MAYA-114335 - Extended Python schema API wrappers #2027
Conversation
Goal was to reach unit test parity with C++. The Python wrappers can now implement the full Bullet simulation unit test originally implemented as a C++ USD plugin.
| GfVec2d v = newValue.Get<GfVec2d>(); | ||
| MFnNumericData data; | ||
| MObject dataObj = data.create(MFnNumericData::k2Double); | ||
| MObject dataObj = data.create(MFnNumericData::k2Double); |
There was a problem hiding this comment.
The values directly translated from a Python tuple will be a Vec2d. Do some conversion if assigning to a Vec2f value.
| @@ -231,21 +328,41 @@ void wrapAdaptor() | |||
| .def("RegisterTypedSchemaConversion", &::RegisterTypedSchemaConversion) | |||
| .staticmethod("RegisterTypedSchemaConversion"); | |||
|
|
|||
| class_<UsdMayaSchemaAdaptor, UsdMayaSchemaAdaptorPtr>("SchemaAdaptor") | |||
| class_<UsdMayaAttributeAdaptor>("AttributeAdaptor") | |||
There was a problem hiding this comment.
The attribute adaptor class was moved from below. Unchanged.
| .def("Set", _AttributeAdaptor_Set) | ||
| .def("GetAttributeDefinition", &UsdMayaAttributeAdaptor::GetAttributeDefinition); | ||
|
|
||
| class_<SchemaAdaptorWrapper, boost::noncopyable> c("SchemaAdaptor", boost::python::no_init); |
There was a problem hiding this comment.
The schema adaptor needs to expose its virtual functions if we want the Python wrapper for schema API adaptor to be able to use it as base class in Python-land.
| .def_readonly("materialConversion", &UsdMayaJobImportArgs::ShadingMode::materialConversion); | ||
|
|
||
| c.add_property( | ||
| "assemblyRep", |
There was a problem hiding this comment.
Now returning the full JobImportArgs struct as read-only properties. Some USD classes could not be returned by reference, so a make_getter was used to properly implement the required pass-by-value.
| { | ||
| return boost::python::object( | ||
| return MayaUsdLibSparseValueWriter( |
There was a problem hiding this comment.
Wrapping the sparse value writer because that class can not be passed-by-value, which is the default in Python. See the wrapper include file for details.
| .add_property("convertMaterialsTo", &::getconvertMaterialsTo); | ||
| using namespace boost::python; | ||
|
|
||
| class_<UsdMayaJobExportArgs>("JobExportArgs", no_init) |
There was a problem hiding this comment.
Exporting the whole JobExportArgs struct.
| @@ -80,6 +82,26 @@ static void _SetMayaAttrKeyableState(const std::string& attrPath, const SdfVaria | |||
| UsdMayaReadUtil::SetMayaAttrKeyableState(plug, variability); | |||
| } | |||
|
|
|||
| bool _ReadUsdAttribute( | |||
| const UsdAttribute& usdAttr, | |||
| const MObject& obj, | |||
There was a problem hiding this comment.
Passing a MObject instead of a MFnDependencyNode which has no copy semantics.
| // Python will create copies of classes to prevent dangerous access to ephemeral pointers. This does | ||
| // not work well for USD's UsdUtilsSparseValueWriter because it builds up a memory of all previously | ||
| // set values, which means it must be passed at least by reference. We will wrap the USD class in a | ||
| // way that allows seamless copies in Python land without duplicating the USD class. |
There was a problem hiding this comment.
Of course this means keeping a sparse value writer in Python land past its life scope in C++ can lead to a crash. Should I try using a weak_ptr somehow as a way to survive this?
There was a problem hiding this comment.
That would make it cleaner, if it doesn't make the binding more complex.
There was a problem hiding this comment.
You are right. Too complex and requires the tracked class to be managed via shared_ptr.
| mayaUsdLib.SchemaApiAdaptor.Register(TestBulletRigidBodyShemaAdaptor, "shape", "PhysicsRigidBodyAPI") | ||
|
|
||
| # Build a scene (and exercise the adaptor in a freeform context) | ||
| cmds.file(f=True, new=True) |
There was a problem hiding this comment.
Same code as found in test\lib\usd\translators\testUsdExportSchemaApi.py meaning the Python code can do what the C++ code does.
| @@ -188,7 +188,8 @@ def __testUsdPreviewSurfaceRoundtrip(self, | |||
| imported_path = cmds.getAttr(file_node+".fileTextureName") | |||
| # imported path will be absolute: | |||
| self.assertFalse(imported_path.startswith("..")) | |||
| self.assertEqual(imported_path.lower(), original_path.lower()) | |||
| self.assertEqual(os.path.normpath(imported_path.lower()), | |||
| os.path.normpath(original_path.lower())) | |||
There was a problem hiding this comment.
On some Windows configurations, normpath can be used to make sure both paths use the same directory separators.
Goal was to reach unit test parity with C++. The Python wrappers can now implement the full Bullet simulation unit test originally implemented as a C++ USD plugin.