Fuzzy and simulated input compiler tests should have full coverage
Created by: stephanemagnenat
Currently there are 4 classes of compiler tests:
- normal tests, that can fail or succeed
- simulated input tests, using only a subset of tests that can succeed
- fuzzy-input tests, using only a subset of tests that can succeed
- valgrind tests, collecting all tests through a Python script
This presents a number of inconsistencies:
- There is no reason why 2 and 3 only use a subset of tests in 1, and only the succeeding ones.
- While 4 collects all tests, 2 and 3 lists them one by one.
A potential solution would be to sort tests by subdirectory, one for succeeding tests, one for failing tests. Then the CMakeLists.txt could automatically extract all tests from this subdirectory, to build 1, 2 and 3. Similarly, 4 could also use CMake rather then the Python script to run the tests individually.