repro: https://playground.babylonjs.com/#CP2RN9#214
Switch to 6.49.0 to see the difference. On local dev, I managed to repro in 7.40.3, 7.40.2 does not show the bug. Unless this is a breaking change…
Basically, cloning after setting thin instance fouls up the position of the clone. Did not test rotations/scaling, I presume the transform matrix should be borked somehow?
I’m not sure what we should see in the PG, but we fixed a bug lately wrt to mesh cloning and thin instances:
Note that you can test a PG with different versions of Babylon by adding ?version=XXX in the url:
I cloned the mesh (p1), placed it at (1,0,0). Then added a thin instance to mesh at position(2,2,2). Then I cloned same mesh again (p2), placed it at (2,0,0). In 7.40.2, the clones and thin instance are independent of each other, so I can clone and add thin instances w/o affecting each other.
In 7.40.3, the p2 clone’s position is not at (2,0,0), it seems to be offset from the thin instance’s. I have to place p2 at p2.position.copyFromFloats(0,-2,-2); to get identical output as 7.40.2. Also, if you change rotation of the thin instance with const rotmat = BABYLON.Matrix.RotationY(Math.PI);, the result becomes more obvious.
I tried to build a simple repro according to your description:
White is original mesh, red is p1 and green is p2.
p2 is at (2,0,0) in this PG. The difference with 7.40.2 (https://playground.babylonjs.com/?version=7.40.2#CP2RN9#216) is that thin instances are not lost when cloning, so p2 also has an additional thin instance:
Thin instances of p2 are independent from thin instances of mesh, so you can remove/add thin instances to one mesh without affecting the other one.
Before 7.40.3, when you clone a mesh, the behavior is that: if said mesh has thin instances, the clone does not have thin instances. This is the default. I can have 1 million thin instances and clone mesh twice, visually, I will see 1 million+2, right?
After 7.40.3, when you clone a mesh: if said mesh has thin instances, any subsequent cloning of said mesh will also generate thin instances. This is the new behavior. So, if I have 1 million thin instances and clone mesh twice, suddenly, I will see 3 million?! c.f., https://playground.babylonjs.com/#CP2RN9#217
I just want to get the original behavior (pre-7.40.3) back. What should I do such that when I clone a mesh (that already has thin instances), it does not generate visuals I never asked for/wanted?
Yes, it’s the new behavior, because the old one was a bug: thin instances pertain to a mesh, if you clone a mesh, you should have the thin instances too.
If you don’t want them, you can simply set mesh.thinInstanceCount = 0:
Awesome! Thanks, I got it now!
edit: nvm, I figured it out.
Everything is carried over.
There’s no light in the PG you’ve linked, which is why it’s the emissive color that’s defined, not the diffuse color:
You replied before my edit 
. In hindsight, I think this is a breaking change and the doc should show? I have to hit the sack , will do more testing tom, cheers !
I’ve updated the comment of my PR to indicate it could be a breaking change and how to fix it.
So after sleep and testing on my local dev, I made some observations:
a) mesh.thinInstanceCount = 0; vs mesh.clone(name,newParent?,doNotCloneChildren?,clonePhysicsImpostor?,doNotGenThinInstances?)
I’d prefer for a doNotGenThinInstances flag (default:true) on the clone method to bypass the new code/retain old behavior for some potential savings in mem/cycles. What do you think?
b) If cloning is supposed to retain thin instances, then it should also do the same for normal instances?
c) As I kept testing, this question kept popping up: If clones are supposed to retain thin instances, why not just add thin instances to the buffers of the mesh in the first place?
M1 is the master mesh, C1 is the clone. In Fig1: As per 7.40.3, we are retaining thin instances at creation time which is equivalent to Fig2 where the user should be adding to the buffers of M1 if he/she so require thin instances?
Isn’t 1 set of buffers easier to manage and maintain then passing buffers down to clones? I tot that was the whole point of thin instances in the first place? There must be some burning reason I’m missing…hrm…