I found an error in the code, it seems to be a bug

[hwh875496]

Within the rayCast function definition in the AABB.ts file, around line 196:
In this section,

for (let f: "x" | "y" = "x"; f !== null; f = (f === "x" ? "y" : null)) {
      if (absD.x < EPSILON) {
        // Parallel.
        if (p[f] < this.lowerBound[f] || this.upperBound[f] < p[f]) {
          return false;
        }
      }
      ...
}

the for loop iterates over both "x" and "y" to analyze the directions along the x-axis and y-axis.
However, at line 209,

if (absD.x < EPSILON) {
  ...
}

it appears that only the value of absD in the x-axis direction is being considered.

I've examined the source code of Box2D, specifically the b2AABB.prototype.RayCast function definition. In it, there are checks for both directions of absD to see if they are less than Number.MIN_VALUE.

b2AABB.prototype.RayCast = function (output, input) {

...

   var absDX = Math.abs(dX);
   var absDY = Math.abs(dY);
   
   if (absDX < Number.MIN_VALUE) {
      if (pX < this.lowerBound.x || this.upperBound.x < pX) return false;
      ...
   }
   
   if (absDY < Number.MIN_VALUE) {
      if (pY < this.lowerBound.y || this.upperBound.y < pY) return false;
      ...
   }

...
}

I'm thinking that in Planck, we should modify line 209 of AABB.ts to if (absD[f] < EPSILON).
I'm not sure if this is the correct approach.

[shakiba]

Hmm, yeah seems like a mistake. Do you have an example to reproduce the bug and verify your fix?

And would you be able to make a PR to fix it?

Also I think instead of the for loop it's better to just duplicate the code for x and y.