nightly CI failures due to finalizers on BigFloats

[lgoettgens]

Just as a heads-up and point to reference to. The issue is already reported upstream at JuliaLang/julia#55965

[thofma]

Not sure if being mutable is part of the API. Anyway, we can adjust the test.

[lgoettgens]

According to the comment in the test file, this test should test JuliaLang/julia#26939. But since this was about BigFloats, and BigFloats can seemingly no longer be used as keys in WeakKeyIdDicts, I think we should just remove the test

[lgoettgens]

Hmm, so JuliaLang/julia#55906 changed the corresponding test in julia base for WeakKeyDict to test with BigInts rather than BigFloats (which then no longer corresponds to JuliaLang/julia#26939...). Let's wait for an answer in the julia issue

[fingolfin]

Well we had that test in there precisely because someone was trying to use a BigFloat as a key there... the old BigFloat implementation had a bug there, which was fixed.

Perhaps WeakKeyDict and WeakKeyIdDict should be enhanced to also support immutable keys (to prevent breakage when changing the implementation of a type, like happened here).

But this all really isn't our concern: we don't use bigfloat keys in weak key (id) dicts, and frankly I don't see why anyone would. So I'd just remove the failing test and be done with this.

[fingolfin]

(Actually I am not even sure it would be possible to have WeakKeyDict work with an immutable key, but again, that's really not important for us here)