It's a good practice to call CachedResource::removeClient(client) when the client destructing. We need to do this for InconLoader to prevent m_resource from keeping dangling client in case m_resource is referenced by someone else in the future.
Created attachment 210141 [details] Patch
Comment on attachment 210141 [details] Patch I am really concerned that we are doing these fixes without regression tests. Over the years we have found the extra effort to demonstrate the problem with a test is well worth it.
(In reply to comment #2) > (From update of attachment 210141 [details]) > I am really concerned that we are doing these fixes without regression tests. Over the years we have found the extra effort to demonstrate the problem with a test is well worth it. I understand the importance of regression tests. But for this specific change it's just defensive and following the common good practice of removing client when destructing. At the moment this is not a bug without the patch because m_resource is going to be destroyed soon anyway.
Comment on attachment 210141 [details] Patch Clearing flags on attachment: 210141 Committed r154905: <http://trac.webkit.org/changeset/154905>
All reviewed patches have been landed. Closing bug.
*** Bug 106663 has been marked as a duplicate of this bug. ***