Re: Back to the GC stall in atomic()

[Niklas Frykholm <niklas@...>]

Roberto Ierusalimschy wrote:
> > 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.
>
> Even if you promise not to touch other Lua objects, Lua itself needs
> the metatable of the userdata in order to call its finalizer. If the
> metatable is collected (because the last reference to it came from
> the userdata and it was not marked in the atomic phase), you are in
> trouble...

I'm holding references to all the metatables on the C side, so I should 
be safe from that at least. If there are no other possible issues, it 
might work....

I'll try it. I really need to get rid of those stalls.

// Niklas