Before, models (of traditional lighting models) with specularity/glossiness would be completely flat when exported to metallicRoughness. These changes approximate glossiness (as an inverse of roughness, with specular intensity as a multiplier) both reading from gltf2 and writing to gltf2.
The changes here (which only apply to reading from or writing to pbrSpecularGlossiness) will:
- store and read specular color on `AI_MATKEY_COLOR_SPECULAR ` rather than `AI_MATKEY_GLTF_PBRSPECULARGLOSSINESS_SPECULAR_FACTOR`
- store and read specular texture from `aiTextureType_SPECULAR` rather than `AI_MATKEY_GLTF_PBRSPECULARGLOSSINESS_SPECULARGLOSSINESS_TEXTURE`. Even though pbrSG’s specularGlossiness texture uses the alpha channel for glossiness, it will still work well enough with just the RGB channels of the image
These changes do a better of importing and exporting baseColor colors and textures, as well as diffuse colors and textures (in the case of pbrSpecularGlossiness)
- baseColorFactor will be stored on both `$clr.diffuse` and `$mat.gltf.pbrMetallicRoughness.baseColorFactor`, and will be extracted from `$mat.gltf.pbrMetallicRoughness.baseColorFactor` first, and falling back to `$clr.diffuse`. The behaviour for baseColorTexture is similar
- pbrSG’s diffuseFactor will now only be store on `$clr.diffuse` (overwriting any previous assignments to `$clr.diffuse`, e.g. from metallicRoughness’ baseColorFactor, as diffuseFactor is more analogous to diffuse color than baseColor), and will only extract from `$clr.diffuse`
Text exporters are using string streams to hold the file content first and then write them to the file in a single pass. If for whatever reason the stream has the fail bit set, tellp() will return pos_type(-1), which in turn makes the subsequent write crash - at least on Windows systems. One reason for the stream being in fail state is when its size exceeds 2^31 bytes, even on 64-bit systems (i.e., when very large scenes get exported).
The fix is checking the fail() before even opening the file.
