Re: Version script help needed

[Franz Sirl <Franz dot Sirl-kernel at lauterbach dot com>]

At 22:00 03.08.2002, H. J. Lu wrote:
> On Sat, Aug 03, 2002 at 07:04:18PM +0200, Franz Sirl wrote:
> > On Samstag, 3. August 2002 18:33, H. J. Lu wrote:
> > > On Sat, Aug 03, 2002 at 05:57:37PM +0200, Franz Sirl wrote:
> > > > One thing makes me wonder though, if I use .symver and a version 
> script,
> > > > why are versioned symbols matched? E. if I use .symver and the 
> following
> > > > version script
> > > >
> > > > VEROLD {
> > > >      local:
> > > >      *;
> > > > };
> > > >
> > > > VERNEW {
> > > >      local:
> > > >      *;
> > > > } VEROLD;
> > > >
> > > > divdi3@VEROLD and divdi3@@VERNEW will be changed from GLOBAL scope to
> > > > LOCAL scope. On the other side version scripts prevent me from from 
> using
> > > > versioned symbols in there? Is this the right behaviour?
> > >
> > > I don't see there is a problem.
> >
> > Hmm, I seem to miss the logic then. If you don't strip mylibgcc2.so, 
> you'll
> > see:
> >
> > [fsirl@enzo:~/test-versioning]$ readelf -a mylibgcc2.so|grep divdi3
> >     31: 00000708    36 FUNC    GLOBAL DEFAULT    9 divdi3@@VERNEW
> >     69: 00000000     0 FILE    LOCAL  DEFAULT  ABS divdi3-old.c
> >     70: 00000000     0 FILE    LOCAL  DEFAULT  ABS divdi3-new.c
> >     78: 000006e4    36 FUNC    LOCAL  DEFAULT    9 divdi3@VEROLD
> >    190: 000006e4    36 FUNC    LOCAL  DEFAULT    9 divdi3_old
> >    200: 00000708    36 FUNC    GLOBAL DEFAULT    9 divdi3
> >
> > So divdi3 got versioned as VEROLD, but it wasn't globalized. On the 
> other hand
> > if I mention divdi3 under VEROLD too in the script (and you can't directly
> > list divdi3@VEROLD in there), I don't get divdi3 at VERNEW, even though 
> it is
> > mentioned there.
> >
> > I guess what I'm trying to say, why isn't the symbol matching in version
> > scripts limited to unversioned symbols and symbols matching the current
> > version node for a particular version node?
>
> I can't guess what you really want. Why not tell us what you want in
> the resulting DSO?
I still want the same as in my first mail in the DSO:

    31: 00000708    36 FUNC    GLOBAL DEFAULT    9 divdi3@@VERNEW
    78: 000006e4    36 FUNC    GLOBAL DEFAULT    9 divdi3@VEROLD

And I wanted to accomplish that without having an explicit .symver for the 
VERNEW symbol version.

I think I just don't understand how the symbol matching in a version script 
works. With this script:

VEROLD {
        global:
          divdi3;
};

VERNEW {
        global:
          divdi3;
        local:
          *;
} VEROLD;

I would have expected matching pseudo code like that:

        for(sym in (all symbols in all objects))
        {
                if (sym exactmatch divdi3@VEROLD)
                {
                        globalize sym;
                        continue;
                }
                if (sym exactmatch divdi3@@VEROLD)
                {
                        globalize sym;
                        continue;
                }
                if (sym exactmatch divdi3@@VERNEW)
                {
                        globalize sym;
                        continue;
                }
                /* check the youngest version child first */
                if (sym exactmatch (divdi3 in global: VERNEW list))
                {
                        attach default version @@VERNEW to sym;
                        globalize sym;
                        continue;
                }
                if (sym exactmatch (divdi3 in global: VEROLD list))
                {
                        attach default version @@VEROLD to sym;
                        globalize sym;
                        continue;
                }
                if (sym exactmatch (local: symbols))
                {
                        localize sym;
                        continue;
                }
                else
                        pass thru sym;
        }

Or maybe a bit simplified:

        for(sym in (all symbols in all objects))
        {
                if (sym is already versioned)
                {
                        pass thru sym;
                        continue;
                }
                /* check the youngest version child first */
                if (sym exactmatch (divdi3 in global: VERNEW list))
                {
                        attach default version @@VERNEW to sym;
                        globalize sym;
                        continue;
                }
                if (sym exactmatch (divdi3 in global: VEROLD list))
                {
                        attach default version @@VEROLD to sym;
                        globalize sym;
                        continue;
                }
                if (sym exactmatch (local: symbols))
                {
                        localize sym;
                        continue;
                }
                else
                        pass thru sym;
        }

I guess the behaviour of symbol matching in version scripts is just badly 
documented.

Anyway, I found relatively low-inasive solution for my GCC versioning 
problem, so I think we'll just wait for the next guy stumbling into 
problems :-).

Franz.