Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Perform validation in const_eval_raw, return valtree from const_eval and decouple MIR-embedded consts from the rest #72396

Closed
RalfJung opened this issue May 20, 2020 · 25 comments
Assignees
Labels
A-const-eval Area: constant evaluation (mir interpretation) A-valtree Area: Value trees or fixed by value trees C-cleanup Category: PRs that clean code up or issues documenting cleanup. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. T-lang Relevant to the language team, which will review and decide on the PR/issue.

Comments

@RalfJung
Copy link
Member

RalfJung commented May 20, 2020

Right now, we have two "const_eval" queries: const_eval_raw is internal and returns a "by-reference" constant (MPlaceTy) pointing to some global allocation in tcx.alloc_map. And const_eval is the entry point supposed to be used by the rest of the compiler, and it returns a ty::Const, representing some constants of some types "by-value".

Beyond the by-ref to by-val conversion, const_eval also performs validation of the constant to make sure it matches its type. The fact that const_eval_raw does not do that is a constant source of concern for me, e.g. we have some delay_span_bug in that query that basically rely on validation happening eventually and I think some of those might actually be "exploitable" ICEs.

Furthermore, "validated or not" as a question seems separate from "by-val or by-ref". By-val representation is appropriate for many uses of const (in particular around array lengths and pattern matching), but AFAIK never makes any sense for static. We currently force statics to go through ty::Const which seems odd. In fact, in the context of rust-lang/const-eval#42 and #70889, we determined that pattern matching and const generics likely want a different, even more higher-level representation of const values than what we currently have for ty::Const.

So I propose we perform validation in const_eval_raw (whose name should be changed then... maybe const_eval_ref?). Then all these delay_span_bug issues disappear. But the query should still return an MPlaceTy. The idea is that this is the query used for static and other clients (like Miri) that can handle constants of any type and only care about their value at the byte level.

Then we can have a second query which, like current const_eval (new name maybe const_eval_val?), returns ty::Const. This would only be used for things like array lengths, pattern matching, const generics -- things which need a more "structured" view of consts than MPlaceTy provides. I think this would be a great place to put some check that "all allocations reachable from a const are read-only" (as proposed here). Also, my hypothesis is that this query should only ever be called on "structural match" types -- maybe we can assert that? Long-term, this could become the "integer tree" representation @eddyb proposed.

The main technical hurdle is that we have to avoid query cycles. That's why validation was split out into a separate pass in the first place. We can achieve that by stopping validation when we hit a GlobalAlloc::Static(DefId) allocation. I think that is reasonable -- we already stop there if the DefId is in a different crate, so we cannot rely on this for soundness anyway.

@rust-lang/wg-const-eval does this all make sense?

I think if this works it quite be a really nice cleanup, and I would finally be much less bothered by the existence of two queries and two different representations of consts. :D Some time ago I thought we could kill ty::Constand use by-ref consts for everything, and I think @oli-obk even tried that and it went nowhere. But "by-ref" just seemed like the canonical representations of consts, and we ended up with ugly hack where whether we are in a static affects how we convert MPlaceTy into ty::Const... ugh. Now I finally understand what that is the case: we have two different kinds of consumers of consts in the compiler, with different needs, and they should use different queries with different return types. Some are happy with MPlaceTy, others require a more high-level view. I feel like we finally found the right structure to represent this. :D

Here's that design worked out a bit more.

@jonas-schievink jonas-schievink added A-const-eval Area: constant evaluation (mir interpretation) C-cleanup Category: PRs that clean code up or issues documenting cleanup. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels May 20, 2020
@oli-obk oli-obk self-assigned this May 27, 2020
@RalfJung
Copy link
Member Author

RalfJung commented Jun 20, 2020

With #73513 we have a ty::consts module that could be a good place for ConstValue and anything else specific to ty::Const (possibly even that type itself, which currently lives in ty::sty).

@oli-obk
Copy link
Contributor

oli-obk commented Jul 6, 2020

MCP for the const representation change: rust-lang/compiler-team#323

@RalfJung
Copy link
Member Author

@oli-obk did we agree on names for the two different types representing consts and the corresponding queries? For the types I think we toyed with mir::Const and ty::Const, but that does not leave an obviously good name for the queries. eval_const to return mir::Const and eval_const_for_ty to return ty::Const? The latter is also heavily used for pattern matching though, right?

@oli-obk
Copy link
Contributor

oli-obk commented Jul 27, 2020

the latter query will be rarely invoked I think. Mostly you'll be using the eval method on ty::Const, so your suggestions are fine with me.

@RalfJung
Copy link
Member Author

eval_const_to_valtree would be another option (anticipating rust-lang/compiler-team#323).

@RalfJung RalfJung changed the title Perform validation in const_eval_raw Perform validation in const_eval_raw and return valtree from const_eval Aug 10, 2020
@RalfJung
Copy link
Member Author

My assumption with this plan was that for const items, codegen would work pretty similar to static items, as both have to work with all types (not just valtree-compatible types). But @oli-obk points out in #74949 that e.g. for consts of integer type, we want to emit LLVM integer constants.

Does this also apply to larger valtree-representable types, e.g. a big struct full of integers? We could try to get a valtree and fall back to a raw allocation when that fails. Or maybe we just want to do with this scalars, in which case we could check the layout before deciding if we want to ask for a valtree or a raw allocation.

@RalfJung
Copy link
Member Author

In another thread, @oli-obk wrote something that confused me a lot:

Since we still have (and maybe forever will have) ty::Const as an optimization for literals, even in mir::Const, we need to support this anyway, so I think for now we should just do this for statics where it's unambiguous.

I do not understand this statement. Are you saying mir::Const will not just be an AllocId? I think that is the wrong approach, I'd rather make the caller that has these weird special needs (i.e., codegen for consts) deal with "I want raw representation except sometimes", than making everyone deal with it.

@oli-obk, could you clarify? :)

@oli-obk
Copy link
Contributor

oli-obk commented Aug 10, 2020

mir::Const is a hypothetical thing that we haven't specified, and I think we both have different pictures of it and are thus talking past each other.

We don't need it at all if I'm understanding your proposal correctly, as it would just be a &'tcx Allocation (or AllocId), so we can use that directly instead of having a wrapper.

What we do need is for the mir::Body to contain both ty::Const (for const generics, unevaluated and literals) as well as &'tcx Allocation/AllocId for successfully evaluated constants. This is the main thing I was talking about, essentially the literal field of mir::Constant.

@RalfJung
Copy link
Member Author

RalfJung commented Aug 10, 2020

mir::Const is a hypothetical thing that we haven't specified, and I think we both have different pictures of it and are thus talking past each other.

Yeah that sounds accurate.^^

We don't need it at all if I'm understanding your proposal correctly, as it would just be a &'tcx Allocation (or AllocId), so we can use that directly instead of having a wrapper.

Indeed, I think that's what I had in mind.

What we do need is for the mir::Body to contain both ty::Const (for const generics, unevaluated and literals) as well as &'tcx Allocation/AllocId for successfully evaluated constants. This is the main thing I was talking about, essentially the literal field of mir::Constant.

Okay, so we actually have two places where we do not want to use ty::Const (valtree) or &'tcx Allocation: in mir::Operand, because when constructing we often have a known scalar (or ScalarPair?) value that we want to embed directly without having to create an allocation; and in codegen, because LLVM likewise has a more efficient way to put scalar (ScalarPair?) literals into the code (rather than always using a raw allocation).

I would treat those separately. In particular, codegen is just one of several consumers of the final value of a const, and different backends could have different things that they can more efficiently represent in the target IR. So the LLVM-specific knowledge of what should become an IR literal vs a raw allocation should be handled in codegen.

For mir::Operand::Constant, indeed maybe that's the thing that we should call mir::Const. It would basically be either a valtree or an &Allocation, right? Hm, or it is unevaluated or a paremeter... so a lot like the current ty::ConstValue. So maybe we should have ty::ConstValtree, and then ty::Const is the thing that is "unevaluated or a parameter or an error or a valtree"? And mir::Const would be ty::Const or &Allocation?

But there's still a problem with that... I'd think a ty::Const is something that is evaluable to a valtree, so we shouldn't use it for all general mir::Const that are not evaluated yet. OTOH it might be too confusing to also have an "unevaluated" variant in mir::Const itself. Not sure what is best here. Maybe we need a type like

enum NotSureWhatToCallThis<T> {
  Unevaluated(...),
  Param(...),
  Error(...),
  Evaluated(T)
}

mod ty { type Const = NotSureWhatToCallThis<Valtree>; }
mod mir { type Const = NotSureWhatToCallThis<&'tcx Allocation>; }

Then use ty::Const in types and mir::Const in mir::Operand.

@oli-obk
Copy link
Contributor

oli-obk commented Aug 10, 2020

. I'd think a ty::Const is something that is evaluable to a valtree

I think we're in agreement here, but I think this wording is more precise: It's something that should be evaluable to a valtree and will error if trying to do so fails.

OTOH it might be too confusing to also have an "unevaluated" variant in mir::Const itself. Not sure what is best here. Maybe we need a type like

This is similar to my first attempt, but it failed badly because getting there is very hard with the way ty::Const is hardwired into things like https://doc.rust-lang.org/nightly/nightly-rustc/rustc_middle/mir/visit/trait.MutVisitor.html#method.visit_const and https://doc.rust-lang.org/nightly/nightly-rustc/rustc_middle/ty/fold/trait.TypeVisitor.html#method.visit_const

I'm not sure we can just use NotSureWhatToCallThis<&'tcx Allocation> for mir::Const, as we also have to allow for literals that came from the AST to be in an "optimized" representation. It may be ok to allocate a &'tcx Allocation for every single one of them, but when we tried to do that for ty::Const the regressions were very bad (though that may just be because they are used in the type system). So we can try again, but we should likely make that a step that's completely independent of any other changes.

@RalfJung
Copy link
Member Author

RalfJung commented Aug 10, 2020

It's something that should be evaluable to a valtree and will error if trying to do so fails.

👍

This is similar to my first attempt, but it failed badly because getting there is very hard with the way ty::Const is hardwired into things like

Those would be two different types then, ty::Const and mir::Const respectively for the ty and mir visitors. What is the problem?

I'm not sure we can just use NotSureWhatToCallThis<&'tcx Allocation> for mir::Const, as we also have to allow for literals that came from the AST to be in an "optimized" representation.

Good point, I had forgotten about that. It would be more like

enum MirConstValue {
  Literal(Valtree),
  Alloc(&'tcx Allocation),
}

mod ty { type Const = NotSureWhatToCallThis<Valtree>; }
mod mir { type Const = NotSureWhatToCallThis<MirConstValue>; }

@RalfJung
Copy link
Member Author

RalfJung commented Aug 10, 2020

We don't really need the full Valtree in MirConstValue though, we just need Scalar and ScalarPair. I am not sure what is easier for producers and/or consumers here. I think if we can make it actually just be Scalar and ScalarPair then codegen and Miri do not have to handle valtrees, which would be good. So maybe more like

//! The evaluated result of a constant in the MIR.
enum MirConstValue {
  Scalar(...),
  ScalarPair(...), // or "Slice" or so -- I suppose we cannot entirely avoid this
  Alloc(&'tcx Allocation),
}

mod ty { type Const = NotSureWhatToCallThis<Valtree>; }
mod mir { type Const = NotSureWhatToCallThis<MirConstValue>; }

@RalfJung
Copy link
Member Author

RalfJung commented Aug 10, 2020

I just realized MirConstValue is identical to our current ConstValue. 🤣 The new thing in this design is that this is never used as the return value of a query, it is just used to store constants in the MIR. The queries return either an &Allocation or a Valtree.

@oli-obk
Copy link
Contributor

oli-obk commented Aug 11, 2020

What is the problem?

the change is massive if I do this, I was hoping to give reviewable PRs, but it's a huge huge change with lots of moving parts. I can split it into reviewable commits, but it won't be like each of the commits actually builds successfully

@oli-obk oli-obk closed this as completed Aug 11, 2020
@oli-obk oli-obk reopened this Aug 11, 2020
@RalfJung
Copy link
Member Author

Hm, I don't think I understand why the change would become massive, or rather why it would become more massive than any other proposal which starts distinguishing mir::Const and ty::Const (such that at least one of the visitors has to change). What is it about NotSureWhatToCallThis that makes this harder?

If we want to not distinguish mir::Const and ty::Const, then another thing we could do is replace ConstValue by Valtree, and have an &'tcx Allocation option in the Valtree. That would permit efficient embedding of const literals in the MIR, and it might help pattern matching to deal with non-valtree patterns, but it has several drawbacks:

  • There is no longer a static guarantee that consts embedded in types are "pure" valtrees.
  • Codegen and Miri either have to be able to consume arbitrary valtrees and recursively turn them into raw memory, or we have another not-statically-checked invariant saying that the valtrees in mir::Operand are "shallow" in the sense that they are all of a shape matching the current ConstValue. I'd prefer the second option, but it's not great either.

@oli-obk
Copy link
Contributor

oli-obk commented Aug 11, 2020

What is it about NotSureWhatToCallThis that makes this harder?

I don't think that proposal is per-se harder, but I am currently trying to separate the generics parts of NotSureWhatToCallThis into separate methods, so we don't duplicate code, but maybe I should just go for NotSureWhatToCallThis directly, but without introducing mir::Const. This could be a good first step

but it has several drawbacks:

Yea, and all of these are reasons to go with a scheme like you described, I'm just trying to find a way to get there that isn't as bad as the miri merger (where I did the change from the old constants to the new ones in one go, which caused lots of regressions)

@RalfJung
Copy link
Member Author

RalfJung commented Aug 13, 2020

I think the names we agreed on now are const_eval_alloc(ation) to the low-level query that returns an AllocId, and const_eval_valtree for the higher-level query that returns a valtree (and fails if no valtree can be created).

What is not clear to me yet is how to handle the problem that we permit consts in patterns that cannot be converted to valtrees -- raw pointers and function pointers. I assume here that pattern matching needs to use the valtree query? Well I guess we could keep the "destructure" query (on "allocation" constants) around and use it for patterns... we'll need some code like this anyway for creating the valtree in the first place.

@RalfJung
Copy link
Member Author

@oli-obk and me discussed a bit more the MirConstValue type. Basically, its non-&Allocation variants would only exist to simplify translation from HIR literals. This means they are either integer scalars, an AllocId, or a wide ptr of AllocId+length. That is simpler than what we have currently, which supports arbitrary pointers (which is needed for pattern matching).

@RalfJung RalfJung changed the title Perform validation in const_eval_raw and return valtree from const_eval Perform validation in const_eval_raw, return valtree from const_eval and decouple MIR-embedded consts from the rest Aug 14, 2020
@RalfJung
Copy link
Member Author

RalfJung commented Aug 14, 2020

I had another long chat with @oli-obk, here's the design that we are at right now as I understood it. Most of the open questions are around pattern matching really -- we want to make sure that patterns can share infrastructure with const generics as much as is possible and makes sense.

@RalfJung RalfJung added the T-lang Relevant to the language team, which will review and decide on the PR/issue. label Aug 14, 2020
Dylan-DPC-zz pushed a commit to Dylan-DPC-zz/rust that referenced this issue Sep 19, 2020
…RalfJung

Validate constants during `const_eval_raw`

This PR implements the groundwork for rust-lang#72396

* constants are now validated during `const_eval_raw`
* to prevent cycle errors, we do not validate references to statics anymore beyond the fact that they are not dangling
* the `const_eval` query ICEs if used on `static` items
* as a side effect promoteds are now evaluated to `ConstValue::Scalar` again (since they are just a reference to the actual promoted allocation in most cases).
bors added a commit to rust-lang-ci/rust that referenced this issue Sep 20, 2020
…lfJung

Validate constants during `const_eval_raw`

This PR implements the groundwork for rust-lang#72396

* constants are now validated during `const_eval_raw`
* to prevent cycle errors, we do not validate references to statics anymore beyond the fact that they are not dangling
* the `const_eval` query ICEs if used on `static` items
* as a side effect promoteds are now evaluated to `ConstValue::Scalar` again (since they are just a reference to the actual promoted allocation in most cases).
@oli-obk
Copy link
Contributor

oli-obk commented Sep 26, 2020

I'm wondering whether we should also have LitBytes ([u8; N]) and LitSlice (&[u8] or &str) variants on ValTree, as I'm considering it very likely that we'll want to use strings in const generics. While we could encode them as aggregates, that would be very inefficient (storage wise), and probably annoying to handle. That said, we can start out with encoding them as aggregates and seeing how it goes. If anyone starts putting huge strings in their types, we may just give up (or set a configurable limit like with the recursion depth checks).

@RalfJung
Copy link
Member Author

RalfJung commented Sep 27, 2020

I think having LitByteSlice at least makes sense... that could cover both arrays and slices (both byte and string slices) I think.

We then need to be careful to never use the normal aggregate representation for these types, as that would break equality checking.

@oli-obk
Copy link
Contributor

oli-obk commented Sep 28, 2020

Ugh yea, I forgot about the equality part. I guess let's go with full expansion at first and then check whether doing an optimization like that actually has any benefits

@RalfJung
Copy link
Member Author

I would be very surprised if this was not necessary to avoid some enormous regressions, but let's see.

@oli-obk
Copy link
Contributor

oli-obk commented Sep 28, 2020

Since this will only be used for constants in the type system, I don't see what this could regress. We don't have any constants but usize in the type system right now (modulo const generics), and const generics other than usize aren't stable or perf tested.

bors added a commit to rust-lang-ci/rust that referenced this issue Nov 4, 2020
Refactorings in preparation for const value trees

cc rust-lang#72396

This PR changes the `Scalar::Bits { data: u128, size: u8 }` variant to `Scalar::Bits(ScalarInt)` where `ScalarInt` contains the same information, but is `repr(packed)`. The reason for using a packed struct is to allow enum variant packing to keep the original size of `Scalar` instead of adding another word to its size due to padding.
Other than that the PR just gets rid of all the inspection of the internal fields of `Scalar::Bits` which were frankly scary. These fields have invariants that we need to uphold and we can't do that without making the fields private.

r? `@ghost`
bors added a commit to rust-lang-ci/rust that referenced this issue Dec 27, 2020
…ination, r=varkor

stabilize `#![feature(min_const_generics)]` in 1.51

*A new Kind*
*A Sort long Prophesized*
*Once Fragile, now Eternal*

blocked on rust-lang#79073.

# Stabilization report

This is the stabilization report for `#![feature(min_const_generics)]` (tracking issue rust-lang#74878), a subset of `#![feature(const_generics)]` (tracking issue rust-lang#44580), based on rust-lang/rfcs#2000.

The [version target](https://forge.rust-lang.org/#current-release-versions) is ~~1.50 (2020-12-31 => beta, 2021-02-11 => stable)~~ 1.51 (2021-02-111 => beta, 2021-03-25 => stable).

This report is a collaborative effort of `@varkor,` `@shepmaster` and `@lcnr.`

## Summary

It is currently possible to parameterize functions, type aliases, types, traits and implementations by types and lifetimes.
With `#![feature(min_const_generics)]`, it becomes possible, in addition, to parameterize these by constants.

This is done using the syntax `const IDENT: Type` in the parameter listing. Unlike full const generics, `min_const_generics` is limited to parameterization by integers, and constants of type `char` or `bool`.

We already use `#![feature(min_const_generics)]` on stable to implement many common traits for arrays. See [the documentation](https://doc.rust-lang.org/nightly/std/primitive.array.html) for specific examples.

Generic const arguments, for now, are not permitted to involve computations depending on generic parameters. This means that const parameters may only be instantiated using either:

1. const expressions that do not depend on any generic parameters, e.g. `{ foo() + 1 }`, where `foo` is a `const fn`
1. standalone const parameters, e.g. `{N}`

### Example

```rust
#![feature(min_const_generics)]

trait Foo<const N: usize> {
    fn method<const M: usize>(&mut self, arr: [[u8; M]; N]);
}

struct Bar<T, const N: usize> {
    inner: [T; N],
}

impl<const N: usize> Foo<N> for Bar<u8, N> {
    fn method<const M: usize>(&mut self, arr: [[u8; M]; N]) {
        for (elem, s) in self.inner.iter_mut().zip(arr.iter()) {
            for &x in s {
                *elem &= x;
            }
        }
    }
}

fn function<const N: u16>() -> u16 {
    // Const parameters can be used freely inside of functions.
    (N + 1) / 2 * N
}

fn main() {
    let mut bar = Bar { inner: [0xff; 3] };
    // This infers the value of `M` from the type of the function argument.
    bar.method([[0b11_00, 0b01_00], [0b00_11, 0b00_01], [0b11_00, 0b00_11]]);
    assert_eq!(bar.inner, [0b01_00, 0b00_01, 0b00_00]);

    // You can also explicitly specify the value of `N`.
    assert_eq!(function::<17>(), 153);
}
```

## Motivation

Rust has the built-in array type, which is parametric over a constant. Without const generics, this type can be quite cumbersome to use as it is not possible to generically implement a trait for arrays of different lengths. For example, this meant that, for a long time, the standard library only contained trait implementations for arrays up to a length of 32. This restriction has since been lifted through the use of const generics.

Const parameters allow users to naturally specify variants of a generic type which are more naturally parameterized by values, rather than by types. For example, using const generics, many of the uses of the crate [typenum](https://crates.io/crates/typenum) may now be replaced with const parameters, improving compilation time as well as code readability and diagnostics.

The subset described by `min_const_generics` is self-contained, but extensive enough to help with the most frequent issues: implementing traits for arrays and using arbitrarily-sized arrays inside of other types. Furthermore, it extends naturally to full `const_generics` once the remaining design and implementation questions have been resolved.

## In-depth feature description

### Declaring const parameters

*Const parameters* are allowed in all places where types and lifetimes are supported. They use the syntax `const IDENT: Type`. Currently, const parameters must be declared after lifetime and type parameters. Their scope is equal to the scope of other generic parameters. They live in the value namespace.

`Type` must be one of `u8`, `u16`, `u32`, `u64`, `u128`, `usize`, `i8`, `i16`, `i32`, `i64`, `i128`, `isize`, `char` and `bool`. This restriction is implemented in two places:

1. during name resolution, where we forbid generic parameters
1. during well-formedness checking, where we only allow the types listed above

The updated syntax of parameter listings is:

```
GenericParams:
    (OuterAttr* LifetimeParam),* (OuterAttr* TypeParam),* (OuterAttr* ConstParam),*

OuterAttr: '#[' ... ']'
LifetimeParam: ...
TypeParam: ...
ConstParam: 'const' IDENT ':' Type
```

Unlike type and lifetime parameters, const parameters of types can be used without being mentioned inside of a parameterized type because const parameters do not have issues concerning variance. This means that the following types are allowed:

```rust
struct Foo<const N: usize>;
enum Bar<const M: usize> { A, B }
```

### Const arguments

Const parameters are instantiated using *const arguments*. Any concrete const expression or const parameter as a standalone argument can be used. When applying an expression as const parameter, most expressions must be contained within a block, with two exceptions:

1. literals and single-segment path expressions
1. array lengths

This syntactic restriction is necessary to avoid ambiguity, or requiring infinite lookahead when parsing an expression as a generic argument.

In the cases where a generic argument could be resolved as either a type or const argument, we always interpret it as a type. This causes the following test to fail:

```rust
type N = u32;
struct Foo<const N: usize>;
fn foo<const N: usize>() -> Foo<N> { todo!() } // ERR
```

To circumvent this, the user may wrap the const parameter with braces, at which point it is unambiguously accepted.

```rust
type N = u32;
struct Foo<const N: usize>;
fn bar<const N: usize>() -> Foo<{ N }> { todo!() } // ok
```

Operations depending on generic parameters are **not** allowed, which is enforced during well-formedness checking. Allowing generic unevaluated constants would require a way to check if they would always evaluate successfully to prevent errors that are not caught at declaration time. This ability forms part of `#![feature(const_evaluatable_checked)]`, which is not yet being stabilised.

Since we are not yet stabilizing `#![feature(lazy_normalization_consts)]`, we must not supply the parent generics to anonymous constants except for repeat expressions. Doing so can cause cycle errors for arrays used in `where`-bounds. Not supplying the parent generics can however lead to ICEs occurring before well-formedness checking when trying to use a generic parameter. See rust-lang#56445 for details.

Since we expect cases like this to occur more frequently once `min_const_generics` is stabilized, we have chosen to forbid generic parameters in anonymous constants during name resolution. While this changes the ICE in the situation above to an ordinary error, this is theoretically a breaking change, as early-bound lifetimes were previously permitted in repeat expressions but now are disallowed, causing the following snippet to break:

```rust
fn late_bound<'a>() {
    let _ = [0; {
        let _: &'a (); // ICE ==> ERR
        3
    }];
}

fn early_bound<'a>() where &'a (): Sized {
    let _ = [0; {
        let _: &'a (); // ok ==> ERR
        3
    }];
}
```

### Using const parameters

Const parameters can be used almost everywhere ordinary constants are allowed, except that they may not be used in the construction of consts, statics, functions, or types inside a function body and are subject to the generic argument restrictions mentioned above.

Expressions containing const parameters are eligible for promotion:

```rust
fn test<const N: usize>() -> &'static usize {
    &(3 + N)
}
```

### Symbol mangling

See the [Rust symbol name mangling RFC](https://rust-lang.github.io/rfcs/2603-rust-symbol-name-mangling-v0.html) for an overview. Generic const parameters take the form `K[type][value]` when the value is known, or `Kp` where the value is not known, where:
- `[type]` is any integral type, `bool`, or `char`.
- `[value]` is the unsigned hex value for integers, preceded by `n` when negative; is `0` or `1` for `bool`; is the hex value for `char`.

### Exhaustiveness checking

We do not check the exhaustiveness of impls, meaning that the following example does **not** compile:

```rust
struct Foo<const B: bool>;
trait Bar {}
impl Bar for Foo<true> {}
impl Bar for Foo<false> {}

fn needs_bar(_: impl Bar) {}
fn generic<const B: bool>() {
    let v = Foo::<B>;
    needs_bar(v);
}
```

### Type inference

The value of const parameters can be inferred during typeck. One interesting case is the length of generic arrays, which can also be inferred from patterns (implemented in rust-lang#70562). Practical usage of this can be seen in rust-lang#76825.

### Equality of constants

`#![feature(min_const_generics)]` only permits generic parameters to be used as standalone generic arguments. We compare two parameters to be equal if they are literally the same generic parameter.

### Associated constants

Associated constants can use const parameters without restriction, see rust-lang#79135 (comment) for more details.

## Future work

As this is a limited subset of rust-lang/rfcs#2000, there are quite a few extensions we will be looking into next.

### Lazy normalization of constants

Stabilizing `#![feature(lazy_normalization_consts)]` (tracking issue rust-lang#72219) will remove some special cases that are currently necessary for `min_const_generics`, and unblocks operations on const parameters.

### Relaxing ordering requirements between const and type parameters

We currently restrict the order of generic parameters so that types must come before consts. We could relax this, as is currently done with `const_generics`. Without this it is not possible to use both type defaults and const parameters at the same time.

Unrestricting the order will require us to improve some diagnostics that expect there to be a strict order between type and const parameters.

### Allowing more parameter types

We would like to support const parameters of more types, especially`&str` and user-defined types. Both are blocked on [valtrees]. There are also open questions regarding the design of `structural_match` concerning the latter. Supporting generic const parameter types such as `struct Foo<T, const N: T>` will be a lot harder and is unlikely to be implemented in the near future.

### Default values of const parameters

We do not yet support default values for const parameters. There is work in progress to enable this on nightly (see rust-lang#75384).

### Generic const operations

With `#![feature(min_const_generics)]`, only concrete const expressions and parameters as standalone arguments are allowed in types and repeat expressions. However, supporting generic const operations, such as `N + 1` or `std::mem::size_of::<T>()` is highly desirable. This feature is in early development under `#![feature(const_evaluatable_checked)]`.

## Implementation history

Many people have contributed to the design and implementation of const generics over the last three years. See rust-lang#44580 (comment) for a summary. Once again thank you to everybody who helped out here!

[valtrees]: rust-lang#72396

---

r? `@varkor`
@oli-obk oli-obk added the A-valtree Area: Value trees or fixed by value trees label Mar 15, 2021
@RalfJung
Copy link
Member Author

I think this is resolved? The main const-eval query is eval_to_allocation_raw, it returns an AllocId and it performs validation. We have "type" constants whose normal form is a valtree and "MIR" constants whose normal form is a ConstValue. ConstValue is not just an AllocId so that it can efficiently hold a Scalar (used a lot everywhere) or a (byte) str literal (used for MIR building for said literals).

There is more cleanup that I can imagine, but it should probably be in a new issue.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
A-const-eval Area: constant evaluation (mir interpretation) A-valtree Area: Value trees or fixed by value trees C-cleanup Category: PRs that clean code up or issues documenting cleanup. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. T-lang Relevant to the language team, which will review and decide on the PR/issue.
Projects
None yet
Development

No branches or pull requests

3 participants