Dealing with huge timedelta values about datetime-fortran HOT 5 CLOSED

This is fixed in latest relase: https://github.com/wavebitscientific/datetime-fortran/releases/tag/v1.6.1. Thanks @alexavr.

from datetime-fortran.

Hi alexavr, great catch! I can reproduce this.

No compiler promotion for all integers is necessary. The addition of integers goes left-to-right here (even though not defined by the standard and compiler-dependent), so integer overflow occurs before we get to multiplying with milliseconds.

The correct and easy solution IMO is to multiply with a 64-bit float at the start:

  total_seconds = 1._real64 * self % days*86400 + self % hours*3600&
                + self % minutes*60 + self % seconds   &
                + self % milliseconds*1e-3_real64

I tested this with your program. With this solution, we can work with 32-bit integers and floats for most stuff, and use 64-bit where needed.

from datetime-fortran.

Very good idea!
Works for me.
Appreciated!

from datetime-fortran.

Do you mind submitting this fix as a PR?

1. Fork this repo
2. git clone https://github.com/alexavr/datetime-fortran
3. Make your changes
4. git add, commit, and push to alexavr/datetime-fortran
5. Open Pull Request from your repo page.

from datetime-fortran.

Are you kidding? I'll be honored!
Give me some little time to learn how to do that LOL

from datetime-fortran.