Expose the internal `requires_caller_location` query through
rustc_public so tools can detect when an instance has an implicit
extra `&'static Location<'static>` argument in its ABI that is
not present in the MIR body signature.

Fixes rust-lang/rustc_public#123

Add an API to construct the `&'static Location<'static>` constant
that must be passed as the implicit extra argument when calling a
`#[track_caller]` function. Users provide the span of the call
site (or the callee's definition span for reify shim fallback).

Fixes rust-lang/rustc_public#62

Since cfg_select is stable since 1.95, this can be enabled.

Co-authored-by: Noah Lev <[email protected]>

This is probably not what the user wants:

    error: nested Markdown emphasis in HTML `style` tag
      --> $DIR/invalid-html-tags-correct-script-style.rs:11:16
       |
    LL | /// One <style>*emph*</style>
       |                ^^^^^^ Markdown translates this into HTML, but the browser parses it as CSS
       |
    help: to turn off Markdown parsing, put the tag at the start of the line
       |
    LL ~ /// One
    LL ~ /// <style>*emph*</style>
       |

The `test_dir_remove_file` test currently fails under Windows 7 on:

```
thread 'fs::tests::test_dir_remove_file' (2028) panicked at library/std/src/fs/tests.rs:2590:5:
dir.remove_file("foo.txt") failed with: The parameter is incorrect. (os error 87)
```

Looking into it, this is because the `FILE_DISPOSITION_INFO_EX` variant
of the `SetFileInformationByHandle` API is used while it is only
available starting from Windows 10 1607. However,
`FILE_DISPOSITION_INFO` is still available since Vista.

This is therefore fixed by introducing a fallback to the latter API if
the former fails. This is achieved by re-using some logic that is
already available through `File::delete` and that does exactly this.

Signed-off-by: Paul Mabileau <[email protected]>

Add a method Body::caller_location that resolves the correct caller
location constant for a call to a #[track_caller] function, walking
inlined source scopes to handle MIR inlining correctly.

Body now stores source scope data as a public field. Body::new
populates it with empty entries (no inlining info); bodies obtained
from the compiler have full scope data populated.

Span::as_caller_location is now pub(crate) since users should use
Body::caller_location instead, which handles inlining correctly.

In favor of passing raw pointers to inline asm.

Keeping up-to-date

…e lint

The new handling for non-local macros in
`semicolon_in_expressions_from_macros` produces new deny-by-default
lints on various existing crates, which have not previously received
warnings because prior Rust versions suppressed them. Split those out
into a new `semicolon_in_expressions_from_non_local_macros` lint, and
make that one warn-by-default (and FCW warn-in-deps), rather than
deny-by-default, to give the ecosystem time to adapt.

This reintroduces the `is_local` tracking from commit
9192337.

…ros`

The automatic mechanism to produce lint examples won't work, since this
lint needs a separate crate, so add a `rust,ignore` example and manually
provide the compiler output.

This is currently a no-op, but will be useful when const in `global_asm!`
can be pointers.

Currently global_asm already have symbol names when using v0 scheme, this
makes them obtain symbols with legacy scheme too.

This gives the asm-const code the basic ability to deal wiht pointer and
provenances, which lays the ground work for asm_const_ptr.

Note that `SymStatic` is not fully removed, a specialized is kept and
renamed as `SymThreadLocalStatic`, for `#[thread_local]` statics where CTFE
does not support naming. The `#[thread_local]` is unstable feature and it's
not clear if we want to support this in `sym`, but removal of it should be
a separate PR.

With the previous commit, now we can see there are some code duplication
for the handling of `GlobalAlloc` inside backends. Do some clean up to
unify them.

CTFE pointers created via type ID, `without_provenance` or pointers to const
ZSTs can now be codegenned with all 3 backends. These pointers are generated
in the same way as integers.

…_attrs enabled