lru_cache is a very useful method but it does not work well with coroutines since they can only be executed once.

Take for example, the attached code (test-case.py) - It will throw a RuntimeError because you cannot reuse an already awaited coroutine.

A solution would be to call `asyncio.ensure_future` on the result of the coroutine if detected.

A coroutine detection is a relatively slow check.
I don't think we need to do it in `functools.lru_cache`.

There is a specialized asyncio compatible version: https://github.com/aio-libs/async_lru
Please use it.

Making this a feature request.

Brett please elaborate.
Do you want to incorporate async_lru library into CPython Core?

I was just saying that this is an enhancement request, no judgment about whether we want to solve the enhancement request.

> A coroutine detection is a relatively slow check.
> I don't think we need to do it in `functools.lru_cache`.

Wouldn't a coroutine check only happen during decoration time? To successfully solve this easily and efficiently, we only really need to wrap the coroutine with `asyncio.ensure_future` if the decorated function is a coroutine, and it will only happen when a result comes back from the decorated function which would have minimal impact.

Of course, I don't know much about the internals of `lru_cache` so my assumptions could be wrong. I should familiar myself with the implementation and figure out how doable it would be.