In "specify type" suggestion, skip type params that are already known
When we suggest specifying a type for an expression or pattern, like in a `let` binding, we previously would print the entire type as the type system knew it. We now look at the params that have *no* inference variables, so they are fully known to the type system which means that they don't need to be specified.
This helps in suggestions for types that are really long, because we can usually skip most of the type params and make the annotation as short as possible:
```
error[E0282]: type annotations needed for `Result<_, ((..., ..., ..., ...), ..., ..., ...)>`
--> $DIR/really-long-type-in-let-binding-without-sufficient-type-info.rs:7:9
|
LL | let y = Err(x);
| ^ ------ type must be known at this point
|
help: consider giving `y` an explicit type, where the type for type parameter `T` is specified
|
LL | let y: Result<T, _> = Err(x);
| ++++++++++++++
```
Fix#135919.
Document some safety constraints and use more safe wrappers
Lots of unsafe codegen_llvm code has safe wrappers already, so I used some of them and added some where applicable. I stopped here because this diff is large enough and should probably be reviewed independently of other changes.
These were a way to ensure hashes were stable over time for ExternAbi,
but simply hashing the strings is more stable in the face of changes.
As a result, we can do away with them.
Directly map each ExternAbi variant to its string and back again.
This has a few advantages:
- By making the ABIs compare equal to their strings, we can easily
lexicographically sort them and use that sorted slice at runtime.
- We no longer need a workaround to make sure the hashes remain stable,
as they already naturally are (by being the hashes of unique strings).
- The compiler can carry around less &str wide pointers
Properly deeply normalize in the next solver
Turn deep normalization into a `TypeOp`. In the old solver, just dispatch to the `Normalize` type op, but in the new solver call `deeply_normalize`. I chose to separate it into a different type op b/c some normalization is a no-op in the new solver, so this distinguishes just the normalization we need for correctness.
Then use `DeeplyNormalize` in the callsites we used to be using a `CustomTypeOp` (for normalizing known type outlives obligations), and also use it to normalize function args and impl headers in the new solver.
Finally, use it to normalize signatures for WF checks in the new solver as well. This addresses https://github.com/rust-lang/trait-system-refactor-initiative/issues/146.
```
error[E0109]: type arguments are not allowed on tuple variant `TSVariant`
--> $DIR/enum-variant-generic-args.rs:54:29
|
LL | Enum::<()>::TSVariant::<()>(());
| --------- ^^ type argument not allowed
| |
| not allowed on tuple variant `TSVariant`
|
= note: generic arguments are not allowed on both an enum and its variant's path segments simultaneously; they are only valid in one place or the other
help: remove the generics arguments from one of the path segments
|
LL - Enum::<()>::TSVariant::<()>(());
LL + Enum::<()>::TSVariant(());
|
```
```
error[E0109]: type arguments are not allowed on enum `Enum` and tuple variant `TSVariant`
--> $DIR/enum-variant-generic-args.rs:54:12
|
LL | Enum::<()>::TSVariant::<()>(());
| ---- ^^ --------- ^^ type argument not allowed
| | |
| | not allowed on tuple variant `TSVariant`
| not allowed on enum `Enum`
|
= note: generic arguments are not allowed on both an enum and its variant's path segments simultaneously; they are only valid in one place or the other
help: remove the generics arguments from one of the path segments
|
LL - Enum::<()>::TSVariant::<()>(());
LL + Enum::<()>::TSVariant(());
|
```
Fix#93993.
Simplify intra-crate qualifiers.
The following is a weird pattern for a file within `rustc_middle`:
```
use rustc_middle::aaa;
use crate::bbb;
```
More sensible and standard would be this:
```
use crate::{aaa, bbb};
```
I.e. we generally prefer using `crate::` to using a crate's own name. (Exceptions are things like in macros where `crate::` doesn't work because the macro is used in multiple crates.)
This commit fixes a bunch of these weird qualifiers.
r? `@jieyouxu`
compiler: die immediately instead of handling unknown target codegen
We cannot produce anything useful if asked to compile unknown targets. We should handle the error immediately at the point of discovery instead of propagating it upward, and preferably in the simplest way: Die.
This allows cleaning up our "error-handling" spread across 5 crates.
show supported register classes in error message
a simple diagnostic change that shows the supported register classes when an invalid one is found.
This information can be hard to find (especially for unstable targets), and this message now gives at least something to try or search for. I've followed the pattern for invalid clobber ABIs.
`@rustbot` label +A-inline-assembly
fix ensure_monomorphic_enough
When polymorphization was still a thing, the visitor was used to only recurse into *used generic parameters* of function/closure/coroutine types and allow unused parameters (i.e. the polymorphized parameters) to remain generic.
When polymorphization got removed, this got changed to always treat all parameters as polymorphic and never recurse into them: https://github.com/rust-lang/rust/pull/133883/files#diff-210c59e321070d0ca4625c04e9fb064bf43ddc34082e7e33a7ee8a6c577e95afL44-L62
This is clearly wrong and can cause MIR opts to misbehave, for example this currently prints "false" in release mode:
```rust
#![feature(core_intrinsics)]
fn generic<T>() {}
const fn type_id_of_val<T: 'static>(_: &T) -> u128 {
std::intrinsics::type_id::<T>()
}
fn cursed_is_i32<T: 'static>() -> bool {
(const { type_id_of_val(&generic::<T>) } == const { type_id_of_val(&generic::<i32>) })
}
fn main() {
dbg!(cursed_is_i32::<i32>());
}
```
This PR reverts to the old behavior of always treating all types that contain type parameters as too generic, like we used to do without `-Zpolymorphize` before.
~~I'm not including the above as a test case here, because I think there is little value in testing code paths that have been removed and this seems unlikely to regress in a way that would be caught by a regression test, but let me know if you disagree and want me to add a test anyway.~~
Overhaul how contracts are lowered on fn-like bodies
Consolidates all of the contracts lowering logic into `lower_fn_body`, rather than having it be split between `lower_item_kind` and `lower_fn_body`. This should fix#136683.
r? celinval
Stop using span hack for contracts feature gating
The contracts machinery is a pretty straightforward case of an *external* feature using a (perma-unstable) *internal* feature within its implementation. There's no reason why it needs to be implemented any differently than other features by using global span tracking hacks to change whether the internals are gated behind the `contracts` or `contracts_internals` feature gate -- for the case of macro expansions we already have `allow_internal_unstable` for exactly this situation.
This PR changes the internal, perma-unstable AST syntax to use the `contracts_internals` gate always, and adjusts the macro expansion to use the right spans so that `allow_internal_unstable` works correctly.
As a follow-up, there's really no reason to have `contracts` be a *compiler feature* since it's at this point fully a *library feature*; the only reason it's a compiler feature today is so we can mark it as incomplete, but that seems like a weak reason. I didn't do anything in this PR for this.
r? ``@celinval``
cg_llvm: Reduce visibility of some items outside the `llvm` module
Next piece of #135502
This reduces the visibility of items (other than those in the `llvm` module) so that dead code analysis will correctly identify unused items.
The following is a weird pattern for a file within `rustc_middle`:
```
use rustc_middle::aaa;
use crate::bbb;
```
More sensible and standard would be this:
```
use crate::{aaa, bbb};
```
I.e. we generally prefer using `crate::` to using a crate's own name.
(Exceptions are things like in macros where `crate::` doesn't work
because the macro is used in multiple crates.)
This commit fixes a bunch of these weird qualifiers.
Show diff suggestion format on verbose replacement
```
error[E0610]: `{integer}` is a primitive type and therefore doesn't have fields
--> $DIR/attempted-access-non-fatal.rs:7:15
|
LL | let _ = 2.l;
| ^
|
help: if intended to be a floating point literal, consider adding a `0` after the period and a `f64` suffix
|
LL - let _ = 2.l;
LL + let _ = 2.0f64;
|
```
before:
```
error[E0610]: `{integer}` is a primitive type and therefore doesn't have fields
--> $DIR/attempted-access-non-fatal.rs:7:15
|
LL | let _ = 2.l;
| ^
|
help: if intended to be a floating point literal, consider adding a `0` after the period and a `f64` suffix
|
LL + let _ = 2.0f64;
| ~~~~
```
r? `@oli-obk`
compiler: gate `extern "{abi}"` in ast_lowering
I don't believe low-level crates like `rustc_abi` should have to know or care about higher-level concerns like whether the ABI string is stable for users. These implementation details can be made less open to public inspection. This way the code that governs stability is near the code that enforces stability, and compiled together.
It also abstracts away certain error messages instead of constantly repeating them.
A few error messages are simply deleted outright, instead of made uniform, because they are either too dated to be useful or redundant with other diagnostic improvements we could make. These can be pursued in followups: my first concern was making sure there wasn't unnecessary diagnostics-related code in `rustc_abi`, which is not well-positioned to understand what kind of errors are going to be generated based on how it is used.
r? ``@ghost``
Prevent generic pattern types from being used in libstd
Pattern types should follow the same rules that patterns follow. So a pattern type range must not wrap and not be empty. While we reject such invalid ranges at layout computation time, that only happens during monomorphization in the case of const generics. This is the exact same issue as other const generic math has, and since there's no solution there yet, I put these pattern types behind a separate incomplete feature.
These are not necessary for the pattern types MVP (replacing the layout range attributes in libcore and rustc).
cc #136574 (new tracking issue for the `generic_pattern_types` feature gate)
r? ``@lcnr``
cc ``@scottmcm`` ``@joshtriplett``
Delay bug when method confirmation cannot upcast object pick of self
Justification is on the test comment. Simply delays a bug that we were previously ICEing on.
cc ``@adetaylor`` since this is a `arbitrary_self_types` ICE.
Introduce CoercePointeeWellformed for coherence checks at typeck stage
Fix#135206
This is the first PR to introduce the "wellformedness" check for `derive(CoercePointee)`.
This patch introduces a new error code to cover all the prerequisites of the said macro. The checks that is enforced with this patch is whether the data is indeed `struct` and whether the layout is set to `repr(transparent)`.
A following series of patch will arrive later to address the following concern.
1. #135217 so that we would only admit one single coercion on one type parameter, and leave the rest for future consideration in tandem of development of other coercion rules.
1. Enforcement of data field requirements.
**An open question** is whether there is a good schema to encode the `#[pointee]` as well, so that we could also check if the `#[pointee]` type parameter is indeed `?Sized`.
``@rustbot`` label F-derive_coerce_pointee
Pointers for variables all need to be in the same address space for
correct compilation. Therefore ensure that even if an `alloca` is
created in a different address space, it is casted to the default
address space before its value is used.
This is necessary for the amdgpu target and others where the default
address space for `alloca`s is not 0.
For example the following code compiles incorrectly when not casting the
address space to the default one:
```rust
fn f(p: *const i8 /* addrspace(0) */) -> *const i8 /* addrspace(0) */ {
let local = 0i8; /* addrspace(5) */
let res = if cond { p } else { &raw const local };
res
}
```
results in
```llvm
%local = alloca addrspace(5) i8
%res = alloca addrspace(5) ptr
if:
; Store 64-bit flat pointer
store ptr %p, ptr addrspace(5) %res
else:
; Store 32-bit scratch pointer
store ptr addrspace(5) %local, ptr addrspace(5) %res
ret:
; Load and return 64-bit flat pointer
%res.load = load ptr, ptr addrspace(5) %res
ret ptr %res.load
```
For amdgpu, `addrspace(0)` are 64-bit pointers, `addrspace(5)` are
32-bit pointers.
The above code may store a 32-bit pointer and read it back as a 64-bit
pointer, which is obviously wrong and cannot work. Instead, we need to
`addrspacecast %local to ptr addrspace(0)`, then we store and load the
correct type.
```
error[E0610]: `{integer}` is a primitive type and therefore doesn't have fields
--> $DIR/attempted-access-non-fatal.rs:7:15
|
LL | let _ = 2.l;
| ^
|
help: if intended to be a floating point literal, consider adding a `0` after the period and a `f64` suffix
|
LL - let _ = 2.l;
LL + let _ = 2.0f64;
|
```
We cannot produce anything useful if asked to compile unknown targets.
We should handle the error immediately at the point of discovery instead
of propagating it upward, and preferably in the simplest way: Die.
This allows cleaning up our "error-handling" spread across 5 crates.
Disable DWARF in linker options for i686-unknown-uefi
This fixes an lld warning:
> warning: linker stderr: rust-lld: section name .debug_frame is longer than 8 characters and will use a non-standard string table
See https://reviews.llvm.org/D69594 for details of where the warning was added.
This warning only occurs with the i686 UEFI target, not x86_64 or aarch64. The x86_64 target uses an LLVM target of
`x86_64-unknown-windows` and aarch64 uses `aarch64-unknown-windows`, but i686 uses `i686-unknown-windows-gnu` (note the `-gnu`). See comments in `i686_unknown_uefi.rs` for details of why.
The `.debug_frame` section should not actually be needed; UEFI targets provide a separate PDB file for debugging. Disable DWARF (and by extension the `.debug_frame` section) by passing `/DEBUG:NODWARF` to lld.
Tested with:
```
export RUSTC_LOG=rustc_codegen_ssa:🔙:link=info
cargo +stage1 build --release --target i686-unknown-uefi
```
This issue was originally raised here: https://github.com/rust-lang/rust/pull/119286#issuecomment-2612746162. See also https://github.com/rust-lang/rust/issues/136096. It was suggested to file an LLVM bug, but I don't think LLVM is actually doing anything wrong as such.
CC `@dvdhrm` `@jyn514` let me know if you have any feedback on this approach
rustc_middle: parallel: TyCtxt: remove "unsafe impl DynSend/DynSync"
rustc_middle: parallel: TyCtxt: remove "unsafe impl DynSend/DynSync"
We don't need to "short circuit trait resolution", because DynSend and DynSync are auto traits and thus coinductive
cc "Parallel Rustc Front-end" https://github.com/rust-lang/rust/issues/113349
r? SparrowLii
``@rustbot`` label: +WG-compiler-parallel
(rustbot sometimes ignores me and doesn't attach labels on my behalf. rustbot banned me?)
adding autodiff tests
I'd like to get started with upstreaming some tests, even though I'm still waiting for an answer on how to best integrate the enzyme pass. Can we therefore temporarily support the -Z llvm-plugins here without too much effort? And in that case, how would that work? I saw you can do remapping, e.g. `rust-src-base`, but I don't think that will give me the path to libEnzyme.so. Do you have another suggestion?
Other than that this test simply checks that the derivative of `x*x` is `2.0 * x`, which in this case is computed as
`%0 = fadd fast double %x.0.val, %x.0.val`
(I'll add a few more tests and move it to an autodiff folder if we can use the -Z flag)
r? ``@jieyouxu``
Locally at least `-Zllvm-plugins=${PWD}/build/x86_64-unknown-linux-gnu/enzyme/build/Enzyme/libEnzyme-19.so` seems to work if I copy the command I get from x.py test and run it manually. However, running x.py test itself fails.
Tracking:
- https://github.com/rust-lang/rust/issues/124509
Zulip discussion: https://rust-lang.zulipchat.com/#narrow/channel/326414-t-infra.2Fbootstrap/topic/Enzyme.20build.20changes
The wording unsafe pointer is less common and not mentioned in a lot of
places, instead this is usually called a "raw pointer". For the sake of
uniformity, we rename this method.
This came up during the review of
https://github.com/rust-lang/rust/pull/134424.
Removed dependency on the field-offset crate, alternate approach
This is an alternate approach to reach the same goals as #136003. As it touches the core of the query system, this too probably should be evaluated for performance.
r? ``@Mark-Simulacrum``
coverage: Defer part of counter-creation until codegen
Follow-up to #135481 and #135873.
One of the pleasant properties of the new counter-assignment algorithm is that we can stop partway through the process, store the intermediate state in MIR, and then resume the rest of the algorithm during codegen. This lets it take into account which parts of the control-flow graph were eliminated by MIR opts, resulting in fewer physical counters and simpler counter expressions.
Those improvements end up completely obsoleting much larger chunks of code that were previously responsible for cleaning up the coverage metadata after MIR opts, while also doing a more thorough cleanup job.
(That change also unlocks some further simplifications that I've kept out of this PR to limit its scope.)
It is speculated that these two can be conceptually merged, and it can
start by ripping out rustc's notion of the PtxKernel call convention.
Leave the ExternAbi for now, but the nvptx target now should see it as
just a different way to spell Conv::GpuKernel.
Add amdgpu target
Add amdgpu target to rustc and enable the LLVM target.
Fix compiling `core` with the amdgpu:
The amdgpu backend makes heavy use of different address spaces. This
leads to situations, where a pointer in one addrspace needs to be casted
to a pointer in a different addrspace. `bitcast` is invalid for this
case, `addrspacecast` needs to be used.
Fix compilation failures that created bitcasts for such cases by
creating pointer casts (which creates an `addrspacecast` under the hood)
instead.
MCP: https://github.com/rust-lang/compiler-team/issues/823
Tracking issue: #135024
Kinda related to the original amdgpu tracking issue #51575 (though that one has been closed for a while).
These are either residue of a long-term migration away from something,
or are simply trying too hard to be specifically useful:
nearest-match suggestions for ABI strings should handle this.
By moving this stability check into AST lowering, we effectively make
it impossible to accidentally miss, as it must happen to generate HIR.
Also, we put the ABI-stability code next to code that actually uses it!
This allows code that wants to reason about backend ABI implementations
to stop worrying about high-level concerns like syntax stability,
while still leaving it as the authority on what ABIs actually exist.
It also makes it easy to refactor things to have more consistent errors.
For now, we only apply this to generalize the existing messages a bit.
This fixes an lld warning:
> warning: linker stderr: rust-lld: section name .debug_frame is longer
> than 8 characters and will use a non-standard string table
See https://reviews.llvm.org/D69594 for details of where the warning was
added.
This warning only occurs with the i686 UEFI target, not x86_64 or
aarch64. The x86_64 target uses an LLVM target of
`x86_64-unknown-windows` and aarch64 uses `aarch64-unknown-windows`, but
i686 uses `i686-unknown-windows-gnu` (note the `-gnu`). See comments in
`i686_unknown_uefi.rs` for details of why.
The `.debug_frame` section should not actually be needed; UEFI targets
provide a separate PDB file for debugging. Disable DWARF (and by
extension the `.debug_frame` section) by passing `/DEBUG:NODWARF` to lld.
Tested with:
export RUSTC_LOG=rustc_codegen_ssa:🔙:link=info
cargo +stage1 build --release --target i686-unknown-uefi
Visit all debug info in MIR Visitor
I've been experimenting with simplifying debug info in MIR inliner, and discovered that MIR Visitor doesn't reliably visit all spans. This PR adds the missing visitor calls.
DWARF 1 is very different than DWARF 2+ (see the commentary in
https://gcc.gnu.org/onlinedocs/gcc/Debugging-Options.html#index-gdwarf)
and LLVM does not really seem to support DWARF 1 as Clang does not offer
a `-gdwarf-1` flag and `llc` will just generate DWARF 2 with the version
set to 1: https://godbolt.org/z/s85d87n3a.
Since this isn't actually supported (and it's not clear it would be
useful anyway), report that DWARF 1 is not supported if it is requested.
Also add a help message to the error saying which versions are supported.
Update bootstrap compiler and rustfmt
The rustfmt version we previously used formats things differently from what the latest nightly rustfmt does. This causes issues for subtrees that get formatted both in-tree and in their own repo. Updating the rustfmt used in-tree solves those issues. Also bumped the bootstrap compiler as the stage0 update command always updates both at the same
time.
