Re: [emacs-bidi] mirroring of glyphs

[Behdad Esfahbod]

On Tue, 20 Nov 2001, Eli Zaretskii wrote:

> > From: Behdad Esfahbod <address@hidden>
> > Date: Tue, 20 Nov 2001 00:33:29 +0330 (IRT)
> > 
> > Warning: To test your algorithm, use test cases from everywhere you 
> > want, but just compare the output, with fribidi's output
> 
> That, of course, ignores the possibility of bugs in FriBidi ;-)
>
> Some site (which I forget for a moment) has a small number of patches
> to FriBidi, which indeed fixes a couple of bugs in the last official
> release.
> 
> So, I'd say, FriBidi alone is not a definitive standard for
> comparison.  No offence.

Well, some other points:

1. The CVS version of fribidi has no known bugs (As Roozeh said I'm so 
lazy to release it, but when released it will hold a version number 
near 1.0 [perhaps 9.9 or 1.0pre1])

2. For near a whole term I waste all my nights reading fribidi's small 
source code, redesigning, implementing, for example, I found that the 
5 loops implementing the Weaks part is so buggy and nonefficient, I 
drew the dependency graph of rules W1 .. W7, and found the algorithm 
with minimum number of loops for them, and proved its correctness, and 
implemented it, then be sure that the algorithm is provably the same 
as UAX#9, and the code is audited hundreds of hours...

3. And about the implementation bugs, In the new year holidays, I made 
on of my computers at home (Celeron 600), to do the loop of generating 
fully random test cases of length 65000 (Yes! any random sequence of 
bytes between 1 and 127), and let it run for about 48 hours, no non 
conforming test found!

4. With this non reference non standard code, I found a bug in 
Implementation notes of UAX#9, that Roozbeh reported and it's now 
fixed in Unicode 3.1.


Ok, cooled down ;), this was just why I said 'just test with fribidi...'

-- 
Behdad
29 Aban 1380, 2001 Nov 20

[Finger for Geek Code]