How to compute a unicode string which bidirectional representation is specified?
- by valdo
Hello, fellows. I have a rather pervert question. Please forgive me :)
There's an official algorithm that describes how bidirectional unicode text should be presented.
http://www.unicode.org/reports/tr9/tr9-15.html
I receive a string (from some 3rd-party source), which contains latin/hebrew characters, as well as digits, white-spaces, punctuation symbols and etc.
The problem is that the string that I receive is already in the representation form. I.e. - the sequence of characters that I receive should just be presented from left to right.
Now, my goal is to find the unicode string which representation is exactly the same. Means - I need to pass that string to another entity; it would then render this string according to the official algorithm, and the result should be the same.
Assuming the following:
The default text direction (of the rendering entity) is RTL.
I don't want to inject "special unicode characters" that explicitly override the text direction (such as RLO, RLE, etc.)
I suspect there may exist several solutions. If so - I'd like to preserve the RTL-looking of the string as much as possible. The string usually consists of hebrew words mostly. I'd like to preserve the correct order of those words, and characters inside those words. Whereas other character sequences may (and should) be transposed.
One naive way to solve this is just to swap the whole string (this takes care of the hebrew words), and then swap inside it sequences of non-hebrew characters. This however doesn't always produce correct results, because actual rules of representation are rather complex.
The only comprehensive algorithm that I see so far is brute-force check. The string can be divided into sequences of same-class characters. Those sequences may be joined in random order, plus any of them may be reversed. I can check all those combinations to obtain the correct result.
Plus this technique may be optimized. For instance the order of hebrew words is known, so we only have to check different combinations of their "joining" sequences.
Any better ideas? If you have an idea, not necessarily the whole solution - it's ok. I'll appreciate any idea.
Thanks in advance.