If you are implementing a compiler you do what makes most sense for the target language/machine. On the other hand, if you specify exactly how each ill-defined

If you are implementing a compiler you do what makes most sense for the target language/machine. On the other hand, if you specify exactly how each ill-defined operation should be handled you 1. restrict the freedom of compiler writers. On very small systems, for instance, array bound checking may not be implemented, 2. make programmer rely on non-obvious behaviour/results which makes programs harder to understand and maintain, 3. make the language specification more complicated.