Honestly, I don't know. I've decided that I'm not going to make any more demos. I think the 2 I made are high enough quality and completely done, and will be launching on Steam in a few weeks, so I will just support those of the time being and probably finish my scaler. So I'm planning on doing a full game, which will likely take at least 1 year or more. I have some story and concepts, but I won't be sharing anything until much later.
Anyway, I'm still trying to decide if it would be worth investing in Godot 4.0 for the game. As I said, I think it will take around 1 or 2 years (or around there) meaning it could potentially be a big launch title for Godot 4.0 stable, which would be exciting. But given how broken the alpha is, it might take me longer since basic stuff doesn't work (again you cannot Tab in the inspector, it does nothing, pressing Control + A to create a node doesn't work the first time, menus and tooltips are black unless you drag the whole editor window, etc.). These seem like simple things that will be fixed in a few months. But there are also many rendering errors and glitches, as well as crashes and project corruption. So it's a gamble.
The graphics are substantially better, especially the shadows, they look as good as any other engine now. The global illumination is great, it has some glitches still but you can tweak the settings, it works well enough. All the post processing seems like an improvement. It is a different look. That is why I am not going to port Ella, because it looks completely different, it wouldn't be fair to compare them (or be a flattering comparison, given the stylistic choices I made). But for a new project, the new settings are easier to use, less options to tweak, and they work better. For example, the GI has very little to no light leaking, the SSAO is substantially better quality, etc. And I was not able to do an apples to apples performance comparison, because I couldn't get my demos to work fully. But with what I did get working, it appears that Vulkan can be around 50% faster (I was getting 90 fps on 4.0 on the same scene in 3.x which would get 60 fps). So it's going to be great, I just don't know if it's too soon.
In terms of porting, the code changes are not too bad, and a lot can be fixed with find and replace all. Most are subtle changes (like Spatial is now Node3D, but works the same), some are huge (Tween has been completely rewritten with none of the same functions), some are in-between like Quaternion, which is mostly the same, but methods have been removed and switched to operators. I ported about half of the code of Ella in 3 hours, just enough to fly the camera around. I still can't figure out the Tween functions, but it took another hour to fix the menu, as the UI layout is different and some properties have been renamed. On a full game I would expect the 3.x to 4.0 port to not take longer than a month (or around there give or take). But it could be longer if you have a lot of assets. Materials need to be remade, you might have to reimport textures with different formats (I was having problems with PNG but JPG seems to work perfect). The auto-converter doesn't fix everything, and can break stuff. Which is why I made a new scene in 4.0 fresh and abandoned porting Ella, cause I'm not sure it was fixable.
And the other advantage, for the community, is that if we have people working on real games in 4.0, we can find all the bugs faster and make reports, so that the beta and eventual stable release will be more stable and potentially arrive sooner. Which benefits everyone. But it's a tough call and it will depend on your game. For my case, I know I will need Godot 4.0 at some point (like I said, I'm planning for 1 or 2 years out) so the only question is whether I jump to 4.0 now and deal with the issues, or hold out on 3.x for better stability and productivity, knowing that the port could potentially break a ton of stuff, or take months, or force me to redo work that was already finished (not to mention testing, as when you rewrite code, you have no guarantee that it works 100% anymore). So again, a huge gamble there too.
