Koji, any idea about the heritage/relationship between those tests?

css3/flexbox is all original tests, at least. I don't know the history of the other tests.

css1/units/rounding.html is also original

Assigning to Koji to answer the questions in comments 0,1.

I don't know of the history of these. As far as quickly looked at them:

- css1/ isn't clear, it may be a different set, or just because tests in csswg has evolved since then.
https://github.com/w3c/csswg-test/tree/master/css1

- css2.1/ looks like an old import in 2005.
https://chromium.googlesource.com/chromium/src/+/bea5137f66afd81b15227009962e56ffd972b3ad
Somehow identifying tests that were not modified since then and remove them (or blindly is good enough?) then importing css21
https://github.com/w3c/csswg-test/tree/master/css21
looks reasonable to me. This will increase the number of tests significantly (493 -> 14000), I'm not sure how it impacts our bots capacity.

- css3/ isn't import as far as I know. Maybe it was when originally created, but I know many tests were locally added to the directory. It should be better organized though. Reorganize, upstream, and re-import looks good direction to me.

In addition, writing-modes has both local and imports. The situation is similar to flexbox; some of them may be testing the same thing, but written separately. Eliminating duplicated test cases isn't easy.

Does this answer your question? I think importing css21 is a good thing to do. If you can do that, that'd be super helpful for the layout team.

Yep, that answers my question :-) Filed bug 656669 for that possible task, closing this bug.