3 `./configure.py` generates the `build.ninja` files used to build
4 ninja. It accepts various flags to adjust build parameters.
5 Run './configure.py --help' for more configuration options.
7 The primary build target of interest is `ninja`, but when hacking on
8 Ninja your changes should be testable so it's more useful to build and
9 run `ninja_test` when developing.
13 Ninja is built using itself. To bootstrap the first binary, run the
14 configure script as `./configure.py --bootstrap`. It This first
15 compiles all non-test source files together, then re-builds Ninja
16 using itself. You should end up with a `ninja' binary (or
17 `ninja.exe`) in the source root.
21 On Windows, you'll need to install Python to run `configure.py`, and
22 run everything under a Visual Studio Tools Command Prompt (or after
23 running `vcvarsall` in a normal command prompt). See below if you
24 want to use mingw or some other compiler instead using Visual Studio.
26 ### Adjusting build flags
28 Build in "debug" mode while developing (disables optimizations and builds
29 way faster on Windows):
31 ./configure.py --debug
33 To use clang, set `CXX`:
35 CXX=clang++ ./configure.py
37 ## How to successfully make changes to Ninja
39 Github pull requests are convenient for me to merge (I can just click
40 a button and it's all handled server-side), but I'm also comfortable
41 accepting pre-github git patches (via `send-email` etc.).
43 Good pull requests have all of these attributes:
45 * Are scoped to one specific issue
46 * Include a test to demonstrate their correctness
47 * Update the docs where relevant
48 * Match the Ninja coding style (see below)
49 * Don't include a mess of "oops, fix typo" commits
51 These are typically merged without hesitation. If a change is lacking
52 any of the above I usually will ask you to fix it, though there are
53 obvious exceptions (fixing typos in comments don't need tests).
55 I am very wary of changes that increase the complexity of Ninja (in
56 particular, new build file syntax or command-line flags) or increase
57 the maintenance burden of Ninja. Ninja is already successfully in use
58 by hundreds of developers for large projects and it already achieves
59 (most of) the goals I set out for it to do. It's probably best to
60 discuss new feature ideas on the mailing list before I shoot down your
65 ### Test-driven development
67 Set your build command to
69 ./ninja ninja_test && ./ninja_test --gtest_filter=MyTest.Name
71 now you can repeatedly run that while developing until the tests pass
72 (I frequently set it as my compilation command in Emacs). Remember to
73 build "all" before committing to verify the other source still works!
75 ## Testing performance impact of changes
77 If you have a Chrome build handy, it's a good test case. Otherwise,
78 [the github downoads page](https://github.com/martine/ninja/downloads)
79 has a copy of the Chrome build files (and depfiles). You can untar
82 path/to/my/ninja chrome
84 and compare that against a baseline Ninja.
86 There's a script at `misc/measure.py` that repeatedly runs a command like
87 the above (to address variance) and summarizes its runtime. E.g.
89 path/to/misc/measure.py path/to/my/ninja chrome
91 For changing the depfile parser, you can also build `parser_perftest`
92 and run that directly on some representative input files.
96 Generally it's the [Google C++ coding style][], but in brief:
98 * Function name are camelcase.
99 * Member methods are camelcase, expect for trivial getters which are
100 underscore separated.
101 * Local variables are underscore separated.
102 * Member variables are underscore separated and suffixed by an extra
104 * Two spaces indentation.
105 * Opening braces is at the end of line.
106 * Lines are 80 columns maximum.
107 * All source files should have the Google Inc. license header.
109 [Google C++ coding style]: http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml
115 * Use `///` for doxygen.
116 * Use `\a` to refer to arguments.
117 * It's not necessary to document each argument, especially when they're
118 relatively self-evident (e.g. in `CanonicalizePath(string* path, string* err)`,
119 the arguments are hopefully obvious)
121 ### Building the manual
123 sudo apt-get install asciidoc --no-install-recommends
126 ### Building the code documentation
128 sudo apt-get install doxygen
131 ## Building for Windows
133 While developing, it's helpful to copy `ninja.exe` to another name like
134 `n.exe`; otherwise, rebuilds will be unable to write `ninja.exe` because
135 it's locked while in use.
137 ### Via Visual Studio
139 * Install Visual Studio (Express is fine), [Python for Windows][],
140 and (if making changes) googletest (see above instructions)
141 * In a Visual Studio command prompt: `python configure.py --bootstrap`
143 [Python for Windows]: http://www.python.org/getit/windows/
145 ### Via mingw on Windows (not well supported)
147 * Install mingw, msys, and python
148 * In the mingw shell, put Python in your path, and
149 `python configure.py --bootstrap`
150 * To reconfigure, run `python configure.py`
151 * Remember to strip the resulting executable if size matters to you
153 ### Via mingw on Linux (not well supported)
155 Setup on Ubuntu Lucid:
156 * `sudo apt-get install gcc-mingw32 wine`
157 * `export CC=i586-mingw32msvc-cc CXX=i586-mingw32msvc-c++ AR=i586-mingw32msvc-ar`
159 Setup on Ubuntu Precise:
160 * `sudo apt-get install gcc-mingw-w64-i686 g++-mingw-w64-i686 wine`
161 * `export CC=i686-w64-mingw32-gcc CXX=i686-w64-mingw32-g++ AR=i686-w64-mingw32-ar`
164 * Uncomment the `[multilib]` section of `/etc/pacman.conf` and `sudo pacman -Sy`.
165 * `sudo pacman -S mingw-w64-gcc wine`
166 * `export CC=x86_64-w64-mingw32-cc CXX=x86_64-w64-mingw32-c++ AR=x86_64-w64-mingw32-ar`
167 * `export CFLAGS=-I/usr/x86_64-w64-mingw32/include`
170 * `./configure.py --platform=mingw --host=linux`
171 * Build `ninja.exe` using a Linux ninja binary: `/path/to/linux/ninja`
172 * Run: `./ninja.exe` (implicitly runs through wine(!))
174 ### Using Microsoft compilers on Linux (extremely flaky)
176 The trick is to install just the compilers, and not all of Visual Studio,
177 by following [these instructions][win7sdk].
179 [win7sdk]: http://www.kegel.com/wine/cl-howto-win7sdk.html
183 Do a clean debug build with the right flags:
185 CFLAGS=-coverage LDFLAGS=-coverage ./configure.py --debug
186 ninja -t clean ninja_test && ninja ninja_test
188 Run the test binary to generate `.gcda` and `.gcno` files in the build
189 directory, then run gcov on the .o files to generate `.gcov` files in the
195 Look at the generated `.gcov` files directly, or use your favorit gcov viewer.
199 Build with afl-clang++:
201 CXX=path/to/afl-1.20b/afl-clang++ ./configure.py
204 Then run afl-fuzz like so:
206 afl-fuzz -i misc/afl-fuzz -o /tmp/afl-fuzz-out ./ninja -n -f @@
208 You can pass `-x misc/afl-fuzz-tokens` to use the token dictionary. In my
209 testing, that did not seem more effective though.
211 #### Using afl-fuzz with asan
213 If you want to use asan (the `isysroot` bit is only needed on OS X; if clang
214 can't find C++ standard headers make sure your LLVM checkout includes a libc++
215 checkout and has libc++ installed in the build directory):
217 CFLAGS="-fsanitize=address -isysroot $(xcrun -show-sdk-path)" \
218 LDFLAGS=-fsanitize=address CXX=path/to/afl-1.20b/afl-clang++ \
220 AFL_CXX=path/to/clang++ ninja
222 Make sure ninja can find the asan runtime:
224 DYLD_LIBRARY_PATH=path/to//lib/clang/3.7.0/lib/darwin/ \
225 afl-fuzz -i misc/afl-fuzz -o /tmp/afl-fuzz-out ./ninja -n -f @@