SheenCloth sample model crashes gltf_viewer with assertion

[bejado]

Describe the bug Debug builds of gltf_viewer crash when loading the SheenCloth.gltf sample model.

in void utils::panic(const char *, const char *, int, const char *):32
in file ../../libs/utils/src/debug.cpp
reason: ../../libs/gltfio/src/FilamentAsset.cpp:130: failed assertion `info.bindings.size() == 0 || info.flags == flags'

To Reproduce Steps to reproduce the behavior:

1. Build gltf_viewer for debug
2. Download the SheenCloth glTF 2.0 sample model.
3. Load the SheenCloth.gltf model into gltf_viewer

Investigation

A comment above the assertion reads:

    // All bindings for a particular glTF texture must have the same transform function.

However, as seen with SheenCloth.gltf, this isn't always the case.

In the SheenCloth model, sheenColorMap (which wants a sRGB texture), and sheenRoughnessMap (which wants a linear texture) could both map to the same glTF texture.

In that case, one solution is to create two copies of the texture: one sRGB, and one linear. This, however, doubles the required texture memory.

Desktop (please complete the following information):

• OS: macOS
• Backend: All backends

[pixelflinger]

Or maybe we could do the conversion in the shader for one of the input when they are mixed. With spec constants, this could even be "free" when both match.

[bejado]

Yeah, I was thinking of something along those lines could work too. I think this is a somewhat rare edge-case.

[pixelflinger]

maybe in the meantime, we could do something to avoid the crash (and I guess get bad roughnessMap instead), and downgrade this bug to "correctness" from "crash"?

[bejado]

Yeah I'm fine with that