Maybe we should revisit that decision?  I think we did it in the belief that 
some of the debugger interfaces required it (e.g., globalValue).  But recently 
André found a work-around for another case where we were relying on all classes 
being public (the implementation of subclassof) which allows it to work even 
for non-public classes.  In the interest of minimizing the perturbation of 
debugging, perhaps we should remove the 'auto-public' feature (assuming we can 
still debug)?

On 2010-05-27, at 22:53, Henry Minsky wrote:

> I thought we'd be able to let users link their lzo's against
> 'external' lzo's in swf10, even if the compiler flags didn't
> match, but it turns out that if you try to link a non-debug version
> against a debug version, you get compiler errors
> because the debug versions of classes always forces methods to  be
> 'public' , and that means they have different signatures than
> the non-debug versions, as far as the flex compiler is concerned.
> 
> This just means that when someone compiles a swf10 lzo that is
> subclassing stuff from another  external lzo, they'll have
> to make sure they are linking against an lzo which was compiled with
> the same compiler flags.
> 
> I guess forcing all methods to be public is not such a great idea, or
> rather, it will have consequences when
> sublassing from precompiled libraries.
> 
> 
> 
> -- 
> Henry Minsky
> Software Architect
> [email protected]


Reply via email to