Earclipping issue

[Blossomical]

there's an issue where trimesh returns a version of the obj without Earclipping, image 1 is the input and image 2 is the output
the only thing that i did here is change the normals and that's all

the code used:

async def FlatNormals(model: discord.Attachment):
    async with manager.session.get(model.url) as response:
        mesh: trimesh.Trimesh = await asyncio.to_thread(
            trimesh.load_mesh, io.BytesIO(await response.read()), force="mesh", file_type="obj"
        )

    def job():
        mesh.vertex_normals = np.tile([0, 0, 1], (len(mesh.vertices), 1))
        mesh.face_normals = np.tile([0, 0, 1], (len(mesh.faces), 1))

        obj_bytes = mesh.export(file_type="obj")
        return io.BytesIO(obj_bytes)

    return await asyncio.to_thread(job) # ignore how i'm using to_thread alot, the bot gets blocked if i don't do this despite being put in an asyncio task

input:

output:

[mikedh]

Thanks for the report! Are you able to reproduce on a model in our tests set? This looked ok to me:

import trimesh
import numpy as np

if __name__ == '__main__':
    mesh = trimesh.load_mesh('models/bunny.ply')

    trimesh.util.attach_to_log()
    
    mesh.vertex_normals = np.tile([0, 0, 1], (len(mesh.vertices), 1))
    mesh.face_normals = np.tile([0, 0, 1], (len(mesh.faces), 1))

    mesh.export('broken.obj')

[Blossomical]

@mikedh for sure! i used models/wallhole.obj since it's an ngon that most likely uses earclipping
output:

input:

this is the model i used before, which is a cutout that uses ngon faces (used .txt as .obj is not supported here)

medkitdeer.obj.txt

[Blossomical]

i tested just the export and it still gave me the same result with the ear clipping issue

mesh = trimesh.load_mesh('medkitdeer.obj')
mesh.export('medkitdeer-output.obj')

the input: 74 KB
the output: 242 KB
despite having no changes