Entry 103

The golf ball's collision step

2026-10-155mplay

Filescrates/rugrats/src/levels/golf.rs crates/rugrats/src/world/collide.rs crates/rugrats/tests/play.rs

With the segment test ported (99), the golf ball's collision step (0x80086ef0, docs/levels/golf.md "Collision step") comes across as documented, in place of the per-axis stand-in. Each frame:

• p' = p + (v dt) >> 6.
• The leading edge e = p' + unit(v) R >> 12, with R = 104.
• The segment from the old position, bent along the floor's plane there: e = s + P d >> 12. P is 0x800763d0(-v, n), the direction along the plane; d is 0x80075f58, 16384 when any axis is that far.
• The segment test on it. A stop (1) plays 137 and bounces off the normal (0x8007666c): with c = N . unit(v) < 0, r = u - (c N >> 11); p' = H - u R >> 12; v = r (|v| 3000 >> 12) >> 12.
• Then the roll as before (0x80086cd0).

collide gains the vector routines it needs:

• unit (0x8007652c: VectorNormal after the 12-bit pre-shift);
• length (0x80076784);
• fix (x y >> 12 toward zero);
• along_plane (0x800763d0's direction);
• floor_normal.

normalize (0x8009f278) is public now.

The doc's one oracle sample, a wall at x 39900 turning (771, 0, 71) into (-551, 0, 49): by hand, the step plus a friction step gives about (-548, 0, 49). The old per-axis bounce gave (-549, 0, 69), so z now agrees. The shot list for golf_course_is_played was searched again (golf_course_search, 68 s). Every hole is played at or under par: 2, 2, 3, 2, 3, 2, 3, 4 and 1 strokes.