Tuesday, December 27, 2011

branch/newcollide : buggy

Voilà environ 3 mois que j'ai entammé la révision du moteur de collision de Bilou, avec pour objectif de permettre la gestion de blocs, plate-formes et autres. Je n'ai pas avancé très vite, mais le code compile et fait tourner AppleAssault ... enfin, presque. Voyez plutôt ....

3 monthes to get a first prototype of the new collision engine. I haven't been very quick on that, but at last, Apple Assault mostly work again ... well ... sort of. I'll let you judge that.
-- btw, I wonder whether desmume-cli --record-movie would be easier to use than byzanz-record for those posts. Please, allow me to debug that on-line:

# stopped by monster-player collision.
statW0->statF2 on found1 (v0 ~ :0 200 ~ :1)
statW1->statF3 on found1 (v0 ~ :0 200 ~ :1)
Makes the appleman bounce when it hurts Bilou
state4..7->statH15 on hit0
\\ [wc 1 ? we 0 < &] (256 ~ :1 512 ~ :0 0 :6 x1)
state4..7->statH16 on hit0
\\ [wc 1 ? we 0 >= &] (256 ~ :1 512 :0 0 :6 x1)

The collision halfly works, on the video above: the applemen indeed bounces and the collision is triggered. But Bilou isn't affected. Afaik, that's because something got wrong in the guardian expression (conditions between the square braces) that can no longer be true in the new collision engine. Before we actually take a transition to the "hit-to-left" (H16) or "hit-to-right" (H15) state, one must detect where the collision came from (test on 'we', other dude's variable #14, which holds centrum-to-centrum horizontal distance), and whether it was actually something that hurts (test on 'wc', the other dude's variable that indicates the matching collision bits on both active and passive collisions area). For some reason,
if (sp>=STACKSIZE-1) return oops("vCTX");
if (c[1].data==0) return oops("!CTX");
if ((op&0xf)>=0xc) stack[sp++]=xcontext[op&0x3];
else stack[sp++]=c[1].data[op&0xf];

Did return at the oops("!CTX") line. In that case, the expression is never true. Bilou won't ever got hurt. I must be missing something that swaps contexts in the code that manages collisions, at the root. Some gc[2]=gc[0]; on line 1742 could do the trick...

that and some silly boolean inversion in the c[1].data =? NULL condition ^^"

No comments: