A version of this message with some graphics is available here http://www.exnet.btinternet.co.uk/KhandaWeb/extending.htm
The function of ZWJ and ZWNJ in regard to Indic scripts is to alter the shaping of a preceding consonant+Virama but in some Indic scripts (Bengali Oriya & possibly Traditional Malayalam) a device to control the shaping of a proceeding Virama+consonant may be desired. Examples of this are found when the first consonant is the letter Ra. For example, the sequence, Ra+Virama+Ya may be rendered as reph+Ya or Ra+Ya.secondaryForm. This secondary form may or may not legate with the preceding character. For the purposes of this discussion, I will use Bengali examples, as that is what I know best. In Bengali, the secondary form of Ya is called jophola (AKA zofola, jofola, japhala, & yaphala). Jophola is often used to transcribe sounds foreign to Bengali. For example to write the English word 'rat', one would use the sequence Ra+Virama+Ya+VowelSignA+TTa and would expect it to be rendered as Ra+Jophola+VowelSignA+TTa. The question is, how would a rendering device know which form of Ra+Virama+Ya was intended (Ya+reph or Ra+Jophola)? My thoughts were to put a ZWNJ after the Ra to indicate that is not to form a Reph e.g. Ra+ZWNJ+Virama+Ya = Ra+Jophola Then I remembered that in some font designs, secondary forms such as jophola can form a conjunct ligature with the preceding consonant. However, I think that a ZWNJ would imply that Ra and Ya should not legate. My second Idea is to use the primary semantic of ZWJ as used in non-Indic scripts e.g. In the sequence, Ra+Virama+ZWJ+Ya the ZWJ would imply that Virama and Ya should combine to make the secondary form. This rule would only apply if the 'base' consonant were the letter Ra. (This rule would not apply to devanagri as: 1. it is not an affected script and 2. For devanagri, Ra+Virama+ZWJ = eyelash_Ra) Andy
