Make fewer types generic over QueryContext
While trying to refactor `rustc_query_system::query::QueryContext` to make it dyn-safe, I noticed some smaller things:
* QueryConfig doesn't need to be generic over QueryContext
* ~~The `kind` field on QueryJobId is unused~~
* Some unnecessary where clauses
* Many types in `job.rs` where generic over `QueryContext` but only needed `QueryContext::Query`.
If handle_cycle_error() could be refactored to not take `error: CycleError<CTX::Query>`, all those bounds could be removed as well.
Changing `find_cycle_in_stack()` in job.rs to not take a `tcx` argument is the only functional change here. Everything else is just updating type signatures. (aka compile-error driven development ^^)
~~Currently there is a weird bug where memory usage suddenly skyrockets when running UI tests. I'll investigate that tomorrow.
A perf run probably won't make sense before that is fixed.~~
EDIT: `kind` actually is used by `Eq`, and re-adding it fixed the memory issue.
Use `DroplessArena` where we know the type doesn't need drop
This PR uses a single `DroplessArena` in resolve instead of three separate `TypedArena`s.
`DroplessArena` checks that the type indeed doesn't need drop, so in case the types change, this will result in visible failures.
Rollup of 10 pull requests
Successful merges:
- #77420 (Unify const-checking structured errors for `&mut` and `&raw mut`)
- #77554 (Support signed integers and `char` in v0 mangling)
- #77976 (Mark inout asm! operands as used in liveness pass)
- #78009 (Haiku: explicitly set CMAKE_SYSTEM_NAME when cross-compiling)
- #78084 (Greatly improve display for small mobile devices screens)
- #78155 (Fix two small issues in compiler/rustc_lint/src/types.rs)
- #78156 (Fixed build failure of `rustfmt`)
- #78172 (Add test case for #77062)
- #78188 (Add tracking issue number for pin_static_ref)
- #78200 (Add `ControlFlow::is_{break,continue}` methods)
Failed merges:
r? `@ghost`
Mark inout asm! operands as used in liveness pass
Variables used in `inout` operands in inline assembly (that is, they're used as both input and output to some arbitrary assembly instruction) are being marked as read and written, but are not marked as being used in the RWU table during the liveness pass. This can result in such expressions triggering an unused variable lint warning. This is incorrect behavior- reads without uses are currently only used for compound assignments. We conservatively assume that an `inout` operand is being read and used in the context of the assembly instruction.
Closes#77915
Support signed integers and `char` in v0 mangling
Likely we want more tests, to check the output is correct too: however, I wasn't sure what kind of test we needed, so I just added one similar to that added in https://github.com/rust-lang/rust/pull/77452 for now.
r? @eddyb
Calculate visibilities once in resolve
Then use them through a query based on resolver outputs.
Item visibilities were previously calculated in three places - initially in `rustc_resolve`, then in `rustc_privacy` during type privacy checkin, and then in `rustc_metadata` during metadata encoding.
The visibility logic is not entirely trivial, especially for things like constructors or enum variants, and all of it was duplicated.
This PR deduplicates all the visibility calculations, visibilities are determined once during early name resolution and then stored in `ResolverOutputs` and are later available through `tcx` as a query `tcx.visibility(def_id)`.
(This query existed previously, but only worked for other crates.)
Some special cases (e.g. visibilities for closure types, which are needed for type privacy checking) are not processed in resolve, but deferred and performed directly in the query instead.
Rewrite `collect_tokens` implementations to use a flattened buffer
Instead of trying to collect tokens at each depth, we 'flatten' the
stream as we go allong, pushing open/close delimiters to our buffer
just like regular tokens. One capturing is complete, we reconstruct a
nested `TokenTree::Delimited` structure, producing a normal
`TokenStream`.
The reconstructed `TokenStream` is not created immediately - instead, it is
produced on-demand by a closure (wrapped in a new `LazyTokenStream` type). This
closure stores a clone of the original `TokenCursor`, plus a record of the
number of calls to `next()/next_desugared()`. This is sufficient to reconstruct
the tokenstream seen by the callback without storing any additional state. If
the tokenstream is never used (e.g. when a captured `macro_rules!` argument is
never passed to a proc macro), we never actually create a `TokenStream`.
This implementation has a number of advantages over the previous one:
* It is significantly simpler, with no edge cases around capturing the
start/end of a delimited group.
* It can be easily extended to allow replacing tokens an an arbitrary
'depth' by just using `Vec::splice` at the proper position. This is
important for PR #76130, which requires us to track information about
attributes along with tokens.
* The lazy approach to `TokenStream` construction allows us to easily
parse an AST struct, and then decide after the fact whether we need a
`TokenStream`. This will be useful when we start collecting tokens for
`Attribute` - we can discard the `LazyTokenStream` if the parsed
attribute doesn't need tokens (e.g. is a builtin attribute).
The performance impact seems to be neglibile (see
https://github.com/rust-lang/rust/pull/77250#issuecomment-703960604). There is a
small slowdown on a few benchmarks, but it only rises above 1% for incremental
builds, where it represents a larger fraction of the much smaller instruction
count. There a ~1% speedup on a few other incremental benchmarks - my guess is
that the speedups and slowdowns will usually cancel out in practice.
Improve wording of "cannot multiply" type error
For example, if you had this code:
fn foo(x: i32, y: f32) -> f32 {
x * y
}
You would get this error:
error[E0277]: cannot multiply `f32` to `i32`
--> src/lib.rs:2:7
|
2 | x * y
| ^ no implementation for `i32 * f32`
|
= help: the trait `Mul<f32>` is not implemented for `i32`
However, that's not usually how people describe multiplication. People
usually describe multiplication like how the division error words it:
error[E0277]: cannot divide `i32` by `f32`
--> src/lib.rs:2:7
|
2 | x / y
| ^ no implementation for `i32 / f32`
|
= help: the trait `Div<f32>` is not implemented for `i32`
So that's what this change does. It changes this:
error[E0277]: cannot multiply `f32` to `i32`
--> src/lib.rs:2:7
|
2 | x * y
| ^ no implementation for `i32 * f32`
|
= help: the trait `Mul<f32>` is not implemented for `i32`
To this:
error[E0277]: cannot multiply `i32` by `f32`
--> src/lib.rs:2:7
|
2 | x * y
| ^ no implementation for `i32 * f32`
|
= help: the trait `Mul<f32>` is not implemented for `i32`
Drop unneeded `mut`
These parameters don't get modified.
Note that `trailing_comment` is pub and gets exported from `rustc_ast_pretty`. Is that considered to be a stable API? If yes, and you want to reserve the right to modify `self` in `trailing_comment` in the future, that hunk would need to be dropped.
Don't update `entries` in `TypedArena` if T does not need drop
As far as I can tell, `entries` is only used when dropping `TypedArenaChunk`s and their contents. It is already ignored there, if T is not `mem::needs_drop`, this PR just skips updating it's value.
You can see `TypedArenaChunk` ignoring the entry count in L71. The reasoning is similar to what you can find in `DroplessArena`.
r? @oli-obk
Improve `skip_binder` usage during FlagComputation
It looks like there was previously a bug around `ExistentialPredicate::Projection` here, don't know how to best trigger that one to add a regression test though.
Trait predicate ambiguities are not always in `Self`
When reporting ambiguities in trait predicates, the compiler incorrectly assumed the ambiguity was always in the type the trait should be implemented on, and never the generic parameters of the trait. This caused silly suggestions for predicates like `<KnownType as Trait<_>>`, such as giving explicit types to completely unrelated variables that happened to be of type `KnownType`.
This also reverts #73027, which worked around this issue in some cases and does not appear to be necessary any more.
fixes#77982fixes#78055
This optimization can result in unsoundness, because it introduces
additional uses of a place holding the discriminant value without
ensuring that it is valid to do so.
Instead of trying to collect tokens at each depth, we 'flatten' the
stream as we go allong, pushing open/close delimiters to our buffer
just like regular tokens. One capturing is complete, we reconstruct a
nested `TokenTree::Delimited` structure, producing a normal
`TokenStream`.
The reconstructed `TokenStream` is not created immediately - instead, it is
produced on-demand by a closure (wrapped in a new `LazyTokenStream` type). This
closure stores a clone of the original `TokenCursor`, plus a record of the
number of calls to `next()/next_desugared()`. This is sufficient to reconstruct
the tokenstream seen by the callback without storing any additional state. If
the tokenstream is never used (e.g. when a captured `macro_rules!` argument is
never passed to a proc macro), we never actually create a `TokenStream`.
This implementation has a number of advantages over the previous one:
* It is significantly simpler, with no edge cases around capturing the
start/end of a delimited group.
* It can be easily extended to allow replacing tokens an an arbitrary
'depth' by just using `Vec::splice` at the proper position. This is
important for PR #76130, which requires us to track information about
attributes along with tokens.
* The lazy approach to `TokenStream` construction allows us to easily
parse an AST struct, and then decide after the fact whether we need a
`TokenStream`. This will be useful when we start collecting tokens for
`Attribute` - we can discard the `LazyTokenStream` if the parsed
attribute doesn't need tokens (e.g. is a builtin attribute).
The performance impact seems to be neglibile (see
https://github.com/rust-lang/rust/pull/77250#issuecomment-703960604). There is a
small slowdown on a few benchmarks, but it only rises above 1% for incremental
builds, where it represents a larger fraction of the much smaller instruction
count. There a ~1% speedup on a few other incremental benchmarks - my guess is
that the speedups and slowdowns will usually cancel out in practice.
