This was added to cover up a lazy extra semicolon in #35849, but does
not actually make sense. This is removed as a part of the stabilization
of `never_type`.
Make accesses to fields of packed structs unsafe
To handle packed structs with destructors (which you'll think are a rare
case, but the `#[repr(packed)] struct Packed<T>(T);` pattern is
ever-popular, which requires handling packed structs with destructors to
avoid monomorphization-time errors), drops of subfields of packed
structs should drop a local move of the field instead of the original
one.
That's it, I think I'll use a strategy suggested by @Zoxc, where this mir
```
drop(packed_struct.field)
```
is replaced by
```
tmp0 = packed_struct.field;
drop tmp0
```
cc #27060 - this should deal with that issue after codegen of drop glue
is updated.
The new errors need to be changed to future-compatibility warnings, but
I'll rather do a crater run first with them as errors to assess the
impact.
cc @eddyb
Things which still need to be done for this:
- [ ] - handle `repr(packed)` structs in `derive` the same way I did in `Span`, and use derive there again
- [ ] - implement the "fix packed drops" pass and call it in both the MIR shim and validated MIR pipelines
- [ ] - do a crater run
- [ ] - convert the errors to compatibility warnings
Fix the derive implementation for repr(packed) structs to move the
fields out instead of calling functions on references to each subfield.
That's it, `#[derive(PartialEq)]` on a packed struct now does:
```Rust
fn eq(&self, other: &Self) {
let field_0 = self.0;
let other_field_0 = other.0;
&field_0 == &other_field_0
}
```
Instead of
```Rust
fn eq(&self, other: &Self) {
let ref field_0 = self.0;
let ref other_field_0 = other.0;
&*field_0 == &*other_field_0
}
```
Taking (unaligned) references to each subfield is undefined, unsound and
is an error with MIR effectck, so it had to be prevented. This causes
a borrowck error when a `repr(packed)` struct has a non-Copy field (and
therefore is a [breaking-change]), but I don't see a sound way to avoid
that error.
To handle packed structs with destructors (which you'll think are a rare
case, but the `#[repr(packed)] struct Packed<T>(T);` pattern is
ever-popular, which requires handling packed structs with destructors to
avoid monomorphization-time errors), drops of subfields of packed
structs should drop a local move of the field instead of the original
one.
cc #27060 - this should deal with that issue after codegen of drop glue
is updated.
The new errors need to be changed to future-compatibility warnings, but
I'll rather do a crater run first with them as errors to assess the
impact.
Turns out ThinLTO was internalizing this symbol and eliminating it. Worse yet if
you compiled with LTO turns out no TLS destructors would run on Windows! The
`#[used]` annotation should be a more bulletproof implementation (in the face of
LTO) of preserving this symbol all the way through in LLVM and ensuring it makes
it all the way to the linker which will take care of it.
move closure kind, signature into `ClosureSubsts`
Instead of using side-tables, store the closure-kind and signature in the substitutions themselves. This has two key effects:
- It means that the closure's type changes as inference finds out more things, which is very nice.
- As a result, it avoids the need for the `freshen_closure_like` code (though we still use it for generators).
- It avoids cyclic closures calls.
- These were never meant to be supported, precisely because they make a lot of the fancy inference that we do much more complicated. However, due to an oversight, it was previously possible -- if challenging -- to create a setup where a closure *directly* called itself (see e.g. #21410).
We have to see what the effect of this change is, though. Needs a crater run. Marking as [WIP] until that has been assessed.
r? @arielb1
impl Trait Lifetime Handling
This PR implements the updated strategy for handling `impl Trait` lifetimes, as described in [RFC 1951](https://github.com/rust-lang/rfcs/blob/master/text/1951-expand-impl-trait.md) (cc #42183).
With this PR, the `impl Trait` desugaring works as follows:
```rust
fn foo<T, 'a, 'b, 'c>(...) -> impl Foo<'a, 'b> { ... }
// desugars to
exists type MyFoo<ParentT, 'parent_a, 'parent_b, 'parent_c, 'a, 'b>: Foo<'a, 'b>;
fn foo<T, 'a, 'b, 'c>(...) -> MyFoo<T, 'static, 'static, 'static, 'a, 'b> { ... }
```
All of the in-scope (parent) generics are listed as parent generics of the anonymous type, with parent regions being replaced by `'static`. Parent regions referenced in the `impl Trait` return type are duplicated into the anonymous type's generics and mapped appropriately.
One case came up that wasn't specified in the RFC: it's possible to write a return type that contains multiple regions, neither of which outlives the other. In that case, it's not clear what the required lifetime of the output type should be, so we generate an error.
There's one remaining FIXME in one of the tests: `-> impl Foo<'a, 'b> + 'c` should be able to outlive both `'a` and `'b`, but not `'c`. Currently, it can't outlive any of them. @nikomatsakis and I have discussed this, and there are some complex interactions here if we ever allow `impl<'a, 'b> SomeTrait for AnonType<'a, 'b> { ... }`, so the plan is to hold off on this until we've got a better idea of what the interactions are here.
cc #34511.
Fixes#44727.
std: Add a new wasm32-unknown-unknown target
This commit adds a new target to the compiler: wasm32-unknown-unknown. This target is a reimagining of what it looks like to generate WebAssembly code from Rust. Instead of using Emscripten which can bring with it a weighty runtime this instead is a target which uses only the LLVM backend for WebAssembly and a "custom linker" for now which will hopefully one day be direct calls to lld.
Notable features of this target include:
* There is zero runtime footprint. The target assumes nothing exists other than the wasm32 instruction set.
* There is zero toolchain footprint beyond adding the target. No custom linker is needed, rustc contains everything.
* Very small wasm modules can be generated directly from Rust code using this target.
* Most of the standard library is stubbed out to return an error, but anything related to allocation works (aka `HashMap`, `Vec`, etc).
* Naturally, any `#[no_std]` crate should be 100% compatible with this new target.
This target is currently somewhat janky due to how linking works. The "linking" is currently unconditional whole program LTO (aka LLVM is being used as a linker). Naturally that means compiling programs is pretty slow! Eventually though this target should have a linker.
This target is also intended to be quite experimental. I'm hoping that this can act as a catalyst for further experimentation in Rust with WebAssembly. Breaking changes are very likely to land to this target, so it's not recommended to rely on it in any critical capacity yet. We'll let you know when it's "production ready".
### Building yourself
First you'll need to configure the build of LLVM and enable this target
```
$ ./configure --target=wasm32-unknown-unknown --set llvm.experimental-targets=WebAssembly
```
Next you'll want to remove any previously compiled LLVM as it needs to be rebuilt with WebAssembly support. You can do that with:
```
$ rm -rf build
```
And then you're good to go! A `./x.py build` should give you a rustc with the appropriate libstd target.
### Test support
Currently testing-wise this target is looking pretty good but isn't complete. I've got almost the entire `run-pass` test suite working with this target (lots of tests ignored, but many passing as well). The `core` test suite is [still getting LLVM bugs fixed](https://reviews.llvm.org/D39866) to get that working and will take some time. Relatively simple programs all seem to work though!
In general I've only tested this with a local fork that makes use of LLVM 5 rather than our current LLVM 4 on master. The LLVM 4 WebAssembly backend AFAIK isn't broken per se but is likely missing bug fixes available on LLVM 5. I'm hoping though that we can decouple the LLVM 5 upgrade and adding this wasm target!
### But the modules generated are huge!
It's worth nothing that you may not immediately see the "smallest possible wasm module" for the input you feed to rustc. For various reasons it's very difficult to get rid of the final "bloat" in vanilla rustc (again, a real linker should fix all this). For now what you'll have to do is:
cargo install --git https://github.com/alexcrichton/wasm-gc
wasm-gc foo.wasm bar.wasm
And then `bar.wasm` should be the smallest we can get it!
---
In any case for now I'd love feedback on this, particularly on the various integration points if you've got better ideas of how to approach them!
This commit adds a new target to the compiler: wasm32-unknown-unknown. This
target is a reimagining of what it looks like to generate WebAssembly code from
Rust. Instead of using Emscripten which can bring with it a weighty runtime this
instead is a target which uses only the LLVM backend for WebAssembly and a
"custom linker" for now which will hopefully one day be direct calls to lld.
Notable features of this target include:
* There is zero runtime footprint. The target assumes nothing exists other than
the wasm32 instruction set.
* There is zero toolchain footprint beyond adding the target. No custom linker
is needed, rustc contains everything.
* Very small wasm modules can be generated directly from Rust code using this
target.
* Most of the standard library is stubbed out to return an error, but anything
related to allocation works (aka `HashMap`, `Vec`, etc).
* Naturally, any `#[no_std]` crate should be 100% compatible with this new
target.
This target is currently somewhat janky due to how linking works. The "linking"
is currently unconditional whole program LTO (aka LLVM is being used as a
linker). Naturally that means compiling programs is pretty slow! Eventually
though this target should have a linker.
This target is also intended to be quite experimental. I'm hoping that this can
act as a catalyst for further experimentation in Rust with WebAssembly. Breaking
changes are very likely to land to this target, so it's not recommended to rely
on it in any critical capacity yet. We'll let you know when it's "production
ready".
---
Currently testing-wise this target is looking pretty good but isn't complete.
I've got almost the entire `run-pass` test suite working with this target (lots
of tests ignored, but many passing as well). The `core` test suite is still
getting LLVM bugs fixed to get that working and will take some time. Relatively
simple programs all seem to work though!
---
It's worth nothing that you may not immediately see the "smallest possible wasm
module" for the input you feed to rustc. For various reasons it's very difficult
to get rid of the final "bloat" in vanilla rustc (again, a real linker should
fix all this). For now what you'll have to do is:
cargo install --git https://github.com/alexcrichton/wasm-gc
wasm-gc foo.wasm bar.wasm
And then `bar.wasm` should be the smallest we can get it!
---
In any case for now I'd love feedback on this, particularly on the various
integration points if you've got better ideas of how to approach them!
Refactor type memory layouts and ABIs, to be more general and easier to optimize.
To combat combinatorial explosion, type layouts are now described through 3 orthogonal properties:
* `Variants` describes the plurality of sum types (where applicable)
* `Single` is for one inhabited/active variant, including all C `struct`s and `union`s
* `Tagged` has its variants discriminated by an integer tag, including C `enum`s
* `NicheFilling` uses otherwise-invalid values ("niches") for all but one of its inhabited variants
* `FieldPlacement` describes the number and memory offsets of fields (if any)
* `Union` has all its fields at offset `0`
* `Array` has offsets that are a multiple of its `stride`; guarantees all fields have one type
* `Arbitrary` records all the field offsets, which can be out-of-order
* `Abi` describes how values of the type should be passed around, including for FFI
* `Uninhabited` corresponds to no values, associated with unreachable control-flow
* `Scalar` is ABI-identical to its only integer/floating-point/pointer "scalar component"
* `ScalarPair` has two "scalar components", but only applies to the Rust ABI
* `Vector` is for SIMD vectors, typically `#[repr(simd)]` `struct`s in Rust
* `Aggregate` has arbitrary contents, including all non-transparent C `struct`s and `union`s
Size optimizations implemented so far:
* ignoring uninhabited variants (i.e. containing uninhabited fields), e.g.:
* `Option<!>` is 0 bytes
* `Result<T, !>` has the same size as `T`
* using arbitrary niches, not just `0`, to represent a data-less variant, e.g.:
* `Option<bool>`, `Option<Option<bool>>`, `Option<Ordering>` are all 1 byte
* `Option<char>` is 4 bytes
* using a range of niches to represent *multiple* data-less variants, e.g.:
* `enum E { A(bool), B, C, D }` is 1 byte
Code generation now takes advantage of `Scalar` and `ScalarPair` to, in more cases, pass around scalar components as immediates instead of indirectly, through pointers into temporary memory, while avoiding LLVM's "first-class aggregates", and there's more untapped potential here.
Closes#44426, fixes#5977, fixes#14540, fixes#43278.
Simplify higher-ranked LUB/GLB
This is a better version of https://github.com/rust-lang/rust/pull/44211. It still makes higher-ranked LUB/GLB into a hard equality test, however, it does try to identify that something changed and issue a notice to the user. I wroteup https://github.com/rust-lang/rust/issues/45852 as a tracking issue for this change.
Currently, this moves straight to a hard-error, on the basis that the crater run in #44211 saw no impact. It might be good to retest -- or perhaps to try for a warning period. Trying to do the latter in a precise way would be somewhat painful, but an imprecise way might suffice -- that is, we could issue warning *whenever* a LUB/GLB operation succeeds that will later fail, even if it doesn't ultimately impact the type check. I could experiment with this.
~~I am *mildly* wary about landing this independently of other code that moves to a universe-based system. In particular, I was nervous that this change would make coherence accepts new pairs of impls that will later be errors. I have the code for the universe-based approach available, I hope to open an PR and run some tests on its impact very shortly.~~ @arielb1 points out that I was being silly.
r? @arielb1
After this change, impl Trait existentials are
desugared to a new `abstract type` definition
paired with a set of lifetimes to apply.
In-scope generics are included as parents of the
`abstract type` generics. Parent regions are
replaced with static, and parent regions
referenced in the `impl Trait` type are duplicated
at the end of the `abstract type`'s generics.
integrate MIR type-checker with NLL inference
This branch refactors NLL type inference so that it uses the MIR type-checker to gather constraints. Along the way, it also refactors how region constraints are gathered in the normal inference context mildly. The new setup is like this:
- What used to be `region_inference` is split into two parts:
- `region_constraints`, which just collects up sets of constraints
- `lexical_region_resolve`, which does the iterative, lexical region resolution
- When `resolve_regions_and_report_errors` is invoked, the inference engine converts the constraints into final values.
- In the MIR type checker, however, we do not invoke this method, but instead periodically take the region constraints and package them up for the NLL solver to use later.
- This allows us to track when and where those constraints were incurred.
- We also remove the central fulfillment context from the MIR type checker, instead instantiating new fulfillment contexts at each point. This allows us to capture the set of obligations that occurred at a particular point, and also to ensure that if the same obligation arises at two points, we will enforce the region constraints at both locations.
- The MIR type checker is also enhanced to instantiate late-bound-regions with fresh variables and handle a few other corner cases that arose.
- I also extracted some of the 'outlives' logic from the regionck, which will be needed later (see future work) to handle the type-outlives relationships.
One concern I have with this branch: since the MIR type checker is used even without the `-Znll` switch, I'm not sure if it will impact performance. One simple fix here would be to only enable the MIR type-checker if debug-assertions are enabled, since it just serves to validate the MIR. Longer term I hope to address this by improving the interface to the trait solver to be more query-based (ongoing work).
There is plenty of future work left. Here are two things that leap to mind:
- **Type-region outlives.** Currently, the NLL solver will ICE if it is required to handle a constraint like `T: 'a`. Fixing this will require a small amount of refactoring to extract the implied bounds code. I plan to follow a file-up bug on this (hopefully with mentoring instructions).
- **Testing.** It's a good idea to enumerate some of the tricky scenarios that need testing, but I think it'd be nice to try and parallelize some of the actual test writing (and resulting bug fixing):
- Same obligation occurring at two points.
- Well-formedness and trait obligations of various kinds (which are not all processed by the current MIR type-checker).
- More tests for how subtyping and region inferencing interact.
- More suggestions welcome!
r? @arielb1
This restores the behavior of regionck with respect to the
free-region-map: that is, it collects all the relations from the fn
and its closures. This feels a bit fishy but it's the behavior we've
had for some time, and it will go away with NLL, so seems best to just
keep it.
We already disallowed them to be in the arg list, such as
Fn(impl Debug), but now we disallow Fn() -> impl Debug.
Also remove the ImplTraitContext argument from the function
lower_parenthesized_parameter_data as it is now unused.
Comment out part of test run-pass/impl-trait/xcrate.rs that now fails.
Make saturating u128 -> f32 casts the default behavior
... rather than being gated by `-Z saturating-float-casts`. There are several reasons for this:
1. Const eval already implements this behavior.
2. Unlike with float->int casts, this behavior is uncontroversially the right behavior and it is not as performance critical. Thus there is no particular need to make the bug fix for u128->f32 casts opt-in.
3. Having two orthogonal features under one flag is silly, and never should have happened in the first place.
4. Benchmarking float->int casts with the -Z flag should not pick up performance changes due to the u128->f32 casts (assuming there are any).
Fixes#41799
Implement arbitrary_self_types
r? @arielb1
cc @nikomatsakis
Partial implementation of #44874. Supports trait and struct methods with arbitrary self types, as long as the type derefs (transitively) to `Self`. Doesn't support raw-pointer `self` yet.
Methods with non-standard self types (i.e. anything other than `&self, &mut self, and Box<Self>`) are not object safe, because dynamic dispatch hasn't been implemented for them yet.
I believe this is also a (partial) fix for #27941.
Fix checking of auto trait bounds in trait objects.
Any auto trait is allowed in trait object bounds. Fix duplicate check of type and lifetime parameter count, which we were [emitting twice](https://play.rust-lang.org/?gist=37dbbdbbec62dec423bb8f6d92f137cc&version=stable).
Note: This was the last use of `Send` in the compiler, meaning after a new `stage0` we could remove the `send` lang item.
... rather than being gated by -Z saturating-float-casts.
There are several reasons for this:
1. Const eval already implements this behavior.
2. Unlike with float->int casts, this behavior is uncontroversially the
right behavior and it is not as performance critical. Thus there is no
particular need to make the bug fix for u128->f32 casts opt-in.
3. Having two orthogonal features under one flag is silly, and never
should have happened in the first place.
4. Benchmarking float->int casts with the -Z flag should not pick up
performance changes due to the u128->f32 casts (assuming there are any).
Fixes#41799
Allow a trailling comma in assert_eq/ne macro
From Rust beginners IRC:
<???> It sure does annoy me that assert_eq!() does not accept a trailing comma after the last argument.
<???> ???: File an issue against https://github.com/rust-lang/rust and CC @rust-lang/libs
Figured that might as well submit it. Will become insta-stable after merging (danger zone).
cc @rust-lang/libs
