Support defining C compatible variadic functions
## Summary
Add support for defining C compatible variadic functions in unsafe rust with
`extern "C"` according to [RFC 2137].
## Details
### Parsing
When parsing a user defined function that is `unsafe` and `extern "C"` allow
variadic signatures and inject a "spoofed" `VaList` in the new functions
signature. This allows the user to interact with the variadic arguments via a
`VaList` instead of manually using `va_start` and `va_end` (See [RFC 2137] for
details).
### Codegen
When running codegen for a variadic function, remove the "spoofed" `VaList`
from the function signature and inject `va_start` when the arg local
references are created for the function and `va_end` on return.
## TODO
- [x] Get feedback on injecting `va_start/va_end` in MIR vs codegen
- [x] Properly inject `va_end` - It seems like it should be possible to inject
`va_end` on the `TerminatorKind::Return`. I just need to figure out how
to get the `LocalRef` here.
- [x] Properly call Rust defined C variadic functions in Rust - The spoofed
`VaList` causes problems here.
Related to: #44930
r? @ghost
[RFC 2137]: https://github.com/rust-lang/rfcs/blob/master/text/2137-variadic.md
Function signatures with the `variadic` member set are actually
C-variadic functions. Make this a little more explicit by renaming the
`variadic` boolean value, `c_variadic`.
Deny `async fn` in 2015 edition
This commit prevents code using `async fn` from being compiled in Rust 2015 edition.
Compiling code of the form:
```rust
async fn foo() {}
```
Will now result in the error:
```
error[E0670]: `async fn` is not permitted in the 2015 edition
--> async.rs:1:1
|
1 | async fn foo() {}
| ^^^^^
error: aborting due to error
For more information about an error, try `rustc --explain E0670`.
```
This resolves#58652 and also resolves#53714.
r? @varkor
Fix for issue #58050
Hi,
a quick PR to mention in the compiler error message that `?` is a macro operator, as according to issue #58050
It passed `python x.py test src/tools/tidy` locally, as well as the recommendation to run `/x.py test src/test/ui --stage 1 --bless`.
Let me know if anything else is needed.
Stabilize `unrestricted_attribute_tokens`
In accordance with a plan described in https://internals.rust-lang.org/t/unrestricted-attribute-tokens-feature-status/8561/3.
Delimited non-macro non-builtin attributes now support the same syntax as macro attributes:
```
PATH
PATH `(` TOKEN_STREAM `)`
PATH `[` TOKEN_STREAM `]`
PATH `{` TOKEN_STREAM `}`
```
Such attributes mostly serve as inert proc macro helpers or tool attributes.
To some extent these attributes are de-facto stable due to a hole in feature gate checking (feature gating is done too late - after macro expansion.)
So if macro *removes* such helper attributes during expansion (and it must remove them, unless it's a derive macro), then the code will work on stable.
Key-value non-macro non-builtin attributes are now restricted to bare minimum required to support what we support on stable - unsuffixed literals (https://github.com/rust-lang/rust/issues/34981).
```
PATH `=` LITERAL
```
(Key-value macro attributes are not supported at all right now.)
Crater run in https://github.com/rust-lang/rust/pull/57321 found no regressions for this change.
There are multiple possible ways to extend key-value attributes (https://github.com/rust-lang/rust/pull/57321#issuecomment-451574065), but I'd expect an RFC for that and it's not a pressing enough issue to block stabilization of delimited attributes.
Built-in attributes are still restricted to the "classic" meta-item syntax, nothing changes here.
https://github.com/rust-lang/rust/pull/57321 goes further and adds some additional restrictions (more consistent input checking) to built-in attributes.
Closes https://github.com/rust-lang/rust/issues/55208
Deny the `overflowing_literals` lint for all editions
The `overflowing_literals` was made deny by default for the 2018 edition by #54507, however I'm not aware of any reason it can't be made deny by default for the 2015 edition as well.
Use normal mutable borrows in matches
`ref mut` borrows are currently two-phase with NLL enabled. This changes them to be proper mutable borrows. To accommodate this, first the position of fake borrows is changed:
```text
[ 1. Pre-match ]
|
[ (old create fake borrows) ]
[ 2. Discriminant testing -- check discriminants ] <-+
| |
| (once a specific arm is chosen) |
| |
[ (old read fake borrows) ] |
[ 3. Create "guard bindings" for arm ] |
[ (create fake borrows) ] |
| |
[ 4. Execute guard code ] |
[ (read fake borrows) ] --(guard is false)-----------+
|
| (guard results in true)
|
[ 5. Create real bindings and execute arm ]
|
[ Exit match ]
```
The following additional changes are made to accommodate `ref mut` bindings:
* We no longer create fake `Shared` borrows. These borrows are no longer needed for soundness, just to avoid some arguably strange cases.
* `Shallow` borrows no longer conflict with existing borrows, avoiding conflicting access between the guard borrow access and the `ref mut` borrow.
There is some further clean up done in this PR:
* Avoid the "later used here" note for Shallow borrows (since it's not relevant with the message provided)
* Make any use of a two-phase borrow activate it.
* Simplify the cleanup_post_borrowck passes into a single pass.
cc #56254
r? @nikomatsakis
Special suggestion for illegal unicode curly quote pairs
Fixes#58436
Did not end up expanding the error message span to include the full string literal since I figured the start of the token was the issue, while the help suggestion span would include up to the closing quotation mark.
The look ahead logic does not affect the reader position, not sure if that is an issue (if eg it should still continue to parse after the closing quote without erroring out).
AFAICT, we do not have the same const-eval issues that we used to when
rust-lang/rust#23926 was filed. (Probably because of the switch to
miri for const-evaluation.)
Add a note about 2018e if someone uses `try {` in 2015e
Inspired by https://github.com/rust-lang/rust/issues/58491, where a `try_blocks` example was accidentally run in 2015, which of course produces a bunch of errors.
What's the philosophy about gating for this? The keyword is stably a keyword in 2018, so I haven't gated it for now but am not mentioning what the keyword _does_. Let me know if I should do differently.
Resolves#53672
make generalization code create new variables in correct universe
In our type inference system, when we "generalize" a type T to become
a suitable value for a type variable V, we sometimes wind up creating
new inference variables. So, for example, if we are making V be some
subtype of `&'X u32`, then we might instantiate V with `&'Y u32`.
This generalized type is then related `&'Y u32 <: &'X u32`, resulting
in a region constriant `'Y: 'X`. Previously, however, we were making
these fresh variables like `'Y` in the "current universe", but they
should be created in the universe of V. Moreover, we sometimes cheat
in an invariant context and avoid creating fresh variables if we know
the result must be equal -- we can only do that when the universes
work out.
Fixes#57843
r? @pnkfelix
