At 19:23 01/10/19 +0900, Soobok Lee wrote: > > I didn't talk about unassigned code points. > > What I mean is this: If you have two frequency tables, > > and they differ by e.g. one forgetting a particular > > character, then virtually all your compressed results > > are different. If you have a wrong mapping in a > > NFC or NFKC algorithm, then only that character is > > affected. > >Never. The REORDERING table for a script has *NO* >missing characters because it applies idempotent >mappings for even unassigned code points.
What I meant is that e.g. somebody goes ahead and implements reordering based on one of your tables. Then e.g. the IETF includes reordering in IDN (Don't misunderstand me, this is only an example, of course I think this should never happen.), but with a different table based on newer statistics. The original implementer doesn't catch that he has a different table, and as a result, we have total chaos, nothing matches. In comparison, if somebody by chance gets an old NFC/KC table, a few labels won't match, but it won't be a whole script or more that gets ruined. Regards, Martin.
