On Mon, 24 Mar 2025 21:30:50 GMT, Andy Goryachev <ango...@openjdk.org> wrote:
>> Yes, sorry, this was addressed in the larger discussion. I've left this >> as-is to keep the PR focus'd on one thing. >> >> The calculation here is using 3 snapped values, and one can reasonably >> assume the result is "nearly" snapped. If this value is used later with a >> ceiling function though, then it might ceil to the next higher value if the >> result is slightly too high due to floating point errors. This is why it >> might be a good idea to adjust how our ceiling functions work in all cases; >> instead of using a tiny epsilon (or ulp), use a much larger value but still >> tiny in terms of pixels (like 1/10000th of a pixel). Any "near" snapped >> values won't accidentally get rounded up to the next higher pixel then when >> ceil is used. > >> Any "near" snapped values won't accidentally get rounded up to the next >> higher pixel > > I like this idea! I agree. This sounds promising. ------------- PR Review Comment: https://git.openjdk.org/jfx/pull/1723#discussion_r2011043278
