https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59016
--- Comment #16 from Dominique d'Humieres <dominiq at lps dot ens.fr> --- > As it seems the problem with the program bug.f90 is that the generic > attribute is set in a symbol as the parser tries to match a declaration > type specification; but finally, the statement isn't matched as such > and the generic attribute wasn't cleared. > > So, the attached patch makes the cleanup, without causing the regressions > listed in comment 9. IMO the patch addresses the root of the problem and prevents the ICEs in the PRs listed in comment 9. However it causes 3000+ failures in the test suite, so the filter you apply is not narrow enough. > The question here is if gfortran had to print an error message for > this program: > > module types > public :: undefined > end module types > write (*,*) "undefined types!" > end Why should it without "implicit none"? Your patches do not follow the GNU coding standards: http://www.gnu.org/prep/standards/ The last patch should look as --- ../_clean/gcc/fortran/decl.c 2015-03-25 14:07:04.000000000 +0100 +++ gcc/fortran/decl.c 2015-04-05 13:02:38.000000000 +0200 @@ -4427,6 +4427,31 @@ ok: gfc_free_data_all (gfc_current_ns); cleanup: + //<pr59016> in gfc_match_data_decl; cleanup the garbages + gfc_symbol *csym = NULL; + if (current_ts.u.derived && current_ts.u.derived->name) + { + char iname[129]; + iname[128] = 0; + for (int i = 0; i < 128; i++) + { + iname[i] = current_ts.u.derived->name[i]; + if (current_ts.u.derived->name[i] == 0) + break; + } + iname[0] = TOLOWER(iname[0]); + for (int i = 0; i < 4; i++) + { + gfc_find_symbol (iname, NULL, i, &csym) ; + if (csym && csym->generic + && (csym->generic->sym == current_ts.u.derived)) + { + csym->generic = NULL; + break; + } + } + } + //</pr59016> gfc_free_array_spec (current_as); current_as = NULL; return m; (note that I may have done some errors myself). In addition you should comment what you are trying to do.
