We discussed this further, made some simplifications, and Claudia will merge
this now.

I left a comment for the pull request.

Pull request for the fix: https://github.com/aibasel/downward/pull/280

Not sure if I have the time for it today or tomorrow but I would look into it
during the sprint next week.

Thanks! I can now reproduce this.

The cause of the unusual output is that this is not an "error" as such -- a bad input that we catch and report to the user -- but a crash (an unhandled exception in the translator). Someone decided that for crashes we print things with "repr", which is less user-friendly but more informative and reliable for developers than printing with "str". (It does not hide any aspects of the output, including whitespace, and it cannot trigger a separate crash due to unicode encoding issues.) I don't remember why we decided to do this, but there are good reasons for it. These messages are never meant to be seen by a user, so we care less about how pretty it looks.

So the real problem is that the translator crashes in the first place and this error is not caught. I'm changing the title of the issue.

Do you want to look into this?

That makes sense in hindsight, I just realized that I also adjusted the problem
file and not only the domain file. I attached the modified problem file. (I
simply added the newly introduced derived predicate to the goal.)

It doesn't crash for me. Perhaps this is because I used the wrong problem file? Can you also add a problem file?

Error messages produced by the translator are single-line but should be multi-
line. The error message string contains '\n' characters, so these are apparently
not correctly translated into line breaks when displayed on the command line.

I
attached a modified blocksworld domain file that contains unstratifiable axioms
and makes the translator crash. The resulting error message is displayed single-
line instead of multi-line.