physics: reconcile exact vehicle collision helper reconstruction

docs/object_matching_logs/claude_physics_vehicle_collision_reconciliation_20260926.md

@@ -0,0 +1,97 @@
+# Physics vehicle collision: independent W6 reconciliation
+
+Canonical baseline: `07fa92af698467a25a2409a56c0e95a51ba65f71`.
+Donor: `09f5208f8a2b8be564519b800f7bdafb20fa19fc`, branch
+`claude/compiler-application-20260925`. The owner requested reconciliation
+and publication if the landing passed. Only the 23-line delta in
+`source/physics/physics.c` is imported, plus this reconciliation record.
+Q13 and every other held source/scorer/ownership packet remain excluded.
+
+## Result
+
+`_physics_compute_vehicle_collision` is independently strict exact:
+**1,160 meaningful code bytes / 1,168 padded bytes / 14 relocations**.
+Normalized SHA256:
+`5aae4029c87d46546c56a16dd66295a34483988ce63e9c2e5ee4db7fe12b6333`.
+
+Halo code advances **1,590,066 -> 1,591,226 / 1,770,166 (89.89%)**;
+credited Halo functions **7,458 -> 7,459 / 7,574**. Complete objects stay
+**388/468**, credited data stays **2,587,011**. Physics has 14 exact and
+3 residual target functions; no whole-object completion is claimed.
+The previously reconciled main update-time function earns no duplicate credit.
+
+## Source and first-party evidence
+
+Independent review inspected the supplied later unoptimized executable as
+data, resolving the helper thunks and inspecting their actual bodies.
+At function `0x7BB7D0`:
+
+- Four calls at `+0x96`, `+0xBC`, `+0xE2`, `+0x108` initialize the live force
+  and torque accumulators with `set_real_vector3d` before their first use.
+- `+0x27C..+0x2A1` corroborates the expression
+  `2.0f * mass_scale * global_gravity / global_physics_collision_depth * penetration`.
+- `+0x320` calls the genuine four-argument `point_from_line3d`, replacing
+  canonical's three hand-expanded component assignments.
+- `+0x3F8` stores the collision flag after both cross products and all four
+  force/torque additions, at the end of the block.
+
+These are meaningful source corrections. No casts, macros, dead locals,
+qualifiers, layouts, artificial dependencies, compiler directives or new UB
+are introduced. The separate initialization is evidenced, not a dummy store.
+The arithmetic change is NOT justified as universally equivalent FP algebra:
+its provenance is independently attested and the final compiler output
+matches January exactly. Later-build evidence does not establish unique
+January source text. The compiler scheduling/VN analysis explains a mechanism,
+not independent source provenance. Coupled changes are justified separately.
+
+Donor evidence remains in
+`research/compiler_application_20260925/workers/W6/`, including V1/V2 cards
+and the failed controls. Supplied `physics.patch` SHA256:
+`86e88ac96dd02b7a611222c8aea2a6fa8c536044331429646adb90182ba42a34`.
+Integrated source equals the donor commit exactly; source SHA256:
+`49cc368300ee9f798af0fef6a268915e41b36501074ae5f774a351bf7d6838e7`.
+
+## Folded inline helper and ownership
+
+The only new defined symbol is `_point_from_line3d`: 48 bytes, zero
+relocations, public storage class 2, function type 0x20, entry offset zero.
+Its raw and normalized bytes equal both January's selected action_charge
+copy and the current compiled action_charge provider. SHA256:
+`9b763841f8519177ac2d3fd471226972b9fa92cf1e9061e9f92359077dc82741`.
+Section flags agree (`0x60501020`). January's physics split neither defines
+nor references it: this uses the existing all-inlined shared-header class,
+not a new ownership exception. The exact caller and full gates satisfy its
+admission conditions; the surplus helper earns no additional code credit.
+
+The compiled caller/provider copies use SELECT_ANY (2), and bounded
+selected-provider duplicate-definition checks pass in both input orders.
+No `/FORCE` is used. Unresolved dependencies are expected in the two-object
+probe; this is not a successful whole-program link or boot claim.
+For completeness, an additional diagnostic linking the raw synthetic January
+split reports six duplicate helpers in either order: its selected sections
+use NODUPLICATES (1). That diagnostic is explicitly NOT recorded as passing
+and does not substitute for the current-source provider checks.
+
+## Fresh gates
+
+- Full ninja before and after: pass; actual objdiff remains 3.3.1.
+- Whole-board stable snapshot: 8,252 owners, **7,630 -> 7,631 exact**;
+  one gain, zero regressions.
+- Whole-TU non-debug inventory: 82 -> 83 sections. Only the target changes,
+  only the disclosed helper is added; all other 81 sections remain equal.
+  All 42 non-code sections, empty COMMON set, and 86 pre-existing named
+  definitions retain their storage/type/value and section flags/selection.
+- Parks: **76 active / 0 stale / 0 invalid**, unchanged. The target had no
+  park to retire; `config/parked.json` remains untouched.
+- Admission unchanged: 10 candidates / 0 contradicted / 1 rejected / 0 revoked.
+- `/W3` diagnostic census unchanged: C4146 x1, C4244 x11, all inherited.
+- Fake scan is identical before/after: 26 inherited board-wide leads.
+- Tools tests before and after: **1,161 passed, 5 skipped, 26 subtests passed**.
+- `git diff --check` clean; source CRLF preserved. No header, config, tooling,
+  compiler/scorer or data changes.
+
+Private receipts are in `scratch/reconcile_physics_collision_20260926/`:
+before source/object, build logs, gates, stable snapshots, semantic reports,
+warning census, parks/admission/fake/test logs, and independent
+`verification.json` with hashes, full section census and both provider orders.
+Reference executables and compiler assets are not included in publication.

source/physics/physics.c

@@ -1093,12 +1093,17 @@ static boolean physics_compute_vehicle_collision(
 	struct vehicle_datum *object0 = vehicle_datum_get(instance0->object_index);
 	struct vehicle_datum *object1 = vehicle_datum_get(instance1->object_index);
 	real mass_scale = square_root(instance0->physics->mass * instance1->physics->mass);
-	real_vector3d force0 = { 0.0f, 0.0f, 0.0f };
-	real_vector3d force1 = { 0.0f, 0.0f, 0.0f };
-	real_vector3d torque0 = { 0.0f, 0.0f, 0.0f };
-	real_vector3d torque1 = { 0.0f, 0.0f, 0.0f };
+	real_vector3d force0;
+	real_vector3d force1;
+	real_vector3d torque0;
+	real_vector3d torque1;
 	short mass_point0_index;
 
+	set_real_vector3d(&force0, 0.0f, 0.0f, 0.0f);
+	set_real_vector3d(&force1, 0.0f, 0.0f, 0.0f);
+	set_real_vector3d(&torque0, 0.0f, 0.0f, 0.0f);
+	set_real_vector3d(&torque1, 0.0f, 0.0f, 0.0f);
+
 	for (mass_point0_index = 0;
 		mass_point0_index < instance0->physics->mass_points.count;
 		mass_point0_index++)
@@ -1132,8 +1137,8 @@ static boolean physics_compute_vehicle_collision(
 			if (distance < radius && distance > 0.0f)
 			{
 				real penetration = (radius - distance) * 0.5f;
-				real force_magnitude = (global_gravity / global_physics_collision_depth) *
-					penetration * mass_scale * 2.0f;
+				real force_magnitude = 2.0f * mass_scale * global_gravity /
+					global_physics_collision_depth * penetration;
 				real_vector3d collision_force0;
 				real_vector3d collision_force1;
 				real_point3d collision_point;
@@ -1144,18 +1149,16 @@ static boolean physics_compute_vehicle_collision(
 
 				scale_vector3d(&direction, -force_magnitude, &collision_force0);
 				scale_vector3d(&direction, force_magnitude, &collision_force1);
-				collision_point.x = point0.x + direction.i * (mass_point0->radius - penetration);
-				collision_point.y = point0.y + direction.j * (mass_point0->radius - penetration);
-				collision_point.z = point0.z + direction.k * (mass_point0->radius - penetration);
+				point_from_line3d(&point0, &direction, mass_point0->radius - penetration, &collision_point);
 				vector_from_points3d(&object0->object.position, &collision_point, &radius0);
 				vector_from_points3d(&object1->object.position, &collision_point, &radius1);
 				cross_product3d(&radius0, &collision_force0, &collision_torque0);
-				collision = TRUE;
 				cross_product3d(&radius1, &collision_force1, &collision_torque1);
 				add_vectors3d(&force0, &collision_force0, &force0);
 				add_vectors3d(&force1, &collision_force1, &force1);
 				add_vectors3d(&torque0, &collision_torque0, &torque0);
 				add_vectors3d(&torque1, &collision_torque1, &torque1);
+				collision = TRUE;
 			}
 		}
 	}