13
13
# You should have received a copy of the GNU General Public License
14
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
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
17
"""A simple least-recently-used (LRU) cache."""
19
from collections import deque
26
class _LRUNode(object):
27
"""This maintains the linked-list which is the lru internals."""
29
__slots__ = ('prev', 'next_key', 'key', 'value', 'cleanup', 'size')
31
def __init__(self, key, value, cleanup=None):
33
self.next_key = _null_key
36
self.cleanup = cleanup
37
# TODO: We could compute this 'on-the-fly' like we used to, and remove
38
# one pointer from this object, we just need to decide if it
39
# actually costs us much of anything in normal usage
46
prev_key = self.prev.key
47
return '%s(%r n:%r p:%r)' % (self.__class__.__name__, self.key,
48
self.next_key, prev_key)
50
def run_cleanup(self):
52
if self.cleanup is not None:
53
self.cleanup(self.key, self.value)
55
# cleanup might raise an exception, but we want to make sure
56
# to break refcycles, etc
23
61
class LRUCache(object):
24
62
"""A class which manages a cache of entries, removing unused ones."""
26
def __init__(self, max_cache=100, after_cleanup_size=None):
27
self._max_cache = max_cache
28
if after_cleanup_size is None:
29
self._after_cleanup_size = self._max_cache
31
self._after_cleanup_size = min(after_cleanup_size, self._max_cache)
33
self._compact_queue_length = 4*self._max_cache
64
def __init__(self, max_cache=100, after_cleanup_count=None,
65
after_cleanup_size=symbol_versioning.DEPRECATED_PARAMETER):
66
if symbol_versioning.deprecated_passed(after_cleanup_size):
67
symbol_versioning.warn('LRUCache.__init__(after_cleanup_size) was'
68
' deprecated in 1.11. Use'
69
' after_cleanup_count instead.',
71
after_cleanup_count = after_cleanup_size
37
self._queue = deque() # Track when things are accessed
38
self._refcount = {} # number of entries in self._queue for each key
73
# The "HEAD" of the lru linked list
74
self._most_recently_used = None
75
# The "TAIL" of the lru linked list
76
self._least_recently_used = None
77
self._update_max_cache(max_cache, after_cleanup_count)
40
79
def __contains__(self, key):
41
80
return key in self._cache
43
82
def __getitem__(self, key):
44
val = self._cache[key]
45
self._record_access(key)
85
# Inlined from _record_access to decrease the overhead of __getitem__
86
# We also have more knowledge about structure if __getitem__ is
87
# succeeding, then we know that self._most_recently_used must not be
89
mru = self._most_recently_used
91
# Nothing to do, this node is already at the head of the queue
93
# Remove this node from the old location
95
next_key = node.next_key
96
# benchmarking shows that the lookup of _null_key in globals is faster
97
# than the attribute lookup for (node is self._least_recently_used)
98
if next_key is _null_key:
99
# 'node' is the _least_recently_used, because it doesn't have a
100
# 'next' item. So move the current lru to the previous node.
101
self._least_recently_used = node_prev
103
node_next = cache[next_key]
104
node_next.prev = node_prev
105
node_prev.next_key = next_key
106
# Insert this node at the front of the list
107
node.next_key = mru.key
109
self._most_recently_used = node
48
113
def __len__(self):
49
114
return len(self._cache)
117
"""Walk the LRU list, only meant to be used in tests."""
118
node = self._most_recently_used
120
if node.prev is not None:
121
raise AssertionError('the _most_recently_used entry is not'
122
' supposed to have a previous entry'
124
while node is not None:
125
if node.next_key is _null_key:
126
if node is not self._least_recently_used:
127
raise AssertionError('only the last node should have'
128
' no next value: %s' % (node,))
131
node_next = self._cache[node.next_key]
132
if node_next.prev is not node:
133
raise AssertionError('inconsistency found, node.next.prev'
134
' != node: %s' % (node,))
135
if node.prev is None:
136
if node is not self._most_recently_used:
137
raise AssertionError('only the _most_recently_used should'
138
' not have a previous node: %s'
141
if node.prev.next_key != node.key:
142
raise AssertionError('inconsistency found, node.prev.next'
143
' != node: %s' % (node,))
51
147
def add(self, key, value, cleanup=None):
52
148
"""Add a new value to the cache.
54
Also, if the entry is ever removed from the queue, call cleanup.
55
Passing it the key and value being removed.
150
Also, if the entry is ever removed from the cache, call
57
153
:param key: The key to store it under
58
154
:param value: The object to store
59
155
:param cleanup: None or a function taking (key, value) to indicate
60
'value' sohuld be cleaned up.
156
'value' should be cleaned up.
159
raise ValueError('cannot use _null_key as a key')
62
160
if key in self._cache:
64
self._cache[key] = value
65
self._cleanup[key] = cleanup
66
self._record_access(key)
161
node = self._cache[key]
165
# Maintain the LRU properties, even if cleanup raises an
168
node.cleanup = cleanup
169
self._record_access(node)
171
node = _LRUNode(key, value, cleanup=cleanup)
172
self._cache[key] = node
173
self._record_access(node)
68
175
if len(self._cache) > self._max_cache:
69
176
# Trigger the cleanup
179
def cache_size(self):
180
"""Get the number of entries we will cache."""
181
return self._max_cache
72
183
def get(self, key, default=None):
73
if key in self._cache:
184
node = self._cache.get(key, None)
187
self._record_access(node)
191
"""Get the list of keys currently cached.
193
Note that values returned here may not be available by the time you
194
request them later. This is simply meant as a peak into the current
197
:return: An unordered list of keys that are currently cached.
199
return self._cache.keys()
202
"""Get the key:value pairs as a dict."""
203
return dict((k, n.value) for k, n in self._cache.iteritems())
77
205
def cleanup(self):
78
206
"""Clear the cache until it shrinks to the requested size.
80
208
This does not completely wipe the cache, just makes sure it is under
81
the after_cleanup_size.
209
the after_cleanup_count.
83
211
# Make sure the cache is shrunk to the correct size
84
while len(self._cache) > self._after_cleanup_size:
212
while len(self._cache) > self._after_cleanup_count:
85
213
self._remove_lru()
87
215
def __setitem__(self, key, value):
88
216
"""Add a value to the cache, there will be no cleanup function."""
89
217
self.add(key, value, cleanup=None)
91
def _record_access(self, key):
219
def _record_access(self, node):
92
220
"""Record that key was accessed."""
93
self._queue.append(key)
94
# Can't use setdefault because you can't += 1 the result
95
self._refcount[key] = self._refcount.get(key, 0) + 1
97
# If our access queue is too large, clean it up too
98
if len(self._queue) > self._compact_queue_length:
101
def _compact_queue(self):
102
"""Compact the queue, leaving things in sorted last appended order."""
104
for item in self._queue:
105
if self._refcount[item] == 1:
106
new_queue.append(item)
108
self._refcount[item] -= 1
109
self._queue = new_queue
110
# All entries should be of the same size. There should be one entry in
111
# queue for each entry in cache, and all refcounts should == 1
112
if not (len(self._queue) == len(self._cache) ==
113
len(self._refcount) == sum(self._refcount.itervalues())):
114
raise AssertionError()
116
def _remove(self, key):
117
"""Remove an entry, making sure to maintain the invariants."""
118
cleanup = self._cleanup.pop(key)
119
val = self._cache.pop(key)
120
if cleanup is not None:
221
# Move 'node' to the front of the queue
222
if self._most_recently_used is None:
223
self._most_recently_used = node
224
self._least_recently_used = node
226
elif node is self._most_recently_used:
227
# Nothing to do, this node is already at the head of the queue
229
# We've taken care of the tail pointer, remove the node, and insert it
232
if node is self._least_recently_used:
233
self._least_recently_used = node.prev
234
if node.prev is not None:
235
node.prev.next_key = node.next_key
236
if node.next_key is not _null_key:
237
node_next = self._cache[node.next_key]
238
node_next.prev = node.prev
240
node.next_key = self._most_recently_used.key
241
self._most_recently_used.prev = node
242
self._most_recently_used = node
245
def _remove_node(self, node):
246
if node is self._least_recently_used:
247
self._least_recently_used = node.prev
248
self._cache.pop(node.key)
249
# If we have removed all entries, remove the head pointer as well
250
if self._least_recently_used is None:
251
self._most_recently_used = None
255
# cleanup might raise an exception, but we want to make sure to
256
# maintain the linked list
257
if node.prev is not None:
258
node.prev.next_key = node.next_key
259
if node.next_key is not _null_key:
260
node_next = self._cache[node.next_key]
261
node_next.prev = node.prev
262
# And remove this node's pointers
264
node.next_key = _null_key
124
266
def _remove_lru(self):
125
267
"""Remove one entry from the lru, and handle consequences.
164
317
The function should take the form "compute_size(value) => integer".
165
318
If not supplied, it defaults to 'len()'
167
# This approximates that texts are > 0.5k in size. It only really
168
# effects when we clean up the queue, so we don't want it to be too
170
LRUCache.__init__(self, max_cache=int(max_size/512))
171
self._max_size = max_size
172
if after_cleanup_size is None:
173
self._after_cleanup_size = self._max_size
175
self._after_cleanup_size = min(after_cleanup_size, self._max_size)
177
320
self._value_size = 0
178
321
self._compute_size = compute_size
179
322
if compute_size is None:
180
323
self._compute_size = len
324
self._update_max_size(max_size, after_cleanup_size=after_cleanup_size)
325
LRUCache.__init__(self, max_cache=max(int(max_size/512), 1))
182
327
def add(self, key, value, cleanup=None):
183
328
"""Add a new value to the cache.
185
Also, if the entry is ever removed from the queue, call cleanup.
186
Passing it the key and value being removed.
330
Also, if the entry is ever removed from the cache, call
188
333
:param key: The key to store it under
189
334
:param value: The object to store
190
335
:param cleanup: None or a function taking (key, value) to indicate
191
'value' sohuld be cleaned up.
336
'value' should be cleaned up.
193
if key in self._cache:
339
raise ValueError('cannot use _null_key as a key')
340
node = self._cache.get(key, None)
195
341
value_len = self._compute_size(value)
196
342
if value_len >= self._after_cleanup_size:
343
# The new value is 'too big to fit', as it would fill up/overflow
344
# the cache all by itself
345
trace.mutter('Adding the key %r to an LRUSizeCache failed.'
346
' value %d is too big to fit in a the cache'
347
' with size %d %d', key, value_len,
348
self._after_cleanup_size, self._max_size)
350
# We won't be replacing the old node, so just remove it
351
self._remove_node(node)
352
if cleanup is not None:
356
node = _LRUNode(key, value, cleanup=cleanup)
357
self._cache[key] = node
359
self._value_size -= node.size
360
node.size = value_len
198
361
self._value_size += value_len
199
self._cache[key] = value
200
self._cleanup[key] = cleanup
201
self._record_access(key)
362
self._record_access(node)
203
364
if self._value_size > self._max_size:
204
365
# Time to cleanup
added
removed
bzrlib/lru_cache.py
