${ac_fc_srcext-f} considered harmful

[Bert Wesarg]

,Dear all,

AC_LANG(Fortran) uses the above parameter expansion to set `ac_ext`, see here:

    
http://git.savannah.gnu.org/cgit/autoconf.git/tree/lib/autoconf/fortran.m4#n223

Though I consider it harmful, because if $ac_fc_srcext is set but null
the value of ac_ext is also null.

That problem arises, if one wants to use the usual 'save=$var; : do
something; var=$save' idiom. Which is done in fortran.m4 in
AC_FC_PP_DEFINE here too:

    
http://git.savannah.gnu.org/cgit/autoconf.git/tree/lib/autoconf/fortran.m4#n1302

It does:

AC_LANG_PUSH([Fortran])dnl
ac_fc_pp_define_srcext_save=$ac_fc_srcext
AC_FC_PP_SRCEXT([F])
AC_CACHE_CHECK(…)
ac_fc_srcext=$ac_fc_pp_define_srcext_save

If $ac_fc_srcext was unset before, its now set but null, thus the next
ac_ext=${ac_fc_srcext-f} is null too.

On a side note, AC_FC_PP_SRCEXT may also set $ac_fcflags_srcext, but
that is not saved/restored by AC_FC_PP_DEFINE.

Can someone please have a llok at my reasoning?

Thanks.

Best,
Bert