Re: [css-flexbox] Policy for reftests that require features from other specs?

On Wed, May 21, 2014 at 6:46 PM, Daniel Holbert <dholbert@mozilla.com> wrote:
> While I was clicking through the flexbox tests liked off of the ED[1], I
> noticed that some of the tests served by the testsuite (at the moment)
> are really tests for interactions between flexbox & regions.[2]
> Obviously, these tests fail in UAs that don't implement regions.
>
> Is there any sort of policy on where to put tests like these ones, that
> depend on features from several in-development mostly-unrelated CSS
> specs?  On the one hand, it's absolutely worthwhile to test the
> interaction of features. But given that we're grouping testsuites in a
> per-spec way right now, it seems odd that we'd implicitly mandate that
> UAs implement an in-development "Spec B" before they can pass the
> testsuite for a mature very-close-to-CR "Spec A". (except perhaps in
> cases where there was a tight coupling between the two specs)
>
> Thoughts? Is this already laid out somewhere?  Am I reading too much
> into what the testsuites mean?
>
> Thanks,
> ~Daniel
>
> [1] http://test.csswg.org/harness/test/css-flexbox-1_dev/
>
> [2] e.g.:
> http://test.csswg.org/harness/test/css-flexbox-1_dev/autoheight-flexbox-001/
> (through 004)
> http://test.csswg.org/harness/test/css-flexbox-1_dev/autoheight-regions-in-autoheight-flexbox-001/
> (through 004)
> http://test.csswg.org/harness/test/css-flexbox-1_dev/autoheight-regions-in-fixed-sized-flexbox-001/
> (through 008)

Pinging this for other people to answer, because I don't know our policy.

We definitely don't want to discourage cross-spec tests, because
treating specs like silos is just a really cool way to have untested
bugs.

~TJ

Received on Tuesday, 10 June 2014 00:46:22 UTC