Re: Pushing a UserData pointer?

[Rici Lake <lua@...>]

On 16-May-05, at 10:43 PM, Chris Marrin wrote:

> Rici Lake wrote:
>
> > ...
> > If you don't want to go hacking the Lua core, I have another 
> > solution...
> > You say you're reffing the new userdata anyway; this is going to 
> > store a reference in the registry. There's no magic to luaL_ref other 
> > than that. So why not kill two birds with one stone?
> > Instead of using luaL_ref to store a reference in the registry, store 
> > your own reference in the registry using a lightuserdata of the  this 
> >  pointer as a key. Then you can push the lua object onto the stack 
> > using the  this  pointer, without breaking the API and without 
> > incurring much overhead (since the hash of a lud is essentially the 
> > address).
>
> The problem with that is now I would always have a pointer to the 
> userdata and it would never GC, right? When I used my own table to 
> store the lightuserdata key/userdata value pair, I made the values in 
> that table weak so they would get collected properly.
>
> Is there another way to avoid the GC problem?

One of us is not being clear :)

You said:

> This leaves the userdata on the stack, so my Object ctor fiddles with 
> it (gives it a common metatable and refs it so it doesn't get GC'ed). 
> Then it gets popped.

I assumed you meant by "refs it" that you used luaL_ref to create a 
reference to it so that it *wouldn't* be collected. Now you seem to be 
saying that you *want* it to be collected.

You could do either of those things, as Mark points out, by using both 
a weak and a strong table with the weak table backing the strong table 
via an __index meta

But perhaps I still don't understand...