Reports

When using @lru_cache avoid returning mutable, i.e. use tuples instead of lists.
E.g.


from functools import lru_cache

@lru_cache
def fib_lru(n):
    # Fibonacci sequence
    if n < 2:
        res = [1, 1]
    else:
        res = fib_lru(n - 1)
        res.append(res[-1] + res[-2])
    return res

fib_lru(3)   # [1, 1, 2, 3]
fib_lru(9)   # [1, 1, 2, 3, 5, 8, 13, 21, 34, 55]
fib_lru(3)   # [1, 1, 2, 3, 5, 8, 13, 21, 34, 55]    ! oops!  previous state returned :(
fib_lru(11)  # [1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144]
fib_lru(9)   # [1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144]   ! oops!  previous state returned :(


# fix it by replacing lists with tuples
@lru_cache
def fib_lru_t(n):
    if n < 2:
        res = (1, 1)
    else:
        res = fib_lru_t(n - 1)
        res = *res, res[-1] + res[-2]
    return res

fib_lru_t(3)   # (1, 1, 2, 3)
fib_lru_t(9)   # (1, 1, 2, 3, 5, 8, 13, 21, 34, 55)
fib_lru_t(3)   # (1, 1, 2, 3)    OK!
fib_lru_t(11)  # (1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144)
fib_lru_t(9)   # (1, 1, 2, 3, 5, 8, 13, 21, 34, 55)    OK!

Btw using `lru_cache` makes a real difference:

%timeit fib(100)        
# 9.49  μs  ± 151 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)
%timeit fib_lru_t(100)  
# 50.1 _ns_ ± 0.464 ns per loop (mean ± std. dev. of 7 runs, 10,000,000 loops each)

79816008