Namely closures and `async` blocks. We have to make a few modifications
to closures to make them compile.

We now require that projection candidates are applicable with the
idenitity substs of the trait, rather than allowing predicates that are
only applicable for certain substs.

This was currently only happening due to eager normalization, which
isn't possible if there's specialization or bound variables.

Given `trait X { type U; }` the bound `<Self as X>::U` now lives
on the type, rather than the trait. This is feature gated on
`feature(generic_associated_types)` for now until more testing can
be done.

The also enabled type-generic associated types since we no longer
need "implies bounds".

This should run much faster.

There are also some drive-by cleanups here to try to simplify things.
Also, the paths for in-tree crates are now displayed as relative
in `x.py test -h -v`.

This commit intends to fix an accidental regression from rust-lang#70846. The
goal of rust-lang#70846 was to build compiler-builtins with a maximal number of
CGUs to ensure that each module in the source corresponds to an object
file. This high degree of control for compiler-builtins is desirable to
ensure that there's at most one exported symbol per CGU, ideally
enabling compiler-builtins to not conflict with the system libgcc as
often.

In rust-lang#70846, however, only part of the compiler understands that
compiler-builtins is built with many CGUs. The rest of the compiler
thinks it's building with `sess.codegen_units()`. Notably the
calculation of `sess.lto()` consults `sess.codegen_units()`, which when
there's only one CGU it disables ThinLTO. This means that
compiler-builtins is built without ThinLTO, which is quite harmful to
performance! This is the root of the cause from rust-lang#73135 where intrinsics
were found to not be inlining trivial functions.

The fix applied in this commit is to remove the special-casing of
compiler-builtins in the compiler. Instead the build system is now
responsible for special-casing compiler-builtins. It doesn't know
exactly how many CGUs will be needed but it passes a large number that
is assumed to be much greater than the number of source-level modules
needed. After reading the various locations in the compiler source, this
seemed like the best solution rather than adding more and more special
casing in the compiler for compiler-builtins.

Closes rust-lang#73135

- Suggest borrowing expression if it would allow cast to work.
- Suggest using `<Type>::from(<expr>)` when appropriate.
- Minor tweak to `;` typo suggestion.

Partily address rust-lang#47136.

… return types

Co-authored-by: lzutao <taolzu@gmail.com>

…akis

add raw_ref macros

In rust-lang#64490, various people were in favor of exposing `&raw` as a macro first before making the actual syntax stable. So this PR (unstably) introduces those macros.

I'll create the tracking issue if we're okay moving forward with this.

…ing-drop, r=matthewjasper,nikomatsakis

Report error when casting an C-like enum implementing Drop

Following approach described in rust-lang#35941

tag/niche terminology cleanup

The term "discriminant" was used in two ways throughout the compiler:
* every enum variant has a corresponding discriminant, that can be given explicitly with `Variant = N`.
* that discriminant is then encoded in memory to store which variant is active -- but this encoded form of the discriminant was also often called "discriminant", even though it is conceptually quite different (e.g., it can be smaller in size, or even use niche-filling).

After discussion with @eddyb, this renames the second term to "tag". The way the tag is encoded can be either `TagEncoding::Direct` (formerly `DiscriminantKind::Tag`) or `TagEncoding::Niche` (formerly `DiscrimianntKind::Niche`).

This finally resolves some long-standing confusion I had about the handling of variant indices and discriminants, which surfaced in rust-lang#72419.

(There is also a `DiscriminantKind` type in libcore, it remains unaffected. I think this corresponds to the discriminant, not the tag, so that seems all right.)

r? @eddyb

…dation, r=nikomatsakis

Projection bound validation

During selection we use bounds declared on associated types (e.g. `type X: Copy`) to satisfy trait/projection bounds. This would be fine so long as those bounds are checked on any impls/trait objects. For simple cases they are because the bound `Self::X: Copy` gets normalized when we check the impl.

However, for default values with specialization and higher-ranked bounds from GATs or otherwise, we can't normalize when checking the impl, and so we use the bound from the trait to prove that the bound applies to the impl, which is clearly unsound.

This PR makes 2 fixes for this:

1. Requiring that the bounds on the trait apply to a projection type with the corresponding substs, so a bound `for<'a> <Self as X<'a>>::U: Copy` on the trait cannot be used to prove `<T as X<'_>>::U: Copy`.
2. Actually checking that the bounds that we still allow apply to generic/default associated types.

Opening for a crater run.

Closes rust-lang#68641
Closes rust-lang#68642
Closes rust-lang#68643
Closes rust-lang#68644
Closes rust-lang#68645
Closes rust-lang#68656

r? @ghost

…, r=nikomatsakis

Further tweak lifetime errors involving `dyn Trait` and `impl Trait` in return position

* Suggest substituting `'static` lifetime in impl/dyn `Trait + 'static` instead of `Trait + 'static + '_`
* When `'static` is explicit, also suggest constraining argument with it
* Reduce verbosity of suggestion message and mention lifetime in label
* Tweak output for overlapping required/captured spans
* Give these errors an error code

Follow up to rust-lang#72543.

r? @nikomatsakis

Fix link error with #[thread_local] introduced by rust-lang#71192

r? @oli-obk

…op, r=pnkfelix

Preserve `Expr`s that have `DefId`s in `ReplaceBodyWithLoop`

This PR fixes rust-lang#71820 as well as the last part of rust-lang#71104 by preserving expressions that are assigned their own `DefId`s (closures and `async` blocks) when passing them to `rustdoc`. This avoids having a `DefId` without a corresponding `HirId`.

The first commit in this PR makes `-Zunpretty=everybody_loops` actually work again, and the subsequent two are miscellaneous cleanup. They should probably get merged regardless of what we end up doing here.

Sample input:
```rust
fn foo() -> Box<i32> {
    let x = |a: i64| {
        const FOO: i64 = 1;
    };

    let a = 4;
    Box::new(a)
}
```

Sample output:
```rust
fn foo() -> Box<i32> {
    || -> !
        {
            const FOO: i64 = 1;
            loop  { }
        };
    loop  { }
}
```

r? @ghost

…ins, r=Mark-Simulacrum

Change how compiler-builtins gets many CGUs

This commit intends to fix an accidental regression from rust-lang#70846. The
goal of rust-lang#70846 was to build compiler-builtins with a maximal number of
CGUs to ensure that each module in the source corresponds to an object
file. This high degree of control for compiler-builtins is desirable to
ensure that there's at most one exported symbol per CGU, ideally
enabling compiler-builtins to not conflict with the system libgcc as
often.

In rust-lang#70846, however, only part of the compiler understands that
compiler-builtins is built with many CGUs. The rest of the compiler
thinks it's building with `sess.codegen_units()`. Notably the
calculation of `sess.lto()` consults `sess.codegen_units()`, which when
there's only one CGU it disables ThinLTO. This means that
compiler-builtins is built without ThinLTO, which is quite harmful to
performance! This is the root of the cause from rust-lang#73135 where intrinsics
were found to not be inlining trivial functions.

The fix applied in this commit is to remove the special-casing of
compiler-builtins in the compiler. Instead the build system is now
responsible for special-casing compiler-builtins. It doesn't know
exactly how many CGUs will be needed but it passes a large number that
is assumed to be much greater than the number of source-level modules
needed. After reading the various locations in the compiler source, this
seemed like the best solution rather than adding more and more special
casing in the compiler for compiler-builtins.

Closes rust-lang#73135

…mulacrum

Speed up bootstrap a little.

The bootstrap script was calling `cargo metadata` 3 times (or 6 with `-v`). This is a very expensive operation, and this attempts to avoid the extra calls. On my system, a simple command like `./x.py test -h -v` goes from about 3 seconds to 0.4.

An overview of the changes:

- Call `cargo metadata` only once with `--no-deps`. Optional dependencies are filtered in `in_tree_crates` (handling `profiler_builtins` and `rustc_codegen_llvm` which are driven by the config).
- Remove a duplicate call to `metadata::build` when using `-v`. I'm not sure why it was there, it looks like a mistake or vestigial from previous behavior.
- Remove check for `_shim`, I believe all the `_shim` crates are now gone.
- Remove check for `rustc_` and `*san` for `test::Crate::should_run`, these are no longer dependencies in the `test` tree.
- Use relative paths in `./x.py test -h -v` output.
- Some code cleanup (remove unnecessary `find_compiler_crates`, etc.).
- Show suite paths (`src/test/ui/...`) in `./x.py test -h -v` output.
- Some doc comments.

…dtwco

Tweak "non-primitive cast" error

- Suggest borrowing expression if it would allow cast to work.
- Suggest using `<Type>::from(<expr>)` when appropriate.
- Minor tweak to `;` typo suggestion.

Partily address rust-lang#47136.