https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63233
--- Comment #5 from leis at in dot tum.de --- Thanks. I did notice that I get warning with optimization but not without, which is strange indeed. However, obviously my real problem is more complicated, and I'm trying to understand if I violate the standard. If I pick apart the offending line (foo.a[1] = 99;): int* p1 = foo.a; //1 int* p2 = p1 + 1; //2 *p2 = 99; //3 Which of these lines causes the undefined behavior? Why should pointer p2 be invalid? Fundamentally, what I'm really trying to do, is to have two arrays (of different types) in a fixed-sized struct. One array grows from the front, and one from the end. Dynamically I make sure that they do not overlap, but the sizes of the two arrays are not known statically. Is it really violating the standard?