top-level functions pick up a `tempCtor` superclass

[angular]

top-level functions pick up a `tempCtor` superclass

[alexeagle]

This seems like a bug in closure compiler. Looking into the type at runtime, it claims to have kind==CONSTRUCTOR when the @type is present. If it's removed, it goes back to kind==ORDINARY.

[alexeagle]

Ben says
"/** @type {!Function} */ is new-able as well as callable, so in some
sense it's both a constructor and an ordinary function."

[rkirov]

I didn't know that in closure all functions are newable. TS plain functions are not new-able, so we have to make it explicitly new-able with

interface ClosureFunction {
  (...args: any[]):any;
  new (...args: any[]): any;
}

var abstractMethod: ClosureFunction; 

[mprobst]

In Closure, function() is a different type from Function. The first is a specific function that can only be called in a specific way, with particular arguments, e.g. function() being a function that takes no args and is void, function(string): number being a function taking a single string, returning a number. A function that's constructable but restricts it's arguments looks like this: function(new:Foo, number, string). Similarly, you can also specify function(this:SomeType).

Function is a weak type assertion that something is any subtype of Function, so could or could not be callable with new and/or any args. JSCompiler allows calling these in any way.

[alexeagle]

The problem seems to be that the information was lost. Introspecting the type system tells us only that the thing is modelled as a class, but not whether it was declared with @constructor.
If we make all these things a class, they will appear to be newable but not callable.
Maybe we need to emit all of them as callable-newable like @rkirov wrote above.

[mprobst]

Can we get to the JSDocInfo field? That has the @constructor thing in it, I'd hope.

Otherwise we could also just special case on tempCtor and fail if somebody really calls their super class tempCtor (which violates the style guide and common sense ;-)).

[rkirov]

I am changing my original proposal, since this is a rare pattern and honestly new-able + callable sounds like trouble. I suggest ignoring the new-able semantics and just outputting it as var foo: Function; for now.

We have only seen 2 places where this is used (goog.reflect.sinkValue and goog.abstractMethod) and neither is very useful when called with new.

[mprobst]

Yeah, this seems like a corner case, and as you can specify explicitly when something should be newable, the intent would probably never be to actually call new on one of these.