How do current implementations handle normalization for IDNA2003?
There were two corrigenda to normalization issued after Unicode 3.2. Formally speaking, an implementation applying IDNA2003 would disregard these corrigenda, but browsers do not consistently implement this behavior. In practice this makes no difference, since the characters and character sequences involved are not found except in specially-devised test cases, so it is understandable that systems may not want to maintain the extra code necessary to duplicate the broken Unicode 3.2 behavior.
