Make YSort stable

[Pennycook]

Keeps track of the order in which items are collected by
_collect_ysort_children, and uses that order to break
ties between items with similar Y positions.

Closes #33230.

[aaronfranke]

This is extremely likely to conflict with #42282. I'm marking this as a draft for now to avoid accidental merging.

[akien-mga]

This is extremely likely to conflict with #42282. I'm marking this as a draft for now to avoid accidental merging.

And this is extremely likely to be merged first, so there's not reason to do that.
We don't operate with First In, First Merged :)

[akien-mga]

As this is the solution that we discussed and approved on IRC, merging.

[akien-mga]

Thanks! And congrats for your first merged Godot contribution 🎉

[Pennycook]

Thanks! And congrats for your first merged Godot contribution 🎉

Thanks! Hopefully the first of many. 😄

[andriyDev]

This is extremely likely to conflict with #42282. I'm marking this as a draft for now to avoid accidental merging.

Actually, the code for the current version of #42282 does not modify RenderingServer at all, which means these changes should be completely independent and will work on top of each other. Also congrats!

[akien-mga]

Cherry-picked for 3.2.4.

[Skuriles]

Just to give some feedback:
This seem to work in 3.2.4 rc1, but the problem is it only works for tiles when they cover only one cell. If you use singel tile objects which are bigger than one cell only the first one is shown.

Example: first one with tilemaps and player in YDort, second just all as child nodes

[Pennycook]

@Skuriles: Is this something that worked differently before the patch?

I've observed that using multi-cell tiles with YSort is really tricky -- especially if the tiles don't fill the cells completely -- and it's something that I want to take another look at once godotengine/godot-proposals#1769 is implemented.

[Skuriles]

@Pennycook in Godot 3.2.3 it was completely random (some tiles were shown above the other, some not even if they were from same tilemap). No the randomness is gone, but it still does not work for tiles with more than one cell.

Im fact: I had some hope to use tilemaps to just place floor, walls and all static objects from tilemaps, but this won't work. I gave up using Y-Sort with tilemaps and only render the floor and and outer walls by tilemaps.

Everything else will be a StaticBody or similar object.
I didn't event got simple tilemap with kinematicbody with multiples child sprites to work with Y-Sort.
So I will now just handle z-index with trigger and scripts and single objects.

[Calamander]

@Skuriles Have you tried setting Tile Origin to "Bottom Left" for tilemap with bigger tiles?
@Pennycook Checked this in 3.3 rc6. Looks like there is minor inconsistency (probably also exists in earlier revisions):
With two tilemaps with YSort, one parent, one child. When placing new tiles for parent and child, child tiles are sorted behind parent tiles (no matter which tile was placed first - parent or child) until after scene is saved and reloaded.

[Pennycook]

@Skuriles: I think this has been fixed by #48918. You can now set the offset and Y-Sort origin of individual tiles to any value you'd like, and in the testing I've done so far with big tiles everything seems to work.

@Calamander: Do you still see this behavior with the new TileMap editor? I'm having trouble reproducing it there.

[Calamander]

@Pennycook which one is new TileMap? Groud's or 3.3.2?..

[Zireael07]

@Calamander: Groud's, I'm guessing.

[Pennycook]

Yes, sorry -- I meant Groud's. My thinking: if the issue is fixed in the new editor, it might be easier to identify why it isn't working in the old editor.

[Calamander]

@Pennycook Does Godot 4.0 use Vulkan 1.2? I can't seem to run it and my videocard doesn't support Vulkan 1.2 while it does support 1.1 so I guess I can't test it.

[Calinou]

Does Godot 4.0 use Vulkan 1.2?

IIRC, it requires Vulkan 1.0 and optionally uses Vulkan 1.1 features.

However, Intel Haswell IGPs are not supported because they don't fully support Vulkan 1.0 in the first place. Any Intel IGP after Haswell should work (Broadwell and later).

[Calamander]

@Calinou Hmm, I have GeForce 660M (with latest available driver installed). Vulkaninfo shows my api version as 1.1.95 so then it should probably be enough. I initially thought that problem lies with Vulkan because it crashes before opening project manager right after reporting errors about unfreed shaders and shows nothing else, even in verbose mode.

I guess I'll try compiling and running editor myself sometime later to see if I can determine the cause of the crash.