Re: Latent functions

["Jay Carlson" <nop@...>]

"Thatcher Ulrich" <tu@tulrich.com> writes:

> I was just reading through the list archives about coroutines and
> cooperative multitasking.  Unfortunately, the external C-level
> coroutine approach of 4.1 won't work for me, and I think I'm in the
> same boat with many other game developers.  The problem is, the state
> of Lua scripts must be serializable, and a C-level coroutine library
> (or something using setjmp/longjmp) makes this unworkable.  I'm OK
> with the restricted sleep() as described by Bret Mogilefsky.

This would be useful to me as well.  The LambdaMOO system is based on a
persistent workspace, and it's implemented with a periodic checkpoint of
world state.  It serializes task (thread) state, and I believe that some
servers have been running particular tasks for several years, across
hardware and OS reboots.

I think LambdaMOO is at a local maximum, and I've been toying with the idea
of building something in the spirit of MOO using Lua.  Although MOO survived
many years without coroutines, I think any new system in the niche should
have a plan for how to support tasks.

> I guess what's needed is a power-patch to make OP_CALL non-recursive,
> and remove OP_TAILCALL from the compiler, and then a few support
> routines.  It looks like Bret left some good pointers for how to do
> it.  Before I duplicate this effort for the umpteenth time, does
> anybody have any actual code to share?

Why do you need to remove OP_TAILCALL?

Also, C code that wants to call back into Lua will have to play nice with
this scheme.  Instead of executing lua_call, it will need to push the
function and args on the stack and *return* to the bytecode engine, and let
the engine do the call.

Jay