Panic in `SystemTime::now()` when `clock_gettime()` fails

[pitdicker]

We got a report in chronotope/chrono#634 that I don't know what to do with.

Apparently it is possible for clock_gettime() to fail on Unix with EPERM, maybe when running inside a container or sandbox (?).

Relevant code:
https://github.com/rust-lang/rust/blob/1.72.0/library/std/src/sys/unix/time.rs#L272

Should this just remain a panic, or should something like UNIX_EPOCH be returned instead?

[jtmoon79]

Should this just remain a panic, or should something like UNIX_EPOCH be returned instead?

I think it's entirely reasonable to provide users of the standard to choose either to panic (support current behavior) or not panic (return None; let users decide if something should be panicked over). The different approaches could be wrapped in different public functions (or maybe some other approach; but provide users the option).

/// non-panic implementation of [`now`]
pub fn now_opt() -> Option<SystemTime>

[the8472]

Before doing anything on the rust side I'd suggesting adjusting the container's seccomp filters.

[Enselic]

Triage: Can this still be reproduced after adjusting seccomp filters?