[css-text] Test for text-tranform:full-size-kana #13461
Conversation
|
This is just a suggestion: I changed the format for these kinds of tests recently to make it easier for manual reviewers to do the comparisons (and slightly easier to create the ref tests). You can see an example at https://w3c.github.io/i18n-tests/run?base=css-text&batch=text-transform&test=text-transform/text-transform-fullwidth-005.html Basically, results are superimposed such that divergeances appear in red. Makes it very quick to spot fails. Just mentioning it in case you think it's a good idea, since your tests are similar. Fwiw, here's an example that will probably contain failures on Firefox: https://w3c.github.io/i18n-tests/run?base=css-text&batch=text-transform&test=text-transform/text-transform-fullwidth-005.html |
|
I considered doing something like this, and decided against because it's really hard to decide if the test is passing of failing when you have sub-pixel or single pixel differences, and just a tiny little bit of red showing. In the full size kana case, a real failure would involve a lot more red than that, but the reviewer does not know that, so when font rendering artifact show a little bit of red, it's not clear. Case in point, with the test you shared, on my machine, chrome is a clear failure: Firefox, on the other hand, is maybe a failure, maybe not. I think it is meant to be a pass, but I do see some red: A human reviewer would be unsure, and an automated ref checker would fail it, even though it is fine. |
|
Yes. I thought a lot about that too, but came to the opposite conclusion ;-). I didn't think the subpixel shadow would be picked up on the ref test - i could be wrong. The instructions on some of the other tests i did like this actually mention that subpixel shadows don't count. |
|
To be rebased on top of #13504, as discussed in https://bugzilla.mozilla.org/show_bug.cgi?id=1498148#c41 |
|
@dpino Rebased on top of your tests from #13504 to try and get the best of both worlds:
Review appreciated. |
Rework the tests from web-platform-tests#13504, so that: * They fit within 600px × 600px (as generally required for reftests) * They are easier for a human reader to review: being asked whether two characters are the same when the only difference expected is size is much easier if you have a reference for the right and wrong size. Only doing this in tests 001 to 004, as for the rest there's not (always) a corresponding small character. * The test that check that nothing happens on already full-size kana only applies the (ineffective) transform to one character in each pair as otherwise, it isn't really testing anything
tests for w3c/csswg-drafts#3143