`from_secs_f64` panics

[leontoeides]

Was getting panic:

thread 'tests::test_wraparound_correctness' panicked at C:\Users\User\.cargo\registry\src\index.crates.io-1949cf8c6b5b557f\jiff-0.2.6\src\signed_duration.rs:473:9:
assertion failed: nanos <= 999_999_999

Had to write helper method:

fn safe_signed_duration_from_secs_f64(seconds: f64) -> SignedDuration {
    let total_nanos = (seconds * 1_000_000_000.0).round() as i64;
    let secs = total_nanos / 1_000_000_000;
    let nanos = total_nanos % 1_000_000_000;

    SignedDuration::from_secs(secs) + SignedDuration::from_nanos(nanos)
}

[BurntSushi]

Can you please provide an MRE?

[BurntSushi]

I've got one:

    /// Regression test for a case where this routine could panic, even though
    /// it is fallible and should never panic.
    ///
    /// This occurred when rounding the fractional part of f64 could result in
    /// a number of nanoseconds equivalent to 1 second. This was then fed to
    /// a `SignedDuration` constructor that expected no nanosecond overflow.
    /// And this triggered a panic in debug mode (and an incorrect result in
    /// release mode).
    ///
    /// See: https://github.com/BurntSushi/jiff/issues/324
    #[test]
    fn panic_try_from_secs_f64() {
        let sdur = SignedDuration::try_from_secs_f64(0.9999999999999).unwrap();
        assert_eq!(sdur, SignedDuration::from_secs(1));
    }

Note that the fact that SignedDuration::from_secs_f64 panics is, on its own, not a bug. It is documented to panic in a subset of cases in which std::time::Duration::from_secs_f64 panics. The intent is to use SignedDuration::try_from_secs_f64 if one wants a fallible routine. The bug here is that try_from_secs_f64 can be made to panic, as shown above.

[BurntSushi]

This should be fixed on crates.io in jiff 0.2.8.