#175 Incorrect math kerning algorithm

[David M. Jones]

This is a sequel to issue #173

The current implementation of get_ot_math_kern() in XeTeXOTMath.cpp has a number of flaws. Attached are

1. opentype_kerning.pdf: a worked example showing where the implementation goes wrong
2. opentype_kerning.patch: a new implementation
3. bad_kerns.pdf: a comparision of XeTeX's rendering with and without my patches compared to the output of Word showing that with the patches, the output matches Word.

These were shared with Sergey Malkin who agreed that the new version of the formula looks correct and confirmed that in the OpenType Spec description of the MathKern algorithm, "take the minimum" means the smallest in the absolute sense.

[Akira Kakuto]

I have applied the patch in the TeX Live: revision 55737.