NRWT doesn't give temp directory to DRT. See this bug for details: https://bugs.webkit.org/show_bug.cgi?id=36899
Created attachment 66632 [details] proposed patch
Comment on attachment 66632 [details] proposed patch OK. I assume this makes Qt's DRT work better when running multiple copies? I see the ORWT has similar code.
(In reply to comment #2) > (From update of attachment 66632 [details]) > OK. I assume this makes Qt's DRT work better when running multiple copies? I see the ORWT has similar code. Yes, you're right. It is important when we run multiple run-webkit-tests. If we didn't separate temp directories, the local storage and icon database could be damage because of concurent file accesses. This method works correctly for a long time with ORWT.
Comment on attachment 66632 [details] proposed patch Clearing flags on attachment: 66632 Committed r66863: <http://trac.webkit.org/changeset/66863>
All reviewed patches have been landed. Closing bug.
Hmm. It's not great that we're doing this work in the constructor and destructor; I've found that can lead to indeterminate results in general (although it's fine in this case). There's actually a port.setup_test_run() method that would be the appropriate place to create the temp dir - it's called only when we're sure we actually want to start running tests, but before DumpRenderTree is ever launched. Unfortunately, there isn't a cleanup_test_run() method to match it. I'll add both of those and refactor this (in a separate bug/change), unless there's an objection ...
(In reply to comment #6) > Hmm. It's not great that we're doing this work in the constructor and destructor; I've found that can lead to indeterminate results in general (although it's fine in this case). > > There's actually a port.setup_test_run() method that would be the appropriate place to create the temp dir - it's called only when we're sure we actually want to start running tests, but before DumpRenderTree is ever launched. Unfortunately, there isn't a cleanup_test_run() method to match it. > > I'll add both of those and refactor this (in a separate bug/change), unless there's an objection ... Ok, you're right. We don't know when the gc delete this temp directory, so maybe a better way is your idea.