[DGD] DGD MP and a TLS idea

[Felix A. Croes]

Steve Wooster <sw_mlists at xprt.net> wrote:

>      It occurred to me that what I meant by my description of call_outs 
> being "effectively single-threaded" might not have been clear... From what 
> I can tell, only one call_out can occur at a time. You can't have two 
> different call_outs running on two different processors, since they'll 
> always conflict with each other when checking if they've been suspended. 
> (or is this not true, since they don't necessarily write to variables?)

It's all right to have two threads simultaneously access the same object,
as long as neither modifies it.  But suspension of all callouts in a mud
should be a sufficiently rare event that there isn't much point in
optimizing for it.


>[...]
> I figured it might be useful for cases where you want data stored 
> per-object, but don't want two threads with the same object to conflict. 
> ...maybe I don't know enough about what will or won't cause two threads to 
> conflict with each other.

I am deliberately going to keep this vague since implementation details
which might well change should not influence mudlib design to that extend.
But you should assume that it's fine for a thread to modify the object it
starts in.  Beyond that, the fewer objects modified the better.

Regards,
Dworkin