Re: [PATCH] PR ld/20828: Move symbol version processing ahead of GC symbol sweep

[Alan Modra <amodra at gmail dot com>]

On Wed, Feb 15, 2017 at 12:24:51PM +1030, Alan Modra wrote:
> On Thu, Feb 09, 2017 at 07:48:50PM +0000, Maciej W. Rozycki wrote:
> > 1. Executable is being linked.
> > 
> > 2. Symbol has been defined locally.
> > 
> > 3. Symbol is being dynamically exported.
> > 
> > 4. Symbol will be garbage-collected.
> > 
> > however #4 cannot happen together with #1-#3 at once, because meeting 
> > #1-#3 means the symbol cannot be garbage-collected.  IOW, if we get to 
> > here, then either `->dynindx == -1' and we don't care or `->dynindx != -1' 
> > and this won't change.  So it looks like most of my update can be dropped 
> > and only the part below retained.
> > 
> >  Have I missed anything here?
> 
> I think you are correct, and this update makes the original patch OK
> to apply.  I'm willing to approve it.

So I went and tested the patch, and found failures on x86_64 and other
architectures.

FAIL: Build rdynamic-1
FAIL: vers4a
FAIL: vers9

This says the patch is not OK.

-- 
Alan Modra
Australia Development Lab, IBM