This is part of #13315, split out for easier reviewing. It adds an ocamltest; action in ocamltest, which allows to test ocamltest scripts. This sounds like a generally useful feature that I think we should integrate no matter what happens to #13315.

For example, suppose I want to test that the not <test>; feature from #13315 works properly. I write the following test script:

(* TEST
{ not pass; fail; }
{ not skip; pass; }
*)

The problem is that if I just add this to the testsuite, I will see that this test script succeeds, but I don't know if the feature works as intended. Maybe not foo; always skips, and therefore this test passes without doing what I think it does. I can check that this does not happen myself, by tweaking the test to say { not skip; fail } instead of { not skip; pass }, and checking that that fails. But I cannot commit this to the testsuite. How do I leave evidence in the testsuite that the pass action after the not skip indeed runs?

My solution is that instead of writing this as a negation.ml test, called with the usual driver, I write my test above as a negation.test file, with a semantics.ml driver test that reads as follows:

(* TEST
{
  readonly_files="negation.test negation.reference";
  program="negation.test";
  output="negation.result";
  flags="-e";
  ocamltest;
  reference="negation.reference";
  check-program-output;
}
*)

And then in negation.reference I put the result of running ocamltest -e negation.test, which is mostly unreadable but changes if I change the test or the not implementation. At this point I'm happy.