New Scale Component?

Version 5:

With the same results as posted above, on a Mac OS 10.9.5, here is the version 4 scale 0 filter:

ScaleVersion4.ffxml

They probably changed the internal precision (not just for that component)...

We didn't change anything in the logic of the Scale component. The change you see is caused by the fact that the Scale component at Scale = 0 is very sensitive to the slightest minuscule changes in sample coordinates.

For example, try the "version 5 scale 0" filter above in Filter Forge 5.007 on different settings of Progressive Preview (e.g. 3 / 4 / 5 passes, Off, Legacy, all with various Degradation settings). You'll see different results - just due to the fact that each preview setting goes through a different codepath.

As soon as you replace Scale = 0 with Scale = 0,000000000000001 or such, the problem disappears and the output becomes consistent across all Progressive Preview settings.

I'm afraid we can't fix or version this. The solution, to me, is to release a new version of the Scale component that disallows Scale ~== 0 by introducing a safeguard limit around 0. Currently, there's already an Epsilon there, but it seems that we need a larger safety buffer.