Patch to revert varasm.c change, fix PRs 10100 & 10053

[Franz Sirl]

At 22:20 30.03.2003, Kaveh R. Ghazi wrote:
>  > From: Mark Mitchell <mark@codesourcery.com>
>  >
>  > > A fair course seems to be to start the patch reversion 48-hour clock
>  > > right now.  That should give someone time enough to take a look and if
>  > > no solution is forthcoming we revert until it can be fixed properly.
>  > >
>  > > Agreed?
>  >
>  > Yes, this is a fine course of action.  Also, if someone will send me
>  > preprocessed source that gets messed up by this change, I will try to
>  > understand why.  I'll also need a command line, and a symbol name that
>  > is handled differently with and without the change.
>
>Thanks for your offer to look into it.
>
>PR 10100 contains a .ii file, target triplet, command line and error
>message (which refers to the problematic symbol.)

Hi,

sorry for the delay, I was on vacation and I'm slowly catching up now.

It turns out that even though I had this patch in my sources for a long 
time, this codepath isn't exercised at all on a lot of platforms cause 
there is yet another macro involved in this, namely MAKE_DECL_ONE_ONLY 
which does a simple (DECL_WEAK (DECL) = 1) on most platforms. Changing this 
macro in config/elfos.h to (declare_weak (DECL)) lets a bootstrap fail on 
linux too.
I can't reconstruct how I missed this back then, but somehow it happened, 
arghhh!

Since I don't think I can debug the underlying problem during the next few 
days, it's fully OK with me if this patch is reverted for now. I'll do it 
myself in a few hours if nobody else beats me.

Franz.