Comment by tomsmeding

You do have a point. However, if I'm allowed to move the goalposts a little: not all changes in semantics are equal. If you take a program that crashes for certain inputs and turn it into one that is semantically equivalent except that in some of those crashing cases, it actually continues running (as if on a machine with infinite time and/or memory), then that is not quite as bad as one that changes a non-crashing result into a different non-crashing result, or one that turns a non-crashing result into a crash.

With this kind of "benign" change, all programs that worked before still work, and some that didn't work before now work. I would argue this is a good thing.