Closed wadudmiah closed 6 years ago
Absolutely agree, good catch! Can you submit a PR with this fix?
Thanks, milan
I don't know what PR means. Can you elaborate please? I think you might know Arjen Markus - you can ask him how good the NAG compiler is at catching bugs :-)
Can you please fork the repo, commit your changes to fix this in your repo, than submit a Pull Request (PR)? Thanks!
Hi Milan,
Yeah, sure. I can do it this weekend as I'm a bit tied up this week. Hope that's okay. I will run it past the NAG compiler with all error catching flags.
Cheers, Wadud.
Hi Milan, I did a git clone
but was unable to create the configure
script:
$ ./mkdist.sh
Running mkdist.sh
Checking versioning consistency
configure.ac reports version: '1.6.1'
datetime.f90 reports version: ''
FAIL: Versions do not match. Aborting
Hi again Milan, could you please let me know the process of pull requests? Thanks.
Hi Wadud,
Sorry for the late response. If you git clone, run autoreconf -i
and this should generate the configure
script for you. Otherwise, you can also build using cmake (see the README).
For contributing, I recommend the following:
git clone https://github.com/wadudmiah/datetime-fortran
Thanks!
Hi Milan,
I have created the pull request. Please review the changes. The tests passed. By the way, in your INSTALL file, I would put the command "autoreconf -i" to create the configure script.
The statement below (and similar ones) are incorrect:
real(kind=real64),parameter :: d2s = 86400_real64
the reason being is that 86400 is an integer literal, and the REAL64 kind is for real types. This might work if the REAL64 kind is the same as integer kinds, but will not work otherwise. The safe option is to convert it to real, i.e.
real(kind=real64),parameter :: d2s = 86400.0_real64
I came across this when trying to build it with the NAG Fortran compiler. Might be worth building it with the NAG compiler to catch any other potential bugs.
Cheers, Wadud.