Reject argument name in function signature for inconsistency

[bombless]

Function signature is not even documented in the reference.
For a function signature

type T = fn(arg1: c_void, c_void);

the argument names can be ignored, and there's two drawbacks here:

• Fn/ FnOnce/ FnMut traits don't have argument names, so it leads to inconsistency
• it's harder to parse if argument names are optional, for syntax-highlighting and other purposes

And the syntax of extern functions should be changed accordingly for obvious reason.

So this will break current code, thus need to be landed before 1.0, or even before beta.

cc @nikomatsakis @AngryLawyer @mvdnes

[brendanzab]

I'm not sure I agree - the argument names are very useful for documentation purposes.

[bombless]

@bjz

useful for documentation purposes

But it's just signature. Different functions can have same signature with totally different meanings.
So maybe some time you add an argument name by means of self-documentation, then you use the second function with same signature, former argument name may become nonsense.

So the type of the argument is all it is about, an optional name here can even be harmful for documentation.

[Diggsey]

I don't agree either - a common case is that you alias the function signature for a specific purpose (for example a callback) the argument names are then incredibly useful as documentation, for both for someone implementing the callback and for someone invoking the callback.

The alias contains more information than just the signature, it also corresponds to one specific use for that signature, even if there's no way for the type system to enforce that.

[bombless]

I just realized that signatures in traits also have optional argument names, so this may not be a good idea.