Instance registry

[uvlad7]

It's not a some specific bug, actually, I just want to discuss instance registry problems highlighted in the docs.

There could be a window where a Ruby object is marked for deletion but the underlying C++ object is returned back to Ruby

I feel like it's possible to reuse ObjectSpace._id2ref logic here, and take a look on how gc frees T_DATA. Or maybe use weakmap.

Ruby instance may be freed on the C++ side

Is it possible to enable the registry only for Ruby-managed instances? Maybe use Ruby's xmalloc somehow?

[cfis]

The trickiest problem is when a C++ object is freed and a new one is created using the same address. I don't see way that Rice would be able to figure that out.

Any C++ created objects would not use Ruby's xmalloc call on creation - instead they will use the malloc call that they are linked against in the std c++ library.

[uvlad7]

when a C++ object is freed

I don't get it. A C++ object must not be freed as long as Ruby object that wraps it isn't freed, right? And when Ruby object is freed, it's deleted from the registry.

If the same C++ object is returned to Ruby again, a new Ruby object will be created.

[cfis]

Yes in theory. But either Jason or I wrote that text based on what we are seeing.

Maybe give it a try and see what you find.