# 2025-03-19

## rust-analyzer check in

* "so far things seem to be working fine"
* nikomatsakis is terrified
* Lukas:
* tried getting rid of the multi-argument query interning thing
* BUT https://github.com/salsa-rs/salsa/pull/748
* we now have multiple keys
* takeaway:
* actually Niko's scheme is very invasive and perhaps not a good idea
* problem:
* when you invoke `foo.bar(...)` we need to add a dependency into the current task
* currently that dependency is (trackedfningredient, id-of-some-salsa-struct)
* where the salsa struct is either `foo` if `...` is empty set or an interned value
* what would we use instead? dependencies become more complex to express
* alternative:
* don't be so naive about interning
* look into actual patterns to optimize
* maybe drop eagerly when `foo` (or other tracked structs that appear in the arguments...) is deleted
* `impl Foo { fn bar(&self, other: Baz<'db>) }`
* where `Foo` and `Baz` are both salsa structs
* deleting either one should delete the interned value
* (and hence the memos associated with it)
* this actually applies to any interned value
* Fixed point... landed!
* holy potatoes!
* encountered a hang issue in certain files
* some caching is missing, Carl is working on a PR
* Carl is terrified (Niko assumes)
* wasm playground:
* overall working
* but for tracked structs we reach an unreachable branch
* might be related to the aliasing issue that Lukas identified last time

#[salsa::tracked(serialize)]
struct Foo<'db> {
    ...
}

* GC questions
* How to decide the "limit" for reusing things
* conclusion:
* no right answer, we'll need to do dynamic tuning
* possible approaches
* when interning exceeds its capacity, bump it for the future
* we would have to be able to drop it down later
* running a GC
* would be useful to occasionally "stop the world" and do a cleanup of things that haven't been used recently
* prob easier not to do it "concurrently" with other computation but to require `&mut self`
* but might want to be able to cancel it
* https://hackmd.io/nNmrVvHfRYO_EpwmsQyYTQ#Dumping-fn-contents
* Contributing to salsa book
* why no direct pointer?
* because ids allow us to
* do speculative execution
* lazy deserialization
* and a bunch of other fancy stuff
* "intra-trait cache solving discussion"
* "scratch space" for queries
* https://rust-lang.zulipchat.com/#narrow/channel/185405-t-compiler.2Frust-analyzer/topic/Chalk.20support.20for.20trait.20solver.20cache/near/506567936
* `plumbing::AsId`
* we should add a method here
* and mark `plumbing` as `doc(hidden)`
* lobby IDEs to RustRover
* niko learns about r-a
* `rust-analyzer.completion.privateEditable.enable`
* workspace crates? current crate? something.
* Niko queue:
* serialization