bug#21846: 25.0.50; emacs hebrew script bidi nikud rendering bug

[Benjamin Riefenstahl <b.riefenstahl@turtle-trading.net>]

Benjamin Riefenstahl writes:
> With m17n-edit, the text is run through COMBINING.flt (according to
> MDEBUG_FLT=3), with Emacs, no FLT is triggered at all.  Which makes me
> suspect that Emacs does not use m17n for this font at all (?).

So I armed myself with a debugger and tried to trace what happens.  My
command line is

  gdb --args ./emacs -Q --eval '(set-default-font "Aharoni CLM 20")' test.txt

The first wrong turn is in hebrew-shape-gstring
(lisp/language/hebrew.el).  This only calls the font-based shaping
function, if the font indicates that it supports OpenType (OTF) features
for Hebrew.  As a fallback, limited shaping is done in ELisp by looking
up precomposed characters.

When I stub this out for further testing, we get until ftfont_shape
(ftfont.c).  That function checks if there is any OTF support in the
font and only when it is there, it calls libm17n.  The next function in
the chain (ftfont_shape_by_flt) actually uses that OTF object that we
did not get in ftfont_shape to pass some font information.

There seem to be three possibilities:

* Don't fix, PostScript fonts are just not supported here.  Maybe instead "Aharoni CLM" could be converted to OpenType.
* Try to support non-OTF fonts in ftfont_shape_by_flt.  See how the example programs for libm17n do it.
* Reimplement what COMBINING.flt does in hebrew.el.