Re: [glob2-devel] Threads ?

[Bo Lorentsen]

Nuage wrote:

> Hello, thank you for your interrest.
>  
I like the AI framework, and like to learn something more about this 
(and maybe play along, too), so it's purely for my own benefit :-)

> > From a technical poit of view, we could still use sepatate threads and
> display every frame exactly once. The idea would be to start the displaying
> process just once the first computation step is finished, and to start the
> next displaying and computation process step only once both are finished.
>               __compute(1)__    __compute(2)__
> __compute(0)__/              \__/              \__
>              \__display(0)__/  \__display(1)__/  \__display(2)__/
>  
I guess this is more and less what I imagined, but I was thinking about 
some kind of display event que or hints, to prevent it
from any kind of unnessesary updates. Only when some elements in the map 
have changed state (I know I need to use more glob2 terms), the display 
will refresh.

The reason for this was also to separate the compute and display thread 
more clearly, but it is very possible that I may have misunderstood some 
of the flow in the code.

> But this would need some more architecture. More important, we would maybe
> loose the greater feature: be able to use "valgrind" wouldn't we ? Anyone
> exactly know, because we really need it!
>  
I have used valgrind on applications that was threaded, and it worked OK.

The other thing to say is that we may implement it in a way that both 
models are possible (two kind of game loops).

Btw. There is a "TODO" list in the CVS, and some of the content is not 
in English ! Is this an outdated TODO list, or is it just secret :-)

/BL