This mainly aids in error recovery but also makes it a bit easier to handle lifetime resolution.
While doing so it also came apparent that we were not actually lowering lifetime outlives relationships within lifetime parameter declaration bounds, so this fixes that.
Implement diagnostics in all places left: generics (predicates, defaults, const params' types), fields, and type aliases.
Unfortunately this results in a 20mb addition in `analysis-stats .` due to many type methods returning an addition diagnostics result now (even if it's `None` in most cases). I'm not sure if this can be improved.
An alternative strategy that can prevent the memory usage growth is to never produce diagnostics in hir-ty methods. Instead, lower all types in the hir crate when computing diagnostics from scratch (with diagnostics this time). But this has two serious disadvantages:
1. This can cause code duplication (although it can probably be not that bad, it will still mean a lot more code).
2. I believe we eventually want to compute diagnostics for the *entire* workspace (either on-type or on-save or something alike), so users can know when they have diagnostics even in inactive files. Choosing this approach will mean we lose all precomputed salsa queries. For one file this is fine, for the whole workspace this will be very slow.
The diagnostic implemented is a simple one (E0109). It serves as a test for the new foundation.
This commit only implements diagnostics for type in bodies and body-carrying signatures; the next commit will include diagnostics in the rest of the things.
Also fix one weird bug that was detected when implementing this that caused `Fn::(A, B) -> C` (which is a valid, if bizarre, alternative syntax to `Fn(A, B) -> C` to lower incorrectly.
And also fix a maybe-bug where parentheses were sneaked into a code string needlessly; this was not detected until now because the parentheses were removed (by the make-AST family API), but with a change in this commit they are now inserted. So fix that too.
Most paths are types and therefore already are in the source map, but the trait in impl trait and in bounds are not.
We do this by storing them basically as `TypeRef`s. For convenience, I created a wrapper around `TypeRefId` called `PathId` that always stores a path, and implemented indexing from the types map to it.
Fortunately, this change impacts memory usage negligibly (adds 2mb to `analysis-stats .`, but that could be just fluff). Probably because there aren't that many trait bounds and impl traits, and this also shrinks `TypeBound` by 8 bytes.
I also added an accessor to `TypesSourceMap` to get the source code, which will be needed for diagnostics.
We add union fields access (in both expressions and patterns) and inline assembly.
That completes the unsafe check (there are some other unsafe things but they are unstable), and so also opens the door to reporting unused unsafe without annoying people about their not-unused unsafe blocks.
When a glob import overriding the visibility of a previous glob import was not properly resolved when the items are only available in the next fixpoint iteration.
The bug was hidden until #18390.
I.e. the following situation:
```
fn foo() {
mod bar {
fn qux() {
// Prelude path here (e.g. macro use prelude or extern prelude).
}
}
}
```
Those were previously unresolved, because, in order to support `self` and `super` properly, since #15148 we do not ascend block paths when there is a module in between, but only crate def maps register preludes, not block def maps, and we can't change this because block def map prelude can always be overridden by another block. E.g.
```
fn foo() {
struct WithTheSameNameAsPreludeItem;
{
WithTheSameNameAsPreludeItem
}
}
```
Here `WithTheSameNameAsPreludeItem` refer to the item from the top block, but if we would register prelude items in each block the child block would overwrite it incorrectly.
Thanks to the observation (supported by counting) that the vast majority paths have neither generics no type anchors, and thanks to a new datastructure `ThinVecWithHeader` that is essentially `(T, Box<[U]>)` but with the size of a single pointer, we are able to reach this feat.
This (together with `ThinVecWithHeader`) makes the possibility to shrink `TypeRef`, because most types are paths.
So that given a `TypeRef` we will be able to trace it back to source code.
This is necessary to be able to provide diagnostics for lowering to chalk tys, since the input to that is `TypeRef`.
This means that `TypeRef`s now have an identity, which means storing them in arena and not interning them, which is an unfortunate (but necessary) loss but also a pretty massive change. Luckily, because of the separation layer we have for IDE and HIR, this change never crosses the IDE boundary.
E.g.:
```rust
let v;
macro_rules! m { () => { v }; }
```
This was an existing bug, but it was less severe because unless the variable was shadowed it would be correctly resolved. With hygiene however, without this fix the variable is never resolved.
Or macro_rules hygiene, or mixed site hygiene. In other words, hygiene for variables and labels but not items.
The realization that made me implement this was that while "full" hygiene (aka. def site hygiene) is really hard for us to implement, and will likely involve intrusive changes and performance losses, since every `Name` will have to carry hygiene, mixed site hygiene is very local: it applies only to bodies, and we very well can save it in a side map with minor losses.
This fixes one diagnostic in r-a that was about `izip!()` using hygiene (yay!) but it introduces a huge number of others, because of #18262. Up until now this issue wasn't a major problem because it only affected few cases, but with hygiene identifiers referred by macros like that are not resolved at all. The next commit will fix that.
