93
92
:param key_elements: The number of bytestrings in each key.
95
94
self.reference_lists = reference_lists
96
96
# A dict of {key: (absent, ref_lists, value)}
98
# Keys that are referenced but not actually present in this index
99
self._absent_keys = set()
100
98
self._nodes_by_key = None
101
99
self._key_length = key_elements
102
100
self._optimize_for_size = False
167
165
key_dict = self._nodes_by_key
168
166
if self.reference_lists:
169
key_value = StaticTuple(key, value, node_refs)
167
key_value = key, value, node_refs
171
key_value = StaticTuple(key, value)
169
key_value = key, value
172
170
for subkey in key[:-1]:
173
171
key_dict = key_dict.setdefault(subkey, {})
174
172
key_dict[key[-1]] = key_value
205
202
if reference not in self._nodes:
206
203
self._check_key(reference)
207
204
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
205
node_refs.append(tuple(reference_list))
206
return tuple(node_refs), absent_references
213
208
def add_node(self, key, value, references=()):
214
209
"""Add a node to the index.
229
224
# There may be duplicates, but I don't think it is worth worrying
231
226
self._nodes[reference] = ('a', (), '')
232
self._absent_keys.update(absent_references)
233
self._absent_keys.discard(key)
234
227
self._nodes[key] = ('', node_refs, value)
235
229
if self._nodes_by_key is not None and self._key_length > 1:
236
230
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'
245
232
def finish(self):
246
233
lines = [_SIGNATURE]
247
234
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
248
235
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')
236
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
251
237
prefix_length = sum(len(x) for x in lines)
252
238
# references are byte offsets. To avoid having to do nasty
253
239
# polynomial work to resolve offsets (references to later in the
382
368
suitable for production use. :XXX
385
def __init__(self, transport, name, size, unlimited_cache=False, offset=0):
371
def __init__(self, transport, name, size):
386
372
"""Open an index called name on transport.
388
374
:param transport: A bzrlib.transport.Transport.
447
430
mutter('Reading entire index %s', self._transport.abspath(self._name))
448
431
if stream is None:
449
432
stream = self._transport.get(self._name)
450
if self._base_offset != 0:
451
# This is wasteful, but it is better than dealing with
452
# adjusting all the offsets, etc.
453
stream = StringIO(stream.read()[self._base_offset:])
454
433
self._read_prefix(stream)
455
434
self._expected_elements = 3 + self._key_length
476
453
node_value = value
477
454
self._nodes[key] = node_value
478
455
# cache the keys for quick set intersections
456
self._keys = set(self._nodes)
479
457
if trailers != 1:
480
458
# there must be one line - the empty trailer line.
481
459
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
461
def external_references(self, ref_list_num):
492
462
"""Return references that are not present in this index.
496
466
raise ValueError('No ref list %d, index has %d ref lists'
497
467
% (ref_list_num, self.node_ref_lists))
500
for key, (value, ref_lists) in nodes.iteritems():
469
for key, (value, ref_lists) in self._nodes.iteritems():
501
470
ref_list = ref_lists[ref_list_num]
502
refs.update([ref for ref in ref_list if ref not in nodes])
471
refs.update(ref_list)
472
return refs - self._keys
505
474
def _get_nodes_by_key(self):
506
475
if self._nodes_by_key is None:
634
603
def _iter_entries_from_total_buffer(self, keys):
635
604
"""Iterate over keys when the entire index is parsed."""
636
# Note: See the note in BTreeBuilder.iter_entries for why we don't use
637
# .intersection() here
639
keys = [key for key in keys if key in nodes]
605
keys = keys.intersection(self._keys)
640
606
if self.node_ref_lists:
642
value, node_refs = nodes[key]
608
value, node_refs = self._nodes[key]
643
609
yield self, key, value, node_refs
646
yield self, key, nodes[key]
612
yield self, key, self._nodes[key]
648
614
def iter_entries(self, keys):
649
615
"""Iterate over keys within the index.
1198
1164
self._buffer_all()
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
1167
readv_data = self._transport.readv(self._name, readv_ranges, True,
1207
self._size + self._base_offset)
1209
1170
for offset, data in readv_data:
1210
offset -= base_offset
1211
1171
self._bytes_read += len(data)
1213
# transport.readv() expanded to extra data which isn't part of
1215
data = data[-offset:]
1217
1172
if offset == 0 and len(data) == self._size:
1218
1173
# We read the whole range, most likely because the
1219
1174
# Transport upcast our readv ranges into one long request
1248
1203
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
1204
and then the second and so on. The order of index's can thus influence
1250
1205
performance significantly. For example, if one index is on local disk and a
1251
1206
second on a remote server, the local disk index should be before the other
1252
1207
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
1210
def __init__(self, indices, reload_func=None):
1268
1218
self._indices = indices
1269
1219
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
1221
def __repr__(self):
1279
1222
return "%s(%s)" % (
1280
1223
self.__class__.__name__,
1281
1224
', '.join(map(repr, self._indices)))
1283
def clear_cache(self):
1284
"""See GraphIndex.clear_cache()"""
1285
for index in self._indices:
1288
1226
def get_parent_map(self, keys):
1289
1227
"""See graph.StackedParentsProvider.get_parent_map"""
1290
1228
search_keys = set(keys)
1303
1241
has_key = _has_key_from_parent_map
1305
def insert_index(self, pos, index, name=None):
1243
def insert_index(self, pos, index):
1306
1244
"""Insert a new index in the list of indices to query.
1308
1246
:param pos: The position to insert the index.
1309
1247
:param index: The index to insert.
1310
:param name: a name for this index, e.g. a pack name. These names can
1311
be used to reflect index reorderings to related CombinedGraphIndex
1312
instances that use the same names. (see set_sibling_indices)
1314
1249
self._indices.insert(pos, index)
1315
self._index_names.insert(pos, name)
1317
1251
def iter_all_entries(self):
1318
1252
"""Iterate over all keys within the index
1343
1277
value and are only reported once.
1345
1279
:param keys: An iterable providing the keys to be retrieved.
1346
:return: An iterable of (index, key, reference_lists, value). There is
1347
no defined order for the result iteration - it will be in the most
1280
:return: An iterable of (index, key, reference_lists, value). There is no
1281
defined order for the result iteration - it will be in the most
1348
1282
efficient order for the index.
1350
1284
keys = set(keys)
1354
1287
for index in self._indices:
1358
1290
for node in index.iter_entries(keys):
1359
1291
keys.remove(node[1])
1363
hit_indices.append(index)
1365
1294
except errors.NoSuchFile:
1366
1295
self._reload_or_raise()
1367
self._move_to_front(hit_indices)
1369
1297
def iter_entries_prefix(self, keys):
1370
1298
"""Iterate over keys within the index using prefix matching.
1392
1320
seen_keys = set()
1396
1323
for index in self._indices:
1398
1324
for node in index.iter_entries_prefix(keys):
1399
1325
if node[1] in seen_keys:
1401
1327
seen_keys.add(node[1])
1405
hit_indices.append(index)
1407
1330
except errors.NoSuchFile:
1408
1331
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
1333
def find_ancestry(self, keys, ref_list_num):
1478
1334
"""Find the complete ancestry for the given set of keys.
1634
1484
defined order for the result iteration - it will be in the most
1635
1485
efficient order for the index (keys iteration order in this case).
1637
# Note: See BTreeBuilder.iter_entries for an explanation of why we
1638
# aren't using set().intersection() here
1640
keys = [key for key in keys if key in nodes]
1641
1488
if self.reference_lists:
1489
for key in keys.intersection(self._keys):
1490
node = self._nodes[key]
1644
1491
if not node[0]:
1645
1492
yield self, key, node[2], node[1]
1494
for key in keys.intersection(self._keys):
1495
node = self._nodes[key]
1649
1496
if not node[0]:
1650
1497
yield self, key, node[2]
added
removed
bzrlib/index.py
