some inspect improvements
split from #114810 because I still want to experiment a bunch with that PR and these changes are self-contained.
r? `@compiler-errors`
This commit adds support for a `#[diagnostic::on_unimplemented]`
attribute with the following options:
* `message` to customize the primary error message
* `note` to add a customized note message to an error message
* `label` to customize the label part of the error message
Co-authored-by: León Orell Valerian Liehr <me@fmease.dev>
Co-authored-by: Michael Goulet <michael@errs.io>
Bubble up opaque <eq> opaque operations instead of picking an order
In case we are in `Bubble` mode (meaning every opaque type that is defined in the current crate is treated as if it were in its defining scope), we don't try to register an opaque type as the hidden type of another opaque type, but instead bubble up an obligation to equate them at the query caller site. Usually that means we have a `DefiningAnchor::Bind` and thus can reliably figure out whether an opaque type is in its defining scope. Where we can't, we'll error out, so the default is sound.
With this change we start using `AliasTyEq` predicates in the old solver, too.
fixes https://github.com/rust-lang/rust/issues/108498
But also regresses `tests/ui/impl-trait/anon_scope_creep.rs`. Our use of `Bubble` for `check_opaque_type_well_formed` is going to keep biting us.
r? `@lcnr` `@compiler-errors`
also handle 2 panics when dumping proof trees for the whole test suite
- need to actually tell the proof tree builder about overflow
- need to handle a recursion_limit of 0 :<
The `Debug` impl for `Ty` just calls the `Display` impl for `Ty`. This
is surprising and annoying. In particular, it means `Debug` doesn't show
as much information as `Debug` for `TyKind` does. And `Debug` is used in
some user-facing error messages, which seems bad.
This commit changes the `Debug` impl for `Ty` to call the `Debug` impl
for `TyKind`. It also does a number of follow-up changes to preserve
existing output, many of which involve inserting
`with_no_trimmed_paths!` calls. It also adds `Display` impls for
`UserType` and `Canonical`.
Some tests have changes to expected output:
- Those that use the `rustc_abi(debug)` attribute.
- Those that use the `EMIT_MIR` annotation.
In each case the output is slightly uglier than before. This isn't
ideal, but it's pretty weird (particularly for the attribute) that the
output is using `Debug` in the first place. They're fairly obscure
attributes (I hadn't heard of them) so I'm not worried by this.
For `async-is-unwindsafe.stderr`, there is one line that now lacks a
full path. This is a consistency improvement, because all the other
mentions of `Context` in this test lack a path.
Fix error report for size overflow from transmute
Fixes#115402
The span in the error reporting always points to the `dst`, this is an old issue, I may open another PR to fix it.
Don't ICE on associated type projection without feature gate in new solver
Self-explanatory, we should avoid ICEs when the feature gate is not enabled. Continue to ICE when the feature gate *is* enabled, though.
Fixes#115500
Do not require associated types with Self: Sized to uphold bounds when confirming object candidate
RPITITs and associated types that have `Self: Sized` bounds are opted out of the `dyn Trait` well-formedness check that happens during confirmation. This ensures that we can actually *use* `dyn Trait`s that have associated types that, e.g., have GATs and RPITITs and other naughty things as long as those are opted-out of object safety via a `Self: Sized` bound.
Fixes#115464
This seems like a natural part of https://github.com/rust-lang/rust/pull/112319#issuecomment-1592574451, and I don't think needs re-litigation.
r? `@oli-obk`
Permit recursive weak type aliases
I saw #63097 and thought "we can do ~~better~~ funnier". So here it is. It's not useful, but it's certainly something. This may actually become feasible with lazy norm (so in 5 years (constant, not reducing over time)).
r? `@estebank`
cc `@GuillaumeGomez`
Fix bors missing a commit when merging #115355
bors incorrectly merged an outdated version of PR #115355 (via rollup #115370):
- it [recorded r+](https://github.com/rust-lang/rust/pull/115355#issuecomment-1698372365) as approving commit 325b585259, and thus merged the original revision 7762ac7bb5
- but the branch at the time was at commit eefa07d69b, so bors missed the `compiler/rustc_trait_selection/src/solve/search_graph/mod.rs` cleanup in commit 0e1e964a34😓
Thankfully the change that bors missed was small, and this new PR corrects the situation (as I'd rather avoid having confusing multiple merge commits of PR #115355 in the git history)
r? ``@compiler-errors``
More precisely detect cycle errors from type_of on opaque
Not sure if this still needs work. Just putting it up for initial impressions, since it seems that a few people are frustrated with the increased error verbosity due to #113320.
Essentially we introduce a new sub-query for `type_of` specifically for opaques which returns a value that is able to distinguish "has errors" from "due to cycle recovery".
Fixes#115188
r? `@oli-obk`
Add an (perma-)unstable option to disable vtable vptr
This flag is intended for evaluation of trait upcasting space cost for embedded use cases.
Compared to the approach in #112355, this option provides a way to evaluate end-to-end cost of trait upcasting. Rationale: https://github.com/rust-lang/rust/issues/112355#issuecomment-1658207769
## How this flag should be used (after merge)
Build your project with and without `-Zno-trait-vptr` flag. If you are using cargo, set `RUSTFLAGS="-Zno-trait-vptr"` in the environment variable. You probably also want to use `-Zbuild-std` or the binary built may be broken. Save both binaries somewhere.
### Evaluate the space cost
The option has a direct and indirect impact on vtable space usage. Directly, it gets rid of the trait vptr entry needed to store a pointer to a vtable of a supertrait. (IMO) this is a small saving usually. The larger saving usually comes with the indirect saving by eliminating the vtable of the supertrait (and its parent).
Both impacts only affects vtables (notably the number of functions monomorphized should , however where vtable reside can depend on your relocation model. If the relocation model is static, then vtable is rodata (usually stored in Flash/ROM together with text in embedded scenario). If the binary is relocatable, however, the vtable will live in `.data` (more specifically, `.data.rel.ro`), and this will need to reside in RAM (which may be a more scarce resource in some cases), together with dynamic relocation info living in readonly segment.
For evaluation, you should run `size` on both binaries, with and without the flag. `size` would output three columns, `text`, `data`, `bss` and the sum `dec` (and it's hex version). As explained above, both `text` and `data` may change. `bss` shouldn't usually change. It'll be useful to see:
* Percentage change in text + data (indicating required flash/ROM size)
* Percentage change in data + bss (indicating required RAM size)
Speed up compilation of `type-system-chess`
[`type-system-chess`](https://github.com/rust-lang/rustc-perf/pull/1680) is an unusual program that implements a compile-time chess position solver in the trait system(!) This PR is about making it compile faster.
r? `@ghost`
Point at return type when it influences non-first `match` arm
When encountering code like
```rust
fn foo() -> i32 {
match 0 {
1 => return 0,
2 => "",
_ => 1,
}
}
```
Point at the return type and not at the prior arm, as that arm has type `!` which isn't influencing the arm corresponding to arm `2`.
Fix#78124.
Separate `consider_unsize_to_dyn_candidate` from other unsize candidates
Move the unsize candidate assembly *just for* `T -> dyn Trait` out of `assemble_candidates_via_self_ty` so that we only consider it once, instead of for every normalization step of the self ty. This makes sure that we don't assemble several candidates that are equal modulo normalization when we really don't care about normalizing the self type of an `T: Unsize<dyn Trait>` goal anyways.
Fixesrust-lang/trait-system-refactor-initiative#57
r? lcnr
Probe when assembling upcast candidates so they don't step on eachother's toes in new solver
Lack of a probe causes one candidate to disqualify the other due to inference side-effects.
r? lcnr
When encountering code like
```rust
fn foo() -> i32 {
match 0 {
1 => return 0,
2 => "",
_ => 1,
}
}
```
Point at the return type and not at the prior arm, as that arm has type
`!` which isn't influencing the arm corresponding to arm `2`.
Fix#78124.
normalize in `trait_ref_is_knowable` in new solver
fixes https://github.com/rust-lang/trait-system-refactor-initiative/issues/51
Alternatively we could avoid normalizing the self type and do this at the end of the `assemble_candidates_via_self_ty` stack by splitting candidates into:
- applicable without normalizing self type
- applicable for aliases, even if they can be normalized
- applicable for stuff which cannot get normalized further
I don't think this would have any significant benefits and it also seems non-trivial to avoid normalizing only the self type in `trait_ref_is_knowable`.
r? `@compiler-errors`
Fix a couple of bad comments
A couple of nits I saw. Sorry, this really should be folded into some other PR of mine, but I will literally forget if I don't put these up now.
Structurally normalize weak and inherent in new solver
It seems pretty obvious to me that we should be normalizing weak and inherent aliases too, since they can always be normalized. This PR still leaves open the question of what to do with opaques, though 💀
**Also**, we need to structurally resolve the target of a coercion, for the UI test to work.
r? `@lcnr`
Store the laziness of type aliases in their `DefKind`
Previously, we would treat paths referring to type aliases as *lazy* type aliases if the current crate had lazy type aliases enabled independently of whether the crate which the alias was defined in had the feature enabled or not.
With this PR, the laziness of a type alias depends on the crate it is defined in. This generally makes more sense to me especially if / once lazy type aliases become the default in a new edition and we need to think about *edition interoperability*:
Consider the hypothetical case where the dependency crate has an older edition (and thus eager type aliases), it exports a type alias with bounds & a where-clause (which are void but technically valid), the dependent crate has the latest edition (and thus lazy type aliases) and it uses that type alias. Arguably, the bounds should *not* be checked since at any time, the dependency crate should be allowed to change the bounds at will with a *non*-major version bump & without negatively affecting downstream crates.
As for the reverse case (dependency: lazy type aliases, dependent: eager type aliases), I guess it rules out anything from slight confusion to mild annoyance from upstream crate authors that would be caused by the compiler ignoring the bounds of their type aliases in downstream crates with older editions.
---
This fixes#114468 since before, my assumption that the type alias associated with a given weak projection was lazy (and therefore had its variances computed) did not necessarily hold in cross-crate scenarios (which [I kinda had a hunch about](https://github.com/rust-lang/rust/pull/114253#discussion_r1278608099)) as outlined above. Now it does hold.
`@rustbot` label F-lazy_type_alias
r? `@oli-obk`
Bubble up nested goals from equation in `predicates_for_object_candidate`
This used to be needed for https://github.com/rust-lang/rust/pull/114036#discussion_r1273987510, but since it's no longer, I'm opening this as a separate PR. This also fixes one ICEing UI test: (`tests/ui/unboxed-closures/issue-53448.rs`)
r? `@lcnr`
update overflow handling in the new trait solver
implements https://hackmd.io/QY0dfEOgSNWwU4oiGnVRLw?view. I want to clean up this doc and add it to the rustc-dev-guide, but I think this PR is ready for merge as is, even without the dev-guide entry.
r? `@compiler-errors`
Improve spans for indexing expressions
fixes#114388
Indexing is similar to method calls in having an arbitrary left-hand-side and then something on the right, which is the main part of the expression. Method calls already have a span for that right part, but indexing does not. This means that long method chains that use indexing have really bad spans, especially when the indexing panics and that span in coverted into a panic location.
This does the same thing as method calls for the AST and HIR, storing an extra span which is then put into the `fn_span` field in THIR.
r? compiler-errors
Indexing is similar to method calls in having an arbitrary
left-hand-side and then something on the right, which is the main part
of the expression. Method calls already have a span for that right part,
but indexing does not. This means that long method chains that use
indexing have really bad spans, especially when the indexing panics and
that span in coverted into a panic location.
This does the same thing as method calls for the AST and HIR, storing an
extra span which is then put into the `fn_span` field in THIR.
Rework upcasting confirmation to support upcasting to fewer projections in target bounds
This PR implements a modified trait upcasting algorithm that is resilient to changes in the number of associated types in the bounds of the source and target trait objects.
It does this by equating each bound of the target trait ref individually against the bounds of the source trait ref, rather than doing them all together by constructing a new trait object.
#### The new way we do trait upcasting confirmation
1. Equate the target trait object's principal trait ref with one of the supertraits of the source trait object's principal.
fdcab310b2/compiler/rustc_trait_selection/src/traits/select/mod.rs (L2509-L2525)
2. Make sure that every auto trait in the *target* trait object is present in the source trait ref's bounds.
fdcab310b2/compiler/rustc_trait_selection/src/traits/select/mod.rs (L2559-L2562)
3. For each projection in the *target* trait object, make sure there is exactly one projection that equates with it in the source trait ref's bound. If there is more than one, bail with ambiguity.
fdcab310b2/compiler/rustc_trait_selection/src/traits/select/mod.rs (L2526-L2557)
* Since there may be more than one that applies, we probe first to check that there is exactly one, then we equate it outside of a probe once we know that it's unique.
4. Make sure the lifetime of the source trait object outlives the lifetime of the target.
<details>
<summary>Meanwhile, this is how we used to do upcasting:</summary>
1. For each supertrait of the source trait object, take that supertrait, append the source object's projection bounds, and the *target* trait object's auto trait bounds, and make this into a new object type:
d12c6e947c/compiler/rustc_trait_selection/src/traits/select/confirmation.rs (L915-L929)
2. Then equate it with the target trait object:
d12c6e947c/compiler/rustc_trait_selection/src/traits/select/confirmation.rs (L936)
This will be a type mismatch if the target trait object has fewer projection bounds, since we compare the bounds structurally in relate:
d12c6e947c/compiler/rustc_middle/src/ty/relate.rs (L696-L698)
</details>
Fixes#114035
Also fixes#114113, because I added a normalize call in the old solver.
r? types
resolve before canonicalization in new solver, ICE if unresolved
Fold the values with a resolver before canonicalization instead of making it happen within canonicalization.
This allows us to filter trivial region constraints from the external constraints.
r? ``@lcnr``
