These are chat archives for Fortran-FOSS-Programmers/General-Discussion

26th
Jun 2017
Jeff Hammond
@jeffhammond
Jun 26 2017 05:06
@szaghi I favor the autotools style of attempting to compile code and making decisions based upon that. Autotools is often too fine-grain for most applications. if an app requires features A B C D, then write a simple source that uses all of them. if that sources compiles, use the compiler. if not, throw and error. while i use autotools for a lot of system software, in most cases, i would be fine to just test if the compiler supports C99 and POSIX, because that is what i need.
a lot of what buildsystems do is punish developers for supporting terrible platforms. i mean, if your computer doesn't have a C99 compiler, maybe you should just throw it in the trash, no? Microsoft refuses to support modern C in MSVC, and their users should revolt rather than work around this nonsense. Intel and Clang support C99 on Windows. anyways, i have strong feelings about crappy programming environments. they are a waste of everyone's time and should be ignored. i recently tried out Flang based on PGI. it's 2017 and they don't support basic features of Fortran 2008. no one who cares about modern Fortran should support this compiler.
Jeff Hammond
@jeffhammond
Jun 26 2017 05:14
of course, i am biased because i work for Intel and Intel compilers are pretty good about the latest standards (and where they are not, i have filed bug reports and communicated directly with the team about the necessity of fixing the issues ASAP), and x86 has great support for GCC and Clang, but when i was supporting IBM Blue Gene and the IBM C++ compiler sucked, i advocated vigorously for Clang. https://www.ibm.com/developerworks/community/blogs/fe313521-2e95-46f2-817d-44a4f27eba32/entry/ibm_xl_compilers_for_little_endian_coming?lang=en may or may not be related ;-)
Stefano Zaghi
@szaghi
Jun 26 2017 07:25
@jeffhammond Dear Jeff, thank you for your insight, I am going along your suggested approach: minimal tests (to be compiled/ran/checked on the fly) to verify if compiler support that feature. I agree with your other ideas, but I am less sharp: flang is currently inadequate for my modern Fortran approach, but I would like to support them if they are aimed to improve the compiler into a FOSS framework :smile:
My best regards