Is kInduction imprecise?

[Novak756]

Hi,
I was wondering if the default kInduction analysis is meant to be precise, because running 2ls --inline --k-induction on this safe program gives VERIFICATION FAILED (result should be SUCCESSFUL).
Meanwhile the default analysis (i.e. just 2ls --inline) at least gives VERIFIFICATION INCONCLUSIVE.

If I had to guess it's the combination of pointers and the Ternary Operator that is causing the problem, but I don't know for sure.
(array_comp(((long *)(array_Q_2)),(((unsigned char) (1U & (FV0))) ? (value_store(((long *)(FV26)),(unsigned int) ((FV1)),(unsigned int) ((BubbleSort_Q_0_Q_temp_Q_1)))) : ((long *)(array_Q_1))),ARRAY_SIZE)))

Version: built from latest commit (c572aa1)

Thanks in advance,
Alex

[viktormalik]

Hi, in general, k-induction should not return VERIFICATION FAILED if it cannot find a counterexample reachable within the unwinding limit. We've usually experienced this behaviour when the program contained constructions not supported by 2LS (e.g. calls to various functions from the standard library) when 2LS over-approximated the effect of the construction which lead to an incorrect counterexample. Briefly looking at your example, I don't see any such construction but the program is rather difficult to read.

[peterschrammel]

The counterexample output when running in plain BMC mode with 2ls --inline --havoc --unwind 3 --trace bug.c.c is wrong. So, my suspicion is that there is a bug in the SSA translation.