Geometry nodes: Wrong instance order when instancing a collection with more than 30 objects #93285

New Issue

Ingo Clemens · 2021-11-22T12:20:51+01:00

Ingo Clemens commented

2021-11-22 12:20:51 +01:00

System Information
Operating system: macOS-11.4-x86_64-i386-64bit 64 Bits
Graphics card: AMD Radeon Pro Vega 48 OpenGL Engine ATI Technologies Inc. 4.1 ATI-4.5.14

Blender Version
Broken: version: 3.1.0 Alpha, branch: master, commit date: 2021-11-14 00:26, hash: 7e82c840b7

Short description of error
When creating a collection instance with more than 31 separate children the resulting order doesn't match the named order of the objects from the collection, as shown in the attached screenshot 1. It also doesn't matter if the source of the points is a bezier curve or a mesh curve.
The index where the unmatched order occurs seems to depend on the scene and objects. In the attached scene the wrong sorting occurs at index 29. The screenshot 2 from another scene shows that the wrong index is 23. The cause for this is unknown and doesn't seem to be name related.
If all objects exeeding the count of 31 are deleted the correct order is retained.

Exact steps for others to reproduce the error
The attached scene should display the issue as shown in screenshot 1.

Steps based on a default scene:

Create a new collection with more than 31 objects. To better see the order rename the objects in ascending order.
Create a curve with more than 31 points.
Create a new geometry node setup for the curve.
Create a Instance in Points node and insert it to the geometry connection. Activate Pick Instance.
Create a Collection Info node and connect the geometry output with the instance input of the Instance on Points node.
Select the collection with the more than 31 objects.
Activate Separate Children on the Collection Info node.
Check the spreadsheet for the named order of the instances in relation to the curve point indices.

wrong_instance_order.blend

**System Information** Operating system: macOS-11.4-x86_64-i386-64bit 64 Bits Graphics card: AMD Radeon Pro Vega 48 OpenGL Engine ATI Technologies Inc. 4.1 ATI-4.5.14 **Blender Version** Broken: version: 3.1.0 Alpha, branch: master, commit date: 2021-11-14 00:26, hash: `7e82c840b7` **Short description of error** When creating a collection instance with more than 31 separate children the resulting order doesn't match the named order of the objects from the collection, as shown in the attached screenshot 1. It also doesn't matter if the source of the points is a bezier curve or a mesh curve. The index where the unmatched order occurs seems to depend on the scene and objects. In the attached scene the wrong sorting occurs at index 29. The screenshot 2 from another scene shows that the wrong index is 23. The cause for this is unknown and doesn't seem to be name related. If all objects exeeding the count of 31 are deleted the correct order is retained. **Exact steps for others to reproduce the error** The attached scene should display the issue as shown in screenshot 1. Steps based on a default scene: 1. Create a new collection with more than 31 objects. To better see the order rename the objects in ascending order. 2. Create a curve with more than 31 points. 4. Create a new geometry node setup for the curve. 5. Create a **Instance in Points** node and insert it to the geometry connection. Activate **Pick Instance**. 6. Create a **Collection Info** node and connect the geometry output with the instance input of the **Instance on Points** node. 7. Select the collection with the more than 31 objects. 8. Activate **Separate Children** on the **Collection Info** node. 9. Check the spreadsheet for the named order of the instances in relation to the curve point indices. [wrong_instance_order.blend](https://archive.blender.org/developer/F11842897/wrong_instance_order.blend) ![wrong_instance_order_1.png](https://archive.blender.org/developer/F11842898/wrong_instance_order_1.png) ![wrong_instance_order_2.png](https://archive.blender.org/developer/F11842900/wrong_instance_order_2.png)

Ingo Clemens commented

2021-11-22 12:20:51 +01:00

Added subscriber: @iclemens

blender-admin commented

2021-11-22 12:20:51 +01:00

#97643 was marked as duplicate of this issue

blender-admin commented

2021-11-22 12:20:51 +01:00

#96310 was marked as duplicate of this issue

Richard Antalik commented

2021-11-24 08:35:53 +01:00

Added subscriber: @iss

Richard Antalik commented

2021-11-24 08:35:53 +01:00

Changed status from 'Needs Triage' to: 'Needs User Info'

Richard Antalik commented

2021-11-24 08:35:53 +01:00

Checked provided file with build 7e42ae7c1a, which is 1 day older than reported as broken and can't rerpoduce the issue. Please check if problem is 100% reproducible and if this is fixed in latest builds

Checked provided file with build 7e42ae7c1a, which is 1 day older than reported as broken and can't rerpoduce the issue. Please check if problem is 100% reproducible and if this is fixed in latest builds

Ingo Clemens commented

2021-11-24 09:20:48 +01:00

Changed status from 'Needs User Info' to: 'Needs Developer To Reproduce'

Ingo Clemens commented

2021-11-24 09:20:48 +01:00

I just downloaded and installed the latest version: 3.1.0 Alpha, branch: master, commit date: 2021-11-22 23:57, hash: d1a4e043bd
Opened the attached test scene and the problem still persists. I also tried with another scene and can reproduce the issue.
Maybe this issue is specific to system and OS version?

I just downloaded and installed the latest version: 3.1.0 Alpha, branch: master, commit date: 2021-11-22 23:57, hash: `d1a4e043bd` Opened the attached test scene and the problem still persists. I also tried with another scene and can reproduce the issue. Maybe this issue is specific to system and OS version? ![wrong_instance_order.png](https://archive.blender.org/developer/F11965876/wrong_instance_order.png)

Richard Antalik commented

2021-11-24 23:18:57 +01:00

Changed status from 'Needs Developer To Reproduce' to: 'Needs Triage'

Richard Antalik commented

2021-11-24 23:18:57 +01:00

Added subscriber: @ankitm

Richard Antalik commented

2021-11-24 23:18:57 +01:00

It's possible, but at this point I can only speculate, will have to have closer look at implementation, but not sure how this would be handled on GPU exactly. Perhaps check with cycles CPU render.

@ankitm Can you repro?

It's possible, but at this point I can only speculate, will have to have closer look at implementation, but not sure how this would be handled on GPU exactly. Perhaps check with cycles CPU render. @ankitm Can you repro?

Ankit Meel commented

2021-11-25 13:21:13 +01:00

Can't redo
Here's the file I created on 827c5b399e

T93285_my_attempt.blend

![image.png](https://archive.blender.org/developer/F12133276/image.png) Can't redo Here's the file I created on 827c5b399e0b [T93285_my_attempt.blend](https://archive.blender.org/developer/F12133494/T93285_my_attempt.blend)

Ankit Meel commented

2021-11-25 13:21:23 +01:00

Changed status from 'Needs Triage' to: 'Needs User Info'

Ingo Clemens commented

2021-11-25 15:12:04 +01:00

Changed status from 'Needs User Info' to: 'Needs Developer To Reproduce'

Ingo Clemens commented

2021-11-25 15:12:04 +01:00

I just tested with the latest version:

3.1.0 Alpha, branch: master, commit date: 2021-11-25 09:22, hash: 5ffb9b6dc4
(download page hash: 5ffb9b6dc4)

and it still fails with the original scene attached as well as the attached scene from Ankit.
The scene T93285_my_attempt.blend appears to be working because it only has 31 objects in the collection. Which means occupying point indices 0-30.
If I duplicate one of them and give it the next number (32) (point index 31) the ordering issue occurs.
The modified scene is attached.

No failure with:
Operating system: Windows-10-10.0.19042-SP0 64 Bits
Graphics card: NVIDIA GeForce GTX TITAN X/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 471.41
version: 3.1.0 Alpha, branch: master, commit date: 2021-11-25 09:36, hash: 1e2376f41f
(download page hash: 1e2376f41f)

Which indicates that the issue is related to macOS.

Update:
Tested also with an older MacBookPro
Operating system: macOS-10.13.6-x86_64-i386-64bit 64 Bits
Graphics card: NVIDIA GeForce GT 650M OpenGL Engine NVIDIA Corporation 4.1 NVIDIA-10.32.0 355.11.10.10.40.102
3.1.0 Alpha, branch: master, commit date: 2021-11-25 09:22, hash: 5ffb9b6dc4

T93285_still_failing.blend

I just tested with the latest version: 3.1.0 Alpha, branch: master, commit date: 2021-11-25 09:22, hash: `5ffb9b6dc4` (download page hash: 5ffb9b6dc45f) and it still fails with the original scene attached as well as the attached scene from Ankit. The scene T93285_my_attempt.blend appears to be working because it only has 31 objects in the collection. Which means occupying point indices 0-30. If I duplicate one of them and give it the next number (32) (point index 31) the ordering issue occurs. The modified scene is attached. **No failure with:** Operating system: Windows-10-10.0.19042-SP0 64 Bits Graphics card: NVIDIA GeForce GTX TITAN X/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 471.41 version: 3.1.0 Alpha, branch: master, commit date: 2021-11-25 09:36, hash: `1e2376f41f` (download page hash: 1e2376f41f5a) Which indicates that the issue is related to macOS. **Update:** Tested also with an older MacBookPro Operating system: macOS-10.13.6-x86_64-i386-64bit 64 Bits Graphics card: NVIDIA GeForce GT 650M OpenGL Engine NVIDIA Corporation 4.1 NVIDIA-10.32.0 355.11.10.10.40.102 3.1.0 Alpha, branch: master, commit date: 2021-11-25 09:22, hash: `5ffb9b6dc4` ![T93285_still_failing.png](https://archive.blender.org/developer/F12136806/T93285_still_failing.png) [T93285_still_failing.blend](https://archive.blender.org/developer/F12136822/T93285_still_failing.blend)

Ankit Meel commented

2021-11-25 15:47:20 +01:00

Changed status from 'Needs Developer To Reproduce' to: 'Confirmed'

Jacques Lucke commented

2021-11-30 15:51:21 +01:00

Added subscriber: @JacquesLucke

Jacques Lucke commented

2021-11-30 15:51:21 +01:00

I can't reproduce it unfortunately. Can someone check if the incorrect ordering exists in the Collection Info node already?

Ankit Meel commented

2021-12-12 06:57:12 +01:00

The diff:

diff --git a/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc b/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc
--- a/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc
+++ b/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc
@@ -131,12 +131,17 @@ static void node_geo_exec(GeoNodeExecParams params)
       }
       entries.append({handle, &(child_object->id.name[2]), transform});
     }
-
+    for (const InstanceListEntry &entry : entries) {
+      std::cout << "******" << entry.name;
+    }
     std::sort(entries.begin(),
               entries.end(),
               [](const InstanceListEntry &a, const InstanceListEntry &b) {
                 return BLI_strcasecmp_natural(a.name, b.name) <= 0;
               });
+    for (const InstanceListEntry &entry : entries) {
+      std::cout << "******" << entry.name;
+    }
     for (const InstanceListEntry &entry : entries) {
       instances.add_instance(entry.handle, entry.transform);
     }

The output:

******1******2******3******4******5******6******7******8******9******10******11******12******13******14******15******16******17******18******19******20******21******22******23******24******25******26******27******28******29******30******31******32******33******1******2******3******4******5******6******7******8******9******10******11******12******13******14******15******16******18******17******19******20******21******22******23******24******25******26******27******28******29******30******31******32******33

In #93285#1262440, @JacquesLucke wrote:
I can't reproduce it unfortunately. Can someone check if the incorrect ordering exists in the Collection Info node already?

@JacquesLucke so I'd guess yes ?

The diff: ``` diff --git a/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc b/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc --- a/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc +++ b/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc @@ -131,12 +131,17 @@ static void node_geo_exec(GeoNodeExecParams params) } entries.append({handle, &(child_object->id.name[2]), transform}); } - + for (const InstanceListEntry &entry : entries) { + std::cout << "******" << entry.name; + } std::sort(entries.begin(), entries.end(), [](const InstanceListEntry &a, const InstanceListEntry &b) { return BLI_strcasecmp_natural(a.name, b.name) <= 0; }); + for (const InstanceListEntry &entry : entries) { + std::cout << "******" << entry.name; + } for (const InstanceListEntry &entry : entries) { instances.add_instance(entry.handle, entry.transform); } ``` The output: ``` ******1******2******3******4******5******6******7******8******9******10******11******12******13******14******15******16******17******18******19******20******21******22******23******24******25******26******27******28******29******30******31******32******33******1******2******3******4******5******6******7******8******9******10******11******12******13******14******15******16******18******17******19******20******21******22******23******24******25******26******27******28******29******30******31******32******33 ``` > In #93285#1262440, @JacquesLucke wrote: > I can't reproduce it unfortunately. Can someone check if the incorrect ordering exists in the Collection Info node already? @JacquesLucke so I'd guess yes ?

Ankit Meel commented

2022-03-22 06:33:44 +01:00

Added subscribers: @simen, @lichtwerk, @mano-wii, @OmarEmaraDev

Pratik Borhade commented

2022-04-27 15:07:34 +02:00

Added subscribers: @himalaya, @PratikPB2123

Germano Cavalcante commented

2022-04-27 15:52:40 +02:00

If nothing against, I'm tempted to commit the fix mentioned in #96310:

In #96310#1327141, @mano-wii wrote:
After a little more testing the problem seems to be in using less than or equal instead of less than:

diff --git a/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc b/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc
index 68d42444afd..54a061993a3 100644
--- a/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc
+++ b/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc
@@ -120,7 +120,7 @@ static void node_geo_exec(GeoNodeExecParams params)
     std::sort(entries.begin(),
               entries.end(),
               - [ ](const InstanceListEntry &a, const InstanceListEntry &b) {
-                return BLI_strcasecmp_natural(a.name, b.name) <= 0;
+                return BLI_strcasecmp_natural(a.name, b.name) < 0;
               });
     for (const InstanceListEntry &entry : entries) {
       instances.add_instance(entry.handle, entry.transform);

If nothing against, I'm tempted to commit the fix mentioned in #96310: > In #96310#1327141, @mano-wii wrote: > After a little more testing the problem seems to be in using `less than or equal` instead of `less than`: > > ``` > diff --git a/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc b/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc > index 68d42444afd..54a061993a3 100644 > --- a/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc > +++ b/source/blender/nodes/geometry/nodes/node_geo_collection_info.cc > @@ -120,7 +120,7 @@ static void node_geo_exec(GeoNodeExecParams params) > std::sort(entries.begin(), > entries.end(), > - [ ](const InstanceListEntry &a, const InstanceListEntry &b) { > - return BLI_strcasecmp_natural(a.name, b.name) <= 0; > + return BLI_strcasecmp_natural(a.name, b.name) < 0; > }); > for (const InstanceListEntry &entry : entries) { > instances.add_instance(entry.handle, entry.transform); > > ```

Hans Goudey commented

2022-04-27 15:56:52 +02:00

Added subscriber: @HooglyBoogly