SMT solvers in clang SA

Hi all,

Currently studying the impact of SMT solvers on Clang SA. Can anyone help in telling more about the SMT solvers use in clang SA ? ( Have already tested Z3 SMT solver implemented in GSoC 2018 , also looking for other SMT solvers if possible )


We are currently implementing the backends for other solvers (you can follow the progress here: So far we got Boolector, MathSAT and Yices ready. CVC4 should be done soon.

When used to refute bugs, they all give roughly the same results: a ~5% speedup if there are refuted bugs or a ~5% slowdown if no bug is refuted.

I’ve only tried to analyze one full project (tmux) with Yices; the CSA ran for 24hrs and it didn’t complete the analyze. For comparison, it takes 90s to analyze it with the ranged constraint manager in the CSA.


Thanks Mikhail. Can you tell while refuting false positives does inclusion of SMT solvers lead to elimination of some genuine issues by mistake ?

It’s possible and it happened before (see:, but I haven’t seen any true positive being removed lately.