Revise never type fallback algorithm
This is a rebase of https://github.com/rust-lang/rust/pull/84573, but dropping the stabilization of never type (and the accompanying large test diff).
Each commit builds & has tests updated alongside it, and could be reviewed in a more or less standalone fashion. But it may make more sense to review the PR as a whole, I'm not sure. It should be noted that tests being updated isn't really a good indicator of final behavior -- never_type_fallback is not enabled by default in this PR, so we can't really see the full effects of the commits here.
This combines the work by Niko, which is [documented in this gist](https://gist.github.com/nikomatsakis/7a07b265dc12f5c3b3bd0422018fa660), with some additional rules largely derived to target specific known patterns that regress with the algorithm solely derived by Niko. We build these from an intuition that:
* In general, fallback to `()` is *sound* in all cases
* But, in general, we *prefer* fallback to `!` as it accepts more code, particularly that written to intentionally use `!` (e.g., Result's with a Infallible/! variant).
When evaluating Niko's proposed algorithm, we find that there are certain cases where fallback to `!` leads to compilation failures in real-world code, and fallback to `()` fixes those errors. In order to allow for stabilization, we need to fix a good portion of these patterns.
The final rule set this PR proposes is that, by default, we fallback from `?T` to `!`, with the following exceptions:
1. `?T: Foo` and `Bar::Baz = ?T` and `(): Foo`, then fallback to `()`
2. Per [Niko's algorithm](https://gist.github.com/nikomatsakis/7a07b265dc12f5c3b3bd0422018fa660#proposal-fallback-chooses-between--and--based-on-the-coercion-graph), the "live" `?T` also fallback to `()`.
The first rule is necessary to address a fairly common pattern which boils down to something like the snippet below. Without rule 1, we do not see the closure's return type as needing a () fallback, which leads to compilation failure.
```rust
#![feature(never_type_fallback)]
trait Bar { }
impl Bar for () { }
impl Bar for u32 { }
fn foo<R: Bar>(_: impl Fn() -> R) {}
fn main() {
foo(|| panic!());
}
```
r? `@jackh726`
Lazy TAIT preparation cleanups
Check that TAIT generics are fully generic in mir typeck instead of wf-check, as wf-check can by definition only check TAIT in return position and not account for TAITs defined in the body of the function
r? `@spastorino` `@nikomatsakis`
Support `#[track_caller]` on closures and generators
## Lang team summary
This PR adds support for placing the `#[track_caller]` attribute on closure and generator expressions. This attribute's addition behaves identically (from a users perspective) to the attribute being placed on the method in impl Fn/FnOnce/FnMut for ... generated by compiler.
The attribute is currently "double" feature gated -- both `stmt_expr_attributes` (preexisting) and `closure_track_caller` (newly added) must be enabled in order to place these attributes on closures.
As the Fn* traits lack a `#[track_caller]` attribute in their definition, caller information does not propagate when invoking closures through dyn Fn*. There is no limitation that this PR adds in supporting this; it can be added in the future.
# Implementation details
This is implemented in the same way as for functions - an extra
location argument is appended to the end of the ABI. For closures,
this argument is *not* part of the 'tupled' argument storing the
parameters - the final closure argument for `#[track_caller]` closures
is no longer a tuple.
For direct (monomorphized) calls, the necessary support was already
implemented - we just needeed to adjust some assertions around checking
the ABI and argument count to take closures into account.
For calls through a trait object, more work was needed.
When creating a `ReifyShim`, we need to create a shim
for the trait method (e.g. `FnOnce::call_mut`) - unlike normal
functions, closures are never invoked directly, and always go through a
trait method.
Additional handling was needed for `InstanceDef::ClosureOnceShim`. In
order to pass location information throgh a direct (monomorphized) call
to `FnOnce::call_once` on an `FnMut` closure, we need to make
`ClosureOnceShim` aware of `#[tracked_caller]`. A new field
`track_caller` is added to `ClosureOnceShim` - this is used by
`InstanceDef::requires_caller` location, allowing codegen to
pass through the extra location argument.
Since `ClosureOnceShim.track_caller` is only used by codegen,
we end up generating two identical MIR shims - one for
`track_caller == true`, and one for `track_caller == false`. However,
these two shims are used by the entire crate (i.e. it's two shims total,
not two shims per unique closure), so this shouldn't a big deal.
This PR allows applying a `#[track_caller]` attribute to a
closure/generator expression. The attribute as interpreted as applying
to the compiler-generated implementation of the corresponding trait
method (`FnOnce::call_once`, `FnMut::call_mut`, `Fn::call`, or
`Generator::resume`).
This feature does not have its own feature gate - however, it requires
`#![feature(stmt_expr_attributes)]` in order to actually apply
an attribute to a closure or generator.
This is implemented in the same way as for functions - an extra
location argument is appended to the end of the ABI. For closures,
this argument is *not* part of the 'tupled' argument storing the
parameters - the final closure argument for `#[track_caller]` closures
is no longer a tuple.
For direct (monomorphized) calls, the necessary support was already
implemented - we just needeed to adjust some assertions around checking
the ABI and argument count to take closures into account.
For calls through a trait object, more work was needed.
When creating a `ReifyShim`, we need to create a shim
for the trait method (e.g. `FnOnce::call_mut`) - unlike normal
functions, closures are never invoked directly, and always go through a
trait method.
Additional handling was needed for `InstanceDef::ClosureOnceShim`. In
order to pass location information throgh a direct (monomorphized) call
to `FnOnce::call_once` on an `FnMut` closure, we need to make
`ClosureOnceShim` aware of `#[tracked_caller]`. A new field
`track_caller` is added to `ClosureOnceShim` - this is used by
`InstanceDef::requires_caller` location, allowing codegen to
pass through the extra location argument.
Since `ClosureOnceShim.track_caller` is only used by codegen,
we end up generating two identical MIR shims - one for
`track_caller == true`, and one for `track_caller == false`. However,
these two shims are used by the entire crate (i.e. it's two shims total,
not two shims per unique closure), so this shouldn't a big deal.
Re-enable the `src/test/debuginfo/mutex.rs` test on Windows
This test required a newer version of cdb than was previously enabled in
CI thus leading to some bitrot in the test since the time it was
originally created. With the update to the `windows-latest` image last
week, we're now running this test in CI and thus uncovered the
regression.
I've updated the test and it now passes.
r? `@ehuss`
"Fix" an overflow in byte position math
r? `@estebank`
help! I fixed the ICE only to brick the diagnostic.
I mean, it was wrong previously (using an already expanded macro span), but it is really bad now XD
Implement `#[must_not_suspend]`
implements #83310
Some notes on the impl:
1. The code that searches for the attribute on the ADT is basically copied from the `must_use` lint. It's not shared, as the logic did diverge
2. The RFC does specify that the attribute can be placed on fn's (and fn-like objects), like `must_use`. I think this is a direct copy from the `must_use` reference definition. This implementation does NOT support this, as I felt that ADT's (+ `impl Trait` + `dyn Trait`) cover the usecase's people actually want on the RFC, and adding an imp for the fn call case would be significantly harder. The `must_use` impl can do a single check at fn call stmt time, but `must_not_suspend` would need to answer the question: "for some value X with type T, find any fn call that COULD have produced this value". That would require significant changes to `generator_interior.rs`, and I would need mentorship on that. `@eholk` and I are discussing it.
3. `@estebank` do you know a way I can make the user-provided `reason` note pop out? right now it seems quite hidden
Also, I am not sure if we should run perf on this
r? `@nikomatsakis`
In suggest_missing_return_type, erase late bound regions after normalizing
Fixes#88360
There might be some hardening that could be done to not error or avoid erroring with LUBing `ReErased` with `ReEmpty`, but this was the most simple fix for this particular case.
r? `@nikomatsakis`
Fix debuginfo for parameters passed via the ScalarPair abi on Windows
Mark all of these as locals so the debugger does not try to interpret
them as being a pointer to the value. This extends the approach used
in #81898.
Fixes#88625
Fix ICE when `indirect_structural_match` is allowed
Fixes#89088. The ICE is caused by `delay_good_path_bug()`, which is called (indirectly) from a `format!()` macro invocation. I have moved the macro invocation into the `decorate` closure of `struct_span_lint_hir()`, so that the macro is only invoked if the lint is not allowed (i.e., causes at least a warning, and thus prevents `delay_good_path_bug()` from firing).
rustc_codegen_llvm: make sse4.2 imply crc32 for LLVM 14
This fixes compiling things like the `snap` crate after
https://reviews.llvm.org/D105462. I added a test that verifies the
additional attribute gets specified, and confirmed that I can build
cargo with both LLVM 13 and 14 with this change applied.
r? `@nagisa` cc `@nikic`
This allows the format_args! macro to keep the pre-expansion code out of
the unsafe block without doing gymnastics with nested `match`
expressions. This reduces codegen.
Don't use projection cache or candidate cache in intercrate mode
Fixes#88969
It appears that *just* disabling the evaluation cache (in #88994)
leads to other issues involving intercrate mode caching. I suspect
that since we now always end up performing the full evaluation
in intercrate mode, we end up 'polluting' the candidate and projection
caches with results that depend on being in intercrate mode in some way.
Previously, we might have hit a cached evaluation (stored during
non-intercrate mode), and skipped doing this extra work in
intercrate mode.
The whole situation with intercrate mode caching is turning into
a mess. Ideally, we would remove intercrate mode entirely - however,
this might require waiting on Chalk.
Register normalization obligations instead of immediately normalizing in opaque type instantiation
For lazy TAIT we will need to instantiate opaque types from within `rustc_infer`, which cannot invoke normalization methods (they are in `rustc_trait_resolution`). So before we move the logic over to `rustc_infer`, we need make sure no normalization happens anymore. This PR resolves that by just registering normalization obligations and continuing.
This PR is best reviewed commit by commit
I also included f7ad36e which is just an independent cleanup that touches the same code and reduces diagnostics noise a bit
r? `@nikomatsakis` cc `@spastorino`
Lower only one HIR owner at a time
Based on https://github.com/rust-lang/rust/pull/83723
Additional diff is here: https://github.com/cjgillot/rust/compare/ownernode...lower-mono
Lowering is very tangled and has a tendency to intertwine the transformation of different items. This PR aims at simplifying the logic by:
- moving global analyses to the resolver (item_generics_num_lifetimes, proc_macros, trait_impls);
- removing a few special cases (non-exported macros and use statements);
- restricting the amount of available information at any one time;
- avoiding back-and-forth between different owners: an item must now be lowered all at once, and its parent cannot refer to its nodes.
I also removed the sorting of bodies by span. The diagnostic ordering changes marginally, since definitions are pretty much sorted already according to the AST. This uncovered a subtlety in thir-unsafeck.
(While these items could logically be in different PRs, the dependency between commits and the amount of conflicts force a monolithic PR.)
Suggest replacing braces for brackets on array-esque invalid block expr
Newcomers may write `{1, 2, 3}` for making arrays, and the current error message is not informative enough to quickly convince them what is needed to fix the error.
This PR implements a diagnostic for this case, and its output looks like this:
```text
error: this code is interpreted as a block expression, not an array
--> src/lib.rs:1:22
|
1 | const FOO: [u8; 3] = {
| ______________________^
2 | | 1, 2, 3
3 | | };
| |_^
|
= note: to define an array, one would use square brackets instead of curly braces
help: try using [] instead of {}
|
1 | const FOO: [u8; 3] = [
2 | 1, 2, 3
3 | ];
|
```
Fix#87672
Fixes#88969
It appears that *just* disabling the evaluation cache (in #88994)
leads to other issues involving intercrate mode caching. I suspect
that since we now always end up performing the full evaluation
in intercrate mode, we end up 'polluting' the candidate and projection
caches with results that depend on being in intercrate mode in some way.
Previously, we might have hit a cached evaluation (stored during
non-intercrate mode), and skipped doing this extra work in
intercrate mode.
The whole situation with intercrate mode caching is turning into
a mess. Ideally, we would remove intercrate mode entirely - however,
this might require waiting on Chalk.
This test required a newer version of cdb than was previously enabled in
CI thus leading to some bitrot in the test since the time it was
originally created. With the update to the `windows-latest` image last
week, we're now running this test in CI and thus uncovered the
regression.
I've updated the test and it now passes.
This fixes compiling things like the `snap` crate after
https://reviews.llvm.org/D105462. I added a test that verifies the
additional attribute gets specified, and confirmed that I can build
cargo with both LLVM 13 and 14 with this change applied.
Fix debuginfo tests for the latest version of the Windows SDK.
Re-enable the tests that were disabled to fix CI.
Changes:
- Cdb now correctly visualizes enums.
- Cdb doesn't render emoji characters in `OSStr` anymore.
- Cdb doesn't always render `str` correctly (#88840)
Add `ConstraintCategory::Usage` for handling aggregate construction
In some cases, we emit borrowcheck diagnostics pointing
at a particular field expression in a struct expression
(e.g. `MyStruct { field: my_expr }`). However, this
behavior currently relies on us choosing the
`ConstraintCategory::Boring` with the 'correct' span.
When adding additional variants to `ConstraintCategory`,
(or changing existing usages away from `ConstraintCategory::Boring`),
the current behavior can easily get broken, since a non-boring
constraint will get chosen over a boring one.
To make the diagnostic output less fragile, this commit
adds a `ConstraintCategory::Usage` variant. We use this variant
for the temporary assignments created for each field of
an aggregate we are constructing.
Using this new variant, we can emit a message mentioning
"this usage", emphasizing the fact that the error message
is related to the specific use site (in the struct expression).
This is preparation for additional work on improving NLL error messages
(see #57374)
Gather module items after lowering.
This avoids having a non-local analysis inside lowering.
By implementing `hir_module_items` using a visitor, we make sure that iterations and visitors are consistent.
Newcomers may write `{1, 2, 3}` for making arrays, and the current error
message is not informative enough to quickly convince them what is
needed to fix the error.
This PR implements a diagnostic for this case, and its output looks like
this:
```text
error: this code is interpreted as a block expression, not an array
--> src/lib.rs:1:22
|
1 | const FOO: [u8; 3] = {
| ______________________^
2 | | 1, 2, 3
3 | | };
| |_^
|
= note: to define an array, one would use square brackets instead of curly braces
help: try using [] instead of {}
|
1 | const FOO: [u8; 3] = [
2 | 1, 2, 3
3 | ];
|
```
Fix#87672
Suggest better place to add call parentheses for method expressions wrapped in parentheses
I wanted to improve the suggestion a bit to both remove the wrapping parentheses **and** add call parentheses by both calling `suggest_method_call` and using `multipart_suggestion`. But I very quickly ran into a problem where multiple overlapping machine applicable suggestions cannot be properly applied together. So I applied the suggestion from the issue and only added the call parentheses directly after the expression.
Fixes: https://github.com/rust-lang/rust/issues/89044
Add a separate error for `dyn Trait` in `const fn`
Previously "trait bounds other than `Sized` on const fn parameters are unstable" error was used for both trait bounds (`<T: Trait>`) and trait objects (`dyn Trait`). This was pretty confusing.
This PR adds a separate error for trait objects: "trait objects in const fn are unstable". The error for trait bounds is otherwise intact.
This is follow up to #88907
r? ``@estebank``
``@rustbot`` label: +A-diagnostics
Fix linting when trailing macro expands to a trailing semi
When a macro is used in the trailing expression position of a block
(e.g. `fn foo() { my_macro!() }`), we currently parse it as an
expression, rather than a statement. As a result, we ended up
using the `NodeId` of the containing statement as our `lint_node_id`,
even though we don't normally do this for macro calls.
If such a macro expands to an expression with a `#[cfg]` attribute,
then the trailing statement can get removed entirely. This lead to
an ICE, since we were usng the `NodeId` of the expression to emit
a lint.
Ths commit makes us skip updating `lint_node_id` when handling
a macro in trailing expression position. This will cause us to
lint at the closest parent of the macro call.
Suggest replacing an inexisting field for an unmentioned field
Fix#87938
This PR adds a suggestion to replace an inexisting field for an
unmentioned field. Given the following code:
```rust
enum Foo {
Bar { alpha: u8, bravo: u8, charlie: u8 },
}
fn foo(foo: Foo) {
match foo {
Foo::Bar {
alpha,
beta, // `bravo` miswritten as `beta` here.
charlie,
} => todo!(),
}
}
```
the compiler now emits the error messages below.
```text
error[E0026]: variant `Foo::Bar` does not have a field named `beta`
--> src/lib.rs:9:13
|
9 | beta, // `bravo` miswritten as `beta` here.
| ^^^^
| |
| variant `Foo::Bar` does not have this field
| help: `Foo::Bar` has a field named `bravo`: `bravo`
```
Note that this suggestion is available iff the number of inexisting
fields and unmentioned fields are both 1.
Propagate coercion cause into `try_coerce`
Currently, `coerce_inner` discards its `ObligationCause`
when calling `try_coerce`. This interfers with other
diagnostc improvements I'm working on, since we will lose
the original span by the time the actual coercion occurs.
Additionally, we now use the span of the trailing expression
(rather than the span of the entire function) when performing
a coercion in `check_return_expr`. This currently has no visible
effect on any of the unit tests, but will unblock future
diagnostic improvements.
Start block is not allowed to have basic block predecessors
* The MIR validator is extended to detect potential violations.
* The start block has no predecessors after building MIR, so no changes are required there.
* The SimplifyCfg could previously violate this requirement when collapsing goto chains, so transformation is disabled for the start block, which also substantially simplifies the implementation.
* The LLVM function entry block also must not have basic block predecessors. Previously, to ensure that code generation had to perform necessary adjustments. This now became unnecessary.
The motivation behind the change is to align with analogous requirement in LLVM, and to avoid potential latent bugs like the one reported in #88043.
Skip single use lifetime lint for generated opaque types
Fix: #77175
The opaque type generated by the desugaring process of an async function uses the lifetimes defined by the originating function. The DefId for the lifetimes in the opaque type are different from the ones in the originating async function - as they should be, as far as I understand, and could therefore be considered a single use lifetimes, this causes the single_use_lifetimes lint to fail compilation if explicitly denied. This fix skips the lint for lifetimes used only once in generated opaque types for an async function that are declared in the parent async function definition.
More info in the comments on the original issue: 1 and 2
