werror fallout for cross-builds (was: Re: [BUILDROBOT][PATCH] Fix mmix (unused variable))

[Hans-Peter Nilsson]

On Thu, 24 Jul 2014, Jan-Benedict Glaw wrote:
> On Tue, 2014-07-22 16:40:31 -0400, Hans-Peter Nilsson <hp@bitrange.com> wrote:
> > Jan-Benedict, which host gcc version do you use when getting
> > most targets to build with config-list.mk?  Maybe we can just
> > set the initial version to that instead of 4.4.4.
>
> darkeye		gcc (Debian 4.8.1-7) 4.8.1
> gccbuild	gcc (Debian 4.8.1-7) 4.8.1
> pluto		gcc (Debian 4.9.1-1) 4.9.1
> gcc20		gcc (Debian 4.4.5-8) 4.4.5
> gcc76		gcc (Debian 4.4.5-8) 4.4.5
> gcc110		gcc (GCC) 4.7.2 20121109 (Red Hat 4.7.2-8)
> gcc111		gcc (GCC) 4.8.1
> 		XL 12.1.0.0 (if I ever get that properly working...)

I tried to repeat that, for the CFarm hosts.  On gcc111 trying
config-list.mk on the 0720 snapshot (and with mpc, mpfr and gmp
in-tree) gives me: "configure: error: GNAT is required to build
ada" already for aarch64-unknown-elf.  Somewhat expected, as I
don't think many of the targets in the config-list.mk LIST have
Ada bits ported, but maybe there are no Ada specific bits needed
to build the GNAT compiler proper, just a host GNAT.

On gcc110 which *has* gnat, I get:

/gcc/o/aarch64-elf/./mpfr -I/home/hp/gcc/gcc/mpfr -I/opt/cfarm/mpc/include  -I../../../gcc/gcc/../libdecnumber -I../../../gcc/gcc/../libdecnumber/dpd -I../libdecnumber -I../../../gcc/gcc/../libbacktrace    -o dwarf2out.o -MT dwarf2out.o -MMD -MP -MF ./.deps/dwarf2out.TPo ../../../gcc/gcc/dwarf2out.c
In file included from ../../../gcc/gcc/real.h:25:0,
                 from ../../../gcc/gcc/rtl.h:27,
                 from ../../../gcc/gcc/dwarf2out.c:62:
../../../gcc/gcc/wide-int.h: In function 'void insert_wide_int(const wide_int&, unsigned char*, int)':
../../../gcc/gcc/wide-int.h:800:48: error: array subscript is above array bounds [-Werror=array-bounds]
cc1plus: all warnings being treated as errors
gmake[2]: *** [dwarf2out.o] Error 1
gmake[2]: Leaving directory `/home/hp/gcc/o/aarch64-elf/gcc'

By that list, did you really mean that you got even 4.4.5 to
work on an unmodified config-list.mk?

Perhaps you have local patches or did you call config-list.mk
with some kind of options?  Maybe you didn't actually use
config-list.mk?  Or just looked to see whether the first failure
for each target was on a target-specific file or the (same)
middle-end bits?  Ok, I'm out of guesses. :)

> > For reference, the patch, which works as intended (-Werror in
> > the gcc build directory for cross-builds by default, not
> > affecting native builds and not at all for gcc < 4.4.4).  (Vax
> > is excepted, see J-B's previous post.)  I'd ask for approval
>
> VAX sould work, it's pdp11 that I said wouldn't.

Sorry I misremembered.

brgds, H-P