All committers have signed the CLA.

Oh, interesting! That's a tricky one cause the loop is at the caller. I hope you're right, this is a pretty risky bit. I suppose it'll shave a few nanoseconds per request 😅

Ha, I knew EXACTLY what this was about from the title even before I opened the PR or the diff.

(We do something similar with wrapRoute btw.)

We need to be really careful with this. How have you verified that this is correct? (I had a hard time writing tests for this back in the day.)

I'm going to spend some time in the coming days to verify this works as intended. I'll post an update as soon as possible!

Hey just circling back on this. Did some testing / debugging on my branch and on a branch with versions of Go <1.22 and I'm not seeing the reference issue that you are referring to. Each next variable is referencing a new address per execution as expected. I think the variable copy issue is only for closures but in this case wrapMiddleware doesn't fit the bill.

I would be curious to see how you are able to detect this duplication issue! Let me know if there is a specific way to test this that you had in mind. Maybe I'm missing something here.

In the case of capturing a copy before mutation, I think this portion of wrapRoute handles it? It sends a the copy of next to the returned func from wrapMiddleware.