NLS reports same diagnostic message multiple times about nickel HOT 3 CLOSED

Sort + deduplication is probably the way to go. We do this for some responses, just because it makes the output deterministic and therefore easy to test. But apparently we aren't (yet) doing it for diagnostics.

from nickel.

I wonder what's the best way to avoid this case in the recursive setting. I think @jneem mentioned somewhere that it's not as simple as black-holing or marking fields to not re-evaluate them many times (otherwise we would do that instead of having a max evaluation depth, and also because I guess one field might end up having different copies in different merge expressions, and thus with a different environment).

Maybe just compute a cheap hash of diagnostics and de-duplicate them is the easiest way forward.

from nickel.

(Or, put differently, store the diagnostics in an ordered set [or maybe we don't event care about ordering?] with a cheap hasher instead of a vector)

from nickel.