Merge pull request #2056 from sgrif/sg-allow-trivial-where-clause-constraints

Allow trivial constraints to appear in where clauses

Author: aturon

text/2056-allow-trivial-where-clause-constraints.md

@@ -0,0 +1,85 @@
+- Feature Name: `allow_trivial_constraints`
+- Start Date: 2017-07-05
+- RFC PR: https://github.com/rust-lang/rfcs/pull/2056
+- Rust Issue: https://github.com/rust-lang/rust/issues/48214
+
+# Summary
+[summary]: #summary
+
+Allow constraints to appear in where clauses which are trivially known to either
+always hold or never hold. This would mean that `impl Foo for Bar where i32:
+Iterator` would become valid, and the impl would never be satisfied.
+
+# Motivation
+[motivation]: #motivation
+
+It may seem strange to ever want to include a constraint that is always known to
+hold or not hold. However, as with many of these cases, allowing this would be
+useful for macros. For example, a custom derive may want to add additional
+functionality if two derives are used together. As another more concrete
+example, Diesel allows the use of normal Rust operators to generate the
+equivalent SQL. Due to coherence rules, we can't actually provide a blanket
+impl, but we'd like to automatically implement `std::ops::Add` for columns when
+they are of a type for which `+` is a valid operator. The generated impl would
+look like:
+
+```rust
+impl<T> std::ops::Add<T> for my_column
+where
+    my_column::SqlType: diesel::types::ops::Add,
+    T: AsExpression<<my_column::SqlType as diesel::types::ops::Add>::Rhs>,
+{
+    // ...
+}
+```
+
+One would never write this impl normally since we always know the type of
+`my_column::SqlType`. However, when you consider the use case of a macro, we
+can't always easily know whether that constraint would hold or not at the time
+when we're generating code.
+
+# Detailed design
+[design]: #detailed-design
+
+Concretely implementing this means the removal of [`E0193`]. Interestingly, as of
+Rust 1.7, that error never actually appears. Instead the current behavior is
+that something like `impl Foo for Bar where i32: Copy` (e.g. anywhere that the
+constraint always holds) compiles fine, and `impl Foo for Bar where i32:
+Iterator` fails to compile by complaining that `i32` does not implement
+`Iterator`. The original error message explicitly forbidding this case does not
+seem to ever appear.
+
+The obvious complication that comes to mind when implementing this feature is
+that it would allow nonsensical projections to appear in the where clause as
+well. For example, when `i32: IntoIterator` appears in a where clause, we would
+also need to allow `i32::Item: SomeTrait` to appear in the same clause, and even
+allow `for _ in 1` to appear in item bodies, and have it all successfully
+compile.
+
+Since code that was caught by this error is usually nonsense outside of macros,
+it would be valuable for the error to continue to live on as a lint. The lint
+`trivial_constraints` would be added, matching the pre-1.7 semantics of E0193,
+and would be set to warn by default.
+
+[`E0193`]: https://doc.rust-lang.org/error-index.html#E0193
+
+# How We Teach This
+[how-we-teach-this]: #how-we-teach-this
+
+This feature does not need to be taught explicitly. Knowing the basic rules of
+where clauses, one would naturally already expect this to work.
+
+# Drawbacks
+[drawbacks]: #drawbacks
+
+- The changes to the compiler could potentially increase complexity quite a bit
+
+# Alternatives
+[alternatives]: #alternatives
+
+n/a
+
+# Unresolved questions
+[unresolved]: #unresolved-questions
+
+Should the lint error by default instead of warn?