Hi, despite my issues with recursive data types, I continued with the implementation. I now added the check if the data type arguments can safely be coerced. I do not scan what is in scope for NT values to use, but rather expect the progammer to promise the required witness when he uses "deriving" and the plug the functions together by himself: https://github.com/nomeata/nt-coerce/blob/ec9b5/tests/LiftAbstract.hs I also added a test suite to show what works already: https://github.com/nomeata/nt-coerce/tree/master/tests what does rightfully not work, due to abstraction: https://github.com/nomeata/nt-coerce/tree/master/tests/failing and what does not yet work, although it conceivably should: https://github.com/nomeata/nt-coerce/tree/master/tests/todo The test suite contains expected output in *.out and *.err, so you can check out what happens without having to run it. The error messages are not yet as nice as they should be. How to proceed from here Is this approach and design good enough to to into GHC, despite not being able to handle _all_ cases where it conceivably could, and the slight unwieldiness when coerctions for data types based on abstract data types? BTW, I find it quite nice to be able to develop such a feature without touching GHC’s source. If that is possible for more contributions, the entry barrier to GHC would be noticeably lowered. Thanks, Joachim -- Joachim “nomeata” Breitner mail@joachim-breitner.de • http://www.joachim-breitner.de/ Jabber: nomeata@joachim-breitner.de • GPG-Key: 0x4743206C Debian Developer: nomeata@debian.org