Implement poll_oneoff in WASI P1

[andreaTP]

Zig, the Go standard library, Rust(in some cases) and c2w require poll_oneoff.
It would be great to have a stable implementation of it.

Spike ref:
https://github.com/dylibso/chicory/pull/530/files#diff-9037d84891e8043e756bfd63e2101686b67d1cad2cc2c4af3b2e6b48675acdf3R1086

[evacchi]

I want to take a look at this one soon

[electrum]

I started looking at this a while ago, but it didn't seem possible to run the WASI testsuite tests due to them using stdin/stdout/stderr which are fundamentally blocking in our implementation as we use InputStream/OutputStream. Would it suffice to simply implement the timer part and return ENOTSUP for other descriptors?

[electrum]

I found that various library functions in Python such as the socket library will fail due to poll_oneoff (even though they would eventually fail with a different error as WASI doesn't support creating sockets).

[evacchi]

Would it suffice to simply implement the timer part and return ENOTSUP for other descriptors?

pretty much. Also, stdin kinda wants to be treated differently, so in wazero we special case that. Basically I've been meaning to port over the logic here https://github.com/tetratelabs/wazero/blob/610c202ec48f3a7c729f2bf11707330127ab3689/imports/wasi_snapshot_preview1/poll.go#L98-L197

but you can get away with less than that for a lot of binaries; a lot of the nonblocking stuff is related to how the Go compiler (not tinygo) wants to treat I/O (it sets the nonblocking flag everywhere)

[andreaTP]

Would it suffice to simply implement the timer part and return ENOTSUP for other descriptors?

Very likely, from what I've seen the Go and Zig code using it are fine with the timer part, just remember to throw meaningful exceptions in case is not so that we can eventually track other usages.

[electrum]

That code is polling for data on stdin. We could do a sleep loop while calling available().

[evacchi]

that might work