Improve memoization and refactor NLL type check
I have a big branch that is refactoring NLL type check with the goal of introducing canonicalization-based memoization for all of the operations it does. This PR contains an initial prefix of that branch which, I believe, stands alone. It does introduce a few smaller optimizations of its own:
- Skip operations that are trivially a no-op
- Cache the results of the dropck-outlives computations done by liveness
- Skip resetting unifications if nothing changed
r? @pnkfelix
Add existential type definitions
Note: this does not allow creating named existential types, it just desugars `impl Trait` to a less (but still very) hacky version of actual `existential type` items.
r? @nikomatsakis
Re-enable trivial bounds
cc #50825
Remove implementations from global bounds in winnowing when there is ambiguity.
This results in the reverse of #24066 happening sometimes. I'm not sure if anything can be done about that though.
cc #48214
r? @nikomatsakis
Rollup of 13 pull requests
Successful merges:
- #50143 (Add deprecation lint for duplicated `macro_export`s)
- #51099 (Fix Issue 38777)
- #51276 (Dedup auto traits in trait objects.)
- #51298 (Stabilize unit tests with non-`()` return type)
- #51360 (Suggest parentheses when a struct literal needs them)
- #51391 (Use spans pointing at the inside of a rustdoc attribute)
- #51394 (Use scope tree depths to speed up `nearest_common_ancestor`.)
- #51396 (Make the size of Option<NonZero*> a documented guarantee.)
- #51401 (Warn on `repr` without hints)
- #51412 (Avoid useless Vec clones in pending_obligations().)
- #51427 (compiletest: autoremove duplicate .nll.* files (#51204))
- #51436 (Do not require stage 2 compiler for rustdoc)
- #51437 (rustbuild: generate full list of dependencies for metadata)
Failed merges:
Suggest parentheses when a struct literal needs them
When writing a struct literal in an expression that expects a block to
be started afterwards (like an `if` statement), do not suggest using the
same struct literal:
```
did you mean `S { /* fields * /}`?
```
Instead, suggest surrounding the expression with parentheses:
```
did you mean `(S { /* fields * /})`?
```
Fix#47360, #50090. Leaving #42982 open to come back to this problem with a better solution.
Dedup auto traits in trait objects.
Fixes#47010
Note that the test file `run-pass/trait-object-auto-dedup.rs` passes before and after this change. It's the `ui` test that changed from compiling to not compiling. Which does make this a breaking change, but I cannot imagine anybody actually being broken by it.
NLL: report type moved from behind borrow of array/slice
When NLL has illegal move due to borrowed content in an array or slice, provide feedback about why the move wasn't a copy.
Drive by: While comparing the resulting `.nll.stderr` files to their old borrowck variants, I noticed that the test for borrowck-vec-pattern-nesting.rs was not signaling some errors under NLL due to the test assuming lexical lifetimes. So I fixed that too.
Fix#51190
Removes extra global bounds at the winnowing stage rather than when
normalizing the param_env. This avoids breaking inference when there is
a global bound.
Deny #[cfg] and #[cfg_attr] on generic parameters.
Fix#51279.
Attributes on generic parameters are not expanded, meaning `#[cfg]`, `#[cfg_attr]` and attribute proc macros are entirely ignored on them.
This PR makes using the first two attributes an error, because if they are correctly expanded will affect the AST and change code behavior.
I'm beta-nominating this, because generic parameter attributes are stabilizing in 1.27, and if we did not reserve their usage, we may never be able to repurpose the meaning of these attributes in the Rust 2015 edition.
Accept `..` in incorrect position to avoid further errors
We currently give a specific message when encountering a `..` anywhere
other than the end of a pattern. Modify the parser to accept it (while
still emitting the error) so that we don't also trigger "missing fields
in pattern" errors afterwards.
Add suggestions to either remove trailing `,` or moving the `..` to the
end.
Follow up to #49268.
