Changes EventHandler<...> to have a this of type void.

[MicahZoltu]

Originally, this was set to E['currentTarget'] which, based on the MDN doc linked in #2166, certainly sounds like the correct thing to do. A couple years later it was changed to never in #3147 with no commentary on the PR or commit indicating why. My guess is that this was changed so arrow functions would work, which I believe are permanently bound and cannot be rebound by the caller.

I believe that never is wrong because this code doesn't compile:

declare const event: JSX.TargetedEvent<HTMLElement, Event>

declare const apple: JSX.EventHandler<JSX.TargetedEvent<HTMLElement, Event>>
apple(event) // error TS2684: The 'this' context of type 'void' is not assignable to method's 'this' of type 'never'.

function Apple(model: Pick<JSX.HTMLAttributes<HTMLElement>, 'onInput'>) {
	model.onInput && model.onInput(event) // error TS2684: The 'this' context of type 'Pick<HTMLAttributes<HTMLElement>, "onInput">' is not assignable to method's 'this' of type 'never'.
}

Changing to void resolves this error and I believe will work everywhere that never worked.

I removed the comment since it no longer applies with this being never or void. It appears that it was just forgotten when the switch from E['currentTarget'] to never was made.

[robertknight]

A couple years later it was changed to never in #3147 with no commentary on the PR or commit indicating why.

@JoviDeCroock can you remember the rationale for the change?

Edit: There is some context in #3137.

Personally I would like to amend the TypeScript types to reflect this, and simply have the context of events be never or some value that implies it is not to be used.

You can't call a function if it's this is set to never, since TS will infer the type as void, so that definitely seems preferable if we wanted to prevent users from relying on using this.

[MicahZoltu]

You can't call a function if it's this is set to never, since TS will infer the type as void, so that definitely seems preferable if we wanted to prevent users from relying on using this.

I believe that void addresses this problem as void can't actually be used for anything, similar to never.

[JoviDeCroock]

Hey, this was a big miss on my behalf. The failing test is what I think lead us down the path of changing this. Going to fish our issues in a bit. When trying it out in the browser logging out this seems to be the Element when using function and window when using () => {} I think that was the reasoning behind making it never.

EDIT: this was one of the original issues #3137 I wonder whether we can make it the element 😅

[MicahZoltu]

I get window for both when the function is not bound to anything.

When you bind an arrow function however, the binding is ignored (this always points to globalThis):

The underlying problem here is that JavaScript functions can be rebound. Just because this is the element initially, doesn't mean that will be the case if someone calls the function elsewhere unless the function is explicitly bound.

The net takeaway here I think is that relying on this being the element could be safe, but preact would need to take care to ensure it doesn't accidentally rebind this along the way (which is easy to do if you are using classes).

[MicahZoltu]

Reviewing the failing test, it appears that the test is incorrect here:

// Test `this` binding on event handlers
function onHandler(this: HTMLInputElement, event: any) {
	return this.value;
}
const foo = <input onChange={onHandler} />;

Specifically, if we want to say that there are no guarantees that this is the element in these callbacks then the test is correctly failing now while it was incorrectly passing previously. In the test case, if this was not an element the user would get an error about trying to access a property on undefined.

On the other hand, if we want to guarantee that this is the element, then we should switch the signature back to E['currentTarget'], and do whatever work is necessary to ensure that guarantee is made.

I'm happy to update the PR to "fix the test" (delete it), or switch back to E['currentTarget'].

[JoviDeCroock]

Let's delete the test, I think we should be good to tell folks not to rely on this here

[coveralls]

Coverage: 99.531%. Remained the same when pulling 150b1b0 on MicahZoltu:patch-2 into 15913ad on preactjs:master.

[MicahZoltu]

Let's delete the test, I think we should be good to tell folks not to rely on this here

Done, and tests passing.