WebKit Core Projects

This list is for WebKit work that is about improving performance, infrastructure and/or the maintainability of the codebase. This is a list of high-priority projects that have minimal dependence on web standards bodies and on buy-in from other WebKit ports.

If you decide to take on one of these projects put your name at the beginning like [ ojan ] and move the line to the "Completed Projects" section once the project is done. Some of the bigger projects could use multiple people working on them.

Talk to Ojan if you have questions about any of these or want to add more things to this list. If you discover other new projects that would fit well here we should add them!

WebCore projects

Split RenderBlock into two classes. Create a base class for a box that is a container of other boxes and is the appropriate superclass for RenderBlock, RenderFlexibleBox, RenderGrid, RenderTable, RenderDeprecatedFlexibleBox. The new RenderBlock will just be about dealing with laying out lineboxes and inlines.

Kill RenderArena. In the past, this has come with a 1-2% performance regression on page cycler tests. The hard part is finding a way to do it without a significant perf hit. A good starter bug here would be https://bugs.webkit.org/show_bug.cgi?id=87523.

Add COMPILE_ASSERTS for the size of all object with bitfields and all other objects that we create a lot of.
Add histograms for measuring the cost of doing a second layout when we go from not needing a scrollbar to needing a scrollbar. Maybe we can invert the heuristic to assume that we will need a scrollbar and pay the second layout cost for cases where we don’t need one.
Improve DOM timer throttling to match Apple’s: https://bugs.webkit.org/show_bug.cgi?id=98474#c48
Make appendChild O(1) when possible

Instrument RenderStyle to see which CSS properties cause us to create StyleRare* classes. We might be able to get memory improvements by moving around what is on RenderStyle and what is on a StyleRare* class.
Address the vertical writing mode FIXMEs in core rendering classes (e.g. in RenderBox/RenderBlock). Note: We currently don’t support vertical writing mode on some elements by design. See “-webkit-writing-mode: horizontal-tb !important;” in http://trac.webkit.org/browser/trunk/Source/WebCore/css/html.css. See https://docs.google.com/a/chromium.org/document/d/1I3RfUbuVVoMB4JY3o6qypiEUv7Y7zfEGmG3V78Fa5Ok/edit for a description of vertical writing modes in WebCore.
Convert shadowDom form controls to use the new flexboxes instead of percentage heights + custom layout code. See http://trac.webkit.org/changeset/131367 as an example.
Convert any internal uses of the old flexboxes to the new flexboxes.
Fix hitTesting on RenderLayers to not be O(n^2): https://bugs.webkit.org/show_bug.cgi?id=99669

Identify and fix scrolling performance problems on this page: http://crbug.com/157016
Setting scrollTop / scrollLeft should not cause a layout: https://bugs.webkit.org/show_bug.cgi?id=75015
potential performance improvement when moving nodes within the DOM: https://bugs.webkit.org/show_bug.cgi?id=102992
Fixing the "don't-measure-text-twice" 10% line-layout win for Chromium wkb.ug/100794
Pave the way for removing the Widget tree by making a parallel "widget-less" iframe/scrollbar/plugin codepath wkb.ug/101094
Moving widget-loading (iframes/plugins) out of the rendering tree wkb.ug/45049
Design a proper architecture for repainting that doesn't involve 5 different ways to do it. In particular, solve the issue of multi-pass layouts that leads to over-repainting (see https://bugs.webkit.org/show_bug.cgi?id=92800).
Find a better architecture for applying an operation to a group of RenderObjects instead of stashing it in RenderLayer. RenderLayer is the bottleneck in several operations (see hit testing above and https://bugs.webkit.org/show_bug.cgi?id=92258)
Move more of RenderObject down into subclasses
Make most of the methods on RenderObject protected/private.
Come up with a naming scheme so that only Renderers have “Render” in their name.
Make RenderMarquee a real renderer and not something that hooks onto RenderLayer (patch attached to https://bugs.webkit.org/show_bug.cgi?id=94754, blocked on proper testing for <marquee>)
Add some pre / post hooks to layout, remove ANY modification of the tree during layout.
Refactor reflection handling to avoid the current architecture where we have a synthetic RenderObject (RenderReplica) while most of the logic is in RenderLayer
Add more invariants during layout to ensure proper behavior (see https://bugs.webkit.org/show_bug.cgi?id=49019 and https://bugs.webkit.org/show_bug.cgi?id=64222)

Performance

[META] RoboHornet (and RoboHornet Pro) related performance issues: https://bugs.webkit.org/show_bug.cgi?id=98798

- tag-based collections are invalidated on every add/remove of an element (25%+ win for RoboHornetPro) wkb.ug/98823

Improve performance of Dromaeo and dom_perf bencharks. For example, <https://bugs.webkit.org/show_bug.cgi?id=99197> makes JSC go 40% faster on dom-traverse. Can we do something similar for V8?
Should report private memory on the perf bots: http://crbug.com/163280
Improve memory usage of the following benchmarks:

Create (and then optimize) benchmarks that accurately represent memory usage in gmail, docs, etc.
Improve performance test suites:

first and second tabs of dromaeo benchmarks show the same data: http://crbug.com/160321
page cyclers should have subgraphs for each individual site: http://crbug.com/161837
add performance trybots: http://crbug.com/120380
Linux lowmem perf bot exhibits bimodal behavior: http://crbug.com/159077
Increased variance in moz total_byte_b around range 163625:163631: http://crbug.com/160529
explore ways to reduce performance test variance: http://crbug.com/161844

Infrastructure

Add more debugging tools to content_shell, e.g. apple’s debugging menu has a bunch of handy things like a live view of the rendertree.
Switch chromium windows bots back to using Apache: https://bugs.webkit.org/show_bug.cgi?id=101373
Improve garden-o-matic:

https://docs.google.com/a/google.com/document/d/1laZ9C6b2J1D55e1aiF7fdJZKYzSB0_VjrxcooTvpPMA/edit
webkit-patch rebaseline does the wrong thing: https://bugs.webkit.org/show_bug.cgi?id=97045
garden-o-matic should let you minimize failures so they're not in your face: https://bugs.webkit.org/show_bug.cgi?id=97949
garden-o-matic should refuse rebaseline for reftests: https://bugs.webkit.org/show_bug.cgi?id=90765

garden-o-matic should show the last revision process by every layout test bot and the last by every non-layout test bot: https://bugs.webkit.org/show_bug.cgi?id=98088
garden-o-matic should delete stale pngs if the bot doesn't generate pngs: https://bugs.webkit.org/show_bug.cgi?id=99219
garden-o-matic shouldn’t hard-code a list of buildbots for build.webkit.org: https://bugs.webkit.org/show_bug.cgi?id=106259

Add a tool to garden-o-matic/flakiness dashboard/code review tool to make it possible to easily view the diff of an expected result to another platform's expected file (e.g. diff between platform/chromium/foo-expected.txt and platform/mac/foo-expected.txt).

Rebaseline tests from the results.html page: https://bugs.webkit.org/show_bug.cgi?id=86797

prepare-Changelog should support updating the list of changed files: https://bugs.webkit.org/show_bug.cgi?id=74358
webkit-patch optimize-baselines thinks it's succeeding, but it's not: https://bugs.webkit.org/show_bug.cgi?id=98215
webkit-patch rebaseline-expectations spits out tons of lint warnings: https://bugs.webkit.org/show_bug.cgi?id=106235

Flakiness dashboard

https://code.google.com/p/chromium/issues/detail?id=224008

Flakiness Dashboard server OOMs when the results.json gets too large: https://bugs.webkit.org/show_bug.cgi?id=97643
Stop having run-webkit-tests generate both incremental_results.json and full_results.json https://bugs.webkit.org/show_bug.cgi?id=83222

Like garden-o-matic, allow you to create a local flakiness dashboard instance that updates your local TestExpectations file with data from the dashboard (the UI for this is mostly done, it just needs to be hooked up to the right garden-o-matic python calls). See http://test-results.appspot.com/dashboards/flakiness_dashboard.html#expectationsUpdate=true.
Once ^^^^^ is done, create a bot like sheriffbot, that can make the changes to TestExpectations for you and post a patch to bugs.webkit.org.
Change run-length encoding to use a single string instead of an array of arrays
Cleanup flakiness_dashboard.js and dashboard_base.js to be more object-oriented and better unittested.
Test results server cleanups

stop stripping jsonp callback now that we only store json
start requiring master name and testtype to serve file content

generate builders.js on the server
Annotate the failing test results shown in the flakiness dashboard with the revision it failed at. Something like: DIFF: Webkit Win (deps)(dbg)(2), r12345
Audio failures don't show up correctly on the flakiness dashboard: https://bugs.webkit.org/show_bug.cgi?id=104632
Stupid type error when uploading results to the test results server json: https://bugs.webkit.org/show_bug.cgi?id=105094

https://bugs.webkit.org/buglist.cgi?quicksearch=ToolsHitList

LayoutTests

Create infrastructure for red-green layout tests. They have no expected result. If the bitmap has only green and white pixels, it passes, otherwise it fails.
add NeedsRebaseline keyword to TestExpectations as a way to hande updating pixel tests: https://bugs.webkit.org/show_bug.cgi?id=100415
Make the text caret not blink in tests: https://bugs.webkit.org/show_bug.cgi?id=97558
using js-test-pre.js should only require a DOCTYPE and js-test-pre.js: https://bugs.webkit.org/show_bug.cgi?id=48344
Hundreds of mac failures are due to a slightly different color repaint background: https://bugs.webkit.org/show_bug.cgi?id=83400
Tests sometimes report missing expectations (probably a real webkit or drt bug) https://bugs.webkit.org/show_bug.cgi?id=72039 and https://bugs.webkit.org/show_bug.cgi?id=63057

Completed projects

Comments

The Chromium Projects

Quick links

Other sites