Note: I made those tests pass using (system "... || true") instead of (run ...) but that makes me a bit sad.

Also, if people were willing to give up on the html output, I think we could just get rid of test.sh and switch everything to dune.

I was a bit confused by the in-repo testsuite when I started hacking on ocamlformat. More specifically, I still don't know how to automatically accept a diff, I currently run ./test.sh --meld from the test directory (which I assume is not the most efficient way to do things).

We have a script tools/update_tests.sh that accepts the filenames of the tests (or --all), but it's indeed probably better to let dune take care of these things.

We have a script tools/update_tests.sh that accepts the filenames of the tests (or --all), but it's indeed probably better to let dune take care of these things

I missed that one, I tried the various options of test/test.sh but somehow that always updated more files than I actually wanted, and I'm not sure why.

Anyway, I pushed some more commits, I removed test/test.sh and decided that only dune would take care of the regtests.
The other tests (running on infer, jsoo, etc) are still done through the makefile. I'm not sure it's worth switching that to dune.

The test failure seems to be because test/Makefile calls dune runtest which will try to build in every context, including coverage which is apparently not currently done on travis.

I guess I want to pass a dune-workspace file manually in that case. (Because I didn't see how to select a specific context from the command line).

And I don't know how to tell dune "it's expected to fail, don't worry". @rgrinberg: is that something that dune knows how to handle? Or can I at least ignore the exit code?

There's no good way to handle this at the moment. There was a suggestion to add an action that checks for a particular exit code, but no concrete ticket/rfc. If you'd like this feature, then please make a ticket.