1
# Copyright (C) 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Indexing facilities."""
26
from cStringIO import StringIO
29
from bzrlib import errors
31
_OPTION_NODE_REFS = "node_ref_lists="
32
_SIGNATURE = "Bazaar Graph Index 1\n"
35
_whitespace_re = re.compile('[\t\n\x0b\x0c\r\x00 ]')
36
_newline_null_re = re.compile('[\n\0]')
39
class GraphIndexBuilder(object):
40
"""A builder that can build a GraphIndex.
42
The resulting graph has the structure:
44
_SIGNATURE OPTIONS NODES NEWLINE
45
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
46
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
48
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
49
KEY := Not-whitespace-utf8
51
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
52
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
53
REFERENCE := DIGITS ; digits is the byte offset in the index of the
55
VALUE := no-newline-no-null-bytes
58
def __init__(self, reference_lists=0):
59
"""Create a GraphIndex builder.
61
:param reference_lists: The number of node references lists for each
64
self.reference_lists = reference_lists
67
def add_node(self, key, references, value):
68
"""Add a node to the index.
70
:param key: The key. keys must be whitespace-free utf8.
71
:param references: An iterable of iterables of keys. Each is a
72
reference to another key.
73
:param value: The value to associate with the key. It may be any
74
bytes as long as it does not contain \0 or \n.
76
if not key or _whitespace_re.search(key) is not None:
77
raise errors.BadIndexKey(key)
78
if _newline_null_re.search(value) is not None:
79
raise errors.BadIndexValue(value)
80
if len(references) != self.reference_lists:
81
raise errors.BadIndexValue(references)
83
for reference_list in references:
84
for reference in reference_list:
85
if _whitespace_re.search(reference) is not None:
86
raise errors.BadIndexKey(reference)
87
if reference not in self._nodes:
88
self._nodes[reference] = ('a', (), '')
89
node_refs.append(tuple(reference_list))
90
if key in self._nodes and self._nodes[key][0] == '':
91
raise errors.BadIndexDuplicateKey(key, self)
92
self._nodes[key] = ('', tuple(node_refs), value)
96
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
97
prefix_length = len(lines[0]) + len(lines[1])
98
# references are byte offsets. To avoid having to do nasty
99
# polynomial work to resolve offsets (references to later in the
100
# file cannot be determined until all the inbetween references have
101
# been calculated too) we pad the offsets with 0's to make them be
102
# of consistent length. Using binary offsets would break the trivial
104
# to calculate the width of zero's needed we do three passes:
105
# one to gather all the non-reference data and the number of references.
106
# one to pad all the data with reference-length and determine entry
110
# forward sorted by key. In future we may consider topological sorting,
111
# at the cost of table scans for direct lookup, or a second index for
113
nodes = sorted(self._nodes.items())
114
# if we do not prepass, we don't know how long it will be up front.
115
expected_bytes = None
116
# we only need to pre-pass if we have reference lists at all.
117
if self.reference_lists:
119
non_ref_bytes = prefix_length
121
# TODO use simple multiplication for the constants in this loop.
122
for key, (absent, references, value) in nodes:
123
# record the offset known *so far* for this key:
124
# the non reference bytes to date, and the total references to
125
# date - saves reaccumulating on the second pass
126
key_offset_info.append((key, non_ref_bytes, total_references))
127
# key is literal, value is literal, there are 3 null's, 1 NL
128
non_ref_bytes += len(key) + len(value) + 3 + 1
129
# one byte for absent if set.
132
elif self.reference_lists:
133
# (ref_lists -1) tabs
134
non_ref_bytes += self.reference_lists - 1
135
# (ref-1 cr's per ref_list)
136
for ref_list in references:
137
# how many references across the whole file?
138
total_references += len(ref_list)
139
# accrue reference separators
141
non_ref_bytes += len(ref_list) - 1
142
# how many digits are needed to represent the total byte count?
144
possible_total_bytes = non_ref_bytes + total_references*digits
145
while 10 ** digits < possible_total_bytes:
147
possible_total_bytes = non_ref_bytes + total_references*digits
148
expected_bytes = possible_total_bytes + 1 # terminating newline
149
# resolve key addresses.
151
for key, non_ref_bytes, total_references in key_offset_info:
152
key_addresses[key] = non_ref_bytes + total_references*digits
154
format_string = '%%0%sd' % digits
155
for key, (absent, references, value) in nodes:
156
flattened_references = []
157
for ref_list in references:
159
for reference in ref_list:
160
ref_addresses.append(format_string % key_addresses[reference])
161
flattened_references.append('\r'.join(ref_addresses))
162
lines.append("%s\0%s\0%s\0%s\n" % (key, absent,
163
'\t'.join(flattened_references), value))
165
result = StringIO(''.join(lines))
166
if expected_bytes and len(result.getvalue()) != expected_bytes:
167
raise errors.BzrError('Failed index creation. Internal error:'
168
' mismatched output length and expected length: %d %d' %
169
(len(result.getvalue()), expected_bytes))
170
return StringIO(''.join(lines))
173
class GraphIndex(object):
174
"""An index for data with embedded graphs.
176
The index maps keys to a list of key reference lists, and a value.
177
Each node has the same number of key reference lists. Each key reference
178
list can be empty or an arbitrary length. The value is an opaque NULL
179
terminated string without any newlines. The storage of the index is
180
hidden in the interface: keys and key references are always bytestrings,
181
never the internal representation (e.g. dictionary offsets).
183
It is presumed that the index will not be mutated - it is static data.
185
Successive iter_all_entries calls will read the entire index each time.
186
Additionally, iter_entries calls will read the index linearly until the
187
desired keys are found. XXX: This must be fixed before the index is
188
suitable for production use. :XXX
191
def __init__(self, transport, name):
192
"""Open an index called name on transport.
194
:param transport: A bzrlib.transport.Transport.
195
:param name: A path to provide to transport API calls.
197
self._transport = transport
200
def iter_all_entries(self):
201
"""Iterate over all keys within the index.
203
:return: An iterable of (key, reference_lists, value). There is no
204
defined order for the result iteration - it will be in the most
205
efficient order for the index.
207
stream = self._transport.get(self._name)
208
self._read_prefix(stream)
210
self.keys_by_offset = {}
213
for line in stream.readlines():
217
key, absent, references, value = line.split('\0')
218
value = value[:-1] # remove the newline
220
for ref_string in references.split('\t'):
221
ref_lists.append(tuple([
222
int(ref) for ref in ref_string.split('\r') if ref
224
ref_lists = tuple(ref_lists)
225
self.keys_by_offset[pos] = (key, absent, ref_lists, value)
227
for key, absent, references, value in self.keys_by_offset.itervalues():
230
# resolve references:
231
if self.node_ref_lists:
233
for ref_list in references:
234
node_refs.append(tuple([self.keys_by_offset[ref][0] for ref in ref_list]))
235
node_refs = tuple(node_refs)
238
yield (key, node_refs, value)
240
# there must be one line - the empty trailer line.
241
raise errors.BadIndexData(self)
243
def _read_prefix(self, stream):
244
signature = stream.read(len(self._signature()))
245
if not signature == self._signature():
246
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
247
options_line = stream.readline()
248
if not options_line.startswith(_OPTION_NODE_REFS):
249
raise errors.BadIndexOptions(self)
251
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
253
raise errors.BadIndexOptions(self)
255
def iter_entries(self, keys):
256
"""Iterate over keys within the index.
258
:param keys: An iterable providing the keys to be retrieved.
259
:return: An iterable of (key, reference_lists, value). There is no
260
defined order for the result iteration - it will be in the most
261
efficient order for the index.
266
for node in self.iter_all_entries():
273
def _signature(self):
274
"""The file signature for this index type."""
278
"""Validate that everything in the index can be accessed."""
279
# iter_all validates completely at the moment, so just do that.
280
for node in self.iter_all_entries():
284
class CombinedGraphIndex(object):
285
"""A GraphIndex made up from smaller GraphIndices.
287
The backing indices must implement GraphIndex, and are presumed to be
290
Queries against the combined index will be made against the first index,
291
and then the second and so on. The order of index's can thus influence
292
performance significantly. For example, if one index is on local disk and a
293
second on a remote server, the local disk index should be before the other
297
def __init__(self, indices):
298
"""Create a CombinedGraphIndex backed by indices.
300
:param indices: An ordered list of indices to query for data.
302
self._indices = indices
304
def insert_index(self, pos, index):
305
"""Insert a new index in the list of indices to query.
307
:param pos: The position to insert the index.
308
:param index: The index to insert.
310
self._indices.insert(pos, index)
312
def iter_all_entries(self):
313
"""Iterate over all keys within the index
315
Duplicate keys across child indices are presumed to have the same
316
value and are only reported once.
318
:return: An iterable of (key, reference_lists, value). There is no
319
defined order for the result iteration - it will be in the most
320
efficient order for the index.
323
for index in self._indices:
324
for node in index.iter_all_entries():
325
if node[0] not in seen_keys:
327
seen_keys.add(node[0])
329
def iter_entries(self, keys):
330
"""Iterate over keys within the index.
332
Duplicate keys across child indices are presumed to have the same
333
value and are only reported once.
335
:param keys: An iterable providing the keys to be retrieved.
336
:return: An iterable of (key, reference_lists, value). There is no
337
defined order for the result iteration - it will be in the most
338
efficient order for the index.
341
for index in self._indices:
344
for node in index.iter_entries(keys):
349
"""Validate that everything in the index can be accessed."""
350
for index in self._indices:
354
class InMemoryGraphIndex(GraphIndexBuilder):
355
"""A GraphIndex which operates entirely out of memory and is mutable.
357
This is designed to allow the accumulation of GraphIndex entries during a
358
single write operation, where the accumulated entries need to be immediately
359
available - for example via a CombinedGraphIndex.
362
def add_nodes(self, nodes):
363
"""Add nodes to the index.
365
:param nodes: An iterable of (key, node_refs, value) entries to add.
367
for (key, node_refs, value) in nodes:
368
self.add_node(key, node_refs, value)
370
def iter_all_entries(self):
371
"""Iterate over all keys within the index
373
:return: An iterable of (key, reference_lists, value). There is no
374
defined order for the result iteration - it will be in the most
375
efficient order for the index (in this case dictionary hash order).
377
for key, (absent, references, value) in self._nodes.iteritems():
379
yield key, references, value
381
def iter_entries(self, keys):
382
"""Iterate over keys within the index.
384
:param keys: An iterable providing the keys to be retrieved.
385
:return: An iterable of (key, reference_lists, value). There is no
386
defined order for the result iteration - it will be in the most
387
efficient order for the index (keys iteration order in this case).
390
for key in keys.intersection(self._nodes):
391
node = self._nodes[key]
393
yield key, node[1], node[2]
396
"""In memory index's have no known corruption at the moment."""