https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69860
kargl at gcc dot gnu.org changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |janus at gcc dot gnu.org --- Comment #13 from kargl at gcc dot gnu.org --- (In reply to kargl from comment #12) > (In reply to Gerhard Steinmetz from comment #11) > > And with "kind=4" instead of "kind=1", i.e. with testfile z1.f90 ? > > > > > > for n in `seq 1 1000` > > do > > gfortran-7-20160821 -O2 -mavx -c z1.f90 > > done > list 2>&1 > > > > grep 'internal compiler' list | wc > > 946 5676 47300 > > > > > > I can also confirm that ICEs are gone for z2.f90 ("kind=1"). > > Ugh. I was testing apparently a kind=1 version. The problem > is that gfortran is inserting __convert_s1_s4 to convert to > the same kind, which of course fails. This un-regression-tested > patch covers up the ICE, but it is perhaps not the correct fix. > Fortunately, I don't use kind=4 character types, so I don't have > a dog in this fight. > This might have been fixed by Janus patch: 2016-12-04 Janus Weil <ja...@gcc.gnu.org> PR fortran/78618 * intrinsic.c (gfc_convert_type_warn): Do not set the full typespec for the conversion symbol, but only type and kind. Set the full typespec for the expression. (gfc_convert_chartype): Ditto. Gerhard can re-do your testing. I seem to have good luck with FreeBSD in testing this.