Re: C++ and Java CNI: Check Java references

[Andrew Haley <aph at cambridge dot redhat dot com>]

Mark Mitchell writes:
 > >>>>> "Andrew" == Andrew Haley <aph@cambridge.redhat.com> writes:
 > 
 >     Andrew> I can see the logic here, but doesn't it amount to a
 >     Andrew> redesign of CNI?
 > 
 > Maybe -- I'm not by any means a CNI expert.

Yeah, that question was really for the wider audience, especially Per.

 > 
 >     Andrew> would become this:
 > 
 >     Andrew> jobjectArray java::io::File::performList
 >     Andrew> (JavaPointer<java::io::FilenameFilter> filter,
 >     Andrew> JavaPointer<java::io::FileFilter> fileFilter,
 >     Andrew> JavaPointer<java::lang::Class> result_type)
 > 
 > Right.
 > 
 > Note, too, that you could leave the interface alone and copy the
 > parameters to local variables with JavaPointer<...> type, if you
 > didn't want to change the interfaces.
 > 
 > Also, of course, you can make the `operator*' not do the check at all
 > if you have a system where you can recover from the signal handler.

Yes, I can see why this is so neat.

 > This kind of pointer-checking was one of the explicit motivations
 > behind operator overloading and member templates in C++; it's
 > certainly something that C++ programers are used to doing at
 > user-level.

Sure.  I guess this would have been good when CNI was designed,
although I don't think we had full template support at that time.  I'm
trying to make this work without having to break anything else.  :-)

Thanks.

Andrew.