Fix mesh corruption of CSG by using elalish/manifold #94321

fire · 2024-07-13T17:14:36Z

The goal is to supersede #91748.

~~Ok to review for style. Trying to debug why the materials are broken.~~

Fixes: #43755
Fixes: #58637
Fixes: #41140

31 · 2024-07-20T22:57:05Z

Thanks for the ping on Discord about this. I finally got a chance to take a look. 🙂

I got materials working (it seems), but all I did was put my old code from #91748 (review) back in and change some names to fit the new place where this implementation lives: 31@5eaf7cb.

Admittedly, my testing is still quite limited. Here's a simple scene that is looking good to me: https://gist.github.com/31/a3e597b32ab31b132ddbc488d8eb0324. The stress-test-ish scene with text also still seems to be working fine: https://gist.github.com/31/c3c4bace42d4ca550ae6badf0f2b7cf9. (Maybe even faster? Been a while.)

I like the change of where this logic lives (vs. #91748). However, I'm not following why the implementation details were changed. It seems significantly more complex and I can't spot what it's fixing. Do you have a test case that this helps with?

fire · 2024-07-21T01:08:58Z

As far as I know the implementation details changed because Mesh became MeshGL in manifold. The rest of the changes were a result of this. I didn't like that fact that we are passing an increasingly growing material dictionary so that was removed, but I caused an error.

Thanks for fixing!

fire · 2024-07-21T01:10:36Z

The developers working at manifold worked hard to remove exceptions and thrust, so we were able to vastly reduce the dependencies by 1/3. The changes here were due to api updates.

modules/csg/csg.h

elalish · 2024-07-21T20:46:01Z

Exciting!

fire · 2024-10-15T18:33:33Z

Horray Manifold recently replaced the GLM dependency with linalg.h elalish/manifold#984

fire · 2024-10-16T13:34:18Z

Looking for team approval and style approval.

modules/csg/csg.cpp

smix8 · 2024-10-18T15:43:03Z

Tested this PR a little and while it in general works fine I had a few random instances where for whatever reason certain parts of the mesh would turn out with wrong normals or smoothing.

It would look like this here with the sphere mesh that touches the box mesh with union.

When I change csg properties or move the objects a little it usually disappears. It is hard to reproduce because when I undo all the the changes and have the csg at the original bug state it does not show the issues again. When I try to save the scene to create an MRP it does not show the bug on loading. Any guess what could cause this?

elalish · 2024-10-18T19:37:40Z

It's off by one because our runIndex has an extra element to store the total size. I'd recommend using mesh.runOriginalID.size() instead. This way you can safely do things like: run = {mesh.runIndex[run_i], mesh.runIndex[run_i + 1]}.

elalish · 2024-10-18T19:42:33Z

As for vertProperties, I'd need some more details on what kind of data you're trying to store/extract to help. First guess is you need to remember to duplicate any property vertices that you want to have different values on different triangles. And remember that the results you get will have those duplicates as well. If you are trying to recover a property that isn't really per-vert but a group of faces, I'd recommend using runOriginalID instead of vertProperties.

elalish · 2024-10-18T20:53:00Z

Didn't you make an example of runOriginalID a while ago? If so can you link it?

It's in this thread: #91748 (comment)

fire · 2024-10-18T20:53:45Z

The difference seems to be that the CSGSphere3D and CSGBox3D don't automatically calculate tangents, but CSGMesh3D does.

fire · 2024-10-18T21:04:27Z

@smix8 The artifact where the shading changes when the mesh goes from touching to intersection is because the below list doesn't calculate tangents from the faces. It is possible and would be highly wanted by others and me, but it could be a different feature request after this is merged.

CSGBox3D
CSGCylinder3D
CSGPolygon3D
CSGSphere3D
CSGTorus3D

What do you think about approving this for now?

A workaround is to use CSGMesh3D which does calculate tangents or add a StandardMaterial and set it to vertex shading.

smix8 · 2024-10-18T21:18:49Z

So you mean this is not a regression with this PR? Just something that was already the case in master with those CSG nodes and I just did not encounter it by luck? In that case, yes I think it would be fine to improve that later in another PR.

fire · 2024-10-18T21:26:52Z

I can confirm it happens in Godot Engine 4.3.stable.

Screen.Recording.2024-10-18.at.2.26.22.PM.mov

smix8 · 2024-10-18T21:59:09Z

When testing CSGMesh3D node with a SphereMesh there is a noticeable performance difference between this PR and master using the same build options.

I mean both versions stutter with a Mesh resource but with this PR I can barely control the mouse while dragging. Is this performance difference expected? What is causing it?

fire · 2024-10-18T22:16:02Z

Will need to go over the code with a profiler to see what is the slowdown, but it's late Friday today, so will be delayed for a bit.

smix8

Encountered a few CSG issues while testing this but apparently they all exist in the current master and are not regressions from this PR.

E.g. specific node types had issues or dont work with certain ops which need to be looked at (e.g. hide the non-working options from the inspector if they can't be fixed), same with some performance differences.

Everything outside of that so far worked fine. I would be ok with pushing the other fixes to followup PRs to keep the ball rolling.

lyuma

I'll admit I don't know much about the manifold library itself, but the change seems to largely consolidate code around manifold packing and unpacking, which is good.

The thirdparty code addition is not too crazy, but that's more a question for maintainers to answer.

The main issue I'm worried about is that the process of converting to/from manifold looks at first glance to be a bit costly, and I'm worried it's doing the conversion too often.

lyuma · 2024-10-19T10:25:42Z

servers/rendering/renderer_rd/storage_rd/texture_storage.cpp

@@ -844,7 +844,7 @@ void TextureStorage::texture_2d_layered_initialize(RID p_texture, const Vector<R
 		}
 	}

-	Texture texture;
+	Texture texture = {};


Texture does not have default initializers, which is bad form, but this should be a separate PR.

I don't think we should be adding a single = {} just in this one place since it doesn't improve the overall code quality.

If there is a bug that we're working around or fixing by doing this, I'd like to be explicit about what it is. I double checked all the class members without default initializers and they should all be getting initialized here, so I'm skeptical that this change is needed.

lyuma · 2024-10-19T10:27:36Z

modules/csg/csg.cpp

+	manifold::Manifold brush_a;
+	_pack_manifold(&p_brush_a, brush_a, mesh_materials, p_default_material, p_vertex_snap);


This is called every time the CSGShape3D is dirty, right?

My main feedback here is that I am worried about continually packing and unpacking the manifold for each operation within CSGShape3D::_get_brush() . this seems like it could be a performance bottleneck.

I'd think that the point of this that we can keep each CSGBrush in manifold form and then a merge operation is now as simple as + or ^ or -.

(and separately, we do unpack once per frame if it's dirty just to submit to the graphics API.)

+ ^ are commutative

- is not commutative

I did some cleanup but wasn't able to rewrite the expressions.

lyuma · 2024-10-19T10:36:51Z

modules/csg/csg.cpp

+		mesh.runIndex.push_back(mesh.triVerts.size());
+
+		// Associate the material with an ID.
+		uint32_t reserved_id = r_manifold.ReserveIDs(1);


This is a static function? If we do run this once per CSG operation, if I have 1000 CSG brushes at 60 Hz, it will overflow the static uint32_t after just 20 hours of gameplay. If it's once per face, this uint32_t could overflow within minutes.

Is an overflow of the static ID generator expected / normal?

Is there a way to manage the allocation ourselves? I'm just feeling a bit uneasy about using a global static function even if it isn't hammered every frame.

are we generating 60 000 * 60 * 60 * 20 boolean operations in a nested tree csg operations?

Also this might be an upstream limit until its changed. @elalish Not sure if doubles changed the run id limit

I'd like to understand why you need to generate that many IDs. My first instinct is they're probably not being used as intended if you create that many. Their point is generally to be a way to map back to e.g. materials. I doubt you're creating that many unique materials, right?

@lyuma As far as I know we only generate one runid per material in a mesh. So I don't know how you are calculating this.

lyuma · 2024-10-19T10:38:48Z

modules/csg/csg.cpp

+	if (err != manifold::Manifold::Error::NoError) {
+		print_error(String("Manifold creation from mesh failed:" + itos((int)err)));
+	}


Shouldn't this return a boolean or Error if it fails? Or is it impossible to trigger an error here (this is more like an assertion)

Might be good to use one of the ERR_ macros.

If the geometry is not manifold able I prefer it disappears rather to crash godot. An error message every frame is not fun either.

Uses MeshGL64 for more floating point precision. Co-Authored-By: 31 <[email protected]> Co-Authored-By: Claudio Z <[email protected]>

AThousandShips added feature proposal topic:3d labels Jul 13, 2024

AThousandShips added this to the 4.x milestone Jul 13, 2024

fire force-pushed the vsk-csg-manifold-update-4.3 branch 4 times, most recently from 5a2e7e0 to ccdfea7 Compare July 13, 2024 17:43

fire mentioned this pull request Jul 13, 2024

Build without exceptions elalish/manifold#804

Closed

fire force-pushed the vsk-csg-manifold-update-4.3 branch 6 times, most recently from eb82d74 to e28f331 Compare July 13, 2024 18:24

31 mentioned this pull request Jul 21, 2024

Add a general-purpose CSGConvexHull3D node godotengine/godot-proposals#9933

Open

fire force-pushed the vsk-csg-manifold-update-4.3 branch 3 times, most recently from f4c93df to 3391f6b Compare July 21, 2024 01:38

fire marked this pull request as ready for review July 21, 2024 01:51

fire requested review from a team as code owners July 21, 2024 01:51

fire mentioned this pull request Jul 21, 2024

Add csg boolean operators using elalish/manifold. #91748

Closed

5 tasks

fire commented Jul 21, 2024

View reviewed changes

modules/csg/csg.h Outdated Show resolved Hide resolved

fire force-pushed the vsk-csg-manifold-update-4.3 branch 2 times, most recently from e77e71d to db8d880 Compare July 21, 2024 03:55

fire force-pushed the vsk-csg-manifold-update-4.3 branch from e0697ae to f4eb21a Compare October 11, 2024 03:16

fire requested a review from a team as a code owner October 11, 2024 03:16

fire force-pushed the vsk-csg-manifold-update-4.3 branch 2 times, most recently from 468c689 to 9fb7797 Compare October 16, 2024 13:16

fire force-pushed the vsk-csg-manifold-update-4.3 branch from 9fb7797 to 983b945 Compare October 16, 2024 13:36

AThousandShips reviewed Oct 16, 2024

View reviewed changes

fire force-pushed the vsk-csg-manifold-update-4.3 branch from 5c7147b to e0f20e4 Compare October 16, 2024 13:47

This comment was marked as outdated.

Sign in to view

fire force-pushed the vsk-csg-manifold-update-4.3 branch from e0f20e4 to 0beceb2 Compare October 18, 2024 20:03

This comment was marked as outdated.

Sign in to view

fire force-pushed the vsk-csg-manifold-update-4.3 branch from 0beceb2 to 56394ac Compare October 18, 2024 20:13

This comment was marked as outdated.

Sign in to view

smix8 approved these changes Oct 18, 2024

View reviewed changes

lyuma reviewed Oct 19, 2024

View reviewed changes

Add csg boolean operators using elalish/manifold.

1020518

Uses MeshGL64 for more floating point precision. Co-Authored-By: 31 <[email protected]> Co-Authored-By: Claudio Z <[email protected]>

fire force-pushed the vsk-csg-manifold-update-4.3 branch from 56394ac to 1020518 Compare October 20, 2024 06:13

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix mesh corruption of CSG by using elalish/manifold #94321

Fix mesh corruption of CSG by using elalish/manifold #94321

fire commented Jul 13, 2024 •

edited

Loading

31 commented Jul 20, 2024 •

edited

Loading

fire commented Jul 21, 2024

fire commented Jul 21, 2024 •

edited

Loading

elalish commented Jul 21, 2024

fire commented Oct 15, 2024

fire commented Oct 16, 2024

smix8 commented Oct 18, 2024 •

edited

Loading

This comment was marked as outdated.

This comment was marked as outdated.

elalish commented Oct 18, 2024

elalish commented Oct 18, 2024

This comment was marked as outdated.

This comment was marked as outdated.

This comment was marked as outdated.

elalish commented Oct 18, 2024

fire commented Oct 18, 2024

fire commented Oct 18, 2024 •

edited

Loading

smix8 commented Oct 18, 2024

fire commented Oct 18, 2024

smix8 commented Oct 18, 2024 •

edited

Loading

fire commented Oct 18, 2024

smix8 left a comment •

edited

Loading

lyuma left a comment

lyuma Oct 19, 2024

lyuma Oct 19, 2024

fire Oct 20, 2024 •

edited

Loading

lyuma Oct 19, 2024

fire Oct 19, 2024 •

edited

Loading

elalish Oct 19, 2024

fire Oct 19, 2024

lyuma Oct 19, 2024

fire Oct 19, 2024

		manifold::Manifold brush_a;
		_pack_manifold(&p_brush_a, brush_a, mesh_materials, p_default_material, p_vertex_snap);

Fix mesh corruption of CSG by using elalish/manifold #94321

Are you sure you want to change the base?

Fix mesh corruption of CSG by using elalish/manifold #94321

Conversation

fire commented Jul 13, 2024 • edited Loading

31 commented Jul 20, 2024 • edited Loading

fire commented Jul 21, 2024

fire commented Jul 21, 2024 • edited Loading

elalish commented Jul 21, 2024

fire commented Oct 15, 2024

fire commented Oct 16, 2024

smix8 commented Oct 18, 2024 • edited Loading

This comment was marked as outdated.

This comment was marked as outdated.

elalish commented Oct 18, 2024

elalish commented Oct 18, 2024

This comment was marked as outdated.

This comment was marked as outdated.

This comment was marked as outdated.

elalish commented Oct 18, 2024

fire commented Oct 18, 2024

fire commented Oct 18, 2024 • edited Loading

smix8 commented Oct 18, 2024

fire commented Oct 18, 2024

smix8 commented Oct 18, 2024 • edited Loading

fire commented Oct 18, 2024

smix8 left a comment • edited Loading

Choose a reason for hiding this comment

lyuma left a comment

Choose a reason for hiding this comment

lyuma Oct 19, 2024

Choose a reason for hiding this comment

lyuma Oct 19, 2024

Choose a reason for hiding this comment

fire Oct 20, 2024 • edited Loading

Choose a reason for hiding this comment

lyuma Oct 19, 2024

Choose a reason for hiding this comment

fire Oct 19, 2024 • edited Loading

Choose a reason for hiding this comment

elalish Oct 19, 2024

Choose a reason for hiding this comment

fire Oct 19, 2024

Choose a reason for hiding this comment

lyuma Oct 19, 2024

Choose a reason for hiding this comment

fire Oct 19, 2024

Choose a reason for hiding this comment

fire commented Jul 13, 2024 •

edited

Loading

31 commented Jul 20, 2024 •

edited

Loading

fire commented Jul 21, 2024 •

edited

Loading

smix8 commented Oct 18, 2024 •

edited

Loading

fire commented Oct 18, 2024 •

edited

Loading

smix8 commented Oct 18, 2024 •

edited

Loading

smix8 left a comment •

edited

Loading

fire Oct 20, 2024 •

edited

Loading

fire Oct 19, 2024 •

edited

Loading