Re: Visibility and C++ Classes/Templates

[Jason Merrill <jason at redhat dot com>]

Geoffrey Keating wrote:

> > [#pragma visibility affecting explicit instantiations]
>
> A consequence of this is that if a user instantiates a template that
> they don't 'own' (that is, a template from a different module), they
> must make sure that no #pragma is in effect, because the other module
> may have a specific idea about what visibility should be used.

Can you give an example?  You can also get different visibility 
implicitly, if a template argument has more restricted visibility than 
the template; the instantiation gets the more restricted one.

> I'm not sure how exactly they can do this.  Is there a visibility
> pragma for "no setting"?

#pragma GCC visibility pop

> It seems like this would be a common enough mistake that there should
> be a way to get a warning about it.

I think it would be intentional more often than it would be a mistake.

> In the traditional declaration/definition model, if you try to change
> the linkage of something you get an error...

Yes.  If you try to change the visibility of something you also get an 
error.  But an instantiation need not have the same visibility as its 
template.

For instance, code coming from VC++ will tend to default to hidden 
visibility, but make the interfaces default visibility.  If a particular 
instantiation of a template is part of the interface of a non-template 
class, it must also get default visibility.

Jason