Failure of CI

[ebraminio]

Per travis, harfbuzz tests got break at some point https://travis-ci.org/behdad/harfbuzz/jobs/97483742 it seems to me it is some minor rounding issue that is happenning somewhere but the fact that the local change made this start to fail is unrelated, it is just limited to a Mongolian script test, and it still works on OS X really surprises me and leads me only to this conclusion that perhaps some server side change happened on Linux machines of Travis (could be resulted by some 32/64 bit cpu architecture switch? I don't know if that would be effective or happened even). So perhaps some test rebaselining or compiler option is needed to solve this. Thank you