Re: minibidi bug

[Ahmad Khalifa <ahmad at khalifa dot ws>]

Behdad Esfahbod wrote:
> Quoting fribidi.c:
>
> /* There are a few little points in pushing into and poping from the status
>    stack:
>    1. when the embedding level is not valid (more than
>    FRIBIDI_BIDI_MAX_EXPLICIT_LEVEL=61), you must reject it, and not to push

>    2. there's a more confusing point too, when the embedding level is exactly
>    FRIBIDI_BIDI_MAX_EXPLICIT_LEVEL-1=60, an LRO or LRE is rejected

The input text having the problem only has 2 explicit marks, so this
is not an issue with reaching the maximum embedding levels :)

Quoting miniBidi's readme
> the failed test is ( ; <-has type RLE. : <- has type LRE)
> - INPUT         THE ;best :ONE and
> - OUTPUT        best ENO and EHT
> - miniBidi      ENO and best EHT
> unfortunately the ';' is resolved at level 3 and replaced by a BN in the
> reference C implementation, minibidi removes the ';' after resolving its
> level. later on 'best' takes over and increases the level to 4. miniBidi
> has no trace of the 3 so it misses a flip.
> I currently dont have any solution for this failed test.

I remove the ';' early on after rules X1-X10.
the embedding levels should be
- THE ;best :ONE and
- 11113444..........

while miniBidi resloves them to
- THE ;best :ONE and
- 1111 222..........
OR
- 1111 444..........
As far as I can remember...

So as you can see, applying rule L2 (reversing any contiguous sequence)
with and without the ';' would result in 2 different outputs

any ideas ?


ak.