legacy-javascript Validation Tool

The LegacyJavaScript audit checks if a page is polyfilling Baseline web features. The vast majority of websites don't need to do that. This tool helps determine the polyfills that should be checked for, and additionally validates that the audit is actually correctly identifying each one (even if minified and with no source maps present). If a polyfill cannot be detected correctly, this tool ensures that it doesn't get included in the list to check against.

This tool creates many projects using specific babel transforms / polyfills (called variants) and aggregates the results of the LegacyJavaScript audit for each.

First, be sure to install this folder's dependencies:

yarn

Updating polyfill data

Parts of this folder is also used to generate the data needed to run the audit. To update this data:

bash update.sh

Validate

Run:

node run.js
# `STAGE=build|audit|all node run.js` to just build the audits or run LegacyJavaScript on them. Defaults to both (`all`).

summary-signals.json - summarizes the signals that LegacyJavaScript finds for each variant. Variants in variantsMissingSignals (excluding core-js-*-preset-env/baseline_true_*) signify a lack of detection for that variant. Full coverage isn't necessary.

summary-sizes.json - lists the size of each minified variant. Useful for understanding how many bytes each polyfill / transform adds.

Additional validation happens in core/test/lib/legacy-javascript-test.js.

Interpreting Results

There are two outputs to this test:

• summary-sizes.txt
• summary-signals.json

summary-sizes.txt lists each of the variants (grouped by type) and sorted by their byte size. This is mostly a diagnostic tool and changes in this can be ignored. This is checked in for purely informative reasons–whenever lockfile is updated, changes in summary-sizes.txt give an indication of how transforms or polyfills might be changing in the developer ecosystem. It's also a quick reference for the relative cost of each of these transform/polyfills.

summary-signals.json is for preventing regressions in the audit. .variantsMissingSignals should at least have the babel-preset-env=true variant (since this whole test is about finding signals when babel-preset-env is NOT used). There may be more missing variants since it's just a heuristic. The number of these should only go down as the pattern matching improves.

For the signals of each variant, the expectation is that the number of them only goes up.

Notes

Digging into core-js: https://gist.github.com/connorjclark/cc583554ff07cba7cdc416c06721fd6a