Introduce FingerId

[daxpedda]

This replaces Touch::id with Touch::finger_id and introduces a new type FingerId instead.

Additionally on Web it exposes PointerEvent.isPrimary which tells the user if the finger is the primary/first one in a multi-touch interaction.
I also went ahead and implemented this for Windows.

@MarijnS95 pointed out how this could be implemented for Android in #3783 (comment).
But other OSs would have to be polyfilled.

We could potentially expose this directly to the user without a platform specific extension trait if we implement enough platforms.

Followup
Entry in #2875: Firefox emits wrong primary fingers (Bugzilla).

[kchibisov]

What's the end goal/motivation with it if you mind sharing? Like it's really not clear how you'd use that aside from touch related stuff?

[daxpedda]

What's the end goal/motivation with it if you mind sharing? Like it's really not clear how you'd use that aside from touch related stuff?

Its only used for touch and multi pointer situations.

I'm considering implementing this cross-platform for touch only and close this.

[daxpedda]

So after looking into how exactly browsers implement isPrimary I discovered that its basically a polyfill where they figure out what the first touch was by tracking finger IDs between touch start and end.

Windows is the only OS that supports detecting it via the TOUCHEVENTF_PRIMARY and POINTER_FLAG_PRIMARY flag.

So I see the following options:

1. Change Touch::id to FingerId and implement platforms traits on it.
2. Add Touch::primary and polyfill it for all platforms except Web and Windows.
3. Add extension traits on Touch, which will require us to implement Default or Touch::new() if we want to keep it user constructible.

Personally I'm preferring to go with solution 1.
WDYT?
Cc @kchibisov, @madsmtm, @MarijnS95, @notgull.

[kchibisov]

But if the primary is the first click couldn't it be entirely user detectable? I also don't really understand how one would use it?

Its only used for touch and multi pointer situations.

But how it's supposed to be utilized by the user and what they'll do with it? Is it some web limitation preventing it detecting multitouch correctly? because you should generally understand the order of touch events, etc?

Like in JS world I'd assume that users can just track amount of fingers pressed at the given time or something.

[daxpedda]

But if the primary is the first click couldn't it be entirely user detectable?

Yes and no, its not easy to reliably detect on Web, e.g. starting touch point could be outside the canvas etc.

But how it's supposed to be utilized by the user and what they'll do with it? Is it some web limitation preventing it detecting multitouch correctly? because you should generally understand the order of touch events, etc?

Like in JS world I'd assume that users can just track amount of fingers pressed at the given time or something.

Its only relevant in scenarios where you want to track the primary finger in a multi-touch interaction. This can be common in games, where the primary finger has a certain function that all other fingers don't have, e.g. movement vs shooting.

[kchibisov]

Its only relevant in scenarios where you want to track the primary finger in a multi-touch interaction. This can be common in games, where the primary finger has a certain function that all other fingers don't have, e.g. movement vs shooting.

@madsmtm @MarijnS95 do you happen to know if something like that exists on ios/android? I'd assume users detect all of that themselves?

In general, I don't really mind having solution 1.

[MarijnS95]

I don't exactly understand what this "primary" pointer is in Web; is it just the first pointer/touch that started?

In that case, on Android, you could keep track of the first pointer_id() that went down.

There's also the mention for AMOTION_EVENT_ACTION_POINTER_DOWN and AMOTION_EVENT_ACTION_POINTER_UP that a non-primary pointer has gone down/up. Probably the fist finger triggers AMOTION_EVENT_ACTION_DOWN/AMOTION_EVENT_ACTION_UP instead?

---

Its only relevant in scenarios where you want to track the primary finger in a multi-touch interaction. This can be common in games, where the primary finger has a certain function that all other fingers don't have, e.g. movement vs shooting.

On an unrelated note that sounds very confusing: the first touch controls movement, the second controls aim? I always thought that it was position-based: left half of the screen is moving, right half is panning the camera?

In any case it's fully trackable by the user at least.

[daxpedda]

I don't exactly understand what this "primary" pointer is in Web; is it just the first pointer/touch that started?

Yes.

On an unrelated note that sounds very confusing: the first touch controls movement, the second controls aim? I always thought that it was position-based: left half of the screen is moving, right half is panning the camera?

I have very little experience with games like that, so I can't say much to that.
For our own use-case, where its not position-based, we tracked it ourselves at start, but because we had some controls outside the canvas (for various reasons not in our control), it interfered sometimes.
So we needed to use isPrimary.

[MarijnS95]

On Android an app is usually fullscreen but it doesn't need to be - we should test what happens when a motion gesture enters and exit the app similar to your canvas example!

[daxpedda]

I went ahead and implemented FingerId, ergo solution 1.

[MarijnS95]

In that case, on Android, you could keep track of the first pointer_id() that went down.

There's also the mention for AMOTION_EVENT_ACTION_POINTER_DOWN and AMOTION_EVENT_ACTION_POINTER_UP that a non-primary pointer has gone down/up. Probably the fist finger triggers AMOTION_EVENT_ACTION_DOWN/AMOTION_EVENT_ACTION_UP instead?

After testing this in winit, the results are somewhat surprising regarding this action() enum discrepancy:

• On first finger down, a Down event is emitted with pointer_id() set to 0;
• On second finger down, a PointerDown event is emitted, as expected for a non-primary. The ID is set to 1;
• When the first finger is removed, which used to be the primary finger, a PointerUp is emitted, surprisingly not primary. This event carries pointer_id() = 0;
• Subsequent moves for the "second finger" keep being emitted with pointer_id() = 1 as expected;
• Finally, removing the second finger emits an Up event.

Note that if, instead of removing the second finger, another finger is added, non-primary PointerUp/PointerDown events are still emitted, but pointer_id() = 0 is reused.

---

I don't know exactly what kind of API winit wants here, but if we want to track the first finger in a series of touches, you'll do something like this:

EDIT: Since this is a platform extension, there's nothing to winit to "want" if the current platform doesn't support it. No need to polyfill, we're better off providing users the actual info that Android provides which is pointer_id() (already in there). Discrepancies between PointerUp vs Up may be interesting to know if this is the first touch to start or the last touch to finish though.

• On Down (primary), store its pointer_id();
• On (Pointer)Up, reset this value if it is equal to pointer_id();
• is_primary() returns self.primary_pointer == Some(pointer_id()).

Meaning that no motion is primary anymore, if the primary finger was removed after some point in time (while a touch is still ongoing).

[daxpedda]

EDIT: Since this is a platform extension, there's nothing to winit to "want" if the current platform doesn't support it. No need to polyfill, we're better off providing users the actual info that Android provides which is pointer_id() (already in there).

I agree!

[MarijnS95]

Looks good on the Android side (very minimal)!

If anyone needs it we could expose whether this is the first ("primary") finger going down or the last (""primary""?) finger to go up. This is also easy for the user to track by keeping a rolling count of how many up and down events they have seen though.

(Which could be a little cheaper if we create a separate extension keyed off the difference between Down and PointerDown, and (Pointer)Up respectively, but would have to store a bool for it anyways)