mathcast_examples

mathcast_examples — Enhancement-273

Several cmath2.c expression routines converted a double operand to int with a plain cast, which is undefined behaviour for a non-finite or out-of-range value:

• cx_mod (the % operator) — 1e30 % 5 tripped UBSan and, on the shipped build, returned a garbage 2 instead of erroring;
• cx_vector / cx_cvector / cx_unitvec — vector(1e30) / unitvec(1e30) saturated the length to ~INT_MAX, then allocated and filled it: a multi-GB calloc (macOS over-commit) plus a billion-iteration loop — an apparent hang on the shipped build, not just under a sanitizer.

Fix (src/maths/cmaths/cmath2.c): range-check before every (int) cast. % rejects an out-of-range operand with the existing argument out of range for mod error; the vector builders reject a non-representable length via a shared cx_veclen() helper (vector length ... is out of range). Valid expressions are unchanged.

Verify

python3 verify_mathcast.py

Five checks: 1e30 % 5, vector(1e30), and unitvec(1e30) each error cleanly and fast; valid 17 % 5 (= 2) and vector(4) (= [0,1,2,3]) still evaluate correctly.