My reading of that statement in 3.2.1 is that it is wrong. The
change proposed in 3.2 /does/ affect the semantics of sizeof with
an operand that is an lvalue of VLA type, and also, AFAICT, does
/not/ affect the semantics of sizeof with any other kind of
operand. As to what may have caused it, I have nothing new to
offer.
<editorial-comment>
The proposal document is poorly written. The changes proposed
are either poorly written or poorly thought out, perhaps both.
Sadly I believe the overall quality of the proposal is indicative
of what is going on in the ISO C standard more generally.
</editorial-comment>