Re: [xmlroff] Re: xmlroff on Windows

[Jesper S. <je...@no...>]

Tony Graham wrote:

....

>  > 
>  > *** If I understand xmlroff correctly you are using pt as 'device-unit'. This 
>  > works ok when using pangopdflib since it too (I assume) use pt as
>  > unit, but my pango-win32 backend use pixel as unit. The result of
>  > this is that all lengths/widths/heights are sometimes measured in
>  > pixel (when they are 'received' from pango) and sometimes in
>  > pt. It's not possible to know when to convert from pt to px and
>  > when not to do it.
> 
> Yes, xmlroff uses pt as the 'device-unit'; i.e., it converts all
> dimensions in the FO expressions into points.
> 
> I haven't looked at the Pango Win32 backend, but Pango uses 1/1024 pt
> as a 'Pango Unit', and defines PANGO_SCALE to simplify converting
> points to Pango units.  

I think that this is the problem. From:
http://developer.gnome.org/doc/API/2.0/pango/pango-Glyph-Storage.html#PangoGlyphUnit

"...The PangoGlyphUnit type is used to store dimensions within Pango. Dimensions 
are stored in 1/PANGO_SCALE of a device unit. (A device unit might be a pixel 
for screen display, or a point on a printer.)..."

note the "...(A device unit might be a pixel for screen display, or a point on a 
printer.)..." i.e. a 'device-unit' may be a pt, but it does not have to.

>Pango uses Pango units except when it doesn't;
> e.g. for font sizes.  Unfortunately, the Pango functions just use
> 'int' for all dimensions rather than defining a type for Pango units.
> (Actually, PangoGlyphUnit is defined in pango-glyph.h but is mostly
> only used in doc comments.)
> 
>  > I can see two solutions to this:
>  > 
>  > 1. All pango-backends compatible with xmlroff must use pt as a unit.
>  > 
>  > 2. Use a 'scale' in xmlroff to convert to pango-device-units, or convert from 
>  > pango-device-units to xmlroff-device-units.
> 
> Or:
> 
> 3. Require the child types of FoDoc (wish I had a better name for it)
>    to handle the conversion from points (i.e., xmlroff units) to
>    Pango backend units.

I don't think this will work. Consider this snippet from fo-block-area.c:

void
fo_block_area_new(...)
{

   ...

   pango_layout_get_extents (pango_layout, NULL, &logical);

   fo_area_area_set_height (*new_area,
			   logical.height / PANGO_SCALE +
			   border_before_width + padding_before +
			   border_after_width + padding_after);
   ...

}

If the pango version in use uses pt as 'device-unit', all is fine. But if it 
uses another unit we have a mix of units.


>    For example, the current fo_doc_pdflib_render_layout() code does
>    the multiplication by PANGO_SCALE to convert xmlroff units to Pango
>    units.
> 
>    And, having complained about Pango, it now seems that xmlroff
>    should define and use a typedef for xmlroff units to save people
>    from repeating your confusion.
> 
>  > 1. will loose precision and will not be compatible with:
>  > http://developer.gnome.org/doc/API/2.0/pango/pango-Glyph-Storage.html#PangoGlyphUnit
>  > 
>  > 2. seems the way to do it. I don't know if there is an API in pango that lets 
>  > you express the device-unit in a 'known' unit, i.e. 'pango_deviceunit_to_twips' 
>  > etc. If not, the 'scale' has to be set explicit by the 'client' of xmlroff/libfo.
> 
> In theory, the FT2 backend requires you to specify the resolution when
> creating a PangoContext (or a PangoFT2Context), but the call to
> FT_Set_Char_Size() in pangoft2.c shows that the resolution isn't
> really used.
> 
> Also, the XSL Recommendation errata says that 1 px = 1/96 inch.

Note that in fo-expr-eval.c pt==px.

#define POINT2DEVICE(x)		x
#define PIXEL2DEVICE(x)		x

Do you want me to file a bug?

Regards,
Jesper.