bjorn3 says:
> On errors we don't finalize the incr comp cache, but non-fatal diagnostics are cached afaik.
Otherwise we would have to replay the query in question, which we may not be able to do if the query
key is not reconstructible from the dep node fingerprint.

So we must track these flags to avoid replaying incorrect diagnostics.

`RegionHighlightMode::force_print_trimmed_def_path` can call
`trimmed_def_paths` even when `tcx.sess.opts.trimmed_def_paths` is
false. Based on the `force` in the method name, it seems this is
deliberate, so I have removed the assertion.

Fixes rust-lang#120035.

To expressly warn against the kind of proc macro implementation that was
broken in rust-lang#119875.

Our "true negative" detection assumes that if at least one std handle is a Windows console then no other handle will be a msys2 tty pipe. This turns out to be a faulty assumption in the case of  `/dev/ptmx`.

…ing, r=WaffleLapkin"

This reverts commit 6d2b84b, reversing
changes made to 73bc121.

Fix tty detection for msys2's `/dev/ptmx`

Our "true negative" detection assumes that if at least one std handle is a Windows console then no other handle will be a msys2 tty pipe. This turns out to be a faulty assumption in the case of redirection to  `/dev/ptmx` in an msys2 shell. Maybe this is an msys2 bug but in any case we should try to make it work.

An alternative to this would be to replace the "true negative" detection with an attempt to detect if we're in an msys environment (e.g. by sniffing environment variables) but that seems like it'd be flaky too.

Fixes rust-lang#119658

…mpiler-errors

never_patterns: Count `!` bindings as diverging

A binding that is a never pattern is not reachable, hence counts as diverging code. This allows in particular `fn foo(!: Void) -> SomeType {}` to typecheck.

r? ``@compiler-errors``

Move cmath into `sys`

Part of rust-lang#117276.

r? ``@ChrisDenton``

…ce-resolve-for-coroutines, r=oli-obk

Consolidate logic around resolving built-in coroutine trait impls

Deduplicates a lot of code. Requires defining a new lang item for `Coroutine::resume` for consistency, but it seems not harmful at worst, and potentially later useful at best.

r? oli-obk

Track `verbose` and `verbose_internals`

`verbose_internals` has been UNTRACKED since it was introduced. When i added `verbose` in rust-lang#119129 i made it UNTRACKED as well.

``@bjorn3`` says: rust-lang#119286 (comment)
> On errors we don't finalize the incr comp cache, but non-fatal diagnostics are cached afaik.
Otherwise we would have to replay the query in question, which we may not be able to do if the query key is not reconstructible from the dep node fingerprint.

So we must track these flags to avoid replaying incorrect diagnostics.

r? incremental

…assertion, r=compiler-errors

Fix a `trimmed_def_paths` assertion failure.

`RegionHighlightMode::force_print_trimmed_def_path` can call `trimmed_def_paths` even when `tcx.sess.opts.trimmed_def_paths` is false. Based on the `force` in the method name, it seems this is deliberate, so I have removed the assertion.

Fixes rust-lang#120035.

r? `@compiler-errors`

…cs, r=petrochenkov

Document `Token{Stream,Tree}::Display` more thoroughly.

To expressly warn against the kind of proc macro implementation that was broken in rust-lang#119875.

r? ``@petrochenkov``

…abilization, r=lcnr

Revert stabilization of trait_upcasting feature

Reverts rust-lang#118133

This reverts commit 6d2b84b, reversing changes made to 73bc121.

The feature has a soundness bug:

* rust-lang#120222

It is unclear to me whether we'll actually want to destabilize, but I thought it was still prudent to open the PR for easy destabilization once we get there.