First step towards the new tracked struct design.
SUBTLE: This is actually broken if the `#[id]`
fields have a broken hash. We'll fix that in the
next commit and add a test.
Requires unsafe, since Rust currently doesn't have built-in
support for guaranteed in-place initialization. Unfortunately,
this unsafety propagates to making `Jar` unsafe to implement
in order to guarantee that the jar is initialized, but since
the preferred way to implement it is via `salsa::jar`, this
won't impact most users.
440: Change the constructor of `salsa::Id` to const fn r=XFFXFF a=Y-Nak
This change is motivated to allow interned structs to have default const instances and prefill them in db initialization.
A specific example is managing symbols as a salsa interned struct in a compiler implementation. By declaring keywords as `const` symbols, we can handle keywords without going through their internal representation as rustc does (ref: [rustc_span::symbol](https://github.com/rust-lang/rust/blob/master/compiler/rustc_span/src/symbol.rs)).
Conceptually, it'd look like the one below.
```rust
#[salsa::interned]
pub struct Symbol {
data: String,
}
const SELF_SYM: Symbol = Symbol(salsa::Id::from_u32(1))
...
/// This function is called in db initialization.
fn prefill(db: &dyn HirDb) {
Symbol::new(db, "self".to_string());
...
}
```
Co-authored-by: Yoshitomo Nakanishi <yurayura.rounin.3@gmail.com>
435: Allow `clippy::needless_lifetimes` on tracked method getters r=XFFXFF a=DropDemBits
The tracked method generation adds an extra `__db` lifetime to the signature, but clippy complains that this lifetime can be elided in some cases, so add an allow to silence this warning.
Unfortunately clippy doesn't lint inside of the warning tests, so those tests don't do anything to check that no clippy warnings are generated.
Co-authored-by: DropDemBits <r3usrlnd@gmail.com>
436: Update trybuild tests r=XFFXFF a=DropDemBits
Rust 1.68 changed how the help suggestion text renders, and shows a diff for the suggested change.
Co-authored-by: DropDemBits <r3usrlnd@gmail.com>
428: salsa-2022: fix hanging cancellations due to cvar not being notified r=nikomatsakis a=manapointer
This PR fixes an issue with the `cvar` condvar field of `Shared<DB>` not being notified when `Arc<Shared<DB>>`s were getting dropped.
Previously, the condvar was being notified here:
```rust
impl<DB> Drop for Shared<DB>
where
DB: HasJars,
{
fn drop(&mut self) {
self.cvar.notify_all();
}
}
```
However, because this is implemented on `Shared<DB>`, the `drop` code only ran after all `Arc<Shared<DB>>`s (including that of the actual database and not just its snapshots) had dropped first, even though the intention was for the `drop` code to run when each individual `Arc<Shared<DB>>` was dropped so that the condvar would be notified each time.
To fix this, I've modified the `Shared<DB>` struct to instead hold `Arc`s to both the jars and the condvar, and `Storage` now holds `Shared<DB>` directly rather than `Arc<Shared<DB>>`. When `Shared<DB>` is dropped, it first drops its `Arc` for the jars, and then notifies the condvar.
## Note
On my local branch I have a test case for this functionality, although it relied on timing using `std:🧵:sleep`. I figured this was less than ideal, so I decided not to include it in this PR - I'd love to get feedback on a better way to test this!
Co-authored-by: manapointer <manapointer@gmail.com>
Co-authored-by: Niko Matsakis <niko@alum.mit.edu>
