothing has changed.'
1571
' Raising original exception.')
1572
raise exc_type, exc_value, exc_traceback
1574
def set_sibling_indices(self, sibling_combined_graph_indices):
1575
"""Set the CombinedGraphIndex objects to reorder after reordering self.
1577
self._sibling_indices = sibling_combined_graph_indices
1117
return sum((index.key_count() for index in self._indices), 0)
1579
1119
def validate(self):
1580
1120
"""Validate that everything in the index can be accessed."""
1583
for index in self._indices:
1586
except errors.NoSuchFile:
1587
self._reload_or_raise()
1121
for index in self._indices:
1590
1125
class InMemoryGraphIndex(GraphIndexBuilder):
added
removed
bzrlib/index.py
