difference operation returns random results

[dlednik]

So I have 2 objects. Object A and B which represents holes I want to cut out of A.

When I run difference like below it gives me random results. Sometimes the result is A-B like expected, but sometimes it returns B-A as the result. Is this expected behaviour?
Engine in this case is blender

self.stl.difference(holes_elements[0], engine=self.ENGINE)

So this is what I expect always to get back:

But like I said sometimes it returns this:

It cuts A from B and returns B-A as result:

[dlednik]

I made a workaround in my code for now:
https://github.com/Mambix/co2tools/blob/master/co2tools/stl/builder.py#L171

It just makes a couple of itterations and saves the result with highest mass.
No ideal, but for the size of objects I'm using is working for now.

[MarwinSc]

I'm having the same problem, I tried different blender versions, but the results of the difference operation is always random.

[dlednik]

I also noticed that sometimes even with the correct A-B result returned there are some holes missing in my parts. scad cuts them correctly, but blender misses certain cuts.

[markstor]

I'm having the same issue, I think that problem is that the objects order in Blender is random. My guess is that the objects are alphabetically ordered, and their names are the random temporary stl files generated by trimesh.

I haven't found an elegant way to reference imported stls in Blender, without having to select "blindly" the object.
I attach a modified blender template that solved the issue for me: modified_blender_boolean.txt

I've tested it minimally with Blender 2.91 and Blender 2.79 with the code below

import sys
import time
from pathlib import Path

import trimesh
from trimesh.interfaces.blender import _blender_executable
from trimesh.util import log

blenders_dir = Path(_blender_executable).parent.parent

def test(iterations=10):
    for blender_exe in blenders_dir.rglob("blender.exe"):
        blender_version = next(blender_exe.parent.glob("*2.*")).name
        log.warning("Testing Blender {}".format(blender_version))
        _blender_executable = str(blender_exe)
        good_bools = 0
        start = time.time()
        for _ in range(iterations):  # Montecarlito
            flat_box = trimesh.primitives.creation.box((50, 50, 10))
            tall_box = trimesh.primitives.creation.box((15, 15, 20))

            donut = flat_box.difference(tall_box, engine="blender")
            # donut should be a doughnut, but it can be two boxes
            if donut.extents[0] < 49:
                log.info("Bad boolean")
            else:
                log.info("Good boolean")
                good_bools += 1
        log.warning("Blender {} succes rate {:0.2f}%. {:0.2f} seconds".format(blender_version, good_bools * 100.0 / iterations, time.time() - start))


def change_bool_script(script_path):
    from trimesh.resources import _cache
    log.warning("Changing script to {}".format(script_path))
    with open(script_path, "r") as f:
        _cache[('blender_boolean.py.template', True, False)] = f.read()


if __name__ == "__main__":
    import logging
    logging.basicConfig(level=logging.WARNING, stream=sys.stdout, format="%(message)s")

    test()
    change_bool_script(r"modified_blender_boolean.txt")
    test()

[mikedh]

I believe this was fixed by #1177 so closing for now.

[dlednik]

@mikedh what version will this fix be included in?
I want to test it out with my code :)

[mikedh]

It's in the latest release, if you do a pip install --upgrade trimesh it should get it.