Hey all,

I'm trying to track down some odd behavior in the boolean node I have in my PDG network. It's an relatively infrequent error. It happens maybe 3 time per 100 work items.

In my network I am subtracting a blobby shape from a tube/column. Occasionally, the boolean will simply not do the subtraction. It doesn't error or specifically fail either but the output from the node is simply the subtraction geo by itself. The geo I wanted is gone. Needless to say this breaks things downstream.

I've looked around the network to see what data might be different on these individual work items (piping the previous PDG output into a non-PDG instance of the HDA). I have tried adding poly-doctor and clean nodes to help catch potential work item specific issues in the topology. I've tried throttling the PDG network to allow fewer concurrent items processed. They have no effect in that the errors still occur on the same set of work items.

Oddly enough, what *DOES* seem to work is a super minor translation in world space. A simple Y translate of +0.01 and the boolean suddenly cuts as expected. However this works for some and not all. Now a different (small) subset of work items will fail.

Anyone have any experience with this? I feel like I'm close to having my pdg pipeline working but this issue has me stumped.

Any help or suggestions appreciated.
Pete

UPDATE from the twitter-verse:

Haven't yet cracked it but I did get a note on twitter that coincident verts between the cutter and the target can cause some boolean weirdness. Still need to investigate but that is my front runner culprit right now. It would account for the relatively low occurrence in the PDG output as well as the mystery nudge in 3D space that seems to “fix” the cut.

I'll report back what i find if anyone (other than me - haha) is curious.

I think the co-incident verts is probably the reason. Maybe detect overlapping points and move the ones on the cutter away from the surface.
You might be better off using Vdbs for this kind of thing to avoid the problem entirely. I don't trust booleans to work correctly unsupervised.