Having once tried to implement hot-reloading for MSVC, by hijacking the jump thunks used by their incremental linking, static variables was indeed what (pretty quickly) made me throw in the towel, so I'm glad it seems to be as messy as I imagined it would be.

Note that since the extension now has to live alongside the built-in Jolt module in Godot 4.4, I've renamed both the engine and the project settings category, to hopefully make things less ambiguous.

Both of those should migrate automatically, so shouldn't require any action in most cases.

There might be some explanation for needing that, but odd discrepancies between the two physics engines like that should generally be considered bugs, so should still be reported.

I haven't tried it myself, but it does sound like a bug of some sort.

If you're able to, I'd appreciate if you reported this as an issue to Godot's GitHub repository, ideally with a minimal repro project as well.

Not detected in what way? When listening to the body_entered signals of either body?

If you're able to construct some kind of repro for it (even if it's not 100% reproducible) it would be great if you reported this issue on Godot's GitHub.

You'll likely run into problems when running the current release of the extension in Godot 4.4, as there are class name collisions between the two implementations. I need to put out a new release that is explicitly compatible with Godot 4.4.

You can always compare against Godot Physics instead.

They should in theory be largely the same, but I submitted a couple of scary optimizations for the Godot module over the holidays that seem to have caused some regressions, so keep an eye out for any issues.

I'm not involved in those decisions myself, but assuming everything goes well with Jolt then it will most likely become the default, and Godot's own 3D physics engine would be deprecated some time after that.

I haven't taken the time to profile them both side-by-side, but I wouldn't expect a massive difference. The GDExtension bindings didn't add that much overhead last I checked.

What might matter is the compiler used, as Clang seemed to do better at optimizing Jolt, whereas Godot compiles with GCC.

It definitely wouldn't have gotten to this point without the help from @jrouwe.bsky.social, so a big thank you for his continued support.

Also, thank you to everyone who took the time to try out the Godot Jolt extension these past two years, especially the ones who reported issues.

Fair enough! 🙂 It's almost certainly 0.12.0 then.

Congratulations on all the success with PGRC in general. I've been meaning to pick it up and try it for myself. I was a big fan of both JellyCar and Where's My Water back in the day.

Congratulations! It seems like the perfect title for something like the Steam Deck.

I'm a little bit curious what version of Godot Jolt you ended up shipping with, if you happen to know?

Don't forget @jrouwe.bsky.social as well. 🙂 The work that happens on the extension side is very much dwarfed by his work on the Jolt side of things, often to accommodate for the Godot extension.