Any async delay operation while an aioquic.client object exists makes it silently fail about aioquic

[ravenblackx]

Ah, apparently this is already fixed, I just needed to be on the new version!

[ravenblackx]

Ah, no, it's not fixed, I just mistakenly did (echo && sleep 1 && echo) | h3_request in a context where h3_request took more than a second to start up so there ended up being no async delay.

[jlaine]

I suggest you take a closer look at the http3_client.py example from which your code seems derived. The issue here has nothing to do with sleeping, your client is not sending its HTTP request and eventually times out.

You're missing the call to self.transmit() to actually send the bytes out to the network:

aioquic/examples/http3_client.py

Line 240 in 20cd4ee

self.transmit()

See:

https://aioquic.readthedocs.io/en/latest/asyncio.html#aioquic.asyncio.QuicConnectionProtocol.transmit

[ravenblackx]

In that case isn't there just the opposite bug - if you remove the short sleep it does send the request and get the response even though self.transmit is not called?

[jlaine]

In that case isn't there just the opposite bug - if you remove the short sleep it does send the request and get the response even though self.transmit is not called?

The behaviour is just not defined if you don't call transmit() yourself after queuing data: transmit() can be triggered by receiving data from the server or a timer going off.

[ravenblackx]

Thanks for the explanation, and sorry for the incorrect noise!

[jlaine]

It's not noise, the feedback is welcome. It made me realise that even though the quic (sans IO) layer is reasonably explicit about how it should be used, the asyncio layer is not, much less the interaction with the h3 layer. I think I'll open an issue on this. Would you mind reviewing a proposal for additional docs?

[ravenblackx]

Cool, improved docs are always good. I would suggest that documenting it in transmit might not be the most helpful though - a mention in each of sendData, sendHeaders etc. would be more discoverable. (The problem for me was that I didn't even know the transmit function existed.)