[FriBidi] partially balanced arabic text

[Jarek M]

Hi. I'm bumping into a following problem. I got a fribidi 0.19.2 running,
used to parse in webkit the contents of TextRun.h (pointer to array passed
as a buffer with arabic characters). When converting (mirroring and
positioning LTR) and later writing to the screen, I get all the characters
mirrored but majority of sentences are written RTL. Attached example and
the code. Any hints how to resolve this?
I tried several methods of setting up paragraph conversion, all producing
pretty much same result. 

The code I used is below.

    	bool lRes = false;
        FriBidiParType pBase;
        FriBidiCharType *pCharTypes;
        FriBidiChar *visualStr;
        int i = 0, cnt = 0;
        FriBidiStrIndex *posLtoVlist;
        FriBidiStrIndex *posVtoLlist;
        FriBidiLevel *embedLevelList;

    	FriBidiChar *fribidi_characters;
        FriBidiStrIndex iLen = 0;
   //     if (m_rtl)
   //         pBase = FRIBIDI_PAR_RTL;
   //     else
   //            pBase = FRIBIDI_PAR_LTR;

        iLen = (m_len + 1) << 2;

        fribidi_characters = (FriBidiChar *)malloc(iLen *
        sizeof(FriBidiChar));
        pCharTypes = (FriBidiCharType*) malloc(iLen *
        sizeof(FriBidiCharType));

        visualStr = (FriBidiChar *)malloc(iLen * sizeof(FriBidiChar));

        posLtoVlist = (FriBidiStrIndex*)malloc(iLen *
        sizeof(FriBidiStrIndex));
        posVtoLlist = (FriBidiStrIndex*)malloc(iLen *
        sizeof(FriBidiStrIndex));
        embedLevelList = (FriBidiLevel*)malloc(iLen *
        sizeof(FriBidiLevel));

        UErrorCode pErrorCode = U_ZERO_ERROR;
        u_strToUTF32((UChar32 *)fribidi_characters, iLen, &iLen,
        m_characters, m_len, &pErrorCode);
       fribidi_get_bidi_types(fribidi_characters,iLen, pCharTypes);
        pBase = fribidi_get_par_direction(pCharTypes, iLen);

        if(m_shaped_characters) // I am not sure who calls it and where,
        not sure who is responsible for memory management
        {
        	delete [] m_shaped_characters;
        }

        lRes = fribidi_log2vis(	/* input */fribidi_characters, iLen,
        &pBase,
        						 /* output */ visualStr, posVtoLlist, posLtoVlist,
        embedLevelList );
        iLen = fribidi_remove_bidi_marks(visualStr, iLen, posVtoLlist,
        posLtoVlist, embedLevelList);

        m_shaped_characters = (UChar *)malloc(sizeof(UChar) * (iLen+1));

        u_strFromUTF32(m_shaped_characters, sizeof(UChar) * (iLen+1), NULL,
        (UChar32 *)visualStr, iLen, &pErrorCode);
        m_shaped_characters[iLen] = '\0';
        m_len  = iLen;

        if(pErrorCode)
        {

        	FILE *file_present =
        fopen("/mnt/ota/system/core/fribidi.txt","a+");
        	fprintf(file_present, "BCTextRunWK.cpp:
        TextRun::doShapeCharacters() failure 2. iLen is:
        %i\n",pErrorCode);
        	fclose(file_present);
           	// LOG AN ERROR AND GET THE HELL OUT OF HERE!!!
            free(posLtoVlist);
            free(posVtoLlist);
            free(embedLevelList);
            free(fribidi_characters);
            free(pCharTypes);
            free(visualStr);
            delete [] m_shaped_characters;
            m_len = 0;
            return;
        }

        free(posLtoVlist);
        free(posVtoLlist);
        free(embedLevelList);
        free(fribidi_characters);
        free(pCharTypes);
        free(visualStr);






I draw       m_shaped_characters using skia, and the results sometimes are drawn correctly, and sometimes LTR with all fonts mirrored. 

Am I using the code correctly? Are there some other methods You would
recommend to use to draw arabic in an older version of webkit? (it is embedded)

Rgards,
Jarek