Version 4 (modified by 5 years ago) ( diff ) | ,
---|
Improving Interop with WPT
by Philip Jägenstedt & Stephen McGruer
- Why does interop matter?
- Users, developers, browser engines
- Users
- Go to website on favorite browser, but broken
- Go find a browser it’s meant to work on
- All the things we do for interop make our lives easier
- Developers
- MDN ran largest survey ever
- Top pain points cross-browser compat/interop/testing
- What to do?
- A shared test suite, for all the web
- 20 commits from WebKit
- Believe it is infrastructure
- High Level View
- Chrome + Firefox + Safari
- 0/3 ~ 5K tests
- 1/3 ~ 6K tests
- 2/3 ~ 6K tests
- 3/3 ~ 20K tests
- Moving to target the browser specific failures (2/3)
- Interop is steady
- Browser specific failures in WPT
- Gap is growing
- Methodology for data for this presentation
- WPT splits into tests and sub-tests
- Sub-tests are within the test file
- Some tests with too many subtests
- Encoding tests have absurd
- Normalized to test level
- Runs on Azure pipelines
- STP runs every 3 hours, takes < 2 hours
- Edge Dev and Canary too
- Thanks Microsoft and @mustjab
- Runs on Taskcluster
- Chrome + Firefox on every commit
- WebKitGTK ally
- Thanks Mozilla and @jgraham
- Room for Improvement
- STP updates
- Reliability
- Thanks Apple and @burg
- https://github.com/web-platform-tests/wpt/projects/4
- Epiphany / WebKit
- Also run subtests on Linux for a second opinion
- WebKit on Linux
- Google partnering with Igalia
- WebKitGTK stable+nightly runs
- Fixing noise in the data
- 2-way Sync
- Script for WebKit exists
- 2-way Sync in Chromium
- Update impl+tests in single review
- Use of WPT by Chromium devs up ~220% (*j.mp/wpt-yes)
- ~55% of impl+test changes now use WPT
- Chromium & Gecko
- Automatic import
- Automatic export AFTER changes land
- WebKit
- Manual import (?)
- Yes it is
- Manual (?) export BEFORE changes land
- Manual import (?)
- Chromium & Gecko
- Room for improvement / collaboration?
- Issue on export PRs?
- A Wishlist
- Collaborate on wit.fyi
- Wish there was a way to browser-specific failures in the same way computed for the graph
- Running web-driver to internal builds
- Collaborate on running WPT on iOS
- Collaborate on MDN survey: developers say interop is a big thing
- Collaborate on wit.fyi
- WPT Right approach
- Importing all the tests could slow us down
- Maybe keep them separate?
- New tests added don’t look like regressions
- WPT: Tracking WPT as a sub-module is not compatible
- Please don't scrape wpt.fyi, use the API
Note:
See TracWiki
for help on using the wiki.