[netcdfgroup] netcdf-4.0: Fortran bugs in configure script

• To: netcdfgroup@xxxxxxxxxxxxxxxx
• Subject: [netcdfgroup] netcdf-4.0: Fortran bugs in configure script
• From: Harald Anlauf <Harald.Anlauf@xxxxxx>
• Date: Mon, 06 Oct 2008 17:14:42 +0200

Hi,

netcdf-4.0 as distributed has some bugs in the Fortran configuration
part of the configure script, see attachment.

Explanation:

integer(kind(1)) is just an synonym for integer, but one probably
just wants the smallest integer kind, which often is integer(kind=1).
For a specification of the minimum number of decimal digits, the
last variant (integer(selected_int_kind(2))) does the proper job,
which is probably what the original authors intended.

Without this patch, I get failures in the Fortran self-tests on
a platform where the Fortran compiler does not support "byte",
although the C compiler does.

It would be nice if this patch could be applied to a future
netcdf distribution.

Regards,
-ha

--
Harald Anlauf
Deutscher Wetterdienst (DWD)           | Phone:  +49 69 8062 4941
FE12 / Datenassimilation               | Fax:    +49 69 8062 4954
63067 Offenbach                        | e-Mail: harald.anlauf@xxxxxx

--- ./acinclude.m4.tmp  2008-10-06 14:51:51.883266000 +0000
+++ ./acinclude.m4      2008-10-06 15:06:22.466403000 +0000
@@ -396,8 +396,8 @@
        ;;
     *)
        AC_REQUIRE([UD_CHECK_FCALLSCSUB])
-       UD_CHECK_FORTRAN_TYPE(NF_INT1_T, byte integer*1 "integer(kind(1))")
-       UD_CHECK_FORTRAN_TYPE(NF_INT2_T, integer*2 "integer(kind(2))")
+       UD_CHECK_FORTRAN_TYPE(NF_INT1_T, byte integer*1 "integer(kind=1)" 
"integer(selected_int_kind(2))")
+       UD_CHECK_FORTRAN_TYPE(NF_INT2_T, integer*2 "integer(kind=2)" 
"integer(selected_int_kind(4))")
 
        case "${NF_INT1_T}" in
            '') ;;