Visual Regression Testing with BackstopJS
The visual regression process contains three steps:
-
- generate the ideal result (how it should look) in order to compare the test result to this one later on.
-
- generate the real result (how it actually looks) to compare this one to the reference.
-
- when the test is fine and the references shall be updated, you may approve them.
git clone git@github.com:ZeitOnline/test_visual_regression.git
cd test_visual_regression
npm i
Resulting in the following directory tree:
.
├── backstop_data
│ └── html_report
│ └── [...] —> The report generated by BackstopJS after testing
├── backstop_tests
│ └── [...] —> Legacy Backstop test files
├── data
│ ├── references
│ │ └── [...] —> Generated reference screenshots (how it should look like)
│ ├── tests
│ └── [...] —> Generated screenshots after testing (how it actually looks like)
├── engine_scripts
│ └── [...] -> scripts for engine (e.g. puppeteer) control
├── scenarios
│ └── [...] —> Backstop test scenarios
├── utils
│ └── [...] —> Utility scripts go here
├── README.md
├── backstop-demo.json —> Demo test config generated by backstop
├── package-lock.json
└── package.json
Generate a reference you want to test against later on. This will create screenshots of all pages/elements, optionally filtered by label. Pass a --filter=<scenarioLabelRegex>
argument to just run scenarios matching your scenario label.
npm run reference
Example: npm run reference -- --filter=zon::video
For further examples take a look at the package.json.
After making some changes to your code, you might want to test if the pages still look the same or if they have been affected by your changes. After running the following command, a HTML report is generated and opened in your default browser.
npm run test
Example: npm run test -- --filter=zmo::
If you are happy with the results after testing, you may promote the test screenshots to be the new reference screenshots.
npm run approve
Example: npm run approve -- --filter=zon-teaser-podcast-lead
Test configs for BackstopJS can be setup using JSON files or JS modules, see BackstopJS Docs.
To generate test scenarios testing the dark theme, you can add an engine script to your test scenario as an onBeforeScript
property. To prevent duplicate testing, it is required that you also add the label
of darkmode
to the scenario. For example:
{
label: 'darkmode',
onBeforeScript: 'prefers-color-scheme-dark.js',
url: '/zeit-online/article/simple',
readySelector: '.nav__ressorts--fitted',
selectors: ['header.header'],
viewports: ['tablet', 'desktop'],
},
By toggling enhanceWithDarkMode
in backstop-settings.js
to true
(default: true
), you can switch on/off the duplication of all tests for darkmode. If switched on however, manually added darkmode szenarion like the one above are omitted to avoid duplication.
If you want to run darkmode tests seperately use the darkmode
key word as filter, e.g.
npm run reference -- --filter=darkmode
The requirements to make working screenshots of hover effects are a bit challenging and poorly documented. To make this work the engine script clickAndHoverHelper.js
needs to be used as an onReadyScript
, which we do by default in the global onReady.js
. Additionally the shot cannot be made by selecting a dom node on the page rather the selectors
property needs to be set to document
. For example:
{
url: '/zeit-online/centerpage/zon-teaser-lead',
hoverSelectors: ['.zon-teaser-leadtopiclink'],
selectors: ['document'],
},
Testing for click events also need the engine script clickAndHoverHelper.js
(supplied by default configuration). You can configure a test with a given list of clickSelectors
. The trick here is to choose the correct selector
to see the resulting action on the screenshot. For instance a click on a bookmark icon (if the user is not logged in) will result in a dialog in the middle of the viewport thus using selectors: ['viewport']
is mandatory to see the result. E.g.:
{
url: '/zeit-online/centerpage/zon-teaser-standard',
clickSelectors: ['.bookmark-icon'],
selectors: ['viewport'],
},
Cookies are described by JSON files (see engine_scripts/cookies.json
). There is a Chrome extension „クッキーJSONファイル出力 for Puppeteer” with which cookies can be exported from the browser in the correct format. Cookies can be loaded with the helper script loadCookies.js
that can be loaded as onBeforeScript
.
You can write to localstorage likewise. There is an example puppeteer script in engine_scripts/localstorage-darkmode-on.js
.
For further possibilities see the example scripts in engine_scripts
and/or refer to the docs on running scripts and the puppeteer api documentation.