Re: first time post with some questions

[Owen Taylor <otaylor redhat com>]

Mark Leisher <mleisher crl nmsu edu> writes:

>     >> It does however, address some odd things that come up in the Unicode
>     >> Bidi algorithm, like positioning trailing parentheses/brackets/etc
>     >> correctly without the need for explicit bidi marks.  I've been using
>     >> variations of this approach since 1995 and it has worked "pretty good"
>     >> :-)
> 
>     Roozbeh> Mark, are you sure this should be done? I mean anything other
>     Roozbeh> than exact standard compliance in this case, will really annoy
>     Roozbeh> the user. She will have something shown on her Windows
>     Roozbeh> application, and something else on his Unix application, for
>     Roozbeh> example.
> 
> The inclusion of explicit bidi marks to fix these small reordering
> glitches in the Unicode algorithm should have no effect on my
> routine.
> 
> The idea behind my approach is to demonstrate that exactly the same
> results as the Unicode Bidi Reordering algorithm can be achieved in
> a much simpler fashion.
>
> Because the Unicode reordering algorithm is the reference, if mine does not
> produce exactly the same results, then it is no good.

I don't really follow - you claim that your algorithm handles 
some cases better, but, on the other hand you say that you want
to get exactly the same results. 

I'm definitely of the opinion that following the Unicode algorithm
exactly is a absolute necessity, which is perhaps the main reason for
using the FriBidi code in Pango.

And FriBidi isn't very complex - as it is currently, without handling
explicit embedding and overrides, the part that Pango uses is about
600 lines of code. Adding {LR,RL}{O,E} shouldn't be more than a few
hundred more.

(I actually think that explicit embeddings and overrides are going
to be fairly rare, but would like to see the support there.)

So, the only sense I could imagine it being complex is that it might
be slow compared to a simpler algorithm.... but compared to shaping,
determination of bidirectional level isn't a significant bottleneck
in Pango. And I know of several more optimizations that could speed up
FriBidi further.
 
Regards,
                                        Owen