wasmtime / PR #14908 serve: accept HTTP/2 connections · git-wasmtime

For something that's roughly intended to be a debugging-ish server this feels like wasmtime serve is starting to sprawl quite a lot -- for example custom I/O implementations, hardcoded HTTP prefixes and such, this all feels pretty intense for what is otherwise intended to be a relatively simple wrapper around hyper and wasmtime together. Is it possible to simplify this and/or externalize the complexity? For example I see that hyper-util has utilities for handling either 1.1 or 2 connections, that might be appropriate?

Also, is this valid? I thought that http/2 required TLS and AFAIK we don't have any configuration to have wasmtime serve serve over TLS.

Wasmtime GitHub notifications bot (Oct 21 2026 at 16:37):

xia-chao commented on PR #14908:

i'm pretty sure h2c is valid ,rfc 9113 has prior knowledge for cleartext, http/2 doesn't need tls. ran curl --http2-prior-knowledge and got a 200.

Wasmtime GitHub notifications bot (Oct 21 2026 at 16:47):

xia-chao commented on PR #14908:

For example I see that hyper-util has utilities for handling either 1.1 or 2 connections, that might be appropriate?

i first tried hyper util, then went dep free (the branch name says nodep). using hyper util would probably be less code to maintain long term ok to add it?

Wasmtime GitHub notifications bot (Oct 22 2026 at 19:31):

fitzgen removed the review request for fitzgen from PR #14908.

Wasmtime GitHub notifications bot (Oct 26 2026 at 14:35):

alexcrichton commented on PR #14908:

Subjectively I'd say yeah pulling in something like hyper-util for this would be reasonable.

Wasmtime GitHub notifications bot (Oct 27 2026 at 10:14):

xia-chao updated PR #14908.

Wasmtime GitHub notifications bot (Oct 27 2026 at 10:19):

xia-chao commented on PR #14908:

switched to hyper util's auto builder and dropped the preface sniffing.

Wasmtime GitHub notifications bot (Oct 27 2026 at 16:20):

:memo: dicej submitted PR review.

Wasmtime GitHub notifications bot (Oct 27 2026 at 16:20):

:speech_balloon: dicej created PR review comment:

Would it make sense to use the hyper_util::server::conn::auto::Builder for both cases, conditionally calling Builder::http1_only in the debug case? Seems like that would reduce duplication a bit.

Wasmtime GitHub notifications bot (Oct 28 2026 at 11:37):

xia-chao updated PR #14908.

Wasmtime GitHub notifications bot (Oct 28 2026 at 11:44):

:memo: xia-chao submitted PR review.

Wasmtime GitHub notifications bot (Oct 28 2026 at 11:44):

:speech_balloon: xia-chao created PR review comment:

auto needs a 'static service and the debug store is borrowed, so that path can't use it.

Wasmtime GitHub notifications bot (Oct 28 2026 at 13:16):

:thumbs_up: dicej submitted PR review:

Thanks, @xia-chao!

Wasmtime GitHub notifications bot (Oct 28 2026 at 13:16):

dicej added PR #14908 serve: accept HTTP/2 connections to the merge queue.

Wasmtime GitHub notifications bot (Oct 28 2026 at 13:39):

:check: dicej merged PR #14908.

Wasmtime GitHub notifications bot (Oct 28 2026 at 13:39):

dicej removed PR #14908 serve: accept HTTP/2 connections from the merge queue.

Last updated: Oct 31 2026 at 23:03 UTC