Good news, everyone! I've been able to play for a bit, however more thorough testing is required.
Turns out on Python side, there was another incompatibility that needed addressing: [*] Previously, the division operator (/) would perform integer division if both operands were integers. That was something we relied upon. Well, no more with Python 3. Luckily, since Python 2.2 there's the floor division operator (//) that will still do the correct thing. [*] More of a concern is another change in SWIGs behaviour, though. Previously, when calling a wrapped C++ method such as connect_callback(PyObject *callback, PyObject *args = NULL) on Python side with only a single argument, args would indeed be NULL. Right now they get set to Py_None. That in turn leads to issues when actually calling the callback, since Py_None is not a valid argument tuple. I've not yet figured out what is causing this change in behaviour. For now I am setting args to NULL if they come as Py_None, but I am not sure if I fixed all possible locations yet. The better thing would be to find the root cause and restore the expected behaviour. Other than that, my TODO list just keeps growing, as other areas that may need changes come to mind. [*] I.e. using the proper directories for configuration and savegames on Linux systems instead of stuffing everything into ~/.adonthell. [*] And then, just today, gcc 6.1 was released. A couple days after I finally got my hands on 5.3 :-(. To make matters worse, it'll enable C++-14 by default, so there definitely will be more changes required. (And I've yet to try and compile with C++-11 turned on!) But we'll be getting there eventually. Weekend is near and then there are some holidays over the next couple weeks as well, so time should be plenty for once. It's actually a quite fun little project for a change. Just hope I am not boring you with these little progress updates. And yes, I should be able to commit the first part of the work this weekend. Regards, Kai P.S: An invaluable resource for the upgrade to Python 3 is http://python3porting.com/, should somebody ever get into a similar situation. On Wed, Apr 27, 2016 at 11:57 AM, Kai Sterker <kai.ster...@gmail.com> wrote: > Actually, that is an easy one. With SWIG 3.x, integers are no longer > converted to booleans automatically. > > There appears to be a define (SWIG_PYTHON_LEGACY_BOOL) to restore the old > behaviour, but it might be better to change the Python scripts and use the > True/False boolean types present since Python 2.2.1. > > I don't think limiting us to Python newer than 2.2 in this day and age > would be big restriction (2.2.1 was released in April 2002). > > Tonight will see the beast fully running again (I hope!) :-) > > Kai > > P.S.: I am also thinking of adding a bit of a version check to the > configure script, to make sure SWIG and Python are compatible. Here's an > incomplete list I came across right now. Need to check newer Pythons and > then the details for the 2.x series, but it's a start: > > swig version: 2.0.3 2.0.4 > > A. python 2.x + + > B. python 3.1 + + > C. python 3.2 - + > >
_______________________________________________ Adonthell-devel mailing list Adonthell-devel@nongnu.org https://lists.nongnu.org/mailman/listinfo/adonthell-devel
