94
112
if not element or _whitespace_re.search(element) is not None:
95
113
raise errors.BadIndexKey(element)
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.
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
108
194
self._check_key(key)
109
195
if _newline_null_re.search(value) is not None:
110
196
raise errors.BadIndexValue(value)
111
197
if len(references) != self.reference_lists:
112
198
raise errors.BadIndexValue(references)
200
absent_references = []
114
201
for reference_list in references:
115
202
for reference in reference_list:
116
self._check_key(reference)
203
# If reference *is* in self._nodes, then we know it has already
117
205
if reference not in self._nodes:
118
self._nodes[reference] = ('a', (), '')
119
node_refs.append(tuple(reference_list))
120
if key in self._nodes and self._nodes[key][0] == '':
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':
121
227
raise errors.BadIndexDuplicateKey(key, self)
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
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'
138
245
def finish(self):
139
246
lines = [_SIGNATURE]
140
247
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
141
248
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
142
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
249
key_count = len(self._nodes) - len(self._absent_keys)
250
lines.append(_OPTION_LEN + str(key_count) + '\n')
143
251
prefix_length = sum(len(x) for x in lines)
144
252
# references are byte offsets. To avoid having to do nasty
145
# polynomial work to resolve offsets (references to later in the
253
# polynomial work to resolve offsets (references to later in the
146
254
# file cannot be determined until all the inbetween references have
147
255
# been calculated too) we pad the offsets with 0's to make them be
148
256
# of consistent length. Using binary offsets would break the trivial
219
327
raise errors.BzrError('Failed index creation. Internal error:'
220
328
' mismatched output length and expected length: %d %d' %
221
329
(len(result.getvalue()), expected_bytes))
222
return StringIO(''.join(lines))
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
225
367
class GraphIndex(object):
226
368
"""An index for data with embedded graphs.
228
370
The index maps keys to a list of key reference lists, and a value.
229
371
Each node has the same number of key reference lists. Each key reference
230
372
list can be empty or an arbitrary length. The value is an opaque NULL
231
terminated string without any newlines. The storage of the index is
373
terminated string without any newlines. The storage of the index is
232
374
hidden in the interface: keys and key references are always tuples of
233
375
bytestrings, never the internal representation (e.g. dictionary offsets).
315
475
node_value = value
316
476
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
331
477
# cache the keys for quick set intersections
332
self._keys = set(self._nodes)
333
478
if trailers != 1:
334
479
# there must be one line - the empty trailer line.
335
480
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
337
522
def iter_all_entries(self):
338
523
"""Iterate over all keys within the index.
947
1185
self._parsed_key_map.insert(index + 1, new_key)
949
1187
def _read_and_parse(self, readv_ranges):
950
"""Read the the ranges and parse the resulting data.
1188
"""Read the ranges and parse the resulting data.
952
1190
:param readv_ranges: A prepared readv range list.
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)
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)
966
1230
def _signature(self):
967
1231
"""The file signature for this index type."""
977
1241
class CombinedGraphIndex(object):
978
1242
"""A GraphIndex made up from smaller GraphIndices.
980
1244
The backing indices must implement GraphIndex, and are presumed to be
983
1247
Queries against the combined index will be made against the first index,
984
and then the second and so on. The order of index's can thus influence
1248
and then the second and so on. The order of indices can thus influence
985
1249
performance significantly. For example, if one index is on local disk and a
986
1250
second on a remote server, the local disk index should be before the other
987
1251
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).
990
def __init__(self, indices):
1259
def __init__(self, indices, reload_func=None):
991
1260
"""Create a CombinedGraphIndex backed by indices.
993
1262
: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.
995
1267
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)
997
1277
def __repr__(self):
998
1278
return "%s(%s)" % (
999
1279
self.__class__.__name__,
1000
1280
', '.join(map(repr, 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]
1282
def clear_cache(self):
1283
"""See GraphIndex.clear_cache()"""
1284
for index in self._indices:
1019
1287
def get_parent_map(self, keys):
1020
"""See graph._StackedParentsProvider.get_parent_map"""
1288
"""See graph.StackedParentsProvider.get_parent_map"""
1021
1289
search_keys = set(keys)
1022
1290
if NULL_REVISION in search_keys:
1023
1291
search_keys.discard(NULL_REVISION)
1101
1391
seen_keys = set()
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])
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
1109
1540
def key_count(self):
1110
1541
"""Return an estimate of the number of keys in this index.
1112
1543
For CombinedGraphIndex this is approximated by the sum of the keys of
1113
1544
the child indices. As child indices may have duplicate keys this can
1114
1545
have a maximum error of the number of child indices * largest number of
1115
1546
keys in any index.
1117
return sum((index.key_count() for index in self._indices), 0)
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
1119
1578
def validate(self):
1120
1579
"""Validate that everything in the index can be accessed."""
1121
for index in self._indices:
1582
for index in self._indices:
1585
except errors.NoSuchFile:
1586
self._reload_or_raise()
1125
1589
class InMemoryGraphIndex(GraphIndexBuilder):