Re: RFC: Don't output symbol version requirement for non-DT_NEEDED libs

[Alan Modra <amodra at gmail dot com>]

On Fri, Nov 28, 2014 at 12:06:09AM +0000, Joseph Myers wrote:
> On Thu, 27 Nov 2014, Carlos O'Donell wrote:
> 
> > On 11/27/2014 03:16 AM, Alan Modra wrote:
> > > So, absent someone implementing a glibc fix, how about we just drop
> > > the symbol versioning for weak symbols, when their defining library
> > > won't be in DT_NEEDED?  Note that if "f" above was a strong symbol,
> > > ld will still complain with "./libb.so: error adding symbols: DSO
> > > missing from command line".
> > 
> > This seems like the wrong thing to do, particularly since it violates
> > the principle of least surprise. I would expect the versioned symbol
> > to stay versioned.
> > 
> > What's wrong with fixing this in glibc?
> 
> Actually, I think it's a linker bug not a glibc bug.  If you don't link 
> with a library providing a symbol you use, I don't think any information 
> at all about how it might be resolved with some library you didn't link 
> against should be embedded in the binary: not a DT_NEEDED entry, and not a 
> version requirement.  I don't think you can presume at static link time, 
> with a weak undefined symbol like that, "this symbol isn't needed, but if 
> defined at runtime it must have this version" (as opposed to "this symbol 
> isn't needed, and might have any version at runtime", which is the safe 
> assumption).

This is a quibble, but in this particular case we're talking about
a weak *defined* symbol with versioning.  The definition came from a
library dependency that the user didn't specify on the command line,
which is why I call this a quibble because arguably the symbol should
be undefined.  (BFD ld looks at library DT_NEEDED and loads those
libraries.  BFD ld has done that for a long time, but gold doesn't.
That's why the testcase I posted works with gold; With gold the symbol
is left undefined and therefore unversioned.)

-- 
Alan Modra
Australia Development Lab, IBM