Re: Closing a VM (or "Root State") when running threads

[Rici Lake <lua@...>]

On 25-Aug-04, at 9:01 AM, Matthew Harmon wrote:

> I, too, am wondering if there is a problem shutting down a state that 
> is
> currently yielded.  I can see how that would lock the ability to GC any
> remaining data from that thread.

That doesn't seem very probable to me.

> Another thing I do, which is probably bad, is frequently redefine 
> functions.
> I'm not sure if that leaves the "old" code available for GC right away 
> or
> not.

It is garbage collectable, should work fine.

Most of the times I've been hit by seg faults during lua_close, it has
been related to either running __gc metamethods in the "wrong order" 
(it would
be better to say that the __gc metamethods were badly written) or to 
other
bugs in userdata __gc metamethods. Some userdata are never garbage 
collected
except at lua_close, so these bugs don't manifest themselves earlier.