Support WASI poll_oneoff for clock events

[clarkmcc]

Is your feature request related to a problem? Please describe.
Related to #271. The following code in Rust causes the following panic.

std::thread::sleep(time::Duration::from_secs(10));

thread '<unnamed>' panicked at 'thread::sleep(): unexpected result of poll_oneoff', library/std/src/sys/wasi/thread.rs:57:22

Describe the solution you'd like
Support for the poll_oneoff WASI method.

[codefromthecrypt]

excellent. I'll help with this unless you are game.

[codefromthecrypt]

It appears this is also used in zig, though not sure the code and compilation instructions that generates the binding https://github.com/ziglang/zig/blob/0e6285c8fc31ff866df96847fe34e660da38b4a9/lib/std/time.zig#L24-L47

[codefromthecrypt]

haha our own @mathetake implemented this in tinygo tinygo-org/tinygo#1373 (comment) https://github.com/tinygo-org/tinygo/blob/cf0b7edc78e42038a0bb522b3f1a5b76928e730e/src/runtime/runtime_wasm_wasi.go#L72-L94

[mathetake]

lol I completely forgot about that :))

[codefromthecrypt]

@clarkmcc as there's more than one way to do it, if you have a moment, can you make a scratch repo to reproduce that error? I used wasm32-wasi locally and it failed but invisibly, as opposed to the error you found. I want to make sure I'm doing the same so I can sort it. Main thing is I want to iterate on this, and firstly solve the thread.sleep thing which is reasonably common across compilers.

[clarkmcc]

@codefromthecrypt here's a reproduction https://github.com/clarkmcc/wazero-issue-625. I believe the key is to pipe stdout and stderr so that the Rust panic gets passed back to Go.

https://github.com/clarkmcc/wazero-issue-625/blob/f36171ac363dafa997109f179a3226c0339caf1f/main.go#L28-L29

[codefromthecrypt]

thanks, @clarkmcc!

self-note: here's the wasi-libc call site for sleep, though there are other uses of poll_oneoff https://github.com/WebAssembly/wasi-libc/blob/659ff414560721b1660a19685110e484a081c3d4/libc-bottom-half/cloudlibc/src/libc/time/clock_nanosleep.c

[codefromthecrypt]

I did some research and the best way to incrementally implement this is to cheat similarly to how wasmtime and wasi-libc do, where they hook into sleep when the poll event is a clock. ex bytecodealliance/wasmtime#2753

[codefromthecrypt]

by cheat I mean use one sleeper for both supported clocks (wall and nano) due to an interpretation of "clock_settime" not affecting relative time https://linux.die.net/man/3/clock_settime

[codefromthecrypt]

#629 works, but I need to backfill tests and check other impls

[codefromthecrypt]

Notes about go-generated wasm, doesn't use wasi, rather implements sleep with nanotime.

---

GOOS=js GOARCH=wasm compiled wasm uses nanotime as that's what time.Sleep does internally (among atomic ops that aren't really atomic in go-generated wasm).

Ex. the following only uses nanotime when compiled with GOOS=js GOARCH=wasm

package main

import "time"

func main() {
	time.Sleep(time.Second * 10)
}

related, GOOS=js GOARCH=wasm uses notetsleepg which calls both nanotime and scheduleTimeoutEvent. I'm putting timeout events out of scope even though there's some future need in wasi-clocks

cc @prep

[codefromthecrypt]

verified tinygo uses what we think it does

[codefromthecrypt]

verified zig thanks to @mathetake for the help gettin ziggy witit

$ zig build-exe -O ReleaseSmall -target wasm32-wasi sleep.zig

const std = @import("std");
pub fn main() !void {
    std.time.sleep(std.time.ns_per_s * 5);
}

[codefromthecrypt]

thanks for the patience. #629 should sort this out and also do so incrementally because implementing all of poll_oneoff would be quite a task both for the coder and reviewer!