Re: [Mesa3d-dev] Build error on current CVS (HEAD) [PATCH]

[Brian P. <br...@va...>]

Richard Guenther wrote:
> 
> Hi!
> 
> I'm unable to build the current mesa on standard Linux/ix86:
> 
> ./configure --disable-ggi-fbdev
> 
> gcc -DHAVE_CONFIG_H -I. -I. -I../../.. -I../../../include
> -I../../../src/GGI/include -I../../../src -g -O2 -Wall
> -fomit-frame-pointer -ffast-math -fexpensive-optimizations -malign-loops=2
> -malign-jumps=2 -malign-functions=2 -D_REENTRANT -Wp,-MD,.deps/stubs.pp -c
> stubs.c  -fPIC -DPIC -o stubs.lo
> stubs.c: In function `MesaGGIdl_stubs':
> stubs.c:415: `GGIFUNC_open' undeclared (first use in this function)
> stubs.c:415: (Each undeclared identifier is reported only once
> stubs.c:415: for each function it appears in.)
> stubs.c:418: `GGIFUNC_exit' undeclared (first use in this function)
> stubs.c:419: `GGIFUNC_close' undeclared (first use in this function)
> stubs.c:416: warning: unreachable code at beginning of switch statement
> make[4]: *** [stubs.lo] Error 1
> 
> It seems that though GGI seems to be detected disabled correctly:
> 
> checking for ggi/ggi.h... (cached) yes
> checking for main in -lggi... (cached) yes
> checking for kgi/kgi.h... (cached) no
> checking if we should build GGIMesa's fbdev target... no
> checking if we should build GGIMesa's generic KGI driver... no
> 
> (but ggi.h installed) is is tried to use it - as I can see from
> conf.h GGI is enabled...
> 
> /* #undef FX */
> /* #undef FX_GLIDE3 */
> #define GGI 1
> #define SVGA 1
> #define USE_XSHM 1
> 
> so perhaps we need a --disable-ggi switch... [see attached patch]
> also --disable-svga should be there for the same reasons (well I
> would vote for all but X11 and GLX drivers to be disabled by default
> and to be enabled by --enable-feature rather than autodetecting such
> stuff anyway).
> 
> Note that the patch includes addition of -fstrict-aliasing as this
> enables a whole range of optimizations of array access and friends
> and makes a huge difference (just checked that here on IRIX) - I hope
> Mesa is safe with the aliasing rules of ANSI C.... at least it seems
> to work with the supplied demos (just picked a few random ones).
> 
> Richard.
> 
> --
> Richard Guenther <ric...@st...>
> WWW: http://www.anatom.uni-tuebingen.de/~richi/
> The GLAME Project: http://www.glame.de/
> 
> Index: configure.in
> ===================================================================
> RCS file: /cvsroot/mesa3d/Mesa/configure.in,v
> retrieving revision 1.27
> diff -u -r1.27 configure.in
> --- configure.in        2000/07/01 08:22:08     1.27
> +++ configure.in        2000/07/19 11:27:18
> @@ -77,7 +77,7 @@
>    dnl Optimization flags
>    if test "x$enable_debug" = xno && test "x$enable_prof" = xno; then
>      if test "x$GCC" = xyes; then
> -      CFLAGS="$CFLAGS -fomit-frame-pointer -ffast-math -fexpensive-optimizations"
> +      CFLAGS="$CFLAGS -fomit-frame-pointer -ffast-math -fexpensive-optimizations -fstrict-aliasing"
>        case "$host" in
>        i*86-*-*) CFLAGS="$CFLAGS -malign-loops=2 -malign-jumps=2 -malign-functions=2";;
>        esac
> @@ -299,7 +299,12 @@
> 
>  dnl ------------------------------------------
>  dnl GGI driver
> -have_ggi=auto
> +AC_ARG_ENABLE(ggi,
> +[  --disable-ggi           Don't build the GGI target at all],
> +have_ggi=$enableval, have_ggi=auto)
> +if test "x$have_ggi" = xyes; then
> +       have_ggi=auto;
> +fi
>  GGI_CFLAGS=""
>  GGI_LIBS=""
>  AC_ARG_WITH(ggi,


OK, I've applied your patch to both the 3.2.1 and 3.3 branches.
But, I removed the -fstrict-aliasing option from 3.2.1 just to be
cautious.  I always avoid pointer aliasing in function calls but
can't guarantee that in anybody else's code.

-Brian