Re: 2.10.91: A problem with LINK_ONCE_DISCARD sections on mipsel-linux

[Alan Modra <alan at linuxcare dot com dot au>]

On Tue, 3 Oct 2000, Maciej W. Rozycki wrote:

> On Tue, 3 Oct 2000, Alan Modra wrote:
> 
> > >  Hmm, maybe we could mark such symbols global?...
> > 
> > Well, that would lead to multiply defined symbol errors for normal
> > (non-linkonce) section symbols.  Actually, after following this path of
> 
>  I meant "section symbols for LINK_ONCE_DISCARD sections only".  I suppose
> I wasn't clear enough.  Due to the way these sections are handled, they
> are not quite local, indeed.  Marking these symbols as global would allow
> them to be looked up in the global symbol hash. 

Yes, that should work.  At least the concept is right.

> > reasoning earlier, I came close to retracting my comment about incorrect
> > output.  Since section symbols must stay local, in some ways it's not
> > unreasonable for the reference to the removed section symbol to resolve to
> > zero.  However, it's confusing, and perhaps against the spirit of
> > link-once sections.  See what happens when we extend your example a
> 
>  There are no symbols within LINK_ONCE_DISCARD sections, normally.  They
> carry a single template each when built by the c++ compiler.  Only the
> section symbol gets referenced from them. 

Hmm, I'm still a little intrigued by why x86 gas behaves as it does.  Oh,
well, something to look at tomorrow.  It's time I hit the hay.

Alan Modra
-- 
Linuxcare.  Support for the Revolution.