Support for python's Buffer protocol

[mkzeender]

The save-stating and screenshotting methods should be able to read from and write to any python object that exposes a buffer.

Writing the data directly into a buffer will greatly increase performance, because you won't have to allocate memory for a bytes object, nor will you have to copy the data twice.

The buffer protocol was made an official protocol as of python 3.11, and all major array-like objects support it (including bytes, bytearray, array.array, numpy.array, etc).

[Felk]

Sounds like a great addition. I'll keep it in mind, thanks! Any concrete ideas how the API would look like? E.g. passing in the buffer to write to?

[mkzeender]

I guess numpy-style would be, like,

from collections.abc import Buffer
@overload
def save_to_bytes(out:Buffer)->None:
   ...
def load_from_bytes(in_:Buffer):
   ... # bytes implements Buffer so this signature doesn't need to be overloaded
def mem_size() -> int: ...

For the savestates. Basically, just expanding the existing api to accept any buffer object.

As for the screenshot, I'm not entirely sure what's the best way. Maybe a way to set a global buffer that the frames are dumped into? That would probably suffice. And a way to get the dimensions of the image to set the size of the buffer.

[Felk]

Just throwing this in here: Might be worth to also explore this for the new framedrawn event, as allocating the bytes object containing the image data has noticeable overhead too