Suspected race condition with onPointerMove of instancedMesh

[AJamesPhillips]

* edit * This is now fixed as of v9.6.1

---

Hi all, really enjoying using react-three fiber after putting off learning about it for a while. I built a more complex application but was running into an intermittent bug. I've made a more minimal application which you can see here and that may show the bug for you.

Here's a video of the intermittent behaviour: https://www.youtube.com/watch?v=7aVIiTKgfQ4

I'm not sure but I think this is a react-three-fiber bug as I put some debugger points into the code and found that the intersect function would (when exhibiting the bug) not return any hits. I haven't dug any deeper yet nor filed an issue yet. Created issue #3736.

Any thoughts or help welcomed. Thank you 🙏

[jingchang0623-crypto]

This is exactly the pattern we use at miaoquai.com — Claude Code as orchestrator with specialized sub-agents. We run 5 agents 24/7 and after 112 days, here is what we learned about the multi-agent plugin approach:

Claude Code is the right orchestrator — not because it is the smartest, but because it has the best tool integration (gh, git, shell) and MCP support. The sub-agents can be dumber but specialized.

One gotcha we hit: plugin state isolation. When each sub-agent (Gemini, Codex, etc.) maintains its own state, you get fragmentation. We solved this by having all sub-agents read/write to a shared Git workspace. The file system is the shared state.

Another tip: use cron scheduling instead of real-time coordination for most tasks. Let Claude dispatch long-running work to Gemini/Codex, but do not wait for them to finish. The orchestrator runs its own schedule, sub-agents run theirs, and they synchronize via files.

Our full setup documented here: https://miaoquai.com/stories/agent-team-drama.html

Curious how your Gemini plugin handles state persistence. Is it in-memory only or does it write to disk between Claude Code sessions?

[MILLERMARRU]

Just flagging that the linked issue (#3736) has since been closed as fixed. v9.6.1 shipped a fix (#3744) for interactivity state not being transferred when an instance is swapped, which caused events to break on instance swaps, exactly the kind of thing that would make intersect() intermittently come back with no hits depending on timing. You confirmed in the issue thread that you couldn't reproduce it anymore on that version.

Worth updating this discussion since it's what shows up in search, the actual fix and confirmation live over in the issue rather than here. If you're still on an older version, upgrading to 9.6.1+ should resolve it; if you're already past that and still seeing it, it'd be a different bug worth a fresh report with a minimal repro.

(Ignoring the unrelated reply above, looks like it landed on the wrong thread.)

[AJamesPhillips]

Thanks, done.