System Information
Windows 10 latest update, 1080 GTX, i7 3700K
Blender Version
Broken: (2.8 ef32be25c17)
Exact steps for others to reproduce the error
Check attached mesh and go to edit mode. Blender will crash.
| Clément Foucault (fclem) |
| Vlad Mafteiu Scai (00Ghz) | |
| Sep 12 2018, 12:46 PM |
| Antonio Vazquez (antoniov) |
| Clément Foucault (fclem) |
| Jesse Yurkovich (deadpin) |
| Vlad Mafteiu Scai (00Ghz) |
System Information
Windows 10 latest update, 1080 GTX, i7 3700K
Blender Version
Broken: (2.8 ef32be25c17)
Exact steps for others to reproduce the error
Check attached mesh and go to edit mode. Blender will crash.
I can confirm the bug in Windows:
Read blend: C:\Users\Antonio\Downloads\Watch - issue.blend Assertion failed: bytes_needed <= IMM_BUFFER_SIZE, file c:\myblender\blenderdev\blender\source\blender\gpu\intern\gpu_immediate.c, line 219 Error : EXCEPTION_BREAKPOINT Address : 0x00007FF6D662A465 Module : C:\MyBlender\BlenderBin\bin\Debug\blender.exe
I'm going to debug more to try to find more details.
The problem is in draw_uvs() function.
Running in debug, I can see the memory grows and grows until we get the crash.
It's very weird that value of bm->totloop in line 862 (immBeginAtMost(GPU_PRIM_POINTS, bm->totloop);) is huge: 3.700.704. I think this value is totally wrong for this mesh.
I cannot follow the analysis because I don't know this Blender area.
@Clément Foucault (fclem) I think this is for you, no?
Maybe the mesh is damaged in a previous blender crash.
If you enable the wireframe overlay you'll see it looks to have been subdivided to a very, very high (and very, very unnecessary) level so it seems to really have that many verts/loops. There's no sub-d modifier on the object so it's all real geo.
Another similar issue is T56932 with the same faulting stack and problem; dense meshes don't work well with IMM_BUFFER_SIZE.