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

[Stephen Berman]

On Fri, 11 Jan 2019 15:30:55 +0200 Eli Zaretskii <eliz@gnu.org> wrote:

>> From: Stephen Berman <stephen.berman@gmx.net>
>> Cc: craven@gmx.net,  34035@debbugs.gnu.org
>> Date: Fri, 11 Jan 2019 11:54:06 +0100
>> 
>> > This clearly says that Emacs uses DejaVu Sans for this grapheme
>> > cluster, so I wonder what does "switch the font to Symbola" mean in
>> > this case.  Can you tell what you did to switch to Symbola?
>> 
>> I started emacs with -Q, then I selected the "Set Default Font..." entry
>> from the Options menu, clicked on "Symbola Regular" and pressed the
>> select button.  After doing that, `C-u C-x =' on a character in
>> *scratch* shows this:
>> 
>> xft:-UFAS-Symbola-normal-normal-semicondensed-*-15-*-*-*-*-0-iso10646-1 
>> (#x37)
>
> Symbola supports only 10 characters from the Arabic block, so Emacs
> selects a different font for these characters.  

That's what I suspected.  And I've now found out what the crucial
difference seems to be: I went through the fonts installed on my system,
and there are only two families that display Arabic characters: DejaVu
and Amiri, and only DejaVu Sans displays the shadda-kasrah combination
correctly; in particular my default font, DejaVu Sans Mono, displays it
incorrectly.  Surprisingly, Amiri, which is specifically designed for
Arabic, also displays it incorrectly, with the kasrah below the sin.
With all the other fonts, Arabic characters are displayed using either
Amiri or DejaVu, but of the latter, only a few (including Symbola) use
DejaVu Sans, most (even non-monospaced fonts) use DejaVu Sans Mono. 

>> I don't know if it's significant, but it's only first glyph line from
>> the harfbuzz build:
>> [0 2 1587 6022 0 1 6 16 -8 [0 3 0]]
>> that differs from the second glyph line from the master build:
>> [0 2 0 6022 0 -15 -10 13 -11 [-16 2 0]]
>> The second glyph line from the harfbuzz build:
>> [0 2 1587 1377 16 0 16 6 4 nil]
>> is identical to the first glyph line from the master build:
>> [0 2 1587 1377 16 0 16 6 4 nil]
>
> The identical line describes the base character (1587 is its Unicode
> codepoint) in the same font, so it's small wonder that it is the same
> with both shaping engines.  The different lines describe the
> shadda-kasra diacritics.  

That's also what I thought.

>                           (I'm mildly bothered by the fact that the
> harfbuzz branch displays the information in reverse order, it might
> mean some bug there.)

I wondered about that too.

> You can read more about the values displayed there in the doc string
> of composition-get-gstring.

Thanks for the pointer.

Steve Berman