Osztrogonac Csaba írta:
Luciano's idea about using rebaseline server for this huge rebasing game
is very good. I did what you suggested, I generated new results on r121970
with old testfonts and then run tests on r122100 (top of trunk this time).
I propose the following method for rebasing:
- Let's concentrate for Qt 5 WebKit1 first (Qt 4.8 and Qt 5.0 WebKit2
results are very similar,
there are only ~500 different test results on them.) After we finished
Qt 5.
- download http://build.webkit.sed.hu/x/layout-test-results.tar.gz ,
untar it where you want: my-favourite-path
(download can be slow a little bit, because it is a 330 Mb sized file.)
- Tools/Scripts/webkit-patch rebaseline-server my-favourite-path
- wait, wait and wait :) (starting server took 5-6 minutes for me
because of this huge number of tests)
- Pick a directory for rebasing
- Add a comment to https://bugs.webkit.org/show_bug.cgi?id=85203, for
example: "I started rebasing css2.1 tests."
(To avoid paralel working on same directory.)
- Select baseline target qt (not qt-5.0, qt-5.0-wk1, ...) at the bottom
of the page.
- You can use the following hotkeys for rebasing:
- q - add the actual test for rebasing queue (mark the result good)
- x - remove the actual test for rebasing queue (if you marked it
accidentally)
- left and right arrow - jump to the previous/next test
- r - rebase queued tests - if you finished a bunch of tests
I missed something from here. First we should concentrate the differenes between
the old and new results. If the are same, only the font changed, just rebase
them.
But if you notice that the results are absolutely wrong, please don't unskip it
now,
but file a new bug report and/or annotate it in the skipped list. (For example:
If you
can see a red square, but the test said "you should see a green rectangle" ) If
you aren't
sure if the result is correct or not, you can check other ports' expected
results at the
bottom of rebaseline-server. And of course, you can leave a test in the skipped
list if
you aren't sure ... We could check it in the future. So try to unskip easy
tests quickly
as much as possible. And then try to invistigate problematic tests.
- After rebase, check if there is old expected files in qt-5.0,
qt-5.0-wk1, qt-5.0-wk2, and remove them
- Unskip rebased tests for qt-5.0 and qt-5.0/Skipped lists -
https://bugs.webkit.org/show_bug.cgi?id=85203 block
- Upload the patch to bugzilla, review, land, ...
After we finished rebasing tests for Qt 5.0 WK1, we can check 4.8 and
5.0 WK2 results too. Most of them (~90%)
will pass and we only need to unskip them. We have to check only
~500-500 tests and commit qt-4.8 and qt-5.0-wk2
specific expected results because of minor pixel differences.
Any opinion? objections?
Ossy
_______________________________________________
webkit-qt mailing list
[email protected]
http://lists.webkit.org/mailman/listinfo/webkit-qt
