CFI: Enable KCFI testing of run-pass tests
This enables KCFI-based testing for all the CFI run-pass tests in the suite today. We can add the test header on top of in-flight CFI tests once they land. This is becoming more important as we get closer to leveraging CFI's multiple type attachment feature, as that is where the implementations will have a divergence.
It also enables KCFI as a sanitizer for x86_64 and aarch64 Linux to make this possible. The sanitizer should likely be available for all aarch64, x86_64, and riscv targets, but that isn't critical for initial testing.
Previously, the documentation for a variant appeared after the documentation
for each of its fields. This was inconsistent with structs and unions, and made
little sense on its own; fields are subordinate to variants and should
therefore appear later in the documentation.
Rollup of 9 pull requests
Successful merges:
- #108675 (Document `adt_const_params` feature in Unstable Book)
- #122120 (Suggest associated type bounds on problematic associated equality bounds)
- #122589 (Fix diagnostics for async block cloning)
- #122835 (Require `DerefMut` and `DerefPure` on `deref!()` patterns when appropriate)
- #123049 (In `ConstructCoroutineInClosureShim`, pass receiver by mut ref, not mut pointer)
- #123055 (enable cargo miri test doctests)
- #123057 (unix fs: Make hurd using explicit new rather than From)
- #123087 (Change `f16` and `f128` clippy stubs to be nonpanicking)
- #123103 (Rename `Inherited` -> `TypeckRootCtxt`)
r? `@ghost`
`@rustbot` modify labels: rollup
Rename `Inherited` -> `TypeckRootCtxt`
`Inherited` is a confusing name. Rename it to `TypeckRootCtxt`.
I don't think this needs a type MCP or anything since it's not nearly as pervasive as `FnCtxt` , for example.
r? `@lcnr` `@oli-obk`
Change `f16` and `f128` clippy stubs to be nonpanicking
It turns out there is a bit of a circular dependency - I cannot add anything to `core` because Clippy fails, and I can't actually add correct Clippy implementations without new implementations from `core`.
Change some of the Clippy stubs from `unimplemented!` to success values and leave a FIXME in their place to mitigate this.
Fixes <https://github.com/rust-lang/rust/issues/122587>
In `ConstructCoroutineInClosureShim`, pass receiver by mut ref, not mut pointer
The receivers were compatible at codegen time, but did not necessarily have the same layouts due to niches, which was caught by miri.
Fixesrust-lang/miri#3400
r? oli-obk
test-cargo-miri: add proc-macro2
This is already in the dependency tree of `serde_derive`, but I guess there is is a host dependency, here it is a target dependency.
The logic is presumably the same as in anyhow, so we don't need both; let's test the one that is more widely used.
It turns out there is a bit of a circular dependency - I cannot add
anything to `core` because Clippy fails, and I can't actually add
correct Clippy implementations without new implementations from `core`.
Change some of the Clippy stubs from `unimplemented!` to success values
and leave a FIXME in their place to mitigate this.
Fixes <https://github.com/rust-lang/rust/issues/122587>
Tree Borrows: Make tree root always be initialized
This PR fixes a slight annoyance we discovered while formally proving that certain optimizations are sound with Tree Borrows. In particular... (copied from the commit message):
There should never be an `Active` but not initialized node in the borrow tree. If such a node exists, and if it has a protector, then on a foreign read, it could become disabled. This leads to some problems when formally proving that read moving optimizations are sound.
The root node is the only node for which this can happen, since all other nodes can only become `Active` when actually used. But special-casing the root node here is annoying to reason about, everything becomes much nicer if we can simply say that *all* `Active` nodes must be initialized. This requires making the root node default-initialized.
This is also more intuitive, since the root arguably becomes initialized during the allocation, which can be considered a write.
There should never be an `Active` but not initialized node in the
borrow tree. If such a node exists, and if it has a protector, then
on a foreign read, it could become disabled. This leads to some
problems when formally proving that read-reordering optimizations are
sound.
The root node is the only node for which this can happen, since all
other nodes can only become `Active` when actually used. But special-
casing the root node here is annoying to reason about, everything
becomes much nicer if we can simply say that *all* `Active` nodes
must be initialized. This requires making the root node default-
initialized.
This is also more intuitive, since the root arguably becomes ini-
tialized during the allocation, which can be considered a write.
This enables KCFI-based testing for all the CFI run-pass tests in the
suite today. We can add the test header on top of in-flight CFI tests
once they land.
It also enables KCFI as a sanitizer for x86_64 and aarch64 Linux to make
this possible. The sanitizer should likely be available for all aarch64,
x86_64, and riscv targets, but that isn't critical for initial testing.
Rollup of 10 pull requests
Successful merges:
- #122707 (Fix a typo in the alloc::string::String docs)
- #122769 (extend comments for reachability set computation)
- #122892 (fix(bootstrap/dist): use versioned dirs when vendoring)
- #122896 (Update stdarch submodule)
- #122923 (In `pretty_print_type()`, print `async fn` futures' paths instead of spans.)
- #122950 (Add regression tests for #101903)
- #123039 (Update books)
- #123042 (Import the 2021 prelude in the core crate)
- #123044 (`Instance` is `Copy`)
- #123051 (did I mention that tests are super cool? )
r? `@ghost`
`@rustbot` modify labels: rollup
Print a backtrace in const eval if interrupted
Demo:
```rust
#![feature(const_eval_limit)]
#![const_eval_limit = "0"]
const OW: u64 = {
let mut res: u64 = 0;
let mut i = 0;
while i < u64::MAX {
res = res.wrapping_add(i);
i += 1;
}
res
};
fn main() {
println!("{}", OW);
}
```
```
╭ ➜ ben@archlinux:~/rust
╰ ➤ rustc +stage1 spin.rs
^Cerror[E0080]: evaluation of constant value failed
--> spin.rs:8:33
|
8 | res = res.wrapping_add(i);
| ^ Compilation was interrupted
note: erroneous constant used
--> spin.rs:15:20
|
15 | println!("{}", OW);
| ^^
note: erroneous constant used
--> spin.rs:15:20
|
15 | println!("{}", OW);
| ^^
|
= note: this note originates in the macro `$crate::format_args_nl` which comes from the expansion of the macro `println` (in Nightly builds, run with -Z macro-backtrace for more info)
error: aborting due to previous error
For more information about this error, try `rustc --explain E0080`.
```
fix(bootstrap/dist): use versioned dirs when vendoring
Currently, if you attempt to run ui tests in a vendored build, you will
see this failure
```
---- [ui] tests/ui/issues/issue-21763.rs stdout ----
diff of stderr:
8 = note: required because it appears within the type `(Rc<()>, Rc<()>)`
9 = note: required for `hashbrown::raw::RawTable<(Rc<()>, Rc<()>)>` to implement `Send`
10 note: required because it appears within the type `hashbrown::map::HashMap<Rc<()>, Rc<()>, RandomState>`
- --> $HASHBROWN_SRC_LOCATION
+ --> /rust/deps/hashbrown/src/map.rs:190:12
12 note: required because it appears within the type `HashMap<Rc<()>, Rc<()>>`
13 --> $SRC_DIR/std/src/collections/hash/map.rs:LL:COL
14 note: required by a bound in `foo`
```
This happens because the code that attempts to remap
`HASHBROWN_SRC_LOCATION` expects it to be under `hashbrown-$version`,
which is the case in a normal cargo registry, but not when vendoring, where
by default crates may not have the version in their directory name.
This change passes `--versioned-dirs` to `cargo vendor` to enforce that
every crate includes the version in the subdir name, which fixes the ui
test and brings `--enable-vendor` builds closer to normal ones.
Rollup of 9 pull requests
Successful merges:
- #122842 (Don't emit an error about failing to produce a file with a specific name if user never gave an explicit name)
- #122881 (Delegation: fix ICE on `bound_vars` divergence)
- #122910 (Validate that we're only matching on unit struct for path pattern)
- #122970 (Use `chunk_by` when building `ReverseSccGraph`)
- #122988 (add even more tests! )
- #122999 (Fix unpretty UI test when /tmp does not exist)
- #123001 (Rename `{enter,exit}_lint_attrs` to `check_attributes{,_post}`)
- #123022 (Add `async-closures/once.rs` back to cranelift tests)
- #123034 (Add a bunch of needs-unwind annotations to tests)
r? `@ghost`
`@rustbot` modify labels: rollup
