[Webkit-unassigned] [Bug 22039] Update LayoutTests/platforms/gtk/Skipped - get the LayoutTests going

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=22039





------- Comment #43 from zandobersek at gmail.com  2009-01-06 14:29 PDT -------
In discussion[1] on the webkit-dev mailing list, I proposed some methods to
push layout tests testing forward a bit. Since this is a Gtk-specific issue and
the thread is not, I'll explain it a bit more here.

The idea is to empty the Skipped file and fill it with all the tests that are
generating new results. We shouldn't start applying these results into trunk
until proper work is done in DRT - complete implementation of EventSender,
AccessibilityUIElement, AccessibilityController GCController etc., proper fonts
for DRT, font size and so on. Only then should a script be assembled to start
semantically diffing expected results for Mac and Gtk ports, port them into
trunk and do troubleshooting if needed.
About `semantically` - pixel-for-pixel rendering between Mac and Gtk port is
improbable. Render dumps would be in such script dissected and checked for
consistence in proper rendering element and content.

After that, many failures and timeouts would be seen on the buildbot. Tests
with a same cause of failure would then be grouped and added to the Skipped
list, and a proper bug report would be filled in. With this, it would be easy
to fix tests and furthermore implement the Gtk port. Prior to grouping, we
should do, as proposed on the list, stressing with randomizing the test
sequences.While this would then only be a bypassed problem that can still occur
to normal users and should be debugged in the engine itself, we would gain a
list of constant failures and add that to the Skipped.

What I'd like is your opinion on this method. What else to do, what not to.
There are at the moment some pain-in-the-arse failures on the buildbot (there
are also JSC regressions, but that's not the subject of discussion here) that
can be fixed in some way. _But there's no fire_, layout tests are not
essential, though they are there to help see regressions and to inform
developers that they've done something wrong or haven't done anything yet about
a specific issue.

There are now more than 10k tests being tested on Mac buildbots. 80+% would be
lovely to see on the Gtk buildbot.

Regards,
Zan Dobersek

[1] https://lists.webkit.org/pipermail/webkit-dev/2009-January/006253.html


-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.