Wrong Allah glyph and ligature rules in the Naskh font about arabic

[roozbehp]

It seems that the lookup for forming Allah-related ligatures is also wrong. Using my own names, it is currently something like this:

lookup 10 { # type=4 flag=0x8
  sub alef.isol lam.init lam.medi shadda_superscriptalef heh.fina by alef_lam_lam_heh;
  sub lam.init lam.medi heh.fina by lam_lam_heh;
}

The first rule is basically eating the shadda and the superscript alef, and the second rule is shortening a lam unnecessarily.

It should instead be something like this:

lookup 10 {
  sub lam.init lam.medi shadda_superscriptalef heh.fina by lam_lam_shadda_superscriptalef_heh;
}

[roozbehp]

/cc @raphlinus

[roozbehp]

The second rule, the current

sub lam.init lam.medi heh.fina by lam_lam_heh;

could probably be replaced by an alternative, that shortens in second lam only when a shadda is present on it, to get words like "lam, lam, shadda, fatha, heh" (pronounced "lellah" in Persian) right. If we do this, we may need an additional glyph, that has a shorter second lam and shadda on that lam, but no superscript alef.

(This specific comment is really just an idea. It's not high priority. But the original report and my first comment are high priority.)

/cc @behdad

[roozbehp]

The lookup flag on lookup 10 may also need to change. Currently, it's set to skip marks, and shadda_superscriptalef is defined as a mark, so it seems to be skipping it.

[roozbehp]

Internal tracker b/21492433

[roozbehp]

Lowering the priority, since we could do a hotfix for the product. We still need a fix, but it's no as urgent anymore.

[khaledhosny]

Note that there are many ways that this ligature can be applied in wrong contexts, like the word خالله (the last part of it is not the name of God and should not use the ligature) and many more. I suggest to either not apply the ligature by default or (if you think it is worth the effort) use some complex contextual rules like the ones used in Amiri (which might still give false positives, only Arabic was considered).

[raphlinus]

@khaledhosny: We're trying to be more careful than that. In our current "hotfix" version, we form the ligature for alif-lam-lam-shadda-daggeralif-ha, but not for alif-lam-lam-ha. The word you pasted does form the ligature on my Mac (Geeza Pro, see http://www.org.uib.no/smi/ksv/ArabicFonts.html), but does not on Android.

There's a good chance it'll make it into the next M developer preview. If you could test it then, it would be deeply appreciated.

[khaledhosny]

I think this should be fine. For completeness, you probably should added a فلله ligature (that is feh-alif-lam-lam-shadda-daggeralif-ha).

[roozbehp]

A related issue found by @behdad: Currently the ligature doesn't have correct MarkLig positioning.

[dougfelt]

OK, version 1.05 of Naskh is in the repo now. MTI says this issue has been fixed by this drop. @roozbehp can you confirm?

[xiangyexiao]

confirmed it is fixed.