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):
248
233
lines = [_SIGNATURE]
249
234
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
250
235
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
251
key_count = len(self._nodes) - len(self._absent_keys)
252
lines.append(_OPTION_LEN + str(key_count) + '\n')
236
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
253
237
prefix_length = sum(len(x) for x in lines)
254
238
# references are byte offsets. To avoid having to do nasty
255
239
# polynomial work to resolve offsets (references to later in the
384
368
suitable for production use. :XXX
387
def __init__(self, transport, name, size, unlimited_cache=False, offset=0):
371
def __init__(self, transport, name, size):
388
372
"""Open an index called name on transport.
390
374
:param transport: A bzrlib.transport.Transport.
446
427
# We already did this
448
429
if 'index' in debug.debug_flags:
449
trace.mutter('Reading entire index %s',
450
self._transport.abspath(self._name))
430
mutter('Reading entire index %s', self._transport.abspath(self._name))
451
431
if stream is None:
452
432
stream = self._transport.get(self._name)
453
if self._base_offset != 0:
454
# This is wasteful, but it is better than dealing with
455
# adjusting all the offsets, etc.
456
stream = StringIO(stream.read()[self._base_offset:])
457
433
self._read_prefix(stream)
458
434
self._expected_elements = 3 + self._key_length
479
453
node_value = value
480
454
self._nodes[key] = node_value
481
455
# cache the keys for quick set intersections
456
self._keys = set(self._nodes)
482
457
if trailers != 1:
483
458
# there must be one line - the empty trailer line.
484
459
raise errors.BadIndexData(self)
486
def clear_cache(self):
487
"""Clear out any cached/memoized values.
489
This can be called at any time, but generally it is used when we have
490
extracted some information, but don't expect to be requesting any more
494
461
def external_references(self, ref_list_num):
495
462
"""Return references that are not present in this index.
499
466
raise ValueError('No ref list %d, index has %d ref lists'
500
467
% (ref_list_num, self.node_ref_lists))
503
for key, (value, ref_lists) in nodes.iteritems():
469
for key, (value, ref_lists) in self._nodes.iteritems():
504
470
ref_list = ref_lists[ref_list_num]
505
refs.update([ref for ref in ref_list if ref not in nodes])
471
refs.update(ref_list)
472
return refs - self._keys
508
474
def _get_nodes_by_key(self):
509
475
if self._nodes_by_key is None:
637
603
def _iter_entries_from_total_buffer(self, keys):
638
604
"""Iterate over keys when the entire index is parsed."""
639
# Note: See the note in BTreeBuilder.iter_entries for why we don't use
640
# .intersection() here
642
keys = [key for key in keys if key in nodes]
605
keys = keys.intersection(self._keys)
643
606
if self.node_ref_lists:
645
value, node_refs = nodes[key]
608
value, node_refs = self._nodes[key]
646
609
yield self, key, value, node_refs
649
yield self, key, nodes[key]
612
yield self, key, self._nodes[key]
651
614
def iter_entries(self, keys):
652
615
"""Iterate over keys within the index.
1201
1164
self._buffer_all()
1204
base_offset = self._base_offset
1205
if base_offset != 0:
1206
# Rewrite the ranges for the offset
1207
readv_ranges = [(start+base_offset, size)
1208
for start, size in readv_ranges]
1209
1167
readv_data = self._transport.readv(self._name, readv_ranges, True,
1210
self._size + self._base_offset)
1212
1170
for offset, data in readv_data:
1213
offset -= base_offset
1214
1171
self._bytes_read += len(data)
1216
# transport.readv() expanded to extra data which isn't part of
1218
data = data[-offset:]
1220
1172
if offset == 0 and len(data) == self._size:
1221
1173
# We read the whole range, most likely because the
1222
1174
# Transport upcast our readv ranges into one long request
1251
1203
Queries against the combined index will be made against the first index,
1252
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
1253
1205
performance significantly. For example, if one index is on local disk and a
1254
1206
second on a remote server, the local disk index should be before the other
1255
1207
in the index list.
1257
Also, queries tend to need results from the same indices as previous
1258
queries. So the indices will be reordered after every query to put the
1259
indices that had the result(s) of that query first (while otherwise
1260
preserving the relative ordering).
1263
1210
def __init__(self, indices, reload_func=None):
1271
1218
self._indices = indices
1272
1219
self._reload_func = reload_func
1273
# Sibling indices are other CombinedGraphIndex that we should call
1274
# _move_to_front_by_name on when we auto-reorder ourself.
1275
self._sibling_indices = []
1276
# A list of names that corresponds to the instances in self._indices,
1277
# so _index_names[0] is always the name for _indices[0], etc. Sibling
1278
# indices must all use the same set of names as each other.
1279
self._index_names = [None] * len(self._indices)
1281
1221
def __repr__(self):
1282
1222
return "%s(%s)" % (
1283
1223
self.__class__.__name__,
1284
1224
', '.join(map(repr, self._indices)))
1286
def clear_cache(self):
1287
"""See GraphIndex.clear_cache()"""
1288
for index in self._indices:
1291
1226
def get_parent_map(self, keys):
1292
1227
"""See graph.StackedParentsProvider.get_parent_map"""
1293
1228
search_keys = set(keys)
1294
if _mod_revision.NULL_REVISION in search_keys:
1295
search_keys.discard(_mod_revision.NULL_REVISION)
1296
found_parents = {_mod_revision.NULL_REVISION:[]}
1229
if NULL_REVISION in search_keys:
1230
search_keys.discard(NULL_REVISION)
1231
found_parents = {NULL_REVISION:[]}
1298
1233
found_parents = {}
1299
1234
for index, key, value, refs in self.iter_entries(search_keys):
1300
1235
parents = refs[0]
1301
1236
if not parents:
1302
parents = (_mod_revision.NULL_REVISION,)
1237
parents = (NULL_REVISION,)
1303
1238
found_parents[key] = parents
1304
1239
return found_parents
1306
1241
has_key = _has_key_from_parent_map
1308
def insert_index(self, pos, index, name=None):
1243
def insert_index(self, pos, index):
1309
1244
"""Insert a new index in the list of indices to query.
1311
1246
:param pos: The position to insert the index.
1312
1247
:param index: The index to insert.
1313
:param name: a name for this index, e.g. a pack name. These names can
1314
be used to reflect index reorderings to related CombinedGraphIndex
1315
instances that use the same names. (see set_sibling_indices)
1317
1249
self._indices.insert(pos, index)
1318
self._index_names.insert(pos, name)
1320
1251
def iter_all_entries(self):
1321
1252
"""Iterate over all keys within the index
1346
1277
value and are only reported once.
1348
1279
:param keys: An iterable providing the keys to be retrieved.
1349
:return: An iterable of (index, key, reference_lists, value). There is
1350
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
1351
1282
efficient order for the index.
1353
1284
keys = set(keys)
1357
1287
for index in self._indices:
1361
1290
for node in index.iter_entries(keys):
1362
1291
keys.remove(node[1])
1366
hit_indices.append(index)
1368
1294
except errors.NoSuchFile:
1369
1295
self._reload_or_raise()
1370
self._move_to_front(hit_indices)
1372
1297
def iter_entries_prefix(self, keys):
1373
1298
"""Iterate over keys within the index using prefix matching.
1395
1320
seen_keys = set()
1399
1323
for index in self._indices:
1401
1324
for node in index.iter_entries_prefix(keys):
1402
1325
if node[1] in seen_keys:
1404
1327
seen_keys.add(node[1])
1408
hit_indices.append(index)
1410
1330
except errors.NoSuchFile:
1411
1331
self._reload_or_raise()
1412
self._move_to_front(hit_indices)
1414
def _move_to_front(self, hit_indices):
1415
"""Rearrange self._indices so that hit_indices are first.
1417
Order is maintained as much as possible, e.g. the first unhit index
1418
will be the first index in _indices after the hit_indices, and the
1419
hit_indices will be present in exactly the order they are passed to
1422
_move_to_front propagates to all objects in self._sibling_indices by
1423
calling _move_to_front_by_name.
1425
if self._indices[:len(hit_indices)] == hit_indices:
1426
# The 'hit_indices' are already at the front (and in the same
1427
# order), no need to re-order
1429
hit_names = self._move_to_front_by_index(hit_indices)
1430
for sibling_idx in self._sibling_indices:
1431
sibling_idx._move_to_front_by_name(hit_names)
1433
def _move_to_front_by_index(self, hit_indices):
1434
"""Core logic for _move_to_front.
1436
Returns a list of names corresponding to the hit_indices param.
1438
indices_info = zip(self._index_names, self._indices)
1439
if 'index' in debug.debug_flags:
1440
trace.mutter('CombinedGraphIndex reordering: currently %r, '
1441
'promoting %r', indices_info, hit_indices)
1444
new_hit_indices = []
1447
for offset, (name, idx) in enumerate(indices_info):
1448
if idx in hit_indices:
1449
hit_names.append(name)
1450
new_hit_indices.append(idx)
1451
if len(new_hit_indices) == len(hit_indices):
1452
# We've found all of the hit entries, everything else is
1454
unhit_names.extend(self._index_names[offset+1:])
1455
unhit_indices.extend(self._indices[offset+1:])
1458
unhit_names.append(name)
1459
unhit_indices.append(idx)
1461
self._indices = new_hit_indices + unhit_indices
1462
self._index_names = hit_names + unhit_names
1463
if 'index' in debug.debug_flags:
1464
trace.mutter('CombinedGraphIndex reordered: %r', self._indices)
1467
def _move_to_front_by_name(self, hit_names):
1468
"""Moves indices named by 'hit_names' to front of the search order, as
1469
described in _move_to_front.
1471
# Translate names to index instances, and then call
1472
# _move_to_front_by_index.
1473
indices_info = zip(self._index_names, self._indices)
1475
for name, idx in indices_info:
1476
if name in hit_names:
1477
hit_indices.append(idx)
1478
self._move_to_front_by_index(hit_indices)
1480
1333
def find_ancestry(self, keys, ref_list_num):
1481
1334
"""Find the complete ancestry for the given set of keys.
1637
1484
defined order for the result iteration - it will be in the most
1638
1485
efficient order for the index (keys iteration order in this case).
1640
# Note: See BTreeBuilder.iter_entries for an explanation of why we
1641
# aren't using set().intersection() here
1643
keys = [key for key in keys if key in nodes]
1644
1488
if self.reference_lists:
1489
for key in keys.intersection(self._keys):
1490
node = self._nodes[key]
1647
1491
if not node[0]:
1648
1492
yield self, key, node[2], node[1]
1494
for key in keys.intersection(self._keys):
1495
node = self._nodes[key]
1652
1496
if not node[0]:
1653
1497
yield self, key, node[2]