I want to write a C++ application with Qt, but build a prototype first using Python and then gradually replace the Python code with C++. Is this the right approach?
That depends on your goals. Having done both, I'd recommend you stay with Python wherever possible and reasonable. Although it takes a bit of discipline, it's very possible to write extremely large applications in Python. But, as you find hotspots and things that can be better handled in C++, you can certainly port relevant parts to C++.
Is there any comprehensive documentation about this process or do I have to learn every single component, and if yes, which ones?
Here's what I'd recommend for the various pieces:
EDITOR/IDE: Use any editor/IDE you're comfortable with, but I'd highly recommend one that supports refactoring. If you're comfortable with Eclipse, use it. If you want to mainly go the C++ route and you're not too familiar with any editors, you might be better off with QtCreator. Eric is an extremely good Python IDE with support for refactoring, unless you're going to be doing lots of C++, take a look at it. Even better, its source code is an example of good PyQt usage and practices.
PROCESS:
The quick summary:
- Write your application in Python using PyQt
- When identified as hotspots, convert decoupled Python classes to C++
- Create bindings for those classes using SIP
- Import the newly defined libraries in Python in place of their Python counterparts
- Enjoy the speed boost
General details:
Write the application in Python using PyQt. Be careful to keep a good separation of concerns so that when you need to port pieces to C++ they will be separate from their dependencies. When you finally need to port something to C++, write it in C++/Qt and then create bindings for it using SIP. SIP has a good reference manual on the process, and you have all of PyQt as an example.
DEPLOYMENT:
C++ - For many applications the dependencies are sufficiently simple that it's not too difficult to create an installer using a tool like NullSoft's Installer or InnoSetup.
Python/PyQt - PyQt applications are a bit more difficult to install because of the dependency on Python and its dependence on the presence of the Qt libraries. One person documented his efforts on this post at ARSTechnica. py2exe works pretty well on Windows and should work fine. IME, freeze.py, which comes with the Python source, sometimes has problems determining which shared libraries are truly necessary and will sometimes end up creating a binary whose dependencies aren't present. Py2app can be made to work on Mac OS X.
But worse, however, is the PyQt/Qt licensing. If you are developing a commercial application, you need to have a commercial PyQt (and Qt) license and make sure to prevent the users from easily modifying the source or otherwise writing code against the PyQt/Qt API because of licensing restrictions. Because of that, the PyQt author created a tool called VendorId (although it has a Python license). Within VendorId is a tool called SIB that can be used to create an executable which depends only on the Python interpreter. But, if you're going to go this far, you might want to install a custom Python along with your application.
DISCLAIMER: I haven't used PySide at all, so I'm not sure how it compares to PyQt. Also, note the following warning on their website:
PySide is a work in progress and is not yet suited for application development requiring production-level stability.
But, on a good note, they intend, at least for the initial release to "maintain API compatibility with PyQt." So, aside from the C++ bindings, you could easily switch between the two later.