Re: 1999-07-12 snapshot - problem with ld on DEC unix

[Olly Betts <olly at muscat dot co dot uk>]

In message <oriu7mv1by.fsf@cupuacu.lsd.dcc.unicamp.br>, Alexandre Oliva writes:
>On Jul 14, 1999, Olly Betts <olly@muscat.co.uk> wrote:
>
>> No symbolic information for: mbchar.o
>
>Are you using `cc' as the bootstrap compiler?

No - I was trying to use a build of egcs 1.1.2 as the bootstrap compiler.

>After the stage1
>compiler is moved to `gcc/stage1' and it is used to compile some
>sources, are you restarting the bootstrap?  I seem to recall error
>messages like this whenever I did that.  I've always had to either
>restart the build from scratch, use gcc as the compiler for stage1 or
>continue from stage2 (with `make bootstrap2')

I've now managed to get a successful build precisely by using cc as the
bootstrap compiler!

`CC=cc ../egcs-19990712/configure' followed by `make bootstrap'.  No need to
restart the build at all.

The problem appears to be that the DEC ld doesn't like an object compiled
from an empty source file generated by the egcs 1.1.2 gcc.  However it has
no problems with an object compiled from an empty source file by the
1999-07-12 snapshot.

Trying nm on the two egcs variants, the snapshot compiled object lists no
symbols while the egcs 1.1.2 compiled file gives:

/usr/bin/nm:
nm: Error: no symbol table in test.o

GNU binutils nm:
No symbols in "test.o".

Delving further, the egcs 1.1.2 build is using GNU as (which I think
binutils said was unsupported last time I tried to build it on the alpha).
So I think this problem was all caused by whoever built our egcs 1.1.2
compiler ignoring that warning.

If my memory is incorrect and GNU as 2.9.1 should work, let me know and I'll
be happy to provide further information.

Cheers,
Olly