Adding support for default interface function members #681

RexJaeschke · 2022-12-05T13:54:36Z

Re the MS proposal: The main challenge when reading it was the (misleading) occurrence of the modifier override in numerous examples and narrative. As best as I can tell, this modifier is not actually permitted on any interface member. Instead, overriding is achieved via explicit implementation.

There is one open question, which has to do with the topic "Base Interface Invocations," for which the decision "Decided on base(N.I1<T>).M(s)" was made. However, as far as I can tell, support for this was not added to the V8 compiler. Can someone please confirm (or refute) that.

BTW, I think the proposal is misnamed: it really applies to all interface function members (properties, indexers, and events as well), not just methods. (it also allows nested types.)

Note: No spec changes are needed to support entry point Main in an interface. Any non-generic type containing a Main method with void/int return and no/one string[] parameter list is already permitted, which not only allows Main in a struct or class, but now also in an interface.

As of V7, the descriptions of the various member kinds has been located in classes.md, with augmenting and/or overriding text in structs.md, with pointers from structs.md back into classes.md, and that continues. However, now that many members kinds can have implementations in interfaces as well, I've added an intro para to the start of most member sections in classes.md, which contains forward pointers to that member kind's occurrence in structs.md and interfaces.md, as appropriate.

RexJaeschke · 2022-12-06T17:15:06Z

Depending how PR #680 is resolved, some tweaks might be needed with this PR.

As at 2025-05-03, that PR is still pending, and is tied to the review of this PR.

BillWagner · 2023-09-25T20:54:24Z

rebased on the latest draft-v8 on 09-23-2023

KalleOlaviNiemitalo · 2023-09-25T21:23:24Z

Should the references to "C# implementations targeting the CLI" be changed somehow because the ECMA standard CLI does not support default interface members? Possible answers include:

Remove them; C# 8 implementations cannot target "the CLI".
Change them to say "targeting a CLI with extensions" or the like.
Keep them; a C# implementation can still target the standard CLI, although the mapping may become complex or inefficient or have restrictions (like not supporting dynamic loading of assemblies that were not provided at build time).

RexJaeschke · 2024-11-12T21:52:19Z

Re @KalleOlaviNiemitalo's concern about the Ecma CLI standard not supporting default interface function members, we previously made a related change to the draft-v8 Introduction, which in V7, said:

... Although Microsoft’s implementation of C# relies on CLI for library and run-time support, other implementations of C# need not, provided they support an alternate way of getting at the minimum CLI features required by this C# standard (see Annex C).

It now says:

... Although Microsoft’s implementation of C# relies on CLI for library and run-time support, other implementations of C# need not, provided they support the features and API required by this C# Standard (see Annex C).

I think that allows us to say nothing more for this feature with regard to the CLI.

gafter · 2024-12-11T00:33:25Z

standard/classes.md

+
+When an interface `IB` extends an interface `IA`, it is a compile-time error for `IA` to depend on `IB`. An interface **directly depends on** its direct base interfaces (if any) and **directly depends on** the type within which it is immediately nested (if any).
+
+Given these definitions, the complete set of types upon which a type depends is the transitive closure of the *directly depends on* relationship.


Suggested change

Given these definitions, the complete set of types upon which a type depends is the transitive closure of the *directly depends on* relationship.

Given these definitions, the complete set of types upon which a type depends is the reflexive and transitive closure of the *directly depends on* relationship.

The suggested change is required to disallow class A : A {}

It been a long day, but does not adding reflexive require “A depends on A” not disallow it?

I'll make the alternative suggestion to delete line 277. I don't think it's needed.

gafter · 2024-12-11T00:43:36Z

standard/interfaces.md

+>
+> *end example*
+
+Every interface and class shall have a most specific override for every virtual member among the overrides appearing in the type or its direct and indirect interfaces. The ***most specific override*** is a unique override that is more specific than every other override. If there is no override, the member itself is considered the most specific override.


What is "an override"?

I would use the term from the feature spec "most specific implementation". The feature spec defines it as

The most specific implementation is a unique implementation that is more specific than every other implementation. If there is no implementation, the member itself is considered the most specific implementation.

gafter · 2024-12-15T01:14:52Z

I don't recall why this is assigned to me. Is it for general review, or to answer the question about base(T).M()? If the latter, see the following:

In short, the base(T).M() syntax has not been added to the language yet.

BillWagner · 2024-12-15T16:13:38Z

@gafter

I don't recall why this is assigned to me. Is it for general review, or to answer the question about base(T).M()?

Going from memory, but I believe as a general reviewer.

RexJaeschke · 2025-05-28T15:16:23Z

closed and reopened to get all tests to run

jskeet · 2025-06-11T13:09:31Z

@RexJaeschke and @BillWagner could you look at resolving the conflict?

I'm not sure whether we can realistically expect most members to have reviewed this before the meeting later today though. (I'll see whether I can find time, but I'm prioritizing #606.)

BillWagner · 2025-06-11T14:07:31Z

@RexJaeschke and @BillWagner could you look at resolving the conflict?

I'm not sure whether we can realistically expect most members to have reviewed this before the meeting later today though. (I'll see whether I can find time, but I'm prioritizing #606.)

@jskeet I did a full rebase.

My main goal in prioritizing this for today's meeting is to get it assigned to someone for next meeting. I agree that there isn't time to get it reviewed thoroughly.

jskeet · 2025-06-25T08:29:15Z

@BillWagner Looking again, did we get as far as assigning it during the last meeting? We spent quite a bit of time talking about prose vs grammar for constraints, but we didn't record the next step...

BillWagner · 2025-06-25T12:18:30Z

@jskeet I have on my list to rebase this PR, and do an edit pass with notes so we have a basis for assigning at this coming meeting.

Nigel-Ecma · 2025-06-25T21:33:35Z

@jskeet & @BillWagner – I have a note to myself to “compare grammar vs. semantic by Jul 3”, no idea whether I will make that…

Enclosing a paragraph in brackets breaks the word converter. It looks for a link.

BillWagner

We know that there's more work coming here. For this next meeting, I'd like to see the committee agree on some key points regarding the direction for this PR:

Terms used. Neal had some earlier comments, I added more on what terms to use for the new concepts in this feature.
The overall strategy of defining members of class, struct and interface declarations. Right now, we mostly define everything in terms of class, then deltas for interfaces and structs. But, sometimes those deltas are in the class clause, sometimes in interfaces or structs. Over the next several releases, the differences continue to shrink (static abstract members in interfaces, struct default constructors and field initializers, and so on.) I see two potential directions:

Continue to define interfaces and structs in terms of a delta from class types. If we follow that, I'd recommend moving all the remaining text for deltas out of the class clause into structs or interfaces (including some added in this PR)
Create a new "Members" clause the defines the common features of members, and then classes, interfaces, and structs can reference that with deltas.

Personally, I think the former is easier because "Members" are hard to define without one of class, struct, interface. And all of class, struct, and interface are incomplete without fully specifying all the members.

/cc @jskeet @RexJaeschke for meeting agenda notes.

BillWagner · 2025-06-27T17:12:46Z

standard/classes.md

+
+When an interface `IB` extends an interface `IA`, it is a compile-time error for `IA` to depend on `IB`. An interface **directly depends on** its direct base interfaces (if any) and **directly depends on** the type within which it is immediately nested (if any).
+
+Given these definitions, the complete set of types upon which a type depends is the transitive closure of the *directly depends on* relationship.


I'll make the alternative suggestion to delete line 277. I don't think it's needed.

BillWagner · 2025-06-27T17:28:45Z

standard/interfaces.md

@@ -4,7 +4,9 @@

 An interface defines a contract. A class or struct that implements an interface shall adhere to its contract. An interface may inherit from multiple base interfaces, and a class or struct may implement multiple interfaces.

-Interfaces can contain methods, properties, events, and indexers. The interface itself does not provide implementations for the members that it declares. The interface merely specifies the members that shall be supplied by classes or structs that implement the interface.
+Interfaces may contain various kinds of members, as described in [§18.4](interfaces.md#184-interface-members). The interface itself may provide ***default implementation***s for some or all of the function members that it declares, in which case, those members are *not* part of the interface contract. For those members for which the interface does not provide default implementations, the interface merely specifies the members that shall be supplied by classes or structs that implement the interface.


I think we should remove this phrase. Any interface members that have a default implementation are part of the interface contract, but they have a default implementation:

Suggested change

Interfaces may contain various kinds of members, as described in [§18.4](interfaces.md#184-interface-members). The interface itself may provide ***default implementation***s for some or all of the function members that it declares, in which case, those members are *not* part of the interface contract. For those members for which the interface does not provide default implementations, the interface merely specifies the members that shall be supplied by classes or structs that implement the interface.

Interfaces may contain various kinds of members, as described in [§18.4](interfaces.md#184-interface-members). The interface itself may provide ***default implementation***s for some or all of the function members that it declares. For those members for which the interface does not provide default implementations, the interface merely specifies the members that shall be supplied by classes or structs that implement the interface.

BillWagner · 2025-06-27T17:31:37Z

standard/interfaces.md

-    | interface_event_declaration
-    | interface_indexer_declaration
+    : constant_declaration
+    | field_declaration


This is incorrect. Interface can't include non-static fields. We'll need to rework the grammar in classes.md to provide a distinction between instance and static data fields.

BillWagner · 2025-06-27T17:33:01Z

standard/interfaces.md

+    | field_declaration
+    | method_declaration
+    | property_declaration
+    | event_declaration


field-like events can't be declared in an interface. This will also require more splitting of the grammar rules for events to distinguish the two syntaxes for them.

BillWagner · 2025-06-27T17:35:00Z

standard/interfaces.md

    ;
 ```

-An interface declaration declares zero or more members. The members of an interface shall be methods, properties, events, or indexers. An interface cannot contain constants, fields, operators, instance constructors, finalizers, or types, nor can an interface contain static members of any kind.
+An interface declaration declares zero or more members. The members of an interface shall be constants, fields, methods, properties, events, indexers, operators, constructors, and types, some of which may be instance, others static, as described in the subclauses for each interface member kind.


This needs to clarify that interfaces can't include instance data fields (including field-like events). It might be the best place to mention that interfaces use the same syntax for auto-implemented properties for an abstract property.

BillWagner · 2025-06-27T17:51:37Z

standard/interfaces.md

@@ -242,24 +255,133 @@ If a `new` modifier is included in a declaration that doesn’t hide an inherite

 The set of members of an interface declared in multiple parts ([§15.2.7](classes.md#1527-partial-type-declarations)) is the union of the members declared in each part. The bodies of all parts of the interface declaration share the same declaration space ([§7.3](basic-concepts.md#73-declarations)), and the scope of each member ([§7.7](basic-concepts.md#77-scopes)) extends to the bodies of all the parts.

+Consider an interface with a default implementation for a member `M`. As `M` is not part of that interface’s contract, outside that interface or any interface derived from it, that name is not visible. How then can it be accessed? The following code shows how:


Two thoughts:

Lines 258 - 294 should probably be non-normative. The normative language should be that an interface member must be accessed as an explicit interface member.

I don't like the phrasing about "not part of the contract". Instead, I'd suggest we phrase this in terms of "visibility" or even "inheritance". An interface member with a default implementation must be accessed through a reference of that interface type. It isn't visible as a member of an implementing class, or derived interface.

BillWagner · 2025-06-27T17:54:18Z

standard/interfaces.md

+- A static declaration that is not extern or abstract shall have a *block* as a *method_body*.
+- A virtual declaration that is not extern shall have a *block* as a *method_body*.
+- A private declaration that is not extern shall have a *block* as a *method_body*.
+- A sealed declaration that is not extern shall have a *block* as a *method_body*.
+- An async declaration shall have a *block* as a *method_body*.


This needs some wordsmithing. I don't think we want "block" here. Expression bodied members are legal.

BillWagner · 2025-06-27T17:56:30Z

standard/interfaces.md

+
+*method_modifier* shall not include `override`.
+
+An interface method declaration that has a *block* as a *method_body* is a default implementation ([§18.1](interfaces.md#181-general)), so it is *not* part of the interface’s contract. A *method_declaration* shall not have *type_parameter_constraints_clause*s unless it also has a *type_parameter_list*.


Two bits here:

Again, my concern with "part of the contract" as the terminology. And, an expression bodied member is legal, so block is the wrong term.

BillWagner · 2025-06-27T18:54:45Z

standard/interfaces.md

+>
+> *end example*
+
+Every interface and class shall have a most specific override for every virtual member among the overrides appearing in the type or its direct and indirect interfaces. The ***most specific override*** is a unique override that is more specific than every other override. If there is no override, the member itself is considered the most specific override.


I would use the term from the feature spec "most specific implementation". The feature spec defines it as

The most specific implementation is a unique implementation that is more specific than every other implementation. If there is no implementation, the member itself is considered the most specific implementation.

RexJaeschke added this to the C# 8.0 milestone Dec 5, 2022

RexJaeschke self-assigned this Dec 5, 2022

RexJaeschke marked this pull request as draft December 5, 2022 13:54

BillWagner force-pushed the default-interface-function-members branch from 84a1703 to caba5c7 Compare February 6, 2023 13:52

RexJaeschke added the type: feature This issue describes a new feature label Jul 22, 2023

BillWagner force-pushed the default-interface-function-members branch from caba5c7 to dc8d7a0 Compare September 25, 2023 20:54

BillWagner mentioned this pull request Sep 29, 2023

Consider reordering and reorganizing sections #959

Open

RexJaeschke mentioned this pull request Oct 2, 2023

New Language Features that Rely on Features NOT in the Current CLI Spec #962

Open

RexJaeschke added the Review: pending Proposal is available for review label Oct 13, 2023

RexJaeschke removed their assignment Oct 22, 2023

RexJaeschke mentioned this pull request Nov 1, 2023

Add support for init accessors #978

Draft

RexJaeschke marked this pull request as ready for review January 11, 2024 14:00

gafter self-assigned this Jan 16, 2024

gafter reviewed Dec 11, 2024

View reviewed changes

RexJaeschke unassigned gafter May 14, 2025

RexJaeschke closed this May 28, 2025

RexJaeschke reopened this May 28, 2025

RexJaeschke added the meeting: priority Review before meeting. Merge, merge with issues, or reject at the next TC49-TC2 meeting label Jun 10, 2025

BillWagner force-pushed the default-interface-function-members branch from 22549fb to efb69e5 Compare June 11, 2025 13:57

RexJaeschke and others added 14 commits June 27, 2025 11:48

Allow interfaces as well

a8d0d39

support default interface function members

f6ba642

support default interface function members

39d4516

Update classes.md

488df8f

Update interfaces.md

306a34a

Update interfaces.md

db30673

add annotation to new examples

e0787ec

first pass at fixing references

bf81857

find link errors part 2

f0b349d

word converter warnings

6907966

Enclosing a paragraph in brackets breaks the word converter. It looks for a link.

minor tweaks

583b1c2

Fix editorial nits

f1033e9

fix link

d0ffad8

fix link

4418683

BillWagner force-pushed the default-interface-function-members branch from efb69e5 to 4418683 Compare June 27, 2025 15:48

BillWagner reviewed Jun 27, 2025

View reviewed changes


		When an interface `IB` extends an interface `IA`, it is a compile-time error for `IA` to depend on `IB`. An interface directly depends on its direct base interfaces (if any) and directly depends on the type within which it is immediately nested (if any).

		Given these definitions, the complete set of types upon which a type depends is the transitive closure of the directly depends on relationship.

		@@ -242,24 +255,133 @@ If a `new` modifier is included in a declaration that doesn’t hide an inherite

		The set of members of an interface declared in multiple parts ([§15.2.7](classes.md#1527-partial-type-declarations)) is the union of the members declared in each part. The bodies of all parts of the interface declaration share the same declaration space ([§7.3](basic-concepts.md#73-declarations)), and the scope of each member ([§7.7](basic-concepts.md#77-scopes)) extends to the bodies of all the parts.

		Consider an interface with a default implementation for a member `M`. As `M` is not part of that interface’s contract, outside that interface or any interface derived from it, that name is not visible. How then can it be accessed? The following code shows how:


		method_modifier shall not include `override`.

		An interface method declaration that has a block as a method_body is a default implementation ([§18.1](interfaces.md#181-general)), so it is not part of the interface’s contract. A method_declaration shall not have type_parameter_constraints_clauses unless it also has a type_parameter_list.

Adding support for default interface function members #681

Are you sure you want to change the base?

Adding support for default interface function members #681

Uh oh!

Conversation

RexJaeschke commented Dec 5, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

RexJaeschke commented Dec 6, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

BillWagner commented Sep 25, 2023

Uh oh!

KalleOlaviNiemitalo commented Sep 25, 2023

Uh oh!

RexJaeschke commented Nov 12, 2024

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gafter commented Dec 15, 2024

Uh oh!

BillWagner commented Dec 15, 2024

Uh oh!

RexJaeschke commented May 28, 2025

Uh oh!

jskeet commented Jun 11, 2025

Uh oh!

BillWagner commented Jun 11, 2025

Uh oh!

jskeet commented Jun 25, 2025

Uh oh!

BillWagner commented Jun 25, 2025

Uh oh!

Nigel-Ecma commented Jun 25, 2025

Uh oh!

BillWagner left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

RexJaeschke commented Dec 5, 2022 •

edited

Loading

RexJaeschke commented Dec 6, 2022 •

edited

Loading

BillWagner left a comment •

edited

Loading