> Does that testcase fail for GCC 2.95.2? If so, that would be a regression, > which would indicate that *some* fix should go to the branch as well. 2.95.2 seems to be able to compile the testcase OK, so it is a regression.