cash_div_int64() rejects division by zero, but not INT64_MIN / -1. On x86-64 that surfaces as a floating-point exception via SIGFPE; on aarch64 it silently returns INT64_MIN. Multiplication by -1 is already guarded (commit 4f96281587). The attached patch handles division by -1 as negation, matching int8div(), and raises "money out of range" when the dividend is PG_INT64_MIN. Regression tests cover int8/int4/int2.
Tristan looked at the patch on -bugs and suggested posting it here.
Patch attached.
I looked into this patch and it looks clean and complete as per the requirements mentioned.
I don't find any commitfest entry for this, maybe you can add it to get timely attention to this.
