Permalink
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
736 lines (653 sloc) 26.4 KB
# Copyright (c) 2002-2010 Zooko "Zooko" Wilcox-O'Hearn
"""
This module offers three implementations of an LRUCache, which is a dict that
drops items according to a Least-Recently-Used policy if the dict exceeds a
fixed maximum size.
Warning: if -O optimizations are not turned on then LRUCache performs
extensive self-analysis in every function call, which can take minutes
and minutes for a large cache. Turn on -O, or comment out ``assert self._assert_invariants()``
"""
import operator
from assertutil import _assert, precondition
from humanreadable import hr
class LRUCache:
"""
An efficient least-recently-used cache. It keeps an LRU queue, and when
the number of items in the cache reaches maxsize, it removes the least
recently used item.
"Looking" at an item, key, or value such as with "has_key()" makes that
item become the most recently used item.
You can also use "refresh()" to explicitly make an item become the most
recently used item.
Adding an item that is already in the dict *does* make it the most-
recently-used item although it does not change the state of the dict
itself.
See also SmallLRUCache (below), which is faster in some cases.
"""
class ItemIterator:
def __init__(self, c):
self.c = c
self.i = c.d[c.hs][2]
def __iter__(self):
return self
def next(self):
if self.i is self.c.ts:
raise StopIteration
k = self.i
precondition(self.c.d.has_key(k), "The iterated LRUCache doesn't have the next key. Most likely this is because someone altered the contents of the LRUCache while the iteration was in progress.", k, self.c)
(v, p, n,) = self.c.d[k]
self.i = n
return (k, v,)
class KeyIterator:
def __init__(self, c):
self.c = c
self.i = c.d[c.hs][2]
def __iter__(self):
return self
def next(self):
if self.i is self.c.ts:
raise StopIteration
k = self.i
precondition(self.c.d.has_key(k), "The iterated LRUCache doesn't have the next key. Most likely this is because someone altered the contents of the LRUCache while the iteration was in progress.", k, self.c)
(v, p, n,) = self.c.d[k]
self.i = n
return k
class ValIterator:
def __init__(self, c):
self.c = c
self.i = c.d[c.hs][2]
def __iter__(self):
return self
def next(self):
if self.i is self.c.ts:
raise StopIteration
precondition(self.c.d.has_key(self.i), "The iterated LRUCache doesn't have the next key. Most likely this is because someone altered the contents of the LRUCache while the iteration was in progress.", self.i, self.c)
(v, p, n,) = self.c.d[self.i]
self.i = n
return v
class Sentinel:
def __init__(self, msg):
self.msg = msg
def __repr__(self):
return "<%s %s>" % (self.__class__.__name__, self.msg,)
def __init__(self, initialdata={}, maxsize=128):
precondition(maxsize > 0)
self.m = maxsize+2 # The +2 is for the head and tail nodes.
self.d = {} # k: k, v: [v, prev, next,] # the dict
self.hs = LRUCache.Sentinel("hs")
self.ts = LRUCache.Sentinel("ts")
self.d[self.hs] = [None, self.hs, self.ts,] # This allows us to use sentinels as normal nodes.
self.d[self.ts] = [None, self.hs, self.ts,] # This allows us to use sentinels as normal nodes.
self.update(initialdata)
assert self._assert_invariants()
def __repr_n__(self, n=None):
s = ["{",]
try:
iter = self.iteritems()
x = iter.next()
s.append(str(x[0])); s.append(": "); s.append(str(x[1]))
i = 1
while (n is None) or (i < n):
x = iter.next()
s.append(", "); s.append(str(x[0])); s.append(": "); s.append(str(x[1]))
except StopIteration:
pass
s.append("}")
return ''.join(s)
def __repr__(self):
return "<%s %s>" % (self.__class__.__name__, self.__repr_n__(),)
def __str__(self):
return "<%s %s>" % (self.__class__.__name__, self.__repr_n__(16),)
def _assert_invariants(self):
_assert(len(self.d) <= self.m, "Size is required to be <= maxsize.", len(self.d), self.m)
_assert((len(self.d) > 2) == (self.d[self.hs][2] is not self.ts) == (self.d[self.ts][1] is not self.hs), "Head and tail point to something other than each other if and only if there is at least one element in the dictionary.", self.hs, self.ts, len(self.d))
foundprevsentinel = 0
foundnextsentinel = 0
for (k, (v, p, n,)) in self.d.iteritems():
_assert(v not in (self.hs, self.ts,))
_assert(p is not self.ts, "A reference to the tail sentinel may not appear in prev.", k, v, p, n)
_assert(n is not self.hs, "A reference to the head sentinel may not appear in next.", k, v, p, n)
_assert(p in self.d, "Each prev is required to appear as a key in the dict.", k, v, p, n)
_assert(n in self.d, "Each next is required to appear as a key in the dict.", k, v, p, n)
if p is self.hs:
foundprevsentinel += 1
_assert(foundprevsentinel <= 2, "No more than two references to the head sentinel may appear as a prev.", k, v, p, n)
if n is self.ts:
foundnextsentinel += 1
_assert(foundnextsentinel <= 2, "No more than one reference to the tail sentinel may appear as a next.", k, v, p, n)
_assert(foundprevsentinel == 2, "A reference to the head sentinel is required appear as a prev (plus a self-referential reference).")
_assert(foundnextsentinel == 2, "A reference to the tail sentinel is required appear as a next (plus a self-referential reference).")
count = 0
for (k, v,) in self.iteritems():
_assert(k not in (self.hs, self.ts,))
count += 1
_assert(count == len(self.d)-2, count, len(self.d)) # -2 for the sentinels
return True
def freshen(self, k, strictkey=False):
assert self._assert_invariants()
if not self.d.has_key(k):
if strictkey:
raise KeyError, k
return
node = self.d[k]
# relink
self.d[node[1]][2] = node[2]
self.d[node[2]][1] = node[1]
# move to front
hnode = self.d[self.hs]
node[1] = self.hs
node[2] = hnode[2]
hnode[2] = k
self.d[node[2]][1] = k
assert self._assert_invariants()
def iteritems(self):
return LRUCache.ItemIterator(self)
def itervalues(self):
return LRUCache.ValIterator(self)
def iterkeys(self):
return self.__iter__()
def __iter__(self):
return LRUCache.KeyIterator(self)
def __getitem__(self, key, default=None, strictkey=True):
node = self.d.get(key)
if not node:
if strictkey:
raise KeyError, key
return default
self.freshen(key)
return node[0]
def __setitem__(self, k, v=None):
assert self._assert_invariants()
node = self.d.get(k)
if node:
node[0] = v
self.freshen(k)
return
if len(self.d) == self.m:
# If this insert is going to increase the size of the cache to
# bigger than maxsize.
self.pop()
hnode = self.d[self.hs]
n = hnode[2]
self.d[k] = [v, self.hs, n,]
hnode[2] = k
self.d[n][1] = k
assert self._assert_invariants()
return v
def __delitem__(self, key, default=None, strictkey=True):
"""
@param strictkey: True if you want a KeyError in the case that
key is not there, False if you want a reference to default
in the case that key is not there
@param default: the object to return if key is not there; This
is ignored if strictkey.
@return: the value removed or default if there is not item by
that key and strictkey is False
"""
assert self._assert_invariants()
if self.d.has_key(key):
node = self.d[key]
# relink
self.d[node[1]][2] = node[2]
self.d[node[2]][1] = node[1]
del self.d[key]
assert self._assert_invariants()
return node[0]
elif strictkey:
assert self._assert_invariants()
raise KeyError, key
else:
assert self._assert_invariants()
return default
def has_key(self, key):
assert self._assert_invariants()
if self.d.has_key(key):
self.freshen(key)
assert self._assert_invariants()
return True
else:
assert self._assert_invariants()
return False
def clear(self):
assert self._assert_invariants()
self.d.clear()
self.d[self.hs] = [None, self.hs, self.ts,] # This allows us to use sentinels as normal nodes.
self.d[self.ts] = [None, self.hs, self.ts,] # This allows us to use sentinels as normal nodes.
assert self._assert_invariants()
def update(self, otherdict):
"""
@return: self
"""
assert self._assert_invariants()
if len(otherdict) >= (self.m-2): # -2 for the sentinel nodes
# optimization
self.clear()
assert self._assert_invariants()
i = otherdict.iteritems()
try:
while len(self.d) < self.m:
(k, v,) = i.next()
assert self._assert_invariants()
self[k] = v
assert self._assert_invariants()
return self
except StopIteration:
_assert(False, "Internal error -- this should never have happened since the while loop should have terminated first.")
return self
for (k, v,) in otherdict.iteritems():
assert self._assert_invariants()
self[k] = v
assert self._assert_invariants()
def pop(self):
assert self._assert_invariants()
if len(self.d) < 2: # the +2 is for the sentinels
raise KeyError, 'popitem(): dictionary is empty'
k = self.d[self.ts][1]
self.remove(k)
assert self._assert_invariants()
return k
def popitem(self):
assert self._assert_invariants()
if len(self.d) < 2: # the +2 is for the sentinels
raise KeyError, 'popitem(): dictionary is empty'
k = self.d[self.ts][1]
val = self.remove(k)
assert self._assert_invariants()
return (k, val,)
def keys_unsorted(self):
assert self._assert_invariants()
t = self.d.copy()
del t[self.hs]
del t[self.ts]
assert self._assert_invariants()
return t.keys()
def keys(self):
res = [None] * len(self)
i = 0
for k in self.iterkeys():
res[i] = k
i += 1
return res
def values_unsorted(self):
assert self._assert_invariants()
t = self.d.copy()
del t[self.hs]
del t[self.ts]
assert self._assert_invariants()
return map(operator.__getitem__, t.values(), [0]*len(t))
def values(self):
res = [None] * len(self)
i = 0
for v in self.itervalues():
res[i] = v
i += 1
return res
def items(self):
res = [None] * len(self)
i = 0
for it in self.iteritems():
res[i] = it
i += 1
return res
def __len__(self):
return len(self.d) - 2
def insert(self, key, val=None):
assert self._assert_invariants()
result = self.__setitem__(key, val)
assert self._assert_invariants()
return result
def setdefault(self, key, default=None):
assert self._assert_invariants()
if not self.has_key(key):
self[key] = default
assert self._assert_invariants()
return self[key]
def get(self, key, default=None):
return self.__getitem__(key, default, strictkey=False)
def remove(self, key, default=None, strictkey=True):
assert self._assert_invariants()
result = self.__delitem__(key, default, strictkey)
assert self._assert_invariants()
return result
class SmallLRUCache(dict):
"""
SmallLRUCache is faster than LRUCache for small sets. How small? That
depends on your machine and which operations you use most often. Use
performance profiling to determine whether the cache class that you are
using makes any difference to the performance of your program, and if it
does, then run "quick_bench()" in test/test_cache.py to see which cache
implementation is faster for the size of your datasets.
A simple least-recently-used cache. It keeps an LRU queue, and
when the number of items in the cache reaches maxsize, it removes
the least recently used item.
"Looking" at an item or a key such as with "has_key()" makes that
item become the most recently used item.
You can also use "refresh()" to explicitly make an item become the most
recently used item.
Adding an item that is already in the dict *does* make it the
most- recently-used item although it does not change the state of
the dict itself.
"""
class ItemIterator:
def __init__(self, c):
self.c = c
self.i = 0
def __iter__(self):
return self
def next(self):
precondition(self.i <= len(self.c._lru), "The iterated SmallLRUCache doesn't have this many elements. Most likely this is because someone altered the contents of the LRUCache while the iteration was in progress.", self.i, self.c)
precondition(dict.has_key(self.c, self.c._lru[self.i]), "The iterated SmallLRUCache doesn't have this key. Most likely this is because someone altered the contents of the LRUCache while the iteration was in progress.", self.i, self.c._lru[self.i], self.c)
if self.i == len(self.c._lru):
raise StopIteration
k = self.i
self.i += 1
return (k, dict.__getitem__(self.c, k),)
class KeyIterator:
def __init__(self, c):
self.c = c
self.i = 0
def __iter__(self):
return self
def next(self):
precondition(self.i <= len(self.c._lru), "The iterated SmallLRUCache doesn't have this many elements. Most likely this is because someone altered the contents of the LRUCache while the iteration was in progress.", self.i, self.c)
precondition(dict.has_key(self.c, self.c._lru[self.i]), "The iterated SmallLRUCache doesn't have this key. Most likely this is because someone altered the contents of the LRUCache while the iteration was in progress.", self.i, self.c._lru[self.i], self.c)
if self.i == len(self.c._lru):
raise StopIteration
k = self.i
self.i += 1
return k
class ValueIterator:
def __init__(self, c):
self.c = c
self.i = 0
def __iter__(self):
return self
def next(self):
precondition(self.i <= len(self.c._lru), "The iterated SmallLRUCache doesn't have this many elements. Most likely this is because someone altered the contents of the LRUCache while the iteration was in progress.", self.i, self.c)
precondition(dict.has_key(self.c, self.c._lru[self.i]), "The iterated SmallLRUCache doesn't have this key. Most likely this is because someone altered the contents of the LRUCache while the iteration was in progress.", self.i, self.c._lru[self.i], self.c)
if self.i == len(self.c._lru):
raise StopIteration
k = self.i
self.i += 1
return dict.__getitem__(self.c, k)
def __init__(self, initialdata={}, maxsize=128):
dict.__init__(self, initialdata)
self._lru = initialdata.keys() # contains keys
self._maxsize = maxsize
over = len(self) - self._maxsize
if over > 0:
map(dict.__delitem__, [self]*over, self._lru[:over])
del self._lru[:over]
assert self._assert_invariants()
def _assert_invariants(self):
_assert(len(self._lru) <= self._maxsize, "Size is required to be <= maxsize.")
_assert(len(filter(lambda x: dict.has_key(self, x), self._lru)) == len(self._lru), "Each key in self._lru is required to be in dict.", filter(lambda x: not dict.has_key(self, x), self._lru), len(self._lru), self._lru, len(self), self)
_assert(len(filter(lambda x: x in self._lru, self.keys())) == len(self), "Each key in dict is required to be in self._lru.", filter(lambda x: x not in self._lru, self.keys()), len(self._lru), self._lru, len(self), self)
_assert(len(self._lru) == len(self), "internal consistency", filter(lambda x: x not in self.keys(), self._lru), len(self._lru), self._lru, len(self), self)
_assert(len(self._lru) <= self._maxsize, "internal consistency", len(self._lru), self._lru, self._maxsize)
return True
def insert(self, key, item=None):
assert self._assert_invariants()
result = self.__setitem__(key, item)
assert self._assert_invariants()
return result
def setdefault(self, key, default=None):
assert self._assert_invariants()
if not self.has_key(key):
self[key] = default
assert self._assert_invariants()
return self[key]
def __setitem__(self, key, item=None):
assert self._assert_invariants()
if dict.has_key(self, key):
self._lru.remove(key)
else:
if len(self._lru) == self._maxsize:
# If this insert is going to increase the size of the cache to bigger than maxsize:
killkey = self._lru.pop(0)
dict.__delitem__(self, killkey)
dict.__setitem__(self, key, item)
self._lru.append(key)
assert self._assert_invariants()
return item
def remove(self, key, default=None, strictkey=True):
assert self._assert_invariants()
result = self.__delitem__(key, default, strictkey)
assert self._assert_invariants()
return result
def __delitem__(self, key, default=None, strictkey=True):
"""
@param strictkey: True if you want a KeyError in the case that
key is not there, False if you want a reference to default
in the case that key is not there
@param default: the object to return if key is not there; This
is ignored if strictkey.
@return: the object removed or default if there is not item by
that key and strictkey is False
"""
assert self._assert_invariants()
if dict.has_key(self, key):
val = dict.__getitem__(self, key)
dict.__delitem__(self, key)
self._lru.remove(key)
assert self._assert_invariants()
return val
elif strictkey:
assert self._assert_invariants()
raise KeyError, key
else:
assert self._assert_invariants()
return default
def clear(self):
assert self._assert_invariants()
dict.clear(self)
self._lru = []
assert self._assert_invariants()
def update(self, otherdict):
"""
@return: self
"""
assert self._assert_invariants()
if len(otherdict) > self._maxsize:
# Handling this special case here makes it possible to implement the
# other more common cases faster below.
dict.clear(self)
self._lru = []
if self._maxsize > (len(otherdict) - self._maxsize):
dict.update(self, otherdict)
while len(self) > self._maxsize:
dict.popitem(self)
else:
for k, v, in otherdict.iteritems():
if len(self) == self._maxsize:
break
dict.__setitem__(self, k, v)
self._lru = dict.keys(self)
assert self._assert_invariants()
return self
for k in otherdict.iterkeys():
if dict.has_key(self, k):
self._lru.remove(k)
self._lru.extend(otherdict.keys())
dict.update(self, otherdict)
over = len(self) - self._maxsize
if over > 0:
map(dict.__delitem__, [self]*over, self._lru[:over])
del self._lru[:over]
assert self._assert_invariants()
return self
def has_key(self, key):
assert self._assert_invariants()
if dict.has_key(self, key):
assert key in self._lru, "key: %s, self._lru: %s" % tuple(map(hr, (key, self._lru,)))
self._lru.remove(key)
self._lru.append(key)
assert self._assert_invariants()
return True
else:
assert self._assert_invariants()
return False
def refresh(self, key, strictkey=True):
"""
@param strictkey: raise a KeyError exception if key isn't present
"""
assert self._assert_invariants()
if not dict.has_key(self, key):
if strictkey:
raise KeyError, key
return
self._lru.remove(key)
self._lru.append(key)
def popitem(self):
if not self._lru:
raise KeyError, 'popitem(): dictionary is empty'
k = self._lru[-1]
obj = self.remove(k)
return (k, obj,)
class LinkedListLRUCache:
"""
This is slower and less featureful than LRUCache. It is included
here for comparison purposes.
Implementation of a length-limited O(1) LRU queue.
Built for and used by PyPE:
http://pype.sourceforge.net
original Copyright 2003 Josiah Carlson.
useful methods and _assert_invariant added by Zooko for testing and benchmarking purposes
"""
class Node:
def __init__(self, prev, me):
self.prev = prev
self.me = me
self.next = None
def __init__(self, initialdata={}, maxsize=128):
self._maxsize = max(maxsize, 1)
self.d = {}
self.first = None
self.last = None
for key, value in initialdata.iteritems():
self[key] = value
def clear(self):
self.d = {}
self.first = None
self.last = None
def update(self, otherdict):
for (k, v,) in otherdict.iteritems():
self[k] = v
def setdefault(self, key, default=None):
if not self.has_key(key):
self[key] = default
return self[key]
def _assert_invariants(self):
def lliterkeys(self):
cur = self.first
while cur != None:
cur2 = cur.next
yield cur.me[0]
cur = cur2
def lllen(self):
# Ugh.
acc = 0
for x in lliterkeys(self):
acc += 1
return acc
def llhaskey(self, key):
# Ugh.
for x in lliterkeys(self):
if x is key:
return True
return False
for k in lliterkeys(self):
_assert(self.d.has_key(k), "Each key in the linked list is required to be in the dict.", k)
for k in self.d.iterkeys():
_assert(llhaskey(self, k), "Each key in the dict is required to be in the linked list.", k)
_assert(lllen(self) == len(self.d), "internal consistency", self, self.d)
_assert(len(self.d) <= self._maxsize, "Size is required to be <= maxsize.")
return True
def __contains__(self, obj):
return obj in self.d
def has_key(self, key):
return self.__contains__(key)
def __getitem__(self, obj):
a = self.d[obj].me
self[a[0]] = a[1]
return a[1]
def get(self, key, default=None, strictkey=False):
if not self.has_key(key) and strictkey:
raise KeyError, key
if self.has_key(key):
return self.__getitem__(key)
else:
return default
def __setitem__(self, obj, val):
if obj in self.d:
del self[obj]
nobj = self.Node(self.last, (obj, val))
if self.first is None:
self.first = nobj
if self.last:
self.last.next = nobj
self.last = nobj
self.d[obj] = nobj
if len(self.d) > self._maxsize:
if self.first == self.last:
self.first = None
self.last = None
return
a = self.first
a.next.prev = None
self.first = a.next
a.next = None
del self.d[a.me[0]]
del a
def insert(self, key, item=None):
return self.__setitem__(key, item)
def __delitem__(self, obj, default=None, strictkey=True):
if self.d.has_key(obj):
nobj = self.d[obj]
if nobj.prev:
nobj.prev.next = nobj.next
else:
self.first = nobj.next
if nobj.next:
nobj.next.prev = nobj.prev
else:
self.last = nobj.prev
val = self.d[obj]
del self.d[obj]
return val.me[1]
elif strictkey:
raise KeyError, obj
else:
return default
def remove(self, obj, default=None, strictkey=True):
return self.__delitem__(obj, default=default, strictkey=strictkey)
def __iter__(self):
cur = self.first
while cur != None:
cur2 = cur.next
yield cur.me[1]
cur = cur2
def iteritems(self):
cur = self.first
while cur != None:
cur2 = cur.next
yield cur.me
cur = cur2
def iterkeys(self):
return iter(self.d)
def itervalues(self):
for i,j in self.iteritems():
yield j
def values(self):
l = []
for v in self.itervalues():
l.append(v)
return l
def keys(self):
return self.d.keys()
def __len__(self):
return self.d.__len__()
def popitem(self):
i = self.last.me
obj = self.remove(i[0])
return obj