> From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf > Of Peter Kirk
> The option < ta, ZWJ, virama > is mentioned in the document, but > dismissed without proper argument although it would seem to me that this > is a far more logical encoding than < ta, virama, ZWJ >. After all, the > character in question can easily be understood as a ligature of ta and > virama, but certainly not as ta followed by a ligature of virama with > the following character. I had indeed thought of < ta, ZWJ, virama > because of the fact that the khanda ta is kind of like a ligature of ta and virama. But the generic use of ZWJ for requesting more-ligated forms is *not* applicable to Indic scripts. (If it were, < C, virama, C > should produce a half form and < C, virama, ZWJ, C > should be required to generate the conjuct form.) It would *not* lead to more reliable implementations and better usability to mix usages of ZWJ like this unless absolutely necessary. > While I can understand the objection that this > "involve[s] innovations into the general Indic encoding model", there > does come a time when such innovations are preferable to kludges of the > existing model. Using < ta, virama, ZWJ > for khanda ta is hardly a kludge. While khanda ta does not have behaviours typical of a half form wrt clustering (and so is probably best not referred to as a "half form"), it *is* referred to as such by some, including some Bengalis. The Indic model specifies the use of < C, virama, C > normally and < C, virama, ZWJ, C > and < C, virama, ZWNJ, C > for explicit overrides, and this is precisely what is being proposed here. > Another alternative which should be considered is use of a variation > selector. None of the stakeholders on this issue has suggested that option, and I suspect would reject it outright. There is no need to introduce a variation selector; it would constitute yet another innovation in the Indic model and would only lead to more confusion. While the notion that a different presentation form for what is in some sense the same thing does provide some motivation for the suggestion, the Indic model already has mechanisms for dealing with this in the context of Indic scripts. In this context, then, this would be a far greater kludge than a minor deviation from prototypical behaviour of ZWJ wrt clustering. I was aware of these other possibilities; I left them out of the discussion for a reason: they would only serve to make the document longer with no real benefit. Peter Peter Constable Globalization Infrastructure and Font Technologies Microsoft Windows Division