Hm, yes that IS weird then. It must be another bug then. For some reason I had the internalised groups go into a transform component and it had just some move transforms going in. The output of the transform was all NULLs. It seems to work better now with the geometry internalised ungrouped.
Yes, I am doing the grouping very conciously. Since we already have a data structure of 2 levels even with the groups, we are just trying to keep it manageable. It’s very difficult in GH once you start going over 2 levels of hierarchy, as you have to start duplicating all your transforms and so on. With groups at least it stays manageable. Its the drawback of not having any looping in Shapediver, your data structures have to become much more intricate. But its not really a shapediver issue, it doesn’t make sense until Grasshopper supports it natively.
Now we have hit another roadblock though. I am now displaying the mesh using just the “speed” settings for MeshBrep. But we actually want to output the file for the 3D print, which has to be very high quality. With longer word combinations the computation time of 10s simply isnt enough on your servers. On my computer it calculates the long text in 3.3s though. I have reduced the quality slightly now and then it can just about do it on Shapediver, while locally it executes in 2.2s. I have a i7 9700k, so it seems that the servers you use are quite a bit slower than that (single core). But I would assume that you are using Amazon EC2 as well!?
So it looks like we need to either extend the limit, but which would mean going to the Enterprise account, right? Or we just switch over to Rhino compute and set up our own server on Amazon EC2 and use that, since we only need backend API access anyways and don’t need the viewer or UI elements. I think its a matter of cost in the end and if we can somehow work around the (time) limitations of Shapediver.