The latest pinned Qt5 hash works fine - 1baa51c3d06dfaf08afe16e3e7dd77a63052a419
I tried to do update it to 59eebee2600505b2d207d651fb5aa0da979f0f6a, but I got
a strange font related problem. Almost all tests ~5500 fail, because the
default font is changed for some reason. :-/ We should find the reason
and fix it before updating. I don't thinks if skipping ~5500 tests is
a good idea and we can't check ~5500 manually and then maintain ~5500
expected files for Qt 4.8 and Qt 5.0 parallel. :(
Simon, Balázs, you are font experts. Could you check what happened?
I'll attach png results soon.
Created attachment 128559 [details]
png with the lateset pinned Qt5 hash
Created attachment 128561 [details]
png with the lateset Qt5 hash
I forgot to mention, if you would like to debug, you have to apply the buildfix from bug79328 for the newer Qt5.
I spent a bit of time looking into this and this is what I think happened:
There are two differences that we can observe, a difference in _metrics_ (*-expected.txt) as well as in the _visual rendering_ (attached png files).
I do believe the difference in the png files is _not_ caused by any _recent_ changes (i.e. from last week's qt5 pinned hash to current one), but by some more general changes in Qt 5. I don't think we should do anything about them until we drop support for Qt 4.
The difference in _metrics_ I believe is caused by an intentional behavioural change in Qt 5 http://codereview.qt-project.org/#change,15169 (or by one of its follow-up patches).
I'm looking into compensation for this in WebKit to give us Qt 4 like results. In the future perhaps we can make the Qt 4 rendering behave like Qt 5 (in terms of metrics) and rebase all our results (or drop qt4 and rebase anyway).
Created attachment 128701 [details]
Comment on attachment 128701 [details]
Clearing flags on attachment: 128701
Committed r108788: <http://trac.webkit.org/changeset/108788>
All reviewed patches have been landed. Closing bug.
> I do believe the difference in the png files is _not_ caused by any _recent_ changes (i.e. from last week's qt5 pinned hash to current one), but by some more general changes in Qt 5. I don't think we should do anything about them until we drop support for Qt 4.
I believe the png files are just very out-of-date, AFAIK pixel tests do not pass even with Qt4.