1
# Copyright (C) 2006, 2008, 2009 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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
"""A simple least-recently-used (LRU) cache."""
25
class _LRUNode(object):
26
"""This maintains the linked-list which is the lru internals."""
28
__slots__ = ('prev', 'next_key', 'key', 'value', 'cleanup', 'size')
30
def __init__(self, key, value, cleanup=None):
32
self.next_key = _null_key
35
self.cleanup = cleanup
36
# TODO: We could compute this 'on-the-fly' like we used to, and remove
37
# one pointer from this object, we just need to decide if it
38
# actually costs us much of anything in normal usage
45
prev_key = self.prev.key
46
return '%s(%r n:%r p:%r)' % (self.__class__.__name__, self.key,
47
self.next_key, prev_key)
49
def run_cleanup(self):
51
if self.cleanup is not None:
52
self.cleanup(self.key, self.value)
54
# cleanup might raise an exception, but we want to make sure
55
# to break refcycles, etc
60
class LRUCache(object):
61
"""A class which manages a cache of entries, removing unused ones."""
63
def __init__(self, max_cache=100, after_cleanup_count=None):
65
# The "HEAD" of the lru linked list
66
self._most_recently_used = None
67
# The "TAIL" of the lru linked list
68
self._least_recently_used = None
69
self._update_max_cache(max_cache, after_cleanup_count)
71
def __contains__(self, key):
72
return key in self._cache
74
def __getitem__(self, key):
77
# Inlined from _record_access to decrease the overhead of __getitem__
78
# We also have more knowledge about structure if __getitem__ is
79
# succeeding, then we know that self._most_recently_used must not be
81
mru = self._most_recently_used
83
# Nothing to do, this node is already at the head of the queue
85
# Remove this node from the old location
87
next_key = node.next_key
88
# benchmarking shows that the lookup of _null_key in globals is faster
89
# than the attribute lookup for (node is self._least_recently_used)
90
if next_key is _null_key:
91
# 'node' is the _least_recently_used, because it doesn't have a
92
# 'next' item. So move the current lru to the previous node.
93
self._least_recently_used = node_prev
95
node_next = cache[next_key]
96
node_next.prev = node_prev
97
node_prev.next_key = next_key
98
# Insert this node at the front of the list
99
node.next_key = mru.key
101
self._most_recently_used = node
106
return len(self._cache)
109
"""Walk the LRU list, only meant to be used in tests."""
110
node = self._most_recently_used
112
if node.prev is not None:
113
raise AssertionError('the _most_recently_used entry is not'
114
' supposed to have a previous entry'
116
while node is not None:
117
if node.next_key is _null_key:
118
if node is not self._least_recently_used:
119
raise AssertionError('only the last node should have'
120
' no next value: %s' % (node,))
123
node_next = self._cache[node.next_key]
124
if node_next.prev is not node:
125
raise AssertionError('inconsistency found, node.next.prev'
126
' != node: %s' % (node,))
127
if node.prev is None:
128
if node is not self._most_recently_used:
129
raise AssertionError('only the _most_recently_used should'
130
' not have a previous node: %s'
133
if node.prev.next_key != node.key:
134
raise AssertionError('inconsistency found, node.prev.next'
135
' != node: %s' % (node,))
139
def add(self, key, value, cleanup=None):
140
"""Add a new value to the cache.
142
Also, if the entry is ever removed from the cache, call
145
:param key: The key to store it under
146
:param value: The object to store
147
:param cleanup: None or a function taking (key, value) to indicate
148
'value' should be cleaned up.
151
raise ValueError('cannot use _null_key as a key')
152
if key in self._cache:
153
node = self._cache[key]
157
# Maintain the LRU properties, even if cleanup raises an
160
node.cleanup = cleanup
161
self._record_access(node)
163
node = _LRUNode(key, value, cleanup=cleanup)
164
self._cache[key] = node
165
self._record_access(node)
167
if len(self._cache) > self._max_cache:
168
# Trigger the cleanup
171
def cache_size(self):
172
"""Get the number of entries we will cache."""
173
return self._max_cache
175
def get(self, key, default=None):
176
node = self._cache.get(key, None)
179
self._record_access(node)
183
"""Get the list of keys currently cached.
185
Note that values returned here may not be available by the time you
186
request them later. This is simply meant as a peak into the current
189
:return: An unordered list of keys that are currently cached.
191
return self._cache.keys()
194
"""Get the key:value pairs as a dict."""
195
return dict((k, n.value) for k, n in self._cache.iteritems())
198
"""Clear the cache until it shrinks to the requested size.
200
This does not completely wipe the cache, just makes sure it is under
201
the after_cleanup_count.
203
# Make sure the cache is shrunk to the correct size
204
while len(self._cache) > self._after_cleanup_count:
207
def __setitem__(self, key, value):
208
"""Add a value to the cache, there will be no cleanup function."""
209
self.add(key, value, cleanup=None)
211
def _record_access(self, node):
212
"""Record that key was accessed."""
213
# Move 'node' to the front of the queue
214
if self._most_recently_used is None:
215
self._most_recently_used = node
216
self._least_recently_used = node
218
elif node is self._most_recently_used:
219
# Nothing to do, this node is already at the head of the queue
221
# We've taken care of the tail pointer, remove the node, and insert it
224
if node is self._least_recently_used:
225
self._least_recently_used = node.prev
226
if node.prev is not None:
227
node.prev.next_key = node.next_key
228
if node.next_key is not _null_key:
229
node_next = self._cache[node.next_key]
230
node_next.prev = node.prev
232
node.next_key = self._most_recently_used.key
233
self._most_recently_used.prev = node
234
self._most_recently_used = node
237
def _remove_node(self, node):
238
if node is self._least_recently_used:
239
self._least_recently_used = node.prev
240
self._cache.pop(node.key)
241
# If we have removed all entries, remove the head pointer as well
242
if self._least_recently_used is None:
243
self._most_recently_used = None
247
# cleanup might raise an exception, but we want to make sure to
248
# maintain the linked list
249
if node.prev is not None:
250
node.prev.next_key = node.next_key
251
if node.next_key is not _null_key:
252
node_next = self._cache[node.next_key]
253
node_next.prev = node.prev
254
# And remove this node's pointers
256
node.next_key = _null_key
258
def _remove_lru(self):
259
"""Remove one entry from the lru, and handle consequences.
261
If there are no more references to the lru, then this entry should be
262
removed from the cache.
264
self._remove_node(self._least_recently_used)
267
"""Clear out all of the cache."""
268
# Clean up in LRU order
272
def resize(self, max_cache, after_cleanup_count=None):
273
"""Change the number of entries that will be cached."""
274
self._update_max_cache(max_cache,
275
after_cleanup_count=after_cleanup_count)
277
def _update_max_cache(self, max_cache, after_cleanup_count=None):
278
self._max_cache = max_cache
279
if after_cleanup_count is None:
280
self._after_cleanup_count = self._max_cache * 8 / 10
282
self._after_cleanup_count = min(after_cleanup_count,
287
class LRUSizeCache(LRUCache):
288
"""An LRUCache that removes things based on the size of the values.
290
This differs in that it doesn't care how many actual items there are,
291
it just restricts the cache to be cleaned up after so much data is stored.
293
The size of items added will be computed using compute_size(value), which
294
defaults to len() if not supplied.
297
def __init__(self, max_size=1024*1024, after_cleanup_size=None,
299
"""Create a new LRUSizeCache.
301
:param max_size: The max number of bytes to store before we start
302
clearing out entries.
303
:param after_cleanup_size: After cleaning up, shrink everything to this
305
:param compute_size: A function to compute the size of the values. We
306
use a function here, so that you can pass 'len' if you are just
307
using simple strings, or a more complex function if you are using
308
something like a list of strings, or even a custom object.
309
The function should take the form "compute_size(value) => integer".
310
If not supplied, it defaults to 'len()'
313
self._compute_size = compute_size
314
if compute_size is None:
315
self._compute_size = len
316
self._update_max_size(max_size, after_cleanup_size=after_cleanup_size)
317
LRUCache.__init__(self, max_cache=max(int(max_size/512), 1))
319
def add(self, key, value, cleanup=None):
320
"""Add a new value to the cache.
322
Also, if the entry is ever removed from the cache, call
325
:param key: The key to store it under
326
:param value: The object to store
327
:param cleanup: None or a function taking (key, value) to indicate
328
'value' should be cleaned up.
331
raise ValueError('cannot use _null_key as a key')
332
node = self._cache.get(key, None)
333
value_len = self._compute_size(value)
334
if value_len >= self._after_cleanup_size:
335
# The new value is 'too big to fit', as it would fill up/overflow
336
# the cache all by itself
337
trace.mutter('Adding the key %r to an LRUSizeCache failed.'
338
' value %d is too big to fit in a the cache'
339
' with size %d %d', key, value_len,
340
self._after_cleanup_size, self._max_size)
342
# We won't be replacing the old node, so just remove it
343
self._remove_node(node)
344
if cleanup is not None:
348
node = _LRUNode(key, value, cleanup=cleanup)
349
self._cache[key] = node
351
self._value_size -= node.size
352
node.size = value_len
353
self._value_size += value_len
354
self._record_access(node)
356
if self._value_size > self._max_size:
361
"""Clear the cache until it shrinks to the requested size.
363
This does not completely wipe the cache, just makes sure it is under
364
the after_cleanup_size.
366
# Make sure the cache is shrunk to the correct size
367
while self._value_size > self._after_cleanup_size:
370
def _remove_node(self, node):
371
self._value_size -= node.size
372
LRUCache._remove_node(self, node)
374
def resize(self, max_size, after_cleanup_size=None):
375
"""Change the number of bytes that will be cached."""
376
self._update_max_size(max_size, after_cleanup_size=after_cleanup_size)
377
max_cache = max(int(max_size/512), 1)
378
self._update_max_cache(max_cache)
380
def _update_max_size(self, max_size, after_cleanup_size=None):
381
self._max_size = max_size
382
if after_cleanup_size is None:
383
self._after_cleanup_size = self._max_size * 8 / 10
385
self._after_cleanup_size = min(after_cleanup_size, self._max_size)
added
removed
bzrlib/lru_cache.py
