Full-featured O(1) LRU cache backported from Python3.3. The full Py3.3 API is supported (thread safety, maxsize, keyword args, type checking, __wrapped__, and cache_info). Includes Py3.3 optimizations for better memory utilization, fewer dependencies, and fewer dict lookups.
Download
Copy to clipboard1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 | from collections import namedtuple
from functools import update_wrapper
from threading import RLock
_CacheInfo = namedtuple("CacheInfo", ["hits", "misses", "maxsize", "currsize"])
class _HashedSeq(list):
__slots__ = 'hashvalue'
def __init__(self, tup, hash=hash):
self[:] = tup
self.hashvalue = hash(tup)
def __hash__(self):
return self.hashvalue
def _make_key(args, kwds, typed,
kwd_mark = (object(),),
fasttypes = {int, str, frozenset, type(None)},
sorted=sorted, tuple=tuple, type=type, len=len):
'Make a cache key from optionally typed positional and keyword arguments'
key = args
if kwds:
sorted_items = sorted(kwds.items())
key += kwd_mark
for item in sorted_items:
key += item
if typed:
key += tuple(type(v) for v in args)
if kwds:
key += tuple(type(v) for k, v in sorted_items)
elif len(key) == 1 and type(key[0]) in fasttypes:
return key[0]
return _HashedSeq(key)
def lru_cache(maxsize=100, typed=False):
"""Least-recently-used cache decorator.
If *maxsize* is set to None, the LRU features are disabled and the cache
can grow without bound.
If *typed* is True, arguments of different types will be cached separately.
For example, f(3.0) and f(3) will be treated as distinct calls with
distinct results.
Arguments to the cached function must be hashable.
View the cache statistics named tuple (hits, misses, maxsize, currsize) with
f.cache_info(). Clear the cache and statistics with f.cache_clear().
Access the underlying function with f.__wrapped__.
See: http://en.wikipedia.org/wiki/Cache_algorithms#Least_Recently_Used
"""
# Users should only access the lru_cache through its public API:
# cache_info, cache_clear, and f.__wrapped__
# The internals of the lru_cache are encapsulated for thread safety and
# to allow the implementation to change (including a possible C version).
def decorating_function(user_function):
cache = dict()
stats = [0, 0] # make statistics updateable non-locally
HITS, MISSES = 0, 1 # names for the stats fields
make_key = _make_key
cache_get = cache.get # bound method to lookup key or return None
_len = len # localize the global len() function
lock = RLock() # because linkedlist updates aren't threadsafe
root = [] # root of the circular doubly linked list
root[:] = [root, root, None, None] # initialize by pointing to self
nonlocal_root = [root] # make updateable non-locally
PREV, NEXT, KEY, RESULT = 0, 1, 2, 3 # names for the link fields
if maxsize == 0:
def wrapper(*args, **kwds):
# no caching, just do a statistics update after a successful call
result = user_function(*args, **kwds)
stats[MISSES] += 1
return result
elif maxsize is None:
def wrapper(*args, **kwds):
# simple caching without ordering or size limit
key = make_key(args, kwds, typed)
result = cache_get(key, root) # root used here as a unique not-found sentinel
if result is not root:
stats[HITS] += 1
return result
result = user_function(*args, **kwds)
cache[key] = result
stats[MISSES] += 1
return result
else:
def wrapper(*args, **kwds):
# size limited caching that tracks accesses by recency
key = make_key(args, kwds, typed) if kwds or typed else args
with lock:
link = cache_get(key)
if link is not None:
# record recent use of the key by moving it to the front of the list
root, = nonlocal_root
link_prev, link_next, key, result = link
link_prev[NEXT] = link_next
link_next[PREV] = link_prev
last = root[PREV]
last[NEXT] = root[PREV] = link
link[PREV] = last
link[NEXT] = root
stats[HITS] += 1
return result
result = user_function(*args, **kwds)
with lock:
root, = nonlocal_root
if key in cache:
# getting here means that this same key was added to the
# cache while the lock was released. since the link
# update is already done, we need only return the
# computed result and update the count of misses.
pass
elif _len(cache) >= maxsize:
# use the old root to store the new key and result
oldroot = root
oldroot[KEY] = key
oldroot[RESULT] = result
# empty the oldest link and make it the new root
root = nonlocal_root[0] = oldroot[NEXT]
oldkey = root[KEY]
oldvalue = root[RESULT]
root[KEY] = root[RESULT] = None
# now update the cache dictionary for the new links
del cache[oldkey]
cache[key] = oldroot
else:
# put result in a new link at the front of the list
last = root[PREV]
link = [last, root, key, result]
last[NEXT] = root[PREV] = cache[key] = link
stats[MISSES] += 1
return result
def cache_info():
"""Report cache statistics"""
with lock:
return _CacheInfo(stats[HITS], stats[MISSES], maxsize, len(cache))
def cache_clear():
"""Clear the cache and cache statistics"""
with lock:
cache.clear()
root = nonlocal_root[0]
root[:] = [root, root, None, None]
stats[:] = [0, 0]
wrapper.__wrapped__ = user_function
wrapper.cache_info = cache_info
wrapper.cache_clear = cache_clear
return update_wrapper(wrapper, user_function)
return decorating_function
|
Documentation and examples: http://docs.python.org/dev/library/functools.html#functools.lru_cache
The Python3.3 update added the typed option for differentiating equal arguments differing types. This makes it possible to cache different results for f(1) and for f(1.0).
Python3.3 also added a __wrapped__ attribute that can be used for bypassing or replacing the cache.
The Python 3.3 version was also refactored, beautified, and optimized. It consumes less memory than the 3.2 version. It makes half as many hash calls. And, it inlines the relevant parts of OrderedDict while eliminating unused functionality. It relies exclusively on Python basics such as 1) plain python lists to implement a circular doubly linked list, 2) a plain python dictionary, and 3) a plain tuple used for storing the positional arguments, sorted keyword arguments, and optional type information.
The backport works on Python2.6 or later, Python3.0 or later, Python with Psyco, and with PyPy.
The backport entailed replacing Python 3's nonlocal keyword with nested scope lookup of a mutable list (for the hit/miss statistics). Also,the __wrapped__ attribute was added manually (because older versions functools.wraps didn't support it). The original 3.3 is at: http://hg.python.org/cpython/file/97b0cf9df420/Lib/functools.py#l139
To backport to even earlier versions of Python, replace the named tuple with an equivalent handmade class and replace the with-statement with an equivalent try/finally.
No print-statements were harmed during this backport ;-)
Just in case someone wants a space-based version (i.e.: keys are not pruned based on the number of keys, but based on the total space hold in the cache given a function which calculates the space), there's a version below (and note that I did remove some things I didn't want for speed and simplicity: it has no statistics, it's not thread-safe, must have a max size, doesn't accept kwargs)
line 133: "oldvalue = root[RESULT]" seems unnecessary?
To keep compatibility with python 2.6, you need to replace
with
I think the statement for 2.6 compatibility is:
fasttypes=set([int, str, frozenset, type(None)]),I've packaged the above code, unchanged, as backports.functools_lru_cache: https://pypi.python.org/pypi/backports.functools_lru_cache/1.0