Re: Back to the GC stall in atomic()

[Niklas Frykholm <niklas@...>]

> > My current plan is to see if the finalizer code in Lua can somehow be 
> > simplified. 
>> [...]
> > My finalizers can be 
> > called in any order and at any time during garbage collection, before 
> > the userdata object is freed.
> >
> > I'm thinking (hoping) that with these less strict requirements it would 
> > be possible to get rid of the stall in atomic()
>
> I am not sure this will help. I guess that what causes the stall is
> not that the handling of each userdata is complex, but that all those
> 10,000 userdata must be handled atomically. (The key function here
> is 'luaC_separateudata'.) I think the solution would be to do that
> separation phase incrementally too, but I don't know how to do it.

Yes, it is luaC_separateudate that stalls. But is that separation phase 
really needed if my finalizers "promise" never to create new lua 
objects, touch other lua objects, etc.

Couldn't I just call the finalizer in freeobj()?

(My understanding of Lua's garbage collector is still a bit sketchy.)

// Niklas