Wasmtime GitHub notifications bot (Oct 05 2022 at 14:53): Wasmtime GitHub notifications bot (Oct 05 2022 at 14:53):bjorn3 opened issue #5021:
It causes a miscompilation for cg_clif: https://github.com/bytecodealliance/wasmtime/issues/5018. The merged commit in question is 05cbd667c7e89828a45ebc5760787a6160c55c8d.
cc @alexcrichton (https://github.com/bytecodealliance/wasmtime/pull/5016#pullrequestreview-1131535609)
Wasmtime GitHub notifications bot (Oct 05 2022 at 16:16):cfallin commented on issue #5021:
@bjorn3 I'm looking into the bug, and want to see if we can find it and backport it to the release branch before 2.0 goes out on Oct 20.
However, I don't want to revert #4892: it's a large PR and it's been built on top of by subsequent work, so it would be nontrivial to revert. We suspect that the bug is in handling of struct-ret args, which is a tier-3 feature, and shouldn't hold back work otherwise. I absolutely want to find the bug, but I also want to minimize secondary breakage and complexity from large reverts.
Wasmtime GitHub notifications bot (Oct 05 2022 at 16:22):bjorn3 commented on issue #5021:
A revert of that commit only requires resolving a conflict in
Cargo.lock,cranelift/codegen/Cargo.tomlandsupply-chain/audits.toml(due to a conflict with https://github.com/bytecodealliance/wasmtime/pull/4945 which updated regalloc2) and then all archs except for riscv64 work (riscv64 was merged after that PR and thus doesn't get the revert). There don't seem to be any conflicts with any PR merged after that (apart from #4945). If I keep the update to regalloc2 0.4.1 from 0.3.2, I only need to remove two asserts that check that the stack slot class matches the expectation as the respective field has been removed. This means that the revert is almost trivial in amount of work. I do agree that it is a rather large revert and thus risky.
Wasmtime GitHub notifications bot (Oct 07 2022 at 16:32):bjorn3 commented on issue #5021:
The bug has been fixed by https://github.com/bytecodealliance/wasmtime/pull/5023 which has already been cherry-picked onto the wasmtime 2.0 branch. No need for a revert anymore.
Wasmtime GitHub notifications bot (Oct 07 2022 at 16:32):bjorn3 closed issue #5021:
It causes a miscompilation for cg_clif: https://github.com/bytecodealliance/wasmtime/issues/5018. The merged commit in question is 05cbd667c7e89828a45ebc5760787a6160c55c8d.
cc @alexcrichton (https://github.com/bytecodealliance/wasmtime/pull/5016#pullrequestreview-1131535609)
Last updated: Nov 22 2024 at 16:03 UTC