Remove unnecessary floating point remainder

[CryZe]

Because the code handles all the edge cases already, we can simply cast the integer seconds back to a floating point number to truncate it, instead of using seconds % 1.0 which is not only slower, but also bloats the binary, especially on no_std, as it brings in an implementation of fmod / fmodf.

[codecov[bot]]

Codecov Report

Merging #526 (ba68307) into main (01ca01a) will increase coverage by 0.0%. The diff coverage is 100.0%.

@@          Coverage Diff          @@
##            main    #526   +/-   ##
=====================================
  Coverage   98.4%   98.4%           
=====================================
  Files         76      76           
  Lines       8215    8221    +6     
=====================================
+ Hits        8082    8088    +6     
  Misses       133     133           

Impacted Files	Coverage Δ
time/src/duration.rs	100.0% <100.0%> (ø)

:mega: We’re building smart automated test selection to slash your CI/CD build times. Learn more

[jhpratt]

Neat optimization. I've confirmed that it reduces the amount of generated assembly, so it's definitely a win.

[CryZe]

Another thing to consider might be to port the algorithm from std, which nowadays basically abuses the fact that floating point numbers are really just the mantissa that represents "two integers" separated by a decimal point (through the exponent). So you could just extract those two integers without any loss of precision (as would be the case when trying to represent the fraction as its own floating point number). Though that's of course slower again.

[jhpratt]

Ah right, I forgot the algorithm changed a couple releases ago. Thanks for the reminder!

[jhpratt]

So…I just attempted to adapt that algorithm. It's a bit hacky, as I basically just forced the sign to be what it was supposed to be at the end, and still doesn't differentiate overflow and NaN (for the panic message). But in the end I am measuring a regression. That may be because of the exact benchmarks I am using, but it doesn't seem to be a super obvious gain. As such I'm going to keep your improvement and stop there.