Also known as E0747 and E0107.
And by the way, rewrite how we lower generic arguments and deduplicate it between paths and method calls. The new version is taken almost straight from rustc.
This commit also changes the binders of `generic_defaults()`, to only include the binders of the arguments up to (and not including) the current argument. This make it easier to handle it in the rewritten lowering of generic args. It's also how rustc does it.
And make more queries non-interned.
Also flip the default for queries, now the default is to not intern and to intern a query you need to say `invoke_interned`.
That is, all queries that have a `with_diagnostic` variant.
Them being tracked was (maybe) needed before #19462, because back then diagnostics could refer `AstId`s (for macro types), but now they are no longer needed.
Also fix the `needs_drop()` intrinsic.
Unions also need this information (to err if they have a drop-needing field), but this will come in a follow-up PR.
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.
This commit also adds `tracing` to NotificationDispatcher/RequestDispatcher,
bumps `rust-analyzer-salsa` to 0.17.0-pre.6, `always-assert` to 0.2, and
removes the homegrown `hprof` implementation in favor of a vendored
tracing-span-tree.
MIR episode 5
This PR inits drop support (it is very broken at this stage, some things are dropped multiple time, drop scopes are wrong, ...) and adds stdout support (`println!` doesn't work since its expansion is dummy, but `stdout().write(b"hello world\n")` works if you use `RA_SYSROOT_HACK`) for interpreting. There is no useful unit test that it can interpret yet, but it is a good sign that it didn't hit a major road block yet.
In MIR lowering, it adds support for slice pattern and anonymous const blocks, and some fixes so that we can evaluate `SmolStr::new_inline` in const eval. With these changes, 57 failed mir body remains.
