[makeotf] Non-deterministic output for GPOS type 2 about afdko HOT 5 CLOSED

While @readroberts is the one to provide a definitive answer, as a user of makeotf, I would claim that the 'kern' feature as provided in the features file excerpt should raise a syntax error because the expectation is that a single integer value follow a pair of glyph names, not a four-element array of X and Y placement and advance values. In other words, the font shouldn't have successfully compiled in the first place.

from afdko.

An array of four kern values is frequently seen in RTL scripts, where both advance width and x-position are modified, here is an example:
pos arAlefMadda arTeh <-80 0 -80 0>;

RTL kerning is usually wrapped into its own lookup:

feature kern {
    lookup LATIN_KERNING {
        include (kern_latin.fea)
    } LATIN_KERNING;

    lookup ARABIC_KERNING {
        lookupflag RightToLeft IgnoreMarks;
        include (kern_arabic.fea)
    } ARABIC_KERNING;
} kern;

I don’t think mixing the both kerning styles in the same lookup yields the desired result.
Also – it is very uncommon to do a baseline- and y-adjustment for a Latin pair, is that what you are trying to achieve?

from afdko.

Thank you, @frankrolf, for the clarification and education.

from afdko.

I imagine that this is an uninitialized variable bug. I suspect that once format B has been used, it is used for subsequent records, even if those use format A. The three values that are not specified are then random. I will look at this.

from afdko.

Fixed in makeotfexe

from afdko.