1. 7

  2. 3

    Very interesting work, these tools can really help people take legacy infrastructure forward by not even needing to know how to write the tests, but generating them before you create changes.

    From the results, you can see they get worse coverage, but I think that’s an entirely fair trade-off. Write the hard to discern/annotate tests manually as you create the program, and have things like KLOVER generate as many other tests as possible.

    These symbolic annotations though, wish they had examples of how they are annotated in the original source, they just show what KLOVER expands the functions out to, unless I’m misreading this paper.