Re: Rot13 and letters with accents

Otto Stolz wrote:

> Andreas Prilop wrote:
>> So with respect to rot13, U+00E3 and U0061 U+0303
>> are not equivalent.
>
> This means that a naïve rot13 implementation will not be standard
> conforming w. r. t.
> <http://www.unicode.org/versions/Unicode5.0.0/ch03.pdf>, clause C6.

I wouldn't say so. Clause C6 does not require that canonical equivalents 
be treated as identical. Rather, that you must not rely on others 
treating them as different. I think the standard says this rather 
explicitly.

> But then, rot13 has never claimed to be so.

Rot13 is supposed to be a simple reversible obfuscation method that 
prevents people from _accidentally_ seeing some text they might not want 
to see. I think it does this relatively well for dominantly ASCII text. 
It operates at plain text level, so no special software is needed, 
except for a very trivial converter that often appears as a built-in 
feature in e-mail programs.

> Of course, you could come up with a Unicode-capable, standard
> conforming rot13 implementation -- but what purpose could it ever
> serve?

Well, the same purpose. But rot13 is by definition for a limited 
character repertoire. There are many ways to generalize it. But if the 
generalization needs to coincide with rot13 for ASCII letters, there is 
no method that would result in anything as simple as rot13. One 
possibility would be to give up the rotation idea and simply add 13 to 
each code number. Then the decoding algorithm would not any more be the 
same as the encoding algorithm but would use subtraction. But it 
obfuscate sufficiently and simply - though only at the code number 
level. Doing that to utf-18 data would require more processing.

Jukka K. Korpela ("Yucca")
http://www.cs.tut.fi/~jkorpela/