Paint team OKRs

Q1 2017


Code health #wp-predictability
Owner: paint-dev
  • P1# of open bugs did not increase during the quarter
  • Owner: schenney
  • Starting point: 1350 on Jan 3, 2017
  • P2Reduce # of failing layout tests, layout test fix SLA
  • Owner: schenney
  • Starting point: 58 on Jan 3, 2017
  • P1No regressions reach the M57 or later stable channel
  • Owner: schenney
  • Should this be M56 and later, given it has not gone to stable yet?
Improve quality of implementation #wp-predictability #wp-architecture
Understand, maintain and improve performance #wp-performance
Owner: paint-dev
  • P1Measure performance and prevent any performance regressions from reaching the stable channel
  • Owner: wkorman
Improve paint/compositing interfaces #wp-performance #wp-health #wp-architecture
  • P1SPv2 works for all effects, including mask
  • Owner: trchen
  • P1SPv2 works for scrolling and scrollbars
  • Owner: pdr
  • P1SPv2 works for animation metadata and touch event regions
  • Owner: wkorman
  • P1PaintArtifactCompositor layerization as good as PaintLayerCompositor
  • Owners: chrishtrwangxianzhu
  • P1Finish removing target knowledge from clip and transform trees
  • Owners: ajumajaydasika
  • P1Cache dynamically-determined clips
  • Owner: jaydasika
  • P1Stop requiring a cc::Layer for any property tree nodes
  • P2Build the RenderSurfaceLayerList from effect trees
  • Owner: ajuma
  • P1Launch SlimmingPaintInvalidation
  • Owner: wangxianzhu
  • P1Design and prototype implementation of border-radius clipping in SPv2
  • Owners: ajumachrishtr
  • Stretch additional goal: implement & land it
Support other teams' work
  • Promote more opaque scrollers and fixed-position elements in the presence of LCD text
  • Owners: paint-devthreaded-rendering
  • threaded-rendering: implement
    paint-dev: support

    Fix discovered compositing bugs, re-enable fixed position promotion. Promote scrollers with border-radius, box shadow, and scrollers/sticky/fixed elements containing no text.
  • P1Apply border radius clipping to non-stacking descendants
  • Owners: paint-devthreaded-rendering
  • crbug.com/157218

    paint-dev: implement feature
    threaded-rendering: implement tiled masks for efficiency
  • P2Implement color correction in blink
  • Owners: chrome-gpupaint-dev
  • ccameron: implementation

    paint-dev: support
  • P3Fully threaded scrolling with touch-action
  • Owners: input-dev-eventingpaint-devwkorman
  • paint-dev, input-dev: design for touch-action in SPv2
  • P3Simplify paint usage of Sk*Shaders
  • Owners: chrome-gpupaint-dev
  • This is to support easier implementation of custom display lists
Code Health and Future Design
Support other teams' work
  • Promote more opaque scrollers and fixed-position elements in the presence of LCD text
  • Owners: paint-devthreaded-rendering
  • threaded-rendering: implement
    paint-dev: support

    Fix discovered compositing bugs, re-enable fixed position promotion. Promote scrollers with border-radius, box shadow, and scrollers/sticky/fixed elements containing no text.
  • P1Apply border radius clipping to non-stacking descendants
  • Owners: paint-devthreaded-rendering
  • crbug.com/157218

    paint-dev: implement feature
    threaded-rendering: implement tiled masks for efficiency
  • P2Implement color correction in blink
  • Owners: chrome-gpupaint-dev
  • ccameron: implementation

    paint-dev: support
  • P3Fully threaded scrolling with touch-action
  • Owners: input-dev-eventingpaint-devwkorman
  • paint-dev, input-dev: design for touch-action in SPv2
  • P3Simplify paint usage of Sk*Shaders
  • Owners: chrome-gpupaint-dev
  • This is to support easier implementation of custom display lists


Q4 2016


Code health #wp-health
Owner: paint-dev
  • 1.00P1# of open bugs did not increase during the quarter
  • Owners: schenneyxidachen
  • Mid-quarter update: on track. 1383 -> 1385, including addition of a new component.
    End of quarter: 1383 on Sept 29, 2016 to 1350 on Jan 3, 2017.
  • 0.70P1No regressions reach the M55 or later stable channel
  • Owner: schenney
  • Still on track.

    End of quarter update:
    2 Regression reached M54, not really part of OKR: https://bugs.chromium.org/p/chromium/issues/detail?id=655941https://bugs.chromium.org/p/chromium/issues/detail?id=662378
    One regressions reached M55 specifically. 
    M56 not to stable yet.
  • 0.10P2Reduce # of failing layout tests, layout test fix SLA
  • Owners: schenneyxidachen
  • 1. Reduce # of steady-state failing layout tests by 50%
    2. Fix all new layout test regressions within 2 weeks

    Directories:
    LayoutTests/paint/...
    LayoutTests/compositing/... (except compositing/animations)
    LayoutTests/svg/...
    LayoutTests/hittesting/...
    LayoutTests/fast/images/...
    LayoutTests/css3/images/...
    LayoutTests/canvas/...

    Mid-quarter-update: Negligible overall reduction this quarter. TODO(schenney) check lifetime of new regressions
    End of Quarter: No significant reduction, total count increased significantly due to random order test flakiness. Regression tracking process failed.

    Non-zero score because we did go down except for factors beyond our control. 35 -> 34 before random-order jump.
HistoryCopyLink
0.56P2Improve quality of implementation #wp-predictability #wp-health
Owner: paint-dev
  • 0.56P2Top-starred bugs fixed
  • Owner: chrishtr
  • Fix at least these: 521364, 470669, 608874, 157218, 618324, 226753, 652448, 468497, 622336

    Mid-quarter update: some progress, but needs work. 468497 is done. Review for 470669 currently in review. 521364 still in progress. 157218 in progress, close to patches landing. 652448 in progress, vmpstr doing good work to support.

    End of quarter:
    * 521364: CLs in review
    * 470669 fixed, bug blocked on fix for crbug.com/21798.
    * 608874 no progress
    * 157218 landed Dec 12, reverted Dec 13, re-landed January 5, perf issues and 2 bugs still open.
    * 618324 no progress
    * 226753 investigated, no longer repros on later OS X updates. Probably not going to fix.
    * 652448 prototyped, but concluded design proposal is not feasible in SPv1
    * 468497 fixed
    * 622336 no progress

    Fixed/close to fixed: 3
    mostly there: 2
    WontFix: 1
    No progress: 3

    Total score: (1.0* 3 + 0.75 * 2 + 0 * 3) / 8 = 0.56
  • 1.00P2SVG filter hack in FrameView is gone
  • Owner: fs@opera.com

    Mid-quarter: on track! Code in review.
  • 0.00P2Remove need for FrameView::forceLayoutParentViewIfNeeded in SVG
  • Owner: fs@opera.com

    Mid-quarter: ?
  • 0.00P2Widget hierarchy is gone
  • Owners: szagerwkorman
  • crbug.com/637460

    Mid-quarter: not yet started.
    EOQ: no progress
1.00P1Understand, maintain and improve performance #wp-performance
Owner: paint-dev
  • 1.00P1Measure performance and prevent any performance regressions from reaching the stable channel
  • Owner: wkorman
  • Mid-quarter update: on track. Only thing being tracked is the Linux regression on M54 stable.

    EOQ: still going well
  • 1.00P1Composited opaque scrolling for low-DPI screens
  • Owner: schenney
  • Mid-quarter update: basically done. Launching with M56.

    End of quarter: Landed for M56. Paint update time drops evident in UMA data. No outstanding open issues.
0.58P2Improve paint/compositing interfaces #wp-performance #wp-health #wp-architecture
Owner: paint-dev
  • 0.25P1SPv2 works all effects, including mask
  • Owner: trchen
  • Mid-quarter update: needs more work. implemented for filters. Masks not done.
    Overall, also blocked on creating render surfaces in SPv2.

    EOQ: some progress, but still a ways to go. Some design thought about masks.
  • 0.00P1SPv2 works for scrolling and scrollbars
  • Owner: pdr
  • Mid-quarter update: needs attention. no progress this quarter.

    EOQ: no progress.
  • 0.50P1SPv2 works for animation metadata and touch event regions
  • Owner: wkorman
  • Mid-quarter update: needs attention. no progress this quarter.

    EOQ: design investigation, early implementation.
  • 1.00SPv2 compatible ancestor-dependent and descendant-dependent PaintLayer values
  • Owner: chrishtr
  • Mid-quarter update: needs attention. no progress this quarter.

    EOQ: done! Design and implementation in place.
  • 0.75P2SPv2 paint invalidation test infrastructure
  • Owner: chrishtr
  • Mid-quarter update: landed changes to unify paint invalidation output format between SPv1 and SPv2. Tests not converted yet.

    EOQ: done! Design and implementation in place. Tests not converted yet, but that wasn't in the KR description.
  • 0.60P1PaintArtifactCompositor property tree optimization
  • Owner: chrishtr
  • Passes all compositing tests with equivalently performant layerization

    Mid-quarter update: needs attention. no progress this quarter.

    EOQ: designed implemented the optimizations. But not all compositing tests passing yet.
0.46P1Finish cc property tree implementation #wp-performance #wp-health
  • 0.75P1Finish removing target knowledge from clip and transform trees
  • Owner: jaydasika
  • Mid-quarter update: transforms done! Calculating clips done (behind a flag), just needs caching implementation.'

    EOQ: transforms done, clips getting close.
  • 1.00P1Cache dynamically-determined transforms
  • Owners: ajumasunxd
  • Mid-quarter update: done! TODO(ajuma): check if property update UMA metrics moved due to this.''

    EOQ done.
  • 0.50P1Stop requiring a cc::Layer for any property tree nodes
  • Owners: ajumajaydasika
  • Mid-quarter update: on track. some progress, but work to do.

    EOQ: not done. still several references.
  • 0.00P1Build the RenderSurfaceLayerList from effect trees
  • Owner: ajuma
  • Mid-quarter update: at risk. no progress yet, still consumed with other work.

    EOQ: no progress
  • 0.50Cache dynamically-determined clips
  • Owner: jaydasika
  • Mid-quarter update: no progress yet, work starting soon.

    EOQ: implementation getting close
  • 0.00P2Simplify transform property nodes
  • Owner: ajuma
  • 1. Only one transform per node
    2. No caching (tracked via "Cache dynamically-determined transforms" however)

    Mid-quarter update: no progress.

    EOQ: no progress
P2New features
Owner: paint-dev
  • 0.10P2Drive tentative standards agreement on a raster scale spec #wp-ergonomics
  • Raster scale defines the scale at which to raster composited layers

    Note: crbug.com/652448 tracks doing this automatically for background images, and is tracked in another
    KR.

    Mid-quarter update: no progress.

    EOQ: prototyping of automatic version for images didn't work out. Going to revert code. No progress
    any standards discussion.
  • 0.00P2Fix all bugs blocking backdrop-filter launch #wp-predictability
  • Owner: jaydasika
  • All dependent bugs of crbug.com/497522 fixed.

    Mid-quarter update: no progress so far.

    EOQ update: no progress.
0.93P1Improve paint invalidation correctness and performance #wp-health #wp-performance
  • Mid-quarter update: on track. lots of progress simplifying the code and fixing bugs.
  • 0.75P1Launch slimmingPaintInvalidation
  • Owner: wangxianzhu
  • Mid-quarter update: on track. Close to a complete implementation.

    EOQ: launched to experimental status
  • 1.00P1Implement incremental paint property tree update
  • Owners: pdrwangxianzhu
  • Mid-quarter update: on track. Code mostly done and design complete.

    EOQ: done!
  • 1.00P1Implement all SPv1 invalidation features in slimmingPaintInvalidation mode, or deprecate them if obsolete
  • Owner: wangxianzhu
  • Mid-quarter update: on track. Lots of progress.

    EOQ: done!



Q3 2016

Code health #wp-health
Owners: junovpaint-devschenney
  • 1.00P2# of open bugs did not increase during the quarter
  • Owners: junovschenney
  • Baseline starting point: 1323 open bugs as of July 8 2016 for these labels:
    Blink>Paint, Blink>Paint>Invalidation, Blink>Compositing, Blink>SVG, Blink>Canvas, Blink>Image

    Mid-quarter update: on track. Bugs down to 1295 (2.1% decrease)

    EOQ update: 1311 as of 9/29 (plus 54 for the new HitTesting component)
  • 1.00P17-day SLA for triaging all incoming bugs
  • Owners: chrishtrjunovschenney
  • Another owner: fs@opera.com


    Mid-quarter update: on track. Holding around zero.

    EOQ update: met goal.
  • 0.75P1No regressions reach the M52 or later stable channel.
  • Owner: schenney
  • Definition: bug is a regression that ends up bisected to a commit after the M52 branch point, yet made its way
    to the stable channel.

    Mid-quarter update: at risk. We already have one; crbug.com/635724 might leak to M53 stable.

    EOQ update: crbug.com/646363 made it to stable M53. crbug.com/637556 made it to M52 stable. No bug reached stable due to a triage failure. Problem was late reporting of the bug.
  • 1.00P3Monitor # of failing layout tests
  • Owner: schenney
  • Directories: fast/repaint/*, paint/*, compositing/*, svg/*, fast/canvas/*, canvas/*

    Mid-quarter update: on track.

    EOQ update: met
  • 1.00P2Document an official triage process
  • Owners: chrishtrschenney
  • Mid-quarter update: on track.

    EOQ update: met
0.40Improve quality of implementation #wp-predictability #wp-health
Owner: paint-dev
  • 0.40P1Top-starred bugs fixed
  • Owner: paint-dev
  • Fix at least these: 521364, 470669, 608874, 157218, 618324, 226753, 409868, 470608

    Update: at risk. Only one of these bugs is fixed so far.

    EOQ update:

    - 521364 is close to done
    - 470669 unblocked by dependent bug a few days ago
    - 608874 no progress
    - 157218 has a prototype fix
    - 618324 not clear how to fix yet
    - 226753 no progress
    - 409868 WontFixed
    - 470608 Fixed.

    Score: 2/8 done, 2/8 close to done, 4/8 not done: 2/8 * 1 + 2/8 * 0.75 = 0.43 ~ 0.4
  • 0.60P2SVG filter hack in FrameView is gone
  • Owner: paint-dev
  • Ref: Document::hasSVGFilterElementsRequiringLayerUpdate

    Owner: fs@opera.com

    Update: at risk. More work upcoming.

    EOQ update: patch to finish it coming very soon; some good preparatory refactoring done.
  • 0.20P2Remove need for FrameView::forceLayoutParentViewIfNeeded in SVG
  • Owner: paint-dev
  • Owner: fs@opera.com

    Update: at risk. trchen is also helping with this one.

    EOQ update: some promising prototyping, but a ways t go.
  • 0.40P2Geometry code for layout and paint better-written and documented
  • Owners: blink-layoutszagerwkorman
  • #wp-predictability #wp-health

    Shared between the paint and layout teams.

    Update: at risk. Plan in place for removing the Widget hierarchy, but not much progress yet.

    EOQ update: reworked flipped-block logic in mapToVisualRectInAncestorSpace(); initial exploration to rework Widget/FrameView trees; documentation of writing modes in layout/README.md in flight.
0.97Understand, maintain and improve performance #wp-performance
Owner: paint-dev
  • 1.00P1Measure performance and prevent any performance regressions from reaching the stable channel
  • Owner: wkorman
  • Mid-quarter update: on track.

    EOQ: met
  • 1.00P2Composited scrolling for low-DPI screens
  • Owner: schenney
  • Mid-quarter update: on track. Lots of progress. Should be able to turn it on for M55.

    EOQ: met
  • 1.00P1Use compositor RTrees rather than cached SkPictures
  • Owner: wkorman
  • Mid-quarter update: done!

    EOQ: met
  • 0.80Partial repaint telementry test implemented
  • Mid-quarter update: at risk. No progress yet, work upcoming this quarter.

    EOQ: in code review.
0.67Improve paint/compositing interfaces #wp-performance #wp-health
Owner: paint-dev
  • 0.40P1New (SPv2) compositing integration works for scrolling and non-mask effects
  • Owners: chrishtrpdrtrchen
  • All scrolling (assuming root layer scrolls) and compositing tests pass up to masking.

    Mid-quarter update: on track for scrolling, at risk for effects.

    EOQ update: scrolling almost done. No progress yet on effects.
  • 1.00P1Paint invalidation tree walk within the pre-paint tree walk fully implemented
  • Owners: chrishtrwangxianzhu
  • Mid-quarter update: done!

    EOQ update: met
  • 0.60P1GeometryMapper implements fast paint invalidation, paint clip rects, and interest rects.
  • Owner: chrishtr
  • Mid-quarter update: on track. Lots of progress.

    EOQ update: done for paint invalidation, CL almost done for paint clip rects. Interest rests not yet done.
0.58Finish cc property tree implementation #wp-performance #wp-health
Owner: paint-dev
  • 0.90P1Finish removing target knowledge from clip and transform trees
  • Owners: jaydasikaweiliangc
  • Mid-quarter update: On track for transforms, at risk for clips. Computation of transforms without in-tree target knowledge has been implemented behind a flag; for clips, there are unit test failures to sort out (where the "correct" answer isn't necessarily what the test currently expects), and work is being transitioned to new owners.

    EOQ update: done for transforms (1.0), nearly done for clips (0.8), so 0.9 overall.
  • 0.40P1Cache dynamically-determined draw properties
  • Owner: sunxd
  • Mid-quarter update: On track for transforms, at risk for clips. Currently evaluating performance for transforms, but clips are blocked on the previous KR.

    EOQ update: nearly done for transforms (0.8) and not started for clips (0) so 0.4 overall.
  • 0.00P3(stretch) RenderSurface determination on the impl thread
  • This depends on the previous two KRs being finished, but is not a lot of additional effort.

    Mid-quarter update: no progress.

    EOQ update: not done

    EOQ update: ?
  • 0.60P1Stop requiring a cc::Layer for any property tree nodes
  • Owners: ajumajaydasika
  • Mid-quarter update: On track. Most of the dependencies have been removed.

    EOQ update: Work for routing animation updates directly to property trees is nearly done; most of the direct dependencies of render surfaces on owning layers have been removed, but RSLL iteration still needs to be converted to use layer lists and effect nodes.
1.00New features
  • 1.00P2Support custom paint #wp-future
  • Code reviews, design reviews, etc.

    Mid-quarter update: on track. (nothing to do)

    EOQ update: met
  • 1.00P2Unprefix -webkit-clip-path #wp-predictability
  • owner:fs@opera.com

    Mid-quarter update: on track, code complete already. Sending an intent to ship soon.

    EOQ update: met
Code health #wp-health
Owner: blink-layout

Q2 2016

Code health
Owner: paint-dev
  • 0.50P2Reduce total open bugs by 2%
  • Owners: chrishtr, junov, schenney
  • Mid-quarter update: actual result is up by 1%. Wont Meet.

    EOQ grading notes: actual is up by 1.8%. Score=0.6 for doing ok at fixing bugs, but lots more bugs keep coming in via triage from other labels.
  • 0.90P1Triage all legacy bugs, and triage all new bugs within a 1 week SLA. Enforce priority 1 meaning.
  • Mid-quarter update: triaging SLA is On Track. Priority 1 bug hygiene Needs Work.

    EOQ grading notes: we met the SLA almost every week. Blink>Compositing in particular slipped a couple of times.
  • 0.70P1Prevent any regressions from reaching the M51 or above stable channel
  • Mid-quarter update: M51 just went to stable, so no direct evidence yet of success or failure. Therefore On Track.

    EOQ notes: still not sure about M51. needs some bug digging. Tentative score of 0.7
  • 0.66P2High-priority / most-starred bugs fixed
  • Fix at least these:

    http://crbug.com/596382
    http://crbug.com/521364
    http://crbug.com/470669

    Mid-quarter update: 596382 is fixed (with followup to re-enable safely). The other two bugs have some progress
    but Need Work.

    EOQ update: 596382 is fixed. 521364 is halfway done. 470669 is halfway done. Final score: 0.66 (averaging).
0.40P1Measure performance
Owner: paint-dev
  • 0.00P1Implement a partial repaint Telemetry test
  • http://crbug.com/527189

    Mid-quarter update: no progress. Needs Work.

    EOQ update: nothing done.
  • 0.80P1Monitor UMA stats for paint, raster, paint invalidation and compositing. Don't allow regressions into the stable channel.
  • Owner: wkorman
  • Mid-quarter update: monitoring is On Track, not allowing regressions Needs Work.

    EOQ update: Stats monitored on regular basis, no currently known stable channel regressions M51 (may need more investigation to be 100% sure?)
1.00Performance and complexity improvements
Owner: paint-dev
  • 1.00P2Re-implement reflection using filters
  • Owner: jbroman
  • Launch by end of quarter

    Mid-quarter update: implemented and pretty much working. On Track.

    EOQ update: done, in M53. Score 1.0.
  • 0.70P2Composite scrolling on low-DPI screens when possible
  • Owner: schenney
  • Launch by end of quarter. Tracked in crbug.com/381840

    Mid-quarter update: partially implemented. Need to fix bugs in current implementation and commit it. Needs Work.

    EOQ update: All the necessary work done to set the layer as composited, and resulting Blink bugs fixed, but blocked on enabling background and foreground painting into the same composited layer when the background is opaque in order to actually get LCD text on the layer.
  • 1.00P1New DOM-order-based paint invalidation tree walk
  • Owner: wangxianzhu
  • Launch by end of quarter

    Mid-quarter update: finished on current PaintInvalidationState. Needs work to migrate to PrePaintTreeWalk (PaintPropertyTreeBuilder). On Track.

    EOQ update: Done. Score 1.0.
P1Property trees / SPv2 plumbing
Owners: paint-dev, pdr, trchen
  • 0.50P1Property trees in Blink are complete up to the deficiencies of the paint artifact compositor.
  • Owners: pdr, trchen
  • Transform trees are complete (including backface visibility, flattening, and transform-style).
    Effect trees are complete (including reflection, clip-path, mask).
    Root layer scrolling supported using unittests.
    All of compositing/, fast/, svg/ passing or failing for known&labeled reasons like paint artifact compositor deficiencies.
    This includes multicolumn.
    Implement the scroll tree.

    Mid-quarter update: Needs Work. Some progress on transform tree semantics and SVG clip paint properties, but a lot more work to do.

    EOQ update: lots of measurable progress was made (transform edge-cases, filter tree preparation, html and svg difficult cases) but we still fail a significant number of tests (~3100/16728 of fast/ & compositing/ & svg/).
  • 1.00P1All direct dependencies in cc on the cc::LayerImpl tree are gone, replaced with property trees
  • Owners: jaydasika, sunxd, vollick
  • Mid-quarter update: On Track. Lots of great progress.

    EOQ update: Done!
  • 0.30P2RenderSurface determination is on the impl thread
  • Owners: vollick, weiliangc
  • Update render target info and draw properties after render surface determination.

    Mid-quarter update: Partial work. Rest blocked by target information still on clip/transform tree. Wont Meet

    EOQ update: blocked on other KRs, but also not done.
  • 0.60P1Transformed layers can be sent as a layer list with property trees, rather than a hierarchical layer tree
  • Owners: ajuma, jbroman
  • Mid-quarter update: progressing, with a partial implementation. On Track.

    EOQ update: blink-side: score 0.5 can do trivial translations with a layer list (and trivial trees), but I'm still working on building non-trivial ones. cc side: 1.0. Total: 0.5 (most work is in Blink)
  • 0.40P2SPv2 paint invalidation is in place and works behind the feature flag for SPv2
  • Owner: wangxianzhu
  • Mid-quarter update: no progress yet. Needs Work.

    EOQ update: Plumbed SPv2 paint invalidation onto spv1 paint invalidation and it basically works, but it's still far from the goal of spv2 paint invalidation.
  • 0.50P1Remove target knowledge from Clip tree and Transform tree
  • Owner: weiliangc
  • Remove target_id and *_in_target_space in Clip and Transform tree.
    Remove main thread cc's use to to_target transform, use to_screen transform instead (including visible rects calculation, sublayer scale, snapping)
    Dynamically calculate draw properties with target info from Effect tree.

    Blocks both Blink/cc API and RenderSurface determination on impl thread.

    Mid-quarter update: Progressing. Needs Work.

    EOQ update: progressing, but still needs plenty of work.
  • 0.20P3Cache dynamically calculated draw properties
  • Owner: weiliangc
  • Blocked by Clip and Transform tree clean up and Render Surface determination on impl thread.
    Critical for performance, do not block API implementation.

    Mid-quarter update: Still blocked by dynamic calculation. Wont Meet.

    EOQ update: Only moved a small part of properties.
1.00P2Support goals of other teams
  • 1.00Support custom paint
  • Mid-quarter update: On Track. All custom paint patches are committed.

    EOQ update: Done. Score 1.0.
  • 1.00Support Blink -> Chromium style change, and Onion Soup cleanups
  • Mid-quarter update: On Track. We're reviewing everything coming our way.

    EOQ update: Done. Score 1.0.
0.40P1Geometry code improvements for performance, less redundancy, simplicity and correctness
Owner: paint-dev
  • 0.70P1RTrees used to optimize layer rasterization, instead of SkPicture replay
  • Owners: chrishtr, wkorman
  • Mid-quarter update: down to about 12 failing layout tests, plus a small unknown regression in raster time. On Track.

    EOQ update: Close to landing for M53, work still underway on < 10 failing layout tests.
  • 0.40P2GeometryMapper implemented and used to implement features in Blink
  • Owners: chrishtr, wkorman
  • 1. SPv2-compatible
    2. PaintLayerClipper

    Mid-quarter update: prototype partial implementation of GeometryMapper in progress. Needs Work.

    EOQ update: First patch committed. Still plenty to do. Score 0.4.
  • 1.00P1Partial raster for Blink launched
  • Mid-quarter update: done, On Track.

    EOQ update: On track. Score 1.0.
0.30Improve SVG integration into the document lifecycle
  • 0.50P2SVG filter hack in FrameView is gone
  • Ref: Document::hasSVGFilterElementsRequiringLayerUpdate
    Owner: fs@opera.com

    Mid-quarter update: some progress. Needs Work.

    EOQ update: some progress, but not quite there.
  • 0.10P2Remove need for FrameView::forceLayoutParentViewIfNeeded in SVG
  • Owner: fs@opera.com

    Mid-quarter update: no real progress. Wont Meet.

    EOQ update: some analysis performed, but no work yett

Q1 2016

Code health
Owner: paint-dev
  • Reduce total open bugs by 5%
  • Owners: chrishtr, junov, schenney
  • Document all new features in Markdown and in the code, as we write it
  • Implement pixel capture on Cluster Telemetry
  • Owner: pdr
  • crbug.com/536285
  • Triage all legacy bugs, and triage all new bugs within a 1 week SLA
  • Owners: chrishtr, junov, schenney
Measure and improve performance
Owner: paint-dev
  • Implement a partial repaint Telemetry test
  • Owner: weiliangc
  • crbug.com/527189
  • Improve performance of paint aggregated across Cluster Telemetry by a total of 10%
  • Owner: wangxianzhu
  • Monitor existing UMA stats at the bi-weekly meeting
  • Owner: wkorman
Progress towards Slimming Paint v2
Owner: paint-dev
  • Move RenderSurface determination to the impl thread
  • Owner: weiliangc
  • Finish by M51 branch point
  • Replace all direct dependencies in cc on the cc::LayerImpl tree with property trees
  • Owners: jaydasika, sunxd
  • Finish by M51
  • Complete implementation of paint property trees
  • Owners: pdr, trchen
  • Passes most compositing/ layout tests.
    Passes multicolumn tests.
    Passes scrolling tests for the root frame.
  • Implement a correct (but not yet performance-optimized) version of a new Blink compositing integration
  • Owner: jbroman
  • Passes most compositing/ layout tests.
    Passes multicolumn tests.
    Passes scrolling tests for the root frame.
    Passes video tests.
  • Common implementation of property trees
  • Owner: chrishtr
  • Designdoc completed. Code implemented and working integration with cc (may not yet be committed)
  • Make use of RTrees to optimize layer rasterization, instead of SkPicture replay
  • Owner: wkorman
  • Implement a SPv2-compatible paint invalidation tree walk
  • (..that is integrated into the paint properties or paint tree walk)
Support goals of other teams
Owner: paint-dev
  • Support custom paint implementation
  • Owner: chrishtr
  • Support paint-related loading metrics
  • Owner: chrishtr
  • Investigate OOPIF to see if there are paint-related design problems or bugs
  • Owner: pdr
  • Support conversion of Blink to Chromium style
Code quality and cleanup
Owner: paint-dev
  • Delete CDP
  • Owner: weiliangc
  • Delete selection gap code
  • Owner: wkorman


Q4 2015

Improve bug triage and fix rate (grade: 0.46)
  • This objective apples to the following labels: Cr-Blink-Paint-Invalidation, Cr-Blink-Paint, Cr-Blink-Canvas, and Cr-Blink-Compositing.
  • Reduce open bug count by 10%
  • Triage 100% of all new bugs with a 1-week SLA
  • Note: this goes along with a web-platform-wide OKR with the same goal.
  • Triage or re-triage 50% of all bugs filed before Q3 2015
  • Note: this goes along with a web-platform-wide OKR with the same goal.
  • Define and agree upon an ongoing triage process and standards for the paint team
  • This includes instructions on how to treat priority, Type, Cr labels, etc.
Revision Historycopy
Improve documentation of paint code (grade: 0.13)
  • Migrate all feature documentation on the sites page to Markdown files
  • Add Markdown files describing clearly major architectural pieces of Paint
  • This includes forward-looking documentation of the SPv2 implementation, and important codebase-wide conventions.
Launch components of SlimmingPaintV2 (grade: 0.68)
  • All KRs below are met if they are done and launched in M49.
  • Launch SlimmingPaintSynchronizedPainting
  • Launch SlimmingPaintSubsequenceCaching
  • Launch SlimmingPaintOffsetCaching
  • Launch compositor thread property trees
  • Includes deleting all of CDP
  • Launch removal of DisplayItemList->SkPicture replay
  • Launch the cc effect tree
Implement the new compositing path for Slimming Paint v2 (grade: 0.2)
  • Implement the paint properties design
  • crbug.com/537409

    Includes: paint properties tree walk, definition and generating of property trees, and homogeneous DisplayItemList with just drawings.
  • Implement layer compositing in Blink on top of paint properties and paint chunks
  • (Stretch) Implement squashing and overlap testing in the compositor thread
canvas.toBlob, callback version of the API (grade: 0.8)
  • Have an async implementation that has its scheduling controlled by Blink
  • Refactor the png encoder integration into skia (SkCodec) in a way that allows the workload to be divided into small units of work for the purpose of Idle scheduling.
  • Stress test should show that a long-running png image encode does not interfere with a lightweight 60fps animation on a low spec device (2 core CPU).
  • (Stretch) Refactor JPEG and WebP encoder integrations into skia, and pass the same test described above with these encoders.
  • (Stretch) Ship it
ImageBitmap (grade: 0.94)
  • ImageBitmap is transferable to Workers
  • ImageBitmaps can be created on Workers
  • Calls to createImageBitmap always < 5ms for 1k-by-1k canvas on Android One
  • Async response always < 300ms for 1k-by1k canvas with no animation running (network delays notwithstanding) on Android One
  • (Stretch) Ship it
Ship Display List 2D canvas everywhere (grade: 0.7)
  • Use Display lists in cases where GPU-acceleration is available
  • Google Sheets triggering display list mode on all platforms
  • (Stretch) Google Sheets stays in display list mode while scrolling the document
OffscreenCanvas feature implementation (grade: 0.35)
  • ImageBitmapRenderingContext capable of displaying an ImageBitmap without making a copy of the image data
  • OffscreenCanvas can be created
  • (Stretch) 2D context can be created on an offscreen canvas
  • (Stretch) Functional ImageBitmap flow: OffscreenCanvas on a worker can transfer rendered content to an ImageBitmapRenderingContext on main thread
Canvas product stability (grade: 0.40)
  • Eliminate all OOM crashes caused ImageData APIs
  • Reduce by 50% 2D context losses caused by GPU context resets
Support work to start implementing custom paint (grade: 1.0)
  • Review code and designs for a new canvas rendering context interface for custom paint
  • (Stretch) Review code for implementation of custom paint
  • The Houdini team does not expect to do any work along these lines until December at the earliest.
Support work to define time-to-first paint (grade: 1.0)
  • This is an OKR to support work by the loading team to improve time-to-first-paint metrics. Tracked in crbug.com/520410
    Improve telemetry testing of paint (grade: 0.0)
    • Add a telemetry test for incremental paint
    • Tracked in crbug.com/527189.
    • Add a telemetry benchmark to capture pixels
    • This will allow us to run pixel diff tests on the top 10k sites in Cluster Telemetry. Tracked in crbug.com/536285.

    Q3 2015

    Launch compositor thread property trees (grade: 0.5)
    • Tracked in crbug.com/481585
    • CalcDrawProps not used at runtime
    • Launch in Chrome <=47
    • Stretch: Delete CalcDrawProps and fix unit test dependencies
    • There is a substantial amount of cleanup here, which we should not block launch on.
    Major progress finishing implementation of Slimming Paint phase 2 (grade: 0.24)
    • Tracked in crbug.com/471333

      Fully achieving this OKR will be a stretch!
    • New Blink->cc API complete
    • Tracked in crbug.com/481592
    • Implementation of basic layerization
    • Tracked in crbug.com/472842

      *Not* included: smart code to only re-raster or re-build only the parts of the layer tree that are needed.
    • Implement "full" display lists in Blink
    • Tracked in crbug.com/472782. Includes property trees, diff representation, pixel refs interest rects and RTrees.
    • LayoutObject-based invalidation
    Product excellence: improve bug triage and fix rate (grade: 0.6)
    • Triage all paint bugs (with label Cr-Blink-Paint or Cr-Blink-Paint-Invalidation or paint-related compositor code)
    • Reduce number of unfixed paint bugs by 10%
    Remove selection gap painting (grade: 0.40)
    • Implement representation of newlines in selection
    • Tracked in crbug.com/474759
    • Remove selection gap painting, & resulting code cleanup
    Support implementation of stopping rAF / lifecycle for non-visible, cross-domain iframes (grade: 1.0)
    • Tracked in crbug.com/487937. Primary implementation seems likely to come from scheduler-dev.
      Implement OffscreenCanvas (grade: 0.13)
      • Synchronous mode of operation (using ImageBitmap) functional for 2D contexts.
      • Ship behind experimental flag in M47
      Ship canvas filters (stretch) (grade: 0.38)
      • Can be used by Swiffy to improve efficiency of Flash->JS conversion of Flash code that uses filters.
      • No longer experimental in M47
      • Optimized to apply filter only on the region of interest, in all applicable use cases
      Refactor 2D canvas (grade: 0.9)
      • Remove >1000 lines of code by eliminating dependency on SkDeferredCanvas and simplifying ImageBufferSurface
      • Start using display lists for GPU-accelerated canvases by M47
      Ship ImageBitmap (stretch) (grade: 0.15)
      • ImageBitmap no longer experimental
      • ImageBitmap is transferable to workers
      • ImageBitmap creation is truly async
      Accelerate the use of fonts in 2D canvas (grade: 1.0)
      • When font was previously used (cache hit), setting the 'font' attribute should take < 1 microsecond on all platforms, for all use cases

      Q2 2015

      Launch Slimming Paint phase 1 (grade: 0.9)
      • Tracked in crbug.com/471332
      • Launched on all platforms, with no performance regressions
      • Makes the Chrome 45 cut.
      Working implementation of Slimming Paint phase 2 (grade: 0.0)
      • Tracked in crbug.com/471333
      • Complete implementation of display lists
      • Complete crbug.com/472782 and crbug.com/471264.
      • Implement a working version of layerization in the compositor
      • Complete crbug.com/472842
      • Passing 90% of layout tests
      Launch property trees for all cc main-thread computation (grade: 1.0)
      • M45 or die
      • Launched on all platforms, with no performance regressions
      • Complete crbug.com/386810
      Design for Custom Paint (grade: 0.8)
      • Produce design document
      • Shared with Paint team (https://easyokrs.googleplex.com/view/paint-dev/2015/q2/). This should specify the API, its integration into Blink, fully worked out use cases, and an explanation of how it will benefit performance.
      • Flesh out implementation on a branch

      Collaborators: blink-houdini, paint-dev

      Q1 2015

      Overall Score
      • Solid progress towards a well-implemented Phase 1 launch in Q2. Phase 2 deliverables were way too aggressively stated.
      0.50Implement DisplayItemList backing of existing cc::layers
      • 0.60No performance regressions relative to prior system
      • (Score comment: raster performance appears to match, but recording is still 25% slower.)
      • 0.00On by default in the Chrome Canary channel for at least the last week of the quarter
      • (Score comment: we didn't get this far.)
      • 0.90Passes all layout tests
      • (Score comment: almost all layout tests actually pass, though we don't have all of them whitelisted yet.)
      0.43Use property trees for all cc main-thread computation
      • 0.00On by default in the Chrome Canary channel for at least one week at the end of the quarter
      • (Score comment: we didn't get there yet.)
      • 0.30Finalized design of integration of property trees into display list representation
      • (Score comment: we have a design and partial implementation of transform trees.)
      • 1.00Passes all layout tests
      • (Score comment: they all pass!)
      0.00Make all composited layer decisions in cc
      • Stretch: include DisplayItemList diffs from Blink -> cc

        (Score comment: we didn't get there.)
      • 0.00Implement initial design behind a flag
      • We didn't get there.
      • 0.00Pass layout tests in the compositing/ directory
      • We didn't get there.
      1.00Consider proposals for paint callbacks and decide whether to support implementation
      • Paint callbacks are part of the "explain css" proposals that are part of Glitter.)

        (Score notes: we discussed with the Houdini team about it, and now have much better clarity about its design and justification for ergonomics and potential performance, along with a Q2 OKR to create a detailed design document for the feature.
        0.70DevTools timeline invalidation tracking ready for UI review
        • (Score comment: implemented but not yet committed.)
        Comments