I don't know what happened, but that error doesn't show up anymore. I didn't make any compiler changes, so that bug could happen again. Unfortunately (for bug hunting) there is a pass in GCC to shorten branches. That pass automatically reorganizes the code to make optimally shorter jumps. That means forcing a condition where this bug can occur is tricky.

I ran a bunch of tests to try to recreate this problem, but came up short. I might have to just wait until a repeatable test case is found. That makes me nervous, but I'm not sure what else to do.