Boost Configuration Reference

Contents

Configuring Boost for Your Platform
   Using the default boost configuration
   The <boost\config.hpp> header
   Using the configure script
   User settable options
   Advanced configuration usage
   Testing the boost configuration
Boost Macro Reference
   Macros that describe defects
   Macros that describe optional features
   Boost Helper Macros
   Boost Informational Macros
   Macros for libraries with separate source code
Guidelines for Boost Authors
   Adding New Defect Macros
   Adding New Feature Test Macros
   Modifying the Boost Configuration Headers
Rationale
Acknowledgements

Configuring Boost for Your Platform

Using the default boost configuration

Boost is comes already configured for most common compilers and platforms; you should be able to use boost "as is". Since the compiler is configured separately from the standard library, the default configuration should work even if you replace the compiler's standard library with a third-party standard library (like STLport).

Using boost "as is" without trying to reconfigure is the recommended method for using boost. You can, however, run the configure script if you want to, and there are regression tests provided that allow you to test the current boost configuration with your particular compiler setup.

Boost library users can request support for additional compilers or platforms by visiting our Tracker and submitting a support request.

The <boost/config.hpp> header

Boost library implementations access configuration macros via #include <boost/config.hpp>.  

While Boost library users are not required to include that file directly, or use those configuration macros, such use is acceptable.  The configuration macros are documented as to their purpose, usage, and limitations which makes them usable by both Boost library and user code.

Boost informational or helper macros are designed for use by Boost users as well as for our own internal use.  Note however, that the feature test and defect test macros were designed for internal use by Boost libraries, not user code, so they can change at any time (though no gratuitous changes are made to them). Boost library problems resulting from changes to the configuration macros are caught by the Boost regression tests, so the Boost libraries are updated to account for those changes. By contrast, Boost library user code can be adversely affected by changes to the macros without warning. The best way to keep abreast of changes to the macros used in user code is to monitor the discussions on the Boost developers list.

Using the configure script

If you know that boost is incorrectly configured for your particular setup, and you are on a UNIX like platform, then you may want to try and improve things by running the boost configure script. From a shell command prompt you will need to cd into <boost-root>/libs/config/ and type:

sh ./configure

you will see a list of the items being checked as the script works it way through the regression tests. Note that the configure script only really auto-detects your compiler if it's called g++, c++ or CC. If you are using some other compiler then you will need to set one or more of the following environment variables:

For example to run the configure script with HP aCC, you might use something like:

export CXX="aCC"
export CXXFLAGS="-Aa
-DAportable -D__HPACC_THREAD_SAFE_RB_TREE -DRWSTD_MULTI_THREAD -DRW_MULTI_THREAD -D_REENTRANT -D_THREAD_SAFE" export LDFLAGS="-DAportable"
export LIBS=
"-lpthread" sh
./configure

However you run the configure script, when it finishes you will find a new header - user.hpp - located in the <boost-root/libs/config/> directory. Note that configure does not install this header into your boost include path by default. This header contains all the options generated by the configure script, plus a header-section that contains the user settable options from the default version of user.hpp (located under <boost-root>/boost/config/). There are two ways you can use this header:

Option 1: copy the header into <boost-root>/boost/config/ so that it replaces the default user.hpp provided by boost. This option allows only one configure-generated setup; boost developers should avoid this option, as it incurs the danger of accidentally committing a configure-modified user.hpp to the cvs repository (something you will not be thanked for!).

Option 2: give the header a more memorable name, and place it somewhere convenient, then define the macro BOOST_USER_CONFIG to point to it. For example create a new sub-directory <boost-root>/boost/config/user/, and copy the header there; for example as "multithread-gcc-config.hpp". Then when compiling add the command line option: -DBOOST_USER_CONFIG="boost/config/user/multithread-gcc-config.hpp", and boost will use the new configuration header. This option allows you to generate more than one configuration header, and to keep them separate from the boost source - so that updates to the source do not interfere with your configuration.

User settable options

There are some configuration-options that represent user choices, rather than compiler defects or platform specific options. These are listed in <boost/config/user.hpp> and at the start of a configure-generated user.hpp header. You can define these on the command line, or by editing <boost/config/user.hpp>, they are listed in the following table: 

Advanced configuration usage

By setting various macros on the compiler command line or by editing <boost/config/user.hpp>, the boost configuration setup can be optimised in a variety of ways.

Boost's configuration is structured so that the user-configuration is included first (defaulting to <boost/config/user.hpp> if BOOST_USER_CONFIG is not defined). This sets up any user-defined policies, and gives the user-configuration a chance to influence what happens next.

Next the compiler, standard library, and platform configuration files are included. These are included via macros (BOOST_COMPILER_CONFIG etc, see user settable macros), and if the corresponding macro is undefined then a separate header that detects which compiler/standard library/platform is in use is included in order to set these. The config can be told to ignore these headers altogether if the corresponding BOOST_NO_XXX macro is set (for example BOOST_NO_COMPILER_CONFIG to disable including any compiler configuration file - see user settable macros).

Finally the boost configuration header, includes <boost/config/suffix.hpp>; this header contains any boiler plate configuration code - for example where one boost macro being set implies that another must be set also.

The following usage examples represent just a few of the possibilities:

Example 1, creating our own frozen configuration.

Lets suppose that we're building boost with Visual C++ 6, and STLport 4.0. Lets suppose also that we don't intend to update our compiler or standard library any time soon. In order to avoid breaking dependencies when we update boost, we may want to "freeze" our configuration headers, so that we only have to rebuild our project if the boost code itself has changed, and not because the boost config has been updated for more recent versions of Visual C++ or STLport. We'll start by realising that the configuration files in use are: <boost/config/compiler/visualc.hpp> for the compiler, <boost/config/stdlib/stlport.hpp> for the standard library, and <boost/config/platform/win32.hpp> for the platform. Next we'll create our own private configuration directory: boost/config/mysetup/, and copy the configuration files into there. Finally, open up <boost/config/user.hpp> and edit the following defines:

#define BOOST_COMPILER_CONFIG "boost/config/mysetup/visualc.hpp"
#define BOOST_STDLIB_CONFIG "boost/config/mysetup/stlport.hpp"
#define BOOST_USER_CONFIG "boost/config/mysetup/win32.hpp"

Now when you use boost, its configuration header will go straight to our "frozen" versions, and ignore the default versions, you will now be insulated from any configuration changes when you update boost. This technique is also useful if you want to modify some of the boost configuration files; for example if you are working with a beta compiler release not yet supported by boost.

Example 2: skipping files that you don't need.

Lets suppose that you're using boost with a compiler that is fully conformant with the standard; you're not interested in the fact that older versions of your compiler may have had bugs, because you know that your current version does not need any configuration macros setting. In a case like this, you can define BOOST_NO_COMPILER_CONFIG either on the command line, or in <boost/config/user.hpp>, and miss out the compiler configuration header altogether (actually you miss out two headers, one which works out what the compiler is, and one that configures boost for it). This has two consequences: the first is that less code has to be compiled, and the second that you have removed a dependency on two boost headers.

Example 3: using configure script to freeze the boost configuration.

If you are working on a unix-like platform then you can use the configure script to generate a "frozen" configuration based on your current compiler setup - see using the configure script for more details.

Testing the boost configuration

The boost configuration library provides a full set of regression test programs under the <boost-root>/libs/config/test/ sub-directory:

Although you can run the configuration regression tests as individual test files, there are rather a lot of them, so there are a couple of shortcuts to help you out:

If you have built the boost regression test driver, then you can use this to produce a nice html formatted report of the results using the supplied test file.

Alternatively you can run the configure script like this:

./configure --enable-test

in which case the script will test the current configuration rather than creating a new one from scratch.

If you are reporting the results of these tests for a new platform/library/compiler then please include a log of the full compiler output, the output from config_info.cpp, and the pass/fail test results.

Boost Macro Reference

Macros that describe defects:

The following macros all describe features that are required by the C++ standard, if one of the following macros is defined, then it represents a defect in the compiler's conformance with the standard.

template<typename T>
struct foo : {
   template<typename U>
   struct bar : public U {};
};

template<class T, typename T::type value> 
class X { ... };

template<typename T> void f(T); // #1
template<typename T, typename U> void f(T (*)(U)); // #2

void bar(int);

f(&bar); // should choose #2.

void f() {}
void g() { return f(); }

Macros that describe optional features:

The following macros describe features that are not required by the C++ standard. The macro is only defined if the feature is present.

Boost Helper Macros

The following macros are either simple helpers, or macros that provide workarounds for compiler/standard library defects.

template <class T> void f(T, typename T::type);

template <class T> void f(T, BOOST_DEDUCED_TYPENAME T::type);

struct foo{
   static const int value = 2;
};

struct foo{
   BOOST_STATIC_CONSTANT(int, value = 2);
};

  #include <iostream>
  #include <ostream>
  #include <typeinfo>

  template <int n>
  void f() { std::cout << n << ' '; }

  template <typename T>
  void g() { std::cout << typeid(T).name() << ' '; }

  int main() {
    f<1>();
    f<2>();

    g<int>();
    g<double>();
  }

  template <int n>
  void f(BOOST_EXPLICIT_TEMPLATE_NON_TYPE(int, n)) { ... }

  template <typename T>
  void g(BOOST_EXPLICIT_TEMPLATE_TYPE(T)) { ... }

std::use_facet<Type>(loc);

BOOST_USE_FACET(Type, loc);

std::has_facet<Type>(loc);

BOOST_HAS_FACET(Type, loc);

typedef typename A::template rebind<U> binder;

typedef typename A::BOOST_NESTED_TEMPLATE rebind<U> binder;

Boost Informational Macros

The following macros describe boost features; these are the generally speaking the only boost macros that should be tested in user code.

Macros for libraries with separate source code

The following macros and helper headers are of use to authors whose libraries include separate source code, and are intended to address two issues: fixing the ABI of the compiled library, and selecting which compiled library to link against based upon the compilers settings.

ABI Fixing

When linking against a pre-compiled library it vital that the ABI used by the compiler when building the library matches exactly the ABI used by the code using the library.  In this case ABI means things like the struct packing arrangement used, the name mangling scheme used, or the size of some types (enum types for example).  This is separate from things like threading support, or runtime library variations, which have to be dealt with by build variants.  To put this in perspective there is one compiler (Borland's) that has so many compiler options that make subtle changes to the ABI, that at least in theory there 3200 combinations, and that's without considering runtime library variations.  Fortunately these variations can be managed by #pragma's that tell the compiler what ABI to use for the types declared in your library, in order to avoid sprinkling #pragma's all over the boost headers, there are some prefix and suffix headers that do the job, typical usage would be:

#ifndef MY_INCLUDE_GUARD
#define MY_INCLUDE_GUARD

// all includes go here:
#include <boost/config.hpp>
#include <whatever>

#ifdef BOOST_HAS_ABI_HEADERS
#  include BOOST_ABI_PREFIX
#endif

namespace boost{
// your code goes here
} 

#ifdef BOOST_HAS_ABI_HEADERS
#  include BOOST_ABI_SUFFIX
#endif


#endif // include guard

The user can disable this mechanism by defining BOOST_DISABLE_ABI_HEADERS, or they can define BOOST_ABI_PREFIX and/or BOOST_ABI_SUFFIX to point to their own prefix/suffix headers if they so wish.

Automatic library selection

It is essential that users link to a build of a library which was built against the same runtime library that their application will be built against - if this does not happen then the library will not be binary compatible with their own code - and there is a high likelihood  that their application will experience runtime crashes.  These kinds of problems can be extremely time consuming and difficult to debug, and often lead to frustrated users and authors alike (simply selecting the right library to link against is not as easy as it seems when their are 6-8 of them to chose from, and some users seem to be blissfully unaware that there even are different runtimes available to them).

To solve this issue, some compilers allow source code to contain #pragma's that instruct the linker which library to link against, all the user need do is include the headers they need, place the compiled libraries in their library search path, and the compiler and linker do the rest.  Boost.config supports this via the header <boost/config/auto_link.hpp>, before including this header one or more of the following macros need to be defined:

BOOST_LIB_NAME	Required: An identifier containing the basename of the library, for example 'boost_regex'.
BOOST_DYN_LINK	Optional: when set link to dll rather than static library.
BOOST_LIB_DIAGNOSTIC	Optional: when set the header will print out the name of the library selected (useful for debugging).

If the compiler supports this mechanism, then it will be told to link against the appropriately named library, the actual algorithm used to mangle the name of the library is documented inside <boost/config/auto_link.hpp> and has to match that used to create the libraries via bjam 's install rules.

Typical usage would be:

//
// Don't include auto-linking code if the user has disabled it by
// defining BOOST_WHATEVER_NO_LIB, or if this is one of our own
// source files (signified by BOOST_WHATEVER_SOURCE):
//
#if !defined(BOOST_WHATEVER_NO_LIB) && !defined(BOOST_WHATEVER_SOURCE)
#  define BOOST_LIB_NAME boost_whatever
#  ifdef BOOST_WHATEVER_DYN_LINK
#     define BOOST_DYN_LINK
#  endif
#  include <boost/config/auto_link.hpp>
#endif

Guidelines for Boost Authors

The boost/config.hpp header is used to pass configuration information to other boost files, allowing them to cope with platform dependencies such as arithmetic byte ordering, compiler pragmas, or compiler shortcomings. Without such configuration information, many current compilers would not work with the Boost libraries.

Centralizing configuration information in this header reduces the number of files that must be modified when porting libraries to new platforms, or when compilers are updated. Ideally, no other files would have to be modified when porting to a new platform.

Configuration headers are controversial because some view them as condoning broken compilers and encouraging non-standard subsets. Adding settings for additional platforms and maintaining existing settings can also be a problem. In other words, configuration headers are a necessary evil rather than a desirable feature. The boost config.hpp policy is designed to minimize the problems and maximize the benefits of a configuration header.

Note that:

• Boost library implementers are not required to #include <boost/config.hpp>, and are not required in any way to support compilers that do not comply with the C++ Standard (ISO/IEC 14882).
• If a library implementer wishes to support some non-conforming compiler, or to support some platform specific feature, #include <boost/config.hpp> is the preferred way to obtain configuration information not available from the standard headers such as <climits>, etc.
• If configuration information can be deduced from standard headers such as <climits>, use those standard headers rather than <boost/config.hpp>.
• Boost files that use macros defined in <boost/config.hpp> should have sensible, standard conforming, default behavior if the macro is not defined. This means that the starting point for porting <boost/config.hpp> to a new platform is simply to define nothing at all specific to that platform. In the rare case where there is no sensible default behavior, an #error message should describe the problem.
• If a Boost library implementer wants something added to config.hpp, post a request on the Boost mailing list. There is no guarantee such a request will be honored; the intent is to limit the complexity of config.hpp.
• The intent is to support only compilers which appear on their way to becoming C++ Standard compliant, and only recent releases of those compilers at that.
• The intent is not to disable mainstream features now well-supported by the majority of compilers, such as namespaces, exceptions, RTTI, or templates.

Adding New Defect Macros

When you need to add a new defect macro - either to fix a problem with an existing library, or when adding a new library - distil the issue down to a simple test case, often at this point other (possibly better) workarounds may become apparent. Secondly always post the test case code to the boost mailing list and invite comments; remember that C++ is complex and that sometimes what may appear a defect, may in fact turn out to be a problem with the authors understanding of the standard.

When you name the macro, follow the BOOST_NO_SOMETHING naming convention, so that it's obvious that this is a macro reporting a defect.

Finally, add the test program to the regression tests. You will need to place the test case in a .ipp file with the following comments near the top:

//  MACRO:         BOOST_NO_FOO
//  TITLE:         foo
//  DESCRIPTION:   If the compiler fails to support foo

These comments are processed by the autoconf script, so make sure the format follows the one given. The file should be named "boost_no_foo.ipp", where foo is the defect description - try and keep the file name under the Mac 30 character filename limit though. You will also need to provide a function prototype "int test()" that is declared in a namespace with the same name as the macro, but in all lower case, and which returns zero on success:

namespace boost_no_foo{

int test()
{
   // test code goes here:
   //
   return 0;
}

}

Once the test code is in place, build and run the program "generate.cpp" that you will find in the boost-root/libs/config/tools/ directory. This generates two .cpp test files from the new test code, and adds the tests to the regression test Jamfile, and the config_test.cpp test program. Finally add a new entry to config_info.cpp so that the new macro gets printed out when that program is run.

Adding New Feature Test Macros

When you need to add a macro that describes a feature that the standard does not require, follow the convention for adding a new defect macro (above), but call the macro BOOST_HAS_FOO, and name the test file "boost_has_foo.ipp". Try not to add feature test macros unnecessarily, if there is a platform specific macro that can already be used (for example _WIN32, __BEOS__, or __linux) to identify the feature then use that. Try to keep the macro to a feature group, or header name, rather than one specific API (for example BOOST_HAS_NL_TYPES_H rather than BOOST_HAS_CATOPEN). If the macro describes a POSIX feature group, then add boilerplate code to boost/config/suffix.hpp to auto-detect the feature where possible (if you are wondering why we can't use POSIX feature test macro directly, remember that many of these features can be added by third party libraries, and are not therefore identified inside <unistd.h>).

Modifying the Boost Configuration Headers

The aim of boost's configuration setup is that the configuration headers should be relatively stable - a boost user should not have to recompile their code just because the configuration for some compiler that they're not interested in has changed. Separating the configuration into separate compiler/standard library/platform sections provides for part of this stability, but boost authors require some amount of restraint as well, in particular:

<boost/config.hpp> should never change, don't alter this file.

<boost/config/user.hpp> is included by default, don't add extra code to this file unless you have to. If you do, please remember to update libs/config/tools/configure.in as well.

<boost/config/suffix.hpp> is always included so be careful about modifying this file as it breaks dependencies for everyone. This file should include only "boilerplate" configuration code, and generally should change only when new macros are added.

<boost/config/select_compiler_config.hpp>, <boost/config/select_platform_config.hpp> and <boost/config/select_stdlib_config.hpp> are included by default and should change only if support for a new compiler/standard library/platform is added.

The compiler/platform/standard library selection code is set up so that unknown platforms are ignored and assumed to be fully standards compliant - this gives unknown platforms a "sporting chance" of working "as is" even without running the configure script.

When adding or modifying the individual mini-configs, assume that future, as yet unreleased versions of compilers, have all the defects of the current version. Although this is perhaps unnecessarily pessimistic, it cuts down on the maintenance of these files, and experience suggests that pessimism is better placed than optimism here!

Rationale

The problem with many traditional "textbook" implementations of configuration headers (where all the configuration options are in a single "monolithic" header) is that they violate certain fundamental software engineering principles which would have the effect of making boost more fragile, more difficult to maintain and more difficult to use safely. You can find a description of the principles from the following article.

The problem

Consider a situation in which you are concurrently developing on multiple platforms. Then consider adding a new platform or changing the platform definitions of an existing platform. What happens? Everything, and this does literally mean everything, recompiles.. Isn't it quite absurd that adding a new platform, which has absolutely nothing to do with previously existing platforms, means that all code on all existing platforms needs to be recompiled?

Effectively, there is an imposed physical dependency between platforms that have nothing to do with each other. Essentially, the traditional solution employed by configuration headers does not conform to the Open-Closed Principle:

"A module should be open for extension but closed for modification."

Extending a traditional configuration header implies modifying existing code.

Furthermore, consider the complexity and fragility of the platform detection code. What if a simple change breaks the detection on some minor platform? What if someone accidentally or on purpose (as a workaround for some other problem) defines some platform dependent macros that are used by the detection code? A traditional configuration header is one of the most volatile headers of the entire library, and more stable elements of Boost would depend on it. This violates the Stable Dependencies Principle:

"Depend in the direction of stability."

After even a minor change to a traditional configuration header on one minor platform, almost everything on every platform should be tested if we follow sound software engineering practice.

Another important issue is that it is not always possible to submit changes to <boost/config.hpp>. Some boost users are currently working on platforms using tools and libraries that are under strict Non-Disclosure Agreements. In this situation it is impossible to submit changes to a traditional monolithic configuration header, instead some method by which the user can insert their own configuration code must be provided.

The solution

The approach taken by boost's configuration headers is to separate configuration into three orthogonal parts: the compiler, the standard library and the platform. Each compiler/standard library/platform gets its own mini-configuration header, so that change to one compiler's configuration (for example) does not effect other compilers. In addition there are measures that can be taken both to omit the compiler/standard library/platform detection code (so that adding support to a new platform does not break dependencies), or to freeze the configuration completely; providing almost complete protection against dependency changes.

Acknowledgements

Beman Dawes provided the original config.hpp and part of this document. Vesa Karvonen provided a description of the principles (see rationale) and put together an early version of the current configuration setup. John Maddock put together the configuration current code, the test programs, the configuration script and the reference section of this document. Numerous boost members, past and present, have contributed fixes to boost's configuration.

© Beman Dawes 2001

© Vesa Karvonen 2001

© John Maddock 2001