ILASM/ILDASM support for covariant returns

[MichalStrehovsky]

The shortcut .override Class::Method ILASM syntax is generating unresolvable MemberRefs for covariant returns (only the more wordy .override method instance ReturnType Class::Method(Param1, Param2) syntax works).

ILDASM generates the shortcut syntax, things don't roundtrip. I found this when I was playing with this in the linker yesterday.

To repro, e.g. take the CompatibleWith.il test, ILASM it, then ILDASM it and ILASM it again.

After the first ILDASM:

.method /*06000009*/ public hidebysig newslot virtual 
        instance class [System.Runtime/*23000003*/]System.Collections.Generic.IList`1/*01000002*/<int32> 
        M1() cil managed
{
  .override C1/*02000004*/::M1 /*02000004::06000002*/ 
  // Code size       1 (0x1)
  .maxstack  8
  IL_0000:  ret
} // end of method C2::M1

After the second ILDASM:

.method /*06000009*/ public hidebysig newslot virtual 
        instance class [System.Runtime/*23000003*/]System.Collections.Generic.IList`1/*01000002*/<int32> 
        M1() cil managed
{
  .override C1/*01000006*/::M1 /*01000006::0A00000D*/ 
  // Code size       1 (0x1)
  .maxstack  8
  IL_0000:  ret
} // end of method C2::M1

Notice that the .override is actually using MemberRefs in the second disassembly and I don't think the MemberRefs actually resolve (didn't try with Fadi's/Jan's branch, but I don't think it will work).

The MemberRef token is defined as this:

TypeRef #6 (01000006)
-------------------------------------------------------
Token:             0x01000006
ResolutionScope:   0x00000001
TypeRefName:       C1
	MemberRef #1 (0a00000d)
	-------------------------------------------------------
		Member: (0a00000d) M1: 
		CallCnvntn: [DEFAULT]
		hasThis 
		ReturnType: GenericInst Class System.Collections.Generic.IList`1< I4>
		No arguments.

Notice the return type.

It's also weird there's no warning about this from ILASM, but that's a preexisting issue I guess.

[MichalStrehovsky]

Cc @janvorli

[BruceForstall]

@MichalStrehovsky Someone working on the covariant return feature needs to own this implementation (and take ownership of this bug). Who would that be?

[MichalStrehovsky]

@janvorli who is already Cc'd above