[makeotf] Save space in class-based kerning

[brawer]

Could makeotf possibly save some space when building class-based kerning tables (GPOS type 2, format 2)?

Input feature file:

feature kern {
  pos [T] [comma semicolon period] -100;
} kern;

Output from makeotf v2.0.90:

      <Lookup index="0">
        <!-- LookupType=2 -->
        <LookupFlag value="0"/>
        <!-- SubTableCount=1 -->
        <PairPos index="0" Format="2">
          <Coverage>
            <Glyph value="T"/>
          </Coverage>
          <ValueFormat1 value="4"/>
          <ValueFormat2 value="0"/>
          <ClassDef1>
          </ClassDef1>
          <ClassDef2>
            <ClassDef glyph="comma" class="1"/>
            <ClassDef glyph="period" class="1"/>
            <ClassDef glyph="semicolon" class="1"/>
          </ClassDef2>
          <!-- Class1Count=1 -->
          <!-- Class2Count=2 -->
          <Class1Record index="0">
            <Class2Record index="0">
              <Value1 XAdvance="0"/>
            </Class2Record>
            <Class2Record index="1">
              <Value1 XAdvance="-100"/>
            </Class2Record>
          </Class1Record>
        </PairPos>
      </Lookup>

Wouldn’t the following be equivalent? As in the above version, ClassDef1 would continue to omit entries for glyph class 0 (because for glyphs in the Coverage table, ClassDef1 defaults to 0). However, ClassDef2 would actually assign class 0, so the dummy Class2Records for class 0 would not be needed anymore. There’s one Class2Record for every Class1Record, so the savings might add up in a large font.

      <Lookup index="0">
        <!-- LookupType=2 -->
        <LookupFlag value="0"/>
        <!-- SubTableCount=1 -->
        <PairPos index="0" Format="2">
          <Coverage>
            <Glyph value="T"/>
          </Coverage>
          <ValueFormat1 value="4"/>
          <ValueFormat2 value="0"/>
          <ClassDef1>
          </ClassDef1>
          <ClassDef2>
            <ClassDef glyph="comma" class="0"/>
            <ClassDef glyph="period" class="0"/>
            <ClassDef glyph="semicolon" class="0"/>
          </ClassDef2>
          <!-- Class1Count=1 -->
          <!-- Class2Count=1 -->
          <Class1Record index="0">
            <Class2Record index="0">
              <Value1 XAdvance="-100"/>
            </Class2Record>
          </Class1Record>
        </PairPos>
      </Lookup>

[behdad]

Wouldn’t the following be equivalent? As in the above version, ClassDef1 would continue to omit entries for glyph class 0 (because for glyphs in the Coverage table, ClassDef1 defaults to 0). However, ClassDef2 would actually assign class 0, so the dummy Class2Records for class 0 would not be needed anymore. There’s one Class2Record for every Class1Record, so the savings might add up in a large font.

That would be wrong. Class value 0 means "everything else", so it would match all the other glyphs in the font as well. This is not a problem for ClassDef1, as that one is gated by the Coverage of the lookup, but not ClassDef2.

[brawer]

Thanks for the explanation; closing the issue.