93
93
:param key_elements: The number of bytestrings in each key.
95
95
self.reference_lists = reference_lists
96
97
# 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
99
self._nodes_by_key = None
101
100
self._key_length = key_elements
102
101
self._optimize_for_size = False
167
166
key_dict = self._nodes_by_key
168
167
if self.reference_lists:
169
key_value = StaticTuple(key, value, node_refs)
168
key_value = key, value, node_refs
171
key_value = StaticTuple(key, value)
170
key_value = key, value
172
171
for subkey in key[:-1]:
173
172
key_dict = key_dict.setdefault(subkey, {})
174
173
key_dict[key[-1]] = key_value
205
203
if reference not in self._nodes:
206
204
self._check_key(reference)
207
205
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
207
node_refs.append(tuple(reference_list))
209
return tuple(node_refs), absent_references
213
211
def add_node(self, key, value, references=()):
214
212
"""Add a node to the index.
229
227
# There may be duplicates, but I don't think it is worth worrying
231
229
self._nodes[reference] = ('a', (), '')
232
self._absent_keys.update(absent_references)
233
self._absent_keys.discard(key)
234
230
self._nodes[key] = ('', node_refs, value)
235
232
if self._nodes_by_key is not None and self._key_length > 1:
236
233
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
235
def finish(self):
246
236
lines = [_SIGNATURE]
247
237
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
248
238
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')
239
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
251
240
prefix_length = sum(len(x) for x in lines)
252
241
# references are byte offsets. To avoid having to do nasty
253
242
# polynomial work to resolve offsets (references to later in the
382
371
suitable for production use. :XXX
385
def __init__(self, transport, name, size, unlimited_cache=False, offset=0):
374
def __init__(self, transport, name, size, unlimited_cache=False):
386
375
"""Open an index called name on transport.
388
377
:param transport: A bzrlib.transport.Transport.
447
433
mutter('Reading entire index %s', self._transport.abspath(self._name))
448
434
if stream is None:
449
435
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
436
self._read_prefix(stream)
455
437
self._expected_elements = 3 + self._key_length
475
456
node_value = value
476
457
self._nodes[key] = node_value
477
458
# cache the keys for quick set intersections
459
self._keys = set(self._nodes)
478
460
if trailers != 1:
479
461
# there must be one line - the empty trailer line.
480
462
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
464
def external_references(self, ref_list_num):
491
465
"""Return references that are not present in this index.
495
469
raise ValueError('No ref list %d, index has %d ref lists'
496
470
% (ref_list_num, self.node_ref_lists))
499
for key, (value, ref_lists) in nodes.iteritems():
472
for key, (value, ref_lists) in self._nodes.iteritems():
500
473
ref_list = ref_lists[ref_list_num]
501
refs.update([ref for ref in ref_list if ref not in nodes])
474
refs.update(ref_list)
475
return refs - self._keys
504
477
def _get_nodes_by_key(self):
505
478
if self._nodes_by_key is None:
633
606
def _iter_entries_from_total_buffer(self, keys):
634
607
"""Iterate over keys when the entire index is parsed."""
635
# Note: See the note in BTreeBuilder.iter_entries for why we don't use
636
# .intersection() here
638
keys = [key for key in keys if key in nodes]
608
keys = keys.intersection(self._keys)
639
609
if self.node_ref_lists:
641
value, node_refs = nodes[key]
611
value, node_refs = self._nodes[key]
642
612
yield self, key, value, node_refs
645
yield self, key, nodes[key]
615
yield self, key, self._nodes[key]
647
617
def iter_entries(self, keys):
648
618
"""Iterate over keys within the index.
1197
1167
self._buffer_all()
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
1170
readv_data = self._transport.readv(self._name, readv_ranges, True,
1206
self._size + self._base_offset)
1208
1173
for offset, data in readv_data:
1209
offset -= base_offset
1210
1174
self._bytes_read += len(data)
1212
# transport.readv() expanded to extra data which isn't part of
1214
data = data[-offset:]
1216
1175
if offset == 0 and len(data) == self._size:
1217
1176
# We read the whole range, most likely because the
1218
1177
# Transport upcast our readv ranges into one long request
1247
1206
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
1207
and then the second and so on. The order of index's can thus influence
1249
1208
performance significantly. For example, if one index is on local disk and a
1250
1209
second on a remote server, the local disk index should be before the other
1251
1210
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
1213
def __init__(self, indices, reload_func=None):
1267
1221
self._indices = indices
1268
1222
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
1224
def __repr__(self):
1278
1225
return "%s(%s)" % (
1279
1226
self.__class__.__name__,
1280
1227
', '.join(map(repr, self._indices)))
1282
def clear_cache(self):
1283
"""See GraphIndex.clear_cache()"""
1284
for index in self._indices:
1287
1229
def get_parent_map(self, keys):
1288
1230
"""See graph.StackedParentsProvider.get_parent_map"""
1289
1231
search_keys = set(keys)
1302
1244
has_key = _has_key_from_parent_map
1304
def insert_index(self, pos, index, name=None):
1246
def insert_index(self, pos, index):
1305
1247
"""Insert a new index in the list of indices to query.
1307
1249
:param pos: The position to insert the index.
1308
1250
: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
1252
self._indices.insert(pos, index)
1314
self._index_names.insert(pos, name)
1316
1254
def iter_all_entries(self):
1317
1255
"""Iterate over all keys within the index
1342
1280
value and are only reported once.
1344
1282
:param keys: An iterable providing the keys to be retrieved.
1345
:return: An iterable of (index, key, reference_lists, value). There is
1346
no defined order for the result iteration - it will be in the most
1283
:return: An iterable of (index, key, reference_lists, value). There is no
1284
defined order for the result iteration - it will be in the most
1347
1285
efficient order for the index.
1349
1287
keys = set(keys)
1353
1290
for index in self._indices:
1357
1293
for node in index.iter_entries(keys):
1358
1294
keys.remove(node[1])
1362
hit_indices.append(index)
1364
1297
except errors.NoSuchFile:
1365
1298
self._reload_or_raise()
1366
self._move_to_front(hit_indices)
1368
1300
def iter_entries_prefix(self, keys):
1369
1301
"""Iterate over keys within the index using prefix matching.
1391
1323
seen_keys = set()
1395
1326
for index in self._indices:
1397
1327
for node in index.iter_entries_prefix(keys):
1398
1328
if node[1] in seen_keys:
1400
1330
seen_keys.add(node[1])
1404
hit_indices.append(index)
1406
1333
except errors.NoSuchFile:
1407
1334
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
1336
def find_ancestry(self, keys, ref_list_num):
1477
1337
"""Find the complete ancestry for the given set of keys.
1633
1487
defined order for the result iteration - it will be in the most
1634
1488
efficient order for the index (keys iteration order in this case).
1636
# Note: See BTreeBuilder.iter_entries for an explanation of why we
1637
# aren't using set().intersection() here
1639
keys = [key for key in keys if key in nodes]
1640
1491
if self.reference_lists:
1492
for key in keys.intersection(self._keys):
1493
node = self._nodes[key]
1643
1494
if not node[0]:
1644
1495
yield self, key, node[2], node[1]
1497
for key in keys.intersection(self._keys):
1498
node = self._nodes[key]
1648
1499
if not node[0]:
1649
1500
yield self, key, node[2]
added
removed
bzrlib/index.py
