Jeffrey Hutzelman wrote:
--On Wednesday, July 22, 2009 08:20:34 PM +0200 Simon Josefsson <simon@xxxxxxxxxxxxx> wrote:RFC Errata System <rfc-editor@xxxxxxxxxxxxxx> writes:Section: 2.3 Original Text ------------- Section title is "Prohibited Output" Corrected Text -------------- Section title should be "Prohibited Input" Notes ----- Section 2.3 is talking about prohibited input, not prohibited output.I'm not sure about this -- compare RFC 3454 section 5, the terminology and design is about prohibiting code points in the _output_. Testing for code points in the _output_ may be semantically different from testing in the _input_. Some other IDNA operation may modify code points into one of the prohibited _output_ code points. See RFC 3454 section 2, the map&normalize steps are taken before the prohibit step.I agree with Simon here. Prohibiting a code point in the output usually has the effect of making any input containing that code point result in a failure, but the point is that the _output_ will never contain that code point, even as the result of mapping and normalization -- the operation will fail instead. This distinction was a conscious decision made in stringprep.
Agreed.
There is, however, an error -- the first sentence of this section does in fact claim to be specifying prohibited input characters. The correct fix is to that text, not to the section title.
My understanding is that prohibited characters are never generated during mapping.