W3C

- DRAFT -

Pointer Events W

06 Jul 2016

See also: IRC log

Attendees

Present
dtapuska, Rick_Byers, patrick_h_lauke, teddink, NavidZ, Scott_Gonzalez, shepazu
Regrets
Chair
patrick_h_lauke
Scribe
dtapuska

Contents


<patrick_h_lauke> Meeting: Pointer Events Working Group

<scribe> scribe: dtapuska

https://github.com/w3c/pointerevents/pull/76

NZ: ted seemed to be looking at it and talking to the team

RB: This is a subtle enough space where we have some implementation differences

TD: The PR as it stands today is a little bit different then Edge behaves
... we don't send lost pointer capture until the next event comes in
... there is a synthentic event that we should be suppressing that causes lost pointer capture to occur early
... if you don't move the mouse and wait a few seconds you will get a lost pointer capture
... but we consider that a bug
... on the got pointer capture side; we only send got pointer capture when set pointer capture is called as a result of pointer down
... we did that way back when for interoperability reasons
... unclear whether this is still needed or not.
... still in the midst of doing some of that investigation
... Ricks point is accurate to get the spec closer to the way chrome and edge are currently doing it... then we can discuss in person the subtle differences
... haven't looked at what chrome does in canary

NZ: chrome follows the spec. We delay the got and lost pointer capture until the next event.

RB: And we don't really like that in terms of the developer
... it is hard to reason about

NZ: Right and then we thought about doing the pointer down hack Edge has. But then we really want to send it right away as soon as we can.
... And then aren't really delayed in Edge because of the synthetic event that fires

RB: I'd rather spec something that is coherient/rational. and continue iterating on it post ship

NZ: Right now chrome does what the current spec says; but we can change Chrome to match what this PR says.

RB: I think we will have a set of outstanding issues and we either match the spec (and not edge); or match against a pull request

NZ: We probably should resolve this PR as Mozilla was looking at this too; whether they should get a synthetic event..

RB: its a step in the right direction to land something in the spec
... what do you think ted?

TD: I think I could get behind it; and it is fine for a div that calls set pointer capture; but when something outside of the normal pipeline does it

RB: Can we agree to land something along these lines and then we can match chrome to this spec. And then file a PR that indicates Edge doesn't match the spec

PL: Should we add a highlight in the spec that something may change here?

RB: That depends on how we do that.... some specs do it inline; some don't.
... do we track them inline or use github issues
... I'd prefer to use github because it can confused the reader.

PL: git hub seems fine for me

TD: I agree as well. It is subtle things

RESOLUTION: land PR (in some form), track potential issues/compat worries separately in github until we have more data

DF: First public working draft publishing; do we need issues resolved before hand? Or can we publish the spec as is?

RB: I'm comfortable with that

PL: So that goes FPWD
... do I need to write some pseudo marketing blurb?

DF: w3c usually doesn't do that; but if you guys want to write blog posts about it. Happy to work with you on that if you guys want to.
... we should write a blog post about this spec for the w3c blog to indicate how this spec is great

Removed "pen contact" condition on button/buttons

<patrick_h_lauke> https://github.com/w3c/pointerevents/pull/96

NZ: On this I had a question for ted on his last comment. I didn't understand the css state

<patrick_h_lauke> oh, and i just realised that was actually the topic for Mustaq who's not here sorry

TD: was that 96 or 93?

Clarify click event firing for chorded buttons

<patrick_h_lauke> https://github.com/w3c/pointerevents/issues/93

NZ: So my question is related to this one.. 93

TD: Ya I think I have some questions for dev based on the response I got back..
... he saw an active css class applied

NZ: So going back to the UIEvents spec; for every button primary button we to fire a click event
... and then we discussed about the barrel button and clicking the pen. We ciould to suppress the click event
... maybe we should fire a click for every mousedown and mouseup; but chrome fires a click for every button event not primary

RB: Ya we shouldn't worry what chrome does; we have a separate bug for that. This is a really corner case. If someone is doing something with corded buttons they probably aren't listening to click events
... Perhaps we should spec what edge is doing

NZ: If you release a button while the primary button is down it will cancel the click.

RB: Perhaps we should get the UI Events spec to updated if any mouse button goes down while one is down perhaps click doesn't fire

NZ: That gets complicated if we introduce the auxclick behaviour.
... What is the intent of the user while they have a primary button down

RB: If I lift the release the barrel button after I lift the pen off the screen I don't get a click
... but if it is still on the screen I do get a click

TD: Ya I don't know

RB: Should we just file a spec behaviour or UIEvents?
... I agree getting it intertwined can be confusing
... I think this is orthogonal to pointer events

TD: So test 1 chrome would work like edge
... But for test 2 chrome would send a click for the last pointer up for button 0

RESOLUTION: Rick (?) to file bug on UIEvents spec to ask for clarification, as it's orthogonal to pointer events; Chrome to fix its current bug (test 1)

TD: I think that makes sense; and I'll get a bug open for edge

RB: this is minor

TD: Ya I don't think corded buttons actually really get used

setPointerCapture should say something about iframes

<patrick_h_lauke> https://github.com/w3c/pointerevents/issues/16

RB: So the behaviour we are doing in Chrome matches UIEvents.
... Add a note that any change in other buttons is irrelevant
... Ya I think this is subtle; I don't think it blocks shipping; can save this for a F2F

RESOLUTION: pretty subtle scenario, but not blocking v2 - more discussion on github if needed. nice for consistency perspective, but not a major attack vector concern

TD: Ya our security team feels the way; there is potential something but not high priority

Should a captured pointer send boundary events by default?

<patrick_h_lauke> https://github.com/w3c/pointerevents/issues/61

<patrick_h_lauke> linked to https://github.com/w3c/pointerevents/issues/39 and https://github.com/w3c/pointerevents/issues/8

PL: This is the big topic that we've been pushing around.. I think ted was going to circle back

TD: Ya it isn't that easy we are still trying to figure out what sites to understand then whether if we change something whether we would break them
... as well as mining the bug database and talking to the dev team; but this is still an ongoing effort. Ya if you could find specific sites it would be great to do that in a couple of weeks

RESOLUTION: ongoing work to look at bugs and dev expectations; something to discuss further at F2F; decide if variance is a blocker for shipping in chrome

RB: We may not decide what the right answer is before we ship. I think we should decide as a group whether some features is reasonable for chrome to ship with a different implementation than edge

Outstanding PRs

<patrick_h_lauke> https://github.com/w3c/pointerevents/pulls

NZ: specifically the last two have been open for a longer time; about hit testing and boundary events
... We'd like to do no hit testing when pointer is captured

RB: We need to have chrome match the spec with these Pull requests
... ok to have open pull requests when there are open design concerns

TD: Ya I think the lost pointer capture; needs to be concerned about the hit testing and the performance of the hit testing
... do you have any specific data in terms of how much hit testing costs?
... is hit testing costly in chrome that you need to worry about

RB: There are performance experts that would argue that it is important. And they would block shipping pointer events based on it.
... And so we did some comparisons between edge and chrome based on the cost and it was very similar. But then question is how much is expensive and we have different opinions on that
... we have 16ms per frame; and if it takes 1% of the frame budget then it is too much.
... there are some times that take 100ms for a table with lots and lots of rows; and the code is a big ball of legacy code. So there are concern of making more things dependent on that.
... hit testing is part of the Response in RAIL; and not the Animation conformence

TD: Ya our performance team is also concerned about where we spend our time

RB: We constantly compare web and android; and the user shouldn't notice the difference. They notice all the advantages of the web; but not the performance disadvantages. And so we don't want to introduce dependencies
... This is our biggest open issue. We need a plan on how to ship pointer events in chrome before the end of tpac
... the right people will be there. I'd like the working group to be happy to support us shipping pointer events after tpac; with a set of issues.

DS: There is no hit testing spec. I'm curious if there is some interest in us writing up a spec about hit testing.

RB: I think there are 1000% things that we do as I think it would be really hard to get concensus on the Hit testing spec

<patrick_h_lauke> +1 to test suite first, rather than spec

RB: I've been marking hit testing interop bugs with a certain key... we already are quite interoperable; but perhaps we should start with a test drive for hit testing. Like 100 tests that work for everyone; and then deal with the edge cases

<rbyers> https://bugs.chromium.org/p/chromium/issues/detail?id=549211

RB: ok we have 95 tests that we all agree on and then define the reasoning behind them

DS: That is good; anything that improves interop is great. But I think writing down the things that the spec is based on. And if there are disagreements that we record their behaviour. So something descriptive not prescriptive
... Does that seem like something to do?

RB: I'm not opposed to someone doing it; but I think to do it well it would take many man years. We could spend a lot of effort; but little benefit. Some benefit but in terms of making web developers lives better perhaps not

DS: So going back to a test suite is that something useful?

RB: It is only useful if multiple vendors agree to run them
... this is a dicussion to have a tpac; there is a session on interop

DS: Ya I'll try to attend that; and bring up hit testing
... I agree the cost could be high; but we could see what benefits of it are.

RB: And if anyone has a set of bugs that are interop between vendors it would be good to catalog them.
... or if how it is to design a new engine. like is it difficult for servo

DS: I'll bring it up at tpac and if there are more eyes involved then optimizations might come up...

PL: Only other point implementation status; and we are over time; so we will call it
... same time next week. See you then

Summary of Action Items

Summary of Resolutions

  1. land PR (in some form), track potential issues/compat worries separately in github until we have more data
  2. Rick (?) to file bug on UIEvents spec to ask for clarification, as it's orthogonal to pointer events; Chrome to fix its current bug (test 1)
  3. pretty subtle scenario, but not blocking v2 - more discussion on github if needed. nice for consistency perspective, but not a major attack vector concern
  4. ongoing work to look at bugs and dev expectations; something to discuss further at F2F; decide if variance is a blocker for shipping in chrome
[End of minutes]

Minutes formatted by David Booth's scribe.perl version 1.144 (CVS log)
$Date: 2016/07/06 16:05:42 $

Scribe.perl diagnostic output

[Delete this section before finalizing the minutes.]
This is scribe.perl Revision: 1.144  of Date: 2015/11/17 08:39:34  
Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/

Guessing input format: RRSAgent_Text_Format (score 1.00)

Succeeded: s/does that/doesn't do that/
Succeeded: s/no/not/
Succeeded: s/DG/DS/
Found Scribe: dtapuska
Inferring ScribeNick: dtapuska
Present: dtapuska Rick_Byers patrick_h_lauke teddink NavidZ Scott_Gonzalez shepazu
Got date from IRC log name: 06 Jul 2016
Guessing minutes URL: http://www.w3.org/2016/07/06-pointerevents-minutes.html
People with action items: 

[End of scribe.perl diagnostic output]