Sync to 7.3.0

[LaurenzV]

A couple of points I wasn't sure about / need to discuss, but other than that nothing crazy, luckily:

Status	Commit message	HB
🟢	[MarkBase] Adjust base-finding logic	Link
🟢	[GPOS] Fix indexing in MarkLigPos	Link
🟡	[GSUB] Support SingleSubst in get_glyph_alternates	Link	I guess not relevant? seems more like an API thing, not actually relevant for shaping
🟢	Add Tifinagh to list of both-directions scripts	Link
🟢	[indic] Tighten up base-finding	Link
🟢	[syllabic] Actually clear syllables	Link
🟡	[syllabic] Better fix for previous issue	Link	unsure about this one, our skippy iter only has syllable but not a bool "per_syllable">
🟢	[match_input] Micro-optimize	Link
🟢	[PairPos2] Micro-optimize	Link
🟡	[PairPosFormat2] Micro-optimize and don't kern if class2=0	Link	not sure I can really port this... I remember we ported this in the past and it was a huge pain because ttf-parser doesn't give that much low-level access
🟡	[Ligature] Speed up	Link	do we need to port this since it's behind a compilation flag? probably easier to ignore it
🟡	[Ligature] Minor tweak to recent code	Link
🟡	[Ligature] Micro-optimize	Link
🟡	[Ligature] Micro-optimize more	Link	same as above
🟡	[Ligature] Use slow path if 2 or fewer ligatures	Link
🟡	[GPOS] Optimize iterator reset	Link	we dont have the whole iter thing with init and reset right?
🟡	[layout] Don't init iters successively multiple times	Link	same

[RazrFalcon]

[GSUB] Support SingleSubst in get_glyph_alternates Link I guess not relevant? seems more like an API thing, not actually relevant for shaping

Yep.

[syllabic] Better fix for previous issue Link unsure about this one, our skippy iter only has syllable but not a bool "per_syllable">

Looks like it was added in v4: https://github.com/harfbuzz/harfbuzz/commit/044d7a06db552e1564b8575f4d23798f009d9dde We should add it too. Strange that it works at all without it...

[PairPosFormat2] Micro-optimize and don't kern if class2=0 Link not sure I can really port this... I remember we ported this in the past and it was a huge pain because ttf-parser doesn't give that much low-level access

Let's ignore.

[Ligature] Speed up Link do we need to port this since it's behind a compilation flag? probably easier to ignore it

Looks like it was changed quite a bit in the master. Nice to have, but can leave it for later. Let's add a comment/todo in Rust code explaining why it is different.

[GPOS] Optimize iterator reset Link we dont have the whole iter thing with init and reset right?

No idea... Seems like we never reset a skipping iterator.

[LaurenzV]

Hm, yeah there is definitely something fishy... So this per_syllable thing is only used in the set_syllable function, which in turn is only used in the reset function of the iterator... which for some reason, we don't have, along with init. I searched for its usages, and there are indeed differences in how we interact with the iterator throughout the code. Not sure why it is that way. I guess for now I'll just add a TODO.

[LaurenzV]

There are some more changes regarding how skippy iterators are used, so I'll try to align it in the next PR.

[RazrFalcon]

Got it. I remember it was problematic to implement skipping iterator. Probably because of borrow checker. So no surprises it is different. But per_syllable was still added after my initial port.

[LaurenzV]

I can see why now... the context itself contains a skipping iterator iter_input, and the skipping iterator itself contains a reference to the same context, so it's a circular reference... Yeah this seems pretty hard to deal with. :/

[RazrFalcon]

We should start sending patches to harfbuzz 👀