On Wed, 16 May 2001 [EMAIL PROTECTED] wrote:
> Roozbeh's proposition that a separator with one or more applied NSMs is
> still one character is not preposterous. But how useful is it? I fail to
> find cases where a number would include NSMs and still be a regular
> number. It also seems to me that Roozbeh's proposal complicates
> implementations of the algorithm rather than simplifies it. In
> consequence, unless there is some evidence of usage for these NSMs on
> separators, I would suggest to follow the laws of inertia and not modify
> rule W4.
There is usage for that. The need arised for the exact example I
mentioned, underlining or overcircling a separator. I want to mark a
separator to be distinguishable from other text, to show that to someone,
or to emphasize on it.
Please note that this works for every other case, If I underline anything
by putting a U+0332 after it, it just works fine, with the bidi ordering
preserved. The only offending cases are separators. NSMs over separators
being different from NSMs over anything else are against the sense of the
bidi algorithm.
About complicating implementations, well, it happened that it simplifies
some implementations, as is the case for Fribidi. It uses a linked list to
keep the NSMs that come after any character with the character, and
removes them from the buffer. We needed to add a special case for rule W4
of the bidi algorithm, for not doing anything if there are NSMs over the
character.
--roozbeh
