eagerly compute `sub_unification_table` again
Previously called `sub_relations`. We still only using them for diagnostics right now. This mostly reverts rust-lang/rust#119989. Necessary for type inference guidance due to not-yet defined opaque types, cc https://github.com/rust-lang/trait-system-refactor-initiative/issues/182.
We could use them for cycle detection in generalization and it seems desirable to do so in the future. However, this is unsound with the old trait solver as its cache does not track these `sub_unification_table` in any way.
We now properly track the `sub_unification_table` when canonicalizing so using them in the new solver is totally sound and the performance impact is far more manageable than I thought back in rust-lang/rust#119989.
r? `@compiler-errors`
Migrate more things in the new solver to specific `DefId`s
Continuation of https://github.com/rust-lang/rust/pull/145377. I migrated the rest of the types, except aliases.
Aliases are problematic because opaques and associated types share the same type in the new solver. `@jackh726,` `@lcnr,` `@ShoyuVanilla` I'd like to hear ideas here. Anyway, even if we do nothing with them we already got a substantial improvement.
r? types
cleanup and cache proof tree building
There's some cruft left over from when we had deep proof trees. We never encounter overflow when evaluating proof trees. Even if the recursion limit is `0`, we still only hit the overflow limit when evaluating nested goals of the root. The root goal simply inherits the `root_depth` of the `SearchGraph`.
Split `evaluate_root_goal_for_proof_tree` from the rest of the trait solver. This enables us to simplify the implementation of `evaluate_goal_raw` and the `ProofTreeBuilder` as we no longer need to manually track the state of the builder and can instead use separate types for that. It does require making a few internal methods into associated functions taking a `delegate` and a `span` instead of the `EvalCtxt` itself.
I've also split `SearchGraph::evaluate_goal` and `SearchGraph::evaluate_root_goal_for_proof_tree` for the same reason. Both functions don't actually share too much code, so by splitting them each version gets significantly easier to read.
Add a `query evaluate_root_goal_for_proof_tree_raw` to cache proof tree building. This requires arena allocating `inspect::Probe`. I've added a new type alias `I::ProbeRef` for this. We may need to adapt this for rust-analyzer? It would definitely be easy to remove the `Copy` bound here 🤔
This was done in #145740 and #145947. It is causing problems for people
using r-a on anything that uses the rustc-dev rustup package, e.g. Miri,
clippy.
This repository has lots of submodules and subtrees and various
different projects are carved out of pieces of it. It seems like
`[workspace.dependencies]` will just be more trouble than it's worth.
Switch next solver to use a specific associated type for trait def id
The compiler just puts `DefId` in there, but rust-analyzer uses different types for each kind of item.
See [the Zulip discussion](https://rust-lang.zulipchat.com/#narrow/channel/185405-t-compiler.2Frust-analyzer/topic/Implmentating.20New.20Trait.20Solver/near/534329794). In short, it will be a tremendous help to r-a to use specific associated types, while for the solver and the compiler it's a small change. So I ported `TraitId`, as a proof of concept and it's also likely the most impactful.
r? types
Do not consider a `T: !Sized` candidate to satisfy a `T: !MetaSized` obligation.
This example should fail to compile (and does under this PR, with the old and new solvers), but currently compiles successfully ([playground](https://play.rust-lang.org/?version=nightly&mode=debug&edition=2024&gist=6e0e5d0ae0cdf0571dea97938fb4a86d)), because (IIUC) the old solver's `lazily_elaborate_sizedness_candidate`/callers and the new solver's `TraitPredicate::fast_reject_assumption`/`match_assumption` consider a `T: _ Sized` candidate to satisfy a `T: _ MetaSized` obligation, for either polarity `_`, when that should only hold for positive polarity.
```rs
#![feature(negative_bounds)]
#![feature(sized_hierarchy)]
use std::marker::MetaSized;
fn foo<T: !MetaSized>() {}
fn bar<T: !Sized + MetaSized>() {
foo::<T>();
//~^ ERROR the trait bound `T: !MetaSized` is not satisfied // error under this PR
}
```
Only observable with the internal-only `feature(negative_bounds)`, so might just be "wontfix".
This example is added as a test in this PR (as well as testing that `foo<()>` and `foo<str>` are disallowed for `fn foo<T: !MetaSized`).
cc `@davidtwco` for `feature(sized_hierarchy)`
Maybe similar to 91c53c9 from <https://github.com/rust-lang/rust/pull/143307>
some clauses can be merged together without requiring an attribute for
each trait derived.
also manually impl `Eq` because the `derive_where` generated code is too
much for my comfort
Distinguish prepending and replacing self ty in predicates
There are two kinds of functions called `with_self_ty`:
1. Prepends the `Self` type onto an `ExistentialPredicate` which lacks it in its internal representation.
2. Replaces the `Self` type of an existing predicate, either for diagnostics purposes or in the new trait solver when normalizing that self type.
This PR distinguishes these two because I often want to only grep for one of them. Namely, let's call it `with_replaced_self_ty` when all we're doing is replacing the self type.
Remove the witness type from coroutine *args* (without actually removing the type)
This does as much of rust-lang/rust#144157 as we can without having to break rust-lang/rust#143545 and/or introduce some better way of handling higher ranked assumptions.
Namely, it:
* Stalls coroutines based off of the *coroutine* type rather than the witness type.
* Reworks the dtorck constraint hack to not rely on the witness type.
* Removes the witness type from the args of the coroutine, eagerly creating the type for nested obligations when needed (auto/clone impls).
I'll experiment with actually removing the witness type in a follow-up.
r? lcnr
Deduplicate `IntTy`/`UintTy`/`FloatTy`.
There are identical definitions in `rustc_type_ir` and `rustc_ast`. This commit removes them and places a single definition in `rustc_ast_ir`. This requires adding `rust_span` as a dependency of `rustc_ast_ir`, but means a bunch of silly conversion functions can be removed.
r? `@fmease`
uniquify root goals during HIR typeck
We need to rely on region identity to deal with hangs such as https://github.com/rust-lang/trait-system-refactor-initiative/issues/210 and to keep the current behavior of `fn try_merge_responses`.
This is a problem as borrowck starts by replacing each *occurrence* of a region with a unique inference variable. This frequently splits a single region during HIR typeck into multiple distinct regions. As we assume goals to always succeed during borrowck, relying on two occurances of a region being identical during HIR typeck causes ICE. See the now fixed examples in https://github.com/rust-lang/trait-system-refactor-initiative/issues/27 and rust-lang/rust#139409.
We've previously tried to avoid this issue by always *uniquifying* regions when canonicalizing goals. This prevents caching subtrees during canonicalization which resulted in hangs for very large types. People rely on such types in practice, which caused us to revert our attempt to reinstate `#[type_length_limit]` in https://github.com/rust-lang/rust/pull/127670. The complete list of changes here:
- rust-lang/rust#107981
- rust-lang/rust#110180
- rust-lang/rust#114117
- rust-lang/rust#130821
After more consideration, all occurrences of such large types need to happen outside of typeck/borrowck. We know this as we already walk over all types in the MIR body when replacing their regions with nll vars.
This PR therefore enables us to rely on region identity inside of the trait solver by exclusively **uniquifying root goals during HIR typeck**. These are the only goals we assume to hold during borrowck. This is insufficient as type inference variables may "hide" regions we later uniquify. Because of this, we now stash proven goals which depend on inference variables in HIR typeck and reprove them after writeback. This closes https://github.com/rust-lang/trait-system-refactor-initiative/issues/127.
This was originally part of rust-lang/rust#144258 but I've moved it into a separate PR. While I believe we need to rely on region identity to fix the performance issues in some way, I don't know whether rust-lang/rust#144258 is the best approach to actually do so. Regardless of how we deal with the hangs however, this change is necessary and desirable regardless.
r? `@compiler-errors` or `@BoxyUwU`
Skip walking into param-env component if it has no placeholder/re-var
Although it only provides a minor perf improvement, it seems like it could matter in more pathological cases.
