Extend Miri to correctly pass mutable pointers through FFI
Based off of https://github.com/rust-lang/rust/pull/129684, this PR further extends Miri to execute native calls that make use of pointers to *mutable* memory.
We adapt Miri's bookkeeping of internal state upon any FFI call that gives external code permission to mutate memory.
Native code may now possibly write and therefore initialize and change the pointer provenance of bytes it has access to: Such memory is assumed to be *initialized* afterwards and bytes are given *arbitrary (wildcard) provenance*. This enables programs that correctly use mutating FFI calls to run Miri without errors, at the cost of possibly missing Undefined Behaviour caused by incorrect usage of mutating FFI.
> <details>
>
> <summary> Simple example </summary>
>
> ```rust
> extern "C" {
> fn init_int(ptr: *mut i32);
> }
>
> fn main() {
> let mut x = std::mem::MaybeUninit::<i32>::uninit();
> let x = unsafe {
> init_int(x.as_mut_ptr());
> x.assume_init()
> };
>
> println!("C initialized my memory to: {x}");
> }
> ```
> ```c
> void init_int(int *ptr) {
> *ptr = 42;
> }
> ```
> should now show `C initialized my memory to: 42`.
>
> </details>
r? ``@RalfJung``
improve TagEncoding::Niche docs, sanity check, and UB checks
Turns out the `niche_variants` range can actually contain the `untagged_variant`. We should report this as UB in Miri, so this PR implements that.
Also rename `partially_check_layout` to `layout_sanity_check` for better consistency with how similar functions are called in other parts of the compiler.
Turns out my adjustments to the transmutation logic also fix https://github.com/rust-lang/rust/issues/126267.
the behavior of the type system not only depends on the current
assumptions, but also the currentnphase of the compiler. This is
mostly necessary as we need to decide whether and how to reveal
opaque types. We track this via the `TypingMode`.
remove const-support for align_offset and is_aligned
As part of the recent discussion to stabilize `ptr.is_null()` in const context, the general vibe was that it's okay for a const function to panic when the same operation would work at runtime (that's just a case of "dynamically detecting that something is not supported as a const operation"), but it is *not* okay for a const function to just return a different result.
Following that, `is_aligned` and `is_aligned_to` have their const status revoked in this PR, since they do return actively wrong results at const time. In the future we can consider having a new intrinsic or so that can check whether a pointer is "guaranteed to be aligned", but the current implementation based on `align_offset` does not have the behavior we want.
In fact `align_offset` itself behaves quite strangely in const, and that support needs a bunch of special hacks. That doesn't seem worth it. Instead, the users that can fall back to a different implementation should just use const_eval_select directly, and everything else should not be made const-callable. So this PR does exactly that, and entirely removes const support for align_offset.
Closes some tracking issues by removing the associated features:
Closes https://github.com/rust-lang/rust/issues/90962
Closes https://github.com/rust-lang/rust/issues/104203
Cc `@rust-lang/wg-const-eval` `@rust-lang/libs-api`
compiler: Directly use rustc_abi almost everywhere
Use rustc_abi instead of rustc_target where applicable. This is mostly described by the following substitutions:
```rust
match path_substring {
rustc_target::spec::abi::Abi => rustc_abi::ExternAbi,
rustc_target::abi::call => rustc_target::callconv,
rustc_target::abi => rustc_abi,
}
```
A number of spot-fixes make that not quite the whole story.
The main exception is in 33edc68 where I get a lot more persnickety about how things are imported, especially in `rustc_middle::ty::layout`, not just from where. This includes putting an end to a reexport of `rustc_middle::ty::ReprOptions`, for the same reason that the rest of this change is happening: reexports mostly confound things.
This notably omits rustc_passes and the ast crates, as I'm still examining a question I have about how they do stability checking of `extern "Abi"` strings and if I can simplify their logic. The rustc_abi and rustc_target crates also go untouched because they will be entangled in that cleanup.
r? compiler-errors
Operations like is_aligned would return actively wrong results at compile-time,
i.e. calling it on the same pointer at compiletime and runtime could yield
different results. That's no good.
Instead of having hacks to make align_offset kind-of work in const-eval, just
use const_eval_select in the few places where it makes sense, which also ensures
those places are all aware they need to make sure the fallback behavior is
consistent.
The initial naming of "Abi" was an awful mistake, conveying wrong ideas
about how psABIs worked and even more about what the enum meant.
It was only meant to represent the way the value would be described to
a codegen backend as it was lowered to that intermediate representation.
It was never meant to mean anything about the actual psABI handling!
The conflation is because LLVM typically will associate a certain form
with a certain ABI, but even that does not hold when the special cases
that actually exist arise, plus the IR annotations that modify the ABI.
Reframe `rustc_abi::Abi` as the `BackendRepr` of the type, and rename
`BackendRepr::Aggregate` as `BackendRepr::Memory`. Unfortunately, due to
the persistent misunderstandings, this too is now incorrect:
- Scattered ABI-relevant code is entangled with BackendRepr
- We do not always pre-compute a correct BackendRepr that reflects how
we "actually" want this value to be handled, so we leave the backend
interface to also inject various special-cases here
- In some cases `BackendRepr::Memory` is a "real" aggregate, but in
others it is in fact using memory, and in some cases it is a scalar!
Our rustc-to-backend lowering code handles this sort of thing right now.
That will eventually be addressed by lifting duplicated lowering code
to either rustc_codegen_ssa or rustc_target as appropriate.
Then we can rename the _raw functions to drop their suffix, and instead
explicitly use is_stable_const_fn for the few cases where that is really what
you want.
Fundamentally, we have *three* disjoint categories of functions:
1. const-stable functions
2. private/unstable functions that are meant to be callable from const-stable functions
3. functions that can make use of unstable const features
This PR implements the following system:
- `#[rustc_const_stable]` puts functions in the first category. It may only be applied to `#[stable]` functions.
- `#[rustc_const_unstable]` by default puts functions in the third category. The new attribute `#[rustc_const_stable_indirect]` can be added to such a function to move it into the second category.
- `const fn` without a const stability marker are in the second category if they are still unstable. They automatically inherit the feature gate for regular calls, it can now also be used for const-calls.
Also, several holes in recursive const stability checking are being closed.
There's still one potential hole that is hard to avoid, which is when MIR
building automatically inserts calls to a particular function in stable
functions -- which happens in the panic machinery. Those need to *not* be
`rustc_const_unstable` (or manually get a `rustc_const_stable_indirect`) to be
sure they follow recursive const stability. But that's a fairly rare and special
case so IMO it's fine.
The net effect of this is that a `#[unstable]` or unmarked function can be
constified simply by marking it as `const fn`, and it will then be
const-callable from stable `const fn` and subject to recursive const stability
requirements. If it is publicly reachable (which implies it cannot be unmarked),
it will be const-unstable under the same feature gate. Only if the function ever
becomes `#[stable]` does it need a `#[rustc_const_unstable]` or
`#[rustc_const_stable]` marker to decide if this should also imply
const-stability.
Adding `#[rustc_const_unstable]` is only needed for (a) functions that need to
use unstable const lang features (including intrinsics), or (b) `#[stable]`
functions that are not yet intended to be const-stable. Adding
`#[rustc_const_stable]` is only needed for functions that are actually meant to
be directly callable from stable const code. `#[rustc_const_stable_indirect]` is
used to mark intrinsics as const-callable and for `#[rustc_const_unstable]`
functions that are actually called from other, exposed-on-stable `const fn`. No
other attributes are required.
Continue to get rid of `ty::Const::{try_}eval*`
This PR mostly does:
* Removes all of the `try_eval_*` and `eval_*` helpers from `ty::Const`, and replace their usages with `try_to_*`.
* Remove `ty::Const::eval`.
* Rename `ty::Const::normalize` to `ty::Const::normalize_internal`. This function is still used in the normalization code itself.
* Fix some weirdness around the `TransmuteFrom` goal.
I'm happy to split it out further; for example, I could probably land the first part which removes the helpers, or the changes to codegen which are more obvious than the changes to tools.
r? BoxyUwU
Part of https://github.com/rust-lang/rust/issues/130704
Prevent Deduplication of `LongRunningWarn`
Fixes#118612
As mention in the issue, `LongRunningWarn` is meant to be repeated multiple times.
Therefore, this PR stores a unique number in every instance of `LongRunningWarn` so that it's not hashed into the same value and omitted by the deduplication mechanism.
const-eval interning: accept interior mutable pointers in final value
…but keep rejecting mutable references
This fixes https://github.com/rust-lang/rust/issues/121610 by no longer firing the lint when there is a pointer with interior mutability in the final value of the constant. On stable, such pointers can be created with code like:
```rust
pub enum JsValue {
Undefined,
Object(Cell<bool>),
}
impl Drop for JsValue {
fn drop(&mut self) {}
}
// This does *not* get promoted since `JsValue` has a destructor.
// However, the outer scope rule applies, still giving this 'static lifetime.
const UNDEFINED: &JsValue = &JsValue::Undefined;
```
It's not great to accept such values since people *might* think that it is legal to mutate them with unsafe code. (This is related to how "infectious" `UnsafeCell` is, which is a [wide open question](https://github.com/rust-lang/unsafe-code-guidelines/issues/236).) However, we [explicitly document](https://doc.rust-lang.org/reference/behavior-considered-undefined.html) that things created by `const` are immutable. Furthermore, we also accept the following even more questionable code without any lint today:
```rust
let x: &'static Option<Cell<i32>> = &None;
```
This is even more questionable since it does *not* involve a `const`, and yet still puts the data into immutable memory. We could view this as promotion [potentially introducing UB](https://github.com/rust-lang/unsafe-code-guidelines/issues/493). However, we've accepted this since ~forever and it's [too late to reject this now](https://github.com/rust-lang/rust/pull/122789); the pattern is just too useful.
So basically, if you think that `UnsafeCell` should be tracked fully precisely, then you should want the lint we currently emit to be removed, which this PR does. If you think `UnsafeCell` should "infect" surrounding `enum`s, the big problem is really https://github.com/rust-lang/unsafe-code-guidelines/issues/493 which does not trigger the lint -- the cases the lint triggers on are actually the "harmless" ones as there is an explicit surrounding `const` explaining why things end up being immutable.
What all this goes to show is that the hard error added in https://github.com/rust-lang/rust/pull/118324 (later turned into the future-compat lint that I am now suggesting we remove) was based on some wrong assumptions, at least insofar as it concerns shared references. Furthermore, that lint does not help at all for the most problematic case here where the potential UB is completely implicit. (In fact, the lint is actively in the way of [my preferred long-term strategy](https://github.com/rust-lang/unsafe-code-guidelines/issues/493#issuecomment-2028674105) for dealing with this UB.) So I think we should go back to square one and remove that error/lint for shared references. For mutable references, it does seem to work as intended, so we can keep it. Here it serves as a safety net in case the static checks that try to contain mutable references to the inside of a const initializer are not working as intended; I therefore made the check ICE to encourage users to tell us if that safety net is triggered.
Closes https://github.com/rust-lang/rust/issues/122153 by removing the lint.
Cc `@rust-lang/opsem` `@rust-lang/lang`
make writes_through_immutable_pointer a hard error
This turns the lint added in https://github.com/rust-lang/rust/pull/118324 into a hard error. This has been reported in cargo's future-compat reports since Rust 1.76 (released in February). Given that const_mut_refs is still unstable, it should be impossible to even hit this error on stable: we did accidentally stabilize some functions that can cause this error, but that got reverted in https://github.com/rust-lang/rust/pull/117905. Still, let's do a crater run just to be sure.
Given that this should only affect unstable code, I don't think it needs an FCP, but let's Cc ``@rust-lang/lang`` anyway -- any objection to making this unambiguous UB into a hard error during const-eval? This can be viewed as part of https://github.com/rust-lang/rust/pull/129195 which is already nominated for discussion.
Shrink `TyKind::FnPtr`.
By splitting the `FnSig` within `TyKind::FnPtr` into `FnSigTys` and `FnHeader`, which can be packed more efficiently. This reduces the size of the hot `TyKind` type from 32 bytes to 24 bytes on 64-bit platforms. This reduces peak memory usage by a few percent on some benchmarks. It also reduces cache misses and page faults similarly, though this doesn't translate to clear cycles or wall-time improvements on CI.
r? `@compiler-errors`
By splitting the `FnSig` within `TyKind::FnPtr` into `FnSigTys` and
`FnHeader`, which can be packed more efficiently. This reduces the size
of the hot `TyKind` type from 32 bytes to 24 bytes on 64-bit platforms.
This reduces peak memory usage by a few percent on some benchmarks. It
also reduces cache misses and page faults similarly, though this doesn't
translate to clear cycles or wall-time improvements on CI.
