Barry Smith <[email protected]> writes: >> On Aug 1, 2017, at 11:48 AM, Satish Balay <[email protected]> wrote: >> >> On Tue, 1 Aug 2017, Lisandro Dalcin wrote: >> >>>> Is there a convention for marking deprecation in Fortran? I don't know >>>> a maintainable way to do it. >>> >>> Not that I know, I'm not a Fortran expert. >>> >>>> (Note that the present use of >>>> PETSC_DEPRECATED was already a very hard sell to Barry and Matt.) > > BTW: we don't seem to have a mechanism to remove functions after they have > deprecated sometime deep in the past.
That may be a feature. If you never touch that code again, there is no overhead to leaving those lines in the header. If you are touching the code then the deprecated functions might be "in the way" in which case you can delete them. >>> Well, I fail to see the problem with it, implementation work is >>> minimal, so I'm on your side on this one.... >> >> The hard part is - remembering to do it for all changes [or verifying >> if some change sliped through - without appropriate deprication] >> >>> >>>> If >>>> you know of a good way to provide deprecation stubs in Fortran, we can >>>> add that, but if not I think we should stick with what we've got. >>> >>> So, what would you recommend for deprecated functions? Try hard to >>> keep them available in Fortran? Try hard to remove them from Fortran? >>> Or rather do whatever is easier to implement the deprecation in C/C++? >> >> I don't know what we do now - but for me - if the function works in c >> [in deprecated mode] - it should still work in fortran. > > Currently this would mean writing manual stubs for those functions as > Lisandro pointed out. Kind of an annoying extra burden since before they were > depreciated they likely did not require manual stubs :-( There is a theory that the pain of function removal is much larger when there are libraries between PETSc and the user, as with the likes of Deal.II, FEniCS, and Libmesh. Fortran libraries in this role are much less common. I.e., perhaps it isn't important to offer deprecated Fortran stubs because users can "just update their code" and libraries are rarely written in Fortran. I don't recall hearing requests for a deprecation process from Fortran users, but we had lots (including in a published paper) from users of the C interface.
signature.asc
Description: PGP signature
