Re: Hmm. GCC 2.95 build fails on i586-pc-sco3.2v5.0.2

[Robert Lipe <robertlipe at usa dot net>]

> "  If you choose to configure with --enable-shared you should also
>    specificy --with-gnu-as --disable-multilib even if you are not using
>    the GNU assembler. In doing so you will give up the ability to
>    generate COFF executables as described below. This combination of
>    flags is necessary to suppress negative interactions with multilibing."
> 
> '..even if you are not using the GNU assembler'?  

Oh.   That.    Those instructions are somewhere between misleading
and just plain wrong.

The problem is that --with-gnu-as turns off the code in GCC that
tells the native assembler how to flip between the two supported
object formats.   So now GCC defaults to ELF but the native assembler
defaults to COFF and bad things happen.

We either need to make disable-multilib not disable the part of GCC
that makes it assert the "generate ELF" flag or we just need to doc
that for --enable-shared that requires the GNU assembler.

Kean has a pending patch that might be a better solution.  He now allows
you to invoke a completely different assembler depending upon the
compiler flags.    That's another ticket out of here.


For about six Really Good Beers I'd just go nuts and pull this
multilibbing crap out of OpenServer completely.  If Kean and I knew in
'95 what we know now, we would have killed the COFF beast then and
been free of multilib hell...

> While compiling, I got the following error, but after restarting make all 
> seemed fine.
> xgcc: Internal compiler error: program as got fatal signal 11

I did a lot of builds on a slightly broken SMP system for many months.
The symptom of failure was almost always "random" signal 11's in as when
building EGCS...

RJL