Re: problems with egcs-1.1.1 & i386-pc-sco3.2v5.0.4

[Robert Lipe <robertlipe at usa dot net>]

> I know some of you are successfully using egcs on SCO OSR5, so

I've probably built it a few hundred times on OSR5.   Most of them
were generally successful. :-)

> it on the SCO box, xgcc generates problematic assembly code during
> the bootstrap procedure.

Actually, there are two things happening here.  One is the bug
in EGCS that causes the Dwarf2 debugging to make up references
to non-existent labels.   This bug is normally not triggered during
a bootstrap, but it does show up in testsuite.

The other is a little more subtle.

> Output from ``make bootstrap'' follows.
> 
> ../../egcs-1.1.1/gcc/frame
> xgcc: file path prefix `/usr/local/lib/gcc-lib/elf/egcs-2.91.60/' never used
> make[5]: Entering directory `/usr/macfar/src/dist/egcs/object/gcc'
> ./xgcc -B./ -O2   -DIN_GCC    -O2 -g -O0 -Xa -b elf -I./include   -g1  -DIN_LIBG
> CC2 -D__GCC_FLOAT_NOT_NEEDED    -I. -I../../egcs-1.1.1/gcc -I../../egcs-1.1.1/gc
> c/config \
>   -c ../../egcs-1.1.1/gcc/cp/tinfo.cc
> (EOF):undefined name: .L_T117

The highly trained eye notices the warning from xgcc.   This was
tip one.  The reason for this warning is a bogus '-b' argument.

It can't be coincidence that "-b elf" happens to be the argument to
the native SCO compiler to make it spit out ELF.  HOwever, '-b' means
something completely different to EGCS.   (Proof left as exercise for
the reader.   Hint: reread the warning....)


Now, if you do a 'make bootstrap', you should never ever see a '-b elf'
on the command lines.   You need to find out what you're doing to make
those appear and quit doing that. :-)

RJL