GDEF IVS deltas are not applied to GPOS adjustments?

[simoncozens]

Given the attached Oswald-VF.ttf.gz:

$ hb-shape Oswald-VF.ttf AV --features=-kern
[A=0+492|V=1+505]
$ hb-shape Oswald-VF.ttf AV
[A=0+453|V=1+505]
$ ./target/debug/examples/shape  Oswald-VF.ttf AV
A=0+453|V=1+505

# Getting advance width with variations is OK:

$ hb-shape Oswald-VF.ttf AV --features=-kern --variations=wght=700
[A=0+551|V=1+526]
$ ./target/debug/examples/shape Oswald-VF.ttf AV --features=-kern --variations=wght=700
A=0+551|V=1+526

# However, deltas are not being applied to GPOS adjustments

$ hb-shape Oswald-VF.ttf AV --features=+kern --variations=wght=700
[A=0+517|V=1+526]
$ ./target/debug/examples/shape Oswald-VF.ttf AV --features=+kern --variations=wght=700
A=0+512|V=1+526

As we can see, the font at default location has a 39 unit kern. It looks like Rustybuzz is still applying a 39 unit kern despite the fact that GDEF deltas will apply to the GPOS kerning value. Try it again at wght=200 and we still see a 39 unit kern:

$ ./target/debug/examples/shape Oswald-VF.ttf AV --features=-kern --variations=wght=200
A=0+468|V=1+480
$ ./target/debug/examples/shape Oswald-VF.ttf AV --features=+kern --variations=wght=200
A=0+429|V=1+480

[dfrg]

FWIW, HarfRuzz appears to match HB here so I suspect this is a bug in ttf-parser rather than RustyBuzz.