cannot auto parse right type in these code

[ktangels]

TypeScript Version: 2.1.4

Code

function mergeObject<T extends any>(dst: T, src: T): T
{
 for (var x in src)
 {
 dst[x] = src[x];
 }
 return dst;
}
 
interface Hash<T> { [key: string]: T; }

type EventHandler = (s: BaseEventDealer) => void;
interface BaseEventDealer
{
 GetEventHandler(event: string): EventHandler;
}

class C implements BaseEventDealer
{
 GetEventHandler(event: string) { return C.h[event]; }

 static h : Hash<EventHandler> = {
 a:function(s){}
 }
}

class D extends C
{
 GetEventHandler(event: string) { return D.h[event]; }
 static h = mergeObject<Hash<EventHandler>>(
 {
 b:function(s:D){}
 },C.h);
}

Expected behavior:

type of D.h should be Hash<EventHandler>
type of return value of D.GetEventHandler should be EventHandler

Actual behavior:

type of D.h and return value of D.GetEventHandler both is ImplicitAny
but if you remove D.GetEventHandler or change its name, h will get right type
or you manually declare h:Hash<EventHandler> or GetEventHandler(event: string):EventHandler, they will both get right type

[sandersn]

I tried to fix this with #6118 but it breaks existing code. I gave the reason in this comment.

[sandersn]

Duplicate of #3667, except for statics.

[ktangels]

Not Duplicate of #3667
I don't think we are talking about a same thing...
I mean, if without the code GetEventHandler(event: string) { return D.h[event]; }
type of D.h is already confirmed to right type and D.h[e] is right type
so why add a function returning it will change this fact and let it lose its type?

[sandersn]

Sorry, I missed the second line of the "actual" description where you talked about removing GetEventHandler.

[mhegazy]

This is actually the same issue caused by #10978.

The compiler is building a new type D and it needs to to figure out if D correctly extends C, so

• compare all members of D to C
  • compare GetEventHandler in both D and C
    • compare return type of D.GetEventHandler to C.GetEventHandler
      • Return type of D.GetEventHandler is not defined, infer it from function declaration
        • so what is the type of D.h
          • that is not defined, infer it from initialization
            • so what is the type of that mergeObject call?
              • do overload resolution, and check the arguments
                • compare Hash<EventHandler> to { b: (s:D)=> void }
                  • so what are the properties of D?
                  • There is one D.GetEventHandler, but what is its type?
                  • well, we were just trying to figure this out, since we do not know, call it any.
                  • and tell the user we just did that, i.e. no implicit any error

[ktangels]

well let's consider what will happen when we rename D.GetEventHandlers to D.a like
class D extends C { a(event: string) { return D.h[event]; } static h = mergeObject<Hash<EventHandler>>( { b:function(s:D){} },C.h); }

now we want to know type of D.a
so what's the type of D.a?
so what is the type of D.h
that is not defined, infer it from initialization
so what is the type of that mergeObject call?
do overload resolution, and check the arguments
compare Hash to { b: (s:D)=> void }
so what are the properties of D?
There is one D.a, but what is its type?

it seems like it will meet the same problem, but IT"S NOT

why?

i guess, when we try to compare Hash to { b: (s:D)=> void }
what's type of D? extends BaseEventDealer, is it extends correctly? depends on type of D.GetEventHandler, so we check D.GetEventHandler and boom

but, consider this:

what's type of D? extends BaseEventDealer, is it extends correctly? depends on type of D.GetEventHandler, so we check D.GetEventHandler
we don't know, so we call it any
so is it extends correctly when D.GetEventHandler is any?
yes
so compare Hash to { b: (s:D)=> void }
yes
and so on, finally we got type of D.h
and finally we got type of D.GetEventHandler

I mean, here we got an error of recursive type parsing, but after we temporary consider it as any, everything will go just fine.Will it be a solution?

[ktangels]

consider the same problem in #10978.

`class M {
constructor(fn: () => any) { }
}

const a = new M(() => a); /`

as RyanCavanaugh commented:
Let's figure out the type of a from its initializer
What is the type of new M ?
Resolve M to a constructor function
Which overload, if any, of the function is applicable?
There's one overload, is it applicable?
There's a function here, its argument list OK, is its return type OK?
That depends, is the type of a an OK return type for the overload under consideration?
PANIC, we are trying to see the type of a while resolving a's type!

then we don't know type of a, then we temporary treat is as any,
so when its type is any, is its return type OK?
yes
and finally can we get type of the temporary any?
yes
and resolved

[mhegazy]

I mean, here we got an error of recursive type parsing, but after we temporary consider it as any, everything will go just fine.Will it be a solution?

things can have only one type. having a type any some time and some other types a type C would lead to surprises and phantom errors.

[ktangels]

maybe I should not call it any, we can call it unsolved(which treat as any)
then we cached all unsolved symbols and parse it again after all other symbols, if it still unsolved, then make it implicit any

[mhegazy]

It does not matter what you call it. What matters is the same expression has different types at different places.