On 03/08/2025 08:53, Jonas Smedegaard wrote:
I have tried to avoid benchmarking tests to get rid of the dependency on criterion, but then another no-panic test mysteriously (to me) fails.
I wasn't able to reproduce the no-panic failures. However, I will say that no-panic tests are by their nature fragile. Basically no-panic works by setting the panic handler to a symbol that doesn't exist. Then if the code links successfully it means that the compiler eliminated all panics But there is no guarantee that the compiler will actually eliminate all panics that can be eliminated. Particularly if building in debug mode where the optmiser is much less aggressive.
