W3C home > Mailing lists > Public > public-houdini@w3.org > March 2015

Minutes Sydney F2F 2015-02-08 Part III: Custom Paint, Custom Layout

From: Dael Jackson <daelcss@gmail.com>
Date: Tue, 3 Mar 2015 14:24:35 -0500
Message-ID: <CADhPm3tw36K7vvmAXXFBwLc+MFH6y_U+43WobSUFSYxnRG3sww@mail.gmail.com>
To: public-houdini@w3.org
Custom Paint
------------

  - There was ample conversation about how this spec should be
        constrained to provide maximum opportunities without
        creating too much complexity or problems with complex
        objects, like backgrounds that contain video.
  - RESOLVED: shane, iank to start work on CSS Paint Module

Custom Layout
-------------

  - There was a lot of concern addressed about making sure that the
        properties in this module don't create loops or hangups, so
        the editors will make sure to keep that in consideration as
        they write the spec.
  - Since this is such a large undertaking, it's also likely that it
        will be handled in chunks instead of being solved all at
        once.
  - RESOLVED: gregwitworth, iank, TabAtkins, shane, roc, and Rossen
              to start work on CSS Layout Module

===== FULL MINUTES BELOW ======

Custom Paint
------------
  Scribe: jet

  Rossen: This afternoon's agenda:
  Rossen: Custom paint, custom layout,
  Rossen: Input extensions, async, threading, process.

  Iank: [whiteboard]
  <ChrisL> custom paint
https://c1.staticflickr.com/9/8061/8206144541_4843dd6445.jpg
  Iank: Custom paint assumes invalidation e.g. custom properties.
  iank: Properties that invalidate paint (not layout) should have
        paint callbacks.
  iank: With custom context to be safe,
  iank: To not leave layout in an inconsistent state,
  iank: There's roughly have 3 areas to override:
        1. on BG layer
        2. on content
        3. on outline (potentially)
  dino: Does this include children?
  iank: Not sure.

  roc: There's an argument to be able to modify paint for your
       children, e.g. for effects.
  roc: We should constrain scope.
  iank: We should constrain to just your paint (not children).
  iank: For example in a function(ctx, style) get a canvas context
        and style information.
  iank: ctx.clipPath = path
  iank: ctx.paint(self)

  krit: Can we cover the different paint phases?
  iank: Most effects are covered, per canvas.
  iank: Or we just use the background layer depending on how
        fine-grained.
  krit: What are the use cases for custom paint?
  iank: An example is Google material design.
  iank: It draws custom shadows on buttons.
  iank: Currently we add DOM nodes to get this effect.
  krit: Why do that? Chrome can already draw a canvas element as a
        background.
  iank: [Projects demo on screen]
  <iank> http://www.polymer-project.org/components/paper-button/demo.html
  iank: We can't do this easily with DOM elements today, and need
        script for that.
  iank: When it was first designed, button effects had gradients.

  dino: You already have the context, should be able to set what you
        want, like WebVR.
  dino: You need to know the backing store to cover the element and
        effect.
  dino: It needs more knobs on the API.
  esprehn: Currently it's a read-only API no backing store.
  dino: It may need to be able to get readback (eg from WebGL).

  iank: For example, for getting 3D canvas?
  dino: Don't limit it to 2D API.
  dino: We might need a retained model like SVG.
  dino: Can do similar effects with SVG as input to custom paint.
  esprehn: With 2D canvas you can address a lot of use cases.
  esprehn: We can do 3D and displayList later.
  iank: Initially, specify 2D context, leave it open for later.

  krit: How is it different from BG canvas feature?
  iank: There's problems with webkit BG canvas.
  iank: You can't use the default canvas.
  roc: I don't want to introduce new rendering, just modify existing
       parts.

  esprehn: Want to call in before and after paint.
  esprehn: [shows physics effect on whiteboard]
  iank: You need a known width/height before use for canvas.
  iank: With this proposed API you can get dimensions at paint time.
  iank: Say width is animating, you can do effects without
        synchronous layout flush.
  roc: You can apply an effect, draw yourself, but if content is
       complicated, layerization gets messed up.
  roc: Such as video on the background, you call drawSelf( ) it
       draws the button and video, but now layer decisions get
       complicated.
  esprehn: You'll need to opt into staying on top or below other
           content to force layerization.
  roc: If there's clipping, you need to wrap your content with
       different options.
  esprehn: You need some way to push a clippath or opacity layer.
  esprehn: Probably you'd want a surface area that's closer to a
           traditional drawing API.
  esprehn: Or use a display list.

  krit: Are the 3 context options combine-able?
  iank: No, one per paint phase.
  iank: Depending on what's useful for authors.
  dino: If you're doing something with video, the compositor handles
        video paint.

  krit: Can we handle certain phases first?
  roc: How important is it to modify content now?
  esprehn: Can't mask to arbitrary shapes.
  roc: You can do it with SVG masks.
  esprehn: You can't do it without layout flush.
  roc: Which use case?
  esprehn: I want to mask descendants based on current size.
  esprehn: And to scale the effect as size changes.
  roc: Is that OK to lag? Concern is trying to apply custom effects
       to the contents of an element that aren't accessible easily.
  dino: With video, you'd get a lag unless we do compositor tricks.
  roc: It ramps up the complexity for edge cases.
  esprehn: Just for clippath?
  roc: You can get into very complicated states with multiple
       transforms, or not draw at all, or call drawSelf() twice.
  iank: drawSelf() is constrained before() or after() only.
  roc: The complexity is added by masks, etc.
  shane: It seems pretty standard to invoke custom behavior before
         or after a built-in context,
  esprehn: Add clippath, opacity, transforms with API.
  dino: 3D API's allow hooks into different phases of the rendering
  iank: Mangle with a limited subset within API.

  dino: Does this happen before filters and xforms?
  shane: Filters should apply.
  krit: Filters affect all operations.
  dino: You won't be able to override a CSS clippath with custom
        paint.
  iank: That should be fine.

  shane: Why can't we expose clippath?
  dino: You shouldn't be able to modify the clippath passed into you.
  shane: You can modify local clippaths.
  dino: I'm worried about restrictions per context,
  dino: To avoid touching the backing store.
  shane: We can add other contexts later.

  roc: What kinds of restrictions do we want in the JS environment?
  iank: Initially, just get the context, read the style, get the
        fragment size.
  roc: Do you have persistent global state between paints?
  iank: The style object passes in state.
  esprehn: The features are coupled with custom CSS properties (eg.
           to cover animated physics based on buttonDown).
  esprehn: [whiteboard]

  iank: State comes from style.
  roc: We won't need to count paints without global state.
  roc: How to send/receive messages?
  iank: Via the style object.
  dino: How about images?
  iank: Use an arrayBuffer with the image data.
  roc: Use transferables/structured clones in the API.
  dino: What's in the style object?
  iank: Either all computed style or just what relates to paint.
  iank: The author can specify which properties they want.
  esprehn: You should use the current geometry, not supplied sizes.
  roc: You need to have access to custom and built-in styles, it
       might be good to share custom paint and custom layout
       features.
  esprehn: Except that not all layout causes paint.
  <bkardell> It would be good to sketch out the process /
             relationships of what we are talking about (all of them
             I mean)
  dino: You can specify what you want to invalidate paint.
  dino: It looks a lot like CSS shaders.
  dino: Are you allowed to control how long it should take to paint?

  bkardell: It would be good to list what happens from parse to
            layout to paint and show how to plug into the API.
  esprehn: The specifications are intertwined.
  esprehn: First, a spec on the JS execution (in limited context)
  esprehn: Then specs on the layout and paint.
  <bkardell> +1

  krit: How do you prevent pixel leakage?
  roc: How to prevent timing attacks?
  dino: Limit out drawSelf()?
  esprehn: webGL makes sense above or behind content.
  krit: That would allow limited access to GL pixel data.
  roc: Rendering to GL surfaces will be high overhead.
  dino: I'm more worried about performance issues from usage.

  Rossen: Any concerns re: SVG?
  krit: We don't have z-ordered paint phases in SVG (yet)
  ChrisL: Canvas is expected to work with SVG.
  krit: SVG should be able to work as well.

  Rossen: proposed resolution: shane, iank to work on Custom Paint
          proposal (CSS Painting)

  RESOLVED: shane, iank to work on CSS Paint Module

  rbyers: Do we need an uber spec?
  <bkardell> rbyers: at least a note

Custom Layout
-------------

  shane: Custom Layout, callbacks like for paint,
  shane: Are to read pieces of computed style that's targeted for
         Layout.
  shane: Filter the computed style to the pieces that are relevant
         for Layout.
  shane: This will depend on fragment tree spec.
  shane: API needs to create such fragments.
  dbaron: You want this to be separate from normal layout,
  dbaron: With some new display type to invoke it.
  dbaron: Callbacks will layout its children.
  <bkardell> dbaron that's mostly what we discussed at edge,
  <bkardell> Seems good.
  TabAtkins: new display:--whatever would work.
  dbaron: Mixing custom and built-in layout gets complex. A new
          display type should avoid that.
  shane: Restrictions should make this more implementable.

  Rossen: Declaring an element to have custom layout should take
          over built-in layout.
  dbaron: Descendants that use built-in layouts should be OK.
  dbaron: But don't let that happen with something like table
          descendants that make anonymous boxes.
  <dbaron> Just that using a different 'display' value is useful
           because we have anonymous box construction rules that
           deal with things like avoiding invalid combinations like
           a display:table-row child of something arbitrary.

  roc: What about fragmentation? Do we need to support braking, etc?
  shane: Yes.
  TabAtkins: We need to support printing, etc.
  roc: Interoperable fragmentation/breaking API will be hard.
  shane: We can make that call later.
  roc: We could have a simpler API that doesn't provide for breaking
       and a more complex API that does; that might be better for
       authors anyway.

  plinss: We should provide for breaking, doing so later may get
          harder to add.
  SteveZ: There's a use case for breaking: eg. MathML and text.
  plinss: We break this out into smaller phases.
  plinss: Don't put everything in one call.
  plinss: For example, have a phase for sizing.
  shane: Also, don't make engine-specific assumptions.

  Rossen: How do you prevent hangs? Infinite loops?
  Rossen: Custom layouts often create cycles.
  Rossen: How to prevent that?
  Rossen: Can't keep it to one iteration. eg 13 levels of recursion?
  iank: Constrain things within phases?
  roc: Are we OK with constraining to basic CSS (intrinsic width,
       computed width, height calculations from children, etc.)
  <bkardell> Seems ok - at least for now seems ok.
  Rossen: Flexbox and grid already exceed the basic CSS box
          calculation.
  Rossen: We shouldn't create cycles, but flexbox gets that way
          easily with custom layouts.
  Rossen: Should we treat custom layout as fixed size?
  iank: API will be used by framework developers.
  roc: We need to prevent lockups even from iframe usage,
  roc: Such as Facebook like buttons.

  Rossen: Proposed resolution: greg, iank TabAtkins, shane, Rossen,
          roc to start work on Custom CSS Layout.
  krit: need use cases before API discussions

  RESOLVED: gregwitworth, iank, TabAtkins, shane, roc, Rossen start
            work on CSS Layout Module

  [15 minute break]
Received on Tuesday, 3 March 2015 19:25:11 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 19:53:22 UTC