[spec] Clarify how multiple GlyphClassDef statements are handled

[khaledhosny]

The feature files specification is silent on multiple GlyphClassDef statements. In makeotfexe, the last one overrides all the preceding ones, while in feaLib it is cumulative.

The following file:

table GDEF {
    GlyphClassDef [a], [b], [c], [d];
    GlyphClassDef [e], [f], [g], [h];
    GlyphClassDef [i], [j], [k], [l];
} GDEF;

Produces the following with makeotfexe:

  <GDEF>
    <Version value="0x00010000"/>
    <GlyphClassDef Format="1">
      <ClassDef glyph="i" class="1"/>
      <ClassDef glyph="j" class="2"/>
      <ClassDef glyph="k" class="3"/>
      <ClassDef glyph="l" class="4"/>
    </GlyphClassDef>
  </GDEF>

and the following feaLib:

  <GDEF>
    <Version value="0x00010000"/>
    <GlyphClassDef Format="1">
      <ClassDef glyph="a" class="1"/>
      <ClassDef glyph="b" class="2"/>
      <ClassDef glyph="c" class="3"/>
      <ClassDef glyph="d" class="4"/>
      <ClassDef glyph="e" class="1"/>
      <ClassDef glyph="f" class="2"/>
      <ClassDef glyph="g" class="3"/>
      <ClassDef glyph="h" class="4"/>
      <ClassDef glyph="i" class="1"/>
      <ClassDef glyph="j" class="2"/>
      <ClassDef glyph="k" class="3"/>
      <ClassDef glyph="l" class="4"/>
    </GlyphClassDef>
  </GDEF>

[anthrotype]

I like cumulative. It is in line with the way multiple feature blocks are treated as one, allowing to merge multiple feature files with include statements.

[readroberts]

I like cumulative as well. I vote for fixing the spec, and changing makeotf.