I'm still seeing occasional leaks from the sharedBuffer code when running layout tests. They aren't reproducible 100%. I have a "leaktest" shell script I run to look for leaks in a loop: #!/bin/bash rm /tmp/layout-test-results/DumpRenderTree* until [ -e /tmp/layout-test-results/DumpRenderTree-leaks.txt ]; do WebKitTools/Scripts/run-webkit-tests --debug LayoutTests/fast/events/ --leaks done In my most recent test, it took about 40 iterations before it finally caught the attached leak (not sure why --leaks is non-deterministic). The backtrace (attached) is different every time, but it is always down in the UString::sharedBuffer code.
Created attachment 31310 [details] sample backtrace for leaking allocation
Created attachment 32024 [details] Proposed fix.
Comment on attachment 32024 [details] Proposed fix. r=me
Committed as http://trac.webkit.org/changeset/45360