Le 04/06/2010 16:08, Diederick C. Niehorster a écrit : > And based on your other email announcing the test suite, what is that > fuzz factor exactly? I could imagine big fails that are only a few > pixels.. e.g. -- not being converted to en-dash, or and ft ligatue > failing. Guess (but haven't tried) that output of a given testcase > should be the same given that typesetting should be deterministic, > right? or just moving to a different system would lead to slightly > different output easily (sounds bad intuitively)? Personally, I'd > prefer the human change-detector approach, unless the false negatives > are really common. > On one test, run semi-manually, I got "3" (dunno which unit) as the difference on the "hello world" sample. Will mentioned floating-point and rounding errors. I'd be curious to know more about when this difference comes from (since I naively tend to agree with you that the typesetting should be fully determinisitc), but I'm afraid I don't have the time and necessary background (event the the time to acquire such a background) to invesitgate.
Manuel.
