Re: [idn] Fast nameprep vs. slow nameprep

["D. J. Bernstein" <djb@cr.yp.to>]

There are five possible approaches under discussion:

   (1) UTF-8 with fast nameprep. Names in configuration files are good:
       KC-normalized, no confusing characters, no uppercase characters.
       (Except uppercase ASCII, for compatibility.)

       The keyboard interface has to help users type good names. No
       changes are needed in qmail, djbdns, and many other programs. A
       few programs with 8-bit bugs, notably Sendmail, need to be fixed.

       Patrik and Keith refuse to acknowledge this approach.

   (2) UTF-8 with slow nameprep. Names in configuration files may be
       bad.

       This lets us leave the keyboard interface alone, at the expense
       of putting nameprep into a huge number of programs. Yuck.

       When Patrik said that qmail and djbdns and so on ``have to be
       changed because of the nameprep stage,'' he was talking about #2.

   (3) ACE with fast nameprep and slow encoding. Names in configuration
       files are good UTF-8 names.

       The keyboard interface has to help users type good names.
       Furthermore, a huge number of programs need to learn about ACE.
       Yuck.

       When Patrik recently agreed that qmail and djbdns and so on
       _don't_ have to learn about nameprep, he was talking about #3 and
       #4. He still refuses to acknowledge #1.

   (4) ACE with fast nameprep and fast encoding. Names in configuration
       files are good ACE names.

       The keyboard interface has to help users type good names and
       convert them to ACE. Furthermore, users need to use an extra
       filter to read the ACE names every time they access configuration
       files with tools such as vi, grep, etc. Yuck.

       The big difference between #1 and #4 is that the world is moving
       to UTF-8, not ACE, as the default extension of ASCII. UTF-8
       already works with vi and grep on my system; ACE never will.

   (5) ACE with slow nameprep. Names in configuration files are UTF-8,
       and may be bad.

       This lets us leave the keyboard interface alone. But a huge
       number of programs need to learn about nameprep and ACE. Yuck.

I hope this list is helpful. I wish Patrik and Keith would stop denying
the existence of #1.

---Dan