Load Faster & Use Less RAM

[snake-biscuits]

Use the bpy .foreach_set method with NumPy to speed up mesh assembly. Can also use the new Vertex lookup method from bsp_tool/src/glview.cpp to save some time

VertexReservedClass -> NumPy array will likely be the hardest step

[snake-biscuits]

bpy.types.Mesh.from_pydata might be a better solution¹
Profiling tools do exist,²³ and it is also possible to run Blender "headless" to compare different approaches.
Regardless,: finding a performant way to load large maps (e.g. Apex Legends maps) will make development much easier.

Footnotes

1. b3d.interplanety.org: How to create mesh through the Blender Python API ↩

2. wiki.blender.org: Profiling Python Code ↩

3. devtalk.blender.org: Plugin-Profiler | benchmark / debug your bpy work ↩

[snake-biscuits]

Another option would be to merge meshes for each model,
Blender attaches a lot of metadata to objects, so reducing groups could be helpful

Maintaining the "mesh groups" defined by bsp.LEVEL_INFO for worldspawn should help with visibility etc.
This would allow for hiding or even skipping decals & skybox, as well as clearly detecting transparent geometry & materials

[snake-biscuits]

Another option would be to merge meshes for each model

Specifically, merging meshes into models before generating them, which should improve import speed (theoretically)

[snake-biscuits]

Just wrapped up an update for faster load times in snake-biscuits/bsp_tool#80
Should be able to update to that version of bsp_tool pretty soon
Though I might make it bsp_tool version 0.5.0

[snake-biscuits]

Mesh(...).from_pydata is the way to go (no numpy required)
uvs & vertex colour could get tricky

I do need to take a proper look at vertex colour
Need to check w/ Fifty & others to check VertexReservedX structs are all synced up
That's for another issue though