Want to learn about Boost.Build features? Start with the tutorial and continue with the overview. When you’re ready to try Boost.Build in practice, go to the installation.

Building a project with Boost.Build? See the installation and then read the overview.

Setting up Boost.Build on your project? Take a look at the overview and extender manual.

If there’s anything you find unclear in this documentation, report the problem directly in the issue tracker. For more general questions, please post them to our mailing list (http://boost.org/more/mailing_lists.htm#jamboost). The mailing list is also mirrored to the newsgroup news://news.gmane.org/gmane.comp.lib.boost.build[].

Copyright 2006, 2014 Vladimir Prus. Distributed under the Boost Software License, Version 1.0. (See accompanying file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)

1. Installation

To install Boost.Build from an official release or a nightly build, as available on the official web site, follow these steps:

  1. Unpack the release. On the command line, go to the root of the unpacked tree.

  2. Run either .\bootstrap.bat (on Windows), or ./bootstrap.sh (on other operating systems).

  3. Run

    $ ./b2 install --prefix=PREFIX

    where PREFIX is a directory where you want Boost.Build to be installed.

  4. Optionally, add PREFIX/bin to your PATH environment variable.

If you are not using a Boost.Build package, but rather the version bundled with the Boost C++ Libraries, the above commands should be run in the tools/build directory.

Now that Boost.Build is installed, you can try some of the examples. Copy PREFIX/share/boost-build/examples/hello to a different directory, then change to that directory and run:

$ PREFIX/bin/b2

A simple executable should be built.

2. Tutorial

This section will guide you though the most basic features of Boost.Build. We will start with the “Hello, world” example, learn how to use libraries, and finish with testing and installing features.

2.1. Hello, world

The simplest project that Boost.Build can construct is stored in example/hello/ directory. The project is described by a file called Jamroot that contains:

exe hello : hello.cpp ;

Even with this simple setup, you can do some interesting things. First of all, just invoking b2 will build the hello executable by compiling and linking hello.cpp. By default, the debug variant is built. Now, to build the release variant of hello, invoke

b2 release

Note that the debug and release variants are created in different directories, so you can switch between variants or even build multiple variants at once, without any unnecessary recompilation. Let us extend the example by adding another line to our project’s Jamroot:

exe hello2 : hello.cpp ;

Now let us build both the debug and release variants of our project again:

b2 debug release

Note that two variants of hello2 are linked. Since we have already built both variants of hello, hello.cpp will not be recompiled; instead the existing object files will just be linked into the corresponding variants of hello2. Now let us remove all the built products:

b2 --clean debug release

It is also possible to build or clean specific targets. The following two commands, respectively, build or clean only the debug version of hello2.

b2 hello2
b2 --clean hello2

2.2. Properties

To represent aspects of target configuration such as debug and release variants, or single- and multi-threaded builds portably, Boost.Build uses features with associated values. For example, the debug-symbols feature can have a value of on or off. A property is just a (feature, value) pair. When a user initiates a build, Boost.Build automatically translates the requested properties into appropriate command-line flags for invoking toolset components like compilers and linkers.

There are many built-in features that can be combined to produce arbitrary build configurations. The following command builds the project’s release variant with inlining disabled and debug symbols enabled:

b2 release inlining=off debug-symbols=on

Properties on the command-line are specified with the syntax:


The release and debug that we have seen in b2 invocations are just a shorthand way to specify values of the variant feature. For example, the command above could also have been written this way:

b2 variant=release inlining=off debug-symbols=on

variant is so commonly-used that it has been given special status as an implicit feature—Boost.Build will deduce its identity just from the name of one of its values.

A complete description of features can be found in the section called “Features and properties”.

2.2.1. Build Requests and Target Requirements

The set of properties specified on the command line constitutes a build request—a description of the desired properties for building the requested targets (or, if no targets were explicitly requested, the project in the current directory). The actual properties used for building targets are typically a combination of the build request and properties derived from the project’s Jamroot (and its other Jamfiles, as described in the section called “Project Hierarchies”). For example, the locations of `#include`d header files are normally not specified on the command-line, but described in Jamfiles as target requirements and automatically combined with the build request for those targets. Multithread-enabled compilation is another example of a typical target requirement. The Jamfile fragment below illustrates how these requirements might be specified.

exe hello
    : hello.cpp
    : <include>boost <threading>multi

When hello is built, the two requirements specified above will always be present. If the build request given on the b2 command-line explictly contradicts a target’s requirements, the target requirements usually override (or, in the case of “free” features like <include>, [1] augment) the build request.

The value of the <include> feature is relative to the location of Jamroot where it is used.

2.2.2. Project Attributes

If we want the same requirements for our other target, hello2, we could simply duplicate them. However, as projects grow, that approach leads to a great deal of repeated boilerplate in Jamfiles. Fortunately, there’s a better way. Each project can specify a set of attributes, including requirements:

    : requirements <include>/home/ghost/Work/boost <threading>multi

exe hello : hello.cpp ;
exe hello2 : hello.cpp ;

The effect would be as if we specified the same requirement for both hello and hello2.

2.3. Project Hierarchies

So far we have only considered examples with one project, with one user-written Boost.Jam file, Jamroot. A typical large codebase would be composed of many projects organized into a tree. The top of the tree is called the project root. Every subproject is defined by a file called Jamfile in a descendant directory of the project root. The parent project of a subproject is defined by the nearest Jamfile or Jamroot file in an ancestor directory. For example, in the following directory layout:

  +-- Jamroot
  +-- app/
  |    |
  |    +-- Jamfile
  |    `-- app.cpp
  `-- util/
       +-- foo/
       .    |
       .    +-- Jamfile
       .    `-- bar.cpp

the project root is top/. The projects in top/app/ and top/util/foo/ are immediate children of the root project.

When we refer to a “Jamfile,” set in normal type, we mean a file called either Jamfile or Jamroot. When we need to be more specific, the filename will be set as “Jamfile” or “Jamroot.”

Projects inherit all attributes (such as requirements) from their parents. Inherited requirements are combined with any requirements specified by the subproject. For example, if top/Jamroot has


in its requirements, then all of its subprojects will have it in their requirements, too. Of course, any project can add include paths to those specified by its parents. [2] More details can be found in the section called “Projects”.

Invoking b2 without explicitly specifying any targets on the command line builds the project rooted in the current directory. Building a project does not automatically cause its subprojects to be built unless the parent project’s Jamfile explicitly requests it. In our example, top/Jamroot might contain:

build-project app ;

which would cause the project in top/app/ to be built whenever the project in top/ is built. However, targets in top/util/foo/ will be built only if they are needed by targets in top/ or top/app/.

2.4. Dependent Targets

When building a target X that depends on first building another target Y (such as a library that must be linked with X), Y is called a dependency of X and X is termed a dependent of Y.

To get a feeling of target dependencies, let’s continue the above example and see how top/app/Jamfile can use libraries from top/util/foo. If top/util/foo/Jamfile contains

lib bar : bar.cpp ;

then to use this library in top/app/Jamfile, we can write:

exe app : app.cpp ../util/foo//bar ;

While app.cpp refers to a regular source file, ../util/foo//bar is a reference to another target: a library bar declared in the Jamfile at ../util/foo.

Some other build system have special syntax for listing dependent libraries, for example LIBS variable. In Boost.Build, you just add the library to the list of sources.

Suppose we build app with:

b2 app optimization=full define=USE_ASM

Which properties will be used to build foo? The answer is that some features are propagated — Boost.Build attempts to use dependencies with the same value of propagated features. The <optimization> feature is propagated, so both app and foo will be compiled with full optimization. But <define> is not propagated: its value will be added as-is to the compiler flags for a.cpp, but won’t affect foo.

Let’s improve this project further. The library probably has some headers that must be used when compiling app.cpp. We could manually add the necessary #include paths to app’s requirements as values of the `<include> ` feature, but then this work will be repeated for all programs that use `foo. A better solution is to modify util/foo/Jamfile in this way:

    : usage-requirements <include>.

lib foo : foo.cpp ;

Usage requirements are applied not to the target being declared but to its dependents. In this case, <include>. will be applied to all targets that directly depend on foo.

Another improvement is using symbolic identifiers to refer to the library, as opposed to Jamfile location. In a large project, a library can be used by many targets, and if they all use Jamfile location, a change in directory organization entails much work. The solution is to use project ids—symbolic names not tied to directory layout. First, we need to assign a project id by adding this code to Jamroot:

use-project /library-example/foo : util/foo ;

Second, we modify app/Jamfile to use the project id:

exe app : app.cpp /library-example/foo//bar ;

The /library-example/foo//bar syntax is used to refer to the target bar in the project with id /library-example/foo. We’ve achieved our goal—if the library is moved to a different directory, only Jamroot must be modified. Note that project ids are global—two Jamfiles are not allowed to assign the same project id to different directories.

If you want all applications in some project to link to a certain library, you can avoid having to specify directly the sources of every target by using the <library> property. For example, if /boost/filesystem//fs should be linked to all applications in your project, you can add <library>/boost/filesystem//fs to the project’s requirements, like this:

   : requirements <library>/boost/filesystem//fs

2.5. Static and shared libraries

Libraries can be either static, which means they are included in executable files that use them, or shared (a.k.a. dynamic), which are only referred to from executables, and must be available at run time. Boost.Build can create and use both kinds.

The kind of library produced from a lib target is determined by the value of the link feature. Default value is shared, and to build a static library, the value should be static. You can request a static build either on the command line:

b2 link=static

or in the library’s requirements:

lib l : l.cpp : <link>static ;

We can also use the <link> property to express linking requirements on a per-target basis. For example, if a particular executable can be correctly built only with the static version of a library, we can qualify the executable’s target reference to the library as follows:

exe important : main.cpp helpers/<link>static ;

No matter what arguments are specified on the b2 command line, important will only be linked with the static version of helpers.

Specifying properties in target references is especially useful if you use a library defined in some other project (one you can’t change) but you still want static (or dynamic) linking to that library in all cases. If that library is used by many targets, you could use target references everywhere:

exe e1 : e1.cpp /other_project//bar/<link>static ;
exe e10 : e10.cpp /other_project//bar/<link>static ;

but that’s far from being convenient. A better approach is to introduce a level of indirection. Create a local alias target that refers to the static (or dynamic) version of foo:

alias foo : /other_project//bar/<link>static ;
exe e1 : e1.cpp foo ;
exe e10 : e10.cpp foo ;

The alias rule is specifically used to rename a reference to a target and possibly change the properties.

When one library uses another, you put the second library in the source list of the first. For example:

lib utils : utils.cpp /boost/filesystem//fs ;
lib core : core.cpp utils ;
exe app : app.cpp core ;

This works no matter what kind of linking is used. When core is built as a shared library, links utils directly into it. Static libraries can’t link to other libraries, so when core is built as a static library, its dependency on utils is passed along to core’s dependents, causing `app to be linked with both core and utils.

(Note for non-UNIX system). Typically, shared libraries must be installed to a directory in the dynamic linker’s search path. Otherwise, applications that use shared libraries can’t be started. On Windows, the dynamic linker’s search path is given by the PATH environment variable. This restriction is lifted when you use Boost.Build testing facilities—the PATH variable will be automatically adjusted before running the executable.

2.6. Conditions and alternatives

Sometimes, particular relationships need to be maintained among a target’s build properties. For example, you might want to set specific #define when a library is built as shared, or when a target’s release variant is built. This can be achieved using conditional requirements.

lib network : network.cpp
    : <link>shared:<define>NETWORK_LIB_SHARED

In the example above, whenever network is built with <link>shared, <define>NETWORK_LIB_SHARED will be in its properties, too. Also, whenever its release variant is built, <define>EXTRA_FAST will appear in its properties.

Sometimes the ways a target is built are so different that describing them using conditional requirements would be hard. For example, imagine that a library actually uses different source files depending on the toolset used to build it. We can express this situation using target alternatives:

lib demangler : dummy_demangler.cpp ;                (1)
lib demangler : demangler_gcc.cpp : <toolset>gcc ;   (2)
lib demangler : demangler_msvc.cpp : <toolset>msvc ; (3)

When building demangler, Boost.Build will compare requirements for each alternative with build properties to find the best match. For example, when building with <toolset>gcc alternative (2), will be selected, and when building with <toolset>msvc alternative (3) will be selected. In all other cases, the most generic alternative (1) will be built.

2.7. Prebuilt targets

To link to libraries whose build instructions aren’t given in a Jamfile, you need to create lib targets with an appropriate file property. Target alternatives can be used to associate multiple library files with a single conceptual target. For example:

# util/lib2/Jamfile
lib lib2
    : <file>lib2_release.a <variant>release

lib lib2
    : <file>lib2_debug.a <variant>debug

This example defines two alternatives for lib2, and for each one names a prebuilt file. Naturally, there are no sources. Instead, the <file> feature is used to specify the file name.

Once a prebuilt target has been declared, it can be used just like any other target:

exe app : app.cpp ../util/lib2//lib2 ;

As with any target, the alternative selected depends on the properties propagated from lib2’s dependents. If we build the release and debug versions of `app it will be linked with lib2_release.a and lib2_debug.a, respectively.

System libraries — those that are automatically found by the toolset by searching through some set of predetermined paths — should be declared almost like regular ones:

lib pythonlib : : <name>python22 ;

We again don’t specify any sources, but give a name that should be passed to the compiler. If the gcc toolset were used to link an executable target to pythonlib, -lpython22 would appear in the command line (other compilers may use different options).

We can also specify where the toolset should look for the library:

lib pythonlib : : <name>python22 <search>/opt/lib ;

And, of course, target alternatives can be used in the usual way:

lib pythonlib : : <name>python22 <variant>release ;
lib pythonlib : : <name>python22_d <variant>debug ;

A more advanced use of prebuilt targets is described in the section called “Targets in site-config.jam”.

3. Overview

This section will provide the information necessary to create your own projects using Boost.Build. The information provided here is relatively high-level, and the Reference as well as the on-line help system must be used to obtain low-level documentation (see --help).

Boost.Build has two parts — a build engine with its own interpreted language, and Boost.Build itself, implemented in that language. The chain of events when you type b2 on the command line is as follows:

  1. The Boost.Build executable tries to find Boost.Build modules and loads the top-level module. The exact process is described in the section called “Initialization”

  2. The top-level module loads user-defined configuration files, user-config.jam and site-config.jam, which define available toolsets.

  3. The Jamfile in the current directory is read. That in turn might cause reading of further Jamfiles. As a result, a tree of projects is created, with targets inside projects.

  4. Finally, using the build request specified on the command line, Boost.Build decides which targets should be built and how. That information is passed back to Boost.Jam, which takes care of actually running the scheduled build action commands.

So, to be able to successfully use Boost.Build, you need to know only four things:

3.1. Concepts

Boost.Build has a few unique concepts that are introduced in this section. The best way to explain the concepts is by comparison with more classical build tools.

When using any flavour of make, you directly specify targets and commands that are used to create them from other target. The below example creates a.o from a.c using a hardcoded compiler invocation command.

a.o: a.c
    g++ -o a.o -g a.c

This is a rather low-level description mechanism and it’s hard to adjust commands, options, and sets of created targets depending on the compiler and operating system used.

To improve portability, most modern build system provide a set of higher-level functions that can be used in build description files. Consider this example:

add_program ("a", "a.c")

This is a function call that creates the targets necessary to create an executable file from the source file a.c. Depending on configured properties, different command lines may be used. However, add_program is higher-level, but rather thin level. All targets are created immediately when the build description is parsed, which makes it impossible to perform multi-variant builds. Often, change in any build property requires a complete reconfiguration of the build tree.

In order to support true multivariant builds, Boost.Build introduces the concept of a metatarget definition main target metataget metatarget — an object that is created when the build description is parsed and can be called later with specific build properties to generate actual targets.

Consider an example:

exe a : a.cpp ;

When this declaration is parsed, Boost.Build creates a metatarget, but does not yet decide what files must be created, or what commands must be used. After all build files are parsed, Boost.Build considers the properties requested on the command line. Supposed you have invoked Boost.Build with:

b2 toolset=gcc toolset=msvc

In that case, the metatarget will be called twice, once with toolset=gcc and once with toolset=msvc. Both invocations will produce concrete targets, that will have different extensions and use different command lines.

Another key concept is build property. A build property is a variable that affects the build process. It can be specified on the command line, and is passed when calling a metatarget. While all build tools have a similar mechanism, Boost.Build differs by requiring that all build properties are declared in advance, and providing a large set of properties with portable semantics.

The final concept is property propagation. Boost.Build does not require that every metatarget is called with the same properties. Instead, the "top-level" metatargets are called with the properties specified on the command line. Each metatarget can elect to augment or override some properties (in particular, using the requirements mechanism, see the section called “Requirements”). Then, the dependency metatargets are called with the modified properties and produce concrete targets that are then used in the build process. Of course, dependency metatargets maybe in turn modify build properties and have dependencies of their own.

For a more in-depth treatment of the requirements and concepts, you may refer to SYRCoSE 2009 Boost.Build article.

3.2. Boost.Jam Language

This section will describe the basics of the Boost.Jam language—just enough for writing Jamfiles. For more information, please see the Boost.Jam documentation.

Boost.Jam has an interpreted, procedural language. On the lowest level, a Boost.Jam program consists of variables and rules (the Jam term for functions). They are grouped into modules—there is one global module and a number of named modules. Besides that, a Boost.Jam program contains classes and class instances.

Syntantically, a Boost.Jam program consists of two kinds of elements—keywords (which have a special meaning to Boost.Jam) and literals. Consider this code:

a = b ;

which assigns the value b to the variable a. Here, = and ; are keywords, while a and b are literals.

All syntax elements, even keywords, must be separated by spaces. For example, omitting the space character before ; will lead to a syntax error.

If you want to use a literal value that is the same as some keyword, the value can be quoted:

a = "=" ;

All variables in Boost.Jam have the same type—list of strings. To define a variable one assigns a value to it, like in the previous example. An undefined variable is the same as a variable with an empty value. Variables can be accessed using the $(variable) syntax. For example:

a = $(b) $(c) ;

Rules are defined by specifying the rule name, the parameter names, and the allowed value list size for each parameter.

rule example
     parameter1 :
     parameter2 ? :
     parameter3 + :
     parameter4 *
    # rule body

When this rule is called, the list passed as the first argument must have exactly one value. The list passed as the second argument can either have one value of be empty. The two remaining arguments can be arbitrarily long, but the third argument may not be empty.

The overview of Boost.Jam language statements is given below:

helper 1 : 2 : 3 ;
x = [ helper 1 : 2 : 3 ] ;

This code calls the named rule with the specified arguments. When the result of the call must be used inside some expression, you need to add brackets around the call, like shown on the second line.

if cond { statements } [ else { statements } ]

This is a regular if-statement. The condition is composed of:

  • Literals (true if at least one string is not empty)

  • Comparisons: a operator b where operator is one of =, !=, <, >, or >=. The comparison is done pairwise between each string in the left and the right arguments.

  • Logical operations: ! a, a && b, a || b

  • Grouping: ( cond )

for var in list { statements }

Executes statements for each element in list, setting the variable var to the element value.

while cond { statements }

Repeatedly execute statements while cond remains true upon entry.

return values ;

This statement should be used only inside a rule and returns values to the caller of the rule.

import module ;
import module : rule ;

The first form imports the specified module. All rules from that module are made available using the qualified name: module.rule. The second form imports the specified rules only, and they can be called using unqualified names.

Sometimes, you need to specify the actual command lines to be used when creating targets. In the jam language, you use named actions to do this. For example:

actions create-file-from-another
    create-file-from-another $(<) $(>)

This specifies a named action called create-file-from-another. The text inside braces is the command to invoke. The $(<) variable will be expanded to a list of generated files, and the $(>) variable will be expanded to a list of source files.

To adjust the command line flexibly, you can define a rule with the same name as the action and taking three parameters — targets, sources and properties. For example:

rule create-file-from-another ( targets * : sources * : properties * )
   if <variant>debug in $(properties)
       OPTIONS on $(targets) = --debug ;
actions create-file-from-another
    create-file-from-another $(OPTIONS) $(<) $(>)

In this example, the rule checks if a certain build property is specified. If so, it sets the variable OPTIONS that is then used inside the action. Note that the variables set "on a target" will be visible only inside actions building that target, not globally. Were they set globally, using variable named OPTIONS in two unrelated actions would be impossible.

More details can be found in the Jam reference, the section called “Rules”.

3.3. Configuration

On startup, Boost.Build searches and reads three configuration files: site-config.jam, user-config.jam, and project-config.jam. The first one is usually installed and maintained by a system administrator, and the second is for the user to modify. You can edit the one in the top-level directory of your Boost.Build installation or create a copy in your home directory and edit the copy. The third is used for project specific configuration. The following table explains where the files are searched.

Table 1. Search paths for configuration files
site-config.jam user-config.jam project-config.jam























Any of these files may also be overridden on the command line.

You can use the --debug-configuration option to find which configuration files are actually loaded.

Usually, user-config.jam just defines the available compilers and other tools (see the section called “Targets in site-config.jam” for more advanced usage). A tool is configured using the following syntax:

using tool-name : ... ;

The using rule is given the name of tool, and will make that tool available to Boost.Build. For example,

using gcc ;

will make the GCC compiler available.

All the supported tools are documented in the section called “Builtin tools”, including the specific options they take. Some general notes that apply to most C++ compilers are below.

For all the C++ compiler toolsets that Boost.Build supports out-of-the-box, the list of parameters to using is the same: toolset-name, version, invocation-command, and options.

If you have a single compiler, and the compiler executable

  • has its “usual name” and is in the PATH, or

  • was installed in a standard “installation directory”, or

  • can be found using a global system like the Windows registry.

it can be configured by simply:

using tool-name ;

If the compiler is installed in a custom directory, you should provide the command that invokes the compiler, for example:

using gcc : : g++-3.2 ;
using msvc : : "Z:/Programs/Microsoft Visual Studio/vc98/bin/cl" ;

Some Boost.Build toolsets will use that path to take additional actions required before invoking the compiler, such as calling vendor-supplied scripts to set up its required environment variables. When the compiler executables for C and C are different, the path to the C compiler executable must be specified. The command can be any command allowed by the operating system. For example:

using msvc : : echo Compiling && foo/bar/baz/cl ;

will work.

To configure several versions of a toolset, simply invoke the using rule multiple times:

using gcc : 3.3 ;
using gcc : 3.4 : g++-3.4 ;
using gcc : 3.2 : g++-3.2 ;
using gcc : 5 ;
using clang : 3.9 ;
using msvc : 14.0 ;

Note that in the first call to using, the compiler found in the PATH will be used, and there is no need to explicitly specify the command.

Many of toolsets have an options parameter to fine-tune the configuration. All of Boost.Build’s standard compiler toolsets accept four options cflags, cxxflags, compileflags and linkflags as options specifying flags that will be always passed to the corresponding tools. There must not be a space between the tag for the option name and the value. Values of the cflags feature are passed directly to the C compiler, values of the cxxflags feature are passed directly to the C++ compiler, and values of the compileflags feature are passed to both. For example, to configure a gcc toolset so that it always generates 64-bit code you could write:

using gcc : 3.4 : : <compileflags>-m64 <linkflags>-m64 ;

If multiple of the same type of options are needed, they can be concatenated with quotes or have multiple instances of the option tag.

using gcc : 5 : : <cxxflags>"-std=c++14 -O2" ;
using clang : 3.9 : : <cxxflags>-std=c++14 <cxxflags>-O2 ;

Multiple varaiations of the same tool can be used for most tools. These are deliniated by the version passed in. Because the dash '-' cannot be used here, the convention has become to use the tilde '~' to deliniate variations.

using gcc : 5 : g++-5 : ; # default is C++ 98
using gcc : 5~c++03 : g++-5 : <cxxflags>-std=c++03 ; # C++ 03
using gcc : 5~gnu03 : g++-5 : <cxxflags>-std=gnu++03 ; # C++ 03 with GNU
using gcc : 5~c++11 : g++-5 : <cxxflags>-std=c++11 ; # C++ 11
using gcc : 5~c++14 : g++-5 : <cxxflags>-std=c++14 ; # C++ 14

These are then used as normal toolsets:

b2 toolset=gcc-5 stage
b2 toolset=gcc-5~c++14 stage
Although the syntax used to specify toolset options is very similar to syntax used to specify requirements in Jamfiles, the toolset options are not the same as features. Don’t try to specify a feature value in toolset initialization.

3.4. Invocation

To invoke Boost.Build, type b2 on the command line. Three kinds of command-line tokens are accepted, in any order:


Options start with either one or two dashes. The standard options are listed below, and each project may add additional options


Properties specify details of what you want to build (e.g. debug or release variant). Syntactically, all command line tokens with an equal sign in them are considered to specify properties. In the simplest form, a property looks like feature=value


All tokens that are neither options nor properties specify what targets to build. The available targets entirely depend on the project you are building.

3.4.1. Examples

To build all targets defined in the Jamfile in the current directory with the default properties, run:


To build specific targets, specify them on the command line:

b2 lib1 subproject//lib2

To request a certain value for some property, add property=value to the command line:

b2 toolset=gcc variant=debug optimization=space

3.4.2. Options

Boost.Build recognizes the following command line options.


Invokes the online help system. This prints general information on how to use the help system with additional --help* options.


Cleans all targets in the current directory and in any subprojects. Note that unlike the clean target in make, you can use --clean together with target names to clean specific targets.


Cleans all targets, no matter where they are defined. In particular, it will clean targets in parent Jamfiles, and targets defined under other project roots.


Changes the build directories for all project roots being built. When this option is specified, all Jamroot files must declare a project name. The build directory for the project root will be computed by concatenating the value of the --build-dir option, the project name specified in Jamroot, and the build dir specified in Jamroot (or bin, if none is specified). + The option is primarily useful when building from read-only media, when you can’t modify Jamroot.


Compresses target paths by abbreviating each component. This option is useful to keep paths from becoming longer than the filesystem supports. See also the section called “Target Paths”.


Compresses target paths using an MD5 hash. This option is useful to keep paths from becoming longer than the filesystem supports. This option produces shorter paths than --abbreviate-paths does, but at the cost of making them less understandable. See also the section called “Target Paths”.


Prints information on the Boost.Build and Boost.Jam versions.


Causes all files to be rebuilt.


Do not execute the commands, only print them.


Stop at the first error, as opposed to continuing to build targets that don’t depend on the failed ones.

-j N

Run up to N commands in parallel.


Override all configuration files


Override the default site-config.jam


Override the default user-config.jam


Override the default project-config.jam


Produces debug information about the loading of Boost.Build and toolset files.


Prints what targets are being built and with what properties.


Produces debug output from the generator search process. Useful for debugging custom generators.


Suppress all informational messages.

-d N

Enable cumulative debugging levels from 1 to n. Values are: +

  1. Show the actions taken for building targets, as they are executed (the default).

  2. Show "quiet" actions and display all action text, as they are executed.

  3. Show dependency analysis, and target/source timestamps/paths.

  4. Show arguments and timing of shell invocations.

  5. Show rule invocations and variable expansions.

  6. Show directory/header file/archive scans, and attempts at binding to targets.

  7. Show variable settings.

  8. Show variable fetches, variable expansions, and evaluation of '"if"' expressions.

  9. Show variable manipulation, scanner tokens, and memory usage.

  10. Show profile information for rules, both timing and memory.

  11. Show parsing progress of Jamfiles.

  12. Show graph of target dependencies.

  13. Show change target status (fate).

-d +N

Enable debugging level N.

-o file

Write the updating actions to the specified file instead of running them.

-s var=value

Set the variable var to value in the global scope of the jam language interpreter, overriding variables imported from the environment.

3.4.3. Properties

In the simplest case, the build is performed with a single set of properties, that you specify on the command line with elements in the form feature=value. The complete list of features can be found in the section called “Builtin features”. The most common features are summarized below.

Feature Allowed values Notes





Determines if Boost.Build creates shared or static libraries



Cause the produced binaries to be thread-safe. This requires proper support in the source code itself.



Explicitly request either 32-bit or 64-bit code generation. This typically requires that your compiler is appropriately configured. Please refer to the section called “C++ Compilers” and your compiler documentation in case of problems.


(Depends on configuration)

The C compiler to use. See link:#bbv2.reference.tools.compilers[the section called “C Compilers”] for a detailed list.


(Arbitrary string)

Additional include paths for C and C++ compilers.


(Arbitrary string)

Additional macro definitions for C and C++ compilers. The string should be either SYMBOL or SYMBOL=VALUE


(Arbitrary string)

Custom options to pass to the C++ compiler.


(Arbitrary string)

Custom options to pass to the C compiler.


(Arbitrary string)

Custom options to pass to the C++ linker.



Determines if shared or static version of C and C++ runtimes should be used.

If you have more than one version of a given C++ toolset (e.g. configured in user-config.jam, or autodetected, as happens with msvc), you can request the specific version by passing toolset-version as the value of the toolset feature, for example toolset=msvc-8.0.

If a feature has a fixed set of values it can be specified more than once on the command line. In which case, everything will be built several times — once for each specified value of a feature. For example, if you use

b2 link=static link=shared threading=single threading=multi

Then a total of 4 builds will be performed. For convenience, instead of specifying all requested values of a feature in separate command line elements, you can separate the values with commas, for example:

b2 link=static,shared threading=single,multi

The comma has this special meaning only if the feature has a fixed set of values, so

b2 include=static,shared

is not treated specially.

Multiple features may be grouped by using a forwards slash.

b2 gcc/link=shared msvc/link=static,shared

This will build 3 different variants, altogether.

3.4.4. Targets

All command line elements that are neither options nor properties are the names of the targets to build. See the section called “Target identifiers and references”. If no target is specified, the project in the current directory is built.

3.5. Declaring Targets

A Main target is a user-defined named entity that can be built, for example an executable file. Declaring a main target is usually done using one of the main target rules described in the section called “Builtin rules”. The user can also declare custom main target rules as shown in the section called “Main target rules”.

Most main target rules in Boost.Build have the same common signature:

rule rule-name (
     main-target-name :
     sources + :
     requirements * :
     default-build * :
     usage-requirements * )
  • main-target-name is the name used to request the target on command line and to use it from other main targets. A main target name may contain alphanumeric characters, dashes (‘-’), and underscores (‘_’).

  • sources is the list of source files and other main targets that must be combined.

  • requirements is the list of properties that must always be present when this main target is built.

  • default-build is the list of properties that will be used unless some other value of the same feature is already specified, e.g. on the command line or by propagation from a dependent target.

  • usage-requirements is the list of properties that will be propagated to all main targets that use this one, i.e. to all its dependents.

Some main target rules have a different list of parameters as explicitly stated in their documentation.

The actual requirements for a target are obtained by refining the requirements of the project where the target is declared with the explicitly specified requirements. The same is true for usage-requirements. More details can be found in the section called “Property refinement”.

3.5.1. Name

The name of main target has two purposes. First, it’s used to refer to this target from other targets and from command line. Second, it’s used to compute the names of the generated files. Typically, filenames are obtained from main target name by appending system-dependent suffixes and prefixes.

The name of a main target can contain alphanumeric characters, dashes, undescores and dots. The entire name is significant when resolving references from other targets. For determining filenames, only the part before the first dot is taken. For example:

obj test.release : test.cpp : <variant>release ;
obj test.debug : test.cpp : <variant>debug ;

will generate two files named test.obj (in two different directories), not two files named test.release.obj and test.debug.obj.

3.5.2. Sources

The list of sources specifies what should be processed to get the resulting targets. Most of the time, it’s just a list of files. Sometimes, you’ll want to automatically construct the list of source files rather than having to spell it out manually, in which case you can use the glob rule. Here are two examples:

exe a : a.cpp ; (1)
exe b : [ glob *.cpp ] ; (2)
  1. a.cpp is the only source file

  2. all .cpp files in this directory are sources

Unless you specify a file with an absolute path, the name is considered relative to the source directory — which is typically the directory where the Jamfile is located, but can be changed as described in the section called “Projects”.

The list of sources can also refer to other main targets. Targets in the same project can be referred to by name, while targets in other projects must be qualified with a directory or a symbolic project name. The directory/project name is separated from the target name by a double forward slash. There is no special syntax to distinguish the directory name from the project name—the part before the double slash is first looked up as project name, and then as directory name. For example:

lib helper : helper.cpp ;
exe a : a.cpp helper ;
exe b : b.cpp ..//utils ; (1)
exe c : c.cpp /boost/program_options//program_options ;
  1. Since all project ids start with slash, “..” is a directory name.

The first exe uses the library defined in the same project. The second one uses some target (most likely a library) defined by a Jamfile one level higher. Finally, the third target uses a C++ Boost library, referring to it using its absolute symbolic name. More information about target references can be found in the section called “Dependent Targets” and the section called “Target identifiers and references”.

3.5.3. Requirements

Requirements are the properties that should always be present when building a target. Typically, they are includes and defines:

exe hello : hello.cpp : <include>/opt/boost <define>MY_DEBUG ;

There are a number of other features, listed in the section called “Builtin features”. For example if a library can only be built statically, or a file can’t be compiled with optimization due to a compiler bug, one can use.

lib util : util.cpp : <link>static ;
obj main : main.cpp : <optimization>off ;

Sometimes, particular relationships need to be maintained among a target’s build properties. This can be achieved with conditional requirements. For example, you might want to set specific #defines when a library is built as shared, or when a target’s release variant is built in release mode.

lib network : network.cpp
    : <link>shared:<define>NETWORK_LIB_SHARED

In the example above, whenever network is built with <link>shared, <define>NETWORK_LIB_SHARED will be in its properties, too.

You can use several properties in the condition, for example:

lib network : network.cpp
    : <toolset>gcc,<optimization>speed:<define>USE_INLINE_ASSEMBLER

A more powerful variant of conditional requirements is indirect conditional requirements. You can provide a rule that will be called with the current build properties and can compute additional properties to be added. For example:

lib network : network.cpp
    : <conditional>@my-rule
rule my-rule ( properties * )
    local result ;
    if <toolset>gcc <optimization>speed in $(properties)
        result += <define>USE_INLINE_ASSEMBLER ;
    return $(result) ;

This example is equivalent to the previous one, but for complex cases, indirect conditional requirements can be easier to write and understand.

Requirements explicitly specified for a target are usually combined with the requirements specified for the containing project. You can cause a target to completely ignore a specific project requirement using the syntax by adding a minus sign before the property, for example:

exe main : main.cpp : -<define>UNNECESSARY_DEFINE ;

This syntax is the only way to ignore free properties, such as defines, from a parent. It can be also useful for ordinary properties. Consider this example:

project test : requirements <threading>multi ;
exe test1 : test1.cpp ;
exe test2 : test2.cpp : <threading>single ;
exe test3 : test3.cpp : -<threading>multi ;

Here, test1 inherits the project requirements and will always be built in multi-threaded mode. The test2 target overrides the project’s requirements and will always be built in single-threaded mode. In contrast, the test3 target removes a property from the project requirements and will be built either in single-threaded or multi-threaded mode depending on which variant is requested by the user.

Note that the removal of requirements is completely textual: you need to specify exactly the same property to remove it.

3.5.4. Default Build

The default-build parameter is a set of properties to be used if the build request does not otherwise specify a value for features in the set. For example:

exe hello : hello.cpp : : <threading>multi ;

would build a multi-threaded target unless the user explicitly requests a single-threaded version. The difference between the requirements and the default-build is that the requirements cannot be overridden in any way.

3.5.5. Additional Information

The ways a target is built can be so different that describing them using conditional requirements would be hard. For example, imagine that a library actually uses different source files depending on the toolset used to build it. We can express this situation using target alternatives:

lib demangler : dummy_demangler.cpp ;                # alternative 1
lib demangler : demangler_gcc.cpp : <toolset>gcc ;   # alternative 2
lib demangler : demangler_msvc.cpp : <toolset>msvc ; # alternative 3

In the example above, when built with gcc or msvc, demangler will use a source file specific to the toolset. Otherwise, it will use a generic source file, dummy_demangler.cpp.

It is possible to declare a target inline, i.e. the "sources" parameter may include calls to other main rules. For example:

exe hello : hello.cpp
    [ obj helpers : helpers.cpp : <optimization>off ] ;

Will cause "helpers.cpp" to be always compiled without optimization. When referring to an inline main target, its declared name must be prefixed by its parent target’s name and two dots. In the example above, to build only helpers, one should run b2 hello..helpers.

When no target is requested on the command line, all targets in the current project will be built. If a target should be built only by explicit request, this can be expressed by the explicit rule:

explicit install_programs ;

3.6. Projects

As mentioned before, targets are grouped into projects, and each Jamfile is a separate project. Projects are useful because they allow us to group related targets together, define properties common to all those targets, and assign a symbolic name to the project that can be used in referring to its targets.

Projects are named using the project rule, which has the following syntax:

project id : attributes ;

Here, attributes is a sequence of rule arguments, each of which begins with an attribute-name and is followed by any number of build properties. The list of attribute names along with its handling is also shown in the table below. For example, it is possible to write:

project tennis
    : requirements <threading>multi
    : default-build release

The possible attributes are listed below.

Project id is a short way to denote a project, as opposed to the Jamfile’s pathname. It is a hierarchical path, unrelated to filesystem, such as "boost/thread". Target references make use of project ids to specify a target.

Source location specifies the directory where sources for the project are located.

Project requirements are requirements that apply to all the targets in the projects as well as all subprojects.

Default build is the build request that should be used when no build request is specified explicitly.

The default values for those attributes are given in the table below.

Attribute Name Default value Handling by the project rule

Project id



Assigned from the first parameter of the 'project' rule. It is assumed to denote absolute project id.

Source location


The location of jamfile for the project

Sets to the passed value



The parent’s requirements

The parent’s requirements are refined with the passed requirement and the result is used as the project requirements.

Default build



Sets to the passed value

Build directory


Empty if the parent has no build directory set. Otherwise, the parent’s build directory with the relative path from parent to the current project appended to it.

Sets to the passed value, interpreted as relative to the project’s location.

Besides defining projects and main targets, Jamfiles often invoke various utility rules. For the full list of rules that can be directly used in Jamfile see the section called “Builtin rules”.

Each subproject inherits attributes, constants and rules from its parent project, which is defined by the nearest Jamfile in an ancestor directory above the subproject. The top-level project is declared in a file called Jamroot rather than Jamfile. When loading a project, Boost.Build looks for either Jamroot or Jamfile. They are handled identically, except that if the file is called Jamroot, the search for a parent project is not performed.

Even when building in a subproject directory, parent project files are always loaded before those of their subprojects, so that every definition made in a parent project is always available to its children. The loading order of any other projects is unspecified. Even if one project refers to another via the use-project or a target reference, no specific order should be assumed.

Giving the root project the special name “Jamroot” ensures that Boost.Build won’t misinterpret a directory above it as the project root just because the directory contains a Jamfile.

3.7. The Build Process

When you’ve described your targets, you want Boost.Build to run the right tools and create the needed targets. This section will describe two things: how you specify what to build, and how the main targets are actually constructed.

The most important thing to note is that in Boost.Build, unlike other build tools, the targets you declare do not correspond to specific files. What you declare in a Jamfile is more like a “metatarget.” Depending on the properties you specify on the command line, each metatarget will produce a set of real targets corresponding to the requested properties. It is quite possible that the same metatarget is built several times with different properties, producing different files.

This means that for Boost.Build, you cannot directly obtain a build variant from a Jamfile. There could be several variants requested by the user, and each target can be built with different properties.

3.7.1. Build Request

The command line specifies which targets to build and with which properties. For example:

b2 app1 lib1//lib1 toolset=gcc variant=debug optimization=full

would build two targets, "app1" and "lib1//lib1" with the specified properties. You can refer to any targets, using target id and specify arbitrary properties. Some of the properties are very common, and for them the name of the property can be omitted. For example, the above can be written as:

b2 app1 lib1//lib1 gcc debug optimization=full

The complete syntax, which has some additional shortcuts, is described in the section called “Invocation”.

3.7.2. Building a main target

When you request, directly or indirectly, a build of a main target with specific requirements, the following steps are done. Some brief explanation is provided, and more details are given in the section called “Build process”.

  1. Applying default build. If the default-build property of a target specifies a value of a feature that is not present in the build request, that value is added.

  2. Selecting the main target alternative to use. For each alternative we look how many properties are present both in alternative’s requirements, and in build request. The alternative with largest number of matching properties is selected.

  3. Determining "common" properties. The build request is refined with target’s requirements. The conditional properties in requirements are handled as well. Finally, default values of features are added.

  4. Building targets referred by the sources list and dependency properties. The list of sources and the properties can refer to other target using target references. For each reference, we take all propagated properties, refine them by explicit properties specified in the target reference, and pass the resulting properties as build request to the other target.

  5. Adding the usage requirements produced when building dependencies to the "common" properties. When dependencies are built in the previous step, they return both the set of created "real" targets, and usage requirements. The usage requirements are added to the common properties and the resulting property set will be used for building the current target.

  6. Building the target using generators. To convert the sources to the desired type, Boost.Build uses "generators" — objects that correspond to tools like compilers and linkers. Each generator declares what type of targets it can produce and what type of sources it requires. Using this information, Boost.Build determines which generators must be run to produce a specific target from specific sources. When generators are run, they return the "real" targets.

  7. Computing the usage requirements to be returned. The conditional properties in usage requirements are expanded and the result is returned.

3.7.3. Building a Project

Often, a user builds a complete project, not just one main target. In fact, invoking b2 without arguments builds the project defined in the current directory.

When a project is built, the build request is passed without modification to all main targets in that project. It’s is possible to prevent implicit building of a target in a project with the explicit rule:

explicit hello_test ;

would cause the hello_test target to be built only if explicitly requested by the user or by some other target.

The Jamfile for a project can include a number of build-project rule calls that specify additional projects to be built.

4. Common tasks

This section describes main targets types that Boost.Build supports out-of-the-box. Unless otherwise noted, all mentioned main target rules have the common signature, described in the section called “Declaring Targets”.

4.1. Programs

Programs are created using the exe rule, which follows the common syntax. For example:

exe hello
    : hello.cpp some_library.lib /some_project//library
    : <threading>multi

This will create an executable file from the sources—​in this case, one C file, one library file present in the same directory, and another library that is created by Boost.Build. Generally, sources can include C and C files, object files and libraries. Boost.Build will automatically try to convert targets of other types.

On Windows, if an application uses shared libraries, and both the application and the libraries are built using Boost.Build, it is not possible to immediately run the application, because the PATH environment variable should include the path to the libraries. It means you have to either add the paths manually, or have the build place the application and the libraries into the same directory. See the section called “Installing”.

4.2. Libraries

Library targets are created using the lib rule, which follows the common syntax. For example:

lib helpers : helpers.cpp ;

This will define a library target named helpers built from the helpers.cpp source file. It can be either a static library or a shared library, depending on the value of the <link> feature.

Library targets can represent:

  • Libraries that should be built from source, as in the example above.

  • Prebuilt libraries which already exist on the system. Such libraries can be searched for by the tools using them (typically with the linker’s -l option) or their paths can be known in advance by the build system.

The syntax for prebuilt libraries is given below:

lib z : : <name>z <search>/home/ghost ;
lib compress : : <file>/opt/libs/compress.a ;

The name property specifies the name of the library without the standard prefixes and suffixes. For example, depending on the system, z could refer to a file called z.so, libz.a, or z.lib, etc. The search feature specifies paths in which to search for the library in addition to the default compiler paths. search can be specified several times or it can be omitted, in which case only the default compiler paths will be searched. The file property specifies the file location.

The difference between using the file feature and using a combination of the name and search features is that file is more precise.

The value of the search feature is just added to the linker search path. When linking to multiple libraries, the paths specified by search are combined without regard to which lib target each path came from. Thus, given

lib a : : <name>a <search>/pool/release ;
lib b : : <name>b <search>/pool/debug ;

If /pool/release/a.so, /pool/release/b.so, /pool/debug/a.so, and /pool/release/b.so all exist, the linker will probably take both a and b from the same directory, instead of finding a in /pool/release and b in /pool/debug. If you need to distinguish between multiple libraries with the same name, it’s safer to use file.

For convenience, the following syntax is allowed:

lib z ;
lib gui db aux ;

which has exactly the same effect as:

lib z : : <name>z ;
lib gui : : <name>gui ;
lib db : : <name>db ;
lib aux : : <name>aux ;

When a library references another library you should put that other library in its list of sources. This will do the right thing in all cases. For portability, you should specify library dependencies even for searched and prebuilt libraries, othewise, static linking on Unix will not work. For example:

lib z ;
lib png : z : <name>png ;

When a library has a shared library as a source, or a static library has another static library as a source then any target linking to the first library with automatically link to its source library as well.

On the other hand, when a shared library has a static library as a source then the first library will be built so that it completely includes the second one.

If you do not want a shared library to include all the libraries specified in its sources (especially statically linked ones), you would need to use the following:

lib b : a.cpp ;
lib a : a.cpp : <use>b : : <library>b ;

This specifies that library a uses library b, and causes all executables that link to a to link to b also. In this case, even for shared linking, the a library will not refer to b.

Usage requirements are often very useful for defining library targets. For example, imagine that you want you build a helpers library and its interface is described in its helpers.hpp header file located in the same directory as the helpers.cpp source file. Then you could add the following to the Jamfile located in that same directory:

lib helpers : helpers.cpp : : : <include>. ;

which would automatically add the directory where the target has been defined (and where the library’s header file is located) to the compiler’s include path for all targets using the helpers library. This feature greatly simplifies Jamfiles.

4.3. Alias

The alias rule gives an alternative name to a group of targets. For example, to give the name core to a group of three other targets with the following code:

alias core : im reader writer ;

Using core on the command line, or in the source list of any other target is the same as explicitly using im, reader, and writer.

Another use of the alias rule is to change build properties. For example, if you want to link statically to the Boost Threads library, you can write the following:

alias threads : /boost/thread//boost_thread : <link>static ;

and use only the threads alias in your Jamfiles.

You can also specify usage requirements for the alias target. If you write the following:

alias header_only_library : : : :  <include>/usr/include/header_only_library ;

then using header_only_library in sources will only add an include path. Also note that when an alias has sources, their usage requirements are propagated as well. For example:

lib library1 : library1.cpp : : : <include>/library/include1 ;
lib library2 : library2.cpp : : : <include>/library/include2 ;
alias static_libraries : library1 library2 : <link>static ;
exe main : main.cpp static_libraries ;

will compile main.cpp with additional includes required for using the specified static libraries.

4.4. Installing

This section describes various ways to install built targets and arbitrary files.

4.4.1. Basic install

For installing a built target you should use the install rule, which follows the common syntax. For example:

install dist : hello helpers ;

will cause the targets hello and helpers to be moved to the dist directory, relative to the Jamfile’s directory. The directory can be changed using the location property:

install dist : hello helpers : <location>/usr/bin ;

While you can achieve the same effect by changing the target name to /usr/bin, using the location property is better as it allows you to use a mnemonic target name.

The location property is especially handy when the location is not fixed, but depends on the build variant or environment variables:

install dist : hello helpers :
    <variant>debug:<location>dist/debug ;
install dist2 : hello helpers : <location>$(DIST) ;

4.4.2. Installing with all dependencies

Specifying the names of all libraries to install can be boring. The install allows you to specify only the top-level executable targets to install, and automatically install all dependencies:

install dist : hello :
    <install-dependencies>on <install-type>EXE

will find all targets that hello depends on, and install all of those which are either executables or libraries. More specifically, for each target, other targets that were specified as sources or as dependency properties, will be recursively found. One exception is that targets referred with the use feature are not considered, as that feature is typically used to refer to header-only libraries. If the set of target types is specified, only targets of that type will be installed, otherwise, all found target will be installed.

4.4.3. Preserving Directory Hierarchy

By default, the install rule will strip paths from its sources. So, if sources include a/b/c.hpp, the a/b part will be ignored. To make the install rule preserve the directory hierarchy you need to use the <install-source-root> feature to specify the root of the hierarchy you are installing. Relative paths from that root will be preserved. For example, if you write:

install headers
    : a/b/c.h
    : <location>/tmp <install-source-root>a

the a file named /tmp/b/c.h will be created.

The glob-tree rule can be used to find all files below a given directory, making it easy to install an entire directory tree.

4.4.4. Installing into Several Directories

The alias rule can be used when targets need to be installed into several directories:

alias install : install-bin install-lib ;
install install-bin : applications : /usr/bin ;
install install-lib : helper : /usr/lib ;

Because the install rule just copies targets, most free features [3] have no effect when used in requirements of the install rule. The only two that matter are dependency and, on Unix, dll-path.

(Unix specific) On Unix, executables built using Boost.Build typically contain the list of paths to all used shared libraries. For installing, this is not desired, so Boost.Build relinks the executable with an empty list of paths. You can also specify additional paths for installed executables using the dll-path feature.

4.5. Testing

Boost.Build has convenient support for running unit tests. The simplest way is the unit-test rule, which follows the common syntax. For example:

unit-test helpers_test : helpers_test.cpp helpers ;

The unit-test rule behaves like the exe rule, but after the executable is created it is also run. If the executable returns an error code, the build system will also return an error and will try running the executable on the next invocation until it runs successfully. This behaviour ensures that you can not miss a unit test failure.

There are few specialized testing rules, listed below:

rule compile ( sources : requirements * : target-name ? )
rule compile-fail ( sources : requirements * : target-name ? )
rule link ( sources + : requirements * : target-name ? )
rule link-fail ( sources + : requirements * : target-name ? )

They are given a list of sources and requirements. If the target name is not provided, the name of the first source file is used instead. The compile* tests try to compile the passed source. The link* rules try to compile and link an application from all the passed sources. The compile and link rules expect that compilation/linking succeeds. The compile-fail and link-fail rules expect that the compilation/linking fails.

There are two specialized rules for running executables, which are more powerful than the unit-test rule. The run rule has the following signature:

rule run ( sources + : args * : input-files * : requirements * : target-name ?
    : default-build * )

The rule builds application from the provided sources and runs it, passing args and input-files as command-line arguments. The args parameter is passed verbatim and the values of the input-files parameter are treated as paths relative to containing Jamfile, and are adjusted if b2 is invoked from a different directory. The run-fail rule is identical to the run rule, except that it expects that the run fails.

All rules described in this section, if executed successfully, create a special manifest file to indicate that the test passed. For the unit-test rule the files is named target-name.passed and for the other rules it is called target-name.test. The run* rules also capture all output from the program, and store it in a file named target-name.output.

If the preserve-test-targets feature has the value off, then run and the run-fail rules will remove the executable after running it. This somewhat decreases disk space requirements for continuous testing environments. The default value of preserve-test-targets feature is on.

It is possible to print the list of all test targets (except for unit-test) declared in your project, by passing the --dump-tests command-line option. The output will consist of lines of the form:

boost-test(test-type) path : sources

It is possible to process the list of tests, Boost.Build output and the presense/absense of the *.test files created when test passes into human-readable status table of tests. Such processing utilities are not included in Boost.Build.

The following features adjust behaviour of the testing metatargets.


Defines an argument to be passed to the target when it is executed before the list of input files.

unit-test helpers_test
    : helpers_test.cpp helpers
    : <testing.arg>"--foo bar"

Specifies a file to be passed to the executable on the command line after the arguments. All files must be specified in alphabetical order due to constrainsts in the current implementation.


By default, the executable is run directly. Sometimes, it is desirable to run the executable using some helper command. You should use this property to specify the name of the helper command. For example, if you write:

unit-test helpers_test
    : helpers_test.cpp helpers
    : <testing.launcher>valgrind

The command used to run the executable will be:

valgrind bin/$toolset/debug/helpers_test

A description of the test. This is displayed as part of the --dump-tests command-line option.

4.6. Custom commands

For most main target rules, Boost.Build automatically figures out the commands to run. When you want to use new file types or support new tools, one approach is to extend Boost.Build to support them smoothly, as documented in Extender Manual. However, if the new tool is only used in a single place, it might be easier just to specify the commands to run explicitly.

Three main target rules can be used for that. The make rule allows you to construct a single file from any number of source file, by running a command you specify. The notfile rule allows you to run an arbitrary command, without creating any files. And finaly, the generate rule allows you to describe a transformation using Boost.Build’s virtual targets. This is higher-level than the file names that the make rule operates with and allows you to create more than one target, create differently named targets depending on properties, or use more than one tool.

The make rule is used when you want to create one file from a number of sources using some specific command. The notfile is used to unconditionally run a command.

Suppose you want to create the file file.out from the file file.in by running the command in2out. Here is how you would do this in Boost.Build:

make file.out : file.in : @in2out ;
actions in2out
    in2out $(<) $(>)

If you run b2 and file.out does not exist, Boost.Build will run the in2out command to create that file. For more details on specifying actions, see the section called “Boost.Jam Language”.

It could be that you just want to run some command unconditionally, and that command does not create any specific files. For that you can use the notfile rule. For example:

notfile echo_something : @echo ;
actions echo
    echo "something"

The only difference from the make rule is that the name of the target is not considered a name of a file, so Boost.Build will unconditionally run the action.

The generate rule is used when you want to express transformations using Boost.Build’s virtual targets, as opposed to just filenames. The generate rule has the standard main target rule signature, but you are required to specify the generating-rule property. The value of the property should be in the form @rule-name, the named rule should have the following signature:

rule generating-rule ( project name : property-set : sources * )

and will be called with an instance of the project-target class, the name of the main target, an instance of the property-set class containing build properties, and the list of instances of the virtual-target class corresponding to sources. The rule must return a list of virtual-target instances. The interface of the virtual-target class can be learned by looking at the build/virtual-target.jam file. The generate example contained in the Boost.Build distribution illustrates how the generate rule can be used.

4.7. Precompiled Headers

Precompiled headers is a mechanism to speed up compilation by creating a partially processed version of some header files, and then using that version during compilations rather then repeatedly parsing the original headers. Boost.Build supports precompiled headers with gcc and msvc toolsets.

To use precompiled headers, follow the following steps:

  1. Create a header that includes headers used by your project that you want precompiled. It is better to include only headers that are sufficiently stable — like headers from the compiler and external libraries. Please wrap the header in #ifdef BOOST_BUILD_PCH_ENABLED, so that the potentially expensive inclusion of headers is not done when PCH is not enabled. Include the new header at the top of your source files.

  2. Declare a new Boost.Build target for the precompiled header and add that precompiled header to the sources of the target whose compilation you want to speed up:

    cpp-pch pch : pch.hpp ;
    exe main : main.cpp pch ;

    You can use the c-pch rule if you want to use the precompiled header in C programs.

The pch example in Boost.Build distribution can be used as reference.

Please note the following:

  • The inclusion of the precompiled header must be the first thing in a source file, before any code or preprocessor directives.

  • The build properties used to compile the source files and the precompiled header must be the same. Consider using project requirements to assure this.

  • Precompiled headers must be used purely as a way to improve compilation time, not to save the number of #include statements. If a source file needs to include some header, explicitly include it in the source file, even if the same header is included from the precompiled header. This makes sure that your project will build even if precompiled headers are not supported.

  • On the gcc compiler, the name of the header being precompiled must be equal to the name of the cpp-pch target. This is a gcc requirement.

  • Prior to version 4.2, the gcc compiler did not allow anonymous namespaces in precompiled headers, which limits their utility. See the bug report for details.

4.8. Generated headers

Usually, Boost.Build handles implicit dependendies completely automatically. For example, for C++ files, all #include statements are found and handled. The only aspect where user help might be needed is implicit dependency on generated files.

By default, Boost.Build handles such dependencies within one main target. For example, assume that main target "app" has two sources, "app.cpp" and "parser.y". The latter source is converted into "parser.c" and "parser.h". Then, if "app.cpp" includes "parser.h", Boost.Build will detect this dependency. Moreover, since "parser.h" will be generated into a build directory, the path to that directory will automatically be added to the include path.

Making this mechanism work across main target boundaries is possible, but imposes certain overhead. For that reason, if there is implicit dependency on files from other main targets, the <implicit-dependency> feature must be used, for example:

lib parser : parser.y ;
exe app : app.cpp : <implicit-dependency>parser ;

The above example tells the build system that when scanning all sources of "app" for implicit-dependencies, it should consider targets from "parser" as potential dependencies.

4.9. Cross-compilation

Boost.Build supports cross compilation with the gcc and msvc toolsets.

When using gcc, you first need to specify your cross compiler in user-config.jam (see the section called “Configuration”), for example:

using gcc : arm : arm-none-linux-gnueabi-g++ ;

After that, if the host and target os are the same, for example Linux, you can just request that this compiler version be used:

b2 toolset=gcc-arm

If you want to target a different operating system from the host, you need to additionally specify the value for the target-os feature, for example:

# On windows box
b2 toolset=gcc-arm target-os=linux
# On Linux box
b2 toolset=gcc-mingw target-os=windows

For the complete list of allowed opeating system names, please see the documentation for target-os feature.

When using the msvc compiler, it’s only possible to cross-compile to a 64-bit system on a 32-bit host. Please see the section called “64-bit support” for details.

4.10. Package Managers

Boost.Build support automatic, or manual, loading of generated build files from package managers. For example using the Conan package manager which generates conanbuildinfo.jam files B2 will load that files automatically when it loads the project at the same location. The included file can define targets and other project declarations in the context of the project it’s being loaded into. Control over what package manager file is loaded can be controlled with (in order of priority):

  • With the use-packages rule.

  • Command line argument --use-package-manager=X.

  • Environment variable PACKAGE_MANAGER_BUILD_INFO.

  • Built-in detection of the file. Currently this includes: "conan".

use-packages rule:

rule use-packages ( name-or-glob-pattern ? )

The use-packages rule allows one to specify in the projects themselves kind of package definitions to use either as the ones for a built-in package manager support. For example:

use-packages conan ;

Or to specify a glob pattern to find the file with the definitions. For instance:

use-packages "packages.jam" ;

--use-package-manager command line option:

The --use-package-manager=NAME command line option allows one to non-instrusively specify per invocation which of the built-in package manager types to use.


The PACKAGE_MANAGER_BUILD_INFO variable, which is taken from the environment or defined with the -sX=Y option, specifies a glob pattern to use to find the package definitions.

Built-in detection:

There are a number of built-in glob patterns to support popular package managers. Currently the supported ones are:

  • Conan (conan): currently supports the b2gen generator.

5. Reference

5.1. General information

5.1.1. Initialization

Immediately upon starting, the Boost.Build engine (b2) loads the Jam code that implements the build system. To do this, it searches for a file called boost-build.jam, first in the invocation directory, then in its parent and so forth up to the filesystem root, and finally in the directories specified by the environment variable BOOST_BUILD_PATH. When found, the file is interpreted, and should specify the build system location by calling the boost-build rule:

rule boost-build ( location ? )

If location is a relative path, it is treated as relative to the directory of boost-build.jam. The directory specified by that location and the directories in BOOST_BUILD_PATH are then searched for a file called bootstrap.jam, which is expected to bootstrap the build system. This arrangement allows the build system to work without any command-line or environment variable settings. For example, if the build system files were located in a directory "build-system/" at your project root, you might place a boost-build.jam at the project root containing:

boost-build build-system ;

In this case, running b2 anywhere in the project tree will automatically find the build system.

The default bootstrap.jam, after loading some standard definitions, loads both site-config.jam and user-config.jam.

5.2. Builtin rules

This section contains the list of all rules that can be used in Jamfile—both rules that define new targets and auxiliary rules.


Creates an executable file. See the section called “Programs”.


Creates an library file. See the section called “Libraries”.


Installs built targets and other files. See the section called “Installing”.


Creates an alias for other targets. See the section called “Alias”.


Creates an executable that will be automatically run. See the section called “Testing”.

compile; compile-fail; link; link-fail; run; run-fail

Specialized rules for testing. See the section called “Testing”.


The check-target-builds allows you to conditionally use different properties depending on whether some metatarget builds, or not. This is similar to functionality of configure script in autotools projects. The function signature is:

rule check-target-builds ( target message ? : true-properties * : false-properties * )

This function can only be used when passing requirements or usage requirements to a metatarget rule. For example, to make an application link to a library if it’s available, one has use the following:

exe app : app.cpp : [ check-target-builds has_foo "System has foo" : <library>foo : <define>FOO_MISSING=1 ] ;

For another example, the alias rule can be used to consolidate configuration choices and make them available to other metatargets, like so:

alias foobar : : : : [ check-target-builds has_foo "System has foo" : <library>foo : <library>bar ] ;

Creates an object file. Useful when a single source file must be compiled with special properties.


Creates an preprocessed source file. The arguments follow the common syntax.


The glob rule takes a list shell pattern and returns the list of files in the project’s source directory that match the pattern. For example:

lib tools : [ glob *.cpp ] ;

It is possible to also pass a second argument—the list of exclude patterns. The result will then include the list of files matching any of include patterns, and not matching any of the exclude patterns. For example:

lib tools : [ glob *.cpp : file_to_exclude.cpp bad*.cpp ] ;

The glob-tree is similar to the glob except that it operates recursively from the directory of the containing Jamfile. For example:

ECHO [ glob-tree *.cpp : .svn ] ;

will print the names of all C++ files in your project. The .svn exclude pattern prevents the glob-tree rule from entering administrative directories of the Subversion version control system.


Declares project id and attributes, including project requirements. See the section called “Projects”.


Assigns a symbolic project ID to a project at a given path. This rule must be better documented!


The explicit rule takes a single parameter—​a list of target names. The named targets will be marked explicit, and will be built only if they are explicitly requested on the command line, or if their dependents are built. Compare this to ordinary targets, that are built implicitly when their containing project is built.


The always function takes a single parameter—a list of metatarget names. The top-level targets produced by the named metatargets will be always considered out of date. Consider this example:

exe hello : hello.cpp ;
exe bye : bye.cpp ;
always hello ;

If a build of hello is requested, then the binary will always be relinked. The object files will not be recompiled, though. Note that if a build of hello is not requested, for example you specify just bye on the command line, hello will not be relinked.


Sets project-wide constant. Takes two parameters: variable name and a value and makes the specified variable name accessible in this Jamfile and any child Jamfiles. For example:

constant VERSION : 1.34.0 ;

Same as constant except that the value is treated as path relative to Jamfile location. For example, if b2 is invoked in the current directory, and Jamfile in helper subdirectory has:

path-constant DATA : data/a.txt ;

then the variable DATA will be set to helper/data/a.txt, and if b2 is invoked from the helper directory, then the variable DATA will be set to data/a.txt.


Cause some other project to be built. This rule takes a single parameter—a directory name relative to the containing Jamfile. When the containing Jamfile is built, the project located at that directory will be built as well. At the moment, the parameter to this rule should be a directory name. Project ID or general target references are not allowed.


This rule is deprecated and equivalent to alias.

5.3. Builtin features

This section documents the features that are built-in into Boost.Build. For features with a fixed set of values, that set is provided, with the default value listed first.


Allowed values: debug, release, profile.

A feature combining several low-level features, making it easy to request common build configurations.

The value debug expands to

<optimization>off <debug-symbols>on <inlining>off <runtime-debugging>on

The value release expands to

<optimization>speed <debug-symbols>off <inlining>full <runtime-debugging>off

The value profile expands to the same as release, plus:

<profiling>on <debug-symbols>on

Users can define their own build variants using the variant rule from the common module.

Runtime debugging is on in debug builds to suit the expectations of people used to various IDEs.

Allowed values: shared, static

A feature controlling how libraries are built.


Allowed values: shared, static

Controls if a static or shared C/C++ runtime should be used. There are some restrictions how this feature can be used, for example on some compilers an application using static runtime should not use shared libraries at all, and on some compilers, mixing static and shared runtime requires extreme care. Check your compiler documentation for more details.


Allowed values: single, multi

Controls if the project should be built in multi-threaded mode. This feature does not necessary change code generation in the compiler, but it causes the compiler to link to additional or different runtime libraries, and define additional preprocessor symbols (for example, _MT on Windows and _REENTRANT on Linux). How those symbols affect the compiled code depends on the code itself.


The <source>X feature has the same effect on building a target as putting X in the list of sources. It is useful when you want to add the same source to all targets in the project (you can put <source> in requirements) or to conditionally include a source (using conditional requirements, see the section called “Conditions and alternatives”). See also the <library> feature.


This feature is almost equivalent to the <source> feature, except that it takes effect only for linking. When you want to link all targets in a Jamfile to certain library, the <library> feature is preferred over <source>X--the latter will add the library to all targets, even those that have nothing to do with libraries.


Introduces a dependency on the target named by the value of this feature (so it will be brought up-to-date whenever the target being declared is). The dependency is not used in any other way.


Indicates that the target named by the value of this feature may produce files that are included by the sources of the target being declared. See the section called “Generated headers” for more information.


Introduces a dependency on the target named by the value of this feature (so it will be brought up-to-date whenever the target being declared is), and adds its usage requirements to the build properties of the target being declared. The dependency is not used in any other way. The primary use case is when you want the usage requirements (such as #include paths) of some library to be applied, but do not want to link to it.


Specify an additional directory where the system should look for shared libraries when the executable or shared library is run. This feature only affects Unix compilers. Please see the FAQ: “Why are the dll-path and hardcode-dll-paths properties useful? ” in Frequently Asked Questions for details.


Allowed values: true, false.

Controls automatic generation of dll-path properties.

This property is specific to Unix systems. If an executable is built with <hardcode-dll-paths>true, the generated binary will contain the list of all the paths to the used shared libraries. As the result, the executable can be run without changing system paths to shared libraries or installing the libraries to system paths. This is very convenient during development. Please see the FAQ entry for details. Note that on Mac OSX, the paths are unconditionally hardcoded by the linker, and it is not possible to disable that behaviour.

cflags; cxxflags; linkflags

The value of those features is passed without modification to the corresponding tools. For cflags that is both the C and C compilers, for `cxxflags` that is the C compiler, and for linkflags that is the linker. The features are handy when you are trying to do something special that cannot be achieved by a higher-level feature in Boost.Build.


Specifies an additional include path that is to be passed to C and C++ compilers.


Specifies an preprocessor symbol that should be defined on the command line. You may either specify just the symbol, which will be defined without any value, or both the symbol and the value, separated by equal sign.


The <warnings> feature controls the warning level of compilers. It has the following values:

  • off - disables all warnings.

  • on - enables default warning level for the tool.

  • all - enables all warnings.

    Default value is all.


The <warnings-as-errors> makes it possible to treat warnings as errors and abort compilation on a warning. The value on enables this behaviour. The default value is off.


Allowed values: no

The build feature is used to conditionally disable build of a target. If <build>no is in properties when building a target, build of that target is skipped. Combined with conditional requirements this allows you to skip building some target in configurations where the build is known to fail.


The tag feature is used to customize the name of the generated files. The value should have the form:


where rulename should be a name of a rule with the following signature:

rule tag ( name : type ? : property-set )

The rule will be called for each target with the default name computed by Boost.Build, the type of the target, and property set. The rule can either return a string that must be used as the name of the target, or an empty string, in which case the default name will be used.

Most typical use of the tag feature is to encode build properties, or library version in library target names. You should take care to return non-empty string from the tag rule only for types you care about — otherwise, you might end up modifying names of object files, generated header file and other targets for which changing names does not make sense.


Allowed values: on, off.

The debug-symbols feature specifies if produced object files, executables, and libraries should include debug information. Typically, the value of this feature is implicitly set by the variant feature, but it can be explicitly specified by the user. The most common usage is to build release variant with debugging information.


Allowed values: on, off.

The runtime-debugging feature specifies whether produced object files, executables, and libraries should include behaviour useful only for debugging, such as asserts. Typically, the value of this feature is implicitly set by the variant feature, but it can be explicitly specified by the user. The most common usage is to build release variant with debugging output.


The operating system for which the code is to be generated. The compiler you used should be the compiler for that operating system. This option causes Boost.Build to use naming conventions suitable for that operating system, and adjust build process accordingly. For example, with gcc, it controls if import libraries are produced for shared libraries or not.

The complete list of possible values for this feature is: aix, appletv, bsd, cygwin, darwin, freebsd, hpux, iphone, linux, netbsd, openbsd, osf, qnx, qnxnto, sgi, solaris, unix, unixware, windows.

See the section called “Cross-compilation” for details of crosscompilation.


Allowed values: x86, ia64, sparc, power, mips1, mips2, mips3, mips4, mips32, mips32r2, mips64, parisc, arm, combined, combined-x86-power.

The architecture features specifies the general processor family to generate code for.


Allowed values: depends on the used toolset.

The instruction-set specifies for which specific instruction set the code should be generated. The code in general might not run on processors with older/different instruction sets.

While Boost.Build allows a large set of possible values for this features, whether a given value works depends on which compiler you use. Please see the section called “C++ Compilers” for details.


Allowed values: 32, 64.

The address-model specifies if 32-bit or 64-bit code should be generated by the compiler. Whether this feature works depends on the used compiler, its version, how the compiler is configured, and the values of the architecture instruction-set features. Please see the section called “C++ Compilers” for details.


Allowed values: Any positive integer.

This feature allows configuring a C++ compiler with the maximal template instantiation depth parameter. Specific toolsets may or may not provide support for this feature depending on whether their compilers provide a corresponding command-line option.

Due to some internal details in the current Boost.Build implementation it is not possible to have features whose valid values are all positive integer. As a workaround a large set of allowed values has been defined for this feature and, if a different one is needed, user can easily add it by calling the feature.extend rule.

Allowed values: on, off.

This feature is specific to the msvc toolset (see the section called “Microsoft Visual C++”), and controls whether the manifest files should be embedded inside executables and shared libraries, or placed alongside them. This feature corresponds to the IDE option found in the project settings dialog, under Configuration Properties → Manifest Tool → Input and Output → Embed manifest.


This feature is specific to the msvc toolset (see the section called “Microsoft Visual C++”), and controls which manifest files should be embedded inside executables and shared libraries. This feature corresponds to the IDE option found in the project settings dialog, under Configuration Properties → Manifest Tool → Input and Output → Additional Manifest Files.


Allowed values: the name of any feature.

This feature is used to indicate which other features are relevant for a given target. It is usually not necessary to manage it explicitly, as Boost.Build can deduce it in most cases. Features which are not relevant will not affect target paths, and will not cause conflicts.

  • A feature will be considered relevant if any of the following are true

    • It is referenced by toolset.flags or toolset.uses-features

    • It is used by the requirements of a generator

    • It is a subfeature of a relevant feature

    • It has a subfeature which is relevant

    • It is a composite feature, and any composed feature is relevant

    • It affects target alternative selection for a main target

    • It is a propagated feature and is relevant for any dependency

    • It is relevant for any dependency created by the same main target

    • It is used in the condition of a conditional property and the corresponding value is relevant

    • It is explicitly named as relevent

  • Relevant features cannot be automatically deduced in the following cases:

    • Indirect conditionals. Solution: return properties of the form <relevant>result-feature:<relevant>condition-feature

      This isn’t really a conditional, although for most purposes it functions like one. In particular, it does not support multiple comma-separated elements in the condition, and it does work correctly even in contexts where conditional properties are not allowed
    • Action rules that read properties. Solution: add toolset.uses-features to tell Boost.Build that the feature is actually used.

    • Generators and targets that manipulate property-sets directly. Solution: set <relevant> manually.

5.4. Builtin tools

Boost.Build comes with support for a large number of C++ compilers, and other tools. This section documents how to use those tools.

Before using any tool, you must declare your intention, and possibly specify additional information about the tool’s configuration. This is done by calling the using rule, typically in your user-config.jam, for example:

using gcc ;

additional parameters can be passed just like for other rules, for example:

using gcc : 4.0 : g++-4.0 ;

The options that can be passed to each tool are documented in the subsequent sections.

5.4.1. C++ Compilers

This section lists all Boost.Build modules that support C++ compilers and documents how each one can be initialized. The name of support module for compiler is also the value for the toolset feature that can be used to explicitly request that compiler.

HP aC++ compiler

The acc module supports the HP aC++ compiler for the HP-UX operating system.

The module is initialized using the following syntax:

using acc : [version] : [c++-compile-command] : [compiler options] ;

This statement may be repeated several times, if you want to configure several versions of the compiler.

If the command is not specified, the aCC binary will be searched in PATH.

The following options can be provided, using `<option-name>option-value syntax`:


Specifies additional compiler flags that will be used when compiling C sources.


Specifies additional compiler flags that will be used when compiling C++ sources.


Specifies additional compiler flags that will be used when compiling both C and C++ sources.


Specifies additional command line options that will be passed to the linker.

Borland C++ Compiler

The borland module supports the command line C compiler included in http://www.borland.com/us/products/cbuilder/index.html[C Builder 2006] product and earlier version of it, running on Microsoft Windows.

The supported products are listed below. The version reported by the command lines tools is also listed for reference.:

  • C++ Builder 2006 — 5.8.2

  • CBuilderX — 5.6.5, 5.6.4 (depending on release)

  • CBuilder6 — 5.6.4

  • Free command line tools — 5.5.1

The module is initialized using the following syntax:

using borland : [version] : [c++-compile-command] : [compiler options] ;

This statement may be repeated several times, if you want to configure several versions of the compiler.

If the command is not specified, Boost.Build will search for a binary named bcc32 in PATH.

The following options can be provided, using `<option-name>option-value syntax`:


Specifies additional compiler flags that will be used when compiling C sources.


Specifies additional compiler flags that will be used when compiling C++ sources.


Specifies additional compiler flags that will be used when compiling both C and C++ sources.


Specifies additional command line options that will be passed to the linker.

Comeau C/C++ Compiler

The como-linux and the como-win modules supports the Comeau C/C++ Compiler on Linux and Windows respectively.

The module is initialized using the following syntax:

using como : [version] : [c++-compile-command] : [compiler options] ;

This statement may be repeated several times, if you want to configure several versions of the compiler.

If the command is not specified, Boost.Build will search for a binary named como in PATH.

The following options can be provided, using `<option-name>option-value syntax`:


Specifies additional compiler flags that will be used when compiling C sources.


Specifies additional compiler flags that will be used when compiling C++ sources.


Specifies additional compiler flags that will be used when compiling both C and C++ sources.


Specifies additional command line options that will be passed to the linker.

Before using the Windows version of the compiler, you need to setup necessary environment variables per compiler’s documentation. In particular, the COMO_XXX_INCLUDE variable should be set, where XXX corresponds to the used backend C compiler.

Code Warrior

The cw module support CodeWarrior compiler, originally produced by Metrowerks and presently developed by Freescale. Boost.Build supports only the versions of the compiler that target x86 processors. All such versions were released by Metrowerks before acquisition and are not sold any longer. The last version known to work is 9.4.

The module is initialized using the following syntax:

using cw : [version] : [c++-compile-command] : [compiler options] ;

This statement may be repeated several times, if you want to configure several versions of the compiler.

If the command is not specified, Boost.Build will search for a binary named mwcc in default installation paths and in PATH.

The following options can be provided, using `<option-name>option-value syntax`:


Specifies additional compiler flags that will be used when compiling C sources.


Specifies additional compiler flags that will be used when compiling C++ sources.


Specifies additional compiler flags that will be used when compiling both C and C++ sources.


Specifies additional command line options that will be passed to the linker.


The command that sets up environment variables prior to invoking the compiler. If not specified, cwenv.bat alongside the compiler binary will be used.


The command that compiles C and C++ sources. If not specified, mwcc will be used. The command will be invoked after the setup script was executed and adjusted the PATH variable.


The command that links executables and dynamic libraries. If not specified, mwld will be used. The command will be invoked after the setup script was executed and adjusted the PATH variable.

Digital Mars C/C++ Compiler

The dmc module supports the Digital Mars C++ compiler.

The module is initialized using the following syntax:

using dmc : [version] : [c++-compile-command] : [compiler options] ;

This statement may be repeated several times, if you want to configure several versions of the compiler.

If the command is not specified, Boost.Build will search for a binary named dmc in PATH.

The following options can be provided, using `<option-name>option-value syntax`:


Specifies additional compiler flags that will be used when compiling C sources.


Specifies additional compiler flags that will be used when compiling C++ sources.


Specifies additional compiler flags that will be used when compiling both C and C++ sources.


Specifies additional command line options that will be passed to the linker.


The gcc module supports the GNU C++ compiler on Linux, a number of Unix-like system including SunOS and on Windows (either Cygwin or MinGW).

The gcc module is initialized using the following syntax:

using gcc : [version] : [c++-compile-command] : [compiler options] ;

This statement may be repeated several times, if you want to configure several versions of the compiler.

If the version is not explicitly specified, it will be automatically detected by running the compiler with the -v option. If the command is not specified, the g++ binary will be searched in PATH.

The following options can be provided, using `<option-name>option-value syntax`:


Specifies additional compiler flags that will be used when compiling C sources.


Specifies additional compiler flags that will be used when compiling C++ sources.


Specifies additional compiler flags that will be used when compiling both C and C++ sources.


Specifies additional command line options that will be passed to the linker.


Specifies root directory of the compiler installation. This option is necessary only if it is not possible to detect this information from the compiler command—​for example if the specified compiler command is a user script.


Specifies the archiver command that is used to produce static libraries. Normally, it is autodetected using gcc -print-prog-name option or defaulted to ar, but in some cases you might want to override it, for example to explicitly use a system version instead of one included with gcc.


Specifies the ranlib command that is used to generated symbol table for static libraries. Normally, it is autodetected using gcc -print-prog-name option or defaulted to ranlib, but in some cases you might want to override it, for example to explicitly use a system version instead of one included with gcc.


Specifies the resource compiler command that will be used with the version of gcc that is being configured. This setting makes sense only for Windows and only if you plan to use resource files. By default windres will be used.


Specifies the type of resource compiler. The value can be either windres for msvc resource compiler, or rc for borland’s resource compiler.

In order to compile 64-bit applications, you have to specify address-model=64, and the instruction-set feature should refer to a 64 bit processor. Currently, those include nocona, opteron, athlon64 and athlon-fx.

HP C++ Compiler for Tru64 Unix

The hp_cxx modules supports the HP C++ Compiler for Tru64 Unix.

The module is initialized using the following syntax:

using hp_cxx : [version] : [c++-compile-command] : [compiler options] ;

This statement may be repeated several times, if you want to configure several versions of the compiler.

If the command is not specified, Boost.Build will search for a binary named hp_cxx in PATH.

The following options can be provided, using `<option-name>option-value syntax`:


Specifies additional compiler flags that will be used when compiling C sources.


Specifies additional compiler flags that will be used when compiling C++ sources.


Specifies additional compiler flags that will be used when compiling both C and C++ sources.


Specifies additional command line options that will be passed to the linker.

Intel C++

The intel-* modules support the Intel C++ command-line compiler.

The module is initialized using the following syntax:

using intel : [version] : [c++-compile-command] : [compiler options] ;

This statement may be repeated several times, if you want to configure several versions of the compiler.

If compiler command is not specified, then Boost.Build will look in PATH for an executable icpc (on Linux), or icc.exe (on Windows).

The following options can be provided, using `<option-name>option-value syntax`:


Specifies additional compiler flags that will be used when compiling C sources.


Specifies additional compiler flags that will be used when compiling C++ sources.


Specifies additional compiler flags that will be used when compiling both C and C++ sources.


Specifies additional command line options that will be passed to the linker.

The Linux version supports the following additional options:


Specifies root directory of the compiler installation. This option is necessary only if it is not possible to detect this information from the compiler command — for example if the specified compiler command is a user script.

Microsoft Visual C++

The msvc module supports the Microsoft Visual C++ command-line tools on Microsoft Windows. The supported products and versions of command line tools are listed below:

  • Visual Studio 2017—14.1

  • Visual Studio 2015—14.0

  • Visual Studio 2013—12.0

  • Visual Studio 2012—11.0

  • Visual Studio 2010—10.0

  • Visual Studio 2008—9.0

  • Visual Studio 2005—8.0

  • Visual Studio .NET 2003—7.1

  • Visual Studio .NET—7.0

  • Visual Studio 6.0, Service Pack 5—​6.5

The user would then call the boost build executable with the toolset set equal to msvc-[version number] for example to build with Visual Studio 2017 one could run:

.\b2 toolset=msvc-14.1 target

The msvc module is initialized using the following syntax:

using msvc : [version] : [c++-compile-command] : [compiler options] ;

This statement may be repeated several times, if you want to configure several versions of the compiler.

If the version is not explicitly specified, the most recent version found in the registry will be used instead. If the special value all is passed as the version, all versions found in the registry will be configured. If a version is specified, but the command is not, the compiler binary will be searched in standard installation paths for that version, followed by PATH.

The compiler command should be specified using forward slashes, and quoted.

The following options can be provided, using `<option-name>option-value syntax`:


Specifies additional compiler flags that will be used when compiling C sources.


Specifies additional compiler flags that will be used when compiling C++ sources.


Specifies additional compiler flags that will be used when compiling both C and C++ sources.


Specifies additional command line options that will be passed to the linker.


The command that compiles assembler sources. If not specified, ml will be used. The command will be invoked after the setup script was executed and adjusted the PATH variable.


The command that compiles C and C++ sources. If not specified, cl will be used. The command will be invoked after the setup script was executed and adjusted the PATH variable.


Command through which to pipe the output of running the compiler. For example to pass the output to STLfilt.


The command that compiles Microsoft COM interface definition files. If not specified, midl will be used. The command will be invoked after the setup script was executed and adjusted the PATH variable.


The command that links executables and dynamic libraries. If not specified, link will be used. The command will be invoked after the setup script was executed and adjusted the PATH variable.


The command that compiles Microsoft message catalog files. If not specified, mc will be used. The command will be invoked after the setup script was executed and adjusted the PATH variable.


The command that compiles resource files. If not specified, rc will be used. The command will be invoked after the setup script was executed and adjusted the PATH variable.


The filename of the global environment setup script to run before invoking any of the tools defined in this toolset. Will not be used in case a target platform specific script has been explicitly specified for the current target platform. Used setup script will be passed the target platform identifier (x86, x86_amd64, x86_ia64, amd64 or ia64) as a parameter. If not specified a default script is chosen based on the used compiler binary, e.g. vcvars32.bat or vsvars32.bat.

setup-amd64; setup-i386; setup-ia64

The filename of the target platform specific environment setup script to run before invoking any of the tools defined in this toolset. If not specified the global environment setup script is used.

64-bit support

Starting with version 8.0, Microsoft Visual Studio can generate binaries for 64-bit processor, both 64-bit flavours of x86 (codenamed AMD64/EM64T), and Itanium (codenamed IA64). In addition, compilers that are itself run in 64-bit mode, for better performance, are provided. The complete list of compiler configurations are as follows (we abbreviate AMD64/EM64T to just AMD64):

  • 32-bit x86 host, 32-bit x86 target

  • 32-bit x86 host, 64-bit AMD64 target

  • 32-bit x86 host, 64-bit IA64 target

  • 64-bit AMD64 host, 64-bit AMD64 target

  • 64-bit IA64 host, 64-bit IA64 target

The 32-bit host compilers can be always used, even on 64-bit Windows. On the contrary, 64-bit host compilers require both 64-bit host processor and 64-bit Windows, but can be faster. By default, only 32-bit host, 32-bit target compiler is installed, and additional compilers need to be installed explicitly.

To use 64-bit compilation you should:

  1. Configure you compiler as usual. If you provide a path to the compiler explicitly, provide the path to the 32-bit compiler. If you try to specify the path to any of 64-bit compilers, configuration will not work.

  2. When compiling, use address-model=64, to generate AMD64 code.

  3. To generate IA64 code, use architecture=ia64

The (AMD64 host, AMD64 target) compiler will be used automatically when you are generating AMD64 code and are running 64-bit Windows on AMD64. The (IA64 host, IA64 target) compiler will never be used, since nobody has an IA64 machine to test.

It is believed that AMD64 and EM64T targets are essentially compatible. The compiler options /favor:AMD64 and /favor:EM64T, which are accepted only by AMD64 targeting compilers, cause the generated code to be tuned to a specific flavor of 64-bit x86. Boost.Build will make use of those options depending on the value of the`instruction-set` feature.

Windows Runtime support

Starting with version 11.0, Microsoft Visual Studio can produce binaries for Windows Store and Phone in addition to traditional Win32 desktop. To specify which Windows API set to target, use the windows-api feature. Available options are desktop, store, or phone. If not specified, desktop will be used.

When using store or phone the specified toolset determines what Windows version is targeted. The following options are available:

  • Windows 8.0: toolset=msvc-11.0 windows-api=store

  • Windows 8.1: toolset=msvc-12.0 windows-api=store

  • Windows Phone 8.0: toolset=msvc-11.0 windows-api=phone

  • Windows Phone 8.1: toolset=msvc-12.0 windows-api=phone

For example use the following to build for Windows Store 8.1 with the ARM architecture:

.\b2 toolset=msvc-12.0 windows-api=store architecture=arm

Note that when targeting Windows Phone 8.1, version 12.0 didn’t include the vcvars phone setup scripts. They can be separately downloaded from here.

Sun Studio

The sun module supports the Sun Studio C++ compilers for the Solaris OS.

The module is initialized using the following syntax:

using sun : [version] : [c++-compile-command] : [compiler options] ;

This statement may be repeated several times, if you want to configure several versions of the compiler.

If the command is not specified, Boost.Build will search for a binary named CC in /opt/SUNWspro/bin and in PATH.

When using this compiler on complex C code, such as the http://boost.org[Boost C library], it is recommended to specify the following options when initializing the sun module:

-library=stlport4 -features=tmplife -features=tmplrefstatic

See the Sun C++ Frontend Tales for details.

The following options can be provided, using `<option-name>option-value syntax`:


Specifies additional compiler flags that will be used when compiling C sources.


Specifies additional compiler flags that will be used when compiling C++ sources.


Specifies additional compiler flags that will be used when compiling both C and C++ sources.


Specifies additional command line options that will be passed to the linker.

Starting with Sun Studio 12, you can create 64-bit applications by using the address-model=64 property.

IBM Visual Age

The vacpp module supports the IBM Visual Age C++ Compiler, for the AIX operating system. Versions 7.1 and 8.0 are known to work.

The module is initialized using the following syntax:

using vacpp ;

The module does not accept any initialization options. The compiler should be installed in the /usr/vacpp/bin directory.

Later versions of Visual Age are known as XL C/C++. They were not tested with the the vacpp module.

5.4.2. Third-party libraries

Boost.Build provides special support for some third-party C++ libraries, documented below.

STLport library

The STLport library is an alternative implementation of C++ runtime library. Boost.Build supports using that library on Windows platform. Linux is hampered by different naming of libraries in each STLport version and is not officially supported.

Before using STLport, you need to configure it in user-config.jam using the following syntax:

using stlport : version : header-path : library-path ;

Where version is the version of STLport, for example 5.1.4, headers is the location where STLport headers can be found, and libraries is the location where STLport libraries can be found. The version should always be provided, and the library path should be provided if you’re using STLport’s implementation of iostreams. Note that STLport 5.* always uses its own iostream implementation, so the library path is required.

When STLport is configured, you can build with STLport by requesting stdlib=stlport on the command line.


Provides support for the zlib library. zlib can be configured either to use precompiled binaries or to build the library from source.

zlib can be initialized using the following syntax

using zlib : version : options : condition : is-default ;

Options for using a prebuilt library:


The directory containing the zlib binaries.


Overrides the default library name.


The directory containing the zlib headers.

If none of these options is specified, then the environmental variables ZLIB_LIBRARY_PATH, ZLIB_NAME, and ZLIB_INCLUDE will be used instead.

Options for building zlib from source:


The zlib source directory. Defaults to the environmental variable ZLIB_SOURCE.


Sets the tag property to adjust the file name of the library. Ignored when using precompiled binaries.


The base name to use for the compiled library. Ignored when using precompiled binaries.


# Find zlib in the default system location
using zlib ;
# Build zlib from source
using zlib : 1.2.7 : <source>/home/steven/zlib-1.2.7 ;
# Find zlib in /usr/local
using zlib : 1.2.7 : <include>/usr/local/include <search>/usr/local/lib ;
# Build zlib from source for msvc and find
# prebuilt binaries for gcc.
using zlib : 1.2.7 : <source>C:/Devel/src/zlib-1.2.7 : <toolset>msvc ;
using zlib : 1.2.7 : : <toolset>gcc ;

Provides support for the bzip2 library. bzip2 can be configured either to use precompiled binaries or to build the library from source.

bzip2 can be initialized using the following syntax

using bzip2 : version : options : condition : is-default ;

Options for using a prebuilt library:


The directory containing the bzip2 binaries.


Overrides the default library name.


The directory containing the bzip2 headers.

If none of these options is specified, then the environmental variables BZIP2_LIBRARY_PATH, BZIP2_NAME, and BZIP2_INCLUDE will be used instead.

Options for building bzip2 from source:


The bzip2 source directory. Defaults to the environmental variable BZIP2_SOURCE.


Sets the tag property to adjust the file name of the library. Ignored when using precompiled binaries.


The base name to use for the compiled library. Ignored when using precompiled binaries.


# Find bzip in the default system location
using bzip2 ;
# Build bzip from source
using bzip2 : 1.0.6 : <source>/home/sergey/src/bzip2-1.0.6 ;
# Find bzip in /usr/local
using bzip2 : 1.0.6 : <include>/usr/local/include <search>/usr/local/lib ;
# Build bzip from source for msvc and find
# prebuilt binaries for gcc.
using bzip2 : 1.0.6 : <source>C:/Devel/src/bzip2-1.0.6 : <toolset>msvc ;
using bzip2 : 1.0.6 : : <toolset>gcc ;

Provides support for the python language environment to be linked in as a library.

python can be initialized using the following syntax

using python : [version] : [command-or-prefix] : [includes] : [libraries] : [conditions] : [extension-suffix] ;

Options for using python:


The version of Python to use. Should be in Major.Minor format, for example 2.3. Do not include the subminor version.


Preferably, a command that invokes a Python interpreter. Alternatively, the installation prefix for Python libraries and includes. If empty, will be guessed from the version, the platform’s installation patterns, and the python executables that can be found in PATH.


the include path to Python headers. If empty, will be guessed.


the path to Python library binaries. If empty, will be guessed. On MacOS/Darwin, you can also pass the path of the Python framework.


if specified, should be a set of properties that are matched against the build configuration when Boost.Build selects a Python configuration to use.


A string to append to the name of extension modules before the true filename extension. Ordinarily we would just compute this based on the value of the <python-debugging> feature. However ubuntu’s python-dbg package uses the windows convention of appending _d to debug-build extension modules. We have no way of detecting ubuntu, or of probing python for the "_d" requirement, and if you configure and build python using --with-pydebug, you’ll be using the standard *nix convention. Defaults to "" (or "_d" when targeting windows and <python-debugging> is set).


# Find python in the default system location
using python ;
# 2.7
using python : 2.7 ;
# 3.5
using python : 3.5 ;

# On ubuntu 16.04
using python
: 2.7 # version
: # Interpreter/path to dir
: /usr/include/python2.7 # includes
: /usr/lib/x86_64-linux-gnu # libs
: # conditions

using python
: 3.5 # version
: # Interpreter/path to dir
: /usr/include/python3.5 # includes
: /usr/lib/x86_64-linux-gnu # libs
: # conditions

# On windows
using python
: 2.7 # version
: C:\\Python27-32\\python.exe # Interperter/path to dir
: C:\\Python27-32\\include # includes
: C:\\Python27-32\\libs # libs
: <address-model>32 <address-model> # conditions - both 32 and unspecified

using python
: 2.7 # version
: C:\\Python27-64\\python.exe # Interperter/path to dir
: C:\\Python27-64\\include # includes
: C:\\Python27-64\\libs # libs
: <address-model>64 # conditions

5.4.3. Documentation tools

Boost.Build support for the Boost documentation tools is documented below.


To use xsltproc, you first need to configure it using the following syntax:

using xsltproc : xsltproc ;

Where xsltproc is the xsltproc executable. If xsltproc is not specified, and the variable XSLTPROC is set, the value of XSLTPROC will be used. Otherwise, xsltproc will be searched for in PATH.

The following options can be provided, using `<option-name>option-value syntax`:


Values should have the form name=value


Sets an additional search path for xi:include elements.


A catalog file used to rewrite remote URL’s to a local copy.

The xsltproc module provides the following rules. Note that these operate on jam targets and are intended to be used by another toolset, such as boostbook, rather than directly by users.

rule xslt ( target : source stylesheet : properties * )

Runs xsltproc to create a single output file.

rule xslt-dir ( target : source stylesheet : properties * : dirname )

Runs xsltproc to create multiple outputs in a directory. dirname is unused, but exists for historical reasons. The output directory is determined from the target.


To use boostbook, you first need to configure it using the following syntax:

using boostbook : docbook-xsl-dir : docbook-dtd-dir : boostbook-dir ;

docbook-xsl-dir is the DocBook XSL stylesheet directory. If not provided, we use DOCBOOK_XSL_DIR from the environment (if available) or look in standard locations. Otherwise, we let the XML processor load the stylesheets remotely.

docbook-dtd-dir is the DocBook DTD directory. If not provided, we use DOCBOOK_DTD_DIR From the environment (if available) or look in standard locations. Otherwise, we let the XML processor load the DTD remotely.

boostbook-dir is the BoostBook directory with the DTD and XSL subdirs.

The boostbook module depends on xsltproc. For pdf or ps output, it also depends on fop.

The following options can be provided, using `<option-name>option-value syntax`:


Allowed values: html, xhtml, htmlhelp, onehtml, man, pdf, ps, docbook, fo, tests.

The format feature determines the type of output produced by the boostbook rule.

The boostbook module defines a rule for creating a target following the common syntax.

rule boostbook ( target-name : sources * : requirements * : default-build * )

Creates a boostbook target.


To use doxygen, you first need to configure it using the following syntax:

using doxygen : name ;

name is the doxygen command. If it is not specified, it will be found in the PATH.

The doxygen module depends on the boostbook module when generating BoostBook XML.

The following options can be provided, using `<option-name>option-value syntax`:


All the values of doxygen:param are added to the doxyfile.


Specifies the common prefix of all headers when generating BoostBook XML. Everything before this will be stripped off.


Specifies the title of the library-reference section, when generating BoostBook XML.


When generating BoostBook XML, specifies the directory in which to place the images generated from LaTex formulae.

The path is interpreted relative to the current working directory, not relative to the Jamfile. This is necessary to match the behavior of BoostBook.

The doxygen module defines a rule for creating a target following the common syntax.

rule doxygen ( target : sources * : requirements * : default-build * : usage-requirements * )

Creates a doxygen target. If the target name ends with .html, then this will generate an html directory. Otherwise it will generate BoostBook XML.


The quickbook module provides a generator to convert from Quickbook to BoostBook XML.

To use quickbook, you first need to configure it using the following syntax:

using quickbook : command ;

command is the quickbook executable. If it is not specified, Boost.Build will compile it from source. If it is unable to find the source it will search for a quickbook executable in PATH.


The fop module provides generators to convert from XSL formatting objects to Postscript and PDF.

To use fop, you first need to configure it using the following syntax:

using fop : fop-command : java-home : java ;

fop-command is the command to run fop. If it is not specified, Boost.Build will search for it in PATH and FOP_HOME.

Either java-home or java can be used to specify where to find java.

5.5. Builtin modules

This section describes the modules that are provided by Boost.Build. The import rule allows rules from one module to be used in another module or Jamfile.

5.5.1. modules

The modules module defines basic functionality for handling modules.

A module defines a number of rules that can be used in other modules. Modules can contain code at the top level to initialize the module. This code is executed the first time the module is loaded.

A Jamfile is a special kind of module which is managed by the build system. Although they cannot be loaded directly by users, the other features of modules are still useful for Jamfiles.

Each module has its own namespaces for variables and rules. If two modules A and B both use a variable named X, each one gets its own copy of X. They won’t interfere with each other in any way. Similarly, importing rules into one module has no effect on any other module.

Every module has two special variables. $(file) contains the name of the file that the module was loaded from and $(name) contains the name of the module.

$(file) does not contain the full path to the file. If you need this, use modules.binding.
  1. rule binding ( module-name )

    Returns the filesystem binding of the given module.

    For example, a module can get its own location with:

    me = [ modules.binding $(__name__) ] ;
  2. rule poke ( module-name ? : variables + : value * )

    Sets the module-local value of a variable.

    For example, to set a variable in the global module:

    modules.poke : ZLIB_INCLUDE : /usr/local/include ;
  3. rule peek ( module-name ? : variables + )

    Returns the module-local value of a variable.

    For example, to read a variable from the global module:

    local ZLIB_INCLUDE = [ modules.peek : ZLIB_INCLUDE ] ;
  4. rule call-in ( module-name ? : rule-name args * : * )

    Call the given rule locally in the given module. Use this for rules accepting rule names as arguments, so that the passed rule may be invoked in the context of the rule’s caller (for example, if the rule accesses module globals or is a local rule).

    rules called this way may accept at most 8 parameters.


    rule filter ( f : values * )
        local m = [ CALLER_MODULE ] ;
        local result ;
        for v in $(values)
            if [ modules.call-in $(m) : $(f) $(v) ]
                result += $(v) ;
        return result ;
  5. rule load ( module-name : filename ? : search * )

    Load the indicated module if it is not already loaded.


    Name of module to load.


    (partial) path to file; Defaults to $(module-name).jam


    Directories in which to search for filename. Defaults to $(BOOST_BUILD_PATH).

  6. rule import ( module-names + : rules-opt * : rename-opt * )

    Load the indicated module and import rule names into the current module. Any members of rules-opt will be available without qualification in the caller’s module. Any members of rename-opt will be taken as the names of the rules in the caller’s module, in place of the names they have in the imported module. If rules-opt = '*', all rules from the indicated module are imported into the caller’s module. If rename-opt is supplied, it must have the same number of elements as rules-opt.

    The import rule is available without qualification in all modules.


    import path ;
    import path : * ;
    import path : join ;
    import path : native make : native-path make-path ;
  7. rule clone-rules ( source-module target-module )

    Define exported copies in $(target-module) of all rules exported from $(source-module). Also make them available in the global module with qualification, so that it is just as though the rules were defined originally in $(target-module).

5.5.2. path

Performs various path manipulations. Paths are always in a 'normalized' representation. In it, a path may be either:

  • '.', or

  • ['/'] [ ( '..' '/' )* (token '/')* token ]

In plain english, a path can be rooted, '..' elements are allowed only at the beginning, and it never ends in slash, except for the path consisting of slash only.

  1. rule make ( native )

    Converts the native path into normalized form.

  2. rule native ( path )

    Builds the native representation of the path.

  3. rule is-rooted ( path )

    Tests if a path is rooted.

  4. rule has-parent ( path )

    Tests if a path has a parent.

  5. rule basename ( path )

    Returns the path without any directory components.

  6. rule parent ( path )

    Returns the parent directory of the path. If no parent exists, an error is issued.

  7. rule reverse ( path )

    Returns path2 such that [ join path path2 ] = ".". The path may not contain ".." element or be rooted.

  8. rule join ( elements + )

    Concatenates the passed path elements. Generates an error if any element other than the first one is rooted. Skips any empty or undefined path elements.

  9. rule root ( path root )

    If path is relative, it is rooted at root. Otherwise, it is unchanged.

  10. rule pwd ( )

    Returns the current working directory.

  11. rule glob ( dirs * : patterns + : exclude-patterns * )

    Returns the list of files matching the given pattern in the specified directory. Both directories and patterns are supplied as portable paths. Each pattern should be a non-absolute path, and can’t contain "." or ".." elements. Each slash separated element of a pattern can contain the following special characters:

    • '?' matches any character

    • '*' matches an arbitrary number of characters

      A file $(d)/e1/e2/e3 (where 'd' is in $(dirs)) matches the pattern p1/p2/p3 if and only if e1 matches p1, e2 matches p2 and so on. For example:

      [ glob . : *.cpp ]
      [ glob . : */build/Jamfile ]
  12. rule glob-tree ( roots * : patterns + : exclude-patterns * )

    Recursive version of glob. Builds the glob of files while also searching in the subdirectories of the given roots. An optional set of exclusion patterns will filter out the matching entries from the result. The exclusions also apply to the subdirectory scanning, such that directories that match the exclusion patterns will not be searched.

  13. rule exists ( file )

    Returns true if the specified file exists.

  14. rule all-parents ( path : upper_limit ? : cwd ? )

    Find out the absolute name of path and return the list of all the parents, starting with the immediate one. Parents are returned as relative names. If upper_limit is specified, directories above it will be pruned.

  15. rule glob-in-parents ( dir : patterns + : upper-limit ? )

    Search for patterns in parent directories of dir, up to and including upper_limit, if it is specified, or till the filesystem root otherwise.

  16. rule relative ( child parent : no-error ? )

    Assuming child is a subdirectory of parent, return the relative path from parent to child.

  17. rule relative-to ( path1 path2 )

    Returns the minimal path to path2 that is relative path1.

  18. rule programs-path ( )

    Returns the list of paths which are used by the operating system for looking up programs.

  19. rule makedirs ( path )

    Creates a directory and all parent directories that do not already exist.

5.5.3. regex

Contains rules for string processing using regular expressions.

  • "x*" matches the pattern "x" zero or more times.

  • "x+" matches "x" one or more times.

  • "x?" matches "x" zero or one time.

  • "[abcd]" matches any of the characters, "a", "b", "c", and "d". A character range such as "[a-z]" matches any character between "a" and "z". "[^abc]" matches any character which is not "a", "b", or "c".

  • "x|y" matches either pattern "x" or pattern "y"

  • (x) matches "x" and captures it.

  • "^" matches the beginning of the string.

  • "$" matches the end of the string.

  • "\<" matches the beginning of a word.

  • "\>" matches the end of a word.

    1. rule split ( string separator )

      Returns a list of the following substrings:

      1. from beginning till the first occurrence of separator or till the end,

      2. between each occurrence of separator and the next occurrence,

      3. from the last occurrence of separator till the end.

        If no separator is present, the result will contain only one element.

    2. rule split-list ( list * : separator )

      Returns the concatenated results of applying regex.split to every element of the list using the separator pattern.

    3. rule match ( pattern : string : indices * )

      Match string against pattern, and return the elements indicated by indices.

    4. rule transform ( list * : pattern : indices * )

      Matches all elements of list against the pattern and returns a list of elements indicated by indices of all successful matches. If indices is omitted returns a list of first parenthesized groups of all successful matches.

    5. rule escape ( string : symbols : escape-symbol )

      Escapes all of the characters in symbols using the escape symbol escape-symbol for the given string, and returns the escaped string.

    6. rule replace ( string match replacement )

      Replaces occurrences of a match string in a given string and returns the new string. The match string can be a regex expression.

    7. rule replace-list ( list * : match : replacement )

      Replaces occurrences of a match string in a given list of strings and returns a list of new strings. The match string can be a regex expression.

See also: MATCH

5.5.4. sequence

Various useful list functions. Note that algorithms in this module execute largely in the caller’s module namespace, so that local rules can be used as function objects. Also note that most predicates can be multi-element lists. In that case, all but the first element are prepended to the first argument which is passed to the rule named by the first element.

  1. rule filter ( predicate + : sequence * )

    Return the elements e of $(sequence) for which [ $(predicate) e ] has a non-null value.

  2. rule transform ( function + : sequence * )

    Return a new sequence consisting of [ $(function) $(e) ] for each element e of $(sequence).

  3. rule reverse ( s * )

    Returns the elements of s in reverse order.

  4. rule insertion-sort ( s * : ordered * )

    Insertion-sort s using the BinaryPredicate ordered.

  5. rule merge ( s1 * : s2 * : ordered * )

    Merge two ordered sequences using the BinaryPredicate ordered.

  6. rule join ( s * : joint ? )

    Join the elements of s into one long string. If joint is supplied, it is used as a separator.

  7. rule length ( s * )

    Find the length of any sequence.

  8. rule unique ( list * : stable ? )

    Removes duplicates from list. If stable is passed, then the order of the elements will be unchanged.

  9. rule max-element ( elements + : ordered ? )

    Returns the maximum number in elements. Uses ordered for comparisons or numbers.less if none is provided.

  10. rule select-highest-ranked ( elements * : ranks * )

    Returns all of elements for which the corresponding element in the parallel list rank is equal to the maximum value in rank.

5.5.5. type

Deals with target type declaration and defines target class which supports typed targets.

  1. rule register ( type : suffixes * : base-type ? )

    Registers a target type, possible derived from a base-type. Providing a list of suffixes here is a shortcut for separately calling the register-suffixes rule with the given suffixes and the set-generated-target-suffix rule with the first given suffix.

  2. rule register-suffixes ( suffixes + : type )

    Specifies that files with suffix from suffixes be recognized as targets of type type. Issues an error if a different type is already specified for any of the suffixes.

  3. rule registered ( type )

    Returns true iff type has been registered.

  4. rule validate ( type )

    Issues an error if type is unknown.

  5. rule set-scanner ( type : scanner )

    Sets a scanner class that will be used for this type.

  6. rule get-scanner ( type : property-set )

    Returns a scanner instance appropriate to type and property-set.

  7. rule base ( type )

    Returns a base type for the given type or nothing in case the given type is not derived.

  8. rule all-bases ( type )

    Returns the given type and all of its base types in order of their distance from type.

  9. rule all-derived ( type )

    Returns the given type and all of its derived types in order of their distance from type.

  10. rule is-derived ( type base )

    Returns true if type is equal to base or has base as its direct or indirect base.

  11. rule set-generated-target-suffix ( type : properties * : suffix )

    Sets a file suffix to be used when generating a target of type with the specified properties. Can be called with no properties if no suffix has already been specified for the type. The suffix parameter can be an empty string ("") to indicate that no suffix should be used.

    Note that this does not cause files with suffix to be automatically recognized as being of type. Two different types can use the same suffix for their generated files but only one type can be auto-detected for a file with that suffix. User should explicitly specify which one using the register-suffixes rule.

  12. rule change-generated-target-suffix ( type : properties * : suffix )

    Change the suffix previously registered for this type/properties combination. If suffix is not yet specified, sets it.

  13. rule generated-target-suffix ( type : property-set )

    Returns the suffix used when generating a file of type with the given properties.

  14. rule set-generated-target-prefix ( type : properties * : prefix )

    Sets a target prefix that should be used when generating targets of type with the specified properties. Can be called with empty properties if no prefix for type has been specified yet.

    The prefix parameter can be empty string ("") to indicate that no prefix should be used.

    Usage example: library names use the "lib" prefix on unix.

  15. rule change-generated-target-prefix ( type : properties * : prefix )

    Change the prefix previously registered for this type/properties combination. If prefix is not yet specified, sets it.

  16. rule generated-target-prefix ( type : property-set )

    Returns the prefix used when generating a file of type with the given properties.

  17. rule type ( filename )

    Returns file type given its name. If there are several dots in filename, tries each suffix. E.g. for name of "file.so.1.2" suffixes "2", "1", and "so" will be tried.

5.6. Builtin classes

5.6.1. Class abstract-target

Base class for all abstract targets.

class abstract-target {
    rule __init__ ( name : project )
    rule name ( )
    rule project ( )
    rule location ( )
    rule full-name ( )
    rule generate ( property-set )

Classes derived from abstract-target:

  • project-target

  • main-target

  • basic-target

  1. rule init ( name : project )


    The name of the target in the Jamfile.


    The project to which this target belongs.

  2. rule name ( )

    Returns the name of this target.

  3. rule project ( )

    Returns the project for this target.

  4. rule location ( )

    Returns the location where the target was declared.

  5. rule full-name ( )

    Returns a user-readable name for this target.

  6. rule generate ( property-set )

    Generates virtual targets for this abstract target using the specified properties, unless a different value of some feature is required by the target. This is an abstract method which must be overriden by derived classes.

    On success, returns:

    • a property-set with the usage requirements to be applied to dependents

    • a list of produced virtual targets, which may be empty.

      If property-set is empty, performs the default build of this target, in a way specific to the derived class.

5.6.2. Class project-target

class project-target : abstract-target {
    rule generate ( property-set )
    rule build-dir ( )
    rule main-target ( name )
    rule has-main-target ( name )
    rule find ( id : no-error ? )

    # Methods inherited from abstract-target
    rule name ( )
    rule project ( )
    rule location ( )
    rule full-name ( )

This class has the following responsibilities:

  • Maintaining a list of main targets in this project and building them.

  1. rule generate ( property-set )

    Overrides abstract-target.generate. Generates virtual targets for all the targets contained in this project.

    On success, returns:

    • a property-set with the usage requirements to be applied to dependents

    • a list of produced virtual targets, which may be empty.

  2. rule build-dir ( )

    Returns the root build directory of the project.

  3. rule main-target ( name )

    Returns a main-target class instance corresponding to name. Can only be called after the project has been fully loaded.

  4. rule has-main-target ( name )

    Returns whether a main-target with the specified name exists. Can only be called after the project has been fully loaded.

  5. rule find ( id : no-error ? )

    Find and return the target with the specified id, treated relative to self. Id may specify either a target or a file name with the target taking priority. May report an error or return nothing if the target is not found depending on the no-error parameter.

5.6.3. Class main-target

class main-target : abstract-target {
    rule generate ( property-set )

    # Methods inherited from abstract-target
    rule name ( )
    rule project ( )
    rule location ( )
    rule full-name ( )

A main-target represents a named top-level target in a Jamfile.

  1. rule generate ( property-set )

    Overrides abstract-target.generate. Select an alternative for this main target, by finding all alternatives whose requirements are satisfied by property-set and picking the one with the longest requirements set. Returns the result of calling generate on that alternative.

    On success, returns:

    • a property-set with the usage requirements to be applied to dependents

    • a list of produced virtual targets, which may be empty.

5.6.4. Class basic-target

class basic-target : abstract-target {
    rule __init__ ( name : project : sources * : requirements * : default-build * : usage-requirements * )
    rule generate ( property-set )
    rule construct ( name : source-targets * : property-set )

    # Methods inherited from abstract-target
    rule name ( )
    rule project ( )
    rule location ( )
    rule full-name ( )

Implements the most standard way of constructing main target alternative from sources. Allows sources to be either files or other main targets and handles generation of those dependency targets.

  1. rule init ( name : project : sources * : requirements * : default-build * : usage-requirements * )


    The name of the target


    The project in which the target is declared.

  2. rule generate ( property-set )

    Overrides abstract-target.generate. Determines final build properties, generates sources, and calls construct. This method should not be overridden.

    On success, returns:

    • a property-set with the usage requirements to be applied to dependents

    • a list of produced virtual targets, which may be empty.

  3. rule construct ( name : source-targets * : property-set )

    Constructs virtual targets for this abstract target. Returns a usage-requirements property-set and a list of virtual targets. Should be overriden in derived classes.

5.6.5. Class typed-target

class typed-target : basic-target {
    rule __init__ ( name : project : type : sources * : requirements * : default-build * : usage-requirements * )
    rule type ( )
    rule construct ( name : source-targets * : property-set )

    # Methods inherited from abstract-target
    rule name ( )
    rule project ( )
    rule location ( )
    rule full-name ( )

    # Methods inherited from basic-target
    rule generate ( property-set )

typed-target is the most common kind of target alternative. Rules for creating typed targets are defined automatically for each type.

  1. rule init ( name : project : type : sources * : requirements * : default-build * : usage-requirements * )


    The name of the target


    The project in which the target is declared.


    The type of the target.

  2. rule type ( )

    Returns the type of the target.

  3. rule construct ( name : source-targets * : property-set )

    Implements basic-target.construct. Attempts to create a target of the correct type using generators appropriate for the given property-set. Returns a property-set containing the usage requirements and a list of virtual targets.

    This function is invoked automatically by basic-target.generate and should not be called directly by users.

5.6.6. Class property-set

Class for storing a set of properties.

class property-set {
    rule raw ( )
    rule str ( )
    rule propagated ( )
    rule add ( ps )
    rule add-raw ( properties * )
    rule refine ( ps )
    rule get ( feature )

There is 1<→1 correspondence between identity and value. No two instances of the class are equal. To maintain this property, the 'property-set.create' rule should be used to create new instances. Instances are immutable.

  1. rule raw ( )

    Returns a Jam list of the stored properties.

  2. rule str ( )

    Returns the string repesentation of the stored properties.

  3. rule propagated ( )

    Returns a property-set containing all the propagated properties in this property-set.

  4. rule add ( ps )

    Returns a new property-set containing the union of the properties in this property-set and in ps.

    If ps contains non-free properties that should override the values in this object, use refine instead.
  5. rule add-raw ( properties * )

    Link add, except that it takes a list of properties instead of a property-set.

  6. rule refine ( ps )

    Refines properties by overriding any non-free and non-conditional properties for which a different value is specified in ps. Returns the resulting property-set.

  7. rule get ( feature )

    Returns all the values of feature.

5.7. Build process

The general overview of the build process was given in the user documentation. This section provides additional details, and some specific rules.

To recap, building a target with specific properties includes the following steps:

  1. applying the default build,

  2. selecting the main target alternative to use,

  3. determining the "common" properties,

  4. building targets referred by the the sources list and dependency properties,

  5. adding the usage requirements produced when building dependencies to the "common" properties,

  6. building the target using generators,

  7. computing the usage requirements to be returned.

5.7.1. Alternative selection

When a target has several alternatives, one of them must be selected. The process is as follows:

  1. For each alternative, its condition is defined as the set of base properties in its requirements. Conditional properties are excluded.

  2. An alternative is viable only if all properties in its condition are present in the build request.

  3. If there’s only one viable alternative, it’s choosen. Otherwise, an attempt is made to find the best alternative. An alternative a is better than another alternative b, if the set of properties in b’s condition is a strict subset of the set of properties of a’s condition. If one viable alternative is better than all the others, it’s selected. Otherwise, an error is reported.

5.7.2. Determining common properties

"Common" properties is a somewhat artificial term. This is the intermediate property set from which both the build request for dependencies and the properties for building the target are derived.

Since the default build and alternatives are already handled, we have only two inputs: the build request and the requirements. Here are the rules about common properties.

  1. Non-free features can have only one value

  2. A non-conditional property in the requirements is always present in common properties.

  3. A property in the build request is present in common properties, unless it is overridden by a property in the requirements.

  4. If either the build request, or the requirements (non-conditional or conditional) include an expandable property (either composite, or with a specified subfeature value), the behaviour is equivalent to explicitly adding all the expanded properties to the build request or the requirements respectively.

  5. If the requirements include a conditional property, and the condition of this property is true in the context of common properties, then the conditional property should be in common properties as well.

  6. If no value for a feature is given by other rules here, it has default value in common properties.

These rules are declarative. They don’t specify how to compute the common properties. However, they provide enough information for the user. The important point is the handling of conditional requirements. The condition can be satisfied either by a property in the build request, by non-conditional requirements, or even by another conditional property. For example, the following example works as expected:

exe a : a.cpp
      : <toolset>gcc:<variant>release
        <variant>release:<define>FOO ;

5.7.3. Target Paths

Several factors determine the location of a concrete file target. All files in a project are built under the directory bin unless this is overridden by the build-dir project attribute. Under bin is a path that depends on the properties used to build each target. This path is uniquely determined by all non-free, non-incidental properties. For example, given a property set containing: <toolset>gcc <toolset-gcc:version>4.6.1 <variant>debug <warnings>all <define>_DEBUG <include>/usr/local/include <link>static, the path will be gcc-4.6.1/debug/link-static. <warnings> is an incidental feature and <define> and <include> are free features, so they do not affect the path.

Sometimes the paths produced by Boost.Build can become excessively long. There are a couple of command line options that can help with this. --abbreviate-paths reduces each element to no more than five characters. For example, link-static becomes lnk-sttc. The --hash option reduces the path to a single directory using an MD5 hash.

There are two features that affect the build directory. The <location> feature completely overrides the default build directory. For example,

exe a : a.cpp : <location>. ;

builds all the files produced by a in the directory of the Jamfile. This is generally discouraged, as it precludes variant builds.

The <location-prefix> feature adds a prefix to the path, under the project’s build directory. For example,

exe a : a.cpp : <location-prefix>subdir ;

will create the files for a in bin/subdir/gcc-4.6.1/debug

5.8. Definitions

5.8.1. Features and properties

A feature is a normalized (toolset-independent) aspect of a build configuration, such as whether inlining is enabled. Feature names may not contain the ‘>’ character.

Each feature in a build configuration has one or more associated values. Feature values for non-free features may not contain the ‘<’, ‘:’, or ‘=’ characters. Feature values for free features may not contain the ‘<’ character.

A property is a (feature,value) pair, expressed as <feature>value.

A subfeature is a feature that only exists in the presence of its parent feature, and whose identity can be derived (in the context of its parent) from its value. A subfeature’s parent can never be another subfeature. Thus, features and their subfeatures form a two-level hierarchy.

A value-string for a feature F is a string of the form value-subvalue1-subvalue2…​-subvalueN, where value is a legal value for F and subvalue1…​subvalueN are legal values of some of F's subfeatures. For example, the properties <toolset>gcc <toolset-version>3.0.1 can be expressed more concisely using a value-string, as <toolset>gcc-3.0.1.

A property set is a set of properties (i.e. a collection without duplicates), for instance: <toolset>gcc <runtime-link>static.

A property path is a property set whose elements have been joined into a single string separated by slashes. A property path representation of the previous example would be <toolset>gcc/<runtime-link>static.

A build specification is a property set that fully describes the set of features used to build a target.

5.8.2. Property Validity

For free features, all values are valid. For all other features, the valid values are explicitly specified, and the build system will report an error for the use of an invalid feature-value. Subproperty validity may be restricted so that certain values are valid only in the presence of certain other subproperties. For example, it is possible to specify that the <gcc-target>mingw property is only valid in the presence of <gcc-version>2.95.2.

5.8.3. Feature Attributes

Each feature has a collection of zero or more of the following attributes. Feature attributes are low-level descriptions of how the build system should interpret a feature’s values when they appear in a build request. We also refer to the attributes of properties, so that an incidental property, for example, is one whose feature has the incidental attribute.

  • incidental

    Incidental features are assumed not to affect build products at all. As a consequence, the build system may use the same file for targets whose build specification differs only in incidental features. A feature that controls a compiler’s warning level is one example of a likely incidental feature.

    Non-incidental features are assumed to affect build products, so the files for targets whose build specification differs in non-incidental features are placed in different directories as described in Target Paths.

  • propagated

    Features of this kind are propagated to dependencies. That is, if a main target is built using a propagated property, the build systems attempts to use the same property when building any of its dependencies as part of that main target. For instance, when an optimized executable is requested, one usually wants it to be linked with optimized libraries. Thus, the <optimization> feature is propagated.

  • free

    Most features have a finite set of allowed values, and can only take on a single value from that set in a given build specification. Free features, on the other hand, can have several values at a time and each value can be an arbitrary string. For example, it is possible to have several preprocessor symbols defined simultaneously:

    <define>NDEBUG=1 <define>HAS_CONFIG_H=1
  • optional

    An optional feature is a feature that is not required to appear in a build specification. Every non-optional non-free feature has a default value that is used when a value for the feature is not otherwise specified, either in a target’s requirements or in the user’s build request. [A feature’s default value is given by the first value listed in the feature’s declaration. — move this elsewhere - dwa]

  • symmetric

    Normally a feature only generates a subvariant directory when its value differs from its default value, leading to an asymmetric subvariant directory structure for certain values of the feature. A symmetric feature always generates a corresponding subvariant directory.

  • path

    The value of a path feature specifies a path. The path is treated as relative to the directory of Jamfile where path feature is used and is translated appropriately by the build system when the build is invoked from a different directory

  • implicit

    Values of implicit features alone identify the feature. For example, a user is not required to write "<toolset>gcc", but can simply write "gcc". Implicit feature names also don’t appear in variant paths, although the values do. Thus: bin/gcc/…​ as opposed to bin/toolset-gcc/…​. There should typically be only a few such features, to avoid possible name clashes.

  • composite

    Composite features actually correspond to groups of properties. For example, a build variant is a composite feature. When generating targets from a set of build properties, composite features are recursively expanded and added to the build property set, so rules can find them if necessary. Non-composite non-free features override components of composite features in a build property set.

  • dependency

    The value of a dependency feature is a target reference. When used for building of a main target, the value of dependency feature is treated as additional dependency.

    For example, dependency features allow to state that library A depends on library B. As the result, whenever an application will link to A, it will also link to B. Specifying B as dependency of A is different from adding B to the sources of A.

Features that are neither free nor incidental are called base features.

5.8.4. Feature Declaration

The low-level feature declaration interface is the feature rule from the feature module:

rule feature ( name : allowed-values * : attributes * )

A feature’s allowed-values may be extended with the feature.extend rule.

5.8.5. Property refinement

When a target with certain properties is requested, and that target requires some set of properties, it is needed to find the set of properties to use for building. This process is called property refinement and is performed by these rules

  1. Each property in the required set is added to the original property set

  2. If the original property set includes property with a different value of non free feature, that property is removed.

5.8.6. Conditional properties

Sometime it’s desirable to apply certain requirements only for a specific combination of other properties. For example, one of compilers that you use issues a pointless warning that you want to suppress by passing a command line option to it. You would not want to pass that option to other compilers. Conditional properties allow you to do just that. Their syntax is:

property ( "," property ) * ":" property

For example, the problem above would be solved by:

exe hello : hello.cpp : <toolset>yfc:<cxxflags>-disable-pointless-warning ;

The syntax also allows several properties in the condition, for example:

exe hello : hello.cpp : <os>NT,<toolset>gcc:<link>static ;

5.8.7. Target identifiers and references

Target identifier is used to denote a target. The syntax is:

target-id -> (target-name | file-name | project-id | directory-name)
              | (project-id | directory-name) "//" target-name
project-id -> path
target-name -> path
file-name -> path
directory-name -> path

This grammar allows some elements to be recognized as either

  • name of target declared in current Jamfile (note that target names may include slash).

  • a regular file, denoted by absolute name or name relative to project’s sources location.

  • project id (at this point, all project ids start with slash).

  • the directory of another project, denoted by absolute name or name relative to the current project’s location.

To determine the real meaning the possible interpretations are checked in this order. For example, valid target ids might be:


target in current project


regular file


project "/boost/thread"


target in specific project


project in specific directory

Rationale:Target is separated from project by special separator (not just slash), because:

  • It emphasis that projects and targets are different things.

  • It allows to have main target names with slashes.

Target reference is used to specify a source target, and may additionally specify desired properties for that target. It has this syntax:

target-reference -> target-id [ "/" requested-properties ]
requested-properties -> property-path

For example,

exe compiler : compiler.cpp libs/cmdline/<optimization>space ;

would cause the version of cmdline library, optimized for space, to be linked in even if the compiler executable is build with optimization for speed.

6. Utilities

6.1. Debugger

6.1.1. Overview

Boost.Build comes with a debugger for Jamfiles. To run the debugger, start Boost.Build with b2 -dconsole.

$ b2 -dconsole
(b2db) break gcc.init
Breakpoint 1 set at gcc.init
(b2db) run
Starting program: /usr/bin/b2
Breakpoint 1, gcc.init ( ) at /usr/share/boost-build/tools/gcc.jam:74
74      local tool-command = ;
(b2db) quit

6.1.2. Running the Program

The run command is used to start a new b2 subprocess for debugging. The arguments to run are passed on the command line. If a child process is already running, it will be terminated before the new child is launched.

When the program is paused continue will resume execution. The step command will advance the program by a single statement, stopping on entry to another function or return from the current function. next is like step except that it skips over function calls. finish executes until the current function returns.

The kill command terminates the current child immediately.

6.1.3. Breakpoints

Breakpoints are set using the break command. The location of the breakpoint can be specified as either the name of a function (including the module name) or or a file name and line number of the form file:line. When a breakpoint is created it is given a unique id which is used to identify it for other commands.

(b2db) break Jamfile:10
Breakpoint 1 set at Jamfile:10
(b2db) break msvc.init
Breakpoint 2 set at msvc.init

A breakpoint can be temporarily disabled using the disable command. While a breakpoint is disabled, the child will not stop when it is hit. A disabled breakpoint can be activated again with enable.

(b2db) disable 1
(b2db) enable 1

Breakpoints can be removed permanently with delete or clear. The difference between them is that delete takes the breakpoint id while clear takes the location of the breakpoint as originally specified to break.

(b2db) clear Jamfile:10
Deleted breakpoint 1
(b2db) delete 2

6.1.4. Examining the Stack

The backtrace command will print a summary of every frame on the stack.

The print command can be used to show the value of an expression.

(b2db) print [ modules.peek : ARGV ]
/usr/bin/b2 toolset=msvc install
(b2db) print $(__file__)

6.1.5. Miscellaneous Commands

quit exits the debugger. help describes the available commands.

7. Extender Manual

7.1. Introduction

This section explains how to extend Boost.Build to accomodate your local requirements — primarily to add support for non-standard tools you have. Before we start, be sure you have read and understoon the concept of metatarget, Concepts, which is critical to understanding the remaining material.

The current version of Boost.Build has three levels of targets, listed below.


Object that is created from declarations in Jamfiles. May be called with a set of properties to produce concrete targets.

concrete target

Object that corresponds to a file or an action.

jam target

Low-level concrete target that is specific to Boost.Jam build engine. Essentially a string — most often a name of file.

In most cases, you will only have to deal with concrete targets and the process that creates concrete targets from metatargets. Extending metatarget level is rarely required. The jam targets are typically only used inside the command line patterns.

All of the Boost.Jam target-related builtin functions, like DEPENDS or ALWAYS operate on jam targets. Applying them to metatargets or concrete targets has no effect.

7.1.1. Metatargets

Metatarget is an object that records information specified in Jamfile, such as metatarget kind, name, sources and properties, and can be called with specific properties to generate concrete targets. At the code level it is represented by an instance of class derived from abstract-target. [4]

The generate method takes the build properties (as an instance of the property-set class) and returns a list containing:

  • As front element — Usage-requirements from this invocation (an instance of property-set)

  • As subsequent elements — created concrete targets ( instances of the virtual-target class.)

It’s possible to lookup a metataget by target-id using the targets.resolve-reference function, and the targets.generate-from-reference function can both lookup and generate a metatarget.

The abstract-target class has three immediate derived classes:

  • project-target that corresponds to a project and is not intended for further subclassing. The generate method of this class builds all targets in the project that are not marked as explicit.

  • main-target corresponds to a target in a project and contains one or more target alternatives. This class also should not be subclassed. The generate method of this class selects an alternative to build, and calls the generate method of that alternative.

  • basic-target corresponds to a specific target alternative. This is base class, with a number of derived classes. The generate method processes the target requirements and requested build properties to determine final properties for the target, builds all sources, and finally calls the abstract construct method with the list of source virtual targets, and the final properties.

The instances of the project-target and main-target classes are created implicitly — when loading a new Jamfiles, or when a new target alternative with as-yet unknown name is created. The instances of the classes derived from basic-target are typically created when Jamfile calls a metatarget rule, such as such as exe.

It it permissible to create a custom class derived from basic-target and create new metatarget rule that creates instance of such target. However, in the majority of cases, a specific subclass of basic-target — typed-target is used. That class is associated with a type and relays to generators to construct concrete targets of that type. This process will be explained below. When a new type is declared, a new metatarget rule is automatically defined. That rule creates new instance of type-target, associated with that type.

7.1.2. Concrete targets

Concrete targets are represented by instance of classes derived from virtual-target. The most commonly used subclass is file-target. A file target is associated with an action that creates it —  an instance of the action class. The action, in turn, hold a list of source targets. It also holds the property-set instance with the build properties that should be used for the action.

Here’s an example of creating a target from another target, source

local a = [ new action $(source) : common.copy : $(property-set) ] ;
local t = [ new file-target $(name) : CPP : $(project) : $(a) ] ;

The first line creates an instance of the action class. The first parameter is the list of sources. The second parameter is the name a jam-level action. The third parameter is the property-set applying to this action. The second line creates a target. We specify a name, a type and a project. We also pass the action object created earlier. If the action creates several targets, we can repeat the second line several times.

In some cases, code that creates concrete targets may be invoked more than once with the same properties. Returning two different instances of file-target that correspond to the same file clearly will result in problems. Therefore, whenever returning targets you should pass them via the virtual-target.register function, besides allowing Boost.Build to track which virtual targets got created for each metatarget, this will also replace targets with previously created identical ones, as necessary.[5] Here are a couple of examples:

return [ virtual-target.register $(t) ] ;
return [ sequence.transform virtual-target.register : $(targets) ] ;

7.1.3. Generators

In theory, every kind of metatarget in Boost.Build (like exe, lib or obj) could be implemented by writing a new metatarget class that, independently of the other code, figures what files to produce and what commands to use. However, that would be rather inflexible. For example, adding support for a new compiler would require editing several metatargets.

In practice, most files have specific types, and most tools consume and produce files of specific type. To take advantage of this fact, Boost.Build defines concept of target type and generators generators, and has special metatarget class typed-target. Target type is merely an identifier. It is associated with a set of file extensions that correspond to that type. Generator is an abstraction of a tool. It advertises the types it produces and, if called with a set of input target, tries to construct output targets of the advertised types. Finally, typed-target is associated with specific target type, and relays the generator (or generators) for that type.

A generator is an instance of a class derived from generator. The generator class itself is suitable for common cases. You can define derived classes for custom scenarios.

7.2. Example: 1-to-1 generator

Say you’re writing an application that generates C code. If you ever did this, you know that it's not nice. Embedding large portions of C code in string literals is very awkward. A much better solution is:

  1. Write the template of the code to be generated, leaving placeholders at the points that will change

  2. Access the template in your application and replace placeholders with appropriate text.

  3. Write the result.

It’s quite easy to achieve. You write special verbatim files that are just C++, except that the very first line of the file contains the name of a variable that should be generated. A simple tool is created that takes a verbatim file and creates a cpp file with a single char* variable whose name is taken from the first line of the verbatim file and whose value is the file’s properly quoted content.

Let’s see what Boost.Build can do.

First off, Boost.Build has no idea about "verbatim files". So, you must register a new target type. The following code does it:

import type ;
type.register VERBATIM : verbatim ;

The first parameter to type.register gives the name of the declared type. By convention, it’s uppercase. The second parameter is the suffix for files of this type. So, if Boost.Build sees code.verbatim in a list of sources, it knows that it’s of type VERBATIM.

Next, you tell Boost.Build that the verbatim files can be transformed into C++ files in one build step. A generator is a template for a build step that transforms targets of one type (or set of types) into another. Our generator will be called verbatim.inline-file; it transforms VERBATIM files into CPP files:

import generators ;
generators.register-standard verbatim.inline-file : VERBATIM : CPP ;

Lastly, you have to inform Boost.Build about the shell commands used to make that transformation. That’s done with an actions declaration.

actions inline-file
    "./inline-file.py" $(<) $(>)

Now, we’re ready to tie it all together. Put all the code above in file verbatim.jam, add import verbatim ; to Jamroot.jam, and it’s possible to write the following in your Jamfile:

exe codegen : codegen.cpp class_template.verbatim usage.verbatim ;

The listed verbatim files will be automatically converted into C++ source files, compiled and then linked to the codegen executable.

In subsequent sections, we will extend this example, and review all the mechanisms in detail. The complete code is available in the example/customization directory.

7.3. Target types

The first thing we did in the introduction was declaring a new target type:

import type ;
type.register VERBATIM : verbatim ;

The type is the most important property of a target. Boost.Build can automatically generate necessary build actions only because you specify the desired type (using the different main target rules), and because Boost.Build can guess the type of sources from their extensions.

The first two parameters for the type.register rule are the name of new type and the list of extensions associated with it. A file with an extension from the list will have the given target type. In the case where a target of the declared type is generated from other sources, the first specified extension will be used.

Sometimes you want to change the suffix used for generated targets depending on build properties, such as toolset. For example, some compiler uses extension elf for executable files. You can use the type.set-generated-target-suffix rule:

type.set-generated-target-suffix EXE : <toolset>elf : elf ;

A new target type can be inherited from an existing one.

type.register PLUGIN : : SHARED_LIB ;

The above code defines a new type derived from SHARED_LIB. Initially, the new type inherits all the properties of the base type - in particular generators and suffix. Typically, you’ll change the new type in some way. For example, using type.set-generated-target-suffix you can set the suffix for the new type. Or you can write a special generator for the new type. For example, it can generate additional metainformation for the plugin. In either way, the PLUGIN type can be used whenever SHARED_LIB can. For example, you can directly link plugins to an application.

A type can be defined as "main", in which case Boost.Build will automatically declare a main target rule for building targets of that type. More details can be found later.

7.4. Scanners

Sometimes, a file can refer to other files via some include system. To make Boost.Build track dependencies between included files, you need to provide a scanner. The primary limitation is that only one scanner can be assigned to a target type.

First, we need to declare a new class for the scanner:

class verbatim-scanner : common-scanner
    rule pattern ( )
        return "//###include[ ]*\"([^\"]*)\"" ;

All the complex logic is in the common-scanner class, and you only need to override the method that returns the regular expression to be used for scanning. The parentheses in the regular expression indicate which part of the string is the name of the included file. Only the first parenthesized group in the regular expression will be recognized; if you can’t express everything you want that way, you can return multiple regular expressions, each of which contains a parenthesized group to be matched.

After that, we need to register our scanner class:

scanner.register verbatim-scanner : include ;

The value of the second parameter, in this case include, specifies the properties that contain the list of paths that should be searched for the included files.

Finally, we assign the new scanner to the VERBATIM target type:

type.set-scanner VERBATIM : verbatim-scanner ;

That’s enough for scanning include dependencies.

7.5. Tools and generators

This section will describe how Boost.Build can be extended to support new tools.

For each additional tool, a Boost.Build object called generator must be created. That object has specific types of targets that it accepts and produces. Using that information, Boost.Build is able to automatically invoke the generator. For example, if you declare a generator that takes a target of the type D and produces a target of the type OBJ, when placing a file with extention .d in a list of sources will cause Boost.Build to invoke your generator, and then to link the resulting object file into an application. (Of course, this requires that you specify that the .d extension corresponds to the D type.)

Each generator should be an instance of a class derived from the generator class. In the simplest case, you don’t need to create a derived class, but simply create an instance of the generator class. Let’s review the example we’ve seen in the introduction.

import generators ;
generators.register-standard verbatim.inline-file : VERBATIM : CPP ;
actions inline-file
    "./inline-file.py" $(<) $(>)

We declare a standard generator, specifying its id, the source type and the target type. When invoked, the generator will create a target of type CPP with a source target of type VERBATIM as the only source. But what command will be used to actually generate the file? In Boost.Build, actions are specified using named "actions" blocks and the name of the action block should be specified when creating targets. By convention, generators use the same name of the action block as their own id. So, in above example, the "inline-file" actions block will be used to convert the source into the target.

There are two primary kinds of generators: standard and composing, which are registered with the generators.register-standard and the generators.register-composing rules, respectively. For example:

generators.register-standard verbatim.inline-file : VERBATIM : CPP ;
generators.register-composing mex.mex : CPP LIB : MEX ;

The first (standard) generator takes a single source of type VERBATIM and produces a result. The second (composing) generator takes any number of sources, which can have either the CPP or the LIB type. Composing generators are typically used for generating top-level target type. For example, the first generator invoked when building an exe target is a composing generator corresponding to the proper linker.

You should also know about two specific functions for registering generators: generators.register-c-compiler and generators.register-linker. The first sets up header dependecy scanning for C files, and the seconds handles various complexities like searched libraries. For that reason, you should always use those functions when adding support for compilers and linkers.

(Need a note about UNIX)

Custom generator classes

The standard generators allows you to specify source and target types, an action, and a set of flags. If you need anything more complex, you need to create a new generator class with your own logic. Then, you have to create an instance of that class and register it. Here’s an example how you can create your own generator class:

class custom-generator : generator
    rule __init__ ( * : * )
        generator.__init__ $(1) : $(2) : $(3) : $(4) : $(5) : $(6) : $(7) : $(8) : $(9) ;


  [ new custom-generator verbatim.inline-file : VERBATIM : CPP ] ;

This generator will work exactly like the verbatim.inline-file generator we’ve defined above, but it’s possible to customize the behaviour by overriding methods of the generator class.

There are two methods of interest. The run method is responsible for the overall process - it takes a number of source targets, converts them to the right types, and creates the result. The generated-targets method is called when all sources are converted to the right types to actually create the result.

The generated-targets method can be overridden when you want to add additional properties to the generated targets or use additional sources. For a real-life example, suppose you have a program analysis tool that should be given a name of executable and the list of all sources. Naturally, you don’t want to list all source files manually. Here’s how the generated-targets method can find the list of sources automatically:

class itrace-generator : generator {
    rule generated-targets ( sources + : property-set : project name ? )
        local leaves ;
        local temp = [ virtual-target.traverse $(sources[1]) : : include-sources ] ;
        for local t in $(temp)
            if ! [ $(t).action ]
                leaves += $(t) ;
        return [ generator.generated-targets $(sources) $(leafs)
          : $(property-set) : $(project) $(name) ] ;
generators.register [ new itrace-generator nm.itrace : EXE : ITRACE ] ;

The generated-targets method will be called with a single source target of type EXE. The call to virtual-target.traverse will return all targets the executable depends on, and we further find files that are not produced from anything. The found targets are added to the sources.

The run method can be overriden to completely customize the way the generator works. In particular, the conversion of sources to the desired types can be completely customized. Here’s another real example. Tests for the Boost Python library usually consist of two parts: a Python program and a C file. The C file is compiled to Python extension that is loaded by the Python program. But in the likely case that both files have the same name, the created Python extension must be renamed. Otherwise, the Python program will import itself, not the extension. Here’s how it can be done:

rule run ( project name ? : property-set : sources * )
    local python ;
    for local s in $(sources)
        if [ $(s).type ] = PY
            python = $(s) ;

    local libs ;
    for local s in $(sources)
        if [ type.is-derived [ $(s).type ] LIB ]
            libs += $(s) ;

    local new-sources ;
    for local s in $(sources)
        if [ type.is-derived [ $(s).type ] CPP ]
            local name = [ $(s).name ] ;    # get the target's basename
            if $(name) = [ $(python).name ]
                name = $(name)_ext ;        # rename the target
            new-sources += [ generators.construct $(project) $(name) :
              PYTHON_EXTENSION : $(property-set) : $(s) $(libs) ] ;

    result = [ construct-result $(python) $(new-sources) : $(project) $(name)
                 : $(property-set) ] ;

First, we separate all source into python files, libraries and C sources. For each C source we create a separate Python extension by calling generators.construct and passing the C++ source and the libraries. At this point, we also change the extension’s name, if necessary.

7.6. Features

Often, we need to control the options passed the invoked tools. This is done with features. Consider an example:

# Declare a new free feature
import feature : feature ;
feature verbatim-options : : free ;

# Cause the value of the 'verbatim-options' feature to be
# available as 'OPTIONS' variable inside verbatim.inline-file
import toolset : flags ;
flags verbatim.inline-file OPTIONS <verbatim-options> ;

# Use the "OPTIONS" variable
actions inline-file
    "./inline-file.py" $(OPTIONS) $(<) $(>)

We first define a new feature. Then, the flags invocation says that whenever verbatin.inline-file action is run, the value of the verbatim-options feature will be added to the OPTIONS variable, and can be used inside the action body. You’d need to consult online help (--help) to find all the features of the toolset.flags rule.

Although you can define any set of features and interpret their values in any way, Boost.Build suggests the following coding standard for designing features.

Most features should have a fixed set of values that is portable (tool neutral) across the class of tools they are designed to work with. The user does not have to adjust the values for a exact tool. For example, <optimization>speed has the same meaning for all C++ compilers and the user does not have to worry about the exact options passed to the compiler’s command line.

Besides such portable features there are special 'raw' features that allow the user to pass any value to the command line parameters for a particular tool, if so desired. For example, the <cxxflags> feature allows you to pass any command line options to a C++ compiler. The <include> feature allows you to pass any string preceded by -I and the interpretation is tool-specific. (See Can I get capture external program output using a Boost.Jam variable? for an example of very smart usage of that feature). Of course one should always strive to use portable features, but these are still be provided as a backdoor just to make sure Boost.Build does not take away any control from the user.

Using portable features is a good idea because:

  • When a portable feature is given a fixed set of values, you can build your project with two different settings of the feature and Boost.Build will automatically use two different directories for generated files. Boost.Build does not try to separate targets built with different raw options.

  • Unlike with “raw” features, you don’t need to use specific command-line flags in your Jamfile, and it will be more likely to work with other tools.

Steps for adding a feauture

Adding a feature requires three steps:

  1. Declaring a feature. For that, the "feature.feature" rule is used. You have to decide on the set of feature attributes:

    • if you want a feature value set for one target to automaticaly propagate to its dependant targets then make it “propagated”.

    • if a feature does not have a fixed list of values, it must be “free.” For example, the include feature is a free feature.

    • if a feature is used to refer to a path relative to the Jamfile, it must be a “path” feature. Such features will also get their values automatically converted to Boost.Build’s internal path representation. For example, include is a path feature.

    • if feature is used to refer to some target, it must be a “dependency” feature.

  2. Representing the feature value in a target-specific variable. Build actions are command templates modified by Boost.Jam variable expansions. The toolset.flags rule sets a target-specific variable to the value of a feature.

  3. Using the variable. The variable set in step 2 can be used in a build action to form command parameters or files.

Another example

Here’s another example. Let’s see how we can make a feature that refers to a target. For example, when linking dynamic libraries on Windows, one sometimes needs to specify a "DEF file", telling what functions should be exported. It would be nice to use this file like this:

lib a : a.cpp : <def-file>a.def ;

Actually, this feature is already supported, but anyway…​

  1. Since the feature refers to a target, it must be "dependency".

    feature def-file : : free dependency ;
  2. One of the toolsets that cares about DEF files is msvc. The following line should be added to it.

    flags msvc.link DEF_FILE <def-file> ;
  3. Since the DEF_FILE variable is not used by the msvc.link action, we need to modify it to be:

    actions link bind DEF_FILE
        $(.LD) .... /DEF:$(DEF_FILE) ....

    Note the bind DEF_FILE part. It tells Boost.Build to translate the internal target name in DEF_FILE to a corresponding filename in the link action. Without it the expansion of $(DEF_FILE) would be a strange symbol that is not likely to make sense for the linker.

    We are almost done, except for adding the follwing code to msvc.jam:

    rule link
        DEPENDS $(<) : [ on $(<) return $(DEF_FILE) ] ;

    This is a workaround for a bug in Boost.Build engine, which will hopefully be fixed one day.

Variants and composite features.

Sometimes you want to create a shortcut for some set of features. For example, release is a value of <variant> and is a shortcut for a set of features.

It is possible to define your own build variants. For example:

variant crazy : <optimization>speed <inlining>off
                <debug-symbols>on <profiling>on ;

will define a new variant with the specified set of properties. You can also extend an existing variant:

variant super_release : release : <define>USE_ASM ;

In this case, super_release will expand to all properties specified by release, and the additional one you’ve specified.

You are not restricted to using the variant feature only. Here’s example that defines a brand new feature:

feature parallelism : mpi fake none : composite link-incompatible ;
feature.compose <parallelism>mpi : <library>/mpi//mpi/<parallelism>none ;
feature.compose <parallelism>fake : <library>/mpi//fake/<parallelism>none ;

This will allow you to specify the value of feature parallelism, which will expand to link to the necessary library.

7.7. Main target rules

A main target rule (e.g “exe” Or “lib”) creates a top-level target. It’s quite likely that you’ll want to declare your own and there are two ways to do that.

The first way applies when your target rule should just produce a target of specific type. In that case, a rule is already defined for you! When you define a new type, Boost.Build automatically defines a corresponding rule. The name of the rule is obtained from the name of the type, by downcasing all letters and replacing underscores with dashes. For example, if you create a module obfuscate.jam containing:

import type ;
type.register OBFUSCATED_CPP  : ocpp ;

import generators ;
generators.register-standard obfuscate.file : CPP : OBFUSCATED_CPP ;

and import that module, you’ll be able to use the rule "obfuscated-cpp" in Jamfiles, which will convert source to the OBFUSCATED_CPP type.

The second way is to write a wrapper rule that calls any of the existing rules. For example, suppose you have only one library per directory and want all cpp files in the directory to be compiled into that library. You can achieve this effect using:

lib codegen : [ glob *.cpp ] ;

If you want to make it even simpler, you could add the following definition to the Jamroot.jam file:

rule glib ( name : extra-sources * : requirements * )
    lib $(name) : [ glob *.cpp ] $(extra-sources) : $(requirements) ;

allowing you to reduce the Jamfile to just

glib codegen ;

Note that because you can associate a custom generator with a target type, the logic of building can be rather complicated. For example, the boostbook module declares a target type BOOSTBOOK_MAIN and a custom generator for that type. You can use that as example if your main target rule is non-trivial.

7.8. Toolset modules

If your extensions will be used only on one project, they can be placed in a separate .jam file and imported by your Jamroot.jam. If the extensions will be used on many projects, users will thank you for a finishing touch.

The using rule provides a standard mechanism for loading and configuring extensions. To make it work, your module should provide an init rule. The rule will be called with the same parameters that were passed to the using rule. The set of allowed parameters is determined by you. For example, you can allow the user to specify paths, tool versions, and other options.

Here are some guidelines that help to make Boost.Build more consistent:

  • The init rule should never fail. Even if the user provided an incorrect path, you should emit a warning and go on. Configuration may be shared between different machines, and wrong values on one machine can be OK on another.

  • Prefer specifying the command to be executed to specifying the tool’s installation path. First of all, this gives more control: it’s possible to specify

    time g++

    as the command. Second, while some tools have a logical "installation root", it’s better if the user doesn’t have to remember whether a specific tool requires a full command or a path.

  • Check for multiple initialization. A user can try to initialize the module several times. You need to check for this and decide what to do. Typically, unless you support several versions of a tool, duplicate initialization is a user error. If the tool’s version can be specified during initialization, make sure the version is either always specified, or never specified (in which case the tool is initialied only once). For example, if you allow:

    using yfc ;
    using yfc : 3.3 ;
    using yfc : 3.4 ;

    Then it’s not clear if the first initialization corresponds to version 3.3 of the tool, version 3.4 of the tool, or some other version. This can lead to building twice with the same version.

  • If possible, init must be callable with no parameters. In which case, it should try to autodetect all the necessary information, for example, by looking for a tool in PATH or in common installation locations. Often this is possible and allows the user to simply write:

    using yfc ;
  • Consider using facilities in the tools/common module. You can take a look at how tools/gcc.jam uses that module in the init rule.

8. Frequently Asked Questions

8.1. How do I get the current value of feature in Jamfile?

This is not possible, since Jamfile does not have "current" value of any feature, be it toolset, build variant or anything else. For a single run of Boost.Build, any given main target can be built with several property sets. For example, user can request two build variants on the command line. Or one library is built as shared when used from one application, and as static when used from another. Each Jamfile is read only once so generally there is no single value of a feature you can access in Jamfile.

A feature has a specific value only when building a target, and there are two ways you can use that value:

8.2. I am getting a "Duplicate name of actual target" error. What does that mean?

The most likely case is that you are trying to compile the same file twice, with almost the same, but differing properties. For example:

exe a : a.cpp : <include>/usr/local/include ;
exe b : a.cpp ;

The above snippet requires two different compilations of a.cpp, which differ only in their include property. Since the include feature is declared as free Boost.Build does not create a separate build directory for each of its values and those two builds would both produce object files generated in the same build directory. Ignoring this and compiling the file only once would be dangerous as different includes could potentially cause completely different code to be compiled.

To solve this issue, you need to decide if the file should be compiled once or twice.

  1. To compile the file only once, make sure that properties are the same for both target requests:

    exe a : a.cpp : <include>/usr/local/include ;
    exe b : a.cpp : <include>/usr/local/include ;


    alias a-with-include : a.cpp : <include>/usr/local/include ;
    exe a : a-with-include ;
    exe b : a-with-include ;

    or if you want the includes property not to affect how any other sources added for the built a and b executables would be compiled:

    obj a-obj : a.cpp : <include>/usr/local/include ;
    exe a : a-obj ;
    exe b : a-obj ;

    Note that in both of these cases the include property will be applied only for building these object files and not any other sources that might be added for targets a and b.

  2. To compile the file twice, you can tell Boost.Build to compile it to two separate object files like so:

    obj a_obj : a.cpp : <include>/usr/local/include ;
    obj b_obj : a.cpp ;
    exe a : a_obj ;
    exe b : b_obj ;

    or you can make the object file targets local to the main target:

    exe a : [ obj a_obj : a.cpp : <include>/usr/local/include ] ;
    exe b : [ obj a_obj : a.cpp ] ;

    which will cause Boost.Build to actually change the generated object file names a bit for you and thus avoid any conflicts.

    Note that in both of these cases the include property will be applied only for building these object files and not any other sources that might be added for targets a and b.

A good question is why Boost.Build can not use some of the above approaches automatically. The problem is that such magic would only help in half of the cases, while in the other half it would be silently doing the wrong thing. It is simpler and safer to ask the user to clarify his intention in such cases.

8.3. Accessing environment variables

Many users would like to use environment variables in Jamfiles, for example, to control the location of external libraries. In many cases it is better to declare those external libraries in the site-config.jam file, as documented in the recipes section. However, if the users already have the environment variables set up, it may not be convenient for them to set up their site-config.jam files as well and using the environment variables might be reasonable.

Boost.Jam automatically imports all environment variables into its built-in .ENVIRON module so user can read them from there directly or by using the helper os.environ rule. For example:

import os ;
local unga-unga = [ os.environ UNGA_UNGA ] ;
ECHO $(unga-unga) ;

or a bit more realistic:

import os ;
local SOME_LIBRARY_PATH = [ os.environ SOME_LIBRARY_PATH ] ;
exe a : a.cpp : <include>$(SOME_LIBRARY_PATH) ;

8.4. How to control properties order?

For internal reasons, Boost.Build sorts all the properties alphabetically. This means that if you write:

exe a : a.cpp : <include>b <include>a ;

then the command line with first mention the a include directory, and then b, even though they are specified in the opposite order. In most cases, the user does not care. But sometimes the order of includes, or other properties, is important. For such cases, a special syntax is provided:

exe a : a.cpp : <include>a&&b ;

The && symbols separate property values and specify that their order should be preserved. You are advised to use this feature only when the order of properties really matters and not as a convenient shortcut. Using it everywhere might negatively affect performance.

8.5. How to control the library linking order on Unix?

On Unix-like operating systems, the order in which static libraries are specified when invoking the linker is important, because by default, the linker uses one pass though the libraries list. Passing the libraries in the incorrect order will lead to a link error. Further, this behaviour is often used to make one library override symbols from another. So, sometimes it is necessary to force specific library linking order.

Boost.Build tries to automatically compute the right order. The primary rule is that if library a "uses" library b, then library a will appear on the command line before library b. Library a is considered to use b if b is present either in the a library’s sources or its usage is listed in its requirements. To explicitly specify the use relationship one can use the <use> feature. For example, both of the following lines will cause a to appear before b on the command line:

lib a : a.cpp b ;
lib a : a.cpp : <use>b ;

The same approach works for searched libraries as well:

lib z ;
lib png : : <use>z ;
exe viewer : viewer png z ;

8.6. Can I get capture external program output using a Boost.Jam variable?

The SHELL builtin rule may be used for this purpose:

local gtk_includes = [ SHELL "gtk-config --cflags" ] ;

8.7. How to get the project root (a.k.a. Jamroot) location?

You might want to use your project’s root location in your Jamfiles. To access it just declare a path constant in your Jamroot.jam file using:

path-constant TOP : . ;

After that, the TOP variable can be used in every Jamfile.

8.8. How to change compilation flags for one file?

If one file must be compiled with special options, you need to explicitly declare an obj target for that file and then use that target in your exe or lib target:

exe a : a.cpp b ;
obj b : b.cpp : <optimization>off ;

Of course you can use other properties, for example to specify specific C/C++ compiler options:

exe a : a.cpp b ;
obj b : b.cpp : <cflags>-g ;

You can also use conditional properties for finer control:

exe a : a.cpp b ;
obj b : b.cpp : <variant>release:<optimization>off ;

8.9. Why are the dll-path and hardcode-dll-paths properties useful?

This entry is specific to Unix systems.

Before answering the questions, let us recall a few points about shared libraries. Shared libraries can be used by several applications, or other libraries, without physically including the library in the application which can greatly decrease the total application size. It is also possible to upgrade a shared library when the application is already installed.

However, in order for application depending on shared libraries to be started the OS may need to find the shared library when the application is started. The dynamic linker will search in a system-defined list of paths, load the library and resolve the symbols. Which means that you should either change the system-defined list, given by the LD_LIBRARY_PATH environment variable, or install the libraries to a system location. This can be inconvenient when developing, since the libraries are not yet ready to be installed, and cluttering system paths may be undesirable. Luckily, on Unix there is another way.

An executable can include a list of additional library paths, which will be searched before system paths. This is excellent for development because the build system knows the paths to all libraries and can include them in the executables. That is done when the hardcode-dll-paths feature has the true value, which is the default. When the executables should be installed, the story is different.

Obviously, installed executable should not contain hardcoded paths to your development tree. (The install rule explicitly disables the hardcode-dll-paths feature for that reason.) However, you can use the dll-path feature to add explicit paths manually. For example:

install installed : application : <dll-path>/usr/lib/snake
                                  <location>/usr/bin ;

will allow the application to find libraries placed in the /usr/lib/snake directory.

If you install libraries to a nonstandard location and add an explicit path, you get more control over libraries which will be used. A library of the same name in a system location will not be inadvertently used. If you install libraries to a system location and do not add any paths, the system administrator will have more control. Each library can be individually upgraded, and all applications will use the new library.

Which approach is best depends on your situation. If the libraries are relatively standalone and can be used by third party applications, they should be installed in the system location. If you have lots of libraries which can be used only by your application, it makes sense to install them to a nonstandard directory and add an explicit path, like the example above shows. Please also note that guidelines for different systems differ in this respect. For example, the Debian GNU guidelines prohibit any additional search paths while Solaris guidelines suggest that they should always be used.

8.10. Targets in site-config.jam

It is desirable to declare standard libraries available on a given system. Putting target declaration in a specific project’s Jamfile is not really good, since locations of the libraries can vary between different development machines and then such declarations would need to be duplicated in different projects. The solution is to declare the targets in Boost.Build’s site-config.jam configuration file:

project site-config ;
lib zlib : : <name>z ;

Recall that both site-config.jam and user-config.jam are projects, and everything you can do in a Jamfile you can do in those files as well. So, you declare a project id and a target. Now, one can write:

exe hello : hello.cpp /site-config//zlib ;

in any Jamfile.

8.11. Header-only libraries

In modern C++, libraries often consist of just header files, without any source files to compile. To use such libraries, you need to add proper includes and possibly defines to your project. But with a large number of external libraries it becomes problematic to remember which libraries are header only, and which ones you have to link to. However, with Boost.Build a header-only library can be declared as Boost.Build target and all dependents can use such library without having to remember whether it is a header-only library or not.

Header-only libraries may be declared using the alias rule, specifying their include path as a part of its usage requirements, for example:

alias my-lib
    : # no sources
    : # no build requirements
    : # no default build
    : <include>whatever ;

The includes specified in usage requirements of my-lib are automatically added to all of its dependants' build properties. The dependants need not care if my-lib is a header-only or not, and it is possible to later make my-lib into a regular compiled library without having to add the includes to its dependants' declarations.

If you already have proper usage requirements declared for a project where a header-only library is defined, you do not need to duplicate them for the alias target:

project my : usage-requirements <include>whatever ;
alias mylib ;

8.12. What is the difference between Boost.Build, b2, bjam and Perforce Jam?

Boost.Build is the name of the complete build system. The executable that runs it is b2. That executable is written in C and implements performance-critical algorithms, like traversal of dependency graph and executing commands. It also implements an interpreted language used to implement the rest of Boost.Build. This executable is formally called "Boost.Build engine".

The Boost.Build engine is derived from an earlier build tool called Perforce Jam. Originally, there were just minor changes, and the filename was bjam. Later on, with more and more changes, the similarity of names because a disservice to users, and as of Boost 1.47.0, the official name of the executable was changed to b2. A copy named bjam is still created for compatibility, but you are encouraged to use the new name in all cases.

Perforce Jam was an important foundation, and we gratefully acknowledge its influence, but for users today, these tools share only some basics of the interpreted language.

9. Extra Tools

9.1. Documentation Tools

9.1.1. Asciidoctor

The asciidoctor tool converts the ascidoc documentation format to various backend formats for either viewing or further processing by documentation tools. This tool supports the baseline asciidoctor distribution (i.e. the Ruby based tool).

Feature: asciidoctor-attribute

Defines arbitrary asciidoctor attributes. The value of the feature should be specified with the CLI syntax for attributes. For example to use as a target requirement:

html example : example.adoc :
    <asciidoctor-attribute>idprefix=ex ;

This is a free feature and is not propagated. I.e. it applies only to the target it’s specified on.

Feature: asciidoctor-doctype

Specifies the doctype to use for generating the output format. Allowed doctype values are: article, book, manpage, and inline.

Feature: asciidoctor-backend

Specifies the backend to use to produce output from the source asciidoc. This feature is automatically applied to fit the build target type. For example, when specifying an html target for an asciidoc source:

html example : example.adoc ;

The target will by default acquire the <asciidoctor-backend>html5 requirement. The default for each target type are:

  • html: <asciidoctor-backend>html5

  • docbook: <asciidoctor-backend>docbook45

  • man: <asciidoctor-backend>manpage

  • pdf: <asciidoctor-backend>pdf

To override the defaults you specify it as a requirement on the target:

docbook example : example.adoc :
    <asciidoctor-backend>docbook5 ;

Allowed backend values are: html5, docbook45, docbook5, pdf.


To use the asciidoctor tool you need to declare it in a configuration file with the using rule. The initialiation takes the following arguments:

  • command: The command, with any extra arguments, to execute.

For example you could insert the following in your user-config.jam:

using asciidoctor : "/usr/local/bin/asciidoctor" ;

If no command is given it defaults to just asciidoctor with assumption that the asciidoctor is available in the search PATH.

9.2. Miscellaneous Tools

9.2.1. Sass

This tool converts SASS and SCSS files into CSS. This tool explicitly supports both the version written in C (sassc) and the original Ruby implementation (scss) but other variants might also work. In addition to tool-specific features, described in this section, the tool recognizes features <flags> and <include>.

Feature: sass-style

Sets the output style. Available values are

  • nested: each property is put on its own line, rules are indented based on how deeply they are nested;

  • expanded: each property is put on its own line, rules are not indented;

  • compact: each rule is put on a single line, nested rules occupy adjacent lines, while groups of unrelated rules are separated by newlines;

  • compressed: takes minimum amount of space: all unnecessary whitespace is removed, property values are compressed to have minimal representation.

The feature is optional and is not propagated to dependent targets. If no style is specified, then, if property set contains property <optimization>on, compressed style is selected. Otherwise, nested style is selected.

Feature: sass-line-numbers

Enables emitting comments showing original line numbers for rules. This can be useful for debugging a stylesheet. Available values are on and off. The feature is optional and is not propagated to dependent targets. If no value for this feature is specified, then one is copied from the feature debug-symbols.


To use the sass tool you need to declare it in a configuration file with the using rule. The initialization takes the following arguments:

  • command: the command, with any extra arguments, to execute.

For example you could insert the following in your user-config.jam:

using sass : /usr/local/bin/psass -p2 ; # Perl libsass-based version

If no command is given, sassc is tried, after which scss is tried.

10. Examples

10.1. Introduction

Here we include a collection of simple to complex fully working examples of using Boost Build v2 for various tasks. They show the gamut from simple to advanced features. If you find yourself looking at the examples and not finding something you want to see working please post to our support list and we’ll try and come up with a solution and add it here for others to learn from.

10.2. Hello

This example shows a very basic Boost Build project set up so it compiles a single executable from a single source file:

#include <iostream>

int main()
    std::cout << "Hello!\n";
    return 1;

Our jamroot.jam is minimal and only specifies one exe target for the program:

exe hello : hello.cpp ;

Building the example yields:

> cd /example/hello
> b2
...found 8 targets...
...updating 4 targets...
common.mkdir bin/clang-darwin-4.2.1
common.mkdir bin/clang-darwin-4.2.1/debug
clang-darwin.compile.c++ bin/clang-darwin-4.2.1/debug/hello.o
clang-darwin.link bin/clang-darwin-4.2.1/debug/hello
...updated 4 targets...
> bin/clang-darwin-4.2.1/debug/hello
The actual paths in the bin sub-directory will depend on your toolset.

11. Boost.Jam Documentation

Jam is a make(1) replacement that makes building simple things simple and building complicated things manageable.

11.1. Building B2

Installing B2 after building it is simply a matter of copying the generated executables someplace in your PATH. For building the executables there are a set of build bootstrap scripts to accomodate particular environments. The scripts take one optional argument, the name of the toolset to build with. When the toolset is not given an attempt is made to detect an available toolset and use that. The build scripts accept these arguments:

build [toolset]

Running the scripts without arguments will give you the best chance of success. On Windows platforms from a command console do:

cd jam source location

On Unix type platforms do:

cd jam source location
sh ./build.sh

For the Boost.Jam source included with the Boost distribution the jam source location is BOOST_ROOT/tools/build/src/engine.

If the scripts fail to detect an appropriate toolset to build with your particular toolset may not be auto-detectable. In that case, you can specify the toolset as the first argument, this assumes that the toolset is readily available in the PATH.

The toolset used to build Boost.Jam is independent of the toolsets used for Boost.Build. Only one version of Boost.Jam is needed to use Boost.Build.

The supported toolsets, and whether they are auto-detected, are:

Table 2. Supported Toolsets
Script Platform Toolset Detection and Notes


Windows NT, 2000, and XP

borland, Borland C++Builder (BCC 5.5)

  • Common install location: C:\Borland\BCC55

  • BCC32.EXE in PATH

como, Comeau Computing C/C++

gcc, GNU GCC

gcc-nocygwin, GNU GCC

intel-win32, Intel C++ Compiler for Windows


metrowerks, MetroWerks CodeWarrior C/C++ 7.x, 8.x, 9.x

  • CWFolder variable configured


mingw, GNU GCC as the MinGW configuration

  • Common install location: C:\MinGW

msvc, Microsoft Visual C++ 6.x

  • VCVARS32.BAT already configured

  • %MSVCDir% is present in environment

  • Common install locations: %ProgramFiles%\Microsoft Visual Studio, %ProgramFiles%\Microsoft Visual C++

  • CL.EXE in PATH

vc7, Microsoft Visual C++ 7.x

  • VCVARS32.BAT or VSVARS32.BAT already configured

  • %VS71COMNTOOLS% is present in environment

  • %VCINSTALLDIR% is present in environment

  • Common install locations: %ProgramFiles%\Microsoft Visual Studio .NET, %ProgramFiles%\Microsoft Visual Studio .NET 2003

  • CL.EXE in PATH

vc8 and vc9, Microsoft Visual C++ 8.x and 9.x


  • VCVARSALL.BAT already configured

  • %VS90COMNTOOLS% is present in environment

  • Common install location: %ProgramFiles%\Microsoft Visual Studio 9

  • %VS80COMNTOOLS% is present in environment

  • Common install location: %ProgramFiles%\Microsoft Visual Studio 8

  • CL.EXE in PATH


  • If VCVARSALL.BAT is called to set up the toolset, it is passed all the extra arguments, see below for what those arguments are. This can be used to build, for example, a Win64 specific version of b2. Consult the VisualStudio documentation for what the possible argument values to the VCVARSALL.BAT are.


Unix, Linux, Cygwin, etc.

acc, HP-UX aCC

  • aCC in PATH

  • uname is "HP-UX"

como, Comeau Computing C/C++

  • como in PATH

gcc, GNU GCC

  • gcc in PATH

intel-linux, Intel C++ for Linux

  • icc in PATH

  • Common install locations: /opt/intel/cc/9.0, /opt/intel_cc_80, /opt/intel/compiler70, /opt/intel/compiler60, /opt/intel/compiler50

kcc, Intel KAI C++

  • KCC in PATH

kylix, Borland C++Builder

  • bc++ in PATH

mipspro, SGI MIPSpro C

  • uname is "IRIX" or "IRIX64"

sunpro, Sun Workshop 6 C++

  • Standard install location: /opt/SUNWspro

qcc, QNX Neutrino

  • uname is "QNX" and qcc in PATH

true64cxx, Compaq C++ Compiler for True64 UNIX

  • uname is "OSF1"

vacpp, IBM VisualAge C++

  • xlc in PATH


darwin, Apple MacOS X GCC

  • uname is "Darwin"

Windows NT, 2000, and XP

mingw, GNU GCC as the MinGW configuration with the MSYS shell

  • Common install location: /mingw

The built executables are placed in a subdirectory specific to your platform. For example, in Linux running on an Intel x86 compatible chip, the executables are placed in: bin.linuxx86. The b2[.exe] executable can be used to invoke Boost.Build.

The build scripts support additional invocation arguments for use by developers of Boost.Jam and for additional setup of the toolset. The extra arguments come after the toolset:

  • Arguments not in the form of an option, before option arguments, are used for extra setup to toolset configuration scripts.

  • Arguments of the form --option, which are passed to the build.jam build script.

  • Arguments not in the form of an option, after the options, which are targets for the build.jam script.

build [toolset] [setup*] [--option+ target*]

The arguments immediately after the toolset are passed directly to the setup script of the toolset, if available and if it needs to be invoked. This allows one to configure the toolset ass needed to do non-default builds of b2. For example to build a Win64 version with vc8. See the toolset descriptiona above for when particular toolsets support this.

The arguments starting with the --option forms are passed to the build.jam script and are used to further customize what gets built. Options and targets supported by the build.jam script:


Empty option when one wants to only specify a target.


The default, builds the optimized executable.


Builds debugging versions of the executable. When built they are placed in their own directory bin./platform/.debug.


Normally the Jam language grammar parsing files are not regenerated. This forces building of the grammar, although it may not force the regeneration of the grammar parser. If the parser is out of date it will be regenerated and subsequently built.


Enables Python integration, given a path to the Python libraries.


Enables use of the Boehm Garbage Collector. The build will look for the Boehm-GC source in a "boehm_gc" subdirectory from the b2 sources.


Enables use of the DUMA (Detect Unintended Memory Access) debugging memory allocator. The build expects to find the DUMA source files in a "duma" subdirectory from the b2 sources.


Indicates where the toolset used to build is located. This option is passed in by the bootstrap (build.bat or build.sh) script.


For information, prints out where it will put the built executable.


Disable debug assertions, even if building the debug version of the executable.


Generate packages (compressed archives) as appropriate for distribution in the platform, if possible.


Remove all the built executables and objects.

11.2. Language

B2 has an interpreted, procedural language. Statements in b2 are rule (procedure) definitions, rule invocations, flow-of-control structures, variable assignments, and sundry language support.

11.2.1. Lexical Features

B2 treats its input files as whitespace-separated tokens, with two exceptions: double quotes (") can enclose whitespace to embed it into a token, and everything between the matching curly braces (\{}) in the definition of a rule action is treated as a single string. A backslash (\) can escape a double quote, or any single whitespace character.

B2 requires whitespace (blanks, tabs, or newlines) to surround all tokens, including the colon (:) and semicolon (;) tokens.

B2 keywords (an mentioned in this document) are reserved and generally must be quoted with double quotes (") to be used as arbitrary tokens, such as variable or target names.

Comments start with the # character and extend until the end of line. And block comments start with #| and extend until the next |#.

11.2.2. Targets

The essential b2 data entity is a target. Build targets are files to be updated. Source targets are the files used in updating built targets. Built targets and source targets are collectively referred to as file targets, and frequently built targets are source targets for other built targets. Pseudotargets are symbols representing dependencies on other targets, but which are not themselves associated with any real file.

A file target’s identifier is generally the file’s name, which can be absolutely rooted, relative to the directory of b2`s invocation, or simply local (no directory). Most often it is the last case, and the actual file path is bound using the `$(SEARCH) and $(LOCATE) special variables. See SEARCH and LOCATE Variables below. A local filename is optionally qualified with grist, a string value used to assure uniqueness. A file target with an identifier of the form file(member) is a library member (usually an ar(1) archive on Unix).

Binding Detection

Whenever a target is bound to a location in the filesystem, Boost Jam will look for a variable called BINDRULE (first "on" the target being bound, then in the global module). If non-empty, $(BINDRULE[1]) names a rule which is called with the name of the target and the path it is being bound to. The signature of the rule named by $(BINDRULE[1]) should match the following:

rule bind-rule ( target : path )

This facility is useful for correct header file scanning, since many compilers will search for #include files first in the directory containing the file doing the #include directive. $(BINDRULE) can be used to make a record of that directory.

11.2.3. Rules

The basic b2 language entity is called a rule. A rule is defined in two parts: the procedure and the actions. The procedure is a body of jam statements to be run when the rule is invoked; the actions are the OS shell commands to execute when updating the built targets of the rule.

Rules can return values, which can be expanded into a list with "[ rule args …​ ]". A rule’s value is the value of its last statement, though only the following statements have values: 'if' (value of the leg chosen), 'switch' (value of the case chosen), set (value of the resulting variable), and 'return' (value of its arguments).

The b2 statements for defining and invoking rules are as follows:

Define a rule’s procedure, replacing any previous definition.

rule rulename { statements }

Define a rule’s updating actions, replacing any previous definition.

actions [ modifiers ] rulename { commands }

Invoke a rule.

rulename field1 : field2 : ... : fieldN ;

Invoke a rule under the influence of target’s specific variables..

on target rulename field1 : field2 : ... : fieldN ;

Used as an argument, expands to the return value of the rule invoked.

[ rulename field1 : field2 : ... : fieldN ]
[ on target rulename field1 : field2 : ... : fieldN ]

A rule is invoked with values in field1 through fieldN. They may be referenced in the procedure’s statements as $(1) through $(N) (9 max), and the first two only may be referenced in the action’s commands as $(1) and $(2). $(<) and $(>) are synonymous with $(1) and $(2).

Rules fall into two categories: updating rules (with actions), and pure procedure rules (without actions). Updating rules treat arguments $(1) and $(2) as built targets and sources, respectively, while pure procedure rules can take arbitrary arguments.

When an updating rule is invoked, its updating actions are added to those associated with its built targets ($(1)) before the rule’s procedure is run. Later, to build the targets in the updating phase, commands are passed to the OS command shell, with $(1) and $(2) replaced by bound versions of the target names. See Binding above.

Rule invocation may be indirected through a variable:

$(var) field1 : field2 : ... : fieldN ;

on target $(var) field1 : field2 : ... : fieldN ;

[ $(var) field1 : field2 : ... : fieldN ]
[ on target $(var) field1 : field2 : ... : fieldN ]

The variable’s value names the rule (or rules) to be invoked. A rule is invoked for each element in the list of $(var)`s values. The fields `field1 : field2 : …​ are passed as arguments for each invokation. For the [ …​ ] forms, the return value is the concatenation of the return values for all of the invocations.

Action Modifiers

The following action modifiers are understood:

actions bind vars

$(vars) will be replaced with bound values.

actions existing

$(>) includes only source targets currently existing.

actions ignore

The return status of the commands is ignored.

actions piecemeal

commands are repeatedly invoked with a subset of $(>) small enough to fit in the command buffer on this OS.

actions quietly

The action is not echoed to the standard output.

actions together

The $(>) from multiple invocations of the same action on the same built target are glommed together.

actions updated

$(>) includes only source targets themselves marked for updating.

Argument lists

You can describe the arguments accepted by a rule, and refer to them by name within the rule. For example, the following prints "I’m sorry, Dave" to the console:

rule report ( pronoun index ? : state : names + )
    local he.suffix she.suffix it.suffix = s ;
    local I.suffix = m ;
    local they.suffix you.suffix = re ;
    ECHO $(pronoun)'$($(pronoun).suffix) $(state), $(names[$(index)]) ;
report I 2 : sorry : Joe Dave Pete ;

Each name in a list of formal arguments (separated by : in the rule declaration) is bound to a single element of the corresponding actual argument unless followed by one of these modifiers:

Symbol Semantics of preceding symbol




Bind to zero or more unbound elements of the actual argument. When * appears where an argument name is expected, any number of additional arguments are accepted. This feature can be used to implement "varargs" rules.


Bind to one or more unbound elements of the actual argument.

The actual and formal arguments are checked for inconsistencies, which cause b2 to exit with an error code:

### argument error
# rule report ( pronoun index ?  : state  : names + )
# called with: ( I 2 foo  : sorry  : Joe Dave Pete )
# extra argument foo
### argument error
# rule report ( pronoun index ?  : state  : names + )
# called with: ( I 2  : sorry )
# missing argument names

If you omit the list of formal arguments, all checking is bypassed as in "classic" Jam. Argument lists drastically improve the reliability and readability of your rules, however, and are strongly recommended for any new Jam code you write.

11.2.4. Built-in Rules

B2 has a growing set of built-in rules, all of which are pure procedure rules without updating actions. They are in three groups: the first builds the dependency graph; the second modifies it; and the third are just utility rules.

Dependency Building
rule DEPENDS ( targets1 * : targets2 * )

Builds a direct dependency: makes each of targets1 depend on each of targets2. Generally, targets1 will be rebuilt if targets2 are themselves rebuilt or are newer than targets1.

rule INCLUDES ( targets1 * : targets2 * )

Builds a sibling dependency: makes any target that depends on any of targets1 also depend on each of targets2. This reflects the dependencies that arise when one source file includes another: the object built from the source file depends both on the original and included source file, but the two sources files don’t depend on each other. For example:

DEPENDS foo.o : foo.c ;
INCLUDES foo.c : foo.h ;

foo.o depends on foo.c and foo.h in this example.

Modifying Binding

The six rules ALWAYS, LEAVES, NOCARE, NOTFILE, NOUPDATE, and TEMPORARY modify the dependency graph so that b2 treats the targets differently during its target binding phase. See Binding above. Normally, b2 updates a target if it is missing, if its filesystem modification time is older than any of its dependencies (recursively), or if any of its dependencies are being updated. This basic behavior can be changed by invoking the following rules:

rule ALWAYS ( targets * )

Causes targets to be rebuilt regardless of whether they are up-to-date (they must still be in the dependency graph). This is used for the clean and uninstall targets, as they have no dependencies and would otherwise appear never to need building. It is best applied to targets that are also NOTFILE targets, but it can also be used to force a real file to be updated as well.

rule LEAVES ( targets * )

Makes each of targets depend only on its leaf sources, and not on any intermediate targets. This makes it immune to its dependencies being updated, as the "leaf" dependencies are those without their own dependencies and without updating actions. This allows a target to be updated only if original source files change.

rule NOCARE ( targets * )

Causes b2 to ignore targets that neither can be found nor have updating actions to build them. Normally for such targets b2 issues a warning and then skips other targets that depend on these missing targets. The HdrRule in Jambase uses NOCARE on the header file names found during header file scanning, to let b2 know that the included files may not exist. For example, if an #include is within an #ifdef, the included file may not actually be around.

For targets with build actions: if their build actions exit with a nonzero return code, dependent targets will still be built.
rule NOTFILE ( targets * )

Marks targets as pseudotargets and not real files. No timestamp is checked, and so the actions on such a target are only executed if the target’s dependencies are updated, or if the target is also marked with ALWAYS. The default b2 target all is a pseudotarget. In Jambase, NOTFILE is used to define several addition convenient pseudotargets.

rule NOUPDATE ( targets * )

Causes the timestamps on targets to be ignored. This has two effects: first, once the target has been created it will never be updated; second, manually updating target will not cause other targets to be updated. In Jambase, for example, this rule is applied to directories by the MkDir rule, because MkDir only cares that the target directory exists, not when it has last been updated.

rule TEMPORARY ( targets * )

Marks targets as temporary, allowing them to be removed after other targets that depend upon them have been updated. If a TEMPORARY target is missing, b2 uses the timestamp of the target’s parent. Jambase uses TEMPORARY to mark object files that are archived in a library after they are built, so that they can be deleted after they are archived.

rule FAIL_EXPECTED ( targets * )

For handling targets whose build actions are expected to fail (e.g. when testing that assertions or compile-time type checking work properly), Boost Jam supplies the FAIL_EXPECTED rule in the same style as NOCARE, et. al. During target updating, the return code of the build actions for arguments to FAIL_EXPECTED is inverted: if it fails, building of dependent targets continues as though it succeeded. If it succeeds, dependent targets are skipped.

rule RMOLD ( targets * )

B2 removes any target files that may exist on disk when the rule used to build those targets fails. However, targets whose dependencies fail to build are not removed by default. The RMOLD rule causes its arguments to be removed if any of their dependencies fail to build.

rule ISFILE ( targets * )

ISFILE marks targets as required to be files. This changes the way b2 searches for the target such that it ignores matches for file system items that are not files, like directories. This makes it possible to avoid #include "exception" matching if one happens to have a directory named exception in the header search path.

This is currently not fully implemented.

The two rules ECHO and EXIT are utility rules, used only in `b2`s parsing phase.

rule ECHO ( args * )

Blurts out the message args to stdout.

rule EXIT ( message * : result-value ? )

Blurts out the message to stdout and then exits with a failure status if no result-value is given, otherwise it exits with the given result-value.

Echo, echo, Exit, and exit are accepted as aliases for ECHO and EXIT, since it is hard to tell that these are built-in rules and not part of the language, like include.


The GLOB rule does filename globbing.

rule GLOB ( directories * : patterns * : downcase-opt ? )

Using the same wildcards as for the patterns in the switch statement. It is invoked by being used as an argument to a rule invocation inside of "[ ]". For example: FILES = [ GLOB dir1 dir2 : *.c *.h ] sets FILES to the list of C source and header files in dir1 and dir2. The resulting filenames are the full pathnames, including the directory, but the pattern is applied only to the file name without the directory.

If downcase-opt is supplied, filenames are converted to all-lowercase before matching against the pattern; you can use this to do case-insensitive matching using lowercase patterns. The paths returned will still have mixed case if the OS supplies them. On Windows NT and Cygwin, and OpenVMS, filenames are always downcased before matching.


The GLOB_ARCHIVE rule does name globbing of object archive members.

rule GLOB_ARCHIVE ( archives * : member-patterns * : downcase-opt ? : symbol-patterns ? )

Similarly to GLOB, this rule is used to match names of member files in an archive (static object library). List of successfully matched members is returned or null otherwise. The resulting member names are qualified with pathname of the containing archive in the form archive-path(member-name). Member patterns are for matching member name only; when no wildcards specified — an exact match is assumed. Member names generally correspond to object file names and as such are platform-specific — use of platform-defined object suffix in the matching patterns can allow for portability.

If downcase-opt is supplied, the member names are converted to all-lowercase before matching against the pattern; you can use this to do case-insensitive matching using lowercase patterns. The paths returned will still have mixed case if the OS supplies them. On Windows NT, Cygwin, and OpenVMS, filenames are always downcased before matching.

Additionally, members can be matched with symbol/function patterns on supported platforms (currently, OpenVMS only). In this case, members containing the matching symbols are returned. Member and symbol patterns are applied as OR conditions, with member patterns taking precedence. On unsupported platforms, null is returned when any symbol patterns are specified.


The MATCH rule does pattern matching.

rule MATCH ( regexps + : list * )

Matches the egrep(1) style regular expressions regexps against the strings in list. The result is a list of matching () subexpressions for each string in list, and for each regular expression in regexps.

rule BACKTRACE ( )

Returns a list of quadruples: filename line module rulename…​, describing each shallower level of the call stack. This rule can be used to generate useful diagnostic messages from Jam rules.

rule UPDATE ( targets * )

Classic jam treats any non-option element of command line as a name of target to be updated. This prevented more sophisticated handling of command line. This is now enabled again but with additional changes to the UPDATE rule to allow for the flexibility of changing the list of targets to update. The UPDATE rule has two effects:

  1. It clears the list of targets to update, and

  2. Causes the specified targets to be updated.

If no target was specified with the UPDATE rule, no targets will be updated. To support changing of the update list in more useful ways, the rule also returns the targets previously in the update list. This makes it possible to add targets as such:

local previous-updates = [ UPDATE ] ;
UPDATE $(previous-updates) a-new-target ;
rule W32_GETREG ( path : data ? )

Defined only for win32 platform. It reads the registry of Windows. 'path' is the location of the information, and 'data' is the name of the value which we want to get. If 'data' is omitted, the default value of 'path' will be returned. The 'path' value must conform to MS key path format and must be prefixed with one of the predefined root keys. As usual,

  • HKLM is equivalent to HKEY_LOCAL_MACHINE.

  • HKCU is equivalent to HKEY_CURRENT_USER.

  • HKCR is equivalent to HKEY_CLASSES_ROOT.

Other predefined root keys are not supported.

Currently supported data types : REG_DWORD, REG_SZ, REG_EXPAND_SZ, REG_MULTI_SZ. The data with REG_DWORD type will be turned into a string, REG_MULTI_SZ into a list of strings, and for those with REG_EXPAND_SZ type environment variables in it will be replaced with their defined values. The data with REG_SZ type and other unsupported types will be put into a string without modification. If it can’t receive the value of the data, it just return an empty list. For example,

local PSDK-location =
  [ W32_GETREG HKEY_LOCAL_MACHINE\\SOFTWARE\\Microsoft\\MicrosoftSDK\\Directories : "Install Dir" ] ;
rule W32_GETREGNAMES ( path : result-type )

Defined only for win32 platform. It reads the registry of Windows. 'path' is the location of the information, and 'result-type' is either subkeys or values. For more information on 'path' format and constraints, please see W32_GETREG.

Depending on 'result-type', the rule returns one of the following:


Names of all direct subkeys of 'path'.


Names of values contained in registry key given by 'path'. The "default" value of the key appears in the returned list only if its value has been set in the registry.

If 'result-type' is not recognized, or requested data cannot be retrieved, the rule returns an empty list. Example:

local key = "HKEY_LOCAL_MACHINE\\SOFTWARE\\Microsoft\\Windows\\CurrentVersion\\App Paths" ;
local subkeys = [ W32_GETREGNAMES "$(key)" : subkeys ] ;
for local subkey in $(subkeys)
    local values = [ W32_GETREGNAMES "$(key)\\$(subkey)" : values ] ;
    for local value in $(values)
        local data = [ W32_GETREG "$(key)\\$(subkey)" : "$(value)" ] ;
        ECHO "Registry path: " $(key)\\$(subkey) ":" $(value) "=" $(data) ;
rule SHELL ( command : * )

SHELL executes command, and then returns the standard output of command. SHELL only works on platforms with a popen() function in the C library. On platforms without a working popen() function, SHELL is implemented as a no-op. SHELL works on Unix, MacOS X, and most Windows compilers. SHELL is a no-op on Metrowerks compilers under Windows. There is a variable set of allowed options as additional arguments:


In addition to the output the result status of the executed command is returned as a second element of the result.


Don’t capture the output of the command. Instead an empty ("") string value is returned in place of the output.


Remove trailing end-of-line character from output, if any.

Because the Perforce/Jambase defines a SHELL rule which hides the builtin rule, COMMAND can be used as an alias for SHELL in such a case.

rule MD5 ( string )

MD5 computes the MD5 hash of the string passed as paramater and returns it.

rule SPLIT_BY_CHARACTERS ( string : delimiters )

SPLIT_BY_CHARACTERS splits the specified string on any delimiter character present in delimiters and returns the resulting list.

rule PRECIOUS ( targets * )

The PRECIOUS rule specifies that each of the targets passed as the arguments should not be removed even if the command updating that target fails.

rule PAD ( string : width )

If string is shorter than width characters, pads it with whitespace characters on the right, and returns the result. Otherwise, returns string unmodified.

rule FILE_OPEN ( filename : mode )

The FILE_OPEN rule opens the specified file and returns a file descriptor. The mode parameter can be either "w" or "r". Note that at present, only the UPDATE_NOW rule can use the resulting file descriptor number.

rule UPDATE_NOW ( targets * : log ? : ignore-minus-n ? )

The UPDATE_NOW caused the specified targets to be updated immediately. If update was successfull, non-empty string is returned. The log parameter, if present, specifies a descriptor of a file where all output from building is redirected. If the ignore-minus-n parameter is specified, the targets are updated even if the -n parameter is specified on the command line.

11.2.5. Flow-of-Control

B2 has several simple flow-of-control statements:

for var in list { statements }

Executes statements for each element in list, setting the variable var to the element value.

if cond { statements }
[ else { statements } ]

Does the obvious; the else clause is optional. cond is built of:


true if any a element is a non-zero-length string

a = b

list a matches list b string-for-string

a != b

list a does not match list b

a < b

a[i] string is less than b[i] string, where i is first mismatched element in lists a and b

a <= b

every a string is less than or equal to its b counterpart

a > b

a[i] string is greater than b[i] string, where i is first mismatched element

a >= b

every a string is greater than or equal to its b counterpart

a in b

true if all elements of a can be found in b, or if a has no elements

! cond

condition not true

cond && cond


cond || cond


( cond )

precedence grouping

include file ;

Causes b2 to read the named file. The file is bound like a regular target (see Binding above) but unlike a regular target the include file cannot be built.

The include file is inserted into the input stream during the parsing phase. The primary input file and all the included file(s) are treated as a single file; that is, b2 infers no scope boundaries from included files.

local vars [ = values ] ;

Creates new vars inside to the enclosing {} block, obscuring any previous values they might have. The previous values for vars are restored when the current block ends. Any rule called or file included will see the local and not the previous value (this is sometimes called Dynamic Scoping). The local statement may appear anywhere, even outside of a block (in which case the previous value is restored when the input ends). The vars are initialized to values if present, or left uninitialized otherwise.

return values ;

Within a rule body, the return statement sets the return value for an invocation of the rule and returns to the caller.

switch value
    case pattern1 : statements ;
    case pattern2 : statements ;

The switch statement executes zero or one of the enclosed statements, depending on which, if any, is the first case whose pattern matches value. The pattern values are not variable-expanded. The pattern values may include the following wildcards:


match any single character


match zero or more characters


match any single character in chars


match any single character not in chars


match x (escapes the other wildcards)

while cond { statements }

Repeatedly execute statements while cond remains true upon entry. (See the description of cond expression syntax under if, above).

break ;

Immediately exits the nearest enclosing while or for loop.

continue ;

Jumps to the top of the nearest enclosing while or for loop.

11.2.6. Variables

B2 variables are lists of zero or more elements, with each element being a string value. An undefined variable is indistinguishable from a variable with an empty list, however, a defined variable may have one more elements which are null strings. All variables are referenced as $(variable).

Variables are either global or target-specific. In the latter case, the variable takes on the given value only during the updating of the specific target.

A variable is defined with:

variable = elements ;
variable += elements ;
variable on targets = elements ;
variable on targets += elements ;
variable default = elements ;
variable ?= elements ;

The first two forms set variable globally. The third and forth forms set a target-specific variable. The = operator replaces any previous elements of variable with elements; the += operation adds elements to variable's list of elements. The final two forms are synonymous: they set variable globally, but only if it was previously unset.

Variables referenced in updating commands will be replaced with their values; target-specific values take precedence over global values. Variables passed as arguments ($(1) and $(2)) to actions are replaced with their bound values; the bind modifier can be used on actions to cause other variables to be replaced with bound values. See Action Modifiers above.

B2 variables are not re-exported to the environment of the shell that executes the updating actions, but the updating actions can reference b2 variables with $(variable).

Variable Expansion

During parsing, b2 performs variable expansion on each token that is not a keyword or rule name. Such tokens with embedded variable references are replaced with zero or more tokens. Variable references are of the form $(v) or $(vm), where v is the variable name, and m are optional modifiers.

Variable expansion in a rule’s actions is similar to variable expansion in statements, except that the action string is tokenized at whitespace regardless of quoting.

The result of a token after variable expansion is the product of the components of the token, where each component is a literal substring or a list substituting a variable reference. For example:

$(X) -> a b c
t$(X) -> ta tb tc
$(X)z -> az bz cz
$(X)-$(X) -> a-a a-b a-c b-a b-b b-c c-a c-b c-c

The variable name and modifiers can themselves contain a variable reference, and this partakes of the product as well:

$(X) -> a b c
$(Y) -> 1 2
$(Z) -> X Y
$($(Z)) -> a b c 1 2

Because of this product expansion, if any variable reference in a token is undefined, the result of the expansion is an empty list. If any variable element is a null string, the result propagates the non-null elements:

$(X) -> a ""
$(Y) -> "" 1
$(Z) ->
-$(X)$(Y)- -> -a- -a1- -- -1-
-$(X)$(Z)- ->

A variable element’s string value can be parsed into grist and filename-related components. Modifiers to a variable are used to select elements, select components, and replace components. The modifiers are:


Select element number n (starting at 1). If the variable contains fewer than n elements, the result is a zero-element list. n can be negative in which case the element number n from the last leftward is returned.


Select elements number n through m. n and m can be negative in which case they refer to elements counting from the last leftward.


Select elements number n through the last. n can be negative in which case it refers to the element counting from the last leftward.


+ Select filename base.


Select (last) filename suffix.


Select archive member name.


Select directory path.


Select parent directory.


Select grist.


Replace lowercase characters with uppercase.


Replace uppercase characters with lowercase.


Converts all back-slashes ("\") to forward slashes ("/"). For example

x = "C:\\Program Files\\Borland" ; ECHO $(x:T) ;

prints C:/Program Files/Borland


When invoking Windows-based tools from Cygwin it can be important to pass them true windows-style paths. The :W modifier, under Cygwin only, turns a cygwin path into a Win32 path using the cygwin_conv_to_win32_path function. For example

x = "/cygdrive/c/Program Files/Borland" ; ECHO $(x:W) ;

prints C:\Program Files\Borland on Cygwin

Similarly, when used on OpenVMS, the :W modifier translates a POSIX-style path into native VMS-style format using decc$to_vms CRTL function. This modifier is generally used inside action blocks to properly specify file paths in VMS-specific commands. For example

x = "subdir/filename.c" ; ECHO $(x:W) ;

prints [.subdir]filename.c on OpenVMS

On other platforms, the string is unchanged.


Select the components listed in chars.


Replace grist with grist.


Replace directory with path.


Replace the base part of file name with base.


Replace the suffix of file name with suf.


Replace the archive member name with mem.


Prepend root to the whole file name, if not already rooted.


Assign value to the variable if it is unset.


Concatentate list elements into single element, separated by joinval.

On VMS, $(var:P) is the parent directory of $(var:D).

Local For Loop Variables

Boost Jam allows you to declare a local for loop control variable right in the loop:

x = 1 2 3 ;
y = 4 5 6 ;
for local y in $(x)
    ECHO $(y) ; # prints "1", "2", or "3"
ECHO $(y) ;     # prints "4 5 6"
Generated File Expansion

During expansion of expressions b2 also looks for subexpressions of the form @(filename:E=filecontents) and replaces the expression with filename after creating the given file with the contents set to filecontents. This is useful for creating compiler response files, and other "internal" files. The expansion works both during parsing and action execution. Hence it is possible to create files during any of the three build phases.

Built-in Variables

This section discusses variables that have special meaning to b2. All of these must be defined or used in the global module — using those variables inside a named module will not have the desired effect. See Modules.


These two variables control the binding of file target names to locations in the file system. Generally, $(SEARCH) is used to find existing sources while $(LOCATE) is used to fix the location for built targets.

Rooted (absolute path) file targets are bound as is. Unrooted file target names are also normally bound as is, and thus relative to the current directory, but the settings of $(LOCATE) and $(SEARCH) alter this:

  • If $(LOCATE) is set then the target is bound relative to the first directory in $(LOCATE). Only the first element is used for binding.

  • If $(SEARCH) is set then the target is bound to the first directory in $(SEARCH) where the target file already exists.

  • If the $(SEARCH) search fails, the target is bound relative to the current directory anyhow.

Both $(SEARCH) and $(LOCATE) should be set target-specific and not globally. If they were set globally, b2 would use the same paths for all file binding, which is not likely to produce sane results. When writing your own rules, especially ones not built upon those in Jambase, you may need to set $(SEARCH) or $(LOCATE) directly. Almost all of the rules defined in Jambase set $(SEARCH) and $(LOCATE) to sensible values for sources they are looking for and targets they create, respectively.


These two variables control header file scanning. $(HDRSCAN) is an egrep(1) pattern, with ()'s surrounding the file name, used to find file inclusion statements in source files. Jambase uses $(HDRPATTERN) as the pattern for $(HDRSCAN). $(HDRRULE) is the name of a rule to invoke with the results of the scan: the scanned file is the target, the found files are the sources. This is the only place where b2 invokes a rule through a variable setting.

Both $(HDRSCAN) and $(HDRRULE) must be set for header file scanning to take place, and they should be set target-specific and not globally. If they were set globally, all files, including executables and libraries, would be scanned for header file include statements.

The scanning for header file inclusions is not exact, but it is at least dynamic, so there is no need to run something like makedepend(GNU) to create a static dependency file. The scanning mechanism errs on the side of inclusion (i.e., it is more likely to return filenames that are not actually used by the compiler than to miss include files) because it can’t tell if #include lines are inside #ifdefs or other conditional logic. In Jambase, HdrRule applies the NOCARE rule to each header file found during scanning so that if the file isn’t present yet doesn’t cause the compilation to fail, b2 won’t care.

Also, scanning for regular expressions only works where the included file name is literally in the source file. It can’t handle languages that allow including files using variable names (as the Jam language itself does).


It is sometimes desirable to disallow parallel execution of some actions. For example:

  • Old versions of yacc use files with fixed names. So, running two yacc actions is dangerous.

  • One might want to perform parallel compiling, but not do parallel linking, because linking is i/o bound and only gets slower.

Craig McPeeters has extended Perforce Jam to solve such problems, and that extension was integrated in Boost.Jam.

Any target can be assigned a semaphore, by setting a variable called SEMAPHORE on that target. The value of the variable is the semaphore name. It must be different from names of any declared target, but is arbitrary otherwise.

The semantic of semaphores is that in a group of targets which have the same semaphore, only one can be updated at the moment, regardless of -j option.

Platform Identifier

A number of Jam built-in variables can be used to identify runtime platform:


OS identifier string


Underlying architecture, when applicable


true on MAC platform


true on NT platform


true on OS2 platform


true on Unix platforms


true on VMS platform

Jam Version

Time and date at b2 start-up as an ISO-8601 UTC value.


Ouput of uname(1) command (Unix only)


b2 version, currently "3.1.19"


A predefined global variable with two elements indicates the version number of Boost Jam. Boost Jam versions start at 03 00. Earlier versions of Jam do not automatically define JAM_VERSION.


When b2 executes a rule’s action block, it forks and execs a shell, passing the action block as an argument to the shell. The invocation of the shell can be controlled by $(JAMSHELL). The default on Unix is, for example:

JAMSHELL = /bin/sh -c % ;

The % is replaced with the text of the action block.

B2 does not directly support building in parallel across multiple hosts, since that is heavily dependent on the local environment. To build in parallel across multiple hosts, you need to write your own shell that provides access to the multiple hosts. You then reset $(JAMSHELL) to reference it.

Just as b2 expands a % to be the text of the rule’s action block, it expands a ! to be the multi-process slot number. The slot number varies between 1 and the number of concurrent jobs permitted by the -j flag given on the command line. Armed with this, it is possible to write a multiple host shell. For example:


# This sample JAMSHELL uses the SunOS on(1) command to execute a
# command string with an identical environment on another host.

# Set JAMSHELL = jamshell ! %
# where jamshell is the name of this shell file.
# This version handles up to -j6; after that they get executed
# locally.

case $1 in
1|4) on winken sh -c "$2";;
2|5) on blinken sh -c "$2";;
3|6) on nod sh -c "$2";;
*) eval "$2";;

The __TIMING_RULE__ and __ACTION_RULE__ can be set to the name of a rule for b2 to call after an action completes for a target. They both give diagnostic information about the action that completed. For __TIMING_RULE__ the rule is called as:

rule timing-rule ( args * : target : start end user system )

And __ACTION_RULE__ is called as:

rule action-rule ( args * : target : command status start end user system : output ? )

The arguments for both are:


Any values following the rule name in the __TIMING_RULE__ or __ACTION_RULE__ are passed along here.


The b2 target that was built.


The text of the executed command in the action body.


The integer result of the executed command.


The starting timestamp of the executed command as a ISO-8601 UTC value.


The completion timestamp of the executed command as a ISO-8601 UTC value.


The number of user CPU seconds the executed command spent as a floating point value.


The number of system CPU seconds the executed command spent as a floating point value.


The output of the command as a single string. The content of the output reflects the use of the -pX option.

If both variables are set for a target both are called, first __TIMING_RULE__ then __ACTION_RULE__.

11.2.7. Modules

Boost Jam introduces support for modules, which provide some rudimentary namespace protection for rules and variables. A new keyword, module was also introduced. The features described in this section are primitives, meaning that they are meant to provide the operations needed to write Jam rules which provide a more elegant module interface.

module expression { ... }

Code within the { …​ } executes within the module named by evaluating expression. Rule definitions can be found in the module’s own namespace, and in the namespace of the global module as module-name.rule-name, so within a module, other rules in that module may always be invoked without qualification:

module my_module
    rule salute ( x ) { ECHO $(x), world ; }
    rule greet ( ) { salute hello ; }
    greet ;
my_module.salute goodbye ;

When an invoked rule is not found in the current module’s namespace, it is looked up in the namespace of the global module, so qualified calls work across modules:

module your_module
    rule bedtime ( ) { my_module.salute goodnight ; }
Variable Scope

Each module has its own set of dynamically nested variable scopes. When execution passes from module A to module B, all the variable bindings from A become unavailable, and are replaced by the bindings that belong to B. This applies equally to local and global variables:

module A
    x = 1 ;
    rule f ( )
        local y = 999 ; # becomes visible again when B.f calls A.g
        B.f ;
    rule g ( )
        ECHO $(y) ;     # prints "999"
module B
    y = 2 ;
    rule f ( )
        ECHO $(y) ; # always prints "2"
        A.g ;

The only way to access another module’s variables is by entering that module:

rule peek ( module-name ? : variables + )
    module $(module-name)
        return $($(>)) ;

Note that because existing variable bindings change whenever a new module scope is entered, argument bindings become unavailable. That explains the use of $(>) in the peek rule above.

Local Rules
local rule rulename...

The rule is declared locally to the current module. It is not entered in the global module with qualification, and its name will not appear in the result of:

[ RULENAMES module-name ]
rule RULENAMES ( module ? )

Returns a list of the names of all non-local rules in the given module. If module is omitted, the names of all non-local rules in the global module are returned.

rule VARNAMES ( module ? )

Returns a list of the names of all variable bindings in the given module. If module is omitted, the names of all variable bindings in the global module are returned.

This includes any local variables in rules from the call stack which have not returned at the time of the VARNAMES invocation.

IMPORT allows rule name aliasing across modules:

rule IMPORT ( source_module ? : source_rules *
            : target_module ? : target_rules * )

The IMPORT rule copies rules from the source_module into the target_module as local rules. If either source_module or target_module is not supplied, it refers to the global module. source_rules specifies which rules from the source_module to import; target_rules specifies the names to give those rules in target_module. If source_rules contains a name which doesn’t correspond to a rule in source_module, or if it contains a different number of items than target_rules, an error is issued. For example,

# import m1.rule1 into m2 as local rule m1-rule1.
IMPORT m1 : rule1 : m2 : m1-rule1 ;
# import all non-local rules from m1 into m2
IMPORT m1 : [ RULENAMES m1 ] : m2 : [ RULENAMES m1 ] ;

EXPORT allows rule name aliasing across modules:

rule EXPORT ( module ? : rules * )

The EXPORT rule marks rules from the source_module as non-local (and thus exportable). If an element of rules does not name a rule in module, an error is issued. For example,

module X {
  local rule r { ECHO X.r ; }
IMPORT X : r : : r ; # error - r is local in X
EXPORT X : r ;
IMPORT X : r : : r ; # OK.
rule CALLER_MODULE ( levels ? )

CALLER_MODULE returns the name of the module scope enclosing the call to its caller (if levels is supplied, it is interpreted as an integer number of additional levels of call stack to traverse to locate the module). If the scope belongs to the global module, or if no such module exists, returns the empty list. For example, the following prints "{Y} {X}":

module X {
    rule get-caller { return [ CALLER_MODULE ] ; }
    rule get-caller's-caller { return [ CALLER_MODULE 1 ] ; }
    rule call-Y { return Y.call-X2 ; }
module Y {
    rule call-X { return X.get-caller ; }
    rule call-X2 { return X.get-caller's-caller ; }
callers = [ X.get-caller ] [ Y.call-X ] [ X.call-Y ] ;
ECHO {$(callers)} ;
rule DELETE_MODULE ( module ? )

DELETE_MODULE removes all of the variable bindings and otherwise-unreferenced rules from the given module (or the global module, if no module is supplied), and returns their memory to the system.

Though it won’t affect rules that are currently executing until they complete, DELETE_MODULE should be used with extreme care because it will wipe out any others and all variable (including locals in that module) immediately. Because of the way dynamic binding works, variables which are shadowed by locals will not be destroyed, so the results can be really unpredictable.

11.3. Miscellaneous

11.3.1. Diagnostics

In addition to generic error messages, b2 may emit one of the following:

warning: unknown rule X

A rule was invoked that has not been defined with an actions or rule statement.

using N temp target(s)

Targets marked as being temporary (but nonetheless present) have been found.

updating N target(s)

Targets are out-of-date and will be updated.

can't find N target(s)

Source files can’t be found and there are no actions to create them.

can't make N target(s)

Due to sources not being found, other targets cannot be made.

warning: X depends on itself

A target depends on itself either directly or through its sources.

don't know how to make X

A target is not present and no actions have been defined to create it.

X skipped for lack of Y

A source failed to build, and thus a target cannot be built.

warning: using independent target X

A target that is not a dependency of any other target is being referenced with $(<) or $(>).

X removed

B2 removed a partially built target after being interrupted.

11.3.2. Bugs, Limitations

For parallel building to be successful, the dependencies among files must be properly spelled out, as targets tend to get built in a quickest-first ordering. Also, beware of un-parallelizable commands that drop fixed-named files into the current directory, like yacc(1) does.

A poorly set $(JAMSHELL) is likely to result in silent failure.

11.3.3. Fundamentals

This section is derived from the official Jam documentation and from experience using it and reading the Jambase rules. We repeat the information here mostly because it is essential to understanding and using Jam, but is not consolidated in a single place. Some of it is missing from the official documentation altogether. We hope it will be useful to anyone wishing to become familiar with Jam and the Boost build system.

  • Jam rules are actually simple procedural entities. Think of them as functions. Arguments are separated by colons.

  • A Jam target is an abstract entity identified by an arbitrary string. The build-in DEPENDS rule creates a link in the dependency graph between the named targets.

  • Note that the original Jam documentation for the built-in INCLUDES rule is incorrect: INCLUDES targets1 : targets2 causes everything that depends on a member of targets1 to depend on all members of targets2. It does this in an odd way, by tacking targets2 onto a special tail section in the dependency list of everything in targets1. It seems to be OK to create circular dependencies this way; in fact, it appears to be the "right thing to do" when a single build action produces both targets1 and targets2.

  • When a rule is invoked, if there are actions declared with the same name as the rule, the actions are added to the updating actions for the target identified by the rule’s first argument. It is actually possible to invoke an undeclared rule if corresponding actions are declared: the rule is treated as empty.

  • Targets (other than NOTFILE targets) are associated with paths in the file system through a process called binding. Binding is a process of searching for a file with the same name as the target (sans grist), based on the settings of the target-specific SEARCH and LOCATE variables.

  • In addition to local and global variables, jam allows you to set a variable on a target. Target-specific variable values can usually not be read, and take effect only in the following contexts:

    • In updating actions, variable values are first looked up on the target named by the first argument (the target being updated). Because Jam builds its entire dependency tree before executing actions, Jam rules make target-specific variable settings as a way of supplying parameters to the corresponding actions.

    • Binding is controlled entirely by the target-specific setting of the SEARCH and LOCATE variables, as described here.

    • In the special rule used for header file scanning, variable values are first looked up on the target named by the rule’s first argument (the source file being scanned).

  • The "bound value" of a variable is the path associated with the target named by the variable. In build actions, the first two arguments are automatically replaced with their bound values. Target-specific variables can be selectively replaced by their bound values using the bind action modifier.

  • Note that the term "binding" as used in the Jam documentation indicates a phase of processing that includes three sub-phases: binding (yes!), update determination, and header file scanning. The repetition of the term "binding" can lead to some confusion. In particular, the Modifying Binding section in the Jam documentation should probably be titled "Modifying Update Determination".

  • "Grist" is just a string prefix of the form <characters>. It is used in Jam to create unique target names based on simpler names. For example, the file name test.exe may be used by targets in separate subprojects, or for the debug and release variants of the "same" abstract target. Each distinct target bound to a file called "test.exe" has its own unique grist prefix. The Boost build system also takes full advantage of Jam’s ability to divide strings on grist boundaries, sometimes concatenating multiple gristed elements at the beginning of a string. Grist is used instead of identifying targets with absolute paths for two reasons:

    1. The location of targets cannot always be derived solely from what the user puts in a Jamfile, but sometimes depends also on the binding process. Some mechanism to distinctly identify targets with the same name is still needed.

    2. Grist allows us to use a uniform abstract identifier for each built target, regardless of target file location (as allowed by setting ALL_LOCATE_TARGET).

  • When grist is extracted from a name with $(var:G), the result includes the leading and trailing angle brackets. When grist is added to a name with $(var:G=expr), existing grist is first stripped. Then, if expr is non-empty, leading <s and trailing >s are added if necessary to form an expression of the form <expr2>; <expr2> is then prepended.

  • When Jam is invoked it imports all environment variable settings into corresponding Jam variables, followed by all command-line (-s…​) variable settings. Variables whose name ends in PATH, Path, or path are split into string lists on OS-specific path-list separator boundaries (e.g. ":" for UNIX and ";" for Windows). All other variables are split on space (" ") boundaries. Boost Jam modifies that behavior by allowing variables to be quoted.

  • A variable whose value is an empty list or which consists entirely of empty strings has a negative logical value. Thus, for example, code like the following allows a sensible non-empty default which can easily be overridden by the user:

    MESSAGE ?\= starting jam... ;
    if $(MESSAGE) { ECHO The message is: $(MESSAGE) ; }

    If the user wants a specific message, he invokes jam with -sMESSAGE=message text. If he wants no message, he invokes jam with -sMESSAGE= and nothing at all is printed.

  • The parsing of command line options in Jam can be rather unintuitive, with regards to how other Unix programs accept options. There are two variants accepted as valid for an option:

    1. -xvalue, and

    2. -x value.

11.4. History

11.4.1. Version 3.1.18

After years of bjam developments.. This is going to be the last unbundled release of the 3.1.x series. From this point forward bjam will only be bundled as part of the larger Boost Build system. And hence will likely change name at some point. As a side effect of this move people will get more frequent release of bjam (or whatever it ends up being called).

  • New built-ins, MD5, SPLIT_BY_CHARACTERS, PRECIOUS, PAD, FILE_OPEN, and UPDATE_NOW. — Vladimir P.

  • Ensure all file descriptors are closed when executing actions complete on *nix. — Noel B.

  • Fix warnings, patch from Mateusz Loskot. — Vladimir P.

  • Add KEEP_GOING var to programatically override the '-q' option. — Vladimir P.

  • Add more parameters, up to 19 from 9, to rule invocations. Patch from Jonathan Biggar. — Vladimir P.

  • Print failed command output even if the normally quite '-d0' option. — Vladimir P.

  • Build of bjam with vc10, aka Visual Studio 2010. — Vladimir P.

  • More macros for detection of OSPLAT, patch from John W. Bito. — Vladimir P.

  • Add PARALLELISM var to programatically override the '-j' option. — Vladimir P.

  • Tweak doc building to allow for PDF generation of docs. — John M.

11.4.2. Version 3.1.17

A year in the making this release has many stability improvements and various performance improvements. And because of the efforts of Jurko the code is considerably more readable!

  • Reflect the results of calling bjam from Python. — Rene R.

  • For building on Windows: Rework how arguments are parsed and tested to fix handling of quoted arguments, options arguments, and arguments with "=". — Rene R.

  • Try to work around at least one compiler bug with GCC and variable aliasing that causes crashes with hashing file cache entries. — Rene R.

  • Add -Wc,-fno-strict-aliasing for QCC/QNX to avoid the same aliasing crashes as in the general GCC 4.x series (thanks to Niklas Angare for the fix). — Rene R.

  • On Windows let the child bjam commands inherit stdin, as some commands assume it’s available. — Rene R.

  • On Windows don’t limit bjam output to ASCII as some tools output characters in extended character sets. — Rene R.

  • Isolate running of bjam tests to individual bjam instances to prevent possible spillover errors from one test affecting another test. Separate the bjam used to run the tests vs. the bjam being tested. And add automatic re-building of the bjam being tested. — Rene R.

  • Fix some possible overrun issues revealed by Fortify build. Thanks to Steven Robbins for pointing out the issues. — Rene R.

  • Handle \n and \r escape sequences. — Vladimir P.

  • Minor edits to remove -Wall warnings. — Rene R.

  • Dynamically adjust pwd buffer query size to allow for when PATH_MAX is default defined instead of being provided by the system C library. — Rene R.

  • Minor perf improvement for bjam by replacing hash function with faster version. Only 1% diff for Boost tree. — Rene R.

  • Updated Boost Jam’s error location reporting when parsing Jamfiles. Now it reports the correct error location information when encountering an unexpected EOF. It now also reports where an invalid lexical token being read started instead of finished which makes it much easier to find errors like unclosed quotes or curly braces. — Jurko G.

  • Removed the -xarch=generic architecture from build.jam as this option is unknown so the Sun compilers on Linux. — Noel B.

  • Fixed a bug with T_FATE_ISTMP getting reported as T_FATE_ISTMP & T_FATE_NEEDTMP at the same time due to a missing break in a switch statement. — Jurko G.

  • Fixed a Boost Jam bug causing it to sometimes trigger actions depending on targets that have not been built yet. — Jurko G.

  • Added missing documentation for Boost Jam’s :T variable expansion modifier which converts all back-slashes ('\') to forward slashed ('/'). — Jurko G.

  • Added Boost Jam support for executing command lines longer than 2047 characters (up to 8191) characters when running on Windows XP or later OS version. — Jurko G.

  • Fixed a Boost Jam bug on Windows causing its SHELL command not to work correctly with some commands containing quotes. — Jurko G.

  • Corrected a potential memory leak in Boost Jam’s builtin_shell() function that would appear should Boost Jam ever start to release its allocated string objects. — Jurko G.

  • Made all Boost Jam’s ECHO commands automatically flush the standard output to make that output more promptly displayed to the user. — Jurko G.

  • Made Boost Jam tests quote their bjam executable name when calling it allowing those executables to contain spaces in their name and/or path. — Jurko G.

  • Change execunix.c to always use fork() instead of vfork() on the Mac. This works around known issues with bjam on PPC under Tiger and a problem reported by Rene with bjam on x86 under Leopard. — Noel B.

  • Corrected a bug in Boost Jam’s base Jambase script causing it to trim the error message displayed when its boost-build rule gets called multiple times. — Jurko G.

  • When importing from Python into an module with empty string as name, import into root module. — Vladimir P.

  • Patch for the NORMALIZE_PATH builtin Boost Jam rule as well as an appropriate update for the path.jam Boost Build module where that rule was being used to implement path join and related operations. — Jurko G.

  • Fixed a bug causing Boost Jam not to handle target file names specified as both short and long file names correctly. — Jurko G.

  • Relaxed test, ignoring case of drive letter. — Roland S.

  • Implemented a patch contributed by Igor Nazarenko reimplementing the list_sort() function to use a C qsort() function instead of a hand-crafted merge-sort algorithm. Makes some list sortings (e.g. 1,2,1,2,1,2,1,2, …​) extremely faster, in turn significantly speeding up some project builds. — Jurko G.

  • Fixed a bug with bjam not handling the '' root Windows path correctly without its drive letter being specified. — Jurko G.

  • Solved the problem with child process returning the value 259 (Windows constant STILL_ACTIVE) causing bjam never to detect that it exited and therefore keep running in an endless loop. — Jurko G.

  • Solved the problem with bjam going into an active wait state, hogging up processor resources, when waiting for one of its child processes to terminate while not all of its available child process slots are being used. — Jurko G.

  • Solved a race condition between bjam’s output reading/child process termination detection and the child process’s output generation/termination which could have caused bjam not to collect the terminated process’s final output. — Jurko G.

  • Change from vfork to fork for executing actions on Darwin to improve stability. — Noel B.

  • Code reformatting and cleanups. — Jurko G.

  • Implement ISFILE built-in. — Vladimir P.

11.4.3. Version 3.1.16

This is mostly a bug fix release.

  • Work around some Windows CMD.EXE programs that will fail executing a totally empty batch file. — Rene R.

  • Add support for detection and building with vc9. — John P.

  • Plug memory leak when closing out actions. Thanks to Martin Kortmann for finding this. — _Rene R.

  • Various improvements to __TIMING_RULE__ and __ACTION_RULE__ target variable hooks. — Rene R.

  • Change JAMDATE to use common ISO date format. — Rene R.

  • Add test for result status values of simple actions, i.e. empty actions. — Rene R.

  • Fix buffer overrun bug in expanding @() subexpressions. — Rene R.

  • Check empty string invariants, instead of assuming all strings are allocated. And reset strings when they are freed. — Rene R.

  • Add OSPLAT=PARISC for HP-UX PA-RISC. — Boris G.

  • Make quietly actions really quiet by not printing the command output. The output for the quietly actions is still available through __ACTION_RULE__. — Rene R.

  • Switch intel-win32 to use static multi thread runtime since the single thread static runtime is no longer available. — Rene R.

  • When setting OSPLAT, check _ia64 macro. — _Boris G.

  • Get the unix timing working correctly. — Noel B.

  • Add -fno-strict-aliasing to compilation with gcc. Which works around GCC-4.2 crash problems. — Boris G.

  • Increased support for Python integration. — Vladimir P., Daniel W.

  • Allow specifying options with quotes, i.e. --with-python=xyz, to work around the CMD shell using = as an argument separator. — Rene R.

  • Add values of variables specified with -s to .EVNRION module, so that we can override environment on command line. — Vladimir P.

  • Make NORMALIZE_PATH convert \ to /. — Vladimir P.

11.4.4. Version 3.1.15

This release sees a variety of fixes for long standing Perforce/Jam problems. Most of them relating to running actions in parallel with the -jN option. The end result of the changes is that running parallel actions is now reliably possible in Unix and Windows environments. Many thanks to Noel for joining the effort, to implement and fix the Unix side of stuff.

  • Add support for building bjam with pgi and pathscale toolsets. — Noel B.

  • Implement running action commands through pipes (-p option) to fix jumbled output when using parallel execution with -j option. This is implemented for Unix variants, and Windows (Win32/NT). — Rene R., Noel B.

  • Add "sun" as alias to Sun Workshop compiler tools. — Rene R.

  • Set MAXLINE in jam.h to 23k bytes for AIX. The piecemeal archive action was broken with the default MAXLINE of 102400. Because the AIX shell uses some of the 24k default buffer size for its own use, I reduced it to 23k. — Noel B.

  • Make use of output dir options of msvc to not polute src dir with compiled files. — Rene R.

  • A small fix, so -d+2 will always show the "real" commands being executed instead of casually the name of a temporary batch file. — Roland S.

  • Add test to check 'bjam -n'. — Rene R.

  • Add test to check 'bjam -d2'. — Rene R.

  • Bring back missing output of -n option. The -o option continues to be broken as it has been for a long time now because of the @ file feature. — Rene R.

  • Update GC support to work with Boehm GC 7.0. — Rene R.

  • Revert the BOOST_BUILD_PATH change, since the directory passed to boost-build should be first in searched paths, else project local build system will not be picked correctly. The order had been changed to allow searching of alternate user-config.jam files from boost build. This better should be done with --user-config= switch or similar. — Roland S.

  • Initial support for defining action body from Python. — Vladimir P.

  • Implement @() expansion during parse phase. — Rene R.

  • Define OSPLAT var unconditionally, and more generically, when possible. — Rene R.

  • Fix undeclared INT_MAX on some platforms, i.e. Linux. — Rene R.

  • Modified execunix.c to add support for terminating processes that consume too much cpu or that hang and fail to consume cpu at all. This in support of the bjam -lx option. — Noel B.

  • Add internal dependencies for multi-file generating actions to indicate that the targets all only appear when the first target appears. This fixes the long standing problem Perforce/Jam has with multi-file actions and parallel execution (-jN). — Rene R.

  • Add test of -l limit option now that it’s implemented on windows and unix. — Rene R.

  • Add test for no-op @() expansion. — Rene R.

  • Handle invalid formats of @() as doing a straight substitution instead of erroring out. — Rene R.

  • Various fixes to compile on SGI/Irix. — Noel B.

  • Add output for when actions timeout with -lN option. — Rene R., Noel B.

  • Add needed include (according to XOPEN) for definition of WIFEXITED and WEXITSTATUS. — Markus S.

1. See the section called “Feature Attributes”
2. Many features will be overridden, rather than added-to, in subprojects. See the section called “Feature Attributes” for more information
3. see the definition of "free" in the section called “Feature Attributes”.
4. This name is historic, and will be eventually changed to metatarget
5. This create-then-register pattern is caused by limitations of the Boost.Jam language. Python port is likely to never create duplicate targets.