Entry 308

Outbreak's sqrt.s in the party's distances: Kite's harness at 0

2026-09-30battlevolumes

Filescrates/piney-battle/src/geom.rscrates/piney-battle/src/party_ai.rscrates/piney-battle/src/flow.rscrates/piney-battle/src/kite.rscrates/piney-battle/src/follow.rscrates/piney-battle/src/ai_move.rscrates/piney-battle/src/fellow.rscrates/piney-battle/src/damage.rscrates/piney-battle/src/skill.rscrates/piney-battle/src/drain.rscrates/piney-battle/src/navi.rscrates/piney-battle/src/entry.rscrates/piney-battle/src/ride.rscrates/piney-battle/src/enemy_ai.rscrates/piney-battle/src/boss.rscrates/piney-battle/src/enemy_motion.rscrates/piney-battle/examples/battle_probecrates/piney-battle/examples/ride_probe.rsdocs/engine/battle.md

Worklog 307 left test_battle_kite_rs.py on Outbreak at 10 of 200 mismatches in main_run and 31 in main_run_ai. Infection and Mutation were at 0 on all ten checks, and still are.

One cause

A deep diff of every mismatching case (scratch script, first differing leaf) put all 41 on one field: word 34 of Kite's AI record, distTg, the port above the game by one unit in the last place (up to 32 where the widths come off a larger distance). The writer is ccAI::DistanceToTarget: INF gcmn 0x00582f50 calls sqrtf at 0x00582ff0; OUT 0x005a5eb0 has sqrt.s inline at 0x005a5f54. Worklog 305 found the same in ccGetDist (OUT main 0x001e6e70).

It is the whole executable, not one function. A scan of every gcmn function for jal sqrtf and for sqrt.s: Infection has 70 functions calling sqrtf (112 calls), Mutation 71 (115). Outbreak and Quarantine have 85 functions each with sqrt.s and no sqrtf symbol at all; on OUT the only callers of __ieee754_sqrtf are libm's acosf and asinf.

The fix

geom::sqrt_on(volume, v) is the one place: newlib sqrtf on INF and MUT, the FPU's truncating sqrt on OUT and QUA. geom::plane_dist and geom::length_on take the volume and go through it. Every sqrtf of piney-battle now does: distance_to_target (both DistanceToTargets), SetTargetDist, CheckNote's reach, FollowPlayer, FollowTarget, FollowTargetDirc, FollowTargetTown, MoveP2P, ccSpcChar::HitCheck, the target searches, the area rules (ground_distance, in_area, ccSkillHold), Data Drain's reach, the landmark searches, ccEntryObj::routine's plDist and the Grunty's ControlMove (RideTables keeps the volume). get_dist_on and the bosses' sqrt_of use it too. INF and MUT compute exactly what they did.

Checks

• test_battle_kite_rs.py, 200 cases of all ten checks: 0 on INF, MUT and OUT (OUT was 10 and 31).
• Other harnesses on OUT, 200 cases, before and after: party AI UseSkill 15 to 0; fellow FollowTarget 1, FollowTargetDirc 6, Main 74 and the six Run* 200 each, all to 0; navi ActInTownRun 88, ActInTownFull 81, BrainsDungeon 85, all to 0. Unchanged: flow skill_main 4 and area_item 35, drain drain 176 and side_effect 1. Enemy AI 0 and enemy motion main 2 of 300, as in worklog 305.
• test_ride_rs.py does not start on OUT (AwakeDistantLight__9WORLD_MANFv is not carried).
• Suites: piney-battle 117 + 5, piney-world 88 + 3 + 12, piney-game 171. Clippy and fmt clean. outbreak_whole_story ends at 408,000 and mutation_whole_story at 506,100, both as before.