gcc/libstdc++-v3/doc/xml/manual/using_exceptions.xml

<section xmlns="http://docbook.org/ns/docbook" version="5.0" 
	 xml:id="manual.intro.using.exceptions" xreflabel="Using Exceptions">
<?dbhtml filename="using_exceptions.html"?>

<info><title>Exceptions</title>
  <keywordset>
    <keyword>
      C++
    </keyword>
    <keyword>
      exception
    </keyword>
    <keyword>
      error
    </keyword>
    <keyword>
      exception neutrality
    </keyword>
    <keyword>
      exception safety
    </keyword>
    <keyword>
      exception propagation
    </keyword>
    <keyword>
      -fno-exceptions
    </keyword>
  </keywordset>
</info>

<para>
The C++ language provides language support for stack unwinding
with <literal>try</literal> and <literal>catch</literal> blocks and
the <literal>throw</literal> keyword.
</para>

<para>
These are very powerful constructs, and require some thought when
applied to the standard library in order to yield components that work
efficiently while cleaning up resources when unexpectedly killed via
exceptional circumstances.
</para>

<para>
Two general topics of discussion follow:
exception neutrality and exception safety.
</para>


<section xml:id="intro.using.exception.safety" xreflabel="Exception Safety"><info><title>Exception Safety</title></info>


  <para>
    What is exception-safe code?
  </para>

  <para>
    Will define this as reasonable and well-defined behavior by classes
    and functions from the standard library when used by user-defined
    classes and functions that are themselves exception safe.
  </para>

  <para>
    Please note that using exceptions in combination with templates
    imposes an additional requirement for exception
    safety. Instantiating types are required to have destructors that
    do no throw.
  </para>

  <para>
    Using the layered approach from Abrahams, can classify library
    components as providing set levels of safety. These will be called
    exception guarantees, and can be divided into three categories.
  </para>

<itemizedlist>

  <listitem>
  <para>
    One. Don't throw.
  </para>
  <para>
    As specified in 23.2.1 general container requirements. Applicable
    to container and string classes.
  </para>
  <para>
    Member
    functions <function>erase</function>, <function>pop_back</function>, <function>pop_front</function>, <function>swap</function>, <function>clear</function>. And <type>iterator</type>
    copy constructor and assignment operator.
  </para>
  </listitem>

  <listitem>
  <para>
    Two. Don't leak resources when exceptions are thrown. This is
    also referred to as the <quote>basic</quote> exception safety guarantee.
  </para>

  <para>
    This applicable throughout the standard library.
  </para>
  </listitem>

  <listitem>
  <para>
    Three. Commit-or-rollback semantics.  This is
    referred to as <quote>strong</quote> exception safety guarantee.
  </para>

  <para>
    As specified in 23.2.1 general container requirements. Applicable
    to container and string classes.
  </para>
  <para>
    Member functions <function>insert</function> of a single
    element, <function>push_back</function>, <function>push_front</function>,
    and <function>rehash</function>.
  </para>

  </listitem>
</itemizedlist>

</section>


<section xml:id="intro.using.exception.propagating" xreflabel="Exceptions Neutrality"><info><title>Exception Neutrality</title></info>

  <para>
    Simply put, once thrown an exception object should continue in
    flight unless handled explicitly. In practice, this means
    propagating exceptions should not be swallowed in
    gratuitous <literal>catch(...)</literal> blocks. Instead,
    matching <literal>try</literal> and <literal>catch</literal>
    blocks should have specific catch handlers and allow un-handed
    exception objects to propagate. If a
    terminating <literal>catch(...)</literal> blocks exist then it
    should end with a <literal>throw</literal> to re-throw the current
    exception.
  </para>

  <para>
    Why do this?
  </para>

  <para>
    By allowing exception objects to propagate, a more flexible
    approach to error handling is made possible (although not
    required.) Instead of dealing with an error immediately, one can
    allow the exception to propagate up until sufficient context is
    available and the choice of exiting or retrying can be made in an
    informed manner.
  </para>

  <para>
    Unfortunately, this tends to be more of a guideline than a strict
    rule as applied to the standard library. As such, the following is
    a list of known problem areas where exceptions are not propagated.
  </para>

<itemizedlist>
  <listitem>
    <para>
      Input/Output
    </para>
  <para>
    The destructor <function>ios_base::Init::~Init()</function>
    swallows all exceptions from <function>flush</function> called on
    all open streams at termination.
  </para>

  <para>
    All formatted input in <classname>basic_istream</classname> or
    formatted output in <classname>basic_ostream</classname> can be
    configured to swallow exceptions
    when <function>exceptions</function> is set to
    ignore <type>ios_base::badbit</type>.
  </para>

  <para>
    Functions that have been registered
    with <function>ios_base::register_callback</function> swallow all
    exceptions when called as part of a callback event.
  </para>

  <para>
    When closing the underlying
    file, <function>basic_filebuf::close</function> will swallow
    (non-cancellation) exceptions thrown and return <literal>NULL</literal>.
  </para>
  </listitem>
  <listitem>
    <para>
      Thread
    </para>
    <para>
      The constructors of <classname>thread</classname> that take a
      callable function argument swallow all exceptions resulting from
      executing the function argument.
    </para>
  </listitem>
</itemizedlist>

</section>

<section xml:id="intro.using.exception.no" xreflabel="-fno-exceptions"><info><title>Doing without</title></info>

  <para>
    C++ is a language that strives to be as efficient as is possible
    in delivering features. As such, considerable care is used by both
    language implementer and designers to make sure unused features
    not impose hidden or unexpected costs. The GNU system tries to be
    as flexible and as configurable as possible. So, it should come as
    no surprise that GNU C++ provides an optional language extension,
    spelled <literal>-fno-exceptions</literal>, as a way to excise the
    implicitly generated magic necessary to
    support <literal>try</literal> and <literal>catch</literal> blocks
    and thrown objects. (Language support
    for <literal>-fno-exceptions</literal> is documented in the GNU
    GCC <link xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="http://gcc.gnu.org/onlinedocs/gcc/Code-Gen-Options.html#Code-Gen-Options">manual</link>.)
  </para>

  <para>Before detailing the library support
    for <literal>-fno-exceptions</literal>, first a passing note on
    the things lost when this flag is used: it will break exceptions
    trying to pass through code compiled
    with <literal>-fno-exceptions</literal> whether or not that code
    has any <literal>try</literal> or <literal>catch</literal>
    constructs. If you might have some code that throws, you shouldn't
    use <literal>-fno-exceptions</literal>. If you have some code that
    uses <literal>try</literal> or <literal>catch</literal>, you
    shouldn't use <literal>-fno-exceptions</literal>.
  </para>

  <para>
    And what it to be gained, tinkering in the back alleys with a
    language like this? Exception handling overhead can be measured
    in the size of the executable binary, and varies with the
    capabilities of the underlying operating system and specific
    configuration of the C++ compiler. On recent hardware with GNU
    system software of the same age, the combined code and data size
    overhead for enabling exception handling is around 7%. Of course,
    if code size is of singular concern than using the appropriate
    optimizer setting with exception handling enabled
    (ie, <literal>-Os -fexceptions</literal>) may save up to twice
    that, and preserve error checking.
  </para>

  <para>
    So. Hell bent, we race down the slippery track, knowing the brakes
    are a little soft and that the right front wheel has a tendency to
    wobble at speed. Go on: detail the standard library support
    for <literal>-fno-exceptions</literal>.
  </para>

  <para>
    In sum, valid C++ code with exception handling is transformed into
    a dialect without exception handling. In detailed steps: all use
    of the C++
    keywords <literal>try</literal>, <literal>catch</literal>,
    and <literal>throw</literal> in the standard library have been
    permanently replaced with the pre-processor controlled equivalents
    spelled <literal>__try</literal>, <literal>__catch</literal>,
    and <literal>__throw_exception_again</literal>. They are defined
    as follows.
  </para>

<programlisting>
#ifdef __EXCEPTIONS
# define __try      try
# define __catch(X) catch(X)
# define __throw_exception_again throw
#else
# define __try      if (true)
# define __catch(X) if (false)
# define __throw_exception_again
#endif
</programlisting>

<para>
  In addition, for every object derived from
  class <classname>exception</classname>, there exists a corresponding
  function with C language linkage. An example:
</para>

<programlisting>
#ifdef __EXCEPTIONS
  void __throw_bad_exception(void)
  { throw bad_exception(); }
#else
  void __throw_bad_exception(void)
  { abort(); }
#endif
</programlisting>

<para>
  The last language feature needing to be transformed
  by <literal>-fno-exceptions</literal> is treatment of exception
  specifications on member functions. Fortunately, the compiler deals
  with this by ignoring exception specifications and so no alternate
  source markup is needed.
</para>

<para>
  By using this combination of language re-specification by the
  compiler, and the pre-processor tricks and the functional
  indirection layer for thrown exception objects by the library,
  libstdc++ files can be compiled
  with <literal>-fno-exceptions</literal>.
</para>

<para>
 User code that uses C++ keywords
 like <literal>throw</literal>, <literal>try</literal>,
 and <literal>catch</literal> will produce errors even if the user
 code has included libstdc++ headers and is using constructs
 like <classname>basic_iostream</classname>. Even though the standard
 library has been transformed, user code may need modification. User
  code that attempts or expects to do error checking on standard
  library components compiled with exception handling disabled should
  be evaluated and potentially made conditional.
</para>

<para>
  Some issues remain with this approach (see bugzilla entry
  25191). Code paths are not equivalent, in
  particular <literal>catch</literal> blocks are not evaluated. Also
  problematic are <literal>throw</literal> expressions expecting a
  user-defined throw handler. Known problem areas in the standard
  library include using an instance
  of <classname>basic_istream</classname>
  with <function>exceptions</function> set to specific
  <type>ios_base::iostate</type> conditions, or
  cascading <literal>catch</literal> blocks that dispatch error
  handling or recovery efforts based on the type of exception object
  thrown.
</para>

<para>
  Oh, and by the way: none of this hackery is at all
  special. (Although perhaps well-deserving of a raised eyebrow.)
  Support continues to evolve and may change in the future. Similar
  and even additional techniques are used in other C++ libraries and
  compilers.
</para>

<para>
 C++ hackers with a bent for language and control-flow purity have
  been successfully consoled by grizzled C veterans lamenting the
  substitution of the C language keyword
  <literal>const</literal> with the uglified
  doppelganger <literal>__const</literal>.
</para>


</section>

<section xml:id="intro.using.exception.compat"><info><title>Compatibility</title></info>


<section xml:id="using.exception.compat.c"><info><title>With <literal>C</literal></title></info>

<para>
  C language code that is expecting to interoperate with C++ should be
  compiled with <literal>-fexceptions</literal>. This will make
  debugging a C language function called as part of C++-induced stack
  unwinding possible.
</para>

<para>
  In particular, unwinding into a frame with no exception handling
data will cause a runtime abort. If the unwinder runs out of unwind
info before it finds a handler, <function>std::terminate()</function>
is called.
</para>

<para>
  Please note that most development environments should take care of
  getting these details right. For GNU systems, all appropriate parts
  of the GNU C library are already compiled
  with <literal>-fexceptions</literal>.
</para>

</section>

<section xml:id="using.exception.compat.posix"><info><title>With <literal>POSIX</literal> thread cancellation</title></info>


<para>
  GNU systems re-use some of the exception handling mechanisms to
  track control flow for <literal>POSIX</literal> thread cancellation.
</para>

<para>
  Cancellation points are functions defined by POSIX as worthy of
  special treatment. The standard library may use some of these
  functions to implement parts of the ISO C++ standard or depend on
  them for extensions.
</para>

<para>
  Of note:
</para>

<para>
  <function>nanosleep</function>,
  <function>read</function>, <function>write</function>, <function>open</function>, <function>close</function>,
  and <function>wait</function>.
</para>

<para>
  The parts of libstdc++ that use C library functions marked as
  cancellation points should take pains to be exception neutral.
  Failing this, <literal>catch</literal> blocks have been augmented to
  show that the POSIX cancellation object is in flight.
</para>

<para>
  This augmentation adds a <literal>catch</literal> block
  for <classname>__cxxabiv1::__forced_unwind</classname>, which is the
  object representing the POSIX cancellation object. Like so:
</para>

<programlisting>
  catch(const __cxxabiv1::__forced_unwind&amp;)
  {
    this-&gt;_M_setstate(ios_base::badbit);
    throw;
  }
  catch(...)
  { this-&gt;_M_setstate(ios_base::badbit); }
</programlisting>


</section>
</section>

<bibliography xml:id="using.exceptions.biblio"><info><title>Bibliography</title></info>


  <biblioentry>
    <biblioid xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="http://www.opengroup.org/austin" class="uri">
      </biblioid>
      <citetitle>
	System Interface Definitions, Issue 7 (IEEE Std. 1003.1-2008)
      </citetitle>

    <pagenums>
      2.9.5 Thread Cancellation
    </pagenums>
    <copyright>
      <year>2008</year>
      <holder>
	The Open Group/The Institute of Electrical and Electronics
	Engineers, Inc.
      </holder>
    </copyright>
  </biblioentry>

  <biblioentry>
    <biblioid xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="http://www.boost.org/community/error_handling.html" class="uri">
      </biblioid>
      <citetitle>
	Error and Exception Handling
      </citetitle>
      
    <author><personname><firstname>David</firstname><surname>Abrahams </surname></personname></author>
    <publisher>
      <publishername>
	Boost
      </publishername>
    </publisher>
  </biblioentry>


  <biblioentry>
    <biblioid xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="http://www.boost.org/community/exception_safety.html" class="uri">
      </biblioid>
      <citetitle>
	Exception-Safety in Generic Components
      </citetitle>
      
    <author><personname><firstname>David</firstname><surname>Abrahams</surname></personname></author>
    <publisher>
      <publishername>
	Boost
      </publishername>
    </publisher>
  </biblioentry>

  <biblioentry>
    <biblioid xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="www.open-std.org/jtc1/sc22/wg21/docs/papers/1997/N1077.pdf" class="uri">
      </biblioid>
      <citetitle>
	Standard Library Exception Policy
      </citetitle>
    <author><personname><firstname>Matt</firstname><surname>Austern</surname></personname></author>
    <publisher>
      <publishername>
	WG21 N1077
      </publishername>
    </publisher>
  </biblioentry>

  <biblioentry>
    <biblioid xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="http://gcc.gnu.org/ml/gcc-patches/2001-03/msg00661.html" class="uri">
      </biblioid>
      <citetitle>
	ia64 c++ abi exception handling
      </citetitle>

    <author><personname><firstname>Richard</firstname><surname>Henderson</surname></personname></author>
    <publisher>
      <publishername>
	GNU
      </publishername>
    </publisher>
  </biblioentry>

  <biblioentry>
    <biblioid xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="http://www.research.att.com/~bs/3rd_safe.pdf" class="uri">
      </biblioid>
      <citetitle>
	Appendix E: Standard-Library Exception Safety
      </citetitle>
    <author><personname><firstname>Bjarne</firstname><surname>Stroustrup</surname></personname></author>
  </biblioentry>

  <biblioentry>
    <citetitle>
      Exceptional C++
    </citetitle>
    <pagenums>
      Exception-Safety Issues and Techniques
    </pagenums>
    <author><personname><firstname>Herb</firstname><surname>Sutter</surname></personname></author>
  </biblioentry>

  <biblioentry>
    <biblioid xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="http://gcc.gnu.org/PR25191" class="uri">
    </biblioid>
    <citetitle>
      GCC Bug 25191: exception_defines.h #defines try/catch
    </citetitle>
  </biblioentry>

</bibliography>

</section>