It is very slow but seems to be working. It affines vertex positions,
but normals are not improved. Maybe more work will be done in the future
if that technique becomes more relevant.
Partly based on a prior experiment:
11f716f79d0ade59c0e9547a2333183e96b17b43
- Made a bunch of changes to comply with Godot 4 API
- Use Godot's Vector3i and add the missing stuff with helper functions
- Transvoxel uses custom attributes API, the old way would not work
- Wrap MeshOptimizer in a unique namespace (see build script why)
- Added clang-format file for the module as some rules now differ
- Prevent thirdparty code and lookup tables from being clang-formatted
- Very likely full of runtime bugs that need fixing
With this mode enabled, editing anywhere becomes possible.
All edited blocks are loaded at once when the stream is assigned.
The generator is used on the fly to fill the gaps, instead of being
pre-emptively cached only nearby the viewer to allow editing.
Editing is still on the main thread, so this can cause stalls if
the generator is complex. There is also no controlled limit on how
much data gets loaded, so if edited data is larger than RAM, it will
not be playable. The feature is already quite usable, but more work
may be done for wider viability.
- VoxelGenerator no longer inherit VoxelStream
- VoxelStream is now more focused on files
- Nodes have separate stream and generator properties
- Generators use 2 dedicated threads instead of sharing a single one with streams
- TODO Image.lock() is problematic for multithreading
- TODO Voxel graph can cause RWLock contention if edited while it runs
- TODO Saving generator output no longer works, need to put it back
