On Wed, Oct 19, 2005 at 05:06:40PM -0400, Andrew Pinski wrote: > GCC and gfortran (and gcj) should all play by the normal rules that we > made to GCC so that we don't spend too much time backporting (and > looking backwards) instead of fixing bugs.
gfortran is the replacement for g77, and it still doesn't handle valid code that g77 handles in some cases. So to paraphrase Bill Clinton, it all depends on what we mean be the word "regression". One interpretation is that it's only a regression if some previous release of gfortran (aka g95) worked and the current one does not. But some could make an argument that if g77 handled a case and GNU Fortran now does not, we have a regression. This would not be a hole to allow introduction of missing Fortran 95 or newer features into minor releases, unless g77 already supported the features. Still, such a criterion might lead to too much instability. So maybe there can be a compromise between those two positions that would allow for changes to gfortran to add missing g77 functionality, even for x.y.n+1 releases, as long as only the Fortran front end is affected and the changes are considered highly safe by the maintainers, and nothing else breaks. Andrew, you mention gcj, but that's in a different category; there is no older gcj that was in some ways better and that we still haven't caught up with. I think it's reasonable for the gfortran folks to want to be able to kill off the need for g77 as soon as they can.
