fix regressions on assignment expressions
This is a follow-up PR on #12428. I'm not sure if this is everything I overlooked, so if there are more things that are not right, we may want to revert #12428.
This should also fix the increase of the type mismatches and the unknown types in diesel in the [metrics](https://rust-analyzer.github.io/metrics/?start=2022-06-23&end=2022-07-01) introduced by #12428.
The regressions are:
- some coercions don't work in the ordinary (i.e. non-destructuring) assignments
In order for coercions on ADT fields instantiations to work, lhs type has to be known before inferring rhs. #12428 changed the inference order, making rhs inferred before lhs, breaking the coercion, so I restored the original inference mechanism for the ordinary assignments.
Note that this kind of coercion doesn't happen in destructuring assigments, because when they are desugared, the struct expression is first assigned to a temporary, which is then assigned to the assignee, which is not coercion site anymore.
- type mismatches on individual identifiers are not reported
fix: Simplify macro statement expansion handling
I only meant to fix https://github.com/rust-lang/rust-analyzer/issues/12644 but that somehow turned into a rewrite of the statement handling ... at least this fixes a few more issues in the IDE layer now
fix: improve whitespace insertion in pretty printer
Fixes#12591
The `=>` token in the macro_rules! should be parsed as one fat arrow, but it ["requires a lot of changes in r-a"](143cc528b1), so I left it for the larger refactoring in the future and put a FIXME note.
feat: Show witnesses of non-exhaustiveness in `missing-match-arm` diagnostic
Shamelessly copied from rustc. Thus reporting format is same.
This extends public api `hir::diagnostics::MissingMatchArms` with `uncovered_patterns: String` field. It does not expose data for implementing a quick fix yet.
-----
Worth to note: current implementation does not give a comprehensive list of missing patterns. Also mentioned in [paper](http://moscova.inria.fr/~maranget/papers/warn/warn.pdf):
> One may think that algorithm I should make an additional effort to provide more
> non-matching values, by systematically computing recursive calls on specialized
> matrices when possible, and by returning a list of all pattern vectors returned by
> recursive calls. We can first observe that it is not possible in general to supply the
> users with all non-matching values, since the signature of integers is (potentially)
> infinite.
fix: trailing ':' on empty inactive reasons
## Description
Fixes trailing ':' even when there is no explanation. e.g.
``` sh
code is inactive due to #[cfg] directives:
```
## Issue
Fixes: #12615
feat: implement destructuring assignment
This is an attempt to implement destructuring assignments, or more specifically, type inference for [assignee expressions](https://doc.rust-lang.org/reference/expressions.html#place-expressions-and-value-expressions).
I'm not sure if this is the right approach, so I don't even expect this to be merged (hence the branch name 😉) but rather want to propose one direction we could choose. I don't mind getting merged if this is good enough though!
Some notes on the implementation choices:
- Assignee expressions are **not** desugared on HIR level unlike rustc, but are inferred directly along with other expressions. This matches the processing of other syntaxes that are desugared in rustc but not in r-a. I find this reasonable because r-a only needs to infer types and it's easier to relate AST nodes and HIR nodes, so I followed it.
- Assignee expressions obviously resemble patterns, so type inference for each kind of pattern and its corresponding assignee expressions share a significant amount of logic. I tried to reuse the type inference functions for patterns by introducing `PatLike` trait which generalizes assignee expressions and patterns.
- This is not the most elegant solution I suspect (and I really don't like the name of the trait!), but it's cleaner and the change is smaller than other ways I experimented, like making the functions generic without such trait, or making them take `Either<ExprId, PatId>` in place of `PatId`.
in case this is merged:
Closes#11532Closes#11839Closes#12322
fix: Report proc macro errors in expressions correctly as well
They didn't have a krate before, resulting in the generic "proc macro not found" error.
Also improve error messages a bit more.
