Stream: t-compiler/wg-prioritization/alerts

Topic: #82833 char::encode_utf8 crash on nightly


triagebot (Mar 06 2021 at 12:59, on Zulip):

@WG-prioritization/alerts issue #82833 has been requested for prioritization.

Procedure

apiraino (Mar 06 2021 at 13:40, on Zulip):

hum ... I would say P-high when the nightly triggers an undefined behaviour. The three compilation flags are all on stable (IIUC).
(perhaps I'm missing other mitigation points that would warrant a lower priority)

Léo Lanteri Thauvin (Mar 06 2021 at 15:02, on Zulip):

The flags are indeed stable

Léo Lanteri Thauvin (Mar 06 2021 at 15:08, on Zulip):

Confirmed this is due to LLVM 12 update, pinged LLVM group

Léo Lanteri Thauvin (Mar 06 2021 at 15:13, on Zulip):

Hmm, how often are -Ccodegen-units=1 -Cinline-threshold=0 -Clink-dead-code used in real code? Given that this bug was detected pretty quickly, I assume it's not too rare

Léo Lanteri Thauvin (Mar 06 2021 at 15:13, on Zulip):

So I think this might even be P-critical?

Joshua Nelson (Mar 06 2021 at 17:00, on Zulip):

Léo Lanteri Thauvin said:

Hmm, how often are -Ccodegen-units=1 -Cinline-threshold=0 -Clink-dead-code used in real code? Given that this bug was detected pretty quickly, I assume it's not too rare

FYI codegen unit partitioning is essentially random (or at least unpredictable) so I wouldn't deprioritize this just for -Ccodegen-units=1

Joshua Nelson (Mar 06 2021 at 17:01, on Zulip):

link-dead-code is used frequently for coverage testing

Joshua Nelson (Mar 06 2021 at 17:01, on Zulip):

I don't know what inline-threshold is for

Léo Lanteri Thauvin (Mar 06 2021 at 17:02, on Zulip):

I'd guess it's to prevent inlining?

Joshua Nelson (Mar 06 2021 at 17:02, on Zulip):

looks like it could also be useful for coverage testing: https://doc.rust-lang.org/nightly/rustc/codegen-options/index.html#inline-threshold

Léo Lanteri Thauvin (Mar 06 2021 at 17:03, on Zulip):

Mhh, so you think it's P-critical then?

Joshua Nelson (Mar 06 2021 at 17:03, on Zulip):

given that it breaks coverage testing for pyo3 I think so, yes https://github.com/PyO3/pyo3/pull/1472

Yuki Okushi (Mar 06 2021 at 18:03, on Zulip):

and it's an unsoundness issue, agree with P-critical

triagebot (Mar 06 2021 at 18:05, on Zulip):

Issue #82833's prioritization request has been removed.

Last update: Apr 11 2021 at 19:15UTC