Tests, ESM, TS, dead code

[abignell]

Thank you for reg-cli! We use it extensively for our visual regression testing.

I've recently gone ahead and updated our fork of this repo to:

• Remove flow and use TypeScript
• Replace prettier with biome (for linting and formatting with a close to default config)
• Update to ESM
• Fixed dead code
• Fixed mistakenly commited .only in test?
• Updated CI to run typecheck and biome lint/format checks
• Bumps md5-file to v5 with ESM support

Note: we use yarn v4, which causes some issues when the build-ui.sh script clones the ui report builder and has yarn install --frozen-lockfile run as it wants to update the lock file, so I had to remove the --frozen-lockfile/--immutable option just for that, but it's somewhat negligible.

Tests are only executing 1 test, after removing the .only on the CLI tests everything seems to work, I suspect the .only was mistakenly committed and pushed to the main repo here, but it means that the testing is not currently reliable.

There's some dead code cleanup, for example, when calling copyImages the diffDir is passed in, however copyImages doesn't use it - there are minor issues like this throughout the code. Updating to TypeScript and biome exposes a lot of these problems.

Overall this improves quality and maintainability, albeit it sacrifices support to older Node versions.

You can find the PR where these modernised changes were made:

abignell#4

Feel free to borrow any of the updates / changes, or if anyone else wants to take them on board. 👍

[bokuweb]

@abignell

Thank you very much.
This is very interesting, and I’ll use it as a reference.

Progress is quite slow, but I’m currently working on porting to Rust/WASM on the wasm branch, and in the future I expect the current stack to be significantly reduced.

[abignell]

@abignell

Thank you very much. This is very interesting, and I’ll use it as a reference.

Progress is quite slow, but I’m currently working on porting to Rust/WASM on the wasm branch, and in the future I expect the current stack to be significantly reduced.

Awesome to hear and love this quick response, my main concern here is the tests and that committed .only.

I want to also note the fork I linked, I also added a feature to support per-image threshold overrides, where the config can be passed in as multiple CLI options:

--image-config "relative/output/path/to/image/logo.png:thresholdRate=0.1" \
--image-config "relative/output/path/to/imagescreenshot.png:thresholdPixel=100,matchingThreshold=0.05"

Where the image config is: [relative/output/path/of/image]:thresholdPixel=[overrideValue],matchingThreshold=[value] and can also provide a thresholdRate

Alternatively, you can provide a json file name to the same option:

--image-config ./image-config.json

Where the json is just keyed by the screenshot paths, relative to the (expected/actual) directory, and the same key/values. When compared, the image override values are used and are set to fallback to the global defaults. For us this was handy, as we had some specific edge cases where tweaking per-image allowed us to avoid some false positives and other noise.