Get the version first#
You need the full version, such as 153.0.8010.12, not 153. Record the exact browser version shows where to find it.
Install it with Chrome for Testing#
Chrome for Testing is a Chrome build made for automated testing: it does not auto-update, and each version has its own download. The @puppeteer/browsers command-line tool installs one by version:
npx @puppeteer/browsers install chrome@153.0.8010.12
npx @puppeteer/browsers install chrome@stable # or the current Stable
It prints the path of the binary it installed. Start it with a throwaway profile so your own profile stays untouched:
"$CHROME" --user-data-dir="$(mktemp -d)" https://staging.example.test/orders
The project's availability dashboard lists the versions with downloads, and its JSON endpoints list every known good version. Chrome for Testing ยท @puppeteer/browsers
Not every patch release has a Chrome for Testing build. If the exact one is missing, use the nearest build on each side of it and say so in your result.
Run a reproduction script against it#
Playwright drives a specific binary with executablePath:
const browser = await chromium.launch({ executablePath: process.env.CHROME, headless: false });
Playwright supports the Chromium version it ships with, so an older Chrome may not work with the newest Playwright. Pin an older Playwright release if needed. Puppeteer ties each release to one Chrome version in the same way.
Find the version where it changed#
When the bug appears in one version and not in an earlier one, bisect between them:
- Install the last version known to work and the first known to fail.
- Pick a version halfway between, install it, run the reproduction.
- Keep the half that contains the change, and repeat.
Ten or so steps narrow a range of several hundred builds to two adjacent ones. With two adjacent Chrome for Testing versions, the Chromium project's revision ranges and release notes narrow it further. A bisect that ends between two browser builds is strong evidence for a browser bug report; confirm it before filing.
What the build does not replay#
The binary is one layer. Two copies of one Chrome version can still differ in:
- Field trials. Chrome enables some features for a share of users.
chrome://versionlists the active Variations. Chrome for Testing is built for repeatable automation, so a trial-dependent bug may not appear in it. - Policies set by an organization's administrator, listed at
chrome://policy. - Flags changed at
chrome://flags, and command-line switches inchrome://version. - Extensions and the profile's stored data. Works in incognito separates those.
Record which of these you matched.
Worked example: a date input that changed#
This example is fictional. A tester reports that a date field shows a calendar icon over the text in Chrome. The developer's Chrome, one version newer, looks correct.
The developer installs the tester's chrome@153.0.8010.12 with @puppeteer/browsers, opens the page in a throwaway profile and sees the overlap. The current Stable build does not show it. A short bisect across Chrome for Testing builds put the change between two adjacent builds of the same milestone, where Chrome's release notes listed a fix to date input padding. The team added a CSS rule scoped to the affected rendering and closed the bug with the build range as evidence.
Questions#
Can I do the same for Firefox?
Mozilla publishes older Firefox builds on its archive server, and Firefox can run with a separate profile. Disable updates in that copy so it stays on the version you installed.
And for Safari?
Safari's version follows macOS updates and is harder to pin. Safari Technology Preview runs beside it, and older macOS versions in virtual machines give older Safari versions.
Is Playwright's Chromium the same as Chrome?
It is a Chromium build matching a Chrome version, not branded Chrome. For a bug that may depend on Chrome-only components, such as media codecs, use Chrome for Testing or channel: 'chrome'.
Return to reproducing what someone else saw for the other layers a replay needs.