Basic overview

./ generates the files used to build ninja. It accepts various flags to adjust build parameters. Run ‘./ --help’ for more configuration options.

The primary build target of interest is ninja, but when hacking on Ninja your changes should be testable so it's more useful to build and run ninja_test when developing.


Ninja is built using itself. To bootstrap the first binary, run the configure script as ./ --bootstrap. This first compiles all non-test source files together, then re-builds Ninja using itself. You should end up with a ninja binary (or ninja.exe) in the source root.


On Windows, you'll need to install Python to run, and run everything under a Visual Studio Tools Command Prompt (or after running vcvarsall in a normal command prompt). See below if you want to use mingw or some other compiler instead of Visual Studio.

Adjusting build flags

Build in “debug” mode while developing (disables optimizations and builds way faster on Windows):

./ --debug

To use clang, set CXX:

CXX=clang++ ./

How to successfully make changes to Ninja

Github pull requests are convenient for me to merge (I can just click a button and it‘s all handled server-side), but I’m also comfortable accepting pre-github git patches (via send-email etc.).

Good pull requests have all of these attributes:

  • Are scoped to one specific issue
  • Include a test to demonstrate their correctness
  • Update the docs where relevant
  • Match the Ninja coding style (see below)
  • Don't include a mess of “oops, fix typo” commits

These are typically merged without hesitation. If a change is lacking any of the above I usually will ask you to fix it, though there are obvious exceptions (fixing typos in comments don't need tests).

I am very wary of changes that increase the complexity of Ninja (in particular, new build file syntax or command-line flags) or increase the maintenance burden of Ninja. Ninja is already successfully used by hundreds of developers for large projects and it already achieves (most of) the goals I set out for it to do. It's probably best to discuss new feature ideas on the mailing list before I shoot down your patch.


Test-driven development

Set your build command to

./ninja ninja_test && ./ninja_test --gtest_filter=MyTest.Name

now you can repeatedly run that while developing until the tests pass (I frequently set it as my compilation command in Emacs). Remember to build “all” before committing to verify the other source still works!

Testing performance impact of changes

If you have a Chrome build handy, it's a good test case. Otherwise, the github downoads page has a copy of the Chrome build files (and depfiles). You can untar that, then run

path/to/my/ninja chrome

and compare that against a baseline Ninja.

There's a script at misc/ that repeatedly runs a command like the above (to address variance) and summarizes its runtime. E.g.

path/to/misc/ path/to/my/ninja chrome

For changing the depfile parser, you can also build parser_perftest and run that directly on some representative input files.

Coding guidelines

Generally it's the Google C++ coding style, but in brief:

  • Function name are camelcase.
  • Member methods are camelcase, expect for trivial getters which are underscore separated.
  • Local variables are underscore separated.
  • Member variables are underscore separated and suffixed by an extra underscore.
  • Two spaces indentation.
  • Opening braces is at the end of line.
  • Lines are 80 columns maximum.
  • All source files should have the Google Inc. license header.


Style guidelines

  • Use /// for doxygen.
  • Use \a to refer to arguments.
  • It‘s not necessary to document each argument, especially when they’re relatively self-evident (e.g. in CanonicalizePath(string* path, string* err), the arguments are hopefully obvious)

Building the manual

sudo apt-get install asciidoc --no-install-recommends
./ninja manual

Building the code documentation

sudo apt-get install doxygen
./ninja doxygen

Building for Windows

While developing, it‘s helpful to copy ninja.exe to another name like n.exe; otherwise, rebuilds will be unable to write ninja.exe because it’s locked while in use.

Via Visual Studio

  • Install Visual Studio (Express is fine), Python for Windows, and (if making changes) googletest (see above instructions)
  • In a Visual Studio command prompt: python --bootstrap

Via mingw on Windows (not well supported)

  • Install mingw, msys, and python
  • In the mingw shell, put Python in your path, and python --bootstrap
  • To reconfigure, run python
  • Remember to strip the resulting executable if size matters to you

Via mingw on Linux (not well supported)

Setup on Ubuntu Lucid:

  • sudo apt-get install gcc-mingw32 wine
  • export CC=i586-mingw32msvc-cc CXX=i586-mingw32msvc-c++ AR=i586-mingw32msvc-ar

Setup on Ubuntu Precise:

  • sudo apt-get install gcc-mingw-w64-i686 g++-mingw-w64-i686 wine
  • export CC=i686-w64-mingw32-gcc CXX=i686-w64-mingw32-g++ AR=i686-w64-mingw32-ar

Setup on Arch:

  • Uncomment the [multilib] section of /etc/pacman.conf and sudo pacman -Sy.
  • sudo pacman -S mingw-w64-gcc wine
  • export CC=x86_64-w64-mingw32-cc CXX=x86_64-w64-mingw32-c++ AR=x86_64-w64-mingw32-ar
  • export CFLAGS=-I/usr/x86_64-w64-mingw32/include

Then run:

  • ./ --platform=mingw --host=linux
  • Build ninja.exe using a Linux ninja binary: /path/to/linux/ninja
  • Run: ./ninja.exe (implicitly runs through wine(!))

Using Microsoft compilers on Linux (extremely flaky)

The trick is to install just the compilers, and not all of Visual Studio, by following these instructions.

Using gcov

Do a clean debug build with the right flags:

CFLAGS=-coverage LDFLAGS=-coverage ./ --debug
ninja -t clean ninja_test && ninja ninja_test

Run the test binary to generate .gcda and .gcno files in the build directory, then run gcov on the .o files to generate .gcov files in the root directory:

gcov build/*.o

Look at the generated .gcov files directly, or use your favorite gcov viewer.

Using afl-fuzz

Build with afl-clang++:

CXX=path/to/afl-1.20b/afl-clang++ ./

Then run afl-fuzz like so:

afl-fuzz -i misc/afl-fuzz -o /tmp/afl-fuzz-out ./ninja -n -f @@

You can pass -x misc/afl-fuzz-tokens to use the token dictionary. In my testing, that did not seem more effective though.

Using afl-fuzz with asan

If you want to use asan (the isysroot bit is only needed on OS X; if clang can't find C++ standard headers make sure your LLVM checkout includes a libc++ checkout and has libc++ installed in the build directory):

CFLAGS="-fsanitize=address -isysroot $(xcrun -show-sdk-path)" \
    LDFLAGS=-fsanitize=address CXX=path/to/afl-1.20b/afl-clang++ \
AFL_CXX=path/to/clang++ ninja

Make sure ninja can find the asan runtime:

DYLD_LIBRARY_PATH=path/to//lib/clang/3.7.0/lib/darwin/ \
    afl-fuzz -i misc/afl-fuzz -o /tmp/afl-fuzz-out ./ninja -n -f @@