🚀 go-pugleaf

I have a Compaq 486/25 running Interactive 386/ix version 2.2 and I am
having a horrible time installing gcc 1.38 on it with COFF
encapsulation.

Despite various files posted by people claiming success with the task
I have just mentioned, I cannot seem to get gcc running on this
machine.  I have been through the installation process a gazillion
times. I can get stage1 built, but rebuilding with stage1 is always
the problem.

In my current attempt, I followed these steps:

- Install flex, bison, gas and binutils (no problems)
- run config.gcc i386-sysv-gas
- remove  'gnulib2' from the all: target in the Makefile
- run 'make'.  This uses the native cc, ld and as.
- rename gnulib to gnulib.old, and run 'libconvert gnulib.old gnulib'.
  This puts the converted library in the gcc directory under the name
  gnulib.
- Now run 'make gnulib2'. This should use gcc to compile the object
  code, then gnu's ar from binutils to add it to gnulib. I got this
  output:

[root:ccel486] /usr2/local/gcc-1.38 (62): make gnulib2
        for name in _adddi3 _subdi3 _muldi3 _divdi3 _moddi3 _udivdi3 _umoddi3 _negdi2  _anddi3 _iordi3 _xordi3 _lshrdi3 _lshldi3 _ashldi3 _ashrdi3 _one_cmpldi2 _bdiv _cmpdi2 _ucmpdi2 _fixunsdfdi _fixdfdi _floatdidf _varargs; \
        do \
          echo ${name}; \
          ./gcc -B./ -fstrength-reduce -O -I. -I. -I./config  -c -DL${name} ./gnulib2.c -o ${name}.o; \
          /usr2/local/bin/ar rc gnulib ${name}.o; \
          rm -f ${name}.o; \
        done
_adddi3
ar: bad format (not an object file) in _adddi3.o
sh: 2825 Memory fault - core dumped
*** Error code -117

Stop.

/usr2/local/bin/ar is gnu's ar from binutils.

Just for grins I tried the same command using the native ar instead of
gnu's ar. It worked, but gave me a lot of warnings:

[root:ccel486] /usr2/local/gcc-1.38 (63): ar rc gnulib _adddi3.o
ar: (warning) file _eprintf.o in pre 5.0 format
ar: (warning) file _builtin_new.o in pre 5.0 format
ar: (warning) file _builtin_New.o in pre 5.0 format
ar: (warning) file _builtin_del.o in pre 5.0 format
ar: (warning) file _bb.o in pre 5.0 format
ar: (warning) file _umulsi3.o in pre 5.0 format
ar: (warning) file _mulsi3.o in pre 5.0 format
[ ... many more warnings deleted ... ]

I will continue to hack on it but would appreciate some insights into
what is going on.  As I understand it, I am to build stage1 with the
native tools, then build gcc *with* stage1 with GNU's tools from
binutils.  I assume that the resulting gcc will produce COFF code and
work happily with the native tools.  If I am wrong in these
assumptions please let me know.

Thanks in advance! Please reply through e-mail.

Randy Tidd
rtidd@mwunix.mitre.org
"Follow your path, and let people say
 what they want to."                   - Karl Marx