It turns out that this change badly broke Rend3. See

https://users.rust-lang.org/t/unexpected-borrowed-data-escapes-outside-of-closure-for-generic/119501/2

(Rend3 was abandoned by its author around the time this went in. But I'm stuck using it, and now I have to fix it.)

@John-Nagle This PR loosens the lifetime requirements on the wgpu side. The change to 'static lifetime in egui makes you require forget_lifetime otherwise everything would work the same as usual.

You'll need to call forget_lifetime before you share renderpasses with egui. See emilk/egui#5149 for background and examples.

I get that, mostly. The worst trouble is near here:

https://github.com/BVE-Reborn/rend3/blob/trunk/rend3/src/graph/graph.rs#L403

This code is trying to manage the lifetime of the render pass and encoder from unsafe code.
I'm still trying to figure this out. See the "SAFETY" comment:

// SAFETY: There are two things which borrow this encoder: the render pass and the node's
// encoder reference. Both of these have died by this point.

That's not an assurance that this is safe code. It's a warning that there are lifetime requirements imposed on the caller. Ones which are changing now. The code in that function does considerable unsafe manipulation of encoder and render pass lifetimes. So changes in this area are likely to cause undefined behavior.

I really don't want to touch that code if I can possibly avoid doing so. Is that possible?

@John-Nagle lifetime requirements only got relaxed and more things are now handled as a runtime error rather than a compile time error which is what allows forget_lifetime to be a safe operation. My expectation is that everything still works fine except that there's some lifetime annotations that can just go away - any lifetime annotation that ties a resource (e.g. pipeline) to the lifetime of a render pass should no longer be needed (and in some cases may cause compile errors).
From the little I know about rend3 via its author the unsafe code in this area is exactly because previously the ownership management was only via lifetime annotation & there was no forget_lifetime. But keeping the old unsafe code should still work fine afaik

The unsafe code in this area is exactly because previously the ownership management was only via lifetime annotation & there was no forget_lifetime.

I think that's about right. It looks to me like the old WGPU lifetime system forced complex unsafe ownership management. I'm trying to figure that out now. I think ordinary Rc and RefCell can do the job now. Thanks.