Is skipping ValueFormat1 == 0 in class kerning correct?

[justvanrossum]

In line 250 of getKerningPairsFromOTF.py I see this:

                            elif valueFormat == 0:
                                # valueFormat zero is caused by a value of <0 0 0 0> on a class-class pair; skip these
                                continue

Is this really correct? If valueFormat == 0, no kerning data from the entire subtable is read at all, which makes me wonder what the use of such a subtable is. (I'm looking at a font with a PairPosFormat2 subtable with ValueFormat1 == 0, and I can't quite figure out why it's there and how it should behave.)

[frankrolf]

I wonder if this represents a class-class pair with an explicit zero kerning value and one (or several) non-zero exceptions.

[behdad]

It's possible that ValueFormat2 is nonzero, which means second glyph in pairs is adjusted but no first glyph.

[justvanrossum]

That's not the case here, ValueFormat2 is also 0.

[behdad]

Then yes, that's a degenerate subtable. Is Coverage empty as well?

[justvanrossum]

No, Coverage is not empty, and both ClassDefs are filled, too. It was compiled by AFDKO.

If I do not ignore that subtable (but take those values as 0) I don't get the correct kerning out of the font, so that suggests that getKerningPairsFromOTF.py is doing the right thing. But then, why did AFDKO put it in there...

I may not pursue this further, though, as I have something that seems to works well now. This issue can be labeled "just curious" :)

[behdad]

Then that subtable will act as a stopper. Anything on the coverage gets a zero there and next subtables will NOT be search for those glyphs as the first in a kern pair.

[justvanrossum]

Ok, but then there is after all something wrong with either the font or getKerningPairsFromOTF.py or both.

[behdad]

Not arguing with that :).

[frankrolf]

Just, would it be possible to send over the font? I am curious now!

[justvanrossum]

Emailed you.

[behdad]

I also like to see the font. I read the code and found a few different issues. Filing new issue now.

[behdad]

For reference: #2