What the author wants, essentially, is to dictate that common disable-UB flags (e.g., -fwrapv, -fno-strict-aliasing) be folded into the C standard. Well, he probably knew that the C standards committee isn't going to accept that, so instead he's trying to "only" disable the bits of undefined behavior that don't screw over the user. Except he doesn't have a good enough grasp of the standard to actually effect proper wording:

(1) essentially disables inlining as a side effect, at least if undefined behavior is involved. (2) either is 100% redundant or means that every implementation must strictly execute the C abstract machine with absolutely no optimization permitted. (3) means you can no longer access any type with a char. You also couldn't access the bits of a float by casting to an int.