[flang-commits] [PATCH] D130166: [flang] Adding a guideline for flang design documentation

Thu Jul 21 08:58:07 PDT 2022

clementval added inline comments.

================
Comment at: flang/docs/DesignGuideline.md:75
+  (this does not require writing a design document).
+- Identify if the feature affects compatibility with programs compiled by other
+  Fortran compilers, or if a given solution for flang could not be changed in
----------------
kiranchandramohan wrote:
> awarzynski wrote:
> > jeanPerier wrote:
> > > kiranchandramohan wrote:
> > > > Is there are preference for compatibility with gfortran over other fortran compilers?
> > > I do not want to make it a general rule to avoid inciting people to just do the same as gfortran ABI without thinking it through. But if being compatible with gfortran does not come at extra cost, then I would say it is probably the safe choice.
> > > I do not want to make it a general rule to avoid inciting people to just do the same as gfortran ABI without thinking it through. 
> > 
> > That's a good point, but I think that we also want to avoid a situation in which some parts of "LLVM Flang" align with GFortran, something else with "Classic Flang" etc. To me, it's either "align with GFortran" or "do something custom". Consulting other compilers is important and should be taken into account, but I would still prioritise GFortran.
> > 
> > This is what Clang folks did in the early days. IIRC, `clang` used to be a drop-in replacement for `gcc` for a while. Then, gradually, they started diverging. 
> It might be good to clarify this point further. Given that the descriptor, module format, etc., are not compatible, exactly what compatibility are we referring to here?
> 
> GCC has a dominant position in the C/C++ world, but I am not sure about the prominence of Gfortran in the Fortran world. A lot of our community comes from Classic Flang/nvfortran. It will be good to check compatibility there as well.
+1 on the fact that gfortran has not the same prominence. Especially in scientific computing where users tend to use OpenACC or OpenMP offloading. 

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D130166/new/

https://reviews.llvm.org/D130166