The tool cpp-dependencies creates #include dependency information for C++ source code files, which it derives from scanning a full source tree.

The dependency information is output as .dot files, which can be visualized in, for example, GraphViz.

Happy coding!

Peter Bindels and Rijn Buve

TomTom International BV

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

The tool depends on Boost.Filesystem being available and usable. Installing this should be done with your platform's package management system, such as Apt, Pacman or Brew.

The build configuration is created with CMake. To create the build configuration for your build system (GNU make, MSBuild/Visual Studio) create a build directory outside this source directory and run

cmake <PATH_TO_THIS_SOURCE_DIR>

If you want to use Boost::Filesystem instead of std::filesystem, if your platform does not have a std::filesystem implementation yet or if you prefer it, add -DWITH_BOOST to the invocation of CMake.

To build the tool, either execute

make

for GNU make or open the Visual Studio solution file generated in the build directory.

This creates the executable file cpp-dependencies.

To check if the tool was compiled correctly, execute:

./cpp-dependencies

This provides help information about the tool. More information about its usage is presented in the next paragraph.

As a first thing on a code base is to find out whether it can read the code correctly. From the root of the project, run the command:

cpp-dependencies --stats .

to determine the complexity of the code base and the amount of nodes that are entangled in cycles with other components. In well set-up projects, the cycle count will be equal to zero and the amount of components will be in the same order of size as the logical components you expect.

To investigate a specific component, you can use

cpp-dependencies --info <component> .

for all information the tool has on the component, or:

cpp-dependencies --inout <component> .

to find out who links to and from your component.

In case you have a dependency that you were not expecting, or find out that when rebuilding component A that a supposedly-unrelated component B is built, you can use:

cpp-dependencies --shortest A B .

to determine why there is a link from component A to component B. It will find one of the shortest paths it can find from A to B if there is one.

The tool is also able to provide output in .dot format, which is a format used by GraphViz and other tools to contain graphs. It is a human-readable format and can be used by the dot tool to convert it into a graphical image. To create a graph file, use:

cpp-dependencies --graph mygraph.dot .

to create a mygraph.dot file containing the full component graph.

You can restrict the component graph to either all components beneath a given target (--graph-for <output> <target>) or all components part of a cycle (--graph-cycles).

To make this text-format graph into a viewable graph, use for example:

dot -Tpng mygraph.dot >mygraph.png

to convert it into a PNG file.

The dot program will try to find a way to graphically display the graph output. Note that very large graphs, in particular if many cycles are present, can take hours to render.

In the source tree there's a folder example which contains an empty skeleton project, which does have some dependency information to be extracted from it. To start analyzing it, we first run the tool to extract statistics:

> cpp-dependencies --dir example --stats
6 components with 5 public dependencies, 1 private dependencies
Detected 2 nodes in cycles

This informs us that there is something not quite right with the dependencies. It sees 6 components: the root folder, four libraries and an executable. The simplest way to find out what's wrong is to draw out the graph in a visual way:

> cpp-dependencies --dir example --graph dependencies.dot
> dot -Tpng dependencies.dot >dependencies.png

Then open this PNG file in any tool that can view it, such as a web browser. This shows us the following image:

The light blue links are an implementation-only link, the dark blue ones expose some part of this dependency on their interface. The orange ones are the most interesting ones; they are places where a component can reach itself through some other component. Let's find out why this is there:

> cpp-dependencies --dir example --shortest Engine UI
Engine -> UI
  ./Engine/Engine.h includes ./UI/Display.h
> cpp-dependencies --dir example --shortest UI Engine
UI -> Engine
  ./UI/Display.cpp includes ./Engine/Engine.h
  ./UI/Display.h includes ./Engine/Engine.h

At this point, it's up to the developer or architect to find out which of these two dependencies is the wrong way around and to find a way around that. In the example, the Engine component should not be talking directly to the UI component. Removing this dependency results in the following statistics:

> cpp-dependencies --dir example --stats
6 components with 4 public dependencies, 2 private dependencies
Detected 0 nodes in cycles

The cycle has been removed, and there is one less dependency. We can find out what the shortest path is to the DataAccess component from the executable:

> cpp-dependencies --dir example --shortest main DataAccess
main -> UI
  ./main/main.cpp includes ./UI/Display.h
UI -> Engine
  ./UI/Display.cpp includes ./Engine/Engine.h
  ./UI/Display.h includes ./Engine/Engine.h
Engine -> DataAccess
  ./Engine/Engine.cpp includes ./DataAccess/DA.h

This tells us that there's no path shorter than three steps, and it informs us for each step of the way why it detects this link. In more complicated cycles, this can be a way to isolate the thinnest part of the cycle. In situations where there's an invalid dependency from one component to another - for example, from a unit test of one component to a very far away different component, this can help you identify where on the path from A to B a wrong link is present. It can also be used to explicitly verify that a link is not present, such as the one we just removed:

> cpp-dependencies --dir example --shortest Engine UI
No path could be found from Engine to UI

The graph now also shows proper dependency ordering:

> cpp-dependencies --dir example --graph newdependencies.dot
> dot -Tpng newdependencies.dot >newdependencies.png

We can regenerate the CMakeLists.txt files as well to remove the dependency from the build inputs, so that our build system will also know that the link is no longer present:

> cpp-dependencies --dir example --dryregen
Difference detected at "./Engine"
> cpp-dependencies --dir example --regen

As cpp-dependencies has a lot of analysis it can do on the source tree, there are also some configurable parts to it. The configuration can be found in the file config-cpp-dependencies.txt that should be in your project root. It allows you to customize the colors used in generation, the thresholds for outlier detection and some minor parameters. Please read the documentation in the example config-cpp-dependencies.txt that is in the source distribution for the tool to see all the options.

The tool itself is split up into a few separate files to make it easier to find and extend its functionality. The following files are found:

• main.cpp contains the main functions and help information, as well as the core flow.
• Input.cpp contains the functions that read C++ and CMakeLists files into the information needed by the tool.
• Output.cpp contains functions to write all output files generated, except for the CMakeLists generation.
• CmakeRegen.cpp contains the functionality to write CMakeLists files.
• Analysis.cpp contains all graph processing and navigation functions.
• Component.cpp contains the implementation needed for the struct-like data storage classes.
• generated.cpp contains the function to convert found header files into a lookup map. Also the place to add generated files to the known file list, so that they will be taken into account for components.
• Constants.h contains the constants used throughout the code base.

In general, the root functionality is kept in main.cpp, the structural classes are kept in Component.cpp and any auxiliary functions that are used to do this are split up by domain.

The tool was implemented with the goal of being able to quickly analyze dependencies between components of a complex project, including how the dependency graph changes when some changes are made to the source tree. To accomplish this, choices were made in the direction of more performance at the expense of strict correctness. Specifically:

• It does not use a proper C++ parser to read C++ files, nor a proper CMake parser to read CMake files. Properly parsing these files would increase the full run time of the program by orders of magnitude and make it much less useful.
• strstr is used across the full code base. While profiling, we found that std::string::find was taking over 80% of the full runtime. Replacing it with strstr, which is typically much more optimized, made the whole program twice as fast.

This results in it running on a 1.5GB source code base in about 2.1 seconds -- fast enough for interactive checks and rerunning after any small modification.

The tool was set up to compile on a Ubuntu 12.04 system with the platform default compiler. This means that the sources will use C++11 but will not use anything not available in GCC 4.6. It has been tested and used on Linux (Ubuntu 12.04 - 16.04) and MacOS X (different versions).

It's good practice to set up a personal global .gitignore file on your machine which filters a number of files on your file systems that you do not wish to submit to the Git repository. You can set up your own global ~/.gitignore file by executing: git config --global core.excludesfile ~/.gitignore

In general, add the following file types to ~/.gitignore (each entry should be on a separate line): *.com *.class *.dll *.exe *.o *.so *.log *.sql *.sqlite *.tlog *.epoch *.swp *.hprof *.hprof.index *.releaseBackup *~

The local .gitignore file in the Git repository itself to reflect those file only that are produced by executing regular compile, build or release commands.

If you encounter any problems with this library, don't hesitate to use the Issues session to file your issues. Normally, one of our developers should be able to comment on them and fix.