Hello everyone!
Here's a last batch of fixups and optimizations for PhotonDust. I'm now pretty confident in its state, since the issues have slowed down significantly and have been mostly simple regressions. Because of that, I have important announcement:
IMPORTANT: The Legacy Particle System will be removed this upcoming Monday!
This will finalize the initial PhotonDust implementation and a major milestone towards the big performance update! Thank you everyone who participated in the testing :3
There will be more PhotonDust features in the future, but I'll likely refocus on the next tasks for the performance update first, which will be the audio system.
I'll be updating `prerelease` with the latest fixes in a bit.
Oh and this build is compatible with previous!
New Features:
- Add debug statistics for particle state uploads (currently visible in the Debug dialog on the Worlds page)
Optimizations:
- Skip particle buffer submissions when the particle system is empty (based on report by @orange3134, issue #3301)
- Use separate processing queue for uploading particle simulation results for rendering to improve particle smoothness (based on report by @orange3134, issue #3301)
-- This makes sure that the particle simulation results don't fight for processing time with other asset uploads (meshes, textures, materials...)
-- The particle upload queue also has more time for processing. This means with heavy loads it can lower FPS at bit more (we can tweak how much), but will reduce jerkiness of particle systems in some conditions
Fixes:
- Fixed extremely short particle trails (with only two points) being glitchy and stretching too long in some cases (based on report by @nytra, issue #3716)
-- Note: The legacy particle system actually has a glitchy behavior as well, with the second point of the particle not updating smoothly. PhotonDust fixes this, making it animate smoothly on its end. This is intended change, even though it produces small visual difference.
- Fixed regression causing LineEmitters with fixed emission direction resetting the direction to zero (reported by @orange3134, issue #3381)