When 0.0 and -0.0 occur in the same compilation unit (e.g. function), the first one to occur gets used in lieu of all occurrences of the other. So, I've added a unittest to confirm this bug, and slightly tweaked an optimization performed in ast.c (_exactly_ along the lines of the similar tweak that's already there in the similar optimization performed in peephole.c) to remove the bug (which was not in the latest released 2.4 but keeps resurfacing -- hopefully no longer, w/the new unittest).