Michael Veksler wrote: >The problem that PR6594 tries to fix looks real. > Perhaps. But can you produce from the PR, a short, self contained testcase, which provably leaks memory? Ciao, Paolo.