Enable internal rustc lints

[BD103]

Turns out our #![warn(rustc::internal)] wasn't doing anything, since -Z unstable-options wasn't being passed to the compiler. This PR enables the internal rustc lints in CI, and leaves a comment so people know how to enable it themselves.

With the lint group now enabled, this PR also fixes any leftover warnings. The biggest note right now is that we iterate over a hash map in some places, even though that iteration is not deterministic. There was one spot that I temporarily #[allow(...)]d, even though it wasn't deterministic, since it seemed to complicated to fix right now.

[DaAlbrecht]

@BD103 Are you sure we want to enable rustc lints? Asking because of this conversation rust-lang/rust#138586 (comment)

[BD103]

@BD103 Are you sure we want to enable rustc lints? Asking because of this conversation rust-lang/rust#138586 (comment)

Actually, no, I didn't see that conversation before. I think some lints may still be useful, but maybe it would be good to avoid enabling the entire rustc::internal group.

If you look at the list of lints, which ones do you think are most useful for us?

[DaAlbrecht]

If you look at the list of lints, which ones do you think are most useful for us?

Hmm, good question. I'm fine with just enabling all and seeing how it feels? Otherwise, these sound most useful to me:

• rustc_lint::internal::BAD_OPT_ACCESS
• rustc_lint::internal::DEFAULT_HASH_TYPES
• rustc_lint::internal::USAGE_OF_QUALIFIED_TY
• rustc_lint::internal::POTENTIAL_QUERY_INSTABILITY (interesting read: https://rustc-dev-guide.rust-lang.org/queries/incremental-compilation.html)

[BD103]

I'm fine with just enabling all and seeing how it feels?

I think I'm fine with this as well. We'll experiment with them all on (except usage_of_ty_tykind, which is already disabled) and turn them off if they becomes a hassle. If there are really annoying bugs, we can make our own versions (#325) or fix them upstream.

(Also that link on incremental compilation was really cool!)

[DaAlbrecht]

could we swap to a BTreeMap?

[BD103]

Great idea! I didn't even think of that! I switched it over to BTreeMap. Looks like rustc_data_structures has SortedMap<K, V>, which may be better, but I'll stick to BTreeMap for now.

[DaAlbrecht]

What an interesting issue this has been! Learned a lot :)