Files
airdef/tools
chexit 5737b47dc5
build / mod (push) Failing after 59s
record: thin track points on the caller's grid; extreme-load test; re-pin the stale budget stats
The points budget first gated on (recording tick - first) % step, but a strike
carrier's samples fall on its flight-age grid, offset from the recording's
ticks — the gate skipped them (physical_strike_track_keeps_muzzle... failed).
Count the sampling calls per track instead (Track::due): at STEP it pushes
every call exactly as before, and the thinning doubles the interval without
touching the track's phase. Track::points_len exposes the count for diagnostics.

tests/ground_oom.rs gains the continuous extreme load (eight overlapping
40-Oreshnik waves, 1118 objects airborne at once — well past the crash's >200):
peak 26.3 MiB steady + 5.5 MiB rehash transient, the flight tracks thin to
step 16 (91k points, 2.3 MiB) under their 2 MiB budget, the journal and poses
stay within theirs, and the fight still resolves (157 intercepts). The radar
picture itself (11.5 MiB at 1118 airborne) scales with the sky — it is the
simulation's state, not a cache, and is ~3 MiB at the crash's >200 load.

The WASM budget fixture's final /airdef stats pin was stale at the base commit
already: it was measured on engine 8e711f969, and the base pins acfb379aa,
whose host answers shift the two late Smart raids (417/2463, 7 in the air).
Verified identical on the pristine base and on this branch — the memory bounds
do not move that fight; the tick-2102 checkpoint (184/325/2011) is unchanged.
Re-pinned to the measured string and updated BUDGET_ASSERTED_SHA256 so the
raid/snapshot assertions stay active. test-budgets: 283 steps ok, tick fuel
max 184.7M/300M, 0 failures.

Co-Authored-By: Kimi Code <noreply@moonshot.ai>
2026-11-21 17:57:51 +03:00
..