The implementation of C++ types in shiboken has some issues that block adding more advanced types:

Complexity, duplicated types

• Several types represent a C++ type: TypeInfo (Codemodel), TypeParser::Info (done. removed), AddedFunction::TypeInfo. In addition, the code-model related types are eventually used to populate AbstractMetaType
• Parsing of type code snippets happens in several places
  • abstractmetabuilder.cpp:64 parseTemplateType() for AddedFunction
  • AbstractMetaBuilder::translateType (reparses TypeInfo::toString() back into TypeInfo to normalize signatures, apparently)
  • AddedFunction::TypeInfo::fromSignature() for reading XML, handles only limited subset
  • ShibokenGenerator::buildAbstractMetaTypeFromString()

Representation of const

TypeInfo, AbstractMetaType only have isConstant(), which is true for const int *
which technically (and in Clang representation) is not const, (the pointee is). It is thus not possible to differentiate from const int *const or int *const which is truly const. This causes subtle issues:

• Some signatures for querying the function documentation do not match (doc generator)
• No setters are generated for field types like const QWidget * (QStyleOptionViewItem)
• Patterns for matching rejections and modifications are a bit strange ( bool volatile* instead of volatile bool *)

Plans

A lot of needed information is readily available from Clang. So, introducing a type which represents all the information from Clang would be useful

One issue to consider is that doc-generation (namely the creation of XQueries for the WebXML doc extraction from function signatures) requires using unresolved typedefs (GLEnum instead of unsigned).