…nnecessarily

…utBorrow

The current order is almost perfectly random. This commit puts them into
a predictable order in their own impl block, going from the highest
level (`Block`) to the lowest (`Expect`). Within each level this is the
order:

- struct_err, err
- struct_span_err, span_err
- create_err, emit_err

The first one in each pair creates a diagnostic, the second one creates
*and* emits a diagnostic. Not every method is present for every level.

The diff is messy, but other than moving methods around, the only thing
it does is create the new `impl DiagCtxt` block with its own comment.

- Remove low-value comments about functionality that is obvious.

- Add missing `track_caller` attributes -- every method should have one.

- Adjust `rustc_lint_diagnostic` attributes. Every method involving a
  `impl Into<DiagnosticMessage>` or `impl Into<SubdiangnosticMessage>`
  argument should have one, except for those producing bugs, which
  aren't user-facing.

Now that error counts can't go up and down due to stashing/stealing, we
have a nice property:

  (err_count > 0) iff (an ErrorGuaranteed has been produced)

So we can now record `ErrorGuaranteed`s within `DiagCtxt` and use that
in methods like `has_error`, instead of checking that the count is
greater than 0 and calling `unchecked_error_guaranteed` to create the
`ErrorGuaranteed`.

In fact, we can record a `Vec<ErrorGuaranteed>` and use its length to
count the number, instead of maintaining a separate count.

Because it also has a `DiagnosticBuilder` arg, which contains a `dcx`
reference.

Also rename some `builder` variables as `diag`, because that's the usual
name.

Now that we have both `delayed_bug` and `span_delayed_bug`, it makes
sense to use the generic term "delayed bug" more.

From `Fatal` to `Error`. It has no functional effect, but `Error` makes
more sense and lines up better with the `Warning` level used just above.

…mpiler-errors

Reorder diagnostics API

The totally random ordering of diagnostic methods in `DiagCtxt` has been low-key driving me crazy for a while now.

r? ``@compiler-errors``

…gnostics-cleanups, r=oli-obk

More internal emit diagnostics cleanups

Miscellaneous improvements.

r? ``@oli-obk``

Gracefully handle non-WF alias in `assemble_alias_bound_candidates_recur`

See explanation in test. I think it's fine to delay a bug here -- I don't believe we ever construct a non-wf alias on the good path? If so, then we can just remove the delay.

Fixes rust-lang#120891

r? lcnr

… r=compiler-errors

Remove a bunch of dead parameters in functions

Found this kind of issue when working on rust-lang#119650
I wrote a trivial toy lint and manual review to find these.

…dtwco

Add test for recently fixed issue

Adds a test for issue rust-lang#116864.

check_consts: fix duplicate errors, make importance consistent

This is stuff I noticed while working on rust-lang#120932, but it's orthogonal to that PR.

r? ``@oli-obk``

improve `btree_cursors` functions documentation

As suggested by ``@Amanieu`` (and others) in rust-lang#107540 (rust-lang#107540 (comment))

Improvements:
- Document exact behavior of `{upper/lower}_bound{,_mut}` with each of the three `Bound` types using unambigous words `{greatest,greater,smallest,smaller,before,after}`.
- Added another doc-example for the `Bound::Unbounded` for each of the methods
- Changed doc-example to use From<[T; N]> rather than lots of `insert()`s which requires a mutable map which clutters the example when `mut` may not be required for the method (such as for `{upper,lower}_bound`.
- Removed `# Panics` section from `insert_{before,after}` methods since they were changed to return an error instead a while ago.
- Reworded some phrases to be more consistent with the more regular `BTreeMap` methods such as calling entries "key-value" rather than "element"s.

…-obk

Check that the ABI of the instance we are inlining is correct

When computing the `CallSite` in the mir inliner, double check that the instance of the function that we are inlining is compatible with the signature from the trait definition that we acquire from the MIR.

Fixes rust-lang#120940

r? ``@oli-obk`` or ``@cjgillot``

…r=GuillaumeGomez

Clean inlined type alias with correct param-env

We were cleaning the `hir::Ty` of a type alias item in the param-env of the item that *references* the type alias, and not the alias's own param-env.

Fixes rust-lang#120954

Add myself to library/std review

I'll see how it goes.

…=compiler-errors

fix ICE for deref coercions with type errors

Follow-up to rust-lang#120895, where I made types with errors go through the full coercion code, which is necessary if we want to build MIR for bodies with errors (rust-lang#120550).

The code for coercing `&T` to `&U` currently assumes that autoderef for `&T` will succeed for at least two steps (`&T` and `T`):

https://github.com/rust-lang/rust/blob/b17491c8f6d555386104dfd82004c01bfef09c95/compiler/rustc_hir_typeck/src/coercion.rs#L339-L464

But for types with errors, we previously only returned the no-op autoderef step (`&{type error}` -> `&{type error}`) and then stopped early. This PR changes autoderef for types with errors to still go through the built-in derefs (e.g. `&&{type error}` -> `&{type error}` -> `{type error}`) and only stop early when it would have to go looking for `Deref` trait impls.

fixes rust-lang#120945

r? ``@compiler-errors`` or compiler