Add an exception to the requirement that partial method declarations have the same modifiers

[jnm2]

This code is accepted just fine:

partial class C
{
    async partial void M() { }

    partial void M();
}

And this is by design. It's not allowed for the async modifier to be on both parts. Nor would that make intuitive sense, since async is a keyword that only affects the body. It does not affect the signature.

partial class C
{
    async partial void M() { }

    // ❌ CS1994: The 'async' modifier can only be used in methods that have a body.
    async partial void M();
}

When C# 9 comes along, extern will have to be added as another exception for the same reason.

[jnm2]

@BillWagner @jskeet This looks like a candidate for merging without a meeting discussion.

[Nigel-Ecma]

Nor would that make intuitive sense, since async is a keyword that only affects the body. It does not affect the signature.

Could not an implementation choose to invoke an async method differently than a non-async one? And user code surely does? So why does it only “affect the body”?

While I do not doubt that a certain compiler made this choice intentionally, I’m unsure that an implementation which choose to allow it should be deemed to be in error.

[jnm2]

Could not an implementation choose to invoke an async method differently than a non-async one? And user code surely does? So why does it only “affect the body”?

User code does not invoke an async method differently from a non-async one. Nor would an implementation invoke it differently. The effect of the async keyword is invisible to callers of the method.

What the async keyword does in a nutshell is rewrite the body of the method to use a state machine to produce the Task object (or some task-like object) which is returned by the method, rather than having the user manually return some Task object. The effect of async from a language user perspective is not "up to the implementation" but is rather a semantic mode switch that the user understands.

Either way, the caller invokes the method the same way (a direct method call), and it sees only the Task object which is returned the same way (from that direct method call). The signature is entirely unaffected by the async modifier by design. It can be swapped in and out to suit the method implementer.

[jnm2]

And then, given that the async keyword affects only the body, it would be nonsensical to allow the keyword on a method which has no body. So an implementation which ignored this nonsensical placement should be in error. The thing which the keyword refers to is not there.

[Nigel-Ecma]

What the async keyword does in a nutshell is rewrite the body of the method to use a state machine to produce the Task object (or some task-like object) which is returned by the method, rather than having the user manually return some Task object.

Where in your code:

async partial void M() { }

is this “Task object” you mention specified, the signature seems to say it returns void.

Is it perhaps indicated by the async, which thereby informs that calling this method should be by way of something like an await and it returns a Task object?

So is this method perhaps not called like one taking no arguments and returning nothing as its “signature” might suggest?

Anyway, neither of us should be here today! Happy New Year, see you in 2026 🙂

[jnm2]

Oops, I forgot that part. With async void all the same things hold true about its lack of effect on the signature:

• async enables the await keyword within the body and rewrites the body as a state machine, for example marshaling exceptions elsewhere
• async may be added or removed without affecting the caller in any way- the caller does not react to its presence

But the caller cannot observe when any asynchronous work is completed due to the return type being void. The method is called like one taking no arguments and returning nothing, since that is what it is.

I appreciate that ☺️ Happy 2026 to you as well!