32
32
from bzrlib.lazy_import import lazy_import
33
33
lazy_import(globals(), """
36
revision as _mod_revision,
34
from bzrlib import trace
35
from bzrlib.bisect_multi import bisect_multi_bytes
36
from bzrlib.revision import NULL_REVISION
37
from bzrlib.trace import mutter
40
39
from bzrlib import (
44
from bzrlib.static_tuple import StaticTuple
46
44
_HEADER_READV = (0, 200)
47
45
_OPTION_KEY_ELEMENTS = "key_elements="
70
68
class GraphIndexBuilder(object):
71
69
"""A builder that can build a GraphIndex.
73
The resulting graph has the structure::
71
The resulting graph has the structure:
75
_SIGNATURE OPTIONS NODES NEWLINE
76
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
77
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
79
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
80
KEY := Not-whitespace-utf8
82
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
83
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
84
REFERENCE := DIGITS ; digits is the byte offset in the index of the
86
VALUE := no-newline-no-null-bytes
73
_SIGNATURE OPTIONS NODES NEWLINE
74
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
75
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
77
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
78
KEY := Not-whitespace-utf8
80
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
81
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
82
REFERENCE := DIGITS ; digits is the byte offset in the index of the
84
VALUE := no-newline-no-null-bytes
89
87
def __init__(self, reference_lists=0, key_elements=1):
94
92
:param key_elements: The number of bytestrings in each key.
96
94
self.reference_lists = reference_lists
97
96
# A dict of {key: (absent, ref_lists, value)}
99
# Keys that are referenced but not actually present in this index
100
self._absent_keys = set()
101
98
self._nodes_by_key = None
102
99
self._key_length = key_elements
103
100
self._optimize_for_size = False
168
165
key_dict = self._nodes_by_key
169
166
if self.reference_lists:
170
key_value = StaticTuple(key, value, node_refs)
167
key_value = key, value, node_refs
172
key_value = StaticTuple(key, value)
169
key_value = key, value
173
170
for subkey in key[:-1]:
174
171
key_dict = key_dict.setdefault(subkey, {})
175
172
key_dict[key[-1]] = key_value
185
182
:param value: The value associate with this key. Must not contain
186
183
newlines or null characters.
187
184
:return: (node_refs, absent_references)
189
* node_refs: basically a packed form of 'references' where all
191
* absent_references: reference keys that are not in self._nodes.
192
This may contain duplicates if the same key is referenced in
185
node_refs basically a packed form of 'references' where all
187
absent_references reference keys that are not in self._nodes.
188
This may contain duplicates if the same key is
189
referenced in multiple lists.
195
as_st = StaticTuple.from_sequence
196
191
self._check_key(key)
197
192
if _newline_null_re.search(value) is not None:
198
193
raise errors.BadIndexValue(value)
207
202
if reference not in self._nodes:
208
203
self._check_key(reference)
209
204
absent_references.append(reference)
210
reference_list = as_st([as_st(ref).intern()
211
for ref in reference_list])
212
node_refs.append(reference_list)
213
return as_st(node_refs), absent_references
205
node_refs.append(tuple(reference_list))
206
return tuple(node_refs), absent_references
215
208
def add_node(self, key, value, references=()):
216
209
"""Add a node to the index.
231
224
# There may be duplicates, but I don't think it is worth worrying
233
226
self._nodes[reference] = ('a', (), '')
234
self._absent_keys.update(absent_references)
235
self._absent_keys.discard(key)
236
227
self._nodes[key] = ('', node_refs, value)
237
229
if self._nodes_by_key is not None and self._key_length > 1:
238
230
self._update_nodes_by_key(key, value, node_refs)
240
def clear_cache(self):
241
"""See GraphIndex.clear_cache()
243
This is a no-op, but we need the api to conform to a generic 'Index'
247
232
def finish(self):
250
:returns: cStringIO holding the full context of the index as it
251
should be written to disk.
253
233
lines = [_SIGNATURE]
254
234
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
255
235
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
256
key_count = len(self._nodes) - len(self._absent_keys)
257
lines.append(_OPTION_LEN + str(key_count) + '\n')
236
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
258
237
prefix_length = sum(len(x) for x in lines)
259
238
# references are byte offsets. To avoid having to do nasty
260
239
# polynomial work to resolve offsets (references to later in the
389
368
suitable for production use. :XXX
392
def __init__(self, transport, name, size, unlimited_cache=False, offset=0):
371
def __init__(self, transport, name, size, unlimited_cache=False):
393
372
"""Open an index called name on transport.
395
374
:param transport: A bzrlib.transport.Transport.
451
427
# We already did this
453
429
if 'index' in debug.debug_flags:
454
trace.mutter('Reading entire index %s',
455
self._transport.abspath(self._name))
430
mutter('Reading entire index %s', self._transport.abspath(self._name))
456
431
if stream is None:
457
432
stream = self._transport.get(self._name)
458
if self._base_offset != 0:
459
# This is wasteful, but it is better than dealing with
460
# adjusting all the offsets, etc.
461
stream = StringIO(stream.read()[self._base_offset:])
462
433
self._read_prefix(stream)
463
434
self._expected_elements = 3 + self._key_length
484
453
node_value = value
485
454
self._nodes[key] = node_value
486
455
# cache the keys for quick set intersections
456
self._keys = set(self._nodes)
487
457
if trailers != 1:
488
458
# there must be one line - the empty trailer line.
489
459
raise errors.BadIndexData(self)
491
def clear_cache(self):
492
"""Clear out any cached/memoized values.
494
This can be called at any time, but generally it is used when we have
495
extracted some information, but don't expect to be requesting any more
499
461
def external_references(self, ref_list_num):
500
462
"""Return references that are not present in this index.
504
466
raise ValueError('No ref list %d, index has %d ref lists'
505
467
% (ref_list_num, self.node_ref_lists))
508
for key, (value, ref_lists) in nodes.iteritems():
469
for key, (value, ref_lists) in self._nodes.iteritems():
509
470
ref_list = ref_lists[ref_list_num]
510
refs.update([ref for ref in ref_list if ref not in nodes])
471
refs.update(ref_list)
472
return refs - self._keys
513
474
def _get_nodes_by_key(self):
514
475
if self._nodes_by_key is None:
642
603
def _iter_entries_from_total_buffer(self, keys):
643
604
"""Iterate over keys when the entire index is parsed."""
644
# Note: See the note in BTreeBuilder.iter_entries for why we don't use
645
# .intersection() here
647
keys = [key for key in keys if key in nodes]
605
keys = keys.intersection(self._keys)
648
606
if self.node_ref_lists:
650
value, node_refs = nodes[key]
608
value, node_refs = self._nodes[key]
651
609
yield self, key, value, node_refs
654
yield self, key, nodes[key]
612
yield self, key, self._nodes[key]
656
614
def iter_entries(self, keys):
657
615
"""Iterate over keys within the index.
1206
1164
self._buffer_all()
1209
base_offset = self._base_offset
1210
if base_offset != 0:
1211
# Rewrite the ranges for the offset
1212
readv_ranges = [(start+base_offset, size)
1213
for start, size in readv_ranges]
1214
1167
readv_data = self._transport.readv(self._name, readv_ranges, True,
1215
self._size + self._base_offset)
1217
1170
for offset, data in readv_data:
1218
offset -= base_offset
1219
1171
self._bytes_read += len(data)
1221
# transport.readv() expanded to extra data which isn't part of
1223
data = data[-offset:]
1225
1172
if offset == 0 and len(data) == self._size:
1226
1173
# We read the whole range, most likely because the
1227
1174
# Transport upcast our readv ranges into one long request
1256
1203
Queries against the combined index will be made against the first index,
1257
and then the second and so on. The order of indices can thus influence
1204
and then the second and so on. The order of index's can thus influence
1258
1205
performance significantly. For example, if one index is on local disk and a
1259
1206
second on a remote server, the local disk index should be before the other
1260
1207
in the index list.
1262
Also, queries tend to need results from the same indices as previous
1263
queries. So the indices will be reordered after every query to put the
1264
indices that had the result(s) of that query first (while otherwise
1265
preserving the relative ordering).
1268
1210
def __init__(self, indices, reload_func=None):
1276
1218
self._indices = indices
1277
1219
self._reload_func = reload_func
1278
# Sibling indices are other CombinedGraphIndex that we should call
1279
# _move_to_front_by_name on when we auto-reorder ourself.
1280
self._sibling_indices = []
1281
# A list of names that corresponds to the instances in self._indices,
1282
# so _index_names[0] is always the name for _indices[0], etc. Sibling
1283
# indices must all use the same set of names as each other.
1284
self._index_names = [None] * len(self._indices)
1286
1221
def __repr__(self):
1287
1222
return "%s(%s)" % (
1288
1223
self.__class__.__name__,
1289
1224
', '.join(map(repr, self._indices)))
1291
def clear_cache(self):
1292
"""See GraphIndex.clear_cache()"""
1293
for index in self._indices:
1296
1226
def get_parent_map(self, keys):
1297
1227
"""See graph.StackedParentsProvider.get_parent_map"""
1298
1228
search_keys = set(keys)
1299
if _mod_revision.NULL_REVISION in search_keys:
1300
search_keys.discard(_mod_revision.NULL_REVISION)
1301
found_parents = {_mod_revision.NULL_REVISION:[]}
1229
if NULL_REVISION in search_keys:
1230
search_keys.discard(NULL_REVISION)
1231
found_parents = {NULL_REVISION:[]}
1303
1233
found_parents = {}
1304
1234
for index, key, value, refs in self.iter_entries(search_keys):
1305
1235
parents = refs[0]
1306
1236
if not parents:
1307
parents = (_mod_revision.NULL_REVISION,)
1237
parents = (NULL_REVISION,)
1308
1238
found_parents[key] = parents
1309
1239
return found_parents
1311
1241
has_key = _has_key_from_parent_map
1313
def insert_index(self, pos, index, name=None):
1243
def insert_index(self, pos, index):
1314
1244
"""Insert a new index in the list of indices to query.
1316
1246
:param pos: The position to insert the index.
1317
1247
:param index: The index to insert.
1318
:param name: a name for this index, e.g. a pack name. These names can
1319
be used to reflect index reorderings to related CombinedGraphIndex
1320
instances that use the same names. (see set_sibling_indices)
1322
1249
self._indices.insert(pos, index)
1323
self._index_names.insert(pos, name)
1325
1251
def iter_all_entries(self):
1326
1252
"""Iterate over all keys within the index
1351
1277
value and are only reported once.
1353
1279
:param keys: An iterable providing the keys to be retrieved.
1354
:return: An iterable of (index, key, reference_lists, value). There is
1355
no defined order for the result iteration - it will be in the most
1280
:return: An iterable of (index, key, reference_lists, value). There is no
1281
defined order for the result iteration - it will be in the most
1356
1282
efficient order for the index.
1358
1284
keys = set(keys)
1362
1287
for index in self._indices:
1366
1290
for node in index.iter_entries(keys):
1367
1291
keys.remove(node[1])
1371
hit_indices.append(index)
1373
1294
except errors.NoSuchFile:
1374
1295
self._reload_or_raise()
1375
self._move_to_front(hit_indices)
1377
1297
def iter_entries_prefix(self, keys):
1378
1298
"""Iterate over keys within the index using prefix matching.
1400
1320
seen_keys = set()
1404
1323
for index in self._indices:
1406
1324
for node in index.iter_entries_prefix(keys):
1407
1325
if node[1] in seen_keys:
1409
1327
seen_keys.add(node[1])
1413
hit_indices.append(index)
1415
1330
except errors.NoSuchFile:
1416
1331
self._reload_or_raise()
1417
self._move_to_front(hit_indices)
1419
def _move_to_front(self, hit_indices):
1420
"""Rearrange self._indices so that hit_indices are first.
1422
Order is maintained as much as possible, e.g. the first unhit index
1423
will be the first index in _indices after the hit_indices, and the
1424
hit_indices will be present in exactly the order they are passed to
1427
_move_to_front propagates to all objects in self._sibling_indices by
1428
calling _move_to_front_by_name.
1430
if self._indices[:len(hit_indices)] == hit_indices:
1431
# The 'hit_indices' are already at the front (and in the same
1432
# order), no need to re-order
1434
hit_names = self._move_to_front_by_index(hit_indices)
1435
for sibling_idx in self._sibling_indices:
1436
sibling_idx._move_to_front_by_name(hit_names)
1438
def _move_to_front_by_index(self, hit_indices):
1439
"""Core logic for _move_to_front.
1441
Returns a list of names corresponding to the hit_indices param.
1443
indices_info = zip(self._index_names, self._indices)
1444
if 'index' in debug.debug_flags:
1445
trace.mutter('CombinedGraphIndex reordering: currently %r, '
1446
'promoting %r', indices_info, hit_indices)
1449
new_hit_indices = []
1452
for offset, (name, idx) in enumerate(indices_info):
1453
if idx in hit_indices:
1454
hit_names.append(name)
1455
new_hit_indices.append(idx)
1456
if len(new_hit_indices) == len(hit_indices):
1457
# We've found all of the hit entries, everything else is
1459
unhit_names.extend(self._index_names[offset+1:])
1460
unhit_indices.extend(self._indices[offset+1:])
1463
unhit_names.append(name)
1464
unhit_indices.append(idx)
1466
self._indices = new_hit_indices + unhit_indices
1467
self._index_names = hit_names + unhit_names
1468
if 'index' in debug.debug_flags:
1469
trace.mutter('CombinedGraphIndex reordered: %r', self._indices)
1472
def _move_to_front_by_name(self, hit_names):
1473
"""Moves indices named by 'hit_names' to front of the search order, as
1474
described in _move_to_front.
1476
# Translate names to index instances, and then call
1477
# _move_to_front_by_index.
1478
indices_info = zip(self._index_names, self._indices)
1480
for name, idx in indices_info:
1481
if name in hit_names:
1482
hit_indices.append(idx)
1483
self._move_to_front_by_index(hit_indices)
1485
1333
def find_ancestry(self, keys, ref_list_num):
1486
1334
"""Find the complete ancestry for the given set of keys.
1642
1484
defined order for the result iteration - it will be in the most
1643
1485
efficient order for the index (keys iteration order in this case).
1645
# Note: See BTreeBuilder.iter_entries for an explanation of why we
1646
# aren't using set().intersection() here
1648
keys = [key for key in keys if key in nodes]
1649
1488
if self.reference_lists:
1489
for key in keys.intersection(self._keys):
1490
node = self._nodes[key]
1652
1491
if not node[0]:
1653
1492
yield self, key, node[2], node[1]
1494
for key in keys.intersection(self._keys):
1495
node = self._nodes[key]
1657
1496
if not node[0]:
1658
1497
yield self, key, node[2]