[HarfBuzz] harfbuzz.old applies x advances for rtl in the wrong direction

[Jjgod Jiang]

On Thu, Jan 13, 2011 at 6:37 PM, Khaled Hosny <khaledhosny at eglug.org> wrote:
> X glyph advances should always be applied from left to right regardless
> of text direction, however HarfBuzz code used in Qt applies X advances
> in negative if text direction is right to left which breaks Arabic
> kerning for example.

Are you sure about "X glyph advances should always be applied
from left to right"? Can you please point me out which part of
OpenType spec states that? I was checking [1] but couldn't
find it.

As a side note, harfbuzz-ng is also using the same way as
harfbuzz, see line 1536 to 1545 in hb-ot-layout-gpos-private.hh:

1536       if (HB_DIRECTION_IS_BACKWARD (buffer->props.direction))
1537         for (j = back + 1; j < i + 1; j++) {
1538           pos[i].x_offset += pos[j].x_advance;
1539           pos[i].y_offset += pos[j].y_advance;
1540         }
1541       else
1542         for (j = back; j < i; j++) {
1543           pos[i].x_offset -= pos[j].x_advance;
1544           pos[i].y_offset -= pos[j].y_advance;
1545         }

> This have been reported against Qt:
> http://bugreports.qt.nokia.com/browse/QTBUG-4475
> And though the issue was closed as not being reproducible, I can still
> reproduce it (may be Qt on Windows does not use HB?).

Qt on Windows does use HB for shaping.

> The attached screenshots show the string:
> بربربربربربربربربر
> حوحوحوحوحوحو

Unfortunately I can't even reproduce the "after" screenshot with the
font installed from your link with the Mac OS X native TextEdit, see
[2], while the result looks correct with other Arabic fonts I have
(Geeza Pro, etc.). Could that mean the font itself has some problems?

- Jiang

[1] http://www.adobe.com/devnet/opentype/afdko/topic_feature_file_syntax.html
[2] http://d.pr/piCx