This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
90 % of test time went into checking every pair of craft against each other.
What was tested
Pairs that cannot affect the rules are now skipped before the check. We confirmed identical output on six reference runs and on five reference results with storm, fog and a link.
What was found then
Results were identical to the last digit, and runs with storms on a slow link went 1.7–2.5 times faster. A further step, a separate search radius for each craft, would change the order in which forces are added up, so results would no longer match exactly. It was shelved.
What changed
Skip pairs that cannot matter (speed only; results unchanged).
Numbers and source references
- 90 %
- time in the pair loopREADME.md:867
- six
- anchorsREADME.md:883
- five
- storm/fog/link fingerprintsREADME.md:883
- 1.7–2.5×
- speed-up on storm-link rowsREADME.md:883; git 4c628d5