93
93
:param key_elements: The number of bytestrings in each key.
95
95
self.reference_lists = reference_lists
97
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()
99
100
self._nodes_by_key = None
100
101
self._key_length = key_elements
101
102
self._optimize_for_size = False
166
167
key_dict = self._nodes_by_key
167
168
if self.reference_lists:
168
key_value = key, value, node_refs
169
key_value = StaticTuple(key, value, node_refs)
170
key_value = key, value
171
key_value = StaticTuple(key, value)
171
172
for subkey in key[:-1]:
172
173
key_dict = key_dict.setdefault(subkey, {})
173
174
key_dict[key[-1]] = key_value
203
205
if reference not in self._nodes:
204
206
self._check_key(reference)
205
207
absent_references.append(reference)
207
node_refs.append(tuple(reference_list))
209
return tuple(node_refs), absent_references
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
211
213
def add_node(self, key, value, references=()):
212
214
"""Add a node to the index.
227
229
# There may be duplicates, but I don't think it is worth worrying
229
231
self._nodes[reference] = ('a', (), '')
232
self._absent_keys.update(absent_references)
233
self._absent_keys.discard(key)
230
234
self._nodes[key] = ('', node_refs, value)
232
235
if self._nodes_by_key is not None and self._key_length > 1:
233
236
self._update_nodes_by_key(key, value, node_refs)
243
246
lines = [_SIGNATURE]
244
247
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
245
248
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
246
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')
247
251
prefix_length = sum(len(x) for x in lines)
248
252
# references are byte offsets. To avoid having to do nasty
249
253
# polynomial work to resolve offsets (references to later in the
378
382
suitable for production use. :XXX
381
def __init__(self, transport, name, size, unlimited_cache=False):
385
def __init__(self, transport, name, size, unlimited_cache=False, offset=0):
382
386
"""Open an index called name on transport.
384
388
:param transport: A bzrlib.transport.Transport.
440
447
mutter('Reading entire index %s', self._transport.abspath(self._name))
441
448
if stream is None:
442
449
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:])
443
454
self._read_prefix(stream)
444
455
self._expected_elements = 3 + self._key_length
484
496
raise ValueError('No ref list %d, index has %d ref lists'
485
497
% (ref_list_num, self.node_ref_lists))
487
for key, (value, ref_lists) in self._nodes.iteritems():
500
for key, (value, ref_lists) in nodes.iteritems():
488
501
ref_list = ref_lists[ref_list_num]
489
refs.update(ref_list)
490
return refs - self._keys
502
refs.update([ref for ref in ref_list if ref not in nodes])
492
505
def _get_nodes_by_key(self):
493
506
if self._nodes_by_key is None:
621
634
def _iter_entries_from_total_buffer(self, keys):
622
635
"""Iterate over keys when the entire index is parsed."""
623
keys = keys.intersection(self._keys)
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]
624
640
if self.node_ref_lists:
626
value, node_refs = self._nodes[key]
642
value, node_refs = nodes[key]
627
643
yield self, key, value, node_refs
630
yield self, key, self._nodes[key]
646
yield self, key, nodes[key]
632
648
def iter_entries(self, keys):
633
649
"""Iterate over keys within the index.
1182
1198
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]
1185
1206
readv_data = self._transport.readv(self._name, readv_ranges, True,
1207
self._size + self._base_offset)
1188
1209
for offset, data in readv_data:
1210
offset -= base_offset
1189
1211
self._bytes_read += len(data)
1213
# transport.readv() expanded to extra data which isn't part of
1215
data = data[-offset:]
1190
1217
if offset == 0 and len(data) == self._size:
1191
1218
# We read the whole range, most likely because the
1192
1219
# Transport upcast our readv ranges into one long request
1221
1248
Queries against the combined index will be made against the first index,
1222
and then the second and so on. The order of index's can thus influence
1249
and then the second and so on. The order of indices can thus influence
1223
1250
performance significantly. For example, if one index is on local disk and a
1224
1251
second on a remote server, the local disk index should be before the other
1225
1252
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).
1228
1260
def __init__(self, indices, reload_func=None):
1236
1268
self._indices = indices
1237
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)
1239
1278
def __repr__(self):
1240
1279
return "%s(%s)" % (
1264
1303
has_key = _has_key_from_parent_map
1266
def insert_index(self, pos, index):
1305
def insert_index(self, pos, index, name=None):
1267
1306
"""Insert a new index in the list of indices to query.
1269
1308
:param pos: The position to insert the index.
1270
1309
: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)
1272
1314
self._indices.insert(pos, index)
1315
self._index_names.insert(pos, name)
1274
1317
def iter_all_entries(self):
1275
1318
"""Iterate over all keys within the index
1300
1343
value and are only reported once.
1302
1345
:param keys: An iterable providing the keys to be retrieved.
1303
:return: An iterable of (index, key, reference_lists, value). There is no
1304
defined order for the result iteration - it will be in the most
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
1305
1348
efficient order for the index.
1307
1350
keys = set(keys)
1310
1354
for index in self._indices:
1313
1358
for node in index.iter_entries(keys):
1314
1359
keys.remove(node[1])
1363
hit_indices.append(index)
1317
1365
except errors.NoSuchFile:
1318
1366
self._reload_or_raise()
1367
self._move_to_front(hit_indices)
1320
1369
def iter_entries_prefix(self, keys):
1321
1370
"""Iterate over keys within the index using prefix matching.
1343
1392
seen_keys = set()
1346
1396
for index in self._indices:
1347
1398
for node in index.iter_entries_prefix(keys):
1348
1399
if node[1] in seen_keys:
1350
1401
seen_keys.add(node[1])
1405
hit_indices.append(index)
1353
1407
except errors.NoSuchFile:
1354
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)
1356
1477
def find_ancestry(self, keys, ref_list_num):
1357
1478
"""Find the complete ancestry for the given set of keys.
1507
1634
defined order for the result iteration - it will be in the most
1508
1635
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]
1511
1641
if self.reference_lists:
1512
for key in keys.intersection(self._keys):
1513
node = self._nodes[key]
1514
1644
if not node[0]:
1515
1645
yield self, key, node[2], node[1]
1517
for key in keys.intersection(self._keys):
1518
node = self._nodes[key]
1519
1649
if not node[0]:
1520
1650
yield self, key, node[2]