CC @Kévin Dietrich (kevindietrich).

Is this even a bug? The UV map (CD_MLOOPUV/CD_MTFACE) is being overwritten by a point attribute (with CD_PROP_FLOAT3 type here) so the attribute lookup fails as the UV Map node is requesting a CD_MTFACE (similarly for the vertex color). We could use CD_AUTO_FROM_NAME in the UV Map and Vertex Color nodes to avoid further complicating the attribute lookup logic in the render code, which will fix the issue. But overwriting such standard attributes in Blender is going to make exporters (like Alembic or USD) fail.

Cycles works here because it just requests the attribute by name, with no type constraint.

It could be argued this is a bug in Cycles, though solving it there does not really make it easier for users to set up correct shaders for geometry nodes. As far as I can tell the downside of CD_AUTO_FROM_NAME would be cases where UV maps and vertex loops have the same name. Though Cycles also has that and I think we might as well make it consistent.

In T93179#1256385, @Kévin Dietrich (kevindietrich) wrote:

We could use CD_AUTO_FROM_NAME in the UV Map and Vertex Color nodes to avoid further complicating the attribute lookup logic in the render code, which will fix the issue. But overwriting such standard attributes in Blender is going to make exporters (like Alembic or USD) fail.

It's not clear to me how Alembic and USD are related to this bug. Are you saying that changing the shader nodes would have some negative impact on them? Or that the fact that geometry nodes write to general attributes instead of UV maps and vertex colors is a problem independent of this, and that we should address that as the root cause instead or in addition to this?

If so, I think we should make the shading nodes consistent regardless. The geometry nodes team is aware of the interop issues between general attributes and UV maps and vertex colors, though I'm not sure if there is a concrete plan to tackle that.

In T93179#1258121, @Brecht Van Lommel (brecht) wrote:

It's not clear to me how Alembic and USD are related to this bug. Are you saying that changing the shader nodes would have some negative impact on them? Or that the fact that geometry nodes write to general attributes instead of UV maps and vertex colors is a problem independent of this, and that we should address that as the root cause instead or in addition to this?

I meant the latter, but I was thinking out loud mostly. My two statements were not meant to follow logically each other. I realized while looking into this that changing the UV maps type information with geometry nodes that exporting UV maps will fail in Alembic as it looks for a CD_MLOOPUV of the same name (CD_PROP_FLOAT2 could work, but even that is not well supported by geometry nodes currently).