On Thu, 10 Mar 2011, Rainer Orth wrote:
> Richard Guenther <[email protected]> writes:
>
> > If I read this patch correctly then
> >
> > 1) it doesn't change the condition under which lto-plugin/ is built
> > (good)
>
> Right.
>
> > 2) it makes -fuse-linker-plugin the default for and only for
> > known good linkers (GNU binutils >= 2.21) (good)
>
> Indeed.
>
> > 3) it makes it impossible to use -fuse-linker-plugin explicitly
> > for other linkers or linkers that were not installed during
> > configuring gcc (bad - esp. the latter)
> >
> > can you please try avoiding 3) at this stage? Or is the whole
> > point of this patch 3) to be able to fix PR46944?
>
> That was my goal: the Solaris linker accepts -p <auditlib>, which causes
> confusion when it is called with -plugin.
>
> > Ideally we'd reject broken linkers at runtime, but that would
> > require some major collect2 massaging (eventually falling back
> > to collect2 or simply reporting an error).
>
> What about making LTO_PLUGIN 3-valued?
>
> 2 linker used has full -plugin support, i.e. gld/gold >= 2.21
> 1 linker has some -plugin support, i.e. gold >= 2.20 < 2.21
> 0 linker has no known plugin support, i.e. everything else, in
> particular vendor linkers
>
> We'd default to -fuse-linker-plugin for 2, accept it if given explicitly
> for 1, and reject it for 0.
>
> This would be similar to the first version of my patch, with the
> difference that we don't try to determine the level of -plugin support
> from trying to run the configured linker with -plugin and check if it
> works, but instead hardcode that knowledge.
>
> > That said, I'm not 100% happy with 3) at this point (though
> > 2) is very desirable).
> >
> > Can we to fix 46944 change the dejagnu require-linker-plugin
> > to check if a linker plugin is used by default and not add
> > -fuse-linker-plugin?
>
> That might be involved since it requires parsing gcc -Wl,-debug output.
> I suppose the solution outlined above is simpler and thus more robust.
It might be as simple as
int res;
int main() { int x; asm ("mov res, %0" : x(g)); return x; }
which should fail to link with the plugin only (but yes, requies
target dependent assembly ...).
Or, use -save-temps and check for the existence of the resolution
file for int main() {}. It should be named t.res for gcc -o t t.c -flto.
Richard.
