Thanks for the fix!! (Maybe worth having a regression test in case it changes again? Though I know I'm asking for extra work... Just might be useful to prevent needing to debug this in the future. Might end up saving you more time...)

It's not easy to test...I'd have to set up a fake REPL. Contributions welcome.

If you do ever notice it regressing, just add display(ex) to revise_first and see what changed, then submit a PR.

I think the Linux expect tool is great for this kinda thing. For example one could solve this with:

1. an environment variable set by the tests,
2. => read as a static const DEBUGGING_REPL_EXIT = get(ENV, ...),
3. => enable some kind of tracking around the exit code that gets removed by the compiler otherwise

Then, a REPL test could use Linux expect to: (1) boot up the REPL, then once the prompt appears; (2) trigger an enormous number of code changes (somehow); (3) write "exit()" in the REPL and time (?) the exit time. Or something

In any case, I'm also facing an extreme time deficit so I suppose we shall need to wait for a hero to see this thread and take it up...