bump: github.com/quic-go/quic-go to v0.53.0

[IceCodeNew]

fix #311

[mholt]

Thanks for the PR -- looks like the tests might also need to be updated in order to be passing. I don't have a chance to look into it at the moment but we could probably merge this when they pass.

[IceCodeNew]

I will fix the test this weekend since I’m on holiday. 😊

[jurekl]

Hi,
Is there any serious reason preventing the implementation of this PR?
Thank you and best regards.

[mholt]

Yes, a timeout during matching is a concerning test failure.

[IceCodeNew]

Sorry for taking back my words. Just did not foresee the workloads piling up while I am taking the vacation...
Hopefully, I will get this done this weekend :-(

[IceCodeNew]

I need some help here.
It seems that test cases with ALPN specified (m2, m3) will somehow fail at

caddy-l4/modules/l4quic/matcher.go

Line 147 in 77e8af4

qConn, err = qListener.Accept(qContext)

I tried to put the qListener.Accept() function in a loop, as the documents suggested, but no luck here.

[mholt]

Still a bit swamped here -- I know @marten-seemann is too, but he might be able to help quicker since he understands the updated APIs, which I still haven't had a chance to look into 🙈

[marten-seemann]

What's the failure? All I see is a test timeout.

[marten-seemann]

You might want to try updating in smaller increments. You're going from v0.44.0 to v0.53.0. A lot has changed between these releases, and figuring out which release causes the failure would help us narrow down what could be going wrong here.

[manselmi]

FYI Caddy has since bumped quic-go to v0.54.0.

[manselmi]

Hi, how has this been going?

[jurekl]

I don't think it bothers anyone too much...

[manselmi]

I don't think it bothers anyone too much...

It's unfortunate that I can't use caddy-l4 with caddy master, which has incorporated a fix for #7109: http.request placeholders not resolved in of header directive.

[marten-seemann]

@manselmi This is an open source project, why don’t you contribute the respective code?

[manselmi]

@marten-seemann I would if I knew how. Not a Go dev.

[IceCodeNew]

I am having vacation now. I expect to do the bi-selection in the middle of the month to pinpoint the exact version that breaks the test cases.
Hopefully, I could get some insight into this problem after locating the smelly commit.

[vnxme]

Hi there, I've investigated the issue with l4quic matcher and quic-go dependency, and it has 2 implications:

1. The matcher has a bug when it comes to the moment it tries to close quic.EarlyListener which makes the tests fail if we upgrade dependenices (go v1.24, caddyserver/caddy v2.10.0, quic-go v0.51.1). I will submit a fix and an upgrade PR soon.
2. The quic-go has its syntax changed from v0.51.1 to v0.54.0: quic.EarlyListener.Accept now returns *quic.Conn instead of quic.EarlyConnection. However, we can't update the matcher and modify caddy-l4 go.mod to use the newest quic-go unless the mainline Caddy requiring this quic-go (v0.54.0) is released (since it uses quic.EarlyConnection as well). This prevents automatic builds with Caddy master.

[vnxme]

Actually, we can resolve the issue with building against Caddy master by means of referring to a pseudo-version. Credits to @IceCodeNew for this brilliant idea shown in this PR.