On Thu, 11 Aug 2022 20:51:50 GMT, Marius Hanl <mh...@openjdk.org> wrote:
>> I am also for declaring a constant >> >> >> private static final double EPSILON = .0000001; >> >> >> and using it in on lines 226 and 251. > > Just a question out of curiosity, is there some kind of recommended value for > epsilon? > I remember we also have very small epsilon in `Region` for snapping. This is a *good* question. In this case, we are summing column widths, so we can estimate the worst case scenario by simply adding max errors accumulated after each addition: sum(N * 2 * ε * |w|) where N = number of columns ε = upper bound for rounding error, or epsilon which is about 1e-16 [0] w = column width So, plugging a 100 columns of 200 px wide we get something to the order of 1e-12 .. 1e-11. On a side note, we could have used Unicode here private static final double ε = .0000001; [0] https://en.wikipedia.org/wiki/Machine_epsilon [1] https://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.html ------------- PR: https://git.openjdk.org/jfx/pull/848
