PySide Shiboken Code Injection Semantics

From Qt Wiki
Revision as of 16:25, 14 January 2015 by Maintenance script (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

PySide Shiboken Code Injection Semantics

API Extractor provides the inject-code tag allowing the user to put custom written code to on specific locations of the generated code. Yet this is only part of what is needed to generate proper binding code, where the custom code should be written to depends upon the technology used on the generated binding code.

This is the inject-code tag options that matters to Shiboken.

Conventions

C++ Wrapper

This term refers to a generated C++ class that extends a class from the wrapped library. It is used only when a wrapped C++ class is polymorphic, i.e. it has or inherits any virtual methods.

Python Wrapper

The code that exports the C++ wrapped class to Python. Python wrapper refers to all the code needed to export a C++ class to Python, and Python method/function wrapper means the specific function that calls the C++ method/function on behalf of Python.

Native

This is a possible value for the class attribute of the inject-code tag, it means things more akin to the C++ side.

Target

Another class attribute value, it indicates things more close to the Python side.

inject-code tag

The following table describes the semantics of inject-code tag as used on Shiboken.

Parent Tag Class Position Meaning
value-type, object-type native beginning Write to the beginning of a class wrapper .cpp file, right after the #include clauses. A common use would be to write prototypes for custom functions whose definitions are put on a native/end code injection.
end Write to the end of a class wrapper .cpp file. Could be used to write custom/helper functions definitions for prototypes declared on native/beginning.
target beginning Meaning.
end Meaning.
modify-function native beginning Code here is put on the virtual method override of a C++ wrapper class (the one responsible for passing C++ calls to a Python override, if there is any), right after the C++ arguments have been converted but before the Python call.
end This code injection is put in a virtual method override on the C++ wrapper class, after the call to Python and before dereferencing the Python method and tuple of arguments.
target beginning This code is injected on the Python method wrapper (PyCLASS_METHOD(…)), right after the decisor have found which signature to call and also after the conversion of the arguments to be used, but before the actual call.
end This code is injected on the Python method wrapper (PyCLASS_METHOD(…)), right after the C++ method call, but still inside the scope created by the overload for each signature.
shell beginning Used only for virtual functions. The code is injected when the function does not has a pyhton implementation, then the code is inserted before c++ call.
end Same as above, but the code is inserted after c++ call.
typesystem native beginning Write code to the beginning of the module .cpp file, right after the #include clauses. This position has a similar purpose as the native/beginning position on a wrapper class .cpp file, namely write function prototypes, but not restricted to this use.
end Write code to the end of the module .cpp file. Usually implementations for function prototypes inserted at the beginning of the file with a native/beginning code injection.
target beginning Insert code at the start of the module initialization function (initMODULENAME()), before the calling Py_InitModule.
end Insert code at the end of the module initialization function (initMODULENAME()), but before the checking that emits a fatal error in case of problems importing the module.

Anatomy of Code Injection

To make things clear let’s use a simplified example of generated wrapper code and the places where each kind of code injection goes.

Below is the example C++ class for whom wrapper code will be generated.

From the C++ class, Shiboken will generate a injectcode_wrapper.cpp file with the binding code. The next section will use a simplified version of the generated wrapper code with the injection spots marked with comments.

Noteworthy Cases

The type system description system gives the binding developer a lot of flexibility, which is power, which comes with responsibility. Some modifications to the wrapped API will not be complete without some code injection.

Removing arguments and setting a default values for them

A simple case is when a function have one argument removed, as when the C++ method METHOD is modified to be used from Python as METHOD; of course the binding developer must provide some guidelines to the generator on what to do to call it. The most common solution is to remove the argument and set a default value for it at the same time, so the original C++ method could be called without problems.

Removing arguments and calling the method with your own hands

If the argument is removed and no default value is provided, the generator will not write any call to the method and expect the modify-function – target/beginning code injection to call the original C++ method on its own terms. If even this custom code is not provided the generator will put an #error clause to prevent compilation of erroneus binding code.

Calling the method with your own hands always!

If your custom code to be injected contains a call to the wrapped C++ method, it surely means that you don’t want the generator to write another call to the same method. As expected Shiboken will detect the user written call on the code injection and will not write its own call, but for this to work properly the binding developer must use the template variable %FUNCTION_NAME instead of writing the actual name of the wrapped method/function.

In other words, use

instead of

Code Injection for Functions/Methods

On The Native Side

Notice that this is only used when there is a C++ wrapper, i.e. the wrapped class is polymorphic.

On The Target Side

All the overloads of a method from C++ are gathered together on a single Python method that uses an overload decisor to call the correct C++ method based on the arguments passed by the Python call. Each overloaded method signature has its own beginning and end code injections.

Code Injection for Wrapped Classes

On The Native Side

Those injections go in the body of the CLASSNAME_wrapper.cpp file for the wrapped class.

On The Target Side

Code injections to the class Python initialization function.

Code Injection for Modules

The C++ libraries are wapped as Python modules, a collection of classes, functions, enums and namespaces. Shiboken creates wrapper files for all of them and also one extra MODULENAME_module_wrapper.cpp to register the whole module. Code injection xml tags who have the typesystem tag as parent will be put on this file.

On The Native Side

This works exactly as the class wrapper code injections On The Native Side.

On The Target Side

This is very similar to class wrapper code injections On The Target Side. Notice that the inject code at target/end is inserted before the check for errors to prevent bad custom code to pass unnoticed.

Categories: