Newtype deriving

text/0000-newtype-deriving.md

@@ -0,0 +1,79 @@
+- Start Date: 2014-12-09
+- RFC PR: (leave this empty)
+- Rust Issue: (leave this empty)
+
+# Summary
+
+Introduce a `#[newtype_deriving(...)]` syntax extension that allows a newtype
+to derive traits of the wrapped type.
+
+# Motivation
+
+Newtypes are very useful to avoid mixing similar types with different
+semantics. The typical example is a `struct Centimeters(f64)` and a
+`struct Inches(f64)`.
+
+A big problem with the usage of newtypes is that you need to manually implement
+the traits of the wrapped type in order to use them. In the previous example,
+you cannot add two `Centimeters` unless you implement the `Add` trait for the
+struct (which is trivial, but verbose). This causes a lot of unnecessary
+boilerplate which discourages using newtypes and makes it very painful when you
+have no alternative.
+
+# Detailed design
+
+Introduce a new syntax extension similar to `#[deriving(...)]` but for
+newtypes. A possible name is `#[newtype_deriving(...)]`.
+
+Example:
+
+```rust
+#[newtype_deriving(Add, Sub, Mul, Div)]
+struct Centimeters(f64);
+
+fn do_something(cm: Centimeters) -> Centimeters {
+    cm + cm * cm / cm - cm
+}
+```
+
+Deriving a trait is trivial once you know the required functions and their
+signatures. A first approach would be to hardcode this information like it
+is done in the `#[deriving(...)]` syntax extension. However, if the first
+unresolved question is solved (see below), we could derive any trait including
+user-defined ones.
+
+In case the first approach is chosen, the traits available to be derived would be:
+* All traits in `std::ops`
+* Show
+
+Note that the comparison operators, serialization, Clone, Hash, Rand, Default
+and FromPrimitive don't need to be included since they can be derived using
+`#[deriving(...)]`.
+
+# Drawbacks
+
+It adds complexity to the language.
+
+If a trait is modified, this syntax extension would need to be updated as well.
+In case we can implement *arbitrary trait deriving* this would not be an issue,
+since the syntax extension could look at the trait definition (see unresolved
+questions).
+
+# Alternatives
+
+We could extend the current `#[deriving(...)]` syntax extension to handle the
+case of newtypes. This could cause ambiguity (e.g. `#[deriving(Show)`]
+could produce a `"Centimeter(5.0)"` as output or just `"5.0"`).
+
+Do nothing and let future IDEs generate the boilerplate for us.
+
+# Unresolved questions
+
+1. According to [this comment]
+(https://github.com/rust-lang/rust/issues/19597#issuecomment-65909118),
+a syntax extension has no access to the function signatures of a trait. If
+this changes, we could derive any trait (also user-defined ones) for a newtype.
+Is it possible to allow syntax extensions to see the (required) function
+signatures of a trait? How difficult would it be?
+
+2. In case 1 cannot be solved, which traits should be available to be derived?