[Webkit] Red tests on master after #12747

[Piotrek Koszuliński]

After #12747 there's one failing test in ​http://tests.ckeditor.dev:1030/tests/plugins/magicline/magicline

The first bad commit is ​git:5875b0cf. The change looked good, but I assumed that the code was wrong - that only one of scroll positions matter. Unfortunately, thanks to missing comments and that broken test I lost now this confidence. There's a small possibility that both scroll positions should be taken into account.

[Olek Nowodziński]

It fails in Chrome (Webkit*) only. ​And this is why. We got to branch some code in element.getDocumentPostion() and write simple tests so that we immediately discover when they finally standardise it.

[Piotrek Koszuliński]

Can we implement this in a way that when it is changed in the browser it will still work in CKEditor, so we avoid problems like Facebook? I mean - is it possible to guess which property should be used?

[Olek Nowodziński]

Replying to Reinmar:

Can we implement this in a way that when it is changed in the browser it will still work in CKEditor, so we avoid problems like Facebook? I mean - is it possible to guess which property should be used?

Assuming that

1. document.documentElement.scrollTop and document.body.scrollTop are either 0 or the right value (different than 0), depending on the browser.
2. If both are 0 then it's really 0.
3. One of the above is the right value.
4. scrollTop is a positive integer.

Math.max( document.documentElement.scrollTop, document.body.scrollTop ) or ( document.documentElement.scrollTop || document.body.scrollTop ) should do the trick.

[Olek Nowodziński]

Put ​branch:t/12758 on review.

[Piotrek Koszuliński]

This patch causes regression in #12747. Check the manual test.

Please target the fix to Webkit/Blink browsers only.

[Olek Nowodziński]

​branch:t/12758b

[Piotrek Koszuliński]

I pushed one more commit to ​branch:t/12758b. I think that we should use the value safely on Webkit/Blink, just like you propose previously. I don't know why you reverted that change now.

However:

• the new test fail on Firefox (rounding) and Safari (2px difference) - you can use assert.inNumberInRange.
• I really miss a manual test.

[Artur Delura]

Modified existing test. Couldn't reproduce issue in manual test - already tried in ticket:12788. Merge ​commit.