D045 - Checks are chosen from what emulators have actually got wrong

Status: decided, and validated on the first try.

015-sync exists because an emulator's release notes said it had just rewritten its event flags and fixed a thread pool that destroyed threads twice under churn. Both functions were already in this program's census - reported as existing, never called - so the suite said nothing about either while both were broken. An existence test cannot see a behaviour bug, and behaviour bugs are what emulators have.

Written against that release, the checks found:

• event-flag-round-trip fails. A bit that is set, then cleared, still polls as set. In the release that rewrote event flags.
• thread-churn crashes the emulator, about one run in four, inside its own ExitThread assertion. Forty create-and-join cycles; two would never have seen it.

Neither needed a user report, a bug tracker, or anyone's source.

Poll, never wait. sceKernelWaitEventFlag blocks, and a probe that blocks against a platform whose event flags are broken never returns - taking every check behind it with it. sceKernelPollEventFlag asks the same question and comes back either way. Any check aimed at something suspected of being broken has to be written this way.