stat: truncate fractional pre-epoch timestamps toward zero - #14975

GNU testsuite comparison:

Congrats! The gnu test tests/stat/stat-nanoseconds is no longer failing!

Should we fix format_scaled_decimal instead?

Should we fix format_scaled_decimal instead?

Should we pass the original sign to format_scaled_decimal?
once the value is truncated to zero,
format_scaled_decimal no longer knows whether the original value was negative

I’m not particularly fond of the existing approach. It might be worth considering whether we should use a crate that explicitly supports rounding, such as rust_decimal.

I’m not particularly fond of the existing approach. It might be worth considering whether we should use a crate that explicitly supports rounding, such as rust_decimal.

rust_decimal makes sense if we want to share rounding behavior across utilities like stat, printf, numfmt, and seq.
For stat alone, keeping the current integer-based approach may be simpler.
I wonder if this will affect performance.

Merging this PR will degrade performance by 2.31%

⚡ 2 improved benchmarks
❌ 1 regressed benchmark
✅ 390 untouched benchmarks
⏩ 54 skipped benchmarks¹

Performance Changes

_{Comparing mattsu2020:stat-nanoseconds.sh (6d18249) with main (cacdd8d)}

Benchmark variance tracked by #14921.