AI cars going the wrong way (more often than they should, and not just Bob!)

[scf32]

We're all familiar with the (presumably intended) mechanic of an AI car occasionally going the wrong way to sabotage others.

However in my last 2 or 3 races I've noticed cars from many different teams going crazy and spending almost the whole race going the wrong way. On one occasion I could swear I saw a Reise Wagon turn 180 instantaneously in front of me and start careering towards me on the Free Fall circuit.

Not sure how to reliably reproduce this at the moment but will report in again if it keeps happening.

[Zizin13]

I have not noticed this myself, what OS are you running it on? The AI decisions are chosen with rand() so there may be differences between unix and windows. This decision is made in "changestrategy" in control.c if you want to try debugging it.

[docfo4r]

I know in the compiled version it was mostly, but not always only Bob going the wrong way. I saw almost every team having a member going wrong ways except Zizin and Reise Wagon as far as I remember. Especially Auto Ariel often went the wrong way as well. In rare cases there were 2 AI cars going backwards at the same time.

In the decompiled version, just like Zizin13, I haven't encountered any unexpected sunday driving behaviour.
I took a short look at the code but did not yet get a full overview. As far as I can tell, there is nothing stopping any of the first team member cars to drive the wrong way. I have yet to figure out how it was mostly Bob doing that, as it was stated in the original readme:

Each computer opponent has it's own personality, look out for Bob, he tends to drive backwards and take out other drivers.

[scf32]

I have not noticed this myself, what OS are you running it on? The AI decisions are chosen with rand() so there may be differences between unix and windows. This decision is made in "changestrategy" in control.c if you want to try debugging it.

Built locally on Linux. Thanks I'll have a look there.

[BSzili]

Perhaps rand and srand could be replaced with Watcom's implementation. I remember NBlood had to do this to achieve demo compatibility with the original DOS version.

[Zizin13]

That is a really good idea actually, we should do that.

[Zizin13]

Rand is now basically watcom's implementation, slightly modified to ensure it is identical in unix/windows. Is this still an issue?