...Good question. I was assuming of uppercase, but maybe not.
I also agree that small capitals have a tertiary or quaternary differences, but it's not clear if they are a variant of lowercase when used as a font style for all letters, or of uppercase letters.
Philippe, I would agree with you that this is a sensible thing to do. But, as Ken has said, it does cause some difficulties. Presumably this is because there is no defined tertiary weight for small caps in http://www.unicode.org/reports/tr10/#Tertiary_Weight_Table, and in software which implements this. I don't see why the currently unused and suitably positioned tertiary weight 0x0007 cannot simply be allocated to small caps. But this would mean changing a Unicode Technical Standard, and software which implements it. I would tend to agree with Ken that it is not worth doing this for just a few very rarely used characters. There are defined weights for <small>, but these seem to be for hiragana and katakana only.So I bet they should have the same distinction as between lowercase and uppercase, so that case-insensitive collation (which ignores secondary differences) will work correctly even if tertiary and quaternary difference are kept to sort accents and other minor variants, probably by sorting small capitals between uppercase and lowercase letters at the same collation weight.
On the other hand, it would be possible to assign to these characters individually the tertiary weight 0x0007, or whatever other value might be suitable. The work needed to do this once for all for about 33 characters (in the IPA Extensions and Phonetic Extensions blocks) is hardly burdensome.
Surely quaternary weights are by default code point values, and so automatically have the required gaps. This tailoring can of course be done, but I am not sure why it might be necessary.The simplest way to allow tailoring is to map small capitals as secondary differences between uppercase and lowercase letters, and let a tailoring remap them in one simple operation to quaternary level, where a 1-weight gap has been left in quaternary collation weights to allow such remapping without needing shifts of weights.
-- Peter Kirk [EMAIL PROTECTED] (personal) [EMAIL PROTECTED] (work) http://www.qaya.org/
