Analysis of C and C++ projects based on JSON Compilation Database
- General information
- Running the analysis and getting the report
- How to generate compile_commands.json
One of the ways to represent the structure of a C++ project is the JSON Compilation Database format. It's a file that contains the compilation parameters necessary to create object files from the source code of a project. Usually, the file has the name 'compile_commands.json'. A compilation database in JSON-format consists of an array of "command objects", where each command object specifies one way a translation unit is compiled in the project.
You can use the 'compile_commands.json' file to compile a project or analyze the project by third-party utilities. The PVS-Studio C and C++ analyzer works with this format as well.
Running the analysis and getting the report
To analyze the project on Linux and macOS, you need to use 'pvs-studio-analyzer' utility. To analyze the project on Windows, use 'CompileCommandsAnalyzer.exe' utility. The utility is usually located in the 'C:\Program Files (x86)\PVS-Studio' folder. Read more information about CompileCommandsAnalyzer and pvs-studio-analyzer here.
Important: The project must be successfully compiled and built to be analyzed.
To start the analysis and get the report, you need to run two commands.
The command example for Linux and macOS:
pvs-studio-analyzer analyze -f path_to_compile_commands.json \ -o pvs.log -e excludepath -j<N> plog-converter -a GA:1,2 -t tasklist -o project.tasks pvs.log
The command example for Windows:
CompilerCommandsAnalyzer.exe analyze ^ -f path_to_compile_commands.json ^ -o pvs.log -e exclude-path -j<N> PlogConverter.exe -a GA:1,2 -t Plog -o path_to_output_directory ^ -n analysis_report pvs.log
If you run the analysis from the directory with the 'compile_commands.json' file, you may disable the '-f' flag.
To exclude directories with third-party libraries and/or tests from the analysis, you can use the '-e' flag. If there are several paths, it's necessary to write the '-e' flag for each path:
-e third-party -e tests
The analysis can be parallelized into multiple threads with the help of '-j' flag.
How to generate compile_commands.json
If by default the project does not contain 'compile_commands.json', you can choose one of the ways to generate such a file.
To generate 'compile_commands.json, add one flag to the CMake call:
cmake -DCMAKE_EXPORT_COMPILE_COMMANDS=On .
cmake -DCMAKE_EXPORT_COMPILE_COMMANDS=On -G Ninja . cmake -DCMAKE_EXPORT_COMPILE_COMMANDS=On -G "NMake Makefiles" .
To use the Ninja generator for Windows, it is often required to execute commands from the Visual Studio developer's command line (e.g., 'x64 Native Tools Command Prompt for VS', etc.).
If the project is built directly with Ninja and there is a 'build.ninja' file in the project folder, use the following command to generate 'compile_commands.json':
ninja -t compdb > compile_commands.json
To generate 'compile_commands.json' in a project that use Qt Build System, execute the following command:
qbs generate --generator clangdb
Text Toolkit utility
Having trouble getting the 'compile_commands.json' file using GNU make? Try Text Toolkit. You can generate a compilation database either using the Web interface (only for Linux and macOS), or by launching a Python script. To generate a database online, take the following steps:
- run the command 'make -nwi > output.txt';
- copy the contents of the 'output.txt ' file and paste them to the window on the Text Toolkit website;
- click the 'Generate' button to generate the compilation database in JSON format;
- copy the obtained commands to the 'compile_commands.json' file.
To generate the 'compile_commands.json' using Python, you need to clone a repository from GitHub and run the following command:
ninja -nv | python path_to_texttoolkit_dir\cdg.py
Bear utility (only for Linux and macOS)
The Bear (version 2.4 or higher) utility collects compilation parameters by intercepting the compiler calls during project build. To generate 'compile_commands.json', run the following command
bear -- <build_command>
The 'build_command' can be any build command such as 'make all' or './build.sh'.
intercept-build utility (only for Linux and macOS)
The 'intercept-build' utility in scan-build is similar to the Bear utility. The command to generate 'compile_commands.json':
Compilation Database Generator utility (only for Linux and macOS)
Compile Database Generator (compiledb) is a utility that generates compilation databases for Makefile-based build systems. The example of the 'compile_commands.json' generation:
compiledb -n make
The '-n' flag means that the build won't happen (dry run).
Xcode project (macOS only)
With the xcpretty utility, you can generate 'compile_commands.json'. To do this, run the following command:
xcodebuild [flags] | xcpretty -r json-compilation-database
To generate 'compile_commands.json' in the project that uses qmake, you can use IDE QtCreator version 4.8 or higher. Open the desired project and select 'Build->Generate Compilation Database for %project_name%' in the menu bar:
The generated 'compile_commands.json' file will be in the project's build directory.
Note: this method of obtaining 'compile_commands.json' does not have automation. We recommend to use this method only to evaluate the analyzer.