56
53
_newline_null_re = re.compile('[\n\0]')
59
def _has_key_from_parent_map(self, key):
60
"""Check if this index has one key.
62
If it's possible to check for multiple keys at once through
63
calling get_parent_map that should be faster.
65
return (key in self.get_parent_map([key]))
68
def _missing_keys_from_parent_map(self, keys):
69
return set(keys) - set(self.get_parent_map(keys))
72
56
class GraphIndexBuilder(object):
73
57
"""A builder that can build a GraphIndex.
75
The resulting graph has the structure::
77
_SIGNATURE OPTIONS NODES NEWLINE
78
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
79
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
81
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
82
KEY := Not-whitespace-utf8
84
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
85
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
86
REFERENCE := DIGITS ; digits is the byte offset in the index of the
88
VALUE := no-newline-no-null-bytes
59
The resulting graph has the structure:
61
_SIGNATURE OPTIONS NODES NEWLINE
62
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
63
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
65
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
66
KEY := Not-whitespace-utf8
68
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
69
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
70
REFERENCE := DIGITS ; digits is the byte offset in the index of the
72
VALUE := no-newline-no-null-bytes
91
75
def __init__(self, reference_lists=0, key_elements=1):
115
97
if not element or _whitespace_re.search(element) is not None:
116
98
raise errors.BadIndexKey(element)
118
def _external_references(self):
119
"""Return references that are not present in this index.
123
# TODO: JAM 2008-11-21 This makes an assumption about how the reference
124
# lists are used. It is currently correct for pack-0.92 through
125
# 1.9, which use the node references (3rd column) second
126
# reference list as the compression parent. Perhaps this should
127
# be moved into something higher up the stack, since it
128
# makes assumptions about how the index is used.
129
if self.reference_lists > 1:
130
for node in self.iter_all_entries():
132
refs.update(node[3][1])
135
# If reference_lists == 0 there can be no external references, and
136
# if reference_lists == 1, then there isn't a place to store the
140
100
def _get_nodes_by_key(self):
141
101
if self._nodes_by_key is None:
142
102
nodes_by_key = {}
233
189
# There may be duplicates, but I don't think it is worth worrying
235
191
self._nodes[reference] = ('a', (), '')
236
self._absent_keys.update(absent_references)
237
self._absent_keys.discard(key)
238
192
self._nodes[key] = ('', node_refs, value)
239
194
if self._nodes_by_key is not None and self._key_length > 1:
240
195
self._update_nodes_by_key(key, value, node_refs)
242
def clear_cache(self):
243
"""See GraphIndex.clear_cache()
245
This is a no-op, but we need the api to conform to a generic 'Index'
249
197
def finish(self):
252
:returns: cStringIO holding the full context of the index as it
253
should be written to disk.
255
198
lines = [_SIGNATURE]
256
199
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
257
200
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
258
key_count = len(self._nodes) - len(self._absent_keys)
259
lines.append(_OPTION_LEN + str(key_count) + '\n')
201
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
260
202
prefix_length = sum(len(x) for x in lines)
261
203
# references are byte offsets. To avoid having to do nasty
262
204
# polynomial work to resolve offsets (references to later in the
338
280
(len(result.getvalue()), expected_bytes))
341
def set_optimize(self, for_size=None, combine_backing_indices=None):
283
def set_optimize(self, for_size=True):
342
284
"""Change how the builder tries to optimize the result.
344
286
:param for_size: Tell the builder to try and make the index as small as
346
:param combine_backing_indices: If the builder spills to disk to save
347
memory, should the on-disk indices be combined. Set to True if you
348
are going to be probing the index, but to False if you are not. (If
349
you are not querying, then the time spent combining is wasted.)
352
290
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
353
291
# other builders do.
354
if for_size is not None:
355
self._optimize_for_size = for_size
356
if combine_backing_indices is not None:
357
self._combine_backing_indices = combine_backing_indices
359
def find_ancestry(self, keys, ref_list_num):
360
"""See CombinedGraphIndex.find_ancestry()"""
366
for _, key, value, ref_lists in self.iter_entries(pending):
367
parent_keys = ref_lists[ref_list_num]
368
parent_map[key] = parent_keys
369
next_pending.update([p for p in parent_keys if p not in
371
missing_keys.update(pending.difference(parent_map))
372
pending = next_pending
373
return parent_map, missing_keys
292
self._optimize_for_size = for_size
376
295
class GraphIndex(object):
377
296
"""An index for data with embedded graphs.
379
298
The index maps keys to a list of key reference lists, and a value.
380
299
Each node has the same number of key reference lists. Each key reference
381
300
list can be empty or an arbitrary length. The value is an opaque NULL
382
terminated string without any newlines. The storage of the index is
301
terminated string without any newlines. The storage of the index is
383
302
hidden in the interface: keys and key references are always tuples of
384
303
bytestrings, never the internal representation (e.g. dictionary offsets).
486
395
node_value = value
487
396
self._nodes[key] = node_value
488
397
# cache the keys for quick set intersections
398
self._keys = set(self._nodes)
489
399
if trailers != 1:
490
400
# there must be one line - the empty trailer line.
491
401
raise errors.BadIndexData(self)
493
def clear_cache(self):
494
"""Clear out any cached/memoized values.
496
This can be called at any time, but generally it is used when we have
497
extracted some information, but don't expect to be requesting any more
501
def external_references(self, ref_list_num):
502
"""Return references that are not present in this index.
505
if ref_list_num + 1 > self.node_ref_lists:
506
raise ValueError('No ref list %d, index has %d ref lists'
507
% (ref_list_num, self.node_ref_lists))
510
for key, (value, ref_lists) in nodes.iteritems():
511
ref_list = ref_lists[ref_list_num]
512
refs.update([ref for ref in ref_list if ref not in nodes])
515
403
def _get_nodes_by_key(self):
516
404
if self._nodes_by_key is None:
517
405
nodes_by_key = {}
762
647
# the last thing looked up was a terminal element
763
648
yield (self, ) + key_dict
765
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
766
"""See BTreeIndex._find_ancestors."""
767
# The api can be implemented as a trivial overlay on top of
768
# iter_entries, it is not an efficient implementation, but it at least
772
for index, key, value, refs in self.iter_entries(keys):
773
parent_keys = refs[ref_list_num]
775
parent_map[key] = parent_keys
776
search_keys.update(parent_keys)
777
# Figure out what, if anything, was missing
778
missing_keys.update(set(keys).difference(found_keys))
779
search_keys = search_keys.difference(parent_map)
782
650
def key_count(self):
783
651
"""Return an estimate of the number of keys in this index.
785
653
For GraphIndex the estimate is exact.
787
655
if self._key_count is None:
1278
1130
self._indices = indices
1279
1131
self._reload_func = reload_func
1280
# Sibling indices are other CombinedGraphIndex that we should call
1281
# _move_to_front_by_name on when we auto-reorder ourself.
1282
self._sibling_indices = []
1283
# A list of names that corresponds to the instances in self._indices,
1284
# so _index_names[0] is always the name for _indices[0], etc. Sibling
1285
# indices must all use the same set of names as each other.
1286
self._index_names = [None] * len(self._indices)
1288
1133
def __repr__(self):
1289
1134
return "%s(%s)" % (
1290
1135
self.__class__.__name__,
1291
1136
', '.join(map(repr, self._indices)))
1293
def clear_cache(self):
1294
"""See GraphIndex.clear_cache()"""
1295
for index in self._indices:
1138
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1139
def get_parents(self, revision_ids):
1140
"""See graph._StackedParentsProvider.get_parents.
1142
This implementation thunks the graph.Graph.get_parents api across to
1145
:param revision_ids: An iterable of graph keys for this graph.
1146
:return: A list of parent details for each key in revision_ids.
1147
Each parent details will be one of:
1148
* None when the key was missing
1149
* (NULL_REVISION,) when the key has no parents.
1150
* (parent_key, parent_key...) otherwise.
1152
parent_map = self.get_parent_map(revision_ids)
1153
return [parent_map.get(r, None) for r in revision_ids]
1298
1155
def get_parent_map(self, keys):
1299
"""See graph.StackedParentsProvider.get_parent_map"""
1156
"""See graph._StackedParentsProvider.get_parent_map"""
1300
1157
search_keys = set(keys)
1301
if _mod_revision.NULL_REVISION in search_keys:
1302
search_keys.discard(_mod_revision.NULL_REVISION)
1303
found_parents = {_mod_revision.NULL_REVISION:[]}
1158
if NULL_REVISION in search_keys:
1159
search_keys.discard(NULL_REVISION)
1160
found_parents = {NULL_REVISION:[]}
1305
1162
found_parents = {}
1306
1163
for index, key, value, refs in self.iter_entries(search_keys):
1307
1164
parents = refs[0]
1308
1165
if not parents:
1309
parents = (_mod_revision.NULL_REVISION,)
1166
parents = (NULL_REVISION,)
1310
1167
found_parents[key] = parents
1311
1168
return found_parents
1313
has_key = _has_key_from_parent_map
1315
def insert_index(self, pos, index, name=None):
1170
def insert_index(self, pos, index):
1316
1171
"""Insert a new index in the list of indices to query.
1318
1173
:param pos: The position to insert the index.
1319
1174
:param index: The index to insert.
1320
:param name: a name for this index, e.g. a pack name. These names can
1321
be used to reflect index reorderings to related CombinedGraphIndex
1322
instances that use the same names. (see set_sibling_indices)
1324
1176
self._indices.insert(pos, index)
1325
self._index_names.insert(pos, name)
1327
1178
def iter_all_entries(self):
1328
1179
"""Iterate over all keys within the index
1402
1247
seen_keys = set()
1406
1250
for index in self._indices:
1408
1251
for node in index.iter_entries_prefix(keys):
1409
1252
if node[1] in seen_keys:
1411
1254
seen_keys.add(node[1])
1415
hit_indices.append(index)
1417
1257
except errors.NoSuchFile:
1418
1258
self._reload_or_raise()
1419
self._move_to_front(hit_indices)
1421
def _move_to_front(self, hit_indices):
1422
"""Rearrange self._indices so that hit_indices are first.
1424
Order is maintained as much as possible, e.g. the first unhit index
1425
will be the first index in _indices after the hit_indices, and the
1426
hit_indices will be present in exactly the order they are passed to
1429
_move_to_front propagates to all objects in self._sibling_indices by
1430
calling _move_to_front_by_name.
1432
if self._indices[:len(hit_indices)] == hit_indices:
1433
# The 'hit_indices' are already at the front (and in the same
1434
# order), no need to re-order
1436
hit_names = self._move_to_front_by_index(hit_indices)
1437
for sibling_idx in self._sibling_indices:
1438
sibling_idx._move_to_front_by_name(hit_names)
1440
def _move_to_front_by_index(self, hit_indices):
1441
"""Core logic for _move_to_front.
1443
Returns a list of names corresponding to the hit_indices param.
1445
indices_info = zip(self._index_names, self._indices)
1446
if 'index' in debug.debug_flags:
1447
trace.mutter('CombinedGraphIndex reordering: currently %r, '
1448
'promoting %r', indices_info, hit_indices)
1451
new_hit_indices = []
1454
for offset, (name, idx) in enumerate(indices_info):
1455
if idx in hit_indices:
1456
hit_names.append(name)
1457
new_hit_indices.append(idx)
1458
if len(new_hit_indices) == len(hit_indices):
1459
# We've found all of the hit entries, everything else is
1461
unhit_names.extend(self._index_names[offset+1:])
1462
unhit_indices.extend(self._indices[offset+1:])
1465
unhit_names.append(name)
1466
unhit_indices.append(idx)
1468
self._indices = new_hit_indices + unhit_indices
1469
self._index_names = hit_names + unhit_names
1470
if 'index' in debug.debug_flags:
1471
trace.mutter('CombinedGraphIndex reordered: %r', self._indices)
1474
def _move_to_front_by_name(self, hit_names):
1475
"""Moves indices named by 'hit_names' to front of the search order, as
1476
described in _move_to_front.
1478
# Translate names to index instances, and then call
1479
# _move_to_front_by_index.
1480
indices_info = zip(self._index_names, self._indices)
1482
for name, idx in indices_info:
1483
if name in hit_names:
1484
hit_indices.append(idx)
1485
self._move_to_front_by_index(hit_indices)
1487
def find_ancestry(self, keys, ref_list_num):
1488
"""Find the complete ancestry for the given set of keys.
1490
Note that this is a whole-ancestry request, so it should be used
1493
:param keys: An iterable of keys to look for
1494
:param ref_list_num: The reference list which references the parents
1496
:return: (parent_map, missing_keys)
1498
# XXX: make this call _move_to_front?
1499
missing_keys = set()
1501
keys_to_lookup = set(keys)
1503
while keys_to_lookup:
1504
# keys that *all* indexes claim are missing, stop searching them
1506
all_index_missing = None
1507
# print 'gen\tidx\tsub\tn_keys\tn_pmap\tn_miss'
1508
# print '%4d\t\t\t%4d\t%5d\t%5d' % (generation, len(keys_to_lookup),
1510
# len(missing_keys))
1511
for index_idx, index in enumerate(self._indices):
1512
# TODO: we should probably be doing something with
1513
# 'missing_keys' since we've already determined that
1514
# those revisions have not been found anywhere
1515
index_missing_keys = set()
1516
# Find all of the ancestry we can from this index
1517
# keep looking until the search_keys set is empty, which means
1518
# things we didn't find should be in index_missing_keys
1519
search_keys = keys_to_lookup
1521
# print ' \t%2d\t\t%4d\t%5d\t%5d' % (
1522
# index_idx, len(search_keys),
1523
# len(parent_map), len(index_missing_keys))
1526
# TODO: ref_list_num should really be a parameter, since
1527
# CombinedGraphIndex does not know what the ref lists
1529
search_keys = index._find_ancestors(search_keys,
1530
ref_list_num, parent_map, index_missing_keys)
1531
# print ' \t \t%2d\t%4d\t%5d\t%5d' % (
1532
# sub_generation, len(search_keys),
1533
# len(parent_map), len(index_missing_keys))
1534
# Now set whatever was missing to be searched in the next index
1535
keys_to_lookup = index_missing_keys
1536
if all_index_missing is None:
1537
all_index_missing = set(index_missing_keys)
1539
all_index_missing.intersection_update(index_missing_keys)
1540
if not keys_to_lookup:
1542
if all_index_missing is None:
1543
# There were no indexes, so all search keys are 'missing'
1544
missing_keys.update(keys_to_lookup)
1545
keys_to_lookup = None
1547
missing_keys.update(all_index_missing)
1548
keys_to_lookup.difference_update(all_index_missing)
1549
return parent_map, missing_keys
1551
1260
def key_count(self):
1552
1261
"""Return an estimate of the number of keys in this index.