Re: Irix6 breaking on o32/n32 again... time to consider my patch?

[Zack Weinberg <zack at blastula dot phys dot columbia dot edu>]

On Sat, 06 Feb 1999 10:40:29 -0700, Jeffrey A Law wrote:
>
>  In message <199902061721.MAA27720@caip.rutgers.edu>you write:
>  > Jeff,
>  > 
>  > 	I'm getting new bootstrap failures on mips-sgi-irix6.2 (using cc
>  > for stage1) caused by the o32/n32 incompatibility (already discussed
>  > begining with http://www.cygnus.com/ml/egcs/1998-Dec/0044.html). 
>  > 
>  >  > stage1/xgcc -Bstage1/ -B/usr/local/mips-sgi-irix6.2/bin/ -DIN_GCC -W
>  >  > 	-Wall -O2 -g -DHAVE_CONFIG_H -o gcj gcj.o jvspec.o prefix.o ver
>sion.o
>  >  > 	obstack.o ../libiberty/libiberty.a
>  >  > ld: FATAL 112: cannot link old 32-bit object with -n32 link:
>  >  > 	../libiberty/libiberty.a(pexecute.o).
>  >  > collect2: ld returned 4 exit status
>  >  > make[2]: *** [gcj] Error 1
>  > 
>  > 
>  > 	I believe this error is caused by recent patches like this one.
>Please, let's find a way to make the mips compiler generate code that is
>compatible with gcc.  We don't want to pull apart this stuff yet.
>
>After 1.2 we can reevaluate this as part of the build reorganization.

Autoconf detect support for -n32 in mips cc, and use it
unconditionally if present.  The -n32 compiler is a good order of
magnitude less buggy anyway.

If there is no -n32 switch, you're on old Irix5 and you probably don't
have N32 runtime libraries.  In that case gcc should be generating O32
objects by default.

zw