Boris Schäling

20 July 2009

Boost.Build: Building C++ projects

Boost.Build is a high-level build system to manage C++ and C projects. It is based on configuration files which are simple text files. In contrary to other build systems configuration files are compiler- and platform-independent and contain only as much information as absolutely required for a build system to know how to generate binaries. As Boost.Build supports many compilers and platforms out of the box and translates options in configuration files to command line options expected by a selected compiler configuration files are written once and used everywhere. Programs implemented in standard C++ or standard C can be easily built with Boost.Build on different platforms no matter what compiler is used.

This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 Unported License.

Table of Contents

1. Introduction

Compiler- and platform-independent build system

Boost.Build is a high-level build system which makes it as easy as possible to manage C++ projects. The idea is to specify in configuration files just as much as necessary to build a program. For example it is not required to tell Boost.Build how to use a certain compiler. Boost.Build supports many compilers out of the box and knows how to use them. If you create a configuration file you just need to tell Boost.Build where to find the source files, what the executable should be called and which compiler Boost.Build should use. Boost.Build will then try to find the compiler and automatically build the program.

As Boost.Build supports many compilers configuration files never contain any compiler-specific options. Configuration files are entirely compiler-independent. Of course it is possible to set options like whether code should be optimized. However these options are written in a language only understood by Boost.Build. Once a compiler is picked to build a program Boost.Build translates options in configuration files to command line options expected by the selected compiler. This makes it possible to write configuration files once and build a program on different platforms with different compilers.

As nice as it sounds Boost.Build can only be used for C++ and C projects. Boost.Build doesn't know how to use other compilers like a Java compiler. Although Boost.Build is extensible it makes more sense to use a different build system for programs implemented in other programming languages.

Boost.Build was created to build and install the Boost C++ libraries easily with different compilers on different platforms. Although Boost.Build is part of and shipped with the Boost C++ libraries it can be used separately for any C++ or C project. It's even possible to download only Boost.Build in case you don't want to use the Boost C++ libraries.

This article is an introduction to help you using Boost.Build for your own C++ or C projects. It gives you a basic understanding of how Boost.Build works and how you start using it. After reading the article you should not only be able to use Boost.Build for your own projects. If will also be easier to understand the Boost.Build documentation as you'll know the big picture.

2. Build process

Jamfiles and an interpreter called bjam

The program you use to build a project managed by Boost.Build is called bjam. If you downloaded and built the Boost C++ libraries you have used bjam already. bjam looks for configuration files, reads them and builds a project accordingly. It also accepts various command line options which can be useful for example to show all commands executed by bjam to build a project.

Projects can be large and can consist of many components whose source code is distributed over many directories. Instead of creating one big configuration file for the entire project components typically get their own configuration files. This is no different with Boost.Build: In a large project there will be many configuration files which have to be found and interpreted by bjam.

For Boost.Build every directory with a configuration file is a project: If there is a configuration file in a directory something can be built. Whether it's a component in a subdirectory or a software consisting of many components doesn't make a difference for Boost.Build.

When bjam is started it doesn't run a search for configuration files on the entire file system. It searches for a configuration file in the current working directory only. If it doesn't find a configuration file it doesn't do anything. bjam does not search for configuration files in any other directory if there is no configuration file in the current working directory.

The configuration file bjam is looking for is called Jamfile.jam. Files with the extension jam are called Jamfiles. If bjam finds a Jamfile in the current working directory it searches for more Jamfiles in parent directories. bjam climbs up parent directories until it finds a configuration file called Jamroot.jam. Jamroot.jam is no different from Jamfile.jam. It only indicates that bjam doesn't need to look further.

The reason why bjam looks for Jamfiles in parent directories is that it makes it possible to group settings. If there are some components which should be built with similar settings they can be stored in a Jamfile in a parent directory which will be automatically used if a component in a subdirectory is built.

Please note that bjam must find a file called Jamroot.jam. It is an error if no Jamroot.jam exists. If Jamroot.jam is in the current working directory no other file Jamfile.jam is required. If Jamroot.jam is in a parent directory a file Jamfile.jam must exist in the current working directory - otherwise bjam doesn't do anything.

If you copy bjam to a directory which contains no Jamfiles and start the program you get an error message. However bjam doesn't complain that it can't find a Jamfile. It complains about not finding the build system.

Unable to load Boost.Build: could not find "boost-build.jam"
Attempted search from C:\Users\Boris\Desktop up to the root

Please consult the documentation at ''.

The first thing bjam does is not looking for a Jamfile but loading the build system. But what exactly is the build system?

bjam is an interpreter. It doesn't really know how to build anything. What bjam does is interpreting Jamfiles. Boost.Build is really implemented in Jamfiles. And they contain all the logic which makes Boost.Build such a powerful tool. As bjam only does what it reads in Jamfiles it needs to know where to find the Jamfiles Boost.Build is made of.

When bjam is started it looks for a file boost-build.jam in the current working directory. If it doesn't find the file it searches all parent directories. This file needs to contain only one line to tell bjam where to find the build system.

boost-build C:/boost_1_39_0/tools/build/v2 ; 

The path after boost-build must refer to a directory which contains a file called bootstrap.jam. This is the file bjam needs to load the build system. As the Boost C++ libraries ship Boost.Build you can refer to the subdirectory tools/build/v2 of the root directory of the Boost C++ libraries. And you can always use a slash as a path separator - even if you are on Windows.

Please note that there must be a space between the path and the semicolon at the end of the line. It is an error if the space is missing. You'll learn more about the syntax used in Jamfiles later in this article.

If bjam finds boost-build.jam it uses the path within the file to load the build system. When the build system is loaded it also prepares itself to use a certain compiler, linker and maybe other tools required to build a project. Boost.Build refers to these programs as a toolset. If no command line option is used to start bjam the build system tries to find a toolset it can use automatically. On Windows for example it searches for Visual C++. And if it detects that Visual C++ is installed it uses the toolset msvc.

warning: No toolsets are configured.
warning: Configuring default toolset "msvc".
warning: If the default is wrong, your build may not work correctly.
warning: Use the "toolset=xxxxx" option to override our guess.
warning: For more configuration options, please consult

If you start bjam without specifying which toolset should be used you see a warning. bjam tells you which toolset it detected and decided to use. If you want to suppress the warning you must specify the toolset yourself. For example you tell the build system to use Visual C++ with bjam toolset=msvc. If you want GCC to be used you enter bjam toolset=gcc.

As of today there are more than 10 toolsets supported. There is a good chance that Boost.Build will work with the compiler you use out of the box.

Once the build system has been found, loaded and knows which toolset to use - either because you specified one or the build system detected one automatically - bjam looks for a file Jamfile.jam in the current working directory. If it doesn't find a Jamfile an error message is printed.

error: error: no Jamfile in current directory found, and no target references specified.

If you create an empty file Jamfile.jam and start bjam again another error message is printed.

error: Could not find parent for project at '.'
error: Did not find Jamfile.jam or Jamroot.jam in any parent directory.

bjam is ultimately looking for a Jamfile called Jamroot.jam. If it doesn't exist in the current working directory bjam expects to find it in a parent directory.

If you create an empty file Jamroot.jam and start bjam the error message is gone. Obviously there is nothing done by Boost.Build. But now you know how bjam proceeds to build a program and what the minimum Boost.Build configuration looks like.

Please note that if you work on a small project and you need only one configuration file you can simply call it Jamroot.jam. You don't need another file called Jamfile.jam.

3. Basic tasks

Rules and features

If you look at Jamfiles the syntax might remind you of configuration files used by other build systems. Simple Jamfiles can look like plain old configuration files where for example values seem to be assigned to keys. What is important to understand though is that Jamfiles are really script files. There is a programming language used to write Jamfiles. bjam isn't the core component of Boost.Build which knows how to build programs. The logic of Boost.Build is in the Jamfiles which tell bjam how to build programs.

Even though Boost.Build is based on a programming language you don't need to think of programming when you create Jamfiles. The syntax of the programming language used by Boost.Build tries to remind you more of creating plain old configuration files. The idea is to have the best of two worlds: A powerful and flexible programming language but a simple syntax you might be familiar with from other build systems.

This article doesn't introduce you into the programming language Boost.Build is based on. The programming language is proprietary and not really a joy to use. It is no competitor to popular scripting languages like Javascript or Python. The developers of Boost.Build recognize it and work on another version of Boost.Build based on Python. However all of this shouldn't matter to developers who plan to manage their projects with Boost.Build. It helps to understand the syntax of Jamfiles better once one realizes that there is a programming language inside Boost.Build. But it's not required to learn the details of the programming language.

Let's look at a simple Jamfile which can be used to build an executable hello from a source file hello.cpp.

exe hello : hello.cpp ; 

Boost.Build provides a lot of built-in rules and exe is one of them. While the documentation of Boost.Build refers to exe as a rule you know already that the above Jamfile is actually built using a programming language. As it turns out rules are simply functions. And the Jamfile above contains a function call.

For the majority of tasks which are typically required to build programs Boost.Build provides predefined rules - or functions if you like. As with functions in other programming languages it is possible to pass parameters. In the Jamfile above the function exe is called with the two parameters hello and hello.cpp.

The programming language Boost.Build is based on knows only one data type: Everything is a list of strings. A list can be empty or contain one or more strings. In the Jamfile above the function exe is called with two parameters each one a list containing one string.

exe "hello" : "hello.cpp" ; 

It is possible to use quotes. It's not necessary though as after all every item in a list has the data type string anyway. Quotes are only used if parameters contain spaces.

While there is no special delimiter between a rule and the first parameter a colon must be used to separate other parameters. It is also required to end a line with a semicolon just as you are used to from C++.

Please note that the programming language of Boost.Build requires that there is a space around all tokens. For example there must be a space on the left and on the right of the colon and there must be a space on the left of the semicolon. Without spaces around tokens bjam won't be able to parse Jamfiles correctly.

If bjam is run in a directory which contains the Jamfile above and a source file hello.cpp, and if the msvc toolset is used on Windows a subdirectory bin\msvc-9.0\debug is created to build an executable hello.exe.

...found 9 targets...
...updating 5 targets...
common.mkdir bin
common.mkdir bin\msvc-9.0
common.mkdir bin\msvc-9.0\debug
compile-c-c++ bin\msvc-9.0\debug\hello.obj
hello.cpp bin\msvc-9.0\debug\hello.exe
msvc.manifest bin\msvc-9.0\debug\hello.exe
...updated 5 targets...

As you see it takes only one line in a Jamfile to build an executable from a source file. And if the program is built on Windows there is even the correct file extension exe appended.

The main advantage of Boost.Build is that you specify just as much as necessary for a build system to know how to build a program. Anything Boost.Build can do automatically is done automatically. You don't need to detect the platform a program is built on to decide if a file extension like exe should be appended or not. And you don't need to specify how a compiler like Visual C++ has actually to be invoked to compile source code.

Boost.Build supports a lot of toolsets out of the box. As a program can be built using different toolsets Boost.Build uses toolset-specific directories. This way it is possible to build a program with different toolsets without a toolset constantly overwriting files produced by another toolset.

There are not only toolset-specific directories but also variant-specific directories. A variant is a debug or release version of a program. For each variant another directory is used to build a program - again for the reason not to overwrite files produced by another variant. By default the debug variant is used. That's why the subdirectory bin\msvc-9.0\debug was created. If you want a release version to be created you can specify the variant on the command line with bjam variant=release.

...found 9 targets...
...updating 5 targets...
common.mkdir bin
common.mkdir bin\msvc-9.0
common.mkdir bin\msvc-9.0\release
compile-c-c++ bin\msvc-9.0\release\hello.obj
hello.cpp bin\msvc-9.0\release\hello.exe
msvc.manifest bin\msvc-9.0\release\hello.exe
...updated 5 targets...

With the variant set to release the subdirectory bin\msvc-9.0\release is used to create the executable hello.exe.

Choosing a variant is something which is done so often that it's sufficient to enter bjam release. Boost.Build figures out that release is meant to choose the variant.

If you don't want to specify the variant on the command line but want to build release versions of hello.exe by default the Jamfile has to be changed.

exe hello : hello.cpp : <variant>release ; 

The exe rule - or if you want function - accepts a few more parameters which are optional. The third parameter is a list of requirements. You can think of command line options which are always set and passed to commands run to build an executable.

In order to force a release version to be built the variant has to be set to release just as it was done before on the command line. The syntax to set the variant in a Jamfile is different though.

Boost.Build defines features which look like XML tags. One of the features supported by Boost.Build is <variant>. If a feature should be set to a value it has to be put next to it - without a space in between. Some features are free which means they can be set to any value you want. <variant> is a non-free feature as it can only be set to debug or release. No other value is allowed. If another value is set bjam will report an error.

If you run bjam variant=debug and try to build a debug version of hello.exe it won't work as the Jamfile contains the requirement that hello.exe is built as a release version. If you want to be able to overwrite the feature on the command line you have to pass the feature as the fourth parameter instead of the third.

exe hello : hello.cpp : : <variant>release ; 

The fourth parameter contains features which are used by default but which can be overwritten.

If you want both a debug and a release version of hello.exe to be built by default the <variant> feature needs to be set twice to debug and release.

exe hello : hello.cpp : : <variant>debug <variant>release ; 

It is important that <variant> is set twice in the fourth parameter where default values are specified. If it was the third parameter where requirements are specified bjam would report an error. It is possible to set a feature multiple times in the requirements but only if values are not mutually exclusive. As a program can't be a debug and a release version at the same time <variant> must be set in the default values. Only then Boost.Build understands that two versions of hello.exe should be built.

exe hello : hello.cpp : <define>WIN32 <define>_WIN32 : <variant>debug <variant>release ; 

The above Jamfile is an example for setting a feature multiple times in the requirements. The feature <define> is used to define preprocessor directives. It is no problem to define several preprocessor directives. Thus there are now two versions of hello.exe built both with the two directives WIN32 and _WIN32 defined.

exe hello : hello.cpp : : <variant>debug <variant>release <define>WIN32 <define>_WIN32 ; 

If the definitions are moved to the fourth parameter and you run bjam you get the same two versions of hello.exe built with the two directives WIN32 and _WIN32. As <define> does not expect mutually exclusive values there is no other set of executables generated. The only difference between this Jamfile and the previous one is that directives passed in the fourth parameter are default values which can be dropped while anything passed as a third parameter is an immutable requirement.

Here is another example of a feature whose values are mutually exclusive.

exe hello : hello.cpp : : <variant>debug <variant>release <optimization>speed <optimization>off ; 

bjam creates four versions of hello.exe: A debug version optimized for speed, a debug version with no optimization, a release version optimized for speed and a release version with no optimization. All of these versions are built in seperate directories which are automatically created.

So far the only rule used was exe. But of course Boost.Build provides many more built-in rules. Another important rule is lib. It is used to build a library.

lib world : world.cpp ; 

The above Jamfile builds a shared library from the source file world.cpp. On Windows a file world.dll is created. The usual file extension is again automatically appended by Boost.Build.

By default a shared library is built. If you want a static library to be generated you set the <link> feature to static.

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

Another useful rule is install. After executables and libraries have been built this rule can be used to install them.

exe hello : hello.cpp ; 
install "C:/Program Files/hello" : hello ; 

The above Jamfile installs the executable hello.exe to the directory C:\Program Files\hello. The second parameter hello is a reference to the target hello defined in the first line. Please note that the path has to be put in quotes as it contains a space.

Here concepts known from other build systems shine through: Instead of thinking of function calls every line defines a target. Dependencies are created by referencing other targets. That's how Boost.Build knows in what order it should build targets.

Typically the rule install is written differently though. Instead of passing the installation directory as the first parameter a feature <location> is used to set the installation directory in the third parameter.

exe hello : hello.cpp ; 
install install-bin : hello : <location>"C:/Program Files/hello" ; 

The main reason why it's better to use <location> is that the first parameter always defines a target. Other rules might refer to a target. That's why it is a good idea to use target names which don't have to be changed later. Imagine a program should be installed to a different directory. It's easier to change the installation directory if the <location> feature has been used as no other rules which might refer to install-bin have to be updated.

There is another reason why it makes sense to use a feature. Boost.Build supports conditional properties which make it possible to use different installation directories depending on the platform a program is built on.

exe hello : hello.cpp ; 
install install-bin : hello : <target-os>windows:<location>"C:/Program Files/hello" <target-os>linux:<location>/usr/local/bin ; 

The feature <target-os> is another feature with mutually exclusive values. It can be set for example to windows or linux but not to both.

The feature <location> follows <target-os> only delimited by a colon. Such a construct is called conditional property: Boost.Build selects the installation directory depending on the operating system.

Of course conditional properties can also be used with other rules. It is for example possible to define different preprocessor directives depending on the variant when building a program or a library.

Boost.Build provides many more built-in rules. Another useful rule is glob which makes it possible to use wildcards. In a big project with many source files it's then not required to list them all one by one but refer to all of them with glob.

exe hello : [ glob *.cpp ] ; 

The above Jamfile contains a nested function call: The result of the rule glob is passed as the second parameter to exe. Due to requirements of the programming language Boost.Build is based on brackets must be used for nested function calls.

4. Project management

Multiple Jamfiles

In large projects with many Jamfiles it's necessary to connect Jamfiles somehow. There is typically a Jamroot.jam file in the project's root directory and many Jamfile.jam files in subdirectories. If bjam is run in the root directory developers probably expect that the entire project including all components in subdirectories is built. As bjam looks for Jamfiles in parent directories but not in subdirectories Jamfiles need to refer to Jamfiles in subdirectories explicitly.

build-project hello ; 

If a Jamfile looks like the sample above it refers to a Jamfile in a subdirectory hello. build-project is a rule which expects a path as its sole parameter. The path is then used to lookup a Jamfile.

build-project hello ; 
build-project world ; 

If you want several projects to be built you must use build-project multiple times.

Apart from referring to Jamfiles in subdirectories it makes also sense to group options which should be used when building components in a project.

project : default-build release ; 
build-project hello ; 
build-project world ; 

The project rule accepts various parameters to set options for the Jamfile in the current working directory and in subdirectories.

While other rules like exe and lib expect parameters to be passed in a certain order project uses named arguments. In the sample above the argument's name is default-build. That's why it is possible to pass the value release in a very different parameter.

project : : : : : : : : : default-build release ; 
build-project hello ; 
build-project world ; 

It doesn't make sense to pass release as the tenth parameter. But it works as project doesn't care about the order. As the tenth parameter is called default-build it is accepted.

project supports only a few named arguments. Another one is requirements which can be used to set options which can't be overwritten.

project : requirements <variant>release ; 
build-project hello ; 
build-project world ; 

The Jamfile above builds only release versions. It is not possible to build a debug version anymore as requirements can not be overwritten. That's the difference to the named argument called default-build which was used in the previous sample: It can be overwritten.

When build-project is used Boost.Build assumes that the parameter is a reference to a subdirectory. We had seen another type of reference before.

exe hello : hello.cpp ; 
install install-bin : hello : <location>"C:/Program Files/hello" ; 

In the above Jamfile the install rule refers to the target hello defined in the first line.

In a large project it might be necessary to refer to targets which are defined in Jamfiles in other directories. It is possible to concatenate a path to a Jamfile and a target with a double slash.

install install-bin : subdir//hello : <location>"C:/Program Files/hello" ; 

Now the install rule refers to a target hello in a Jamfile in the subdirectory subdir.

Let's assume that the executable hello depends on a library in another directory world. The library is also built with Boost.Build using the rule lib.

lib world : world.cpp ; 

In the Jamfile to build the executable a reference is required to the Jamfile of the library. It's not necessary to refer to the target world directly as all targets in a Jamfile are built by default.

exe hello : hello.cpp world : : <variant>debug <variant>release ; 

The above Jamfile assumes that the library and its Jamfile are in a subdirectory world.

When the executable is built there are two versions generated - a debug and a release version. The Jamfile of the library however doesn't set the <variant> feature. But Boost.Build assumes that it should build two versions of the library, too. The feature <variant> is said to be propagated.

Propagating features simplify project management as you don't need to set the same features in various Jamfiles. However it also makes it a bit more complicated to understand how components are built as it all depends on what features are propagated. You can assume that Boost.Build knows what it should do. But of course it doesn't mean that you easily understand what it does.

Let's look at another example using the feature <define>.

exe hello : hello.cpp world : <define>WIN32 : <variant>debug <variant>release ; 

The above Jamfile defines a preprocessor directive WIN32 for the program hello. But will WIN32 be defined for the library, too?

It won't as <define> is not a propagating feature. If you wonder how you should know: The only way to find out which features are propagated is to lookup the documentation.

If you installed the Boost C++ libraries you probably want to link against some of them. You somehow have to add a dependency to the respective Boost C++ library to your project's Jamfile. If you didn't delete the directories you had unzipped the source files of the Boost C++ libraries to you can refer to a target in a Jamfile in the root directory.

exe hello : hello.cpp world C:/boost_1_39_0//filesystem/ ; 

Now hello also depends on the Boost.Filesystem library. As the target filesystem is defined in a Jamfile in the root directory of the Boost C++ libraries the exe rule can refer to it. Not only will the appropriate Boost C++ libraries be linked - an include directory is also passed to the compiler to find the header files. If hello.cpp includes boost/filesystem.hpp the header file will be found.

In the above Jamfile the path to the root directory of the Boost C++ libraries is hardcoded. Somehow bjam needs to know where to find the Boost C++ libraries. But it would be better if the path was hardcoded only once in case several components in a project need to link against some Boost C++ libraries.

project : requirements <variant>release ; 
use-project /boost : C:/boost_1_39_0 ; 
build-project hello ; 
build-project world ; 

The use-project rule is used to define an alias to a Jamfile in another directory. Jamfiles in subdirectories use then the alias to refer to a Boost C++ library.

exe hello : hello.cpp world /boost//filesystem ; 

bjam figures out that hello.cpp is a source file, world a subdirectory and /boost//filesystem a reference to a target filesystem in a Jamfile in C:\boost_1_39_0.

Please note that a reference must start with a slash if it should refer to a project.

As libraries can be linked differently it is possible to set features relevant to the linker.

exe hello : hello.cpp world /boost//filesystem/<link>static ; 

By default libraries are linked dynamically. If libraries should be linked statically the feature <link> has to be set to static.

Features can be appended with a slash. If more than one feature should be set it is appended with another slash to the previous feature.

exe hello : hello.cpp world /boost//filesystem/<link>static/<threading>multi ; 

<threading> is another feature which can be set to single or multi. If hello should be linked against the thread-safe version of Boost.Filesystem the feature can be set accordingly.

Linking a Boost C++ library by referencing a Jamfile might not always work. If the Boost C++ libraries were installed differently because they weren't built from source for example there won't be any Jamfile to reference.

lib filesystem : : <name>libboost_filesystem <search>C:/libs ; 
exe hello : hello.cpp world filesystem : <include>C:/include ; 

The lib rule can not only be used to build a library from source. It also has to be used to refer to an existing and pre-built library.

If lib shouldn't build a library from source the second parameter must be empty. Instead in the third parameter the features <name> and <search> are used to specify the library's name and a location where Boost.Build will find the library.

It is important to specify the library's name in a platform-independent way. For example for the Jamfile above Boost.Build will try to find a file libboost_filesystem.lib on Windows. The usual file extension is again automatically appended.

If you want to reference a file by specifying its exact name you can use the <file> feature.

If a system library should be referenced for which you can expect Boost.Build to know where to find it the feature <search> can be dropped.

It is also possible to use the project rule to make sure all targets in a project are automatically linked against a library.

lib filesystem : : <name>libboost_filesystem <search>C:/libs ; 
explicit filesystem ; 
project : requirements <include>C:/include <library>filesystem ; 
lib world : world.cpp ; 

A feature called <library> must be used to add a library dependency to a project rule. <library> must refer to a lib rule which uses the already known features <name> and <search>.

It is now very important to make the lib rule explicit. This is done by using the explicit rule. It is important as by default all targets in a Jamfile are built. As the project rule defines requirements for all targets in the Jamfile they are also requirements for the lib rule. Thus the lib rule refers to itself. If the lib rule is made explicit though it's not built and no recursive reference occurs.

Please note that the order of rules in a Jamfile matters only if a rule refers to a target: Before a target can be referenced it must have been defined.

5. Best practices

How Boost.Build is used by others

As Boost.Build is a high-level build system you benefit most if you keep Jamfiles platform- and compiler-independent. After all the idea is to build your C++ or C projects on any platform with any compiler without being required to modify or maintain several Jamfiles.

A typical problem you'll run into is that third-party libraries you want to use will be installed in different directories. If you want to build your project on Windows and Unix platforms paths also look very different. Furthermore you might need to link against some system libraries on a platform but not on another.

Instead of trying to put paths for various platforms in a project's Jamfiles it is better to rely on configuration files on every system for system-specific settings. As it turns out bjam does indeed look for two more configuration files when it starts.

The file site-config.jam should be used to set options for an entire system. As it is machine-dependent bjam expects to find it in C:\Windows on Windows platforms and in /etc on Unix systems. As site-config.jam is machine-dependent paths to local libraries are no problem.

Users might not be able to create or change site-config.jam though. They would either need to wait for system administrators to update the file or be forced again to add system-specific paths to their own Jamfiles. As neither is a good solution bjam also looks for a file user-config.jam in a user's home directory. On Windows it is a subdirectory of C:\Users, on Unix a subdirecory of /home. As the file user-config.jam can be maintained by users it is probably used more often than site-config.jam.

You use site-config.jam and user-config.jam just like any other Jamfile. As these configuration files do not belong to a project but to a machine or a user on a machine they are allowed to contain machine-specific options. For example they could contain a using rule.

using msvc ; 

The using rule above tells bjam to use the msvc toolset. If you know that there is only Visual C++ installed on a system it makes sense to put this line into a configuration file. Then bjam doesn't need to guess anymore which toolset to use and won't omit a warning.

If you define targets in site-config.jam or user-config.jam and want to refer to these targets in Jamfiles the project rule must be used to set a name.

using msvc ; 
project user-config ; 
lib xml : : <name>libxml <search>C:/lib : : <include>C:/include ; 

The lib rule is used to refer to a pre-built library whose basename is libxml and can be found in C:\lib. A program which uses this XML library probably needs to include header files from this library. That's why in the usage requirements - this is the fifth parameter - the feature <include> is set to C:\include: Whoever uses this rule will inherit the <include> feature.

As the project rule has been used to set the name user-config a Jamfile can refer to the XML library via /user-config//xml.

exe xmlparser : xmlparser.cpp : <library>/user-config//xml ; 

In order to build xmlparser the program must be linked against the XML library. Even though the location of the library and its header files might vary the Jamfile does not contain any system-specific paths. The Jamfile expects to find the target xml in the project user-config. If this is a configuration file it's no problem to use system-specific paths as after all configuration files are bound to a machine or to a user on a machine.

As Boost.Build has been created to build and install the Boost C++ libraries there is built-in support to use pre-built Boost C++ libraries more easily.

using msvc ; 
project user-config ; 
using boost : 1.39 : <include>C:/include/boost-1_39 <library>C:/lib ; 

The using rule must be used to refer to a toolset called boost. This toolset is different from toolsets like msvc which you've read about so far: It doesn't contain any programs which will be run later. As support for pre-built Boost C++ libraries has been implemented in a toolset though it's required to use the using rule.

Just as with other libraries the location of the Boost C++ libraries might vary. Thus it makes sense to put the using rule into one of the two configuration files.

It is possible to pass parameters to the using rule: The first one is the version number, the second a list of options. In the Jamfile above the Boost C++ libraries 1.39 are used which can be found in the directories passed as options.

Once the boost toolset is used it is possible to use Boost C++ libraries without defining targets yourself.

import boost ; 
boost.use-project 1.39 ; 
exe hello : hello.cpp : <library>/boost//thread ; 

If a program uses a Boost C++ library it can refer to targets in a project called boost. In order to recognize the project boost though the boost module must be imported and the rule boost.use-project used: Importing the boost module makes the boost.use-project rule available. This rule expects a version number as its sole argument. As it is possible to use the using rule to refer to various versions of the Boost C++ libraries a project can specify which version it wants to use. In the Jamfile above the program hello uses Boost.Thread from version 1.39.

6. Rule reference

Building blocks for Jamfiles

If you manage a project with Boost.Build and create Jamfiles you use rules all the time. Thus you should know which rules exist and how they are used. The following table gives you an overview about the most important rules.

There is a star, plus sign or question mark behind some parameters. The star means there can be arbitrary many values, the plus sign there must be at least one value and the question mark there must be zero or exactly one value.

Table 1. Rules
Name Parameters Description
alias name : sources * : requirements * : default-build * : usage-requirements * Refer to sources or any other targets via a new name.
build-project dir Refer to a Jamfile in another directory to build a project.
conditional condition + : requirements * Create conditional requirements without using conditional properties.
exe name : sources * : requirements * : default-build * : usage-requirements * Build an executable.
explicit target-names * Make targets explicit.
glob wildcards + : excludes * Reference files in a directory via wildcards.
glob-tree wildcards + : excludes * Reference files in a directory and all subdirectories via wildcards.
install name-and-dir : sources * : requirements * : default-build * Install files to a directory.
lib names + : sources * : requirements * : default-build * : usage-requirements * Build a library.
project id ? : options * : * Set project options.
unit-test target : source : properties * Build and run an executable.
use-project id : where Reference a Jamfile in another directory to use the project id as a target.
using toolset-module : * Select a toolset.

Your Boost.Build version might support more rules than listed above. If you want to find out which rules are supported you should check out the files in the subdirectory build of your Boost.Build installation.

7. Feature reference

Configuration options for the build process

Features allow you to specify exactly how binaries are built. As there are many configuration options available the list of features is pretty long. The following table introduces you to the most important features.

Table 2. Features
Name Values Description
<address-model> 16, 32, 64, 32_64 Generate 16-, 32- or 64-bit code.
<architecture> x86, ia64, sparc, power, mips1, mips2, mips3, mips4, mips32, mips32r2, mips64, parisc, arm, combined, combined-x86-power Set processor family to generate code for.
<c++-template-depth> 1, 2, 3, ... Set maximum template depth.
<cflags> ... Pass flags to C compiler.
<cxxflags> ... Pass flags to C++ compiler
<debug-symbols> on, off Create debug symbols.
<def-file> ... Set path to def file (specific to Windows DLLs).
<define> ... Define preprocessor directives.
<embed-manifest> on, off Embed manifest (specific to msvc toolset).
<host-os> aix, bsd, cygwin, darwin, freebsd, hpux, iphone, linux, netbsd, openbsd, osf, qnx, qnxnto, sgi, solaris, unix, unixware, windows Use in conditional properties if features depend on host operating systems.
<include> ... Set include directories.
<inlining> off, on, full Inline functions.
<library> ... Link to a library (use in project rule).
<link> shared, static Link to shared or static version of a library.
<linkflags> ... Pass flags to linker.
<location> ... Set directory (use in install rule).
<name> ... Set basename of a library (use in lib rule).
<optimization> off, speed, space Generate optimized code.
<profiling> off, on Generate profiled code.
<runtime-link> shared, static Link to singlethreaded or thread-safe runtime library.
<search> ... Set directory to search for libraries (use in lib rule together with <name>).
<source> ... Set source in requirements parameter of project rule or in conditional properties.
<target-os> aix, bsd, cygwin, darwin, freebsd, hpux, iphone, linux, netbsd, openbsd, osf, qnx, qnxnto, sgi, solaris, unix, unixware, windows Use in conditional properties if features depend on target operating systems.
<threading> single, multi Build singlethreaded or thread-safe version.
<toolset> gcc, msvc, intel-linux, intel-win, acc, borland, como-linux, cw, dmc, hp_cxx, sun Use in conditional properties if features depend on toolsets.
<undef> ... Undefine preprocessor directives.
<use> ... Take over only usage requirements of a referenced target but don't do anything else.
<variant> debug, release, profile Build debug, release or profile version.
<warnings> on, all, off Switch off warnings.
<warnings-as-errors> off, on Treat warnings as errors.

For a complete and up-to-date reference of Boost.Build features look up the file builtin.jam in the subdirectory tools of your Boost.Build installation. Search for lines starting with feature.feature - this is the internal rule used to define features.