Page MenuHome

Bmesh boolean modifier fails with co-planar geometry (carve doesn't)
Closed, InvalidPublic


System Information
Windows 10 Radeon HD 4800 series

Blender Version
Broken: 2.79.2 4b206af
Worked: 2.79.2 a3409d3 (or any build prior to carve being removed)

Short description of error
Well, bmesh flat out fails in cases when carve did fine with co-planar geometry.

Exact steps for others to reproduce the error
Just add a boolean modifier to 2 prongs object and that's it



Event Timeline

Howard Trickey (howardt) claimed this task.
Howard Trickey (howardt) triaged this task as Confirmed, Low priority.

This is not something the current bmesh boolean code was intended to handle (coplanar intersections). So one could say that this behavior is "working as intended" (where "intended" = "undefined" in this case). The user can, of course, move one of the planes very slightly to get it into "defined" territory, but understood that that is tedious for the user especially if there are a lot of things that need to move and no easy single command to find and move all of them at once. So I will take this task assigned to me to see if something can be done, but there is not likely to be a quick fix to this.

Is there an easy way to let carve solver stay as it was? Keep bmesh as a main target, but have carve as a backup? Or is it too much a job maintaining the code?

@Artem (Hickz) Carve is no longer maintained which is why it was removed. @Howard Trickey (howardt) Is Carve behaviour defined for coplanar geometry, can that behaviour be copied for BMesh?

Artem (Hickz) added a comment.EditedFeb 12 2018, 4:47 PM

It's not only failing in coplanar geometry, obviously. For example, i needed to boolean that part, bmesh fails any way i try it, carve solves it with 0 non-manifold edges. Take it as an example (in this particular case i wanted to fatten the prongs), i have lots more of those. I'm using blender for jewelry production
this is quite a change for me, actually, and i'm sure i'm not alone on this

Uh, this example works okay with bmesh if you boolean the prongs between themselves prior to booleaning them to the main object.

Mikhail Rachinskiy (alm) closed this task as Invalid.

First case test_bmesh.blend – coplanar geometry, not intended to work with BMesh solver whatsoever.

Second case test_1.blend – geometry joined together before union operation using object join operator, resulted in different shells intersection with each other.
Which actually worked with Carve, but BMesh solver has different behaviour and requires intersecting shells to have uniform volume.
Simply make union operation between all prongs before attempting boolean operation with the main object.

Closing as invalid.

Despite this having been closed, I still intend to spend some cycles seeing if I can make the bmesh solver handle at least some of the coplanar cases. Carve seems to have worked sometimes but not consistently - from what I can read, it is not claimed that it works in those cases but I guess sometimes it did. The algorithm Carve used was completely different from that used in the BMesh code, so pretty hard to just adapt the parts that may have been solving the coplanar problems. As to why not just leave Carve as a backup: existing developers got tire of all the bug reports against Carve (crashes, not working in certain cases); combine that with the general difficulty of dealing with external libraries (dependencies, versions, hard to push fixes upstream, rot when new compiler versions don't work, ...) led to the decision to remove Carve now that BMesh boolean does pretty well in the general (non-coplanar problem) case.