Re: [emacs-bidi] Speed of logical-to-visual conversion

[Eli Zaretskii]

On Wed, 10 Oct 2001, Behdad Esfahbod wrote:

> Well, the speed for determining the bidi levels is linear to the
> number of runs, a run is a maximal substring with all chars of same
> bidi type, and reordering is somehow more time consuming, a character
> on level i, is reordered i times, I have some plans to speed up
> reordering, but this is the current state, it means when you run
> fribidi_benchmark, its about three times slower for a string full of
> explicit bidi marks (without the iconv overhead).

I meant without explicit formatting codes, only implicit levels (that's 
by far the most probable case).

If reordering eats many cycles, then the speed will go down when there 
are lots of direction changes.  In the sequential algorithm I wrote, the 
reordering is not a particularly expensive operation, since, essentially, 
there's no reordering, just jumps between characters in a non-linear 
fashion.