Override and Cleanup
Three code review nits and a follow-up feature. The nits are barely worth a paragraph.
We dropped Math.sqrt from the probe-to-nexus comparison — the assignment only needs relative ordering, and squared distance gives the same result. We cloned the varargs array in setMiningProbesPerBase, which was storing the reference directly — harmless today, wrong at the method boundary. And we added direct test coverage for the zero-nexus case in countProbesPerBase, making it package-private to allow it.
The override problem
The previous entry landed countProbesPerBase on ReplayValidationHarness — ground-truth probe assignment for the replay harness. The next question: what about the emulated game running standalone, after the AI builds a second Nexus?
When the AI builds a second Nexus, miningProbesPerBase stays at {12}. The field doesn’t update; nothing tells it to. Making tick() auto-compute from buildings and units is the obvious fix.
The complication is the replay harness. It injects ground-truth probe counts before each tick via setMiningProbesPerBase. If tick() auto-computes, it clobbers what the harness just set.
Two bad alternatives: a persistent flag callers are supposed to clear (easy to forget), or a separate harness-specific setter (adds a call site that doesn’t need to exist). The clean solution: a one-shot override. setMiningProbesPerBase sets the array and a boolean. tick() checks it — if set, uses the stored value and clears it; if not, auto-computes. The harness calls setMiningProbesPerBase before every tick, so the flag is always set when the harness is in control. Standalone mode never calls it.
if (!miningProbesOverridden) {
miningProbesPerBase = countProbesPerBase(friendly.buildings, friendly.units);
}
miningProbesOverridden = false;
The algorithm moves
With tick() and the harness both needing countProbesPerBase, we moved it to EmulatedGame as a public static method. The harness delegates to it. One copy, correct home.
The test that didn’t read the new contract
zeroProbesYieldsNoMineralGain set the override to zero probes, then ticked 100 times. First tick consumes the override. Ticks 2–100 auto-compute from 12 probes on the initial Nexus — and earn income. The assertion failed.
The test was written when the override persisted indefinitely. It no longer does. One call, one tick, assert.