Some meshes causing crash on file opening since Blender 3.4 #103610

New Issue

Konstantins Visnevskis · 2023-01-03T12:42:43+01:00

2023-01-03 12:42:43 +01:00

System Information
Operating system: Windows-10-10.0.19045-SP0 64 Bits
Graphics card: NVIDIA GeForce RTX 3070 Ti/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 522.30

Blender Version
Broken: version: 3.4.0, branch: blender-v3.4-release, commit date: 2022-12-06 18:46, hash: a95bf1ac01
Worked: 3.3.2

Short description of error
Opening some files in Blender 3.4 was causing a crash. The source was narrowed down to a mesh that might be corrupt in some way, and upon entering Edit Mode crashes 3.3 too, however 3.2 did not crash on opening the file, thus at least allowing to delete such meshes.

Exact steps for others to reproduce the error
Amphitheater-cedar3.blend
Open the file -> Crash

**System Information** Operating system: Windows-10-10.0.19045-SP0 64 Bits Graphics card: NVIDIA GeForce RTX 3070 Ti/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 522.30 **Blender Version** Broken: version: 3.4.0, branch: blender-v3.4-release, commit date: 2022-12-06 18:46, hash: `a95bf1ac01` Worked: 3.3.2 **Short description of error** Opening some files in Blender 3.4 was causing a crash. The source was narrowed down to a mesh that might be corrupt in some way, and upon entering Edit Mode crashes 3.3 too, however 3.2 did not crash on opening the file, thus at least allowing to delete such meshes. **Exact steps for others to reproduce the error** [Amphitheater-cedar3.blend](https://archive.blender.org/developer/F14110663/Amphitheater-cedar3.blend) Open the file -> Crash

Konstantins Visnevskis commented

2023-01-03 12:42:43 +01:00

Added subscriber: @KonstantinsVisnevskis

Philipp Oeser commented

2023-01-03 13:08:04 +01:00

Added subscriber: @lichtwerk

Philipp Oeser commented

2023-01-03 13:08:04 +01:00

Changed status from 'Needs Triage' to: 'Confirmed'

Philipp Oeser commented

2023-01-03 13:08:04 +01:00

Can confirm the crash, but looks like a corrupt mesh indeed, will do a bit of checking...

Philipp Oeser commented

2023-01-03 13:39:40 +01:00

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

Philipp Oeser commented

2023-01-03 13:39:40 +01:00

Added subscriber: @HooglyBoogly

Philipp Oeser commented

2023-01-03 13:39:40 +01:00

Something seems to be off with the CD_MVERT layer of the mesh, it claims to have ~400k verts, but that data layer is missing/borked.

Immediate crash is probably introduced with 12becbf0df which accesses this layer on file load, but as you said, this mesh wasnt really usable in previous versions anyways.

however 3.2 did not crash on opening the file, thus at least allowing to delete such meshes

One way to still get rid of these (via scripting) would probably be to append all meshes (without instancing them to an object) and check len(mesh.vertices) -- if it is zero, you are probably better off dumping the mesh.
(dont call mesh.validate() here, this will also crash).

I am pretty sure this will not be considered a bug, also taking extra security measures in code to always check if CD_MVERT is valid is probably off the table (not worth the overhead?), but maybe @HooglyBoogly has another opinion?

Something seems to be off with the `CD_MVERT` layer of the mesh, it claims to have ~400k verts, but that data layer is missing/borked. Immediate crash is probably introduced with 12becbf0df which accesses this layer on file load, but as you said, this mesh wasnt really usable in previous versions anyways. > however 3.2 did not crash on opening the file, thus at least allowing to delete such meshes One way to still get rid of these (via scripting) would probably be to append all meshes (without instancing them to an object) and check `len(mesh.vertices)` -- if it is zero, you are probably better off dumping the mesh. (dont call `mesh.validate()` here, this will also crash). I am pretty sure this will not be considered a bug, also taking extra security measures in code to always check if `CD_MVERT` is valid is probably off the table (not worth the overhead?), but maybe @HooglyBoogly has another opinion?

Konstantins Visnevskis commented

2023-01-03 13:50:12 +01:00

One way to still get rid of these (via scripting) would probably be to append all meshes (without instancing them to an object) and check len(mesh.vertices)

Not sure on appending - opening the full file alone takes a couple minutes, but this probably will come in handy to automatically filter objects in 3.2 beforehand. Thanks!

> One way to still get rid of these (via scripting) would probably be to append all meshes (without instancing them to an object) and check `len(mesh.vertices)` Not sure on appending - opening the full file alone takes a couple minutes, but this probably will come in handy to automatically filter objects in 3.2 beforehand. Thanks!

Hans Goudey commented

2023-01-03 14:36:02 +01:00

If we know the way a bad mesh like this is created, we can fix it, but just having the file unfortunately isn't enough to point us in a useful direction, besides guessing.

Thr MVert layer is always expected to be valid, that should be ensured everywhere in Blender, that's the goal anyway.

If we know the way a bad mesh like this is created, we can fix it, but just having the file unfortunately isn't enough to point us in a useful direction, besides guessing. Thr MVert layer is always expected to be valid, that should be ensured everywhere in Blender, that's the goal anyway.

Philipp Oeser removed the

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Some meshes causing crash on file opening since Blender 3.4 #103610