3 Evan Martin <martine@danga.com>
9 Ninja is yet another build system. It takes as input the
10 interdependencies of files (typically source code and output
11 executables) and orchestrates building them, _quickly_.
13 Ninja joins a sea of other build systems. Its distinguishing goal is
14 to be fast. It is born from
15 http://neugierig.org/software/chromium/notes/2011/02/ninja.html[my
16 work on the Chromium browser project], which has over 30,000 source
17 files and whose other build systems (including one built from custom
18 non-recursive Makefiles) can take ten seconds to start building after
19 changing one file. Ninja is under a second.
21 Philosophical overview
22 ~~~~~~~~~~~~~~~~~~~~~~
24 Where other build systems are high-level languages, Ninja aims to be
27 Build systems get slow when they need to make decisions. When you are
28 in a edit-compile cycle you want it to be as fast as possible -- you
29 want the build system to do the minimum work necessary to figure out
30 what needs to be built immediately.
32 Ninja contains the barest functionality necessary to describe
33 arbitrary dependency graphs. Its lack of syntax makes it impossible
34 to express complex decisions.
36 Instead, Ninja is intended to be used with a separate program
37 generating its input files. The generator program (like the
38 `./configure` found in autotools projects) can analyze system
39 dependencies and make as many decisions as possible up front so that
40 incremental builds stay fast. Going beyond autotools, even build-time
41 decisions like "which compiler flags should I use?" or "should I
42 build a debug or release-mode binary?" belong in the `.ninja` file
48 Here are the design goals of Ninja:
50 * very fast (i.e., instant) incremental builds, even for very large
53 * very little policy about how code is built. Different projects and
54 higher-level build systems have different opinions about how code
55 should be built; for example, should built objects live alongside
56 the sources or should all build output go into a separate directory?
57 Is there an "package" rule that builds a distributable package of
58 the project? Sidestep these decisions by trying to allow either to
59 be implemented, rather than choosing, even if that results in
62 * get dependencies correct, and in particular situations that are
63 difficult to get right with Makefiles (e.g. outputs need an implicit
64 dependency on the command line used to generate them; to build C
65 source code you need to use gcc's `-M` flags for header
68 * when convenience and speed are in conflict, prefer speed.
70 Some explicit _non-goals_:
72 * convenient syntax for writing build files by hand. _You should
73 generate your ninja files using another program_. This is how we
74 can sidestep many policy decisions.
76 * built-in rules. _Out of the box, Ninja has no rules for
77 e.g. compiling C code._
79 * build-time customization of the build. _Options belong in
80 the program that generates the ninja files_.
82 * build-time decision-making ability such as conditionals or search
83 paths. _Making decisions is slow._
85 To restate, Ninja is faster than other build systems because it is
86 painfully simple. You must tell Ninja exactly what to do when you
87 create your project's `.ninja` files.
92 Ninja is closest in spirit and functionality to make, relying on
93 simple dependencies between file timestamps.
95 But fundamentally, make has a lot of _features_: suffix rules,
96 functions, built-in rules that e.g. search for RCS files when building
97 source. Make's language was designed to be written by humans. Many
98 projects find make alone adequate for their build problems.
100 In contrast, Ninja has almost no features; just those necessary to get
101 builds correct while punting most complexity to generation of the
102 ninja input files. Ninja by itself is unlikely to be useful for most
105 Here are some of the features Ninja adds to make. (These sorts of
106 features can often be implemented using more complicated Makefiles,
107 but they are not part of make itself.)
109 * A Ninja rule may point at a path for extra implicit dependency
110 information. This makes it easy to get header dependencies correct
113 * A build edge may have multiple outputs.
115 * Outputs implicitly depend on the command line that was used to generate
116 them, which means that changing e.g. compilation flags will cause
117 the outputs to rebuild.
119 * Output directories are always implicitly created before running the
120 command that relies on them.
122 * Rules can provide shorter descriptions of the command being run, so
123 you can print e.g. `CC foo.o` instead of a long command line while
126 * Builds are always run in parallel, based by default on the number of
127 CPUs your system has. Underspecified build dependencies will result
130 * Command output is always buffered. This means commands running in
131 parallel don't interleave their output, and when a command fails we
132 can print its failure output next to the full command line that
133 produced the failure.
136 Using Ninja for your project
137 ----------------------------
139 Ninja currently works on Unix-like systems. It's seen the most testing
140 on Linux (and has the best performance there) but it runs fine on Mac
141 OS X and FreeBSD. Ninja has some preliminary Windows support but the
142 full details of the implementation -- like how to get C header
143 interdependencies correct and fast when using MSVC's compiler -- is
146 If your project is small, Ninja's speed impact is likely unnoticeable.
147 Some build timing numbers are included below. (However, even for
148 small projects it sometimes turns out that Ninja's limited syntax
149 forces simpler build rules that result in faster builds.) Another way
150 to say this is that if you're happy with the edit-compile cycle time
151 of your project already then Ninja won't help.
153 There are many other build systems that are more user-friendly or
154 featureful than Ninja itself. For some recommendations: the Ninja
155 author found http://gittup.org/tup/[the tup build system] influential
156 in Ninja's design, and thinks https://github.com/apenwarr/redo[redo]'s
157 design is quite clever.
159 Ninja's benefit comes from using it in conjunction with a smarter
162 http://code.google.com/p/gyp/[gyp]:: The meta-build system used to
163 generate build files for Google Chrome. gyp can generate Ninja files
164 for Linux and Mac and is used by many Chrome developers; support for
165 Windows is in progress. See the
166 http://code.google.com/p/chromium/wiki/NinjaBuild[Chromium Ninja
167 documentation for more details]. gyp is relatively unpopular outside
168 of the Chrome and v8 world.
170 * For Chrome (~30k source files), Ninja reduced no-op builds from
171 around 15 seconds to under one second.
172 * https://plus.google.com/108996039294665965197/posts/SfhrFAhRyyd[A
173 Mozilla developer compares build systems]: "While chromium's full
174 build is 2.15x slower than firefox's, a nop build is 78.2x faster!
175 That is really noticeable during development. No incremental build
176 of firefox can be faster than 57.9s, which means that in practice
177 almost all of them will be over a minute."
179 http://www.cmake.org/[CMake]:: A widely used meta-build system that
180 can generate Ninja files on Linux as of CMake version 2.8.8. (There
181 is some Mac and Windows support -- http://www.reactos.org[ReactOS]
182 uses Ninja on Windows for their buildbots, but those platforms are not
183 yet officially supported by CMake as the full test suite doesn't
186 * For building Blender, one user reported "Single file rebuild is 0.97
187 sec, same on makefiles was 3.7sec."
188 * For building LLVM on Windows, one user reported no-op build times:
189 "ninja: 0.4s / MSBuild: 11s / jom: 53s".
191 others:: Ninja ought to fit perfectly into other meta-build software
192 like http://industriousone.com/premake[premake]. If you do this work,
198 Run `ninja`. By default, it looks for a file named `build.ninja` in
199 the current directory and builds all out-of-date targets. You can
200 specify which targets (files) to build as command line arguments.
202 `ninja -h` prints help output. Many of Ninja's flags intentionally
203 match those of Make; e.g `ninja -C build -j 20` changes into the
204 `build` directory and runs 20 build commands in parallel. (Note that
205 Ninja defaults to running commands in parallel anyway, so typically
206 you don't need to pass `-j`.)
209 Environment variables
210 ~~~~~~~~~~~~~~~~~~~~~
212 Ninja supports one environment variable to control its behavior.
214 `NINJA_STATUS`:: The progress status printed before the rule being run.
215 Several placeholders are available:
216 * `%s`: The number of started edges.
217 * `%t`: The total number of edges that must be run to complete the build.
218 * `%r`: The number of currently running edges.
219 * `%u`: The number of remaining edges to start.
220 * `%f`: The number of finished edges.
221 * `%o`: Overall rate of finished edges per second
222 * `%c`: Current rate of finished edges per second (average over builds specified by -j or its default)
223 * `%%`: A plain `%` character.
224 * The default progress status is `"[%s/%t] "` (note the trailing space
225 to separate from the build rule). Another example of possible progress status
226 could be `"[%u/%r/%f] "`.
231 The `-t` flag on the Ninja command line runs some tools that we have
232 found useful during Ninja's development. The current tools are:
235 `query`:: dump the inputs and outputs of a given target.
237 `browse`:: browse the dependency graph in a web browser. Clicking a
238 file focuses the view on that file, showing inputs and outputs. This
239 feature requires a Python installation.
241 `graph`:: output a file in the syntax used by `graphviz`, a automatic
242 graph layout tool. Use it like:
245 ninja -t graph mytarget | dot -Tpng -ograph.png
248 In the Ninja source tree, `ninja graph.png`
249 generates an image for Ninja itself. If no target is given generate a
250 graph for all root targets.
252 `targets`:: output a list of targets either by rule or by depth. If used
253 like +ninja -t targets rule _name_+ it prints the list of targets
254 using the given rule to be built. If no rule is given, it prints the source
255 files (the leaves of the graph). If used like
256 +ninja -t targets depth _digit_+ it
257 prints the list of targets in a depth-first manner starting by the root
258 targets (the ones with no outputs). Indentation is used to mark dependencies.
259 If the depth is zero it prints all targets. If no arguments are provided
260 +ninja -t targets depth 1+ is assumed. In this mode targets may be listed
261 several times. If used like this +ninja -t targets all+ it
262 prints all the targets available without indentation and it is faster
263 than the _depth_ mode.
265 `rules`:: output the list of all rules with their description if they have
266 one. It can be used to know which rule name to pass to
267 +ninja -t targets rule _name_+.
269 `commands`:: given a list of targets, print a list of commands which, if
270 executed in order, may be used to rebuild those targets, assuming that all
271 output files are out of date.
273 `clean`:: remove built files. By default it removes all built files
274 except for those created by the generator. Adding the `-g` flag also
275 removes built files created by the generator (see <<ref_rule,the rule
276 reference for the +generator+ attribute>>). Additional arguments are
277 targets, which removes the given targets and recursively all files
280 If used like +ninja -t clean -r _rules_+ it removes all files built using
283 Files created but not referenced in the graph are not removed. This
284 tool takes in account the +-v+ and the +-n+ options (note that +-n+
289 Writing your own Ninja files
290 ----------------------------
292 The remainder of this manual is only useful if you are constructing
293 Ninja files yourself: for example, if you're writing a meta-build
294 system or supporting a new language.
299 Ninja evaluates a graph of dependencies between files, and runs
300 whichever commands are necessary to make your build target up to date.
301 If you are familiar with Make, Ninja is very similar.
303 A build file (default name: `build.ninja`) provides a list of _rules_
304 -- short names for longer commands, like how to run the compiler --
305 along with a list of _build_ statements saying how to build files
306 using the rules -- which rule to apply to which inputs to produce
309 Conceptually, `build` statements describe the dependency graph of your
310 project, while `rule` statements describe how to generate the files
311 along a given edge of the graph.
316 Here's a basic `.ninja` file that demonstrates most of the syntax.
317 It will be used as an example for the following sections.
319 ---------------------------------
323 command = gcc $cflags -c $in -o $out
325 build foo.o: cc foo.c
326 ---------------------------------
330 Despite the non-goal of being convenient to write by hand, to keep
331 build files readable (debuggable), Ninja supports declaring shorter
332 reusable names for strings. A declaration like the following
338 can be used on the right side of an equals sign, dereferencing it with
339 a dollar sign, like this:
343 command = gcc $cflags -c $in -o $out
346 Variables can also be referenced using curly braces like `${in}`.
348 Variables might better be called "bindings", in that a given variable
349 cannot be changed, only shadowed. There is more on how shadowing works
350 later in this document.
355 Rules declare a short name for a command line. They begin with a line
356 consisting of the `rule` keyword and a name for the rule. Then
357 follows an indented set of `variable = value` lines.
359 The basic example above declares a new rule named `cc`, along with the
360 command to run. (In the context of a rule, the `command` variable is
361 special and defines the command to run. A full list of special
362 variables is provided in <<ref_rule,the reference>>.)
364 Within the context of a rule, three additional special variables are
365 available: `$in` expands to the list of input files (`foo.c`) and
366 `$out` to the output file (`foo.o`) for the command. For use with
367 `$rspfile_content`, there is also `$in_newline`, which is the same as
368 `$in`, except that multiple inputs are separated by `\n`, rather than
375 Build statements declare a relationship between input and output
376 files. They begin with the `build` keyword, and have the format
377 +build _outputs_: _rulename_ _inputs_+. Such a declaration says that
378 all of the output files are derived from the input files. When the
379 output files are missing or when the inputs change, Ninja will run the
380 rule to regenerate the outputs.
382 The basic example above describes how to build `foo.o`, using the `cc`
385 In the scope of a `build` block (including in the evaluation of its
386 associated `rule`), the variable `$in` is the list of inputs and the
387 variable `$out` is the list of outputs.
389 A build statement may be followed by an indented set of `key = value`
390 pairs, much like a rule. These variables will shadow any variables
391 when evaluating the variables in the command. For example:
394 cflags = -Wall -Werror
396 command = gcc $cflags -c $in -o $out
398 # If left unspecified, builds get the outer $cflags.
399 build foo.o: cc foo.c
401 # But you can can shadow variables like cflags for a particular build.
402 build special.o: cc special.c
405 # The variable was only shadowed for the scope of special.o;
406 # Subsequent build lines get the outer (original) cflags.
407 build bar.o: cc bar.c
411 For more discussion of how scoping works, consult <<ref_scope,the
414 If you need more complicated information passed from the build
415 statement to the rule (for example, if the rule needs "the file
416 extension of the first input"), pass that through as an extra
417 variable, like how `cflags` is passed above.
419 If the top-level Ninja file is specified as an output of any build
420 statement and it is out of date, Ninja will rebuild and reload it
421 before building the targets requested by the user.
426 Pools allow you to allocate one or more rules or edges a finite number
427 of concurrent jobs which is more tightly restricted than the default
430 This can be useful, for example, to restrict a particular expensive rule
431 (like link steps for huge executables), or to restrict particular build
432 statements which you know perform poorly when run concurrently.
434 Each pool has a `depth` variable which is specified in the build file.
435 The pool is then referred to with the `pool` variable on either a rule
436 or a build statement.
438 No matter what pools you specify, ninja will never run more concurrent jobs
439 than the default parallelism, or the number of jobs specified on the command
443 # No more than 4 links at a time.
447 # No more than 1 heavy object at a time.
448 pool heavy_object_pool
458 # The link_pool is used here. Only 4 links will run concurrently.
459 build foo.exe: link input.obj
461 # A build statement can be exempted from its rule's pool by setting an
462 # empty pool. This effectively puts the build statement back into the default
463 # pool, which has infinite depth.
464 build other.exe: link input.obj
467 # A build statement can specify a pool directly.
468 # Only one of these builds will run at a time.
469 build heavy_object1.obj: cc heavy_obj1.cc
470 pool = heavy_object_pool
471 build heavy_object2.obj: cc heavy_obj2.cc
472 pool = heavy_object_pool
477 Generating Ninja files from code
478 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
480 `misc/ninja_syntax.py` in the Ninja distribution is a tiny Python
481 module to facilitate generating Ninja files. It allows you to make
482 Python calls like `ninja.rule(name='foo', command='bar',
483 depfile='$out.d')` and it will generate the appropriate syntax. Feel
484 free to just inline it into your project's build system if it's
494 The special rule name `phony` can be used to create aliases for other
495 targets. For example:
498 build foo: phony some/file/in/a/faraway/subdir/foo
501 This makes `ninja foo` build the longer path. Semantically, the
502 `phony` rule is equivalent to a plain rule where the `command` does
503 nothing, but phony rules are handled specially in that they aren't
504 printed when run, logged (see below), nor do they contribute to the
505 command count printed as part of the build process.
507 `phony` can also be used to create dummy targets for files which
508 may not exist at build time. If a phony build statement is written
509 without any dependencies, the target will be considered out of date if
510 it does not exist. Without a phony build statement, Ninja will report
511 an error if the file does not exist and is required by the build.
514 Default target statements
515 ~~~~~~~~~~~~~~~~~~~~~~~~~
517 By default, if no targets are specified on the command line, Ninja
518 will build every output that is not named as an input elsewhere.
519 You can override this behavior using a default target statement.
520 A default target statement causes Ninja to build only a given subset
521 of output files if none are specified on the command line.
523 Default target statements begin with the `default` keyword, and have
524 the format +default _targets_+. A default target statement must appear
525 after the build statement that declares the target as an output file.
526 They are cumulative, so multiple statements may be used to extend
527 the list of default targets. For example:
534 This causes Ninja to build the `foo`, `bar` and `baz` targets by
541 For each built file, Ninja keeps a log of the command used to build
542 it. Using this log Ninja can know when an existing output was built
543 with a different command line than the build files specify (i.e., the
544 command line changed) and knows to rebuild the file.
546 The log file is kept in the build root in a file called `.ninja_log`.
547 If you provide a variable named `builddir` in the outermost scope,
548 `.ninja_log` will be kept in that directory instead.
554 A file is a series of declarations. A declaration can be one of:
556 1. A rule declaration, which begins with +rule _rulename_+, and
557 then has a series of indented lines defining variables.
559 2. A build edge, which looks like +build _output1_ _output2_:
560 _rulename_ _input1_ _input2_+. +
561 Implicit dependencies may be tacked on the end with +|
562 _dependency1_ _dependency2_+. +
563 Order-only dependencies may be tacked on the end with +||
564 _dependency1_ _dependency2_+. (See <<ref_dependencies,the reference on
567 3. Variable declarations, which look like +_variable_ = _value_+.
569 4. Default target statements, which look like +default _target1_ _target2_+.
571 5. References to more files, which look like +subninja _path_+ or
572 +include _path_+. The difference between these is explained below
573 <<ref_scope,in the discussion about scoping>>.
578 Ninja is mostly encoding agnostic, as long as the bytes Ninja cares
579 about (like slashes in paths) are ASCII. This means e.g. UTF-8 or
580 ISO-8859-1 input files ought to work. (To simplify some code, tabs
581 and carriage returns are currently disallowed; this could be fixed if
582 it really mattered to you.)
584 Comments begin with `#` and extend to the end of the line.
586 Newlines are significant. Statements like `build foo bar` are a set
587 of space-separated tokens that end at the newline. Newlines and
588 spaces within a token must be escaped.
590 There is only one escape character, `$`, and it has the following
594 `$` followed by a newline:: escape the newline (continue the current line
595 across a line break).
597 `$` followed by text:: a variable reference.
599 `${varname}`:: alternate syntax for `$varname`.
601 `$` followed by space:: a space. (This is only necessary in lists of
602 paths, where a space would otherwise separate filenames. See below.)
604 `$:` :: a colon. (This is only necessary in `build` lines, where a colon
605 would otherwise terminate the list of inputs.)
607 `$$`:: a literal `$`.
609 A `build` or `default` statement is first parsed as a space-separated
610 list of filenames and then each name is expanded. This means that
611 spaces within a variable will result in spaces in the expanded
616 build $spaced/baz other$ file: ...
617 # The above build line has two outputs: "foo bar/baz" and "other file".
620 In a `name = value` statement, whitespace at the beginning of a value
621 is always stripped. Whitespace at the beginning of a line after a
622 line continuation is also stripped.
625 two_words_with_one_space = foo $
627 one_word_with_no_space = foo$
631 Other whitespace is only significant if it's at the beginning of a
632 line. If a line is indented more than the previous one, it's
633 considered part of its parent's scope; if it is indented less than the
634 previous one, it closes the previous scope.
640 A `rule` block contains a list of `key = value` declarations that
641 affect the processing of the rule. Here is a full list of special
644 `command` (_required_):: the command line to run. This string (after
645 $variables are expanded) is passed directly to `sh -c` without
646 interpretation by Ninja. Each `rule` may have only one `command`
647 declaration. To specify multiple commands use `&&` (or similar) to
648 concatenate operations.
650 `depfile`:: path to an optional `Makefile` that contains extra
651 _implicit dependencies_ (see <<ref_dependencies,the reference on
652 dependency types>>). This is explicitly to support `gcc` and its `-M`
653 family of flags, which output the list of headers a given `.c` file
656 Use it like in the following example:
661 command = gcc -MMD -MF $out.d [other gcc flags here]
664 When loading a `depfile`, Ninja implicitly adds edges such that it is
665 not an error if the listed dependency is missing. This allows you to
666 delete a depfile-discovered header file and rebuild, without the build
667 aborting due to a missing input.
669 `description`:: a short description of the command, used to pretty-print
670 the command as it's running. The `-v` flag controls whether to print
671 the full command or its description; if a command fails, the full command
672 line will always be printed before the command's output.
674 `generator`:: if present, specifies that this rule is used to
675 re-invoke the generator program. Files built using `generator`
676 rules are treated specially in two ways: firstly, they will not be
677 rebuilt if the command line changes; and secondly, they are not
680 `restat`:: if present, causes Ninja to re-stat the command's outputs
681 after execution of the command. Each output whose modification time
682 the command did not change will be treated as though it had never
683 needed to be built. This may cause the output's reverse
684 dependencies to be removed from the list of pending build actions.
686 `rspfile`, `rspfile_content`:: if present (both), Ninja will use a
687 response file for the given command, i.e. write the selected string
688 (`rspfile_content`) to the given file (`rspfile`) before calling the
689 command and delete the file after successful execution of the
692 This is particularly useful on Windows OS, where the maximal length of
693 a command line is limited and response files must be used instead.
695 Use it like in the following example:
699 command = link.exe /OUT$out [usual link flags here] @$out.rsp
701 rspfile_content = $in
703 build myapp.exe: link a.obj b.obj [possibly many other .obj files]
706 Finally, the special `$in` and `$out` variables expand to the
707 shell-quoted space-separated list of files provided to the `build`
708 line referencing this `rule`.
714 There are three types of build dependencies which are subtly different.
716 1. _Explicit dependencies_, as listed in a build line. These are
717 available as the `$in` variable in the rule. Changes in these files
718 cause the output to be rebuilt; if these file are missing and
719 Ninja doesn't know how to build them, the build is aborted.
721 This is the standard form of dependency to be used for e.g. the
722 source file of a compile command.
724 2. _Implicit dependencies_, either as picked up from
725 a `depfile` attribute on a rule or from the syntax +| _dep1_
726 _dep2_+ on the end of a build line. The semantics are identical to
727 explicit dependencies, the only difference is that implicit dependencies
728 don't show up in the `$in` variable.
730 This is for expressing dependencies that don't show up on the
731 command line of the command; for example, for a rule that runs a
732 script, the script itself should be an implicit dependency, as
733 changes to the script should cause the output to rebuild.
735 Note that dependencies as loaded through depfiles have slightly different
736 semantics, as described in the <<ref_rule,rule reference>>.
738 3. _Order-only dependencies_, expressed with the syntax +|| _dep1_
739 _dep2_+ on the end of a build line. When these are out of date, the
740 output is not rebuilt until they are built, but changes in order-only
741 dependencies alone do not cause the output to be rebuilt.
743 Order-only dependencies can be useful for bootstrapping dependencies
744 that are only discovered during build time: for example, to generate a
745 header file before starting a subsequent compilation step. (Once the
746 header is used in compilation, a generated dependency file will then
747 express the implicit dependency.)
749 Evaluation and scoping
750 ~~~~~~~~~~~~~~~~~~~~~~
753 Top-level variable declarations are scoped to the file they occur in.
755 The `subninja` keyword, used to include another `.ninja` file,
756 introduces a new scope. The included `subninja` file may use the
757 variables from the parent file, and shadow their values for the file's
758 scope, but it won't affect values of the variables in the parent.
760 To include another `.ninja` file in the current scope, much like a C
761 `#include` statement, use `include` instead of `subninja`.
763 Variable declarations indented in a `build` block are scoped to the
764 `build` block. This scope is inherited by the `rule`. The full
765 lookup order for a variable referenced in a rule is:
767 1. Rule-level variables (i.e. `$in`, `$command`).
769 2. Build-level variables from the `build` that references this rule.
771 3. File-level variables from the file that the `build` line was in.
773 4. Variables from the file that included that file using the
779 Variables are expanded in paths (in a `build` or `default` statement)
780 and on the right side of a `name = value` statement.
782 When a `name = value` statement is evaluated, its right-hand side is
783 expanded once (according to the above scoping rules) immediately, and
784 from then on `$name` expands to the static string as the result of the
785 expansion. It is never the case that you'll need to "double-escape" a
786 value to prevent it from getting expanded twice.