add meeting notes

meetings/2021.02.09-lazy-norm.md

@@ -0,0 +1,80 @@
+# cg meeting 09.02.2020
+
+*mostly taken from https://hackmd.io/9LIywezOQj-h5HX8vvEnSw?view*
+
+Lazy norm deals with 2 different issues:
+
+- We must not typeck anonymous constants in some early queries, most notably `query type_of` and `query predicates_of`. Even if we were to use some hacks, there is a limit to how independent we can make this. One hard barrier is `query crate_variances`.
+- Delay unification of constants during typeck. An example for why this is necessary:
+    - not relevant for today, skip
+    - `D` is an anonymous constant with two generic parameters
+    - We can not unify `D<$0, $2>` with `D<$1, $2>`
+        - If `$0 = $1` we know that they are equal
+        - Analogous to associated types
+            - `<$0 as Trait<$2>>::X = <$1 as Trait<$2>>::X`
+    - @lcnr: still not sure if this is actually necessary.
+
+---
+
+When using the `const_generics` feature gate, anonymous constants currently have all the generic parameters of their parent, even if they are unused inside of the constant.
+
+This breaks some programs, some of which already compile on stable.
+
+```rust
+// unused-subst-4
+fn bind<const N: usize>(value: [u8; N]) -> [u8; 3 + 4] {
+    todo!()
+}
+
+fn main() {
+    let mut arr: ?0 = Default::default();
+    arr = bind(arr); //~ ERROR mismatched type
+    // ?0 = [u8; ?c2]
+    // [u8; D<[u8; ?c2]>] <: [u8; ?c2]
+    // D<[u8; ?c2]> = ?c2 
+    // --> occurs check violation
+}
+```
+
+Two solutions:
+- eagerly filter the supplied generic params to the anonymous constant
+    * if we remove parameters that are not explicitly used, have to filter out the where-clauses
+    * generating the `ty::Predicate` from the HIR (i.e., `predicates_of` query)
+    * annoying thing about `T::Item` which is expanded to `<T as Trait<U>>::Item`
+        * but that implies a `T: Trait<U>`, which would bring in `U`
+    * will pull in too many parameters, still prevent some programs from compiling (e.g., `fn foo<T, U>() -> [u8; size_of::<T>()] where T: PartialEq<U>` will pull in `U` but it's not really needed)
+    * back-compat when later improving the filtering? probably fine
+    * additional weirdness`
+```rust
+pub trait Trait<T> {
+    const ASSOC_CONST: usize = 0;
+}
+
+impl Trait<()> for u8 {}
+
+pub fn foo<T>() -> [T; u8::ASSOC_CONST]
+where
+    u8: Trait<T>,
+    // pulling in this predicate changes the way we resolve
+    // `u8::ASSOC_CONST` here.
+    //
+    // Do we even want this, optionally change the way we
+    // resolve associated constants in anonymous constants?
+{
+    todo!()
+}
+```
+- start out with still using all generic parameters. Once we can `typeck` the anonymous constant, filter everything which is unused in the mir of the constant.
+    - captures exactly the parameters which are needed
+    - less self-contained than the first option
+    - will we encounter cycles where type checking the anonymous constants needs to look at the anon const substs of itself? I think this is unlikely, but would prevent us from using this option. How likely is this?
+        - requires further testing/exploration
+    - In general, how do we deal with predicates on anonymous constants? Can we even use anonymous constants in ways which do not also require the predicates of the parent item?
+        - Predicates required for well-formedness which get removed with this approach, not yet an issue, but maybe in the future. (https://rust-lang.zulipchat.com/#narrow/stream/260443-project-const-generics/topic/meeting.2009.2E02.2E2021/near/225715123)
+# end of meeting
+>    - still have to modify `predicates_of` for anonymous constants, can't just use the parent ones, https://rust-lang.zulipchat.com/#narrow/stream/260443-project-const-generics/topic/lazy.20norm.20future.20compat/near/217663295
+>
+># Optional stuff :sparkles:
+>
+>- https://rust-lang.zulipchat.com/#narrow/stream/260443-project-const-generics/topic/lazy.20norm.20recursive.20impls
+>    - eagerly evaluate fully concrete anonymous constants? probably