Re: [PATCH] http: handle both "h2" and "h2h3" in curl info lines

From: Jeff King <peff@peff.net>
Date: Jun 17, 2023, 05:45 UTC
Message-ID: <20230617054519.GF562686@coredump.intra.peff.net>
In-Reply-To: <20230617051559.GD562686@coredump.intra.peff.net>

On Sat, Jun 17, 2023 at 01:15:59AM -0400, Jeff King wrote:

> So here's that patch with a commit message, though note that t5559.30
> seems to reliably fail with a timeout for me on the new version of curl. :(

OK, looks like the timeout problem is due to another change in curl, but it has since been fixed. Here are my notes in case anyone else runs into it (or has to investigate something similar).

My bisection recipe in curl.git is:

  git clean -dx
  autoreconf -fi &&
  ./configure --disable-shared --prefix=/tmp/foo --with-openssl --with-nghttp2 &&
  make install &&
  (cd /path/to/git.git &&
   make CURL_CONFIG=/tmp/foo/bin/curl-config &&
   cd t &&
   ./t5559-http-fetch-smart-http2.sh -v)

You'll need apache with mod_h2 installed so that t5559 can run the server side of the test.

That bisects to 744dcf22f (http2: flow control and buffer improvements, 2023-03-30), which seems like a plausible candidate. Before that commit, t5559 runs to completion. After, t5559.30, which tries to send a large-ish input to the server, hangs for a while and then we eventually get a 408 timeout error.

Reverse-bisecting from 744dcf22f up to the current tip of "master", it looks like the problem goes away in 5c58cb021 (http2: fix EOF handling on uploads with auth negotiation, 2023-05-24), which likewise makes sense. :)

So I think we can ignore that from the git side, and the patch I posted earlier is what we'd want to fix the h2/h2h3 confusion.

-Peff