L2/04-062

Subject: Encoding Bangla Khanda-Ta With Ta+Virama
Date: Sun, 1 Feb 2004 16:50:51 -0800
From: Peter Constable

This proposal doesn't say how TA + VIRAMA + ZWNJ should be rendered, or
how TA + VIRAMA + ZWJ should appear when followed by something other
than a matra or subjoinable C.

The bigger concerns relate to processing and backward compatibility.
First, in a process such as a rendering engine, when TA + VIRAMA have
been encountered, you won't be able to know whether to create a cluster
break until the following characters have been processed, at which point
it is too late; a process would need to be able to look ahead, which may
or may not be feasible or easy to implement. Secondly, the standard has
documented for some time now that khanda ta is represented as TA +
VIRAMA + ZWJ, whereas this proposal involves a different representation
for khanda ta, and a different rendering for TA + VIRAMA + ZWJ, and so
would break any existing implementations or documents.

I agree with some of the suggested merits insofar as khanda ta does not
behave like a half form, which is what is suggested by the TA + VIRAMA +
ZWJ representation. But the processing and back-compat issues are at
least as significant, I think. Khanda ta has exceptional behaviour and
requires exceptional handling no matter what. At this point, an easier
change would be to clarify in the standard what the behaviour of khanda
ta is (current documentation is definitely lacking in this regard).

--------------------