Meshing issue in cylinder

[Wulfsta]

I am encountering an issue in a mesh generated over a cylinder. See the following screenshot of the exported STL:

What information do you need from me to debug this? I could point you at the example sphere in WeekendSlicer (you can actually see this error in the image in the README), but asking you to look there is not a good way to replicate. I don't see Serde on the Tree or Settings to provide these as data. It might not be trivial to produce a piece of code that will replicate this, due to the amount of floating point math done prior to meshing...

[Wulfsta]

This brings up an interesting idea - can meshing be checked via fuzz testing like this? Perform the following:

• Generate some arbitrary tree.
• Intersect this tree with some half plane.
• Evaluate all points that lie on the edge path of that half plane by plugging them into the original tree, and ensure they are zero with some epsilon of error.
• Save any cases that do not satisfy this condition for further evaluation.

[mkeeter]

serde::Serialize is derived on VmData, so you could serialize it (I typically use bincode) and add it to the issue. That's the raw tape being evaluated, so it should be mostly reproducible 🤞🏻 if you also provide the render config (bounds / resolution) .

Mesh fuzzing is definitely possible, but robust meshing is a very hard problem, so I fear that fuzzing would find more issues that we could resolve 😅

For example, there are adversarial cases where you can build a model that will always fail to capture sharp features correctly.

[Wulfsta]

Okay, hopefully this works to replicate! Included in this archive is the STL, the VmData serialized using bincode, and a text file for settings.
bad_cylinder.zip

For example, there are adversarial cases where you can build a model that will always fail to capture sharp features correctly.

I had no idea about this, pretty interesting!

[Wulfsta]

Hey @mkeeter, was the data I provided useful in root causing this? Just want to follow up in case you need more information.

[mkeeter]

I haven't had time to check it out, but the ZIP files look like they should be sufficient – thanks!

[Wulfsta]

If you have any questions about how it was generated, this branch has the code I used.

[mkeeter]

This specific case should be fixed by #232, let me know if you find others.