Arabic Support for Windows CE

Introduction

When Microsoft Windows CE first came to life and the number of PDAs based on it increased rapidly, the software market was opened widely for a new type of applications and we all wanted to write new applications or to port our application already written for Win32 to Windows CE. Unfortunately, for the Arabic market, there is no Arabic language support in WinCE 2.11 and WinCE 3.0.

After a long time some third parties introduced some solutions and each claimed “We have the first and best Arabic Language Support for Windows CE (or PocketPC)”. They offered their solutions for a big amount of what is called “money”. Too big for a poor freelance programmer to handle. And after one of the companies offering the solution asked me $5000 for their solution (don’t tell them I told you it was bad), I started to write the first and best Arabic Language support for Windows CE myself. It was about 14 months ago. And here is what I did.

Problem Identification

The main problem is that the WinCE 2.11 and WinCE 3.0 do not have intrinsic support for BiDi (Bi-Directional) fonts and RTL (Right To Left) window controls. I don’t know why. I think it’s not too difficult and I also think that, it will not take much memory nor processing time to support those features.

Before I go deep with you, I wanted you to notice what an Arabic support solution makes to a string to be displayed the right way on a window.

The Arabic string is stored from left to right as in English. But when it’s to be displayed, the first character in the string should be the right most character. Further, in Arabic, the character can take more than one glyph (figure or shape) depending on its position in the word: at first, in middle, at last, and alone. In Arabic Enabled Windows and Unicode ready Windows (Windows 2000 and up) all this is implemented intrinsically in the GDI. In WinCE we have to do this ourselves.

The solution

First, You have to install an Arabic TrueType font. The Tahoma of Arabic-Enabled Windows is great.

Arabization involves reversing the string, considering that numbers shouldn’t be reversed and perform glyph substitution manually.

To reverse the string I use:

//////////////////////////////////////////////////////////////////////
void ArabicReverse(CString &s)
{
    CString out, rev;
    s.MakeReverse();

    int i=0;
    while(i<s.GetLength())
    {
        if((s[i]>='0' && s[i]<='9'))    // isDigit(s[i]) ?
        {
            rev.Empty();
            while((s[i]>='0' && s[i]<='9'))    // isDigit(s[i]) ?
            {
                rev = rev + s[i];
                ++i;
            }
            rev.MakeReverse();
            out = out + rev;
        }
        else
        {
            out = out + s[i];
            ++i;
        }
    }
    s=out;
}
//////////////////////////////////////////////////////////////////////

The second step is a bit more tricky so be ready. We know that the Arabic string in the memory contains the Unicode characters that take the same form of the isolated glyphs of the letters. We have to make mapping from the character to the right glyphs to be displayed, considering all states of the letter.

I some how could construct an array of all Arabic letters' state. It is not sorted alphabetically. The first row in the table is the first piece of data I could collect about a letter and it was the letter ‘thal’.

As far as I know .. the rules of glyph substitution are somewhere in the TTF file but I found it very complex. To help constructing the rules for glyph substitution, I made two sets of characters according to whether it may need a link to the previous or the next character.

//////////////////////////////////////////////////////////////////////
CString Arabize (LPCTSTR in)
{
    static struct
    {
        WCHAR character;
        WCHAR endGlyph;
        WCHAR iniGlyph;
        WCHAR midGlyph;
        WCHAR isoGlyph;
    }a[N_DISTINCT_CHARACTERS]=
    {
        {0x630, 0xfeac, 0xfeab, 0xfeac, 0xfeab},
        {0x62f, 0xfeaa, 0xfea9, 0xfeaa, 0xfea9},
        {0x62c, 0xfe9e, 0xfe9f, 0xfea0, 0xfe9d},
        {0x62d, 0xfea2, 0xfea3, 0xfea4, 0xfea1},
        {0x62e, 0xfea6, 0xfea7, 0xfea8, 0xfea5},
        {0x647, 0xfeea, 0xfeeb, 0xfeec, 0xfee9},
        {0x639, 0xfeca, 0xfecb, 0xfecc, 0xfec9},
        {0x63a, 0xfece, 0xfecf, 0xfed0, 0xfecd},
        {0x641, 0xfed2, 0xfed3, 0xfed4, 0xfed1},
        {0x642, 0xfed6, 0xfed7, 0xfed8, 0xfed5},
        {0x62b, 0xfe9a, 0xfe9b, 0xfe9c, 0xfe99},
        {0x635, 0xfeba, 0xfebb, 0xfebc, 0xfeb9},
        {0x636, 0xfebe, 0xfebf, 0xfec0, 0xfebd},
        {0x637, 0xfec2, 0xfec3, 0xfec4, 0xfec1},
        {0x643, 0xfeda, 0xfedb, 0xfedc, 0xfed9},
        {0x645, 0xfee2, 0xfee3, 0xfee4, 0xfee1},
        {0x646, 0xfee6, 0xfee7, 0xfee8, 0xfee5},
        {0x62a, 0xfe96, 0xfe97, 0xfe98, 0xfe95},
        {0x627, 0xfe8e, 0xfe8d, 0xfe8e, 0xfe8d},
        {0x644, 0xfede, 0xfedf, 0xfee0, 0xfedd},
        {0x628, 0xfe90, 0xfe91, 0xfe92, 0xfe8f},
        {0x64a, 0xfef2, 0xfef3, 0xfef4, 0xfef1},
        {0x633, 0xfeb2, 0xfeb3, 0xfeb4, 0xfeb1},
        {0x634, 0xfeb6, 0xfeb7, 0xfeb8, 0xfeb5},
        {0x638, 0xfec6, 0xfec7, 0xfec8, 0xfec5},
        {0x632, 0xfeb0, 0xfeaf, 0xfeb0, 0xfeaf},
        {0x648, 0xfeee, 0xfeed, 0xfeee, 0xfeed},
        {0x629, 0xfe94, 0xfe93, 0xfe93, 0xfe93},
        {0x649, 0xfef0, 0xfeef, 0xfef0, 0xfeef},
        {0x631, 0xfeae, 0xfead, 0xfeae, 0xfead},
        {0x624, 0xfe86, 0xfe85, 0xfe86, 0xfe85},
        {0x621, 0xfe80, 0xfe80, 0xfe80, 0xfe7f},
        {0x626, 0xfe8a, 0xfe8b, 0xfe8c, 0xfe89},
        {0x623, 0xfe84, 0xfe83, 0xfe84, 0xfe83},
        {0x622, 0xfe82, 0xfe81, 0xfe82, 0xfe81},
        {0x625, 0xfe88, 0xfe87, 0xfe88, 0xfe87}
    };
    BOOL linkBefore, linkAfter;
    CString out;
    out=in;
    for(UINT i=0; i<_tcslen(in); i++)
    {
        WCHAR ch=in[i];
        if(((ch>=0x0621 && ch<=0x064a)) // is an Arabic character?
        {
            int idx = 0;
            while (idx < N_DISTINCT_CHARACTERS)
            {
                if (a[idx].character == in[i])
                    break;
                ++idx;
            }
            
            if(i == _tcslen(in) - 1)
                linkAfter=0;
            else
                linkAfter = (isFromTheSet1(in[i+1]) || 
                                   isFromTheSet2(in[i+1]));
            if(i == 0)
                linkBefore=0;
            else
                linkBefore=isFromTheSet1(in[i-1]);
    
            if(linkBefore && linkAfter)
                out.SetAt(i, a[idx].midGlyph);
            if(linkBefore && !linkAfter)
                out.SetAt(i, a[idx].endGlyph);
            if(!linkBefore && linkAfter)
                out.SetAt(i, a[idx].iniGlyph);
            if(!linkBefore && !linkAfter)
                out.SetAt(i, a[idx].glyph);
        }
    }
    ArabicReverse (out);
    return out;
}
//////////////////////////////////////////////////////////////////////
BOOL BzArabicRender::isFromTheSet1(WCHAR ch)
{
    static WCHAR theSet1[22]={
        0x62c, 0x62d, 0x62e, 0x647, 0x639, 0x63a, 0x641, 0x642,
        0x62b, 0x635, 0x636, 0x637, 0x643, 0x645, 0x646, 0x62a,
        0x644, 0x628, 0x64a, 0x633, 0x634, 0x638};
    int i = 0;
    while (i < 22)
    {
        if(ch == theSet1[i])
            return TRUE;
        ++i;
    }
    return FALSE;
}
//////////////////////////////////////////////////////////////////////
BOOL BzArabicRender::isFromTheSet2(WCHAR ch)
{
    static WCHAR theSet2[12]={
        0x627, 0x623, 0x625, 0x622, 0x62f, 0x630, 0x631, 0x632,
        0x648, 0x624, 0x629, 0x649};
    int i = 0;
    while (i < 12)
    {
        if(ch == theSet2[i])
            return TRUE;
        ++i;
    }
    return FALSE;
}
//////////////////////////////////////////////////////////////////////

To use this .. just SetWindowText the window you want to the output of the Arabize().

window.SetWindowText(str); //before
window.SetWindowText(Arabize(str)); //after

or

dc.DrawText(str,……); //before
dc.DrawText(Arabize(str),……); //after

… etc

So there is not much change in code if you already have written old code for English.

Of course it’s not the ideal solution. I think the optimum solution is to hook the DrawText function and add all this stuff. But when I wrote this code there was no online material showing how to hook Windows CE APIs.

I will update the document to show how to retrieve the characters from the glyphs to process them soon. Or it may be another article.