Re: Proper way to link Fortran with C++

[Steven G. Johnson]

Ralf Wildenhues wrote:
> > The Automake documentation on the other hand states that I should put $(FLIBS)
> > into the apropriate _LDADD or _LIBADD variable.  That would include the value
> > of FLIBS before the value of LIBS on the linker command line, contradicting
> > the autoconf documentation.
> >
> > Is there a reason Autconf requires FLIBS after LIBS, and not just FLIBS after
> > the library actually using fortran?  That would be
> >
> >    $BLAS_LIBS $FLIBS $LIBS
> >
> > in the case of blas, and it would allow me to fold FLIBS into BLAS_LIBS.  This
> > in turn would make it easier to omit the FLIBS completely in case blas was not
> > detected by configure.
>
> Do you remember why the Autoconf macro specifically recommends
> this ordering?  We should amend it with a rationale.
>
> I couldn't find anything in the archives of the macro sources about it.
> Maybe it is because $LIBS may contain a Fortran library?

Yes.

The only reason to check for FLIBS is to link to Fortran libraries from 
a non-Fortran program.  And the standard way to check for libraries is 
to use AC_CHECK_LIB, which by default adds the library to LIBS, rather 
than using a separate variable ala BLAS_LIBS (and often people who use 
e.g. ACX_BLAS just add $BLAS_LIBS to LIBS).

Also, putting libraries into LIBS is important for several other 
autoconf macros to work normally, e.g. AC_CHECK_FUNC.

So, the point is to allow people to use AC_CHECK_LIB normally after 
calling AC_F77_LIBRARY_LDFLAGS (and possibly AC_F77_FUNC to find the 
appropriate C linker name for a Fortran identifier).

Regards,
Steven G. Johnson