Hebrew kerning interacts badly with text anchors

Regarding the screen shot of the Hebrew kerning example...

 The top row of text boxes include Hebrew text without kerning. The RJ one was constructed first, then the other two were made by changing the text anchor with the GUI (-> CJ or LJ). There is a very slight shift with respect to the bounding box, but overall this was the correct outcome.

The second row of text boxes was the same test as above, but with text which includes kerning. Changing the text-anchor results in the text not falling in the correct location with respect to the bounding box.

Also, if one examines the SVG in the Hebrew example for one of the LJ paragraphs the "x" value is the same for all <tspan>. So the SVG appears to encode a properly justified paragraph, it just does not display that way.

The problem is in Layout-TNG-Compute.cpp, where kerning was implicitly assumed to be LTR. There are several other
unrelated changes in my version of that file at the moment so a patch would be confusing. However, the change to fix this
is easy, In _measureUnbrokenSpan change this:

        if (span->start.iter_span->dx._set && span->start.char_byte == 0){
            span->width += span->start.iter_span->dx.computed;

to this:

        if (span->start.iter_span->dx._set && span->start.char_byte == 0){
            if(para.direction == RIGHT_TO_LEFT){ span->width -= span->start.iter_span->dx.computed; }
            else { span->width += span->start.iter_span->dx.computed; }

The kerning example after the fix of the previous post is applied.

Note I am not at all convinced that the fix will work in the general cases. It will certainly work for when the paragraph is all R->L or all L->R, but heaven only knows what will happen if there is kerning in an English span of a mostly Hebrew paragraph, or vice versa.

Issue reproduced on Windows XP, Inkscape trunk revision 12387.

Patch tested successfully with the SVG files provided comments #3 and #4.