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
347
333
if combine_backing_indices is not None:
348
334
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
367
337
class GraphIndex(object):
368
338
"""An index for data with embedded graphs.
382
352
suitable for production use. :XXX
385
def __init__(self, transport, name, size, unlimited_cache=False, offset=0):
355
def __init__(self, transport, name, size):
386
356
"""Open an index called name on transport.
388
358
:param transport: A bzrlib.transport.Transport.
447
414
mutter('Reading entire index %s', self._transport.abspath(self._name))
448
415
if stream is None:
449
416
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
417
self._read_prefix(stream)
455
418
self._expected_elements = 3 + self._key_length
475
437
node_value = value
476
438
self._nodes[key] = node_value
477
439
# cache the keys for quick set intersections
440
self._keys = set(self._nodes)
478
441
if trailers != 1:
479
442
# there must be one line - the empty trailer line.
480
443
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
445
def external_references(self, ref_list_num):
491
446
"""Return references that are not present in this index.
495
450
raise ValueError('No ref list %d, index has %d ref lists'
496
451
% (ref_list_num, self.node_ref_lists))
499
for key, (value, ref_lists) in nodes.iteritems():
453
for key, (value, ref_lists) in self._nodes.iteritems():
500
454
ref_list = ref_lists[ref_list_num]
501
refs.update([ref for ref in ref_list if ref not in nodes])
455
refs.update(ref_list)
456
return refs - self._keys
504
458
def _get_nodes_by_key(self):
505
459
if self._nodes_by_key is None:
633
587
def _iter_entries_from_total_buffer(self, keys):
634
588
"""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]
589
keys = keys.intersection(self._keys)
639
590
if self.node_ref_lists:
641
value, node_refs = nodes[key]
592
value, node_refs = self._nodes[key]
642
593
yield self, key, value, node_refs
645
yield self, key, nodes[key]
596
yield self, key, self._nodes[key]
647
598
def iter_entries(self, keys):
648
599
"""Iterate over keys within the index.
1197
1148
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
1151
readv_data = self._transport.readv(self._name, readv_ranges, True,
1206
self._size + self._base_offset)
1208
1154
for offset, data in readv_data:
1209
offset -= base_offset
1210
1155
self._bytes_read += len(data)
1212
# transport.readv() expanded to extra data which isn't part of
1214
data = data[-offset:]
1216
1156
if offset == 0 and len(data) == self._size:
1217
1157
# We read the whole range, most likely because the
1218
1158
# Transport upcast our readv ranges into one long request
1247
1187
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
1188
and then the second and so on. The order of index's can thus influence
1249
1189
performance significantly. For example, if one index is on local disk and a
1250
1190
second on a remote server, the local disk index should be before the other
1251
1191
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
1194
def __init__(self, indices, reload_func=None):
1267
1202
self._indices = indices
1268
1203
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
1205
def __repr__(self):
1278
1206
return "%s(%s)" % (
1279
1207
self.__class__.__name__,
1280
1208
', '.join(map(repr, self._indices)))
1282
def clear_cache(self):
1283
"""See GraphIndex.clear_cache()"""
1284
for index in self._indices:
1287
1210
def get_parent_map(self, keys):
1288
1211
"""See graph.StackedParentsProvider.get_parent_map"""
1289
1212
search_keys = set(keys)
1302
1225
has_key = _has_key_from_parent_map
1304
def insert_index(self, pos, index, name=None):
1227
def insert_index(self, pos, index):
1305
1228
"""Insert a new index in the list of indices to query.
1307
1230
:param pos: The position to insert the index.
1308
1231
: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
1233
self._indices.insert(pos, index)
1314
self._index_names.insert(pos, name)
1316
1235
def iter_all_entries(self):
1317
1236
"""Iterate over all keys within the index
1342
1261
value and are only reported once.
1344
1263
: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
1264
:return: An iterable of (index, key, reference_lists, value). There is no
1265
defined order for the result iteration - it will be in the most
1347
1266
efficient order for the index.
1349
1268
keys = set(keys)
1353
1271
for index in self._indices:
1357
1274
for node in index.iter_entries(keys):
1358
1275
keys.remove(node[1])
1362
hit_indices.append(index)
1364
1278
except errors.NoSuchFile:
1365
1279
self._reload_or_raise()
1366
self._move_to_front(hit_indices)
1368
1281
def iter_entries_prefix(self, keys):
1369
1282
"""Iterate over keys within the index using prefix matching.
1391
1304
seen_keys = set()
1395
1307
for index in self._indices:
1397
1308
for node in index.iter_entries_prefix(keys):
1398
1309
if node[1] in seen_keys:
1400
1311
seen_keys.add(node[1])
1404
hit_indices.append(index)
1406
1314
except errors.NoSuchFile:
1407
1315
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
1317
def find_ancestry(self, keys, ref_list_num):
1477
1318
"""Find the complete ancestry for the given set of keys.
1633
1468
defined order for the result iteration - it will be in the most
1634
1469
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
1472
if self.reference_lists:
1473
for key in keys.intersection(self._keys):
1474
node = self._nodes[key]
1643
1475
if not node[0]:
1644
1476
yield self, key, node[2], node[1]
1478
for key in keys.intersection(self._keys):
1479
node = self._nodes[key]
1648
1480
if not node[0]:
1649
1481
yield self, key, node[2]
added
removed
bzrlib/index.py
