backporting the `gc` feature performance fix to 34 · wasmtime

Stream: wasmtime

Topic: backporting the `gc` feature performance fix to 34

Roman Volosatovs (Jun 19 2025 at 10:36):

Currently, disabling gc via feature flag (or otherwise) may carry pretty significant performance penalty, at least in Wasm modules. I know that 34 release has already been branched off, but is there a chance the fix could be back ported? I am happy to open a PR to the release branch to facilitate that

Here's the PR with more details and the fix (which is tiny): https://github.com/bytecodealliance/wasmtime/pull/11068

fix: do not collect backtrace on each `gc_runtime` access by rvolosatovs · Pull Request #11068 · bytecodealliance/wasmtime

This is something I've discovered profiling a Wasmtime embedding with wasm32-unknown-unknown modules. Currently, gc_runtime accessor returns an anyhow::Result, which triggers a backtrace collec...

Roman Volosatovs (Jun 19 2025 at 10:40):

Filed a backport PR for now: https://github.com/bytecodealliance/wasmtime/pull/11074

[backport] fix: do not collect backtrace on each gc_runtime access by rvolosatovs · Pull Request #11074 · bytecodealliance/wasmtime

This is a backport of #11068 to 34.0.0 release branch. I am not entirely sure what's the standard procedure for backports, but hopefully this helps. Zulip discussion: #wasmtime > backporting...

Alex Crichton (Jun 19 2025 at 20:41):

Backport sounds good to me, and thanks for making the PR!

Last updated: Dec 06 2025 at 06:05 UTC