Migrate current wasi-sockets TCP impl to worker thread

[manekinekko]

• ✅preview2_tcp_bind
• ✅ preview2_tcp_sockopts
• ✅ preview2_tcp_states
• ✅ preview2_tcp_connect
• ❌ preview2_tcp_sample_application

Closes #154

[manekinekko]

Currently, running cargo test preview2_tcp_sample_application will block and timeout on preview2_tcp_sample_application.rs#L33:

let data = input.blocking_read(first_message.len() as u64).unwrap();

@guybedford any idea what could trigger a timeout on stream.read() call?

[guybedford]

@manekinekko one common need we have when creating streams is to ensure that the end event triggers the readable event, since we always block on the readable event when reading streams. See eg https://github.com/bytecodealliance/jco/blob/main/packages/preview2-shim/lib/io/worker-thread.js#L432. That's the first thing that comes to mind here, although it could well be something else.

[squillace]

@guybedford @manekinekko if you think I'm letting you get through the day without closing this, you're nuts.

[manekinekko]

Thanks @guybedford I am gonna try that.

[manekinekko]

@squillace we are working on it. We just need one more day :P

[manekinekko]

so, the stream.on("end", () => void stream.emit("readable")); workaround didn't work. I am investigating others options. cc @guybedford

[squillace]

https://www.youtube.com/watch?v=zKVq-P3z5Vg&t=1s.

[manekinekko]

OMG! Now I am listening to Phil Collins all day long!

[squillace]

https://tenor.com/T9CZ.gif

[manekinekko]

@guybedford after investigation, it seems that this promise never resolve, which then make the call to block. Any clue how to debug this?

jco/packages/preview2-shim/lib/io/worker-thread.js

Lines 485 to 487 in 05b016b

	unfinishedPolls.get(
	handle(INPUT_STREAM_SUBSCRIBE | (call & CALL_TYPE_MASK), id)
	)

[guybedford]

I'm working on the TCP calls now, starting with connect streams. Should have a listen / accept flow working early next week, but there's a lot to do here.

[guybedford]

@manekinekko I managed to get basic initial listen and accept supported working in #330. Still a bunch of edge cases to work through, but if you're interested in the implementation approach you can find it there.

[guybedford]

TCP implementation now passing all tests as of #330!