> I also considered that we could just delay all foreign key validations > until phase 3, but I ended up just doing then only when a rewrite is > pending.
I still wonder if it's best to delay the validation of the foreign key regardless of if there's a pending table rewrite, but the patch as it is now only delays if there's a pending rewrite.
Consistency seems the better choice, so I agree we should validate later in all cases. Does changing that have any other effects?