This discussion prompts me to relate something from experience in exegesis of the Ada RM.
In Robert's rules of order, there is an overriding rule that says "none of the other rules in this book can be used to obfuscate" [don't have my copy here, so not an exact quote]. Following that line of thinking, I proposed for the Ada standard that we consider (but perhaps not include explicitly) that the standard has an overarching rule "none of the other rules in this standard can be used to derive obviously incorrect conclusions". The work of interpreting a standard is always one of interpreting intention. Most of the time, we expect that the language clearly indicates that intention, but when it doesn't, we don't blindly follow some misprint or thinko in the standard, we figure out what was intended and makes sense. Of course we still may need to do careful exegesis to see if the standard needs fixing or clarification. My own thoughts in this case, not being a real expert in the C standard, is that the standard seems pretty clear and supports the gcc interpretation. In any case the gcc interpretation is clearly what's intended in my view, so if it can be argued that the standard is inconsistent with this interpretation (I am unconvinced that this burden has been met), then the conclusion is to add a clarification to the standard, not to modify the gcc behavior. Yes, this behavior may surprise, but there are a number of practices in C that are clearly outside the language. Even with the original K&R definition, there were lots of things clearly specified to lead to undefined behavior that people got used to using because they happen to work. Any time you let the actual behavior of a particular be your language tutor, you are in trouble. There is no substitute for knowing the language accurately.
