Re: fribidi question

["Chahine M. Hamila" <mch at chaham dot com>]

Answering what wasn't in the other email,

Behdad Esfahbod a *crit :

> On Sat, 5 Jan 2002, Chahine M. Hamila wrote:
> > Salam,
> > Akka 1.0 is really close, one issue is left before I announce the public
> > testing phase.
> > All but one bugs I am aware of are fixed, endian-related issues aside
> > which I will fix as soon as I get my hands on a GNU/Linux x86, all
> > functionalities scheduled for version 1.0 are now in and running.
>
> I have been curious about that, whats the exact problem? (reply me in
> private if needed)

I will if needed, thanks:) In any case, the problem is I don't have an x86
running GNU/Linux right now to test it, so I can't know if my rendering
engine is right. It seems attributes and character codes are swapped
according to endianness in the screen map, so if confirmed, it would mess up
some of my indices and border conditions. I've been working on a Linux PPC
since the only PC that ran Linux here had a bad hard disk crash, so I'll have
to wait til it is tested on PC in order to commit it to CVS and upload the
new tarball. That's a max of ten days, when I move again, but it would be
silly to wait that long since the program is ready. The other problem is
minor, the one that we discuss next,

> > The bug is in computing the cursor position when approximating bidi
> > (yes, bidi approximation was finally included and Akka has become a very
> > extended superset of Acon).
>
> Good news, I'm really waiting to see what it will do, I know the
> problems of simulating a bidi console well...

I still advise against the use of the bidi approximation;)
I included it with Nadim's encouragement because I thought it could be scary
to run Akka and Acon side by side in case a bidi approximation is needed
(when viewing a web page with Lynx for example) and because it is better to
approach bidi using a UAX#9 conformant implementation like fribidi's than
Acon's one. That said, since it's impossible to have an exact bidi ordering
below the application level and edition in a bidi approximation mode would be
buggy in the general case, I made monodirectional modes the default. So we
still have to study the reverse bidi algorithm offered by ICU in order to be
compatible with monodirectional RTL the way UAX#9 is compatible with
monodirectional LTR, formally specify it the way UAX#9 is, and eventually
integrate a sample implementation in fribidi for example.

> (snip)
> You are right, the last element of L_to_V map is the place of the last
> character of input str, in visual_str.

Thanks Behdad. In that case, I must have it wrong on a border condition,
because the data I obtained was a constant:/ Will check it later.

> (snip)
> postition_L_to_V_list has type FriBidiStrIndex which has been
> typedefed as int, then 32bit, not 16, try to use the new interface
> please.

How long do you think it would take to have the new fribidi in the testing
section of Debian at least? How long for RPM packages? This is a rather
important change, but I want Akka's install to be as user friendly as
possible, or it won't meet any significant audience, so these are important
considerations for me.

Regards,
Chahine