[FS-1031] Allow implementing the same interface at different generic instantiations in the same type

[0x53A]

Suggestion
Details: under discussion
RFC: merged

This now compiles:

type IA<'T> =
    abstract member Get : unit -> obj

type MyClass() =
    interface IA<int> with
        member x.Get() = upcast 1
    interface IA<string> with
        member x.Get() = upcast 2

This also compiles:

type IB<'T> =
    abstract member Get : unit -> 'T

type IC =
    abstract member Get : unit -> int

[<EntryPoint>]
let main argv =
    let x = { new IB<System.Int32> with member x.Get() = 1
              interface IB<string> with member x.Get() = "hello"
              interface IB<float> with member x.Get() = 1.
              interface IC with member x.Get() = 1 }
    0 // return an integer exit code

This does NOT compile, and is an explicit non-goal in the RFC:

type IGet<'T> =
    abstract member Get : unit -> 'T

[<EntryPoint>]
let main argv =
    let x = { new IGet<_> with member x.Get() = 1
              interface IGet<_> with member x.Get() = "hello"
              interface IGet<_> with member x.Get() = 1. }

This is now pretty near finished.

• Build is green

• one unit test is partially commented out, open question whether it should be supported. resolved: not supported

• Implementation needs a little bit of tidying up

• I want to change the error message, here is the one from C#:
  

• REVIEW: can this exception ever be hit? https://github.com/Microsoft/visualfsharp/pull/2867/files/01ae6c0526354e38fcc958d1d187761733a58b1e#diff-5b9ab9dd9d7133aaf23add1048742031R1085

[auduchinok]

Thanks @0x53A! It would be great to see this feature merged.

[0x53A]

Me too ;D

There is a little bit of clean-up missing, I mainly paused because at that time there seemed to be a feature-freeze.

@cartermp when is the cut-off date for F#4.2? Or to be more precise, what is the earliest date this could be merged? I didn't want to have to rebase it for months, that's why I stopped.

[dsyme]

@cartermp when is the cut-off date for F#4.2? Or to be more precise, what is the earliest date this could be merged?

There's no specific cut off as yet.

I didn't want to have to rebase it for months, that's why I stopped.

I'm afraid that's just what needs to be done to keep language additions in the pipe. That said, you have almost no conflicts so a merge+update should be easy enough

[0x53A]

I'm afraid that's just what needs to be done to keep language additions in the pipe.

In general I agree, but if it is clear that it can't be merged, then it doesn't really make sense.

Anyway, regarding this specific PR: I will plan some time this weekend to get it to a reviewable state.

[cartermp]

I agree with @dsyme and I don't have any cutoff in mind, either. At some point, as we keep collecting bug fixes in F# 4.1, we may want to rev, but I don't feel that way right now. Certainly not without a few features like this.

[realvictorprm]

@0x53A should I update this for you? I'm used to this stuff and so it's not a big effort for me 😄

[realvictorprm]

understood @0x53A. You can always ping me and ask me for keeping this up-to-date.

[0x53A]

thanks @realvictorprm for the offer, it was not that the effort of doing it once was so much, more that the effort of doing it every other day for a few months was something I didn't bother to do ;-)

[realvictorprm]

I try to reduce this time with adding features @0x53A 😛 😄

[0x53A]

I looked around for any suspiciously named methods (isObjectExpression and so on), but couldn't find anything. Help appreciated.

[0x53A]

Afaik this code path is only hit with classes, not OEs, but not sure if that is guaranteed

[0x53A]

I would appreciate a review and additional positive as well as negative unit tests, especially combinations of measures and aliases and generics.

[abelbraaksma]

@0x53A, do we have an RFC for this? I have trouble finding it. In fsharp/fslang-design#185 you link to a possible RFC FS-1031, but this link is dead (it goes into your own repo) and I don't seem to be able to find the "official" one, if any exists. Or does the language suggestion #545 in the original post count as RFC?

I'm curious to the expected width of this submission, whether it puts extra strain on the type inference system (i.e., performance-wise), whether it influences type-inferenced signatures in quick-info tooltips or FSI feedback and whether there are certain restrictions we should be aware of.

Also of interest is this subtle bug introduced in VS2015, also related to generic type inheritance, according to Don's comment there, the bug is now a feature (i.e., no rollback will be done), meaning that code snippet should stay the same (and perhaps be made into a unit test).

As many out there, I have been bitten by this restriction in the past, so I very much hope to see it merged into master!

[0x53A]

@abelbraaksma https://github.com/fsharp/fslang-design/blob/master/RFCs/FS-1031-Allow%20implementing%20the%20same%20interface%20at%20different%20generic%20instantiations%20in%20the%20same%20type.md

The scope of this RFC and PR is as small as possible - it just fixes the naming collision, and adapts the check which previously prohibited this.

There are certainly other additions which would round this feature out (e.g. see fsharp/fslang-design#185 (comment)), but that would - if approved by don - be implemented independent of this.

[KevinRansom]

The recent simplification commits -- introduced a bug

// #Conformance #ObjectOrientedTypes #InterfacesAndImplementations 
type MyInt = int
[<Measure>] type kg

type IB<'a> =
    interface 
        abstract X : unit -> int
    end

type C() =
    interface IB<int<kg>> with
        member x.X() = 1
    interface IB<MyInt> with
        member x.X() = 2
    
exit 1

should produce this output:

MultipleInst08.fs(10,6): error FS3360: 'C' cannot implement the interface 'IB<_>' with the two instantiations 'IB<MyInt>' and 'IB<int<kg>>' because they may unify.

actually produces this output:

Microsoft (R) F# Compiler version 10.10.0.0 for F# 4.7
Copyright (c) Microsoft Corporation. All Rights Reserved.
Duplicate in method table is:
 Type index: 5
 Method name: MultipleInst08.IB<System.Int32>.X
 Method arity (num generic params): 0

error FS2014: A problem occurred writing the binary 'c:\kevinransom\fsharp\tests\fsharpqa\Source\Conformance\ObjectOrientedTypeDefinitions\InterfaceTypes\MultipleInst08.exe': Error in pass2 for type MultipleInst08, error: Error in pass2 for type C, error: duplicate entry 'MultipleInst08.IB<System.Int32>.X' in method table

[0x53A]

Thank you for taking care of this!

[abelbraaksma]

Wow, amazing stuff! 👍

[KevinRansom]

@0x53A , thank you for getting it done in the first place 👍