bug#34035: 26.1; Arabic shadda-kasrah renders incorrectly

[Eli Zaretskii]

> From: "Basil L. Contovounesios" <contovob@tcd.ie>
> Cc: Eli Zaretskii <eliz@gnu.org>,  34035@debbugs.gnu.org,  Peter
>  <craven@gmx.net>
> Date: Wed, 19 Aug 2020 09:01:06 +0100
> 
> The kasrah is shown below the sin:
> 
> [...]
> 
> Composed with the following character(s) "ِّ" using this font:
>   xfthb:-PfEd-DejaVu Sans Mono-normal-normal-normal-*-26-*-*-*-m-0-iso10646-1
> by these glyphs:
>   [0 2 1616 1153 16 4 12 0 4 [0 6 0]]
>   [0 2 1617 1154 16 3 12 23 -15 [0 2 0]]
>   [0 2 1587 1129 16 -4 15 10 7 nil]
> 
> Character code properties: customize what to show
>   name: ARABIC LETTER SEEN
>   general-category: Lo (Letter, Other)
>   decomposition: (1587) ('س')
> 
> [...]
> 
> The kasrah is shown above the sin:
> 
> [...]
> 
> Composed with the following character(s) "ِّ" using this font:
>   xfthb:-PfEd-DejaVu 
> Sans-normal-normal-semicondensed-*-26-*-*-*-*-0-iso10646-1
> by these glyphs:
>   [0 2 1616 6022 0 2 10 27 -15 [1 5 0]]
>   [0 2 1617 1377 29 1 27 10 7 nil]
> 
> Character code properties: customize what to show
>   name: ARABIC LETTER SEEN
>   general-category: Lo (Letter, Other)
>   decomposition: (1587) ('س')

Which seems to clearly indicate that this _is_ font-dependent, right?

Moreover, it seems also to hint on the reason for the issue: the
correct display uses only 2 glyphs, whereas the incorrect display uses
3 glyphs.  Which means -- and that matches my observations on my
systems -- that the "good" font has a precomposed glyph for
shadda-kasrah, while the "bad" font doesn't.  And the composition data
in the latter case indicates that we were told to display the kasrah
below the base character (the descent value is positive).

Can someone please see what HarfBuzz's hb-view produces from these
glyphs, with the same fonts as you see in Emacs?  If hb-view produces
the same display for the same fonts, then it's not an Emacs problem,
and we should ask the HarfBuzz developers what, if anything, HarfBuzz
can do better for the problematic fonts.  And if hb-view does better
than Emacs, then we should ask the HarfBuzz developers to help us
understand what we do incorrectly in this case.