Building wxPython on Unix or Unix-like Systems ---------------------------------------------- The basic steps for building wxPython for Unix or Unix-like systems are: 1. Compile and/or install glib and gtk+ 2. Compile and/or install wxGTK 3. Compile and install wxPython We'll go into more detail of each of these steps below, but first a few bits of background information on tools. I use a tool called SWIG (http://www.swig.org) to help generate the C++ sources used in the wxPython extension module. However you don't need to have SWIG unless you want to modify the *.i files. I've made several modifications to SWIG specific to wxPython's needs and so the modified sources are included in the wx CVS at .../wxPython/wxSWIG. If you need to modify the *.i files for wxPython then change to this directory and run: configure make (Do not run "make install" as wxswig is run in-place.) You'll then need to change a flag in the setup.py script as described below so the wxPython build process will use SWIG if needed. I use the new Python Distutils tool to build wxPython. It is included with Python 2.0, but if you want to use Python 1.5.2 or 1.6 then you'll need to download and install Distutils 1.0 from http://www.python.org/sigs/distutils-sig/ Okay, now on the the fun stuff... 1. Compile and/or install glib and gtk+ --------------------------------------- A. First of all, check and see if you've already got glib/gtk+ on your system, all the Linux distributions I know of come with it, at least as an option. Look for libglib.* and libgtk.* in your system's standard library directories. You'll also need the headers and config scripts in order to build things that use glib/gtk. Try running gtk-config: gtk-config --version If you have version 1.2.5 or better then you're all set. You can skip to step #2. B. If your system has a binary package mechanism, (RPMs, debs, whatever...) check and see if binaries for glib abd gtk+ are available. Be sure to get the runtime library package as well as the development package, if they are separate. Install them with your package tool, and skip to step #2. C. If all else fails, you can get the source code for glib and gtk+ at http://www.gtk.org/. Fetch the latest of each in the 1.2.x series. Compile and install each of them like this: gzip -d [package].tar.gz | tar xvf - cd [package] ./configure make make install The last step will probably have to be done as root. Also, if your system needs anything done to update the dynamic loader for shared libraries, (such as running ldconfig on Linux) then do it after each library is installed. 2. Compile and/or install wxGTK ------------------------------- A. You can find the sources and RPMs for wxGTK at http://wxwindows.org/, just follow the download links from the nevigation panel. You can also check out a current snapshot of the sources from the CVS server. (Some information about annonymous CVS access is at http://wxwindows.org/cvs.htm.) The advantage of using CVS is that you can easily update as soon as the developers check in new sources or fixes. The advantage of using a released version is that it usually has had more thorough testing done. You can decide which method is best for you. B. You'll usually want to use a version of wxGTK that has the same version number as the wxPython sources you are using. (Another advantage of using CVS is that you'll get both at the same time.) C. If using the RPMs be sure to get both the wxGTK and wxGTK-devel RPMs (at a minimum) and then install them as root. rpm -Uhv wxGTK-2.2.2-0.i386.rpm wxGTK-devel-2.2.2-0.i386.rpm D. If using the sources (either from the tarball or from CVS) then configure it like this: cd wxWindows # or whatever your top-level directory is called mkdir build cd build ../configure --with-gtk There are gobs and gobs of options for the configure script, run ../configure --help to see them all. I'll describe some that I find useful here. If you have OpenGL or compatible libraries installed, then add the --with-opengl flag. If you are on Solaris and are using a recent version of GCC, then you'll probably want to add the --enable-permissive flag so the compiler won't barf on your broken X11 header files. To make a debugging version of wxGTK, add the --enable-debug flag. This sets the -g flag for the compiler and also activates some special debugging code in wxWindows by defining the __WXDEBUG__ macro. You'll get some extra asserts, failure logging, etc. To make a static library and not make a shared library, use the --disable-shared and --enable-static flags. NOTE: There is a potential type mismatch between Python and wxGTK. This happens if Python defines some flags that turn on 64-bit file offset support and wxGTK does not. This causes some basic types, like off_t, to be typedef'd differently causing the C++ method signatures to be incompatible and giving link errors at runtime. If you get errors upon running a wxPython script that looks something like this: SeekI_13wxInputStream10wxSeekMode: referenced symbol not found then that is probably the issue. This can be fixed in the current code by predefining these flags before wxGTK's configure is run, for example: export CFLAGS="-D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -DHAVE_LARGEFILE_SUPPORT" export CXXFLAGS=$CFLAGS ../configure --with-gtk --with-opengl --enable-debug In the 2.3.3 final release there will be a real configure flag for it, and it should be enabled by default. You will be able to use --enable-largefile or --disable-largefile to control it. If you still get this or a similar error with 2.3.3 then try disabling largefile support in wxGTK. E. Now just compile and install. You need to use GNU make, so if your system has something else get GNU make and build and install it and use it instead of your system's default make command. make make install The last step will probably have to be done as root. Also, if your system needs anything done to update the dynamic loader for shared libraries, (such as running ldconfig on Linux) then do it now. F. You can test your build by changing to one of the directories under build/samples or build/demos, running make and then running the executable that is built. 3. Compile and install wxPython ------------------------------- A. You have the same options (and same advantages/disadvantages) for getting the wxPython source, either a released snapshot or from CVS. The released version file is named wxPython-[version].tar.gz and is available at http://wxpython.org/download.php. If you want to use CVS you'll find wxPython in the wxWindows CVS tree (see above) in the wxWindows/wxPython directory. B. As mentioned previouslly, wxPython is built with the standard Python Distutils tool. If you are using Python 2.0 or later you are all set, otherwise you need to download and install Distutils 1.0 from http://www.python.org/sigs/distutils-sig/. On Unix systems Distutils figures out what commands and flags to use for the compiler and linker by looking in the Makefile that was used to build Python itself. Most of the time this works okay. If it doesn't, there doesn't seem to be a way to override the values that Distutils uses without hacking either Distutils itself, or Python's Makefile. (Complain to the distutils-sig about this please.) For example, on a Solaris system I had to edit /usr/local/lib/python1.5/config/Makefile and replace LDSHARED=ld -G with LDSHARED=gcc -G This particular problem has been fixed in Python 1.6 and beyond, but there may be similar issues on other platforms. While we're on the subject of how Python was built... Since wxPython is a C++ extension some platforms and/or compilers will require that the Python executable was linked with the C++ linker in order for everything to work correctly. If you build and install Python yourself then this is easy to take care of, otherwise you may have to mess with binary packages or bribe your system administrator... In my case on Solaris wxPython applications would core dump on exit. The core file indicated that the fault happened after _exit() was called and the run-time library was trying to execute cleanup code. After relinking the Python executable the problem went away. To build Python to link with the C++ linker do this: cd Python-2.0 # wherever the root of the source tree is rm python # in case it's still there from an old build make LINKCC=g++ # or whatever your C++ command is make install I recently built Python 2.1.3 and Python 2.2.1 on Solaris and did not have to resort to this workaround so apparently thigns are getting better there. I will leave this note here though in case there are similar issues elsewhere. However I did run into a Python build issue that affects the wxPython build when attempting to use SunCC instead of GNU gcc. See the note below titled "Building with non-GNU compilers" if you are interested. C. Change to the root wxPython directory and look at the setup.py file. This is the script that configures and defines all the information that Distutils needs to build wxPython. There are some options near the begining of the script that you may want or need to change based on your system and what options you have selected up to this point, (sources from tar.gz or from CVS, etc.) You can either change these flags directly in setup.py or supply them on the command-line. BUILD_GLCANVAS Set to zero if you don't want to build the Open GL canvas extension module. If you don't have OpenGL or compatible libraries then you'll need to set this to zero. BUILD_OGL Set to zero if you don't want to build the Object Graphics Library extension module. BUILD_STC Set to zero if you don't want to build the wxStyledTextCtrl (the Scintilla wrapper) extension module. USE_SWIG If you have edited any of the *.i files you will need to set this flag to non-zero so SWIG will be executed to regenerate the wrapper C++ and shadow python files. IN_CVS_TREE If you are using the CVS version of the wxWindows and wxPython sources then you will need to set this flag to non-zero. This is needed because some source files from the wxWindows tree are copied to be under the wxPython tree in order to keep Distutils happy. With this flag set then setup.py will automatically keep these copied sources up to date if the original version is ever updated. If you are using the tar.gz version of the Python sources then these copied sources are already present in your source tree. D. To build and install wxPython you simply need to execute the setup.py script. If you have more than one version of Python installed, be sure to execute setup.py with the version you want to build wxPython for. Depending on the permissions on your site-packages directory you may need to be root to run the install command. python setup.py build python setup.py install E. At this point you should be able to change into the wxPython/demo directory and run the demo: python demo.py F. If you would like to make a test build that doesn't overwrite the installed version of wxPython you can do so with this command instead of the install command above: python setup.py build_ext --inplace This will build the wxPython package in the local wxPython directory instead of installing it under your Python installation. To run using this test version just add the base wxPython source directory to the PYTHONPATH: export PYTHONPATH=~/projects/wxWindows/wxPython # or whatever is required for your shell cd ~/projects/wxWindows/wxPython/demo python demo.py 4. Building with non-GNU compilers ---------------------------------- As mentioned above Python's distutils uses whatever compiler Python was compiled with to compile extension modules. It also appears that distutils assumes that this compiler can compile C or C++ sources as distutils makes no differentiation between the two. For builds using GNU gcc and a few other compilers this is not an issue as they will determine the type of source from the file extension. For SunCC (and probably other compilers that came from cfront) it won't work as the C compiler (cc) is totally separate from the C++ compiler (CC). This causes distutils to attempt to compile the wxPython sources with the C compiler, which won't work. There may be better ways to get around this, but here is the workaround I devised. I created a script that will execute either cc or CC based on the file extension given to it. If Python uses this script for its compiler then it will also be used by extensions built with distutils and everybody will be more or less happy. Here is a copy of the script I used. It was a fairly quick rush job so there are probably issues with it but it worked for me. #!/bin/bash #-------------------------------------------------------------- # Try to determine type of file being compiled and then # launch cc for C sources or CC for C++. # args=$@ is_C= for arg in $args; do # is the arg a file that exists? if [ -e $arg ]; then # does it end in ".c"? if [ "${arg:${#arg}-2}" == ".c" ]; then is_C=yes fi fi done # if the flag wasn't set then assume C++ and execute CC, # otherwise execute cc. if [ -z $is_C ]; then exec CC -w $@ else exec cc -w $@ fi #-------------------------------------------------------------- I called it pycc, put it in ${prefix}/bin and set its execute permission bit. The next step is to configure and build Python such that it uses pycc as it's compiler. You can do that by setting CC in your environment before running configure, like this in bash: export CC=pycc configure After making and installing Python with this configuration you should be able to build wxPython as described in the steps above. ----------------- robin@alldunn.com