On Mon, Sep 09, 2019 at 02:52:20PM +0100, Mark Eggleston wrote: > gcc/fortran > > Mark Eggleston <mark.eggles...@codethink.com> > > * arith.c (gfc_arith_concat): Set length field in typespec. > * expr.c (gfc_get_character_expr): Set length field in typespec. > * gfortran.h: Add length field to gfc_typespec for use to allow the > length to available for character literals. > * interface.c (gfc_check_dummy_characteristics): Use gfc_dummy_typename > instead of gfc_typename when constructing error message to allow for > CHARACTER(*) and CHARACTER(*,4). > (compare_parameter): Use gfc_dummy_typename for formal argument when > constructing error message to allow for CHARACTER(*) and > CHARACTER(*,4). > * intrinsic.c (gfc_actual_arglist): Reword error message so that > CHARACTER(*) or CHARACTER(*,4) can be reported as the target type. Use > gfc_dummy_typename for the formal argument. > * misc.c (gfc_typename): Add new local variable length and initialise > with the value from the length field in typespec passed in. If there > is a character structure use the value from there for length. If the > kind is the default character kind construct the type name using length > otherwise use the length followed by kind separated by a comma. > (gfc_dummy_typename): New routine for use with formal arguments, if the > typespec does not have a character length structure then the length is > assumed and * is used for the length, if kind is not the default > character kind follow * with a comma and then the kind. > > gcc/testsuite > > Mark Eggleston <mark.eggles...@codethink.com> > > * gfortran.dg/bad_operands.f90: New test. > * gfortran.dg/character mismatch.f90: New test. > * gfortran.dg/compare_interfaces.f90: New test. > * gfortran.dg/widechar_intrinsics_1.f90: Checked for specific character > type names instead of "Type of argument". > * gfortran.dg/widechar_intrinsics_2.f90: Checked for specific character > type names instead of "Type of argument". > * gfortran.dg/widechar_intrinsics_3.f90: Checked for specific character > type names instead of "Type of argument". >
This looks OK to me. I don't know if anyone else will glance at the patch. You said that the patch has the feeling of a kludge. This is probably a by-product of bolting kind=4 support onto what gfortran inherited long ago. -- steve
