The Gtk+ port manages to hit this assert on fast/history/clicked-link-is-visited.html and some more test cases. A backtrace to illustrate what is happening is inside the patch.
Created attachment 24855 [details] Do not run into an assert Proposed patch to not run into the assert. From what I can see is that the CURL backend is not really guilty. FrameLoader is currently using a RefPtr to keep the DocumentLoader alive, so the other ports might have a similar invocation pattern. The question is if the call to dl->setPrimaryLoadComplete(true) can be omitted or not.
Comment on attachment 24855 [details] Do not run into an assert > The cruelpit is the following. FrameLoader::finishedLoading is calling I think you want "culprit", not "cruelpit" here. :) Sorry, I don't know enough about the loading machinery to review this patch.
Comment on attachment 24855 [details] Do not run into an assert This change seems right. I'm going to say r=me even though this doesn't have a ChangeLog or a test case.
I did land it in r39221. The next step will be to remove some tests from the skipped list.