Outline a structure for Variable Components

[justvanrossum]

Static font using Static Components

glyf table:

• Glyph:
  • Outline --or--
  • list of Components
• Component:
  • BaseGlyph
  • Transformation

Variable font using Static Components

glyf table:

• Glyph:
  • Outline --or--
  • list of Components
• Component:
  • BaseGlyph
  • Transformation

gvar table:

• Deltas for outlines
• Deltas for the offsets of components

fvar table:

• no hidden axes

Main idea of the proposal

• Variable Component sources are regular glyphs, but are unencoded
• They use hidden axes that are not exposed to the user
• Components determine the designspace location for their sources

Static font using (non-nested) Variable Components

glyf table:

• Glyph:
  • Outline --or--
  • list of Components
• Component:
  • BaseGlyph
  • Transformation
  • Designspace Location --> VaCo table

VaCo table:

• Glyph:
  • list of Designspace Location (one for each Component)

gvar table:

• Deltas for outlines for the sources for variable components

fvar table:

• all axes are hidden

Variable font using Variable Components

glyf table:

• Glyph:
  • Outline --or--
  • list of Components
• Component:
  • BaseGlyph
  • Transformation
  • Designspace Location --> VaCo table

VaCo table:

• Glyph:
  • list of Designspace Location (one for each Component)
• GlyphVariations:
  • deltas for components transformation and Designspace Location (anything not covered by gvar)

gvar table:

• Deltas for outlines --or--
• Deltas for the offsets of components

fvar table:

• all axes used for Variable Components are hidden

To be determined: exactly which transform parameters are needed

RoboCJK uses:

• x
• y
• rotation
• scalex
• scaley
• rcenterx (in component source coordinates)
• rcentery (in component source coordinates)

There are two things to consider:

1. which transform parameters are convenient for editing
2. which transform parameters are needed for a font renderer to interpolate the good path between two masters

I am somewhat uncomfortable with the current RoboCJK parameters, as they seem a little arbitrary.

My proposal would be this:

• use the existing 2x2 matrix + offset in the glyf table
• specify interpolation of the 2x2 through matrix decomposition (interpolate decomposed scale, rotation and skew parmameters)
• specify an additional "center of transformation" (in component source coordinates) in the VaCo table

With this, it is possible to exactly replicate GS-CJK at the masters, but the behavior in-between may sometimes be subtly different. I want to do experiments to figure out how much any such differences are, and whether they are show-stoppers or acceptable.

[davelab6]

Since this isn't a table from MS (all caps) or Apple (all lowercase) I propose Cvar as a mixed case "Component Variation" table. (Obviously this is potentially confused with the existing cvar table ;)

[davelab6]

all axes used for Variable Components are hidden

This suggests to me that we will need a 2nd flag (3rd state) for axes that distinguish "advanced user" axes (hidden by default, but useful for some sophisticated/educated users to interact with directly) from "non-user axes" (which require a higher level UI to interact with, or should never be interacted with.)

[davelab6]

I am somewhat uncomfortable with the current RoboCJK parameters

I hope @raphlinus can review these soon too :)

[davelab6]

There should be an interface between variable components and OpenType features, such as smcp or onum or sups (as David Berlow proposed in http://variationsguide.typenetwork.com/#votf)

(It seems David's proposal for "Variable Unicode IDs" (http://variationsguide.typenetwork.com/#vuid) is provisioned for already, by simply having glyphs with the relevant variable component that are correctly encoded.)

[justvanrossum]

(It seems David's proposal for "Variable Unicode IDs" (http://variationsguide.typenetwork.com/#vuid) is provisioned for already, by simply having glyphs with the relevant variable component that are correctly encoded.)

Yes, I think that's indeed a logical (and pleasant) consequence of this proposal.