New server arch

[kazu-yamamoto]

This implements #127

[kazu-yamamoto]

Test fails sometime.
This failure should be fixed with #102.

[kazu-yamamoto]

@khibino Even with this PR, bowline can work well.

[edsko]

If I update my dependencies to use this new version of http2, without changing anything else, my test suite fails. I will investigate why.

[edsko]

All my tests simply deadlock. I will try to figure out where and why.

[kazu-yamamoto]

Thank you but I don't want you to take time to understand the deadlock at this moment.
I know the current code is unstable a bit.
I should at least resolve #102.
For this purpose, I'm now labeling all threads so that listThreads gives me much information.

[edsko]

Ok, then I shall wait :) FWIW, it seems that the client can connect to the server just fine, and the server can respond, and the client gets the initial response headers, but then the (streaming) response body is never started, and the client is left waiting.

[kazu-yamamoto]

If you run h2c-client and h2c-server, they list up the current threads with their status per second. :)

[kazu-yamamoto]

@edsko It would be nice if you tested the current main.

[edsko]

@kazu-yamamoto I just tried with 80de8db , same result as with aa6f979 . My test setup is that the client does a streaming request to the server; this request arrives, and the serves initiates a streaming response. The client receives the headers, but then nothing. If I insert a call to putStrLn in outBodyPush at

http2/Network/HTTP2/Server/Worker.hs

Line 150 in 80de8db

, outBodyPush = \b -> do

then I can see that print statement executed with the data that the client should receive, but the client does not receive anything after the response headers. There is a Wireshark log at well-typed/grapesy#183 where we're tracking updating to the new http2 architecture.

[kazu-yamamoto]

@edsko I cannot reproduce your issue.
If you kindly inserted several putStrLns into Sender.hs, you could see what's happening.