This is the mail archive of the
xconq7@sources.redhat.com
mailing list for the Xconq project.
Re: Fixing repair
>OK, fixing this and another little fix solves the problem, and repair
>seems to work. Enclosed is the new patch.
Excellent. I suggest you check it in.
>> This illustrates the fact that a lot of low level ai code (including
>> the resupply and repair code) really lives in the kernel (plan.c and
>> task.c) rather than in ai.c or the mplayer. It has two
>> consequences. First, we don't need network functions for this code
>> since it runs the same way on all computers.
>
>Is that why the AI doesn't work with network play? Certainly things
>like search_around (called from do_repair_task) involve randomness.
No, randomness should not matter. The random number seed is downloaded from
the master program, so all copies of xconq always get the same result when
they cast the die. Weird but true. I think the main problem with AIs and
network play rather is the fact that they are much faster than humans, and
therefore sometimes cause packet buffer overflow on the master side. In
addition, there is a strange bug in the tcltk version that prevents the AI
from working with non-display sides during network play. I suspect it
involves the event loop, which is rather differently organized from the Mac
version (which lacks this particular bug).
Hans
Hans Ronne
hronne@pp.sbbs.se