Objective

Comparing with a browser implementation is the simplest way to ensure results will match what is expected. This can't be done by fuzzing or prop testing as both would mean recoding a flexbox implementation in tests to generate the assertions to validate the "random" values used.

#141 removed the gentlest script which ensure the tests generated from Chrome are up to date. I strongly disagree with this.

Fixes #65

• I re-added and fixed the gentest script
• I regenerated the tests. There was a few errors that are probably due to changes in Chrome impl since the last time the tests were run (most likely 3 years ago?)
  • rounding of fractional values has been simplified
  • do not use the minimum size to compute the basis value for the size. this caused an issue when specifying a growth factor as the free space was not correctly computed. I also fixed the doc comment on the default value for flex-grow

About the comments on maintaining the generated tests. I believe this was a pain because they were updated manually as the script was failing. Now that it's working again, the tests should not be edited by hand: update the script and regenerate them. They can be ignored during reviews. It would probably be best to commit them in a separate commit to help reviewers
If needed, this could be improved using GitHub actions:

• generate the tests from a GitHub action, and commit them from the action in their own PR
• or just check that the tests in the PR matches the one from a fresh generation to ensure no manual edit happened