Apple: Usdmtlx updates to pass usdchecker #3243
Conversation
…n usdchecker. Fix small error in MaterialXFilter.cpp. Update and add unit tests for UsdMtlx
|
Filed as internal issue #USD-10009 |
|
/AzurePipelines run |
|
Azure Pipelines successfully started running 1 pipeline(s). |
| (*outputs)[_Name(mtlxOutput)] = TfToken(nodeName); | ||
| } | ||
| } | ||
| } |
There was a problem hiding this comment.
Could you explain why you deleted the second clause here?
There was a problem hiding this comment.
Firstly, credit to this goes to Dan Knowlton, who did the investigative work to figure this stuff out...
... So currently if there are "dangling" <output> elements at the top level of a document, then those outputs are never created, but the indirection implied by the <output> is resolved directly in the connections between the shader nodes.
In the PR, the "Nodegraphs" prim, is now created at a UsdShadeNodegraph type, which then causes UsdMtlx to create the "dangling" <output> on the "Nodegraphs" prim itself, so we no longer need this map to track the indirection of the outputs.
Thats not the easiest explanation to grok, so here's a super small contrived example....
> cat example.mtlx
<?xml version="1.0"?>
<materialx version="1.38">
<ramplr name="layered_layer2_gradient" type="color3" />
<output name="o_layered_layer2_gradient" type="color3" nodename="layered_layer2_gradient" />
<nodedef name="ND_layerShader" node="layerShader">
<input name="weight_1" type="color3" />
<input name="weight_2" type="color3" />
<input name="weight_3" type="color3" />
<output name="out" type="surfaceshader" />
</nodedef>
<layerShader name="layered_sr" type="surfaceshader">
<!-- An output coming from an "output" outside of an explicit node graph. -->
<input name="weight_2" type="color3" output="o_layered_layer2_gradient" />
</layerShader>
<surfacematerial name="layered" type="material">
<input name="surfaceshader" type="surfaceshader" nodename="layered_sr" />
</surfacematerial>
</materialx>
In USD 24.08 if we convert this to USD we get...
> usdcat example.mtlx
#usda 1.0
def "MaterialX"
{
def "Materials"
{
def Material "layered"
{
color3f inputs:weight_1
color3f inputs:weight_2
color3f inputs:weight_3
token outputs:mtlx:surface.connect = </MaterialX/Materials/layered/ND_layerShader.outputs:surface>
def Shader "ND_layerShader" (
prepend references = </MaterialX/Shaders/ND_layerShader>
)
{
color3f inputs:weight_1.connect = </MaterialX/Materials/layered.inputs:weight_1>
color3f inputs:weight_2.connect = </MaterialX/Materials/layered/NodeGraphs/layered_layer2_gradient.outputs:out>
color3f inputs:weight_3.connect = </MaterialX/Materials/layered.inputs:weight_3>
}
def "NodeGraphs" (
prepend references = </MaterialX/NodeGraphs>
)
{
}
}
}
def "Shaders"
{
def Shader "ND_layerShader"
{
uniform token info:id = "ND_layerShader"
token outputs:surface
}
}
def "NodeGraphs"
{
def Shader "layered_layer2_gradient"
{
uniform token info:id = "ND_ramplr_color3"
color3f outputs:out
}
}
}
Notice here that there is no output in the "Nodegraphs" prim, but instead the inputs:weight_2 on the ND_layerShader prim is directly connected to the layered_layer2_gradient prim inside the "Nodegraphs" prim.
With this patch applied we now get
> usdcat example.mtlx
#usda 1.0
(
defaultPrim = "MaterialX"
metersPerUnit = 0.01
upAxis = "Y"
)
def "MaterialX"
{
def "Materials"
{
def Material "layered"
{
color3f inputs:weight_1
color3f inputs:weight_2
color3f inputs:weight_3
token outputs:mtlx:surface.connect = </MaterialX/Materials/layered/ND_layerShader.outputs:surface>
def Shader "ND_layerShader" (
prepend references = </MaterialX/Shaders/ND_layerShader>
)
{
color3f inputs:weight_1.connect = </MaterialX/Materials/layered.inputs:weight_1>
color3f inputs:weight_2.connect = </MaterialX/Materials/layered/NodeGraphs.outputs:o_layered_layer2_gradient>
color3f inputs:weight_3.connect = </MaterialX/Materials/layered.inputs:weight_3>
}
def "NodeGraphs" (
prepend references = </MaterialX/NodeGraphs>
)
{
}
}
}
def "Shaders"
{
def Shader "ND_layerShader"
{
uniform token info:id = "ND_layerShader"
token outputs:surface
}
}
def NodeGraph "NodeGraphs"
{
color3f outputs:o_layered_layer2_gradient.connect = </MaterialX/NodeGraphs/layered_layer2_gradient.outputs:out>
def Shader "layered_layer2_gradient"
{
uniform token info:id = "ND_ramplr_color3"
color3f outputs:out
}
}
}
Now with the change, that tags the "Nodegraphs" prim as a UsdShadeNodegraph, and thus allows connections through that prim, we add the outputs:o_layered_layer2_gradient, which itself is connected to the layered_layer2_gradient prim, and then the inputs:weight_2 connection on ND_layerShader, is just connected to the new output on the "Nodegraphs" prim.
Because we're directly creating the same output and connection structure as is represented in the MaterialX document we no longer need the additional map to maintain the indirection.
Hopefully this makes sense - sorry for being so wordy :)
There was a problem hiding this comment.
Thanks for the great explanation!
Description of Change(s)
usdcheckercurrently fails some fairly trivial things that can be fixed when converting a MaterialX document usingUsdMtlx.NodeGraphsprim as a UsdShadeNodeGraphto pass the Connectable Shader test.Also fixed an compiler warning in
materialXFilter.cppwhere a variable was being used before initialized.Fixes Issue(s)
usdchecker currently reports errors with layers created from MaterialX documents.