Inlining call frames in BEFORE_WITH/BEFORE_ASYNC_WITH

[Fidget-Spinner]

Recently I noticed a lot of use patterns like this (especially in aiohttp):

async def f():
    async with req.get():
        ...

I wonder if we should try making __aenter__ and __aexit__ stackless in the BEFORE_ASYNC_WITH instruction and its non asnyc counterpart BEFORE_WITH. This will work similarly to __getitem__ in BINARY_SUBSCR. This will likely show zero speedup in pyperformance, but if may move something like the aiohttp benchmark in Pyston.

Asking around before I invest time into this because it may not be worth the effort.

[markshannon]

I doubt it is worth it adding the complexity.

I'd rather break apart BEFORE_ASYNC_WITH and BEFORE_WITH into lower level instructions:

COPY 1
LOAD_SPECIAL __enter__
SWAP 2
LOAD_SPECIAL __exit__
SWAP 2
CALL 0