[glob2-devel] Checking for nondeterminism

[Erik Søe Sørensen]

Joe Wells skrev:

> And non-determinism can lead to net games being aborted, so this is
> an important consideration.
>  
Yes, I know...
which is why some seriousish regression testing in this area might be a 
good idea.

I propose (in case this doesn't exist already) an automated way to run 
the following test, e.g. when called as  glob2 --twin-test <savegame> 
<number of steps>:
- $n versions of the game is loaded, on the same host as a default, 
where $n is the number of AI players in the savegame.
- The games commence a multiplayer game, loading the savegame and 
running as each of the AIs.
- Besides the normal inter-game traffic, the games `compare notes', 
comparing the exact locations and states of all the entities involved, 
and perhaps other stuff like the gradients. For instance, after each 
turn a random tenth of the entities are picked and compared.

It would be interesting to see if a test like that would catch the error 
I mentioned...

> I don't really know this part of the code.  I think there are two
> versions of “locked”: one for not-reachable by non-swimmers and one
> for not-reachable by swimmers.  I think this information is used by
> some of the AIs.
>  
Yes, looking at the code I figured 'locked' must mean 'reachable'.
But, reachable from where exactly? Any point on the border of the local 
gradient? That would have been my guess, but that's not how I read the code.

/Erik