Always capture tokens for `macro_rules!` arguments
When we invoke a proc-macro, the `TokenStream` we pass to it may contain 'interpolated' AST fragments, represented by `rustc_ast::token::Nonterminal`. In order to correctly, pass a `Nonterminal` to a proc-macro, we need to have 'captured' its `TokenStream` at the time the AST was parsed.
Currently, we perform this capturing when attributes are present on items and expressions, since we will end up using a `Nonterminal` to pass the item/expr to any proc-macro attributes it is annotated with. However, `Nonterminal`s are also introduced by the expansion of metavariables in `macro_rules!` macros. Since these metavariables may be passed to proc-macros, we need to have tokens available to avoid the need to pretty-print and reparse (see https://github.com/rust-lang/rust/issues/43081).
This PR unconditionally performs token capturing for AST items and expressions that are passed to a `macro_rules!` invocation. We cannot know in advance if captured item/expr will be passed to proc-macro, so this is needed to ensure that tokens will always be available when they are needed.
This ensures that proc-macros will receive tokens with proper `Spans` (both location and hygiene) in more cases. Like all work on https://github.com/rust-lang/rust/issues/43081, this will cause regressions in proc-macros that were relying on receiving tokens with dummy spans.
In this case, Crater revealed only one regression: the [Pear](https://github.com/SergioBenitez/Pear) crate (a helper for [rocket](https://github.com/SergioBenitez/Rocket)), which was previously [fixed](https://github.com/SergioBenitez/Pear/pull/25) as part of https://github.com/rust-lang/rust/pull/73084.
This regression manifests itself as the following error:
```
[INFO] [stdout] error: proc macro panicked
[INFO] [stdout] --> /opt/rustwide/cargo-home/registry/src/github.com-1ecc6299db9ec823/rocket_http-0.4.5/src/parse/uri/parser.rs:119:34
[INFO] [stdout] |
[INFO] [stdout] 119 | let path_and_query = pear_try!(path_and_query(is_pchar));
[INFO] [stdout] | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[INFO] [stdout] |
[INFO] [stdout] = help: message: called `Option::unwrap()` on a `None` value
[INFO] [stdout] = note: this error originates in a macro (in Nightly builds, run with -Z macro-backtrace for more info)
```
It can be fixed by running `cargo update -p pear`, which updates your `Cargo.lock` to use the latest version of Pear (which includes a bugfix for the regression).
Split out from https://github.com/rust-lang/rust/pull/73084/
The const propagator cannot trace references.
Thus we avoid propagation of a local the moment we encounter references to it.
fixes#73609
cc @RalfJung
r? @wesleywiser
Point at the call span when overflow occurs during monomorphization
This improves the output for issue #72577, but there's still more work
to be done.
Currently, an overflow error during monomorphization results in an error
that points at the function we were unable to monomorphize. However, we
don't point at the call that caused the monomorphization to happen. In
the overflow occurs in a large recursive function, it may be difficult
to determine where the issue is.
This commit tracks and `Span` information during collection of
`MonoItem`s, which is used when emitting an overflow error. `MonoItem`
itself is unchanged, so this only affects
`src/librustc_mir/monomorphize/collector.rs`
In the following example, an inaccessible path is suggested via
`use foo::bar::X;` whereas an accessible public exported path can
be suggested instead.
```
mod foo {
mod bar {
pub struct X;
}
pub use self::bar::X;
}
fn main() { X; }
```
This fixes the issue.
Add second message for LiveDrop errors
This is an attempt to fix https://github.com/rust-lang/rust/issues/72907 by adding a second message to the `LiveDrop` diagnostics. Changing from this
```
error[E0493]: destructors cannot be evaluated at compile-time
--> src/lib.rs:7:9
|
7 | let mut always_returned = None;
| ^^^^^^^^^^^^^^^^^^^ constants cannot evaluate destructors
error: aborting due to previous error
```
to this
```
error[E0493]: destructors cannot be evaluated at compile-time
--> foo.rs:6:9
|
6 | let mut always_returned = None;
| ^^^^^^^^^^^^^^^^^^^ constants cannot evaluate destructors
...
10 | always_returned = never_returned;
| --------------- value is dropped here
error: aborting due to previous error
```
r? @RalfJung @ecstatic-morse
Account for multiple impl/dyn Trait in return type when suggesting `'_`
Make `impl` and `dyn` Trait lifetime suggestions a bit more resilient.
Follow up to #72804.
r? @nikomatsakis
move leak-check to during coherence, candidate eval
Implementation of MCP https://github.com/rust-lang/compiler-team/issues/295.
I'd like to do a crater run on this.
Note to @rust-lang/lang: This PR is a breaking change (bugfix). It causes tests like the following to go from a future-compatibility warning #56105 to a hard error:
```rust
trait Trait {}
impl Trait for for<'a, 'b> fn(&'a u32, &'b u32) {}
impl Trait for for<'c> fn(&'c u32, &'c u32) {} // now rejected, used to warn
```
I am not aware of any instances of this code in the wild, but that is why we are doing a crater run. The reason for this change is that those two types are, in fact, the same type, and hence the two impls are overlapping.
There will still be impls that trigger #56105 after this lands, however -- I hope that we will eventually just accept those impls without warning, for the most part. One example of such an impl is this pattern, which is used by wasm-bindgen and other crates as well:
```rust
trait Trait {}
impl<T> Trait for fn(&T) { }
impl<T> Trait for fn(T) { } // still accepted, but warns
```
Improve compiler error message for wrong generic parameter order
- Added optional "help" parameter that shows a help message on the compiler error if required.
- Added a simple ordered parameter as a sample help.
@varkor will make more changes as required. Let me know if I'm heading in the right direction.
Fixes#68437
r? @varkor
Currently, rustc uses a heuristic to determine if a range expression is
not a literal based on whether the expression looks like a function call
or struct initialization. This fails for range literals whose
lower/upper bounds are the results of function calls. A possibly-better
heuristic is to check if the expression contains `..`, required in range
literals.
Of course, this is also not perfect; for example, if the range
expression is a struct which includes some text with `..` this will
fail, but in general I believe it is a better heuristic.
A better alternative altogether is to add the `QPath::LangItem` enum
variant suggested in #60607. I would be happy to do this as a precursor
to this patch if someone is able to provide general suggestions on how
usages of `QPath` need to be changed later in the compiler with the
`LangItem` variant.
Closes#73553
Fix spurious 'value moved here in previous iteration of loop' messages
Fixes#46099
Previously, we would check the 'move' and 'use' spans to see if we
should emit this message. However, this can give false positives when
macros are involved, since two distinct expressions may end up with the
same span.
Instead, we check the actual MIR `Location`, which eliminates false
positives.
The bug was revealed by the behavior of the old-lub-glb-hr-noteq1.rs
test. The old-lub-glb-hr-noteq2 test shows the current 'order dependent'
behavior of coercions around higher-ranked functions, at least when
running with `-Zborrowck=mir`.
Also, run compare-mode=nll.
This improves the output for issue #72577, but there's still more work
to be done.
Currently, an overflow error during monomorphization results in an error
that points at the function we were unable to monomorphize. However, we
don't point at the call that caused the monomorphization to happen. In
the overflow occurs in a large recursive function, it may be difficult
to determine where the issue is.
This commit tracks and `Span` information during collection of
`MonoItem`s, which is used when emitting an overflow error. `MonoItem`
itself is unchanged, so this only affects
`src/librustc_mir/monomorphize/collector.rs`
Revert the code that states that upcasting traits requires full
equality and change to require that the source type is a subtype of
the target type, as one would expect. As the comment states, this was
an old bug that we didn't want to fix yet as it interacted poorly with
the old leak-check. This fixes the old-lub-glb-object test, which was
previously reporting too many errors (i.e., in the previous commit).
In particular, it no longer occurs during the subtyping check. This is
important for enabling lazy normalization, because the subtyping check
will be producing sub-obligations that could affect its results.
Consider an example like
for<'a> fn(<&'a as Mirror>::Item) =
fn(&'b u8)
where `<T as Mirror>::Item = T` for all `T`. We will wish to produce a
new subobligation like
<'!1 as Mirror>::Item = &'b u8
This will, after being solved, ultimately yield a constraint that `'!1
= 'b` which will fail. But with the leak-check being performed on
subtyping, there is no opportunity to normalize `<'!1 as
Mirror>::Item` (unless we invoke that normalization directly from
within subtyping, and I would prefer that subtyping and unification
are distinct operations rather than part of the trait solving stack).
The reason to keep the leak check during coherence and trait
evaluation is partly for backwards compatibility. The coherence change
permits impls for `fn(T)` and `fn(&T)` to co-exist, and the trait
evaluation change means that we can distinguish those two cases
without ambiguity errors. It also avoids recreating #57639, where we
were incorrectly choosing a where clause that would have failed the
leak check over the impl which succeeds.
The other reason to keep the leak check in those places is that I
think it is actually close to the model we want. To the point, I think
the trait solver ought to have the job of "breaking down"
higher-ranked region obligation like ``!1: '2` into into region
obligations that operate on things in the root universe, at which
point they should be handed off to polonius. The leak check isn't
*really* doing that -- these obligations are still handed to the
region solver to process -- but if/when we do adopt that model, the
decision to pass/fail would be happening in roughly this part of the
code.
This change had somewhat more side-effects than I anticipated. It
seems like there are cases where the leak-check was not being enforced
during method proving and trait selection. I haven't quite tracked
this down but I think it ought to be documented, so that we know what
precisely we are committing to.
One surprising test was `issue-30786.rs`. The behavior there seems a
bit "fishy" to me, but the problem is not related to the leak check
change as far as I can tell, but more to do with the closure signature
inference code and perhaps the associated type projection, which
together seem to be conspiring to produce an unexpected
signature. Nonetheless, it is an example of where changing the
leak-check can have some unexpected consequences: we're now failing to
resolve a method earlier than we were, which suggests we might change
some method resolutions that would have been ambiguous to be
successful.
TODO:
* figure out remainig test failures
* add new coherence tests for the patterns we ARE disallowing
In the new leak check, instead of getting a list of placeholders to
track, we look for any placeholder that is part of a universe which
was created during the snapshot.
We are looking for the following error patterns:
* P1: P2, where P1 != P2
* P1: R, where R is in some universe that cannot name P1
This new leak check is more precise than before, in that it accepts
this patterns:
* R: P1, even if R cannot name P1, because R = 'static is a valid
sol'n
* R: P1, R: P2, as above
Note that this leak check, when running during subtyping, is less
efficient than before in some sense because it is going to check and
re-check all the universes created since the snapshot. We're going to
move when the leak check runs to try and correct that.
Also, update the affected tests. This seems strictly better but it is
actually more permissive than I initially intended. In particular it
accepts this
```
forall<'a, 'b> {
exists<'intersection> {
'a: 'intersection,
'b: 'intersection,
}
}
```
and I'm not sure I want to accept that. It implies that we have a
`'empty` in the new universe intoduced by the `forall`.
Prefer accessible paths in 'use' suggestions
This PR addresses issue https://github.com/rust-lang/rust/issues/26454, where `use` suggestions are made for paths that don't work. For example:
```rust
mod foo {
mod bar {
struct X;
}
}
fn main() { X; } // suggests `use foo::bar::X;`
```
Fixes#46099
Previously, we would check the 'move' and 'use' spans to see if we
should emit this message. However, this can give false positives when
macros are involved, since two distinct expressions may end up with the
same span.
Instead, we check the actual MIR `Location`, which eliminates false
positives.
Upgrade Chalk
Things done in this PR:
- Upgrade Chalk to `0.11.0`
- Added compare-mode=chalk
- Bump rustc-hash in `librustc_data_structures` to `1.1.0` to match Chalk
- Removed `RustDefId` since the builtin type support is there
- Add a few more `FIXME(chalk)`s for problem spots I hit when running all tests with chalk
- Added some more implementation code for some newer builtin Chalk types (e.g. `FnDef`, `Array`)
- Lower `RegionOutlives` and `ObjectSafe` predicates
- Lower `Dyn` without the region
- Handle `Int`/`Float` `CanonicalVarKind`s
- Uncomment some Chalk tests that actually work now
- Remove the revisions in `src/test/ui/coherence/coherence-subtyping.rs` since they aren't doing anything different
r? @nikomatsakis
This fixes an issue with the following sample:
mod foo {
mod inaccessible {
pub struct X;
}
pub mod avail {
pub struct X;
}
}
fn main() { X; }
Instead of suggesting both `use crate::foo::inaccessible::X;` and `use
crate::foo::avail::X;`, it should only suggest the latter.
It is done by trimming the list of suggestions from inaccessible paths
if accessible paths are present.
Visibility is checked with `is_accessible_from` now instead of being
hard-coded.
-
Some tests fixes are trivial, and others require a bit more explaining,
here are my comments:
src/test/ui/issues/issue-35675.stderr: Only needs to make the enum
public to have the suggestion make sense.
src/test/ui/issues/issue-42944.stderr: Importing the tuple struct won't
help because its constructor is not visible, so the attempted
constructor does not work. In that case, it's better not to suggest it.
The case where the constructor is public is covered in `issue-26545.rs`.
Add a lint to catch clashing `extern` fn declarations.
Closes#69390.
Adds lint `clashing_extern_decl` to detect when, within a single crate, an extern function of the same name is declared with different types. Because two symbols of the same name cannot be resolved to two different functions at link time, and one function cannot possibly have two types, a clashing extern declaration is almost certainly a mistake.
This lint does not run between crates because a project may have dependencies which both rely on the same extern function, but declare it in a different (but valid) way. For example, they may both declare an opaque type for one or more of the arguments (which would end up distinct types), or use types that are valid conversions in the language the extern fn is defined in. In these cases, we can't say that the clashing declaration is incorrect.
r? @eddyb
lint: normalize projections using opaque types
Fixes#73251.
This PR normalizes projections which use opaque types (opaque types are otherwise linted against, which is would have previously made the test cases added in this PR fail).
Projection bound validation
During selection we use bounds declared on associated types (e.g. `type X: Copy`) to satisfy trait/projection bounds. This would be fine so long as those bounds are checked on any impls/trait objects. For simple cases they are because the bound `Self::X: Copy` gets normalized when we check the impl.
However, for default values with specialization and higher-ranked bounds from GATs or otherwise, we can't normalize when checking the impl, and so we use the bound from the trait to prove that the bound applies to the impl, which is clearly unsound.
This PR makes 2 fixes for this:
1. Requiring that the bounds on the trait apply to a projection type with the corresponding substs, so a bound `for<'a> <Self as X<'a>>::U: Copy` on the trait cannot be used to prove `<T as X<'_>>::U: Copy`.
2. Actually checking that the bounds that we still allow apply to generic/default associated types.
Opening for a crater run.
Closes#68641Closes#68642Closes#68643Closes#68644Closes#68645Closes#68656
r? @ghost
Try to suggest dereferences on trait selection failed
Fixes#39029Fixes#62530
This PR consists of two parts:
1. Decouple `Autoderef` with `FnCtxt` and move `Autoderef` to `librustc_trait_selection`.
2. Try to suggest dereferences when trait selection failed.
The first is needed because:
1. For suggesting dereferences, the struct `Autoderef` should be used. But before this PR, it is placed in `librustc_typeck`, which depends on `librustc_trait_selection`. But trait selection error emitting happens in `librustc_trait_selection`, if we want to use `Autoderef` in it, dependency loop is inevitable. So I moved the `Autoderef` to `librustc_trait_selection`.
2. Before this PR, `FnCtxt` is coupled to `Autoderef`, and `FnCtxt` only exists in `librustc_typeck`. So decoupling is needed.
After this PR, we can get suggestion like this:
```
error[E0277]: the trait bound `&Baz: Happy` is not satisfied
--> $DIR/trait-suggest-deferences-multiple.rs:34:9
|
LL | fn foo<T>(_: T) where T: Happy {}
| ----- required by this bound in `foo`
...
LL | foo(&baz);
| ^^^^
| |
| the trait `Happy` is not implemented for `&Baz`
| help: consider adding dereference here: `&***baz`
error: aborting due to previous error
For more information about this error, try `rustc --explain E0277`.
```
r? @estebank
