Don't allow `poly_select` in new solver
I added a `poly_select` call in #140519, but this causes an ICE since the new solver doesn't properly handle the "instantiate binder -> recanonicalize" step in the proof tree visitor.
While we could fix the select visitor to look at the next step in proof tree, it's not really necessary. Instead, let's enforce that all callees call the non-higher-ranked `select` function in the new solver.
Fixes https://github.com/rust-lang/rust/issues/141322
r? lcnr
Normalize aliases to correct kind of error term
Fixes#140642
When normalizing an alias to an error in the old solver, normalize to the same term kind as the alias being normalized instead of always to a type error.
r? lcnr
Fix `FnOnce` impl for `AsyncFn`/`AsyncFnMut` self-borrowing closures in new solver
This only affects closures that are "`AsyncFn`/`AsyncFnMut`" in their calling capability that are being called with the `FnOnce` trait.
fixes https://github.com/rust-lang/trait-system-refactor-initiative/issues/217
r? lcnr
Fast path for processing some obligations in the new solver
Fast path applies to:
- Dyn compatibility predicates
- Region and type outlives predicates
- Trivially sized predicates
dont handle bool transmute
removes `transmute(u8) -> bool` suggestion due to ambiguity, leave it for clippy
elaboration on ambiguity in question:
`transmute::<u8, bool>(x)` will codegen to an `assume(u8 < 2)`;
`_ == 1` or `_ != 0` or `_ % 2 == 0` would remove that assumption
`match _ { x @ (0 | 1) => x == 1, _ => std::hint::unreachable_unchecked() }` is very verbose
`@rustbot` label L-unnecessary_transmutes
Document why we allow escaping bound vars in LTA norm
r? lcnr
followup from the const normalization PR. I think I now understand why free alias norm is funny about binders
`unpretty=thir-tree`: don't require the final expr to be the body's value
Two motivations for this:
- I couldn't find a comment motivating this hard-coding. I can imagine it might be easier to read `unpretty=thir-flat` output if the final expression in the THIR is always the body's value, but if that's the reason, that should be the justification in the source. I can also imagine it's meant to check that all expressions will be visited by the pretty-printer, but the existing check doesn't quite do that either.
- Guard patterns (#129967) contain expressions, so lowering params containing guard patterns may add more expressions to the THIR. Currently a body's params are lowered after its expression, so guard expressions in params would end up last, breaking this. As an alternative, the params could be lowered first (#141356).
lower bodies' params to thir before the body's value
Two motivations for this:
- Lowering params first means errors from lowering the params are emitted before errors from lowering the body's expression. This comes up in [tests/ui/associated-consts/associated-const-type-parameter-pattern.stderr](https://github.com/rust-lang/rust/compare/master...dianne:rust:thir-lower-params-before-body-expr?expand=1#diff-acac6ea10e991af0da91633e08b2739f9f9ca0c8f826401b6ba829914d0806f2), where both the params and expression encounter errors in translating consts to patterns. This change puts the errors in the order they appear in the source file.
- Guard patterns (#129967) contain expressions, so lowering params containing guard patterns may add more expressions to the THIR. However, there's a check for `-Zunpretty=thir-tree` that the final expression in the THIR corresponds to its value [(link)](c43786c9b7/compiler/rustc_mir_build/src/builder/mod.rs (L453-L455)); lowering params last would break this. As an alternative way to get guard patterns to work, I think the pretty-printer could use the expression returned by `thir_body` and the check could be removed or changed (#141357).
incorrectly prefer builtin `dyn` impls :3
This makes #57893 slightly more exploitable with the new solver. It's still strictly better than the old solver and the underlying unsoundness persists in the new one even without this preference.
Properly fixing #57893 is something we've been looking at more deeply recently and discussed at the [Types Meetup during the All-Hands](https://hackmd.io/rz-4ghMzTb2wXOkdLKHaHw#Dyn-traits). Whatever approach we'll end up deciding on will likely require a fairly long transition period and some significant further design work. This should not block `-Znext-solver`.
fixes https://github.com/rust-lang/trait-system-refactor-initiative/issues/183
r? `@compiler-errors` cc `@rust-lang/types`
Rollup of 7 pull requests
Successful merges:
- #141267 (only resolve top-level guard patterns' guards once)
- #141280 (Use Docker cache from the current repository)
- #141296 (Async drop fix for 'broken mir, place has deref as later projection')
- #141328 (When AsyncDrop impl is empty, sync drop generated in elaborator)
- #141332 (Do not eagerly fold consts in `normalize_param_env_or_error` if new solver)
- #141333 (Use `DeepRejectCtxt` in `assemble_inherent_candidates_from_param`)
- #141334 (eagerly check nested obligations when coercing fndefs)
r? `@ghost`
`@rustbot` modify labels: rollup
Do not eagerly fold consts in `normalize_param_env_or_error` if new solver
Fixes https://github.com/rust-lang/trait-system-refactor-initiative/issues/213
Given:
```
trait Trait: Deref<Target = [u8; { 1 + 1 }]> {}
```
when elaborating param env for `Trait`, we have `Self: Trait`, `Self: Deref<Target = [u8; {anon const}]>`.
Before this PR, we would fold the anon consts away *before* elaborating. However, we end up getting another *un-folded* copy of the anon const from elaborating `Self: Trait`. This leads to normalization ambiguity.
r? lcnr
only resolve top-level guard patterns' guards once
We resolve guard patterns' guards in `resolve_pattern_inner`, so to avoid resolving them multiple times, we must avoid doing so earlier. To accomplish this, `LateResolutionVisitor::visit_pat` contains a case for guard patterns that avoids visiting their guards while walking patterns.
This PR fixes#141265, which was due to `visit::walk_pat` being used instead; this meant guards at the top level of a pattern would be visited twice. e.g. it would ICE on `for x if x in [] {}`, but not `for (x if x) in [] {}`. `visit_pat` was already used for the guard pattern in the second example, on account of the top-level pattern being parens.
Match on lang item kind instead of using an if/else chain
Similar to how the new solver does this. Just noticed while I was adding a new entry to the chain 😆
typeck: catch `continue`s pointing to blocks
This taints the typeck results with errors if a `continue` is found not pointing to a loop.
A few things were going wrong here. First, since this wasn't caught in typeck, we'd end up building the THIR and then running liveness lints on ill-formed HIR. Since liveness assumes all `continue`s point to loops, it wasn't setting a live node for the `continue`'s destination. There was a fallback for if it couldn't retrieve that live node, but it was faulty; it would create a new live node to represent an erroneous state after the analysis's RWU table had already been built. This would ICE if the new live node was used in operations, such as merging results from the arms of a match. I've removed this error-recovery since it was buggy, and we should really catch bad labels before liveness.
I've also replaced an outdated comment about when liveness lints are run. At this point, I think the call to `check_liveness` could be moved elsewhere, but if it can be run when the typeck results are tainted by errors, it'll need some slight refactoring so it can bail out in that case. In lieu of that, I've added an assert.
Fixes#113379Fixes#121623
Allow trailing comma after argument in query definition
Don't catastrophically fail the query macro if you put a comma after your query key!
r? oli-obk
Suggest use "{}", self.x instead of {self.x} when resolve x as field of `self`
Fixes#141136
Changes can be seen in the second commit: 9de7fff0d8
r? compiler
collect doc alias as tips during resolution
Close#124273
Collect the symbol in the doc alias attributes and provide a tip when a match is found.
r? `@estebank`
This taints the typeck results with errors if a `continue` is found not
pointing to a loop, which fixes an ICE.
A few things were going wrong here. First, since this wasn't caught in
typeck, we'd end up building the THIR and then running liveness lints on
ill-formed HIR. Since liveness assumes all `continue`s point to loops,
it wasn't setting a live node for the `continue`'s destination. However,
the fallback for this was faulty; it would create a new live node to
represent the erroneous state after the analysis's RWU table had already
been built. This would ICE if the new live node was used in operations,
such as merging results from the arms of a match. I've removed this
error-recovery since it was buggy, and we should really catch bad labels
before liveness.
I've also replaced an outdated comment about when liveness lints are
run. At this point, I think the call to `check_liveness` could be moved
elsewhere, but if it can be run when the typeck results are tainted by
errors, it'll need some slight refactoring so it can bail out in that
case. In lieu of that, I've added an assertion.
Add TRACING_ENABLED to Machine and add enter_trace_span!()
This PR adds the necessary infrastructure to make it possible to do tracing calls from within `rustc_const_eval` when running Miri, while making sure they don't impact the performance of normal compiler execution. This is done by adding a `const` boolean to `Machine`, false by default, but that will be set to true in Miri only. The tracing macro `enter_trace_span!()` checks if it is true before doing anything, and since the value of a `const` is known at compile time, if it it false it the whole tracing call should be optimized out.
I will soon open further PRs to add tracing macro calls similar to this one, so that afterwards it will be possible to learn more about Miri's time spent in the various interpretation steps:
```rs
let _guard = enter_trace_span!(M, "eval_statement", "{:?}", stmt);
```
r? `@RalfJung`
