Re: Ntemacs chooses wrong font.

[Jason Rumney <jasonr@gnu.org>]

Kenichi Handa wrote:
> I currently explicitly generate a unibyte string for font
> names just to avoid the font name encoding problem until the
> font-backend codes gets stable.
>   

Does that mean that the encoding will eventually be done in generic 
code, so I should avoid fixing this in w32font.c?

> Emacs at first checks if a charater is supported by the
> frame font (here "Monaco-10") to avoid unnecessary looking
> up of fontset table .  If supported, the frame font is used.
> And, in your case, the font backend on Windows says that the
> frame font supports it.  That is the problem.
>   

The problem appears to be that the system API used in one of the 
encode_char functions on Windows (I don't know whether it is uniscribe 
or gdi) seems to return a space glyph for unsupported characters in some 
fonts, instead of 0 (which is ".notdef" according to the truetype spec).

Perhaps C-u C-x = should also report which font backend a font belongs 
to, to make tracking these sorts of bugs down easier.

> It seems that you saved the file with some of legacy
> encoding (e.g. euc-cn, big5).  On reading such a file, Emacs
> adds a charset text-property (e.g. chinese-gb2312, big5),
> and if a character has such a property, Emacs doesn't try
> the frame font, but does a normal fontset looking up
> (because `charset' information may change the priority of
> fonts).  So, your fontset setting above takes effect.
>   

Ah, that explains a lot.

> Perhaps, we should not try the frame font for a certain
> group of charcters (e.g. han, indic, ??).
>   

Where is the has_char function used? On Windows, this should work to 
filter out unsuitable fonts, as it checks the character against the 
font's supported scripts.