Reduce until nothing else could be the cause#
Start from a reproduction that fails reliably, then remove things while it still fails:
- Save the page's rendered HTML and the CSS and script it needs into local files.
- Remove the framework: replace components with the plain HTML they render.
- Remove styles and markup in halves, checking after each step that the bug remains.
- Stop when removing anything makes the bug disappear.
What remains should be a page of a few dozen lines that shows the wrong behavior and states the expected one. Cut a reproduction down to one page covers the reduction in detail.
Run it in each engine#
| Engine | Browsers |
|---|---|
| Blink | Chrome, Edge, Opera, Brave, Chrome on Android |
| Gecko | Firefox |
| WebKit | Safari, and every browser on iOS |
Playwright runs Chromium, Firefox and WebKit builds from one script, which makes the comparison quick. For the final evidence, confirm in the real branded browser.
Read the results:
| Result | Likely meaning |
|---|---|
| Fails in one engine only | A bug in that engine, or a behavior the specification leaves open. Read the relevant specification section before filing. |
| Fails in all three | Probably the page or your expectation. Check the specification. |
| Fails in a new version, works in an older one of the same engine | A regression in that engine. Bisecting Chrome versions narrows it. |
Check what the specification says#
Web platform behavior is specified by the WHATWG and W3C standards, and MDN's browser compatibility tables record known differences. The web-platform-tests project holds shared tests that browsers run, and wpt.fyi shows which engines pass each one. A failing test there for your behavior is strong confirmation and often links to an existing bug.
File it where the engine's maintainers look#
- Chromium: issues.chromium.org
- Firefox: bugzilla.mozilla.org
- WebKit: bugs.webkit.org
Search first. A report needs the reduced test case attached or pasted, the expected and actual result, the exact browser version, the OS, and the results from the other engines. Link the specification section.
Work around it meanwhile#
Your users cannot wait for a browser release. Ship a workaround scoped as narrowly as you can, preferably by feature detection rather than by browser name, and link the browser bug in a code comment so it can be removed when the fix ships.
Worked example: a browser bug that every engine shared#
This example is fictional. A developer believes a recent Chrome update broke a sign-in form: clicking the eye icon to show the password submits the form. They plan to file a Chromium bug.
The reduction is eight lines: a <form> with a password input and a <button> holding the icon. It submits in Chromium, Firefox and WebKit alike. The HTML standard gives a button element without a type attribute the submit button state, so every engine was right. The component library's latest release had dropped type="button" from its icon button. The test across engines turned a browser bug report into a one-attribute fix and saved the Chromium team a duplicate report. HTML standard: the button element
Questions#
What if the bug needs a server?
Keep the server part as small as possible, such as a static file with one response header. Attach the header and body. A reduction that needs your whole backend is not one a vendor can run.
Should I use a framework's playground for the reduction?
For a framework bug, yes. For a browser bug, remove the framework entirely, so the vendor sees only the platform.
Is a different result in Playwright's WebKit proof of a Safari bug?
It is a strong hint. Confirm in Safari before filing, since Playwright's WebKit builds are not Safari.
Return to reproducing what someone else saw for the other layers a replay needs.