Re: Speed issue?

[Eugen-Andrei Gavriloaie <shiretu@...>]

On Mar 26, 2008, at 5:58 AM, Graham Wakefield wrote:

> On Mar 25, 2008, at 6:23 PM, Eugen-Andrei Gavriloaie wrote:
>
> > > > 2. I need a method of keeping some kind of "state" or "session"  
> > > > on each connection. This "session" is
> > > > opaque from c++ point of view. Must be opaque because the user of  
> > > > the server must write his own
> > > > Lua scripts and keep his own state particular for each  
> > > > application. For example, I have app1 and app2.
> > > > App1 must keep an username and a passowrd in his "session" and  
> > > > app2 must keep a counter of some kind.
> > >
> > > How about a coroutine per session: lua_newthread() & lua_resume()?  
> > > Session storage would come for free in the form of any local  
> > > variables.
> > One little issue... each lua_newthread() pushes a lua_State on the  
> > stack of the parent context. Right? And they need to stay there,  
> > otherwise
> > will be garbage collected. On the other hand, stack space is  
> > precious and limited. And I will surly have hundreds of  
> > connections... So, I need to
> > pop the lua_State from the stack and keep it somewhere and make Lua  
> > interpreter understand that is not allowed to garbage collect it.
> >
> > Am I right? Any idea how can I keep this lua_State in the parent  
> > interpreter?
> >
> > Thank you very much for your help!
>
> Well, I wouldn't worry too much a few hundred stack slots... but...
Does the stack size increase by itself? If I do for example (I know  
this is a stupid example, but it works as an example):
L1=lua_newthread(L);
L2=lua_newthread(L);
...
L200=lua_newthread(L);
Will the stack be resized automatically? Or should I make something  
special?


> Typically, to avoid collection, new threads can be placed into the  
> registry, or a designated table in the registry.
>
> E.g. to create:
>
> L2 = lua_newthread(L);
> lua_pushvalue(L, true);
Stack status:  elem1 elem2 ..... elemN thread true [top of the stack]
> lua_settable(L, LUA_REGISTRYINDEX); // pops the thread from the  
> parent state
This is translated into:
t[thread]=true. Is this correct?
> .. now do things with L2, e.g. push a function and lua_resume() it.
>
> When the session expires, you can let garbage collection occur again  
> like this:
>
> lua_pushthread(L2);
> lua_pushnil(L2);
Stack status:  elem1 elem2 ..... elemN thread nil [top of the stack]
> lua_settable(L2, LUA_REGISTRYINDEX);
t[thread]=nil. Is this correct? If it is that means that putting nil  
value into a table removes that pair? Or the pair [thread,nil]  
persists there... Because if it persists, I have a problem...
> Also note that garbage collection and LUA_REGISTRYINDEX is shared  
> between the parent and all child states.

If I understood all this correctly, it will be a huge step forward for  
me...

Thank you so much!