There seems to have been some change that makes parts reach a semi floating-point bug very early, which only happens after rotating the camera, either on the client or on the server.
This makes it very annoying when in first person with a gun, as every time you move your camera there will be random jittering.
I have tested this on a brand new fresh baseplate to ensure it is not some weird error with the team create place we are using, and the same effect occurs.
In the examples I’m going to provide the position is 2057.685, 112.985, -3275.223. Which is not that far away from 0, 0, 0, especially on a map for the game we are developing.
The jittering should not occur, as far as I’m aware this has never happened before at such a low deviation in studs from 0,0,0. This is a clip in the same place but right near 0, 0, 0, where you can see the bug no longer occurs.
Not sure if the same thing causing this bug causes the following but I thought I’d add on anyways that a similar thing happens when you export a model as an .obj file.
At 2057.685, 112.985, -3275.223.
Hey there! Thanks for reporting this. We’re curious if you happen to have a sense of when this jittering started to occur? And also if you don’t mind, can you share your repro files, like the one on the baseplate?
Hello, I’m not quite sure exactly when this started to occur as we only noticed this while working on a new scope system, but most likely a few days ago, as before this was not an issue to my knowledge.
BugRepo.rbxl (69.0 KB)
Also included the same set up part for the scope in the clips.
Hey there. It looks like there aren’t any scrips in your repro file, so we’re unable to scope like you’re demonstrating in your videos. Do you mind attaching a repro file where we can reproduce exactly what you’re doing to demonstrate the jittering? If you’re worried about revealing scripts to the public, please take a look at How to post a Bug Report, specifically the section on Reproduction Files, for how to send a repro file with potentially private content.
It’s not a script related issue causing this, it happens when you move your camera around, you can try it with freecam or just by getting very close to the objects in first person. If for whatever reason you still need the scripts please let me know.
Thanks for the info. I played around with your test case a little bit more, now knowing that you’re using the freecam, and I am able to reproduce what you’re seeing. However, after talking to various people about this, it seems like the general consensus is that this is to be expected when one is zoomed in that close to an object that is that far from the origin. If you’d like to discuss this in more detail, I can try and loop in more appropriate staff members to chat with you.
I am experiencing this even with 70 FoV, at a mere distance of 3000 studs. I’m not sure this is intended. This did not happen until a few days before OP made this post…
All you have to do is walk up to a wall and stare at it while close to it. You don’t need any zoom for this.
Are the instances in question Unions / MeshParts by any chance? Could you share the geometry you’re near there?
I suspect your case is related to having a mesh origin which is highly offset, meaning there’s additional floating point error being added (part.Position - vertex.Position + mesh.Origin is small but the vertex positions in the mesh are quite large making the computation lose precision).
I’ve seen that happen before and it looks exactly like what you show in the video.
You’ll have to DM me a more complete repro including some code that actually matches what you’re showing in the videos if you want this to be investigated further.
I don’t know if you intended to include more in the repro than you did but it only includes the geometry and I’m unable to reproduce the issue trying to infer what the debug code you were using was doing.
Like, I can’t tell exactly what perspective you’re looking from in the video, the geometry in the repro file does not seem to match it. If you could include a script which sets the camera up with a view where the issue reproduces for you that should be sufficient.
