Mutate regexp options

[mbj]

[dkubb]

@mbj does rbx expose an AST that represents the regexp itself? I wonder if it would be possible to mutate that.

[dkubb]

Thinking about this a bit more, I think if we could somehow get the underlying regexp representation it could expose many more kinds of bugs with mutation testing. I was reading through http://olabini.com/blog/2008/06/testing-regular-expressions/ by @olabini and it talks about the difficulties of testing regular expressions; if we could mutation test each "branch" we could expose lots of latent problems.

Also another, somewhat related, mutation would be to do things like change the name of $1, $2, etc variables.. or swapping $1 with $2, and vice versa.

[mbj]

Yeah I plan to mutate the $1 things also the nodes are called Rubinius::AST::NthRef. RBX only shows the regexp as #source attribute that returns string. It would be no problem to use an additional regexp parser to mutate the regexp AST (if any) via the same techniques again.

Will open additional issues for the points outlined here.

[dkubb]

@mbj yeah, through a bit of searching I came across https://github.com/ammar/regexp_parser which I haven't tested, but if it could be used to build an AST to represent the regexp, then we can mutate that. We'd need the equivalent of to_source for regexps to take the mutated AST and produce a valid regexp, but I think it's doable ... probably a lot of work, but I think it would be valuable since regexps are almost always very poorly tested.