Right, both constraints work for me as well. I didn't mention that 1st one works because it is defined properly with comma, my point was that 2nd constraint is not ignored. I can not judge the importance of this because I am not aware of other issues severity, but I think this should be either registered as a bug just to know that it exists, OR spec should be updated to mention that comma is optional. For me it is important because I need to parse table definition, but I understand that it can be not important for most of others. And that there can be lot of real tables created in the world without that comma and that also can affect the final decision on this.