These instructions are also found in the live documentation at ReadTheDocs.
We currently use CMake to manage building Tracktable and running its tests. We expect to have Tracktable available on PyPI as of version 1.2.0 so that you can ‘pip install tracktable’. This document is for people who want to (or have to) build from source.
Step 0: Audience
We assume that you are familiar with downloading, compiling and installing software from source as well as with your operating system’s package manager (if any). You will need to know how to set or modify environment variables, run the compiler and find libraries or header files on your system.
Step 1: Dependencies
Tracktable has the following required dependencies:
- Python 2.7, 3.4, 3.5, 3.6, or 3.7 – http://python.org
- NOTE: Tracktable 1.1 is the last version that will support Python 2.7.
- numpy 1.7+ – http://numpy.org
- Matplotlib 2.0+ – http://matplotlib.org
- Basemap (including hi-res data) – http://matplotlib.org/basemap
- NOTE: We’re switching to Cartopy as of Tracktable 1.2.
- PyTZ – https://pypi.python.org/pypi/pytz/
- Shapely – https://pypi.python.org/pypi/Shapely
- Six – https://pypi.org/project/six/
- PyProj – https://pypi.org/project/pyproj/
- Compiler – GCC 4.4.7 or newer (http://gcc.gnu.org), clang 3.5 or newer (http://clang.llvm.org), Visual Studio 14 2015 or newer (https://visualstudio.microsoft.com)
- Boost 1.61 or newer – http://www.boost.org
- GEOS library – http://geos.osgeo.org
- You must build Boost with Boost.Python enabled using the headers from the same Python installation you will use to run Tracktable.
- We rely on the r-tree and distance computation code available in recent versions of Boost. Use 1.61 or newer.
- With respect to C++11: if you want to call Tracktable from code built with C++11 turned on, you must also build Tracktable with C++11 turned on. The implementation of
boost::variant(which we use for
tracktable::PropertyValueT) is entirely different between the two language versions. This causes link errors if you try to mix versions.
- We do not yet use any C++11 features in Tracktable in order to maintain compatibility with a few environments that are still stranded in the age of C++03. We look forward to their arrival in the modern age.
- CMake 3.0+ – http://www.cmake.org
If you want to build documentation you will also need the following packages:
- Sphinx – http://sphinx-doc.org
- napoleon – https://sphinxcontrib-napoleon.readthedocs.org/en/latest
- This is bundled with Sphinx as of version 1.3.
- Breathe – http://breathe.readthedocs.org/en/latest/
- Sphinx Read the Docs theme – https://sphinx-rtd-theme.readthedocs.io/en/latest
- Doxygen – http://www.doxygen.nl/index.html
- Graphviz (for dot executable)- https://www.graphviz.org/
If you want to render movies you will need FFMPEG:
- FFMPEG – https://www.ffmpeg.org – If you build from source please be sure to include the MPEG4 and FFV1 codecs. Both of these are included with the standard FFMPEG download. Tracktable can use other codecs but does not require them. – Windows users can obtain the ffmpeg executable by installing Image Magick (https://www.imagemagick.org)
Build Notes for Dependencies
If you can possibly help it, install all the dependencies using package managers like
conda (Anaconda’s built-in package manager),
pip (comes with Python),
apt-get (both of these are common in Linux environments), MacPorts or Homebrew. The notes in this section are for cases when you have no choice but to build external packages from source.
We need several of Boost’s compiled libraries including chrono, date_time, iostreams, log, random, timer and especially Boost.Python. As with other dependencies, check your operating system’s package manager first. It’s possible that you can install Boost with all its optional components from there.
If you already have a recent Boost installation you can check for Boost.Python by looking for files named
(prefix)boost_python.(suffix) where (prefix) is
lib on Unix-like systems and (suffix) is
.so on Unix systems,
.dylib on Mac OSX and
.lib) on Windows.
If you really do have to build Boost from source — for example, if you had to build your own Python installation — then make sure to configure it to use the proper Python installation. Information about how to do this can be found in the Boost.Python documentation at http://www.boost.org/doc/libs/1_61_0/libs/python/doc/building.html
One final note: We know that it’s a pain to try to keep up with recent versions of a library as big as Boost. We will not require a newer version unless absolutely necessary.
For up-to-date instructions on building FFMPEG please refer to https://trac.ffmpeg.org/wiki/CompilationGuide and choose your OS. We recommend that you compile in support for H264 video (via libx264). While this is not required, it is widely supported by current devices such as iPads, iPhones and Android systems.
You are now ready to configure and build the C++ part of Tracktable. Install the Python dependencies whenever convenient.
Step 2: Configuration
CMake enforces what we call "out-of-source" builds: that is, you cannot build object files alongside source code files. This makes it much easier to manage multiple build configurations. It also means that the first thing you must do is create a build directory. In the rest of this section we will use
TRACKTABLE_HOME to refer to the directory where you unpacked the Tracktable source.:
$ cd TRACKTABLE_HOME $ mkdir build $ cd build
(You can also put your build directory anywhere else you please.)
Next, use CMake’s configuration utility
ccmake (or its GUI tool if you prefer) to configure compile settings:
If you made your build directory inside the source directory:
$ ccmake ..
If you made it someplace else:
$ ccmake TRACKTABLE_HOME/
Once CMake starts you will see a mostly empty screen with the message
EMPTY CACHE. Press ‘c’ (if you use
ccmake) or click ‘Configure’ (if you use the CMake GUI) to start configuration. After a moment, several new options will appear including
BUILD_SHARED_LIBS. Leave these set to
ON. Without these options you will not be able to use any of Tracktable’s Python components. Set the value of
CMAKE_INSTALL_PREFIX to the directory where you want to install the software. Press ‘c’ or click the ‘Configure’ button again to incorporate your choice.
Now you need to set options that are normally hidden. Press ‘t’ or select the Show Advanced Options checkbox. Here are the variables you need to check:
Boost_LIBRARY_DIR. These should point to your Boost 1.61 install with Boost.Python. Filenames for the
boost_pythonlibraries should appear automatically. If you change either of these directories in CMake, press ‘c’ or click ‘Configure’ to make your changes take effect.
PYTHON_INCLUDE_DIRMake sure that all three of these point to the same installation. On Mac OSX with MacPorts in particular, CMake has a habit of using whatever Python executable is first in your path, the include directory from
/System/Library/Frameworks/Python.frameworkand the library from
/usr/lib/. MacPorts installs its Python library in
/opt/local/Library/Frameworks/Python.framework/Versions/3.7with headers in
Headers/and the Python library in
lib/libpython3.7.dylib. Substitute whatever version you have installed in place of 3.7. If you have installed your own Python interpreter then use whatever path you chose for its installation. Note: You must make sure that all three components (interpreter, library and headers) correspond to one another or else the Python code will crash on startup with an unhelpful error message about thread state. If you change any of these variables, press ‘c’ or click Configure’ to make your changes take effect.
Now press ‘g’ or click ‘Generate’ to confirm all of your choices and generate Makefiles, Visual Studio project files or your chosen equivalent.
Gotcha: Boost import targets not found
This happens when your installed version of CMake is too old for your installed version of Boost.
Gotcha: Anaconda does not install ccmake
This is a known bug that has been fixed in
conda-forge but has not yet propagated to the main distribution. Install
cmake from the
conda-forge channel as follows:
$ conda install -c conda-forge cmake
Gotcha: python3 Boost library not found but I’m using Python 2
Check your Python CMake variables as listed in #2 above. They are probably pointing to a Python 3 interpreter.
Some older CMake installations have an odd bug that shows up with certain Linux installations. You may see
Boost_DIR set to something like
/usr/lib64 no matter what value you try to set for
Boost_LIBRARY_DIR. If you experience this, try adding the line:
TRACKTABLE_HOME/tracktable/CMakeLists.txt and then rerun CMake as described above.
Step 3: Build and Test
On Unix-like systems, type
make. For Visual Studio, run
msbuild on a project file, or open up the project files in your IDE (as appropriate).
Once the build process has finished go to your build directory and run
ctest (part of CMake) to run all the tests. Optionally, Windows users can run the test project but this is just a fancy wrapper for ctest in this case. They should all succeed. Some of the later Python tests such as P_Mapmaker may take a minute or two.
If you have multiple cores or processors and your build system supports it, by all means build in parallel. GNU Make will do this when you say
make -j where is the number of compilers you’re willing to run. A bare
make -j will cause it to run as many compiler instances as it believes you have cores or processors. Windows users using msbuild, can use the
/m: option from the command line.
Warning: The Python wrappers, especially the wrappers for DBSCAN, feature vectors and the R-tree, take between 1GB and 1.5GB of memory to compile. Keep this in mind when you run parallel builds. A good rule of thumb is to run no more than 1 process for every 1.5-2GB of main memory in your computer.
CMake error: "cannot find numpy"
This usually arises when CMake detects a different Python installation than the one you actually use. Take a look at the
PYTHON_EXECUTABLEfield in CMake. If it says something like
/usr/bin/pythonand you use a Python distribution like Anaconda or Enthought’s Canopy, that’s the problem. To fix, change
PYTHON_EXECUTABLEto point to the Python interpreter in your environment. For Anaconda under Linux and OS X, this is usually either
~/anaconda3/envs//bin/python. Remember to also change
PYTHON_INCLUDE_DIRto the files inside your Anaconda (or Enthought) directory.
Python tests crashing
If the tests whose names begin with
P_crash, you probably have a mismatch between
PYTHON_LIBRARY. Check their values in
ccmake/ CMake GUI. If your Python executable is in (for example)
/usr/local/python/bin/pythonthen its corresponding library will usually be in
/usr/local/python/lib/libpython2.7.soinstead of halfway across the system.
- Python tests running but failing
- Cause #1: One or more required Python packages missing. Check to make sure you have installed everything listed in the Dependencies section.
- Cause #2: Couldn’t load one or more C++ libraries. Make sure that the directories containing the libraries in question are in your LD_LIBRARY_PATH (DYLD_LIBRARY_PATH for Mac OSX) environment variable or on your PATH in Windows.
Cause #3: The wrong Python interpreter is being invoked.
This really shouldn’t happen: we use the same Python interpreter that you specify in
PYTHONPATHourselves while running tests.
- Nearby stars go nova
- We’re afraid you’re on your own if this happens.
Step 4: Install
You can use Tracktable as-is from its build directory or install it elsewhere on your system. To install it, type
make install in the build directory (or, again, your IDE’s equivalent).
You will also need to add Tracktable to your system’s Python search path, usually stored in an environment variable named
- If you are going to run Tracktable from the directory where you unpacked it then add the directory
- If you installed Tracktable via
make installthen you will need to add
INSTALL_DIRis the directory you specified for installation when running CMake.
Finally, you will need to tell your system where to find the Tracktable C++ libraries.
- If you are running from your build tree (common during development) then the libraries will be in
BUILD/bin(XXX Check where Windows puts its DLLs).
- If you are running from an installed location the libraries will be in
INSTALL_DIR/bin(XXX same check).
- On Windows, add the library directory to your
- On Linux and most Unix-like systems, add the library directory to your
- On Mac OSX, add the library directory to your
On Unix-like systems you can also add the library directory to your system-wide ld.so.conf file. You will need root permissions in order to do so. That is beyond the scope of this document.