[glyf] Support cubic curves

[behdad]

harfbuzz/boring-expansion-spec#41

[behdad]

Can be used with this patch to ufo2ft:

diff --git a/Lib/ufo2ft/preProcessor.py b/Lib/ufo2ft/preProcessor.py
index 1543916a..64c8f0d1 100644
--- a/Lib/ufo2ft/preProcessor.py
+++ b/Lib/ufo2ft/preProcessor.py
@@ -330,13 +330,13 @@ class TTFInterpolatablePreProcessor:
             for func in funcs:
                 func(ufo, glyphSet)
 
-        fonts_to_quadratic(
-            self.glyphSets,
-            max_err=self._conversionErrors,
-            reverse_direction=self._reverseDirection,
-            dump_stats=True,
-            remember_curve_type=self._rememberCurveType and self.inplace,
-        )
+        #fonts_to_quadratic(
+        #    self.glyphSets,
+        #    max_err=self._conversionErrors,
+        #    reverse_direction=self._reverseDirection,
+        #    dump_stats=True,
+        #    remember_curve_type=self._rememberCurveType and self.inplace,
+        #)
 
         # TrueType fonts cannot mix contours and components, so pick out all glyphs
         # that have contours (`bool(len(g)) == True`) and decompose their

tested with the -o=variable output.

With NotoSansArabic-VF, I see 10% size improvement.

[behdad]

See #2989
Currently I'm passing all offcurve point sequences to curveTo, which will interpret them as a "super-Bezier". Not what we want.

[behdad]

HarfBuzz counterpart: harfbuzz/harfbuzz#4107

[behdad]

I've started a spec at:
https://github.com/harfbuzz/boring-expansion-spec/blob/main/glyf1-cubicOutlines.md

The full implementation handling all corner-cases needs more work (and is quite cumbersome). But I like feedback and whether this can move forward.

I like to land the HarfBuzz patch and ship it in HB 7 soon, even if it only handles simple cases.

[behdad]

I also think the qu2cu.cli should have a flag to set Qu2CuPen all_cubic=True if one wants to force the output to be only cubic (could be useful for testing purposes)

I'll add.

cu2qu.cli can also gain an inverse of all_quadratic when I add the support to ufo.py...

[behdad]

how much work would adding support for all_quadratic=False be for cu2qu.ufo.fonts_to_quadratic as well? we can also file an issue and do later if/when needed

Done. Can you look into testing it?

[anthrotype]

both curves_to_quadratic (with all_quadratic=False) and quadratic_to_curves (with all_cubic=False) return list of tuples of either length three (for quads) or lenght 4 (for cubics).. so the caller (be it a pen or the cu2qu.ufo module which also would like to know if it did actually modify a given curve segment) should be able use the lengths of the returned tuples to determine whether or not the function has actually done anything to the input

[anthrotype]

Can you look into testing it?

yes

[behdad]

cu2qu.cli can also gain an inverse of all_quadratic when I add the support to ufo.py...

What should I call that argument? --mixed? Umm.

[anthrotype]

--mixed sounds good

[behdad]

I don't know why test_stats is failing on two bots only.

[behdad]

Nevermind. They are somehow stuck on old code. No idea why. I already fixed the code.

[behdad]

I think this is good to merge now.

[behdad]

Next step would be ufo2ft / fontmake support...

[anthrotype]

ufo2ft / fontmake support...

basically we just need a way to pass down all_quadratic=False to cu2qu fonts_to_quadratic (to get the mixed quad+cubic curves)

[behdad]

ufo2ft / fontmake support...

basically we just need a way to pass down all_quadratic=False to cu2qu fonts_to_quadratic (to get the mixed quad+cubic curves)

Correct..