asyncio generic fails when the buffer is larger than the file contents

[thelvyn]

asyncio generic fails when the buffer is larger than the file contents.

Description

1. testasyncio was updated so it fails when the error occurs using asyncio generic (works fine with liburing, not tested on windows)
2. the fix is trivial: if the read output is <= to the read request size, then we return SDL_ASYNCIO_COMPLETE. This seems to match the expected behavior.

[icculus]

A short read would be an error, if it isn't EOF, though. Let me look at this more closely before we merge.

[thelvyn]

The problem I encountered was that the IO remained in READY state after the read instead of EOF (even if EOF was reached).

My use case is to reuse buffers and don't query the file size before reading, that's usual with primitives like read. My logic is based on the fact that transferred_bytes < buffer_size or transferred_bytes==0 is considered as EOF so I'm done.

I agree that it is not strictly correct in case of spurious failures, I should wait for a EOF flag anyway 👍
An incomplete read is not necessarily a failure or a EOF, just that a second read should be performed to get more data or EOF (or a EAGAIN I suppose).

I can investigate this further if you want, or leave it to you.