Different shaping results compared to harfbuzz

[floppyhammer]

Test text:
ح ب ا حبا
It consists of two parts (separated with a space):
حبا
ح ب ا
The second part is basically the first part with every character delimited with space (in order to cancel ligatures).

Related buffer config:

buffer.set_direction(rustybuzz::Direction::RightToLeft);
buffer.set_language(rustybuzz::Language::from_str("ar").unwrap());
buffer.set_script(rustybuzz::script::ARABIC);

Test case 1 (unifont-15.0.01.ttf)
Shaped glyph id:
harfbuzz 56721 56725 56237 35 56720 35 56722 35 56740
rustybuzz 1578 1579 1584 35 1578 35 1579 35 1584

So firstly, I had tested with unifont, and rustybuzz failed to handle the ligatures. I thought I must had got something wrong, but then I tested with another font.

Test case 2 (HONORSansArabicUI-B.ttf)
Shaped glyph id:
harfbuzz 7 33 97 1 5 1 30 1 93
rustybuzz 7 33 97 1 5 1 30 1 93

And the result is correct and conforms to that of harfbuzz.

There might be something off with rustybuzz.

[behdad]

It is probably because Unifont doesn't have OpenType shaping rules for Arabic, and HarfBuzz synthesizes those, implementing fallback Arabic shaping, which rustybuzz currently doesn't. This is documented behavior.

[behdad]

From README:

• No Arabic fallback shaper, since it requires subsetting.

[floppyhammer]

@behdad Thanks for the answer! I didn't realize that.

[RazrFalcon]

Yes, this particular feature is not implemented. I wasn't expecting it to be that common. I guess I would have to figure it out after all.

[LaurenzV]

Could you further explain why implementing this requires subsetting? I don't understand how a shaping feature would require implementing subsetting?

[RazrFalcon]

It doesn't "require" subsetting per se, but it requires part of it. At least this is how it is implemented in HB.
I do not remember details, it was nearly 5 years ago, but I assume this is hb-ot-shaper-arabic-fallback.hh. And as you can see HB creates temporary tables using subsetting code and them passes them to the shaper.
Meaning we have to implement this part of subsetting as well.

[behdad]

It needs the code to serialize SingleSubst and LigatureSubst tables. Alternatively you can implement the logic in code...

[LaurenzV]

Good to know, thanks! I guess it should be doable...