The rationale for making a variation selector ineligible to apply to combining marks comes from normalization. In normalization all combining character sequences are put in their canonical order, based on their combining class. If we want to allow accents to be placed on base characters to which a variation selector has been applied, then the variation selector needs to come in between the base character and the accent. Unless we make it a combining character, the VS would interrupt the combining character sequence (separating the accent from the base character). But if it is a a combining character, it takes part in the reordering. Therefore, it needs to have combining class 0, so it stays with the base character. And that's the reason it cannot be used to apply to characters whose combining class is not 0.
So you see, the rule is not based on the linguistic nature of the combining character, but on its combining class.
But if the VS has class 0, it stays wherever it is put because it does not participate in reordering. If it is put after a particular combining character, it will stay there. So this argument doesn't really hold. Yes, problems do arise if there is more than one combining character between the base character and the VS and they are not in canonical order. But this is a marginal case which can be avoided by ensuring that canonical order is always used.
So, provided that the need to encode a variant of a combining character can be demonstrated, I can see no good reason not to allow a VS to be used for this, with appropriate notes about ordering of combining character sequences.
An alternative of course would be to define a special VS with the same combining class as the character it applies to, so that the two will always remain together. Thus there would potentially be the need for a considerable set of VSs. But I don't think this is really necessary.
-- Peter Kirk [EMAIL PROTECTED] (personal) [EMAIL PROTECTED] (work) http://www.qaya.org/
