Follow-up to Monday's discussion of digits

[Kenneth Whistler]

Eric Brunner asked:

> Could you point out where the "slightly different *directional* 
> behavior, with distinct bidi properties" is present? I'm reading TR9 but 
> I don't yet see where this matters. Examples of the differences would be 
> nice.
> 

Well, I'm not a bidi expert myself, so I'd appreciate it
if anyone who has actually implemented this stuff could
chime in here...

But my reading of the bidi algorithm would concur that for
*most* contexts, bc=AN versus bc=EN isn't going to make any
difference in the final layout order of lines.

The exception would seem to be in the following contexts:

L N EN --> L N L  --> L L L
L N AN --> L N AN --> L e AN

For L L L you are going to get resolved levels 0 0 0 in a L-to-R
context and 2 2 2 in a R-to-L context. And everything in the
stretch is going to display L-to-R in either.

For L e AN you get resolved levels 0 0 2 in a L-to-R context
and 2 1 2 in a R-to-L context. For that one, the neutrals
in between the L and the AN make a significant difference then
in the final layout result. You reverse both runs of 2's
separately, and then the entire span including the neutrals.
That gives you a very different outcome than if you had
started with an L N EN sequence, instead.

I could be wrong here, though -- so if somebody who knows better
can correct me, that would be great. Also I'm not clear
regarding what about the behavior of neutrals in the context
of strong L strings and Perso-Arabic numbers versus Arabic
numbers led the designers of the bidi algorithm in the first
place to difurcate the properties to have this result.

--Ken