regression: X does not live long enough

[Mark-Simulacrum]

Picked up in crater:

• https://crater-reports.s3.amazonaws.com/beta-1.64-2/beta-2022-08-09/gh/cheif.json2rss/log.txt
• https://crater-reports.s3.amazonaws.com/beta-1.64-2/beta-2022-08-09/gh/jvff.dkr/log.txt
• https://crater-reports.s3.amazonaws.com/beta-1.64-2/beta-2022-08-09/gh/nielsdos.FGS.rs/log.txt
• https://crater-reports.s3.amazonaws.com/beta-1.64-2/beta-2022-08-09/reg/wonnx-0.3.0/log.txt
• https://crater-reports.s3.amazonaws.com/beta-1.64-2/beta-2022-08-09/reg/charcoal-1.1.0/log.txt
• https://crater-reports.s3.amazonaws.com/beta-1.64-2/beta-2022-08-09/reg/ft-cli-0.1.0/log.txt

Slightly different error (dropped while borrowed), probably same root cause but should confirm:

• https://crater-reports.s3.amazonaws.com/beta-1.64-2/beta-2022-08-09/reg/operator-0.6.0/log.txt

[ehuss]

I think these are two separate (unrelated) issues.

#98835 had a crater run where these regressions were identified: #98835 (comment)

#99413 had a crater run where there were quite a lot of regressions, but nobody has reviewed it: #99413 (comment)

There are quite a few regressions in #99413, but I suspect that will need to be accepted since it is a soundness fix.

[estebank]

From #98835 (comment):

The migration is trivial in most cases: removing lifetime annotations is sufficient, however innernet requires replacing Self constructor with the struct name.

I wish this had been flagged as a potential diagnostic issue back when this was identified. Filed #100572.

[jackh726]

In theory, we could back out of #98835, maybe with a future compat warning.

[steffahn]

I bisected charcoal and dkr, both pointing to caee496 which does contain #98835.
And I bisected operator, pointing to af7ab34 which contains #99413.

[steffahn]

json2rss is also #99413
(by the way, I have opened PRs to multiple affected crates, including json2rss and operator, those two are already merged ^^)

FGS regresses at the same nightly as #98835, as does wonnx

ft-cli is most definitely #99413 (the handlebars method “render” was involved in a lot of other #99413-regressions, too). I didn’t bisect or send a patch, since the repo is archived and has lots of dependencies.

@rustbot label -E-needs-bisection

@aliemjay provided a minimal example of the #98835-regression in their PRs:

async fn test<'a>() { // `'a`, being a lifetime parameter, must be longer than the entire function body
    let f = |_: &'a str| {}; // the argument is required to outlive `'a`
    f(&String::new()); // an argument of shorter lifetime is passed!
}

For the #99413-regressions, we don’t need a mcve, since breakage was expected from the soundness fix.

@rustbot label -E-needs-mcve

[seanpianka]

Is this actually a regression? I saw it mentioned that this is a correctness fixed introduced as a part of #98835.

I think this issue is probably safe to close.

[apiraino]

@seanpianka I think at this point I agree with your opinion (also by looking at this comment).

@jackh726 do you agree this can be closed without further action? is it a case of intended behaviour but perhaps poorly communicated to users? other interesting points to note here?

thanks!