Question SkinnedMeshRenderer particles sometimes randomly disappear

I'm seeing an infuriating bug in Unity where particles emitted by SkinnedMeshRenderer sometimes just disappear even though both Animator and SkinnedMeshRenderer is set to update while offscreen.

Anyone have any clue what could cause this?

This is in Unity 2019, built-in pipeline, using the regular (Shuriken) ParticleSystem component.

More details in this video of the issue (contains some gameplay spoilers for Eye of the Temple):

Only clue I have is that the ParticleSystem bounds visualization is missing at the same times as the particles are missing. However, since bounds are set to update all the time, I don't know why or what to do about it.

 

Hi Richard, thanks, I hope you're well as well!

I counted particles with x, y, or z coordinates of positions that are either infinite or NaN but this is all zero every frame, also when the particles are visually missing. Are there other stats I could print out which might help illuminate the issue?

I also tried setting the Rate over Time to 1 (it's normally 0) to purposefully break procedural mode of the particle system, but this made no difference.

I also tried changing simulation space from Local to World which some people have said is a workaround, but the particles still go missing regardless (and it changes the visuals significantly).

Interestingly, when the simulation space is set to World, the moment the particles reappear after having been missing, there's a long particle trail, as if the bird had teleported from a past position to the current one. This seems to imply the issue affects the simulation too, not only the rendering.

I'll look into whether it's feasible to submit a bug report.

 

No problem, thank you too!

I've submitted a bug report with the title "ParticleSystem particles disappear randomly (c/o Richard Kettlewell)".

I'll reply again when I receive the bug tracking number.

 

CASE IN-36781

 

More info: By printing out the world space bounds of the ParticleSystemRenderer each frame, I can see that the bounds continuously change (as expected) when the particles show correctly, whereas when the particles are missing, the bounds don't update with the bird's position but keep staying the same.

 

Is this issue the same perhaps?
https://issuetracker.unity3d.com/is...meshrenderer-and-model-is-moved-in-lateupdate

That has a won't fix note saying "Unfortunately this happens due to how the Particle System and Transform system are designed, and to be totally honest - we're not sure how we can fix it."
But the off thing is that the bug was not there in Unity 5 - it was introduced somewhere in between Unity 5.6 and Unity 2019.4.

Is there really no workaround in 2019.4? It might help if I better understood what the exact reason for the bug is (e.g. what is it that determines when the particles appear and disappear), which the resolution note in the issue above unfortunately does not shed any light on.

To be clear, the issue I'm seeing keeps happening if I update the position of the bird in Update instead of LateUpdate, so I'm not 100% sure if it's the same issue.

 

Ok yeah, not the same then, as the bounds are out of date for sometimes long periods of time (many seconds) before suddenly getting updates again.

 

Ok, I think there is an obvious workaround I hasn't noticed.

The particle system had the property "Culling Mode" set to "Pause And Catch-up". After I changed it to "Always Simulate", I can't reproduce the bug any more.

I guess I never noticed this property before. It didn't exist back in Unity 5.6 which I used to author this particle system effect. I guess it was set to "Pause And Catch-Up" by default when the project was upgraded - not so good when that culling mode is buggy.

But I'm happy it seems there is a solution!

 

All right. The implementation must have been different because the bug I saw here in Unity 2019 didn't happen in Unity 5.

Since the world space bounds seem to not update when the particles are not simulating, it's tempting to think the issue is that the particles updating depends on the bounds being in view, but the bounds being in view also depends on the particles updating; a circular dependency which would certainly cause issues like this. But this would be so stupid that I'm not sure it can realistically be the actual cause.