It looks like around h3Index.c line 344 it tries to take the parent of a res 0 cell at res -1 (res -1 was calculated on line 325), which is where the failure is coming from.

I'm uncertain of this solution, but it passes existing tests and the new failing test.

Basically if the compaction process reaches a base cell, we skip the parent reserved-bit-hit processing and move on to the rest of the logic.

Coverage increased (+0.001%) to 99.035% when pulling f125f68 on jogly:gilley/test-compact-to-res0 into e5209fc on uber:master.