Get set up
- Make a Bugzilla account at http://bugs.webkit.org/.
- Add <your Subversion working copy>/WebKit/Tools/Scripts to your shell path.
Submit a patch
Verify the problem
- Update your build by running update-webkit followed by build-webkit.
- Make sure the problem you're fixing still exists by running run-safari.
Create a testcase
For step by step instructions see CreatingLayoutTests
- See the documentation on running tests, writing new tests, and Writing Layout Tests for DumpRenderTree
- Run run-webkit-tests to see if your bug fix fixes an existing broken testcase. (If it breaks a test that passed before, be sure your fix is correct before submitting it.)
- Otherwise, create a new testcase.
- It's also possible to create a testcase that produces a PNG file and checksum to be compared with an expected result, if you're testing things like colors. For example, see fast/dom/css-rule-functions.html. For that, be sure not to use testRunner.dumpAsText().
- Create the expected result file(s).
- If your fix fixes an existing test (or breaks one, but you're certain it's correct), delete the existing expected result(s) for that test so a new expected result can be generated.
- Run your new test(s). To run only a subset of the full test suite, run run-webkit-tests with the paths of the test directories or files relative to your LayoutTests directory. For example, run-webkit-tests fast/dom/HTMLSelectElement/ will run all the tests in that directory. If you want to generate PNGs and checksum files too, add the -p option to run-webkit-tests.
- After the tests are done, the test system will launch your newly built Safari and offer to show the actual results of any tests that have no expected results, as well as diffs of the expected and actual results when applicable. It will automatically create the expected result files for any new tests, so look at their results and make sure they're correct.
- If you need to make further changes, you can remove an expected result file and re-run its test to generate a new one. Or, you can run the test with the old expected results in place, and copy the new actual result from /tmp to the correct expected result file. (Don't just copy and paste from the window, because it's hard to make sure that all the whitespace is exactly right.)
- When you're happy with the expected result, run the test again to be sure it now passes.
Prepare the patch
- Find, or file, an appropriate bug for your patch in Bugzilla, following the bug-reporting guidelines.
- See the WebKit guide to contributing code.
- svn add any new files you're adding.
- Run prepare-ChangeLog from a directory that contains all of your changes (to be safe at the expense of speed, you can use the top of your working copy).
- Examine the output from that script and manually edit any ChangeLog files it edited. Add bug reference(s) and brief descriptions for each change, following the examples in the file. Be sure to use spaces rather than tabs, just like in source code.
- Make the actual patch by running svn-create-patch > patch.txt. Your testcase changes and the lines you added to ChangeLog(s) will show up as part of the patch, which is what you want. (If they don't, probably you either need to svn add the files or you ran svn-create-patch from too deep a directory.) Alternatively, and preferably, use webkit-patch upload to create and upload the patch to the Bugzilla bug in one step.
- If you used svn-create-patch, then attach your patch to the Bugzilla bug and set review? on it, leaving the requestee field empty. Wait for review.
- If you used webkit-patch upload as described above, then the patch will already have been uploaded as an attachment to the bug. When doing this, a review will automatically be requested unless you add the --no-review option, e.g., by specifying webkit-patch upload --no-review. Note that uploading a new patch will obsolete a previous patch unless you specify the --no-obsolete option. Use webkit-patch help upload to learn about other options.