[Fribidi-discuss] Re: BiDi WINE status and fribidi

[Behdad Esfahbod]

On Sun, 25 Aug 2002, Shachar Shemesh wrote:

> Not looking at the source yet, I may be talking utter bullshit here.
> 
> What I think will be the proper thing to do is to change two locations:
> 1. When classifying the characters, do the surrogate unification, lookup 
> the combined code point, and then mark both parts of the surrogate as 
> the same type.
> 2. When reordering, if the level is odd (right to left), and the char is 
> a surrogate, don't change the order of the pair.
> 
> As far as I can see, these are the only changes required in order to 
> support UTF-16. They can be relatively trivially extended to support UTF-8.
> 
> If anyone who has actually had a look at the code has anything to 
> correct me, please do.
> 
>             Shachar

You are simplifying things too much.  The two easy steps are what 
you told.  But there are some harder ones two:

	* Rule W5:  States that if there is just *one* char of 
type ....   Then you should be aware that a surrogate pair is 
just one character, not two.

	* Rule L3:  NSMs in RTL levels should be reordered to 
come after their base, now the problem is that both the NSM can 
be a surrogate pair, and the base can be a surrogate.  
Headache...

So please please don't talk about UTF-8, thats already enough.

Yours,
-- 
Behdad Esfahbod		3 Shahrivar 1381, 2002 Aug 25 
http://behdad.org/	[Finger for Geek Code]

#define is_persian_leap(y) ((((y)-474)%2820+2820)%2820*31%128<31)