Re: Bugs encountered updating GCC

[Daniel Jacobowitz]

On Fri, Sep 12, 2003 at 01:43:19PM -0700, Paul Eggert wrote:
> Daniel Jacobowitz <address@hidden> writes:
> 
> > It happens that libiberty does
> > not use -Werror, and is always configured first, so _that_ finds
> > malloc.h.
> 
> Can you arrange for libiberty/configure to use -Werror, even if
> libiberty doesn't use -Werror when compiling?  Technically that should
> be fairly easy to do.
> 
> If that's not possible, then perhaps you can arrange for another
> 'configure' to be called first, even before libiberty, so that you can
> test for some things using -Werror.  This 'configure' would set up
> nothing, other than config.cache.

I think the easiest way for me to do this is simply to locally reset
the warnings-are-errors flag for AC_PREPROC_IFELSE.  I don't suppose
you're interested in making this an exported feature?

> 
> > > /* Do not include <memory.h> if compiling with GCC, even if 'configure'
> > >    found it available with the native compiler.  GCC never needs
> > >    <memory.h>, and since we use -Werror in bootstrap mode, including
> > >    <memory.h> here might cause a compilation failure.  */
> > > #ifdef __GNU_C__
> > > # undef HAVE_MEMORY_H
> > > #endif
> > 
> > Unless there's a platform where malloc.h is in fact necessary.
> 
> I doubt that one exists, because of fixincludes etc.  It's quite
> possible that I'm wrong, but if there is one then we can work around
> its problems with further ifdefs (or better yet, by fixing
> fixincludes).
> 

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer