See the stacktrace at http://crbug.com/120453. r76990 introduced RefPtrs to hold SVGFontElement and looks like it is the cause of the leak. I checked removing RefPtrs fixed the leak and svg/custom/use-multiple-on-nested-disallowed-font.html passed successfully (DRT didn't crash). I think zimmermann's recent changes (e.g. r109333) disallow invalid svg contents like the test and the test is no longer effective. jschuh@, zimmermann@, Do you think it is safe to remove RefPtrs which were introduces by r76990?
Created attachment 134254 [details] Patch
Comment on attachment 134254 [details] Patch I think they were added on purpose - don't you see any crashes now?? Maybe something has changed, but I'd be careful here. Can you try running tests under guard malloc? nrwt --tolerance 0 -p svg -g, and/or with --gc-between-tests.
(In reply to comment #2) > (From update of attachment 134254 [details]) > I think they were added on purpose - don't you see any crashes now?? Maybe something has changed, but I'd be careful here. I didn't see any crashes with this patch, but I agree that we should be careful. I'd like to hear jschuh's opinion. > Can you try running tests under guard malloc? nrwt --tolerance 0 -p svg -g, and/or with --gc-between-tests. All svg tests ran as expected with above switches (Debug/Release).
(In reply to comment #0) > See the stacktrace at http://crbug.com/120453. I don't see one.
No, it's not safe to remove the RefPtr without making other changes. This is discussed in bug 66438, along with an explanation of the solution to the memory leak. I just haven't had a chance to do any WebKit work in the last few months. *** This bug has been marked as a duplicate of bug 66438 ***
Comment on attachment 134254 [details] Patch Thanks. Removing r?.