RE: UI Events - Mouse Events and iframe targetting

On the Microsoft side we discussed this and here is a summary of those discussions:

Everyone on the EdgeHTML side prefers the behavior that we have today. The example that came up a couple of times was drag and drop – if events are implicitly captured to the iframe, then there is no easy way to hand off the interaction from one frame to another.

We generally thought that the current approach in EdgeHTML is straightforward – events/hover applies to whatever is under the mouse, regardless of movement over iframes or button presses.

We investigated whether we fire enter/leave when leaving the iframe, and confirmed that we do fire the events on the element, not the document.

We agree that developers likely want a setPointerCapture approach, which gives the web developer flexibility and more control.

Thoughts?

From: rbyers@google.com [mailto:rbyers@google.com] On Behalf Of Rick Byers
Sent: Tuesday, June 23, 2015 11:19 AM
To: Olli Pettay <olli@pettay.fi>
Cc: Dave Tapuska <dtapuska@chromium.org>; www-dom@w3.org; Jacob Rossi <Jacob.Rossi@microsoft.com>
Subject: Re: UI Events - Mouse Events and iframe targetting

On Tue, Jun 23, 2015 at 12:01 PM, Olli Pettay <olli@pettay.fi<mailto:olli@pettay.fi>> wrote:
On 06/22/2015 11:00 AM, Rick Byers wrote:
Thanks Dave,
I've been aware of differences between browsers here for awhile, but only recently been convinced that it's important that we try to fix this. /cc
Jacob and Olli in case they have specific experience/advice for EdgeHTML and Gecko respectively.

So the idea behind "2) Firefox targets the mouse down frame unless you move into a sibling iframe"
is that iframe and top level browsing context would get similar behavior when dealing with mouse event based dragging.
If you have mousedown, you flag the mousedown browsing context as the active and any mouse events in ancestors (or outside the browser window) are
forwarded to the active browsing context.
(This was perhaps slightly based on the single process Firefox architecture where 'chrome browsing context' hosts the top level web browsing contexts).
I do consider the 'move into a sibling iframe' a bug, since the idea was to deal nested iframes only in that way, not also siblings.
The relevant change was done in https://bugzilla.mozilla.org/show_bug.cgi?id=603550 (or more like, that fixed Gecko to behave as it had behaved for ages).

As I said in that bug (5 years ago ;)),
I think I'm leaning over to the behavior where mousedown frame would always get the events, in other words implicit capturing.

Thanks Olli!  If it's sufficiently web compatible then I'd prefer implicit capturing too.  That would often just be the right thing and wouldn't require any additional APIs to solve all the important scenarios I know of.  If other engines were open to trying to move to this model, we could give it a shot in Chrome.

-Olli




In particular, I had a great discussion with the Google Maps team about why this matters to them.  The Google Maps embedded API
<https://developers.google.com/maps/documentation/embed/> lets you host a map in an iframe on your page.  Here when you click on the map and drag to
pan it, the user really expects that it continues to pan even if the cursor leaves the frame.  They've got some ugly UserAgent-specific hacks to make
this work on all browsers and these hacks have gotten in the way of us fixing other interop problems in Chrome.

So, we need some standard interoperable way to enable such a scenario.  But this doesn't mean we have to do it by default if that's hard to agree on
or likely to cause developer confusion.  Is this scenario possible today in IE / Gecko using the non-standard setCapture API
<https://developer.mozilla.org/en-US/docs/Web/API/Element/setCapture>?  Presumably setPointerCapture
<https://w3c.github.io/pointerevents/#pointer-capture> can definitely be used for this purpose (though the spec is not clear
<https://github.com/w3c/pointerevents/issues/16>).  But I'm not sure we want to block a solution to these problems on Pointer Events (eg. there should
be some path for Safari as well).

Has there been any discussion about standardizing setCapture?  If IE and Gecko already have an interoperable implementation here, perhaps we should
consider that a defacto standard and add it to blink too, then change our mouse event behavior to match IE's?

But still, I think the UI Events spec should be updated to be clear on what the default behavior for mouse events should be in this regard.

Thoughts?
    Rick

On Mon, Jun 22, 2015 at 1:21 PM, Dave Tapuska <dtapuska@chromium.org<mailto:dtapuska@chromium.org> <mailto:dtapuska@chromium.org<mailto:dtapuska@chromium.org>>> wrote:

    It was pointed out to me offline that I didn't include the sample page that demonstrates the issues with the vendors; to find it you'd need to dig
    into the chromium bug; for simplicity I've put it on github. See http://dtapuska.github.io/iframe-mouse-target/iframe_outer..html

    <http://dtapuska.github.io/iframe-mouse-target/iframe_outer.html>


    On Tue, Jun 16, 2015 at 4:16 PM, Dave Tapuska <dtapuska@chromium.org<mailto:dtapuska@chromium.org> <mailto:dtapuska@chromium.org<mailto:dtapuska@chromium.org>>> wrote:

        It seems that processing mouse [move|up] events have 3 different implementations when involving an iframe in each rendering engine.

        I'm soliciting comments and hope we can work on defining some behavior and converging our implementations to benefit the web author.

        Specifically when an mouse down event occurs inside an iframe; the subsequent mouse moves/ups are targeted at different frames when the mouse
        moves outside the bounds of the iframe.

        1) Chrome targets the frame that generated the mouse down frame sometimes (but has side effects with prevent default; see
        http://crbug.com/269917; we need to fix this).
        2) Firefox targets the mouse down frame unless you move into a sibling iframe
        3) IE 11 & Edge target the topmost frame under the mouse move regardless of the mouse down operation.

        We need to get some clarity written into the specification. As the spec indicates the target is:

          * |Event.target| <http://www.w3.org/TR/DOM-Level-3-Events/#widl-Event-target>: topmost event target
            <http://www.w3.org/TR/DOM-Level-3-Events/#glossary-topmost-event-target>


        Likewise in Edge/IE 11 you can have a mouse down event in the iframe; but the mouse up event occurs in the main frame.

        All browsers do not generate a mouse leave/enter on the iframe document when the mouse leaves or enters the iframe when a mouse button is
        depressed; this can leave the iframe in an inconsistent state possibly.

        So I would expect that the IE 11/Edge implementation is spec compliant. However this doesn't necessarily seem correct to the web author point
        of view.

        I can see why the target is defined is set to be the iframe that handled the mouse down event because if you are dragging something around you
        might not want to start processing mouse move events on the outer frame until the mouse is released.  The crbug references google maps
        embedded inside a page as a use case that occurs in the field today.

        With respect to hover processing:
        1) Chrome doesn't generate hover events for items outside of the iframe during a mouse down.. (Seems incorrect).
        2) Firefox does the interesting behaviour of generating hover events for items not in the parent frame but in sibling frames.
        3) IE 11 & Edge generate hover on whatever is under the mouse.

        The IE 11/Edge implementation seems straight forward. Perhaps some folks can confirm that it is as I think it to be.

        The Firefox implementation mystifies me why it behaves different between the main frame and a sibling iframe is interesting. Perhaps I'm
        flawed in my reproduction step; but some explanation behind this would be appreciated.

        The Chrome/WebKit implementation makes sense a little bit of sense to me as well; granted it has some weird bugs like hover and prevent default.

        dave.