From: Matitiahu Allouche
Date: 2005-05-05 00:35:33 -0700
Subject: Comments on L2/05-093


I have some comments on document L2/05-093  
(http://www.unicode.org/L2/L2005/05093-karlsson-line-brk-cmts.pdf) by Kent  
Karlsson.

1) Kent (on page 7 of the document) wrote (see the original for better  
formatting of the examples):
< start of quote >
E.g., the input (where . marks space, and uppercase marks Arabic/Hebrew):
CAR.CAR.RACK<ls>
..."<lre>eng.eng.english<ls>
...quote<pdf>"<ls>
RACE
(where the line breaks are LS-es, the bidi controls are
correctly INSIDE of the quote marks) should turn out as:
KCAR.RAC.RAC
eng.eng.english"...
"quote...
ECAR
NOT as (it is currently, unless I'm too confused):
KCAR.RAC.RAC
eng.eng.english"...
"...quote
ECAR
< end of quote >
Then Kents proposes some changes to the Bidi algorithm.

IMHO, this is not required. ÊThe problem in Kent's example stems from bad  
authoring: the right solution for Kent's requirement would be to terminate  
the first part of the english segment with a <pdf> before the <ls>, then to  
open a new <lre> after the leading dots of the third line. ÊRemember that  
we are dealing with explicit line separators, not automatic line splitting,  
so there is nothing preventing the author to do what I suggest.

However, Kent's proposal has its merits for dealing with white space  
around Segment Separators and in the case of automatic line split. ÊThis  
must be weighed against the disruption that would be caused by any change  
to the algorithm.

2) On page 8, Kent writes:
< start of quote >
Furthermore, to get the default paragraph direction better chosen (if the  
default is used), all digits should be considered strong left-to-right in  
rules P2 and P3. I.e., look also for EN and AN as strongly directional L,  
which, if first strongly directional character, should set the (default, if  
there is no higher level protocol saying otherwise) paragraph level to 0.
This affects the (default) treatment of leading and trailing spaces in  
each "segment", esp. for (almost) purely numeric "paragraphs" or columns",  
possibly followed by a unit (in Arabic, I've seen "cm" transcribed into  
Arabic) (logically) after the digits. If one has tried to do poor-man's  
alignment (which is what one does in plain text, which by definition has no  
higher level protocol for alignments) by using leading spaces, those  
spaces should be on the left side of the digits.
< end of quote >

Sorry, but I don't agree. ÊFirst of all, in an Arabic table of numbers,  
columns should proceed from right to left, so the Arabic letters set the  
right paragraph direction. ÊBut even if Kent's use case is justified, his  
proposal would break other cases, like numbered lists of Arabic items,  
where the item numbers must be on the right side of the textual entries.
No, thanks.


Shalom (Regards), ÊMati
Ê Ê Ê Ê Ê Bidi Architect
Ê Ê Ê Ê Ê Globalization Center Of Competency - Bidirectional Scripts
Ê Ê Ê Ê Ê IBM Israel
Ê Ê Ê Ê Ê Phone: +972 2 5888802 Ê ÊFax: +972 2 5870333 Ê ÊMobile: +972 52 2554160