Status report, @bozhidar: In trying to isolate a repro I have found that my nrepl client was sometimes relying on a specific order of messages returned for a request. (Or, not really order, but I was resolving my promises too early). I am pretty sure this is the problem, but I will dig a bit further before I am ready to conclude so. It would at least explain why something might have changed w/o anyone complaining, but me. 😃

Roger that! Looking forward to seeing the conclusion of your investigation.

OK, so that was a good exercise for me. The hard Leiningen dep on 0.6.0 meant I had to use other tools and it was about time I checked out clj/deps and Figwheel main, anyway. It was like I had already sort of found out. When there’s an error in an evaluation the ex and err messages come in different order and I had a check for eval-error that terminated my waiting for done and caused a disturbance in the force. But the real havoc was caused by me not realizing that Leiningen did not care about me injecting some other nrepl version, so your guess about a version mismatch was on target, @bozhidar. Now there’s hope for Calva Jack-in again.

I made a little script that runs a sequence of nrepl evals, first on the CLJ repl, then with the CLJS one. Here are the results from running the script on the two sets of nrepl dependencies that I have been trying to use lately.