| .. | ||
| fail | ||
| pass | ||
| Main.hs | ||
| README.org | ||
Overview
This test suite tests that symbolic execution of AArch32 programs works. It is also a convenient place to test the semantics. The test harness looks for binaries in the pass and fail subdirectories. In both cases, it enumerates the functions whose names are prefixed by test_, each of which are symbolically executed with fully symbolic initial states. It treats the return value of each test function as an assertion that it will attempt to prove (i.e., it checks the satisfiability of the assertion). Test cases in pass are expected to return a value that is always true, while the test cases in fail are expected to not always be true (but note that the precise models are not inspected). The test harness uses yices for path satisfiability checking and z3 to prove goals. Since path satisfiability checking is enabled, test cases can include loops as long as they symbolically terminate.
In addition to proving that the result of a test case is always true (or not), the test harness will attempt to prove all generated side conditions for tests with names prefixed by test_and_verify_. In both the pass and fail directories, these side conditions are expected to pass (or they will cause test failures).
Usage
The test runner has two command line options (beyond the defaults for tasty):
--save-macaw: Saves the Macaw IR for each test case to /tmp for debugging purposes--save-smt: Saves the generated SMTLib for each test to /tmp for debugging purposes
Limitations
- It currently tests both optimized an unoptimized binaries. It is intended that this set will expand and that a wide variety of compilers will be tested.
- Only two solvers are involved in the test, rather than a whole matrix
- Function calls are not supported in test cases
- There are no facilities for generating symbolic data beyond the initial symbolic arguments to each function