[Python-ideas] Using functools.lru_cache only on some arguments of a function

[Chris Angelico]

On Sat, Dec 5, 2015 at 8:44 AM, Bill Winslow <bunslow at gmail.com> wrote:
> def deterministic_recursive_calculation(input, partial_state=None):
>     condition = do_some_calculations(input)
>     if condition:
>         return deterministic_recursive_calculation(reduced_input,
> some_state)
>
> Basically, in calculating the results of the subproblem, the subproblem can
> be calculated quicker by including/sharing some partial results from the
> superproblem. (Calling the subproblem without the partial state still gives
> the same result, but takes substantially longer.)
>
> I want to memoize this function for obvious reasons, but I need the
> lru_cache to ignore the partial_state argument, for its value does not
> affect the output, only the computation expense.

Coming right back to the beginning here...

Is there a reason the some_state argument is being passed around,
instead of being global? If the result of the function depends only on
reduced_input and not on some_state, then wouldn't it be possible to
make use of state from an entirely separate call, not just the current
one? (And if you _can't_ make use of some_state from an unrelated
call, then it _does_ affect the call, and it needs to be incorporated
into the cache key.) ISTM this should be exactly as global as the
cache itself.

ChrisA