mesh_free_data freeing but leaving the now dangling pointer in mesh->runtime seems to be somewhat inviting this class of bugs.
I've seen pip get confused about the exact version of python before, give python -m pip install bpy (could be python3 on linux/mac unsure) a whirl.
Can someone try reproducing it with -t 1 and -t 2?
given the last asan dump was kinda noisy, -t1 asan attached below., my build is a few days old though (28th)
can't repro this, but if either of you is handy with the command prompt, can you try
msiexec /l* blender.log /i blender-4.1.0-windows-x64.msi
and attaching the resulting blender.log here?
asan nabs this one pretty easily.
Read blend: "k:\BlenderGit\2022_lite_msvc\bin\Debug\Dingtotest.blend"
=================================================================
==26996==ERROR:…wait, the unroll is in a loop that concatenates a string? did we have genuine performance concerns there? what!? how?! i don't.....
It's difficult to tell why it properly reports an error Windows, and not other operating systems.
I'm guessing windows does better here since we build python shared there, and extensions are…
I don't know this code well enough, but is eff->pd->seed shared between all threads? could we end up with multiple threads having the same seed?
Py_Finalize destroys the python Interpreter, of course blender is gonna crash..... it's like shooting your self in the foot and complaining it hurts.
The dev doing the collada integration is a volunteer, i have not seen them contribute to other parts of blender, so it be unlikely they'll suddenly start working on USD, that being said, the…