After reading the code of https://github.com/python/cpython/blob/master/Lib/functools.py I thought that lru_cache use hash to build a key from the function arguments, so if I two object have the same hash they should be the same for lru_cache.
If you run this code:
from functools import lru_cache
COUNT=0
@lru_cache(maxsize=None)
def fnc(*args, **kvargs):
global COUNT
COUNT=COUNT+1
return COUNT, hash(args[0]), args ,kvargs
class MyClass:
def __init__(self, *args, **kvargs):
self._init_args=(args, frozenset(kvargs.items()))
def __hash__(self):
return hash(self._init_args)
m1a = MyClass(1)
m2a = MyClass(2)
m1b = MyClass(1)
m2b = MyClass(2)
fnc(m1a) # Output: (1, -7270110455953140331, (<__main__.MyClass object at 0x7fb14b540710>,), {})
fnc(m1a) # Output: (1, -7270110455953140331, (<__main__.MyClass object at 0x7fb14b540710>,), {})
fnc(m2a) # Output: (2, 7567087542259278010, (<__main__.MyClass object at 0x7fb14b540810>,), {})
fnc(m2a) # Output: (2, 7567087542259278010, (<__main__.MyClass object at 0x7fb14b540810>,), {})
fnc(m1b) # Output: (3, -7270110455953140331, (<__main__.MyClass object at 0x7fb14b540610>,), {})
fnc(m2b) # Output: (4, 7567087542259278010, (<__main__.MyClass object at 0x7fb14b540b50>,), {})
you can see that lru_cache is detecting that m1a and m1b are different even they have the same hash.
What can I do in order that lru_cache doesn't differentiate between two instances of MyClass with same __init__ arguments?
lru_cache uses __hash__ at first then it checks __eq__
Add __eq__ method to your class
def __eq__(self, other):
return type(self) == type(other) and self._init_args == other._init_args