92
93
:param key_elements: The number of bytestrings in each key.
94
95
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()
98
100
self._nodes_by_key = None
99
101
self._key_length = key_elements
100
102
self._optimize_for_size = False
165
167
key_dict = self._nodes_by_key
166
168
if self.reference_lists:
167
key_value = key, value, node_refs
169
key_value = StaticTuple(key, value, node_refs)
169
key_value = key, value
171
key_value = StaticTuple(key, value)
170
172
for subkey in key[:-1]:
171
173
key_dict = key_dict.setdefault(subkey, {})
172
174
key_dict[key[-1]] = key_value
202
205
if reference not in self._nodes:
203
206
self._check_key(reference)
204
207
absent_references.append(reference)
205
node_refs.append(tuple(reference_list))
206
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
208
213
def add_node(self, key, value, references=()):
209
214
"""Add a node to the index.
224
229
# There may be duplicates, but I don't think it is worth worrying
226
231
self._nodes[reference] = ('a', (), '')
232
self._absent_keys.update(absent_references)
233
self._absent_keys.discard(key)
227
234
self._nodes[key] = ('', node_refs, value)
229
235
if self._nodes_by_key is not None and self._key_length > 1:
230
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'
232
245
def finish(self):
233
246
lines = [_SIGNATURE]
234
247
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
235
248
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
236
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')
237
251
prefix_length = sum(len(x) for x in lines)
238
252
# references are byte offsets. To avoid having to do nasty
239
253
# polynomial work to resolve offsets (references to later in the
333
347
if combine_backing_indices is not None:
334
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
337
367
class GraphIndex(object):
338
368
"""An index for data with embedded graphs.
352
382
suitable for production use. :XXX
355
def __init__(self, transport, name, size):
385
def __init__(self, transport, name, size, unlimited_cache=False):
356
386
"""Open an index called name on transport.
358
388
:param transport: A bzrlib.transport.Transport.
437
468
node_value = value
438
469
self._nodes[key] = node_value
439
470
# cache the keys for quick set intersections
440
self._keys = set(self._nodes)
441
471
if trailers != 1:
442
472
# there must be one line - the empty trailer line.
443
473
raise errors.BadIndexData(self)
475
def clear_cache(self):
476
"""Clear out any cached/memoized values.
478
This can be called at any time, but generally it is used when we have
479
extracted some information, but don't expect to be requesting any more
445
483
def external_references(self, ref_list_num):
446
484
"""Return references that are not present in this index.
450
488
raise ValueError('No ref list %d, index has %d ref lists'
451
489
% (ref_list_num, self.node_ref_lists))
453
for key, (value, ref_lists) in self._nodes.iteritems():
492
for key, (value, ref_lists) in nodes.iteritems():
454
493
ref_list = ref_lists[ref_list_num]
455
refs.update(ref_list)
456
return refs - self._keys
494
refs.update([ref for ref in ref_list if ref not in nodes])
458
497
def _get_nodes_by_key(self):
459
498
if self._nodes_by_key is None:
587
626
def _iter_entries_from_total_buffer(self, keys):
588
627
"""Iterate over keys when the entire index is parsed."""
589
keys = keys.intersection(self._keys)
628
# Note: See the note in BTreeBuilder.iter_entries for why we don't use
629
# .intersection() here
631
keys = [key for key in keys if key in nodes]
590
632
if self.node_ref_lists:
592
value, node_refs = self._nodes[key]
634
value, node_refs = nodes[key]
593
635
yield self, key, value, node_refs
596
yield self, key, self._nodes[key]
638
yield self, key, nodes[key]
598
640
def iter_entries(self, keys):
599
641
"""Iterate over keys within the index.
702
744
# the last thing looked up was a terminal element
703
745
yield (self, ) + key_dict
747
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
748
"""See BTreeIndex._find_ancestors."""
749
# The api can be implemented as a trivial overlay on top of
750
# iter_entries, it is not an efficient implementation, but it at least
754
for index, key, value, refs in self.iter_entries(keys):
755
parent_keys = refs[ref_list_num]
757
parent_map[key] = parent_keys
758
search_keys.update(parent_keys)
759
# Figure out what, if anything, was missing
760
missing_keys.update(set(keys).difference(found_keys))
761
search_keys = search_keys.difference(parent_map)
705
764
def key_count(self):
706
765
"""Return an estimate of the number of keys in this index.
1297
1361
except errors.NoSuchFile:
1298
1362
self._reload_or_raise()
1364
def find_ancestry(self, keys, ref_list_num):
1365
"""Find the complete ancestry for the given set of keys.
1367
Note that this is a whole-ancestry request, so it should be used
1370
:param keys: An iterable of keys to look for
1371
:param ref_list_num: The reference list which references the parents
1373
:return: (parent_map, missing_keys)
1375
missing_keys = set()
1377
keys_to_lookup = set(keys)
1379
while keys_to_lookup:
1380
# keys that *all* indexes claim are missing, stop searching them
1382
all_index_missing = None
1383
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1384
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1386
# len(missing_keys))
1387
for index_idx, index in enumerate(self._indices):
1388
# TODO: we should probably be doing something with
1389
# 'missing_keys' since we've already determined that
1390
# those revisions have not been found anywhere
1391
index_missing_keys = set()
1392
# Find all of the ancestry we can from this index
1393
# keep looking until the search_keys set is empty, which means
1394
# things we didn't find should be in index_missing_keys
1395
search_keys = keys_to_lookup
1397
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1398
# index_idx, len(search_keys),
1399
# len(parent_map), len(index_missing_keys))
1402
# TODO: ref_list_num should really be a parameter, since
1403
# CombinedGraphIndex does not know what the ref lists
1405
search_keys = index._find_ancestors(search_keys,
1406
ref_list_num, parent_map, index_missing_keys)
1407
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1408
# sub_generation, len(search_keys),
1409
# len(parent_map), len(index_missing_keys))
1410
# Now set whatever was missing to be searched in the next index
1411
keys_to_lookup = index_missing_keys
1412
if all_index_missing is None:
1413
all_index_missing = set(index_missing_keys)
1415
all_index_missing.intersection_update(index_missing_keys)
1416
if not keys_to_lookup:
1418
if all_index_missing is None:
1419
# There were no indexes, so all search keys are 'missing'
1420
missing_keys.update(keys_to_lookup)
1421
keys_to_lookup = None
1423
missing_keys.update(all_index_missing)
1424
keys_to_lookup.difference_update(all_index_missing)
1425
return parent_map, missing_keys
1300
1427
def key_count(self):
1301
1428
"""Return an estimate of the number of keys in this index.
1388
1515
defined order for the result iteration - it will be in the most
1389
1516
efficient order for the index (keys iteration order in this case).
1518
# Note: See BTreeBuilder.iter_entries for an explanation of why we
1519
# aren't using set().intersection() here
1521
keys = [key for key in keys if key in nodes]
1392
1522
if self.reference_lists:
1393
for key in keys.intersection(self._keys):
1394
node = self._nodes[key]
1395
1525
if not node[0]:
1396
1526
yield self, key, node[2], node[1]
1398
for key in keys.intersection(self._keys):
1399
node = self._nodes[key]
1400
1530
if not node[0]:
1401
1531
yield self, key, node[2]