Re: bootstrap on 3.1 (probably 3.0.4) on i586-pc-sco3.2v5.0.2 --with-gnu-as

[Robert Lipe <robertlipe at usa dot net>]

rbrown64@csc.com.au wrote:

> Building on i586-pc-sco3.2v5.0.2 --with-gnu-as fails as follows
> 
> ./xgcc -B./ -B/usr/local/i586-pc-sco3.2v5.0.2/bin/ -isystem
> /usr/local/i586-pc-sco3.2v5.0.2/include -isystem
> /usr/local/i586-pc-sco3.2v5.0.2/sys-include -O2 -DIN_GCC    -W -Wall
> -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -isystem ./include
> -I. -I. -I../../gcc-20020218/gcc -I../../gcc-20020218/gcc/.
> -I../../gcc-20020218/gcc/config -I../../gcc-20020218/gcc/../include  -g0
> -finhibit-size-directive -fno-inline-functions -fno-exceptions -mcoff
> -fno-omit-frame-pointer \
>    -c ../../gcc-20020218/gcc/crtstuff.c -DCRT_BEGIN -DCRTSTUFFS_O \
>   -o crtbeginS.o
> cc1: invalid option `coff'

Bummer.  The clash of the vestigal options. --with-gnu-as really isn't
tested very often on OpenServer becuase in the days when we supported
both ELF and COFF binutils worked terribly when it worked at all.  Now
that COFF sleeps with the fishies, I should revisit my grudge against
it, I suppose.


> The -mcoff is provided by
> 
> gcc/config/i386/t-sco5gas:CRTSTUFF_T_CFLAGS_S = -mcoff
> -fno-omit-frame-pointer
> 
> I worked around it by editting the gcc/Makefile. The obvious fix is to
> remove the -mcoff from
> t-sco5gas .Will try tomorrow when this bootstrap completes.
> 
> Robert, do you agree with removing -mcoff from t-sco5gas and is it
> sufficiently low impact
> to go into 3.0.4?

It now seems pretty clear you're the first one to configure
--with-gnu-as on this target since 3.0.0, so I think anything we do to
this file is low risk. :-)

My quick testing (verifying that --enable-languages=c compiles; nothing
more) shows that deleting CRTSTUFF_T_CFLAGS_S from t-sco5gas completely
is OK.  Since we no longer pretend to support COFF, there's no reason to
keep this spare object around.

If somone that understands CRTSTUFF_T_CFLAGS_S rubber-stamps this as OK,
I'm hip with it.

> xgcc: Internal error: Segmentation fault (program as)
> Please submit a full bug report.
> See <URL:http://www.gnu.org/software/gcc/bugs.html> for instructions.
> make[4]: *** [limits.lo] Error 1
> make[4]: Leaving directory
> `/users/rdb/src/gcc-3.0.3.obj/i586-pc-sco3.2v5.0.2/libstdc++-v3/src'

as dropping core on 5.0.2 on wild section manipulation (such as done
by modernish g++) was a well known problem when 5.0.2 was current.
Installing TLS706, a backport of a later assembler, stands an excellent
chance of curing this.  I think you'll need TLS597, too.  We seem to be
trying to make these hard to find for users (sigh) but I did finally
find them at ftp://stage.caldera.com/TLS/.