I've now officially spent way WAY too much time looking into this, but here's what I've found:
The problems happen when the polygon clearing code tries to remove a region. In this case, a temporary rounded rectangle with the shape of a pad + clearance. A line in the existing polygon (the remaining part of the partially cleared copper polygon) that SHOULD intersect this polygon does not. So a line which should get a new vertex in the middle stays whole. Then, inside some code I still haven't tried to figure out, the attempt to collect the new set of contours and holes goes wrong because of the missing vertex (the line can't be trimmed right). The reason the lines do not intersect is that the FIRST time the pad was cleared (remember, the first thing I found to avoid this was to fix the multi-clearing of pads), two lines fully crossed in an 'X'. A point in an *integer coordinate system* was chosen to be the crossing point. Each line is now two lines, going to that point. So far, the roundoff isn't that scary. In the trimming, X becomes something like > (with some of the segments deleted). Now the duplicate subtraction comes along to intersect > with /, only the point of the > doesn't actually quite reach the /, so the / doesn't get split into two segments. Then you get problems. Trying to bandaid that causes problems by creating duplicate intersections elsewhere. Part of the reason this happens so often is that there are a lot of lines floating around with verrrry shallow angles just off vertical or horizontal. That causes the rounding of the intersection to be much more significant in the shallow axis. Where do these lines come from? Back to the duplicate pad subtraction... It turns out RoundRect (and most things that call frac_circle) generate slightly wonky polygons. frac_circle doesn't turn quite 90 degrees, so the sides of the pad clearances are all slightly twisted. This caught my eye while I was looking at polygons at about 10000% magnification to see intersection points and I didn't realize the mathematical implications of the asthetically displeasing corners for a while. If you fix RoundRect, you can even get away with some double-pad-subtraction that blew up before, becuase the resulting horizontal and vertical lines are less prone to rounding errors when intersected with other lines. If you have boards that still cause problems after applying the multi- pad-sub patch I sent out, mail them to me and I'll see if this fixes them. (by the way, holes in the mask aren't rounded, so there's another reason to tighten up your masks! A perfectly aligned mask equal to the clearance would *always* leave an arc of copper exposed). -- Ben Jackson AD7GD <[EMAIL PROTECTED]> http://www.ben.com/ _______________________________________________ geda-user mailing list [email protected] http://www.seul.org/cgi-bin/mailman/listinfo/geda-user
