[siracozmen01]

Problem

When the blast of m3World_Explode sits inside a shape, its rounding included, PushShape in src/destruction.c pushed the body from the blast toward a point of the shape and applied the impulse there. That point was a sphere's centre, the middle of a capsule's segment, or the average of a hull's vertices (#417). It is not the body's centre of mass when a hull is lopsided or the body has more than one shape, so the push spun the body for a reason with no physical meaning. A blast exactly at that point pushed along the shape's local x axis, a direction nothing about the blast picks.

docs/user/destruction.md said the shape "is pushed out through its centroid", which matched neither case. maul2d pushes such a body toward its centre of mass, and the family should behave the same.

Change

A shape with the blast inside it, its rounding included, now pushes its body in the direction from the blast to the body's centre of mass. The impulse is linear, along a line through the centre of mass, so that push spins nothing. A blast inside it at the centre of mass itself, within 1e-5 m, pushes nothing; the body still wakes. The impulse's size is computed as before: impulsePerArea times the area the shape shows along the push.

What changes for which calls:

• m3World_Explode, and a replay of a journaled explosion, can give a different velocity to a body one of whose shapes holds the blast. A shape the blast is outside of is pushed as before.
• No other call changes. The golden scenes, the warm start scenes and the bench scenes keep every line of test/hashes.txt.

M3_SOLVER_REV stays at 21: an explosion is an API call that acts at once, so a snapshot written before this change still steps to the same bits.

docs/user/destruction.md now says: "A shape with the blast inside it, its rounding included, is pushed in the direction from the blast to its body's centre of mass, so that push spins nothing; a blast inside it at the centre of mass itself pushes nothing."

Verification

• TestBlastInside in test/test_explosion.c has three cases:
  • a square pyramid hull, whose vertex average sits at y = 0.4 and its centre of mass at y = 0.5, with the blast at 0.3, 0.1, 0: it moves toward its centre of mass with no spin;
  • a blast at that hull's centre of mass leaves the world hash as it was;
  • a capsule with a ball above it on the same body, so the centre of mass is off the capsule, with the blast in the capsule's rounding past the segment's end: it moves toward the body's centre of mass with no spin.
• With main's src/destruction.c, all three fail.
• test_golden, test_warmhold and test_bench from a gcc-14 Release build give all 20 lines of test/hashes.txt unchanged, before the two helpers were split out of PushShape; the split moves no arithmetic, and the golden tests pass on this head.
• tools/check.sh pr passes in the main checkout on this head: 78 and 76 tests, and fuzz with no finding.

Closes #417

[siracozmen01]

Review at 6f52206: ready. No findings. Read, not run.

• Inside matches the sentence. inside = !(out.distance > 1e-6f && surface > 1e-6f): the blast is in the core (distance about 0), or within the rounding (surface = distance - radius about 0 or below). That is "inside it, its rounding included", with the 1e-6 m tolerance the old branch already used.
• The push is the sentence's.
  • BlastToCenter takes body position + rotated localCenter - blast, the vector from the blast to the body's center of mass, and normalizes it.
  • m3ApplyLinearImpulseInternal applies it with no lever arm, so it spins nothing.
  • Its size still uses ShapeProjectedArea along that direction, brought into the shape's frame with m3InvRotateVec3(xf.q, ...), and FalloffScale is the old arithmetic moved out unchanged.
  • "At the center of mass itself" is a 1e-5 m ball (dot(d, d) <= 1e-10). The body says so, and that is a fair reading of "itself" for a float position.
  • The body is woken before the branch, so even the no-push case wakes it, as "every body the blast reaches wakes" says.
• Multi-shape bodies: PushShape still runs once per shape and pushes that shape's body once.
  • A shape the blast is inside pushes toward the body's center of mass.
  • A shape it is outside of pushes from its nearest point exactly as before (the !inside branch is the old code).
  • So a blast inside one shape of a two-shape body gives one center-of-mass push plus one ordinary push. A blast at the center of mass inside one shape gives that shape nothing and the other its ordinary push.
• The M3_SOLVER_REV reasoning holds for what the rev guards. Its comment says it is in the snapshot config hash, so that a snapshot does not silently diverge. A snapshot holds state, not a pending explosion, and the step itself is unchanged, so every snapshot steps to the same bits. A journal has its own version and no config hash, so a tape recorded before this change replays an explosion with the new push. The body states that, and the feat changelog entry is where a user sees it.