boost.png (6897 bytes) System Library
Boost Home    Library Home   Tutorial   Reference
Design Rationale


Error conditions originating from the operating system or other low-level application program interfaces (API's) are typically reported via an integer representing an error code. When these low-level API calls are wrapped in portable code, such as in a portable library, some users want to deal with the error codes in portable ways. Other users need to get at the system specific error codes, so they can deal with system specific needs. The Boost System library provides simple, light-weight error_code objects that encapsulate system-specific error code values, yet also provide access to more abstract and portable error conditions via error_condition objects. Because error_code objects can represent errors from sources other than the operating system, including user-defined sources, each error_code and error_condition has an associated error_category.

An exception class,  system_error, is provided. Derived from std::runtime_error, it captures the underlying error_code for the problem causing the exception so that this important information is not lost.

While exceptions are the preferred C++ default error code reporting mechanism, users of libraries dependent on low-level API's often need overloads reporting error conditions via error code arguments and/or return values rather than via throwing exceptions. Otherwise, when errors are not exceptional occurrences and must be dealt with as they arise, programs become littered with try/catch blocks, unreadable, and very inefficient. The Boost System library supports both error reporting by exception and by error code.

In addition to portable errors codes and conditions supported by the error_code.hpp header, system-specific headers support the Cygwin, Linux, and Windows platforms. These headers are effectively no-ops if included for platforms other than their intended target.

The Boost System Library will become part of the C++0x Standard Library. A number of changes, particularly to names, were made by the C++ committee during standardization. The Boost implementation is tracking those changes. See Deprecated names for synonyms provided to prevent breakage of existing user code. See Breaking changes for changes that unavoidably break existing user code. All breaking changes are noisy and will cause compile-time errors.

Design Rationale

Class error_code  and error_condition are designed as a value types so they can be copied without slicing and do not requiring heap allocation, but still have polymorphic behavior based on the error category. This is achieved by abstract base class error_category supplying the polymorphic behavior, and error_code and error_condition containing a pointer to an object of a type derived from error_category.

Many of the detailed design decisions were driven by the requirements that users to be able to add additional error categories, and that it be no more difficult to write portable code than system-specific code.

The operator<< overload for error_code eliminates a misleading conversion to bool in code like cout << ec, where ec is of type error_code. It is also useful in its own right.


N1975, Filesystem Library Proposal for TR2, accepted for Library Technical Report 2 (TR2) at the Berlin meeting, included additional components to supplement the Standard Library's Diagnostics clause. Since then, these error reporting components have received wider public scrutiny and enhancements have been made to the design. The enhanced version has been used by N2054, Networking Library Proposal for TR2, demonstrating that these error reporting components are useful beyond the original Filesystem Library.

The original proposal viewed error categories as a binary choice between errno (i.e. POSIX-style) and the native operating system's error codes. The proposed components now allow as many additional error categories as are needed by either implementations or by users. The need to support additional error categories, for example, occurs in some networking library implementations because they are built on top of the POSIX getaddrinfo API that uses error codes not based on errno.


Christopher Kohlhoff and Peter Dimov made important contributions to the design. Comments and suggestions were also received from Pavel Vozenilek, Gennaro Prota, Dave Abrahams, Jeff Garland, Iain Hanson, Oliver Kowalke, and Oleg Abrosimov. Christopher Kohlhoff suggested several improvements to the N2066 paper. Johan Nilsson's comments led to several of the refinements in N2066 .

Revised May 28, 2010

Copyright Beman Dawes, 1999

Distributed under the Boost Software License, Version 1.0. (See file LICENSE_1_0.txt or