I think I’ve seen this before as an option, but cannot find it neither in Glyphs nor in this forum anymore. I often end up with rudimentary points in interpolated instances due to overlapping shapes in the masters. These overlaps are very precise though, so these points should not appear. please see the following example:
Wow, the LayerGeek does the trick. Never noticed it’s purpose before. Nice one.
Unfortunately I still got the problem with extending shapes, that are totally aligned in the Masters. I cannot wrap my head around what is going on there. I sent you an email with a file and screen shots yesterday (Subject »Interpolation Rounding Error?«)
Recapping this alignment thing: Please see the screen cap (it uses the Show Interpolations Plugin with just some minor tweaks concerning the display, not – nevertheless – any adjustments to the math or code it uses to represent the interpolations)
What I am referring to is still the off-position of lines in interpolations, that are aligned in the masters.
That may be an inherent problem of the interpolation process with a fixed grid. Since both the LSB and all coordinate positions get rounded to integers, little left/right shifts may appear when switching from one instance to the next. What should not happen, however, is the first shift you see in the animated GIF, but I remember I had problems with the rounding of the displayed instances, and I cannot remember how I solved it.
So in general it would be judicious to avoid constructions like these, where overlapping shapes align to each other (here on the right side)? I still want(ed) to refuse to change my habits. It feels awkward to me to just sloppily clash shapes on top of each other (in this case: drawing the top rect so that it overlaps the component just a bit).
But if this is a basic issue with interpolation and rounding, then I’ll try to change this behaviour of course.
Yes, this is rather what I was after. Now I set up a Glyph that has this rounding-thing going on in the reporter Preview, exported and checked the result. The latter seems to be fine. According to the random sample it seems to be just some issue with the reporter then. I can live with that if the export is still fine.