Running very long testcase, that loops through all possible field types and checks results, fails with "bad allocation" at some point. Reducing number of tested types makes it pass(the same point). That would not happen, if that was some single erroneous allocation attempt. And test organized in the way, that memory is supposed to be released for each next type iteration. Another option can be, that some memory is not freed at right time, since seemingly all involved objects are in smart pointers