libstdc++-v3/doc/xml/manual/support.xml

<chapter xmlns="http://docbook.org/ns/docbook" version="5.0"
	 xml:id="std.support" xreflabel="Support">
<?dbhtml filename="support.html"?>

<info><title>
  Support
  <indexterm><primary>Support</primary></indexterm>
</title>
  <keywordset>
    <keyword>ISO C++</keyword>
    <keyword>library</keyword>
  </keywordset>
</info>

  <para>
    This part deals with the functions called and objects created
    automatically during the course of a program's existence.
  </para>

  <para>
    While we can't reproduce the contents of the Standard here (you
    need to get your own copy from your nation's member body; see our
    homepage for help), we can mention a couple of changes in what
    kind of support a C++ program gets from the Standard Library.
  </para>

<section xml:id="std.support.types" xreflabel="Types"><info><title>Types</title></info>
  <?dbhtml filename="fundamental_types.html"?>

  <section xml:id="std.support.types.fundamental" xreflabel="Fundamental Types"><info><title>Fundamental Types</title></info>

    <para>
      C++ has the following builtin types:
    </para>
    <itemizedlist>
      <listitem><para>
	char
      </para></listitem>
      <listitem><para>
	signed char
      </para></listitem>
      <listitem><para>
	unsigned char
      </para></listitem>
      <listitem><para>
	signed short
      </para></listitem>
      <listitem><para>
	signed int
      </para></listitem>
      <listitem><para>
	signed long
      </para></listitem>
      <listitem><para>
	unsigned short
      </para></listitem>
      <listitem><para>
	unsigned int
      </para></listitem>
      <listitem><para>
	unsigned long
      </para></listitem>
      <listitem><para>
	bool
      </para></listitem>
      <listitem><para>
	wchar_t
      </para></listitem>
      <listitem><para>
	float
      </para></listitem>
      <listitem><para>
	double
      </para></listitem>
      <listitem><para>
	long double
      </para></listitem>
    </itemizedlist>

    <para>
      These fundamental types are always available, without having to
      include a header file. These types are exactly the same in
      either C++ or in C.
    </para>

    <para>
      Specializing parts of the library on these types is prohibited:
      instead, use a POD.
    </para>

  </section>
  <section xml:id="std.support.types.numeric_limits" xreflabel="Numeric Properties"><info><title>Numeric Properties</title></info>

    <para>
    The header <filename class="headerfile">&lt;limits&gt;</filename> defines
    traits classes to give access to various implementation
    defined-aspects of the fundamental types. The traits classes --
    fourteen in total -- are all specializations of the class template
    <classname>numeric_limits</classname>
    and defined as follows:
    </para>

<programlisting>
   template&lt;typename T&gt;
     struct class
     {
       static const bool is_specialized;
       static T max() throw();
       static T min() throw();

       static const int digits;
       static const int digits10;
       static const bool is_signed;
       static const bool is_integer;
       static const bool is_exact;
       static const int radix;
       static T epsilon() throw();
       static T round_error() throw();

       static const int min_exponent;
       static const int min_exponent10;
       static const int max_exponent;
       static const int max_exponent10;

       static const bool has_infinity;
       static const bool has_quiet_NaN;
       static const bool has_signaling_NaN;
       static const float_denorm_style has_denorm;
       static const bool has_denorm_loss;
       static T infinity() throw();
       static T quiet_NaN() throw();
       static T denorm_min() throw();

       static const bool is_iec559;
       static const bool is_bounded;
       static const bool is_modulo;

       static const bool traps;
       static const bool tinyness_before;
       static const float_round_style round_style;
     };
</programlisting>
  </section>

  <section xml:id="std.support.types.null" xreflabel="NULL"><info><title>NULL</title></info>

    <para>
     The only change that might affect people is the type of
     <constant>NULL</constant>: while it is required to be a macro,
     the definition of that macro is <emphasis>not</emphasis> allowed
     to be an expression with pointer type such as
     <constant>(void*)0</constant>, which is often used in C.
    </para>

    <para>
     For <command>g++</command>, <constant>NULL</constant> is
     <code>#define</code>'d to be
     <constant>__null</constant>, a magic keyword extension of
     <command>g++</command> that is slightly safer than a plain integer.
    </para>

    <para>
     The biggest problem of #defining <constant>NULL</constant> to be
     something like <quote>0L</quote> is that the compiler will view
     that as a long integer before it views it as a pointer, so
     overloading won't do what you expect. It might not even have the
     same size as a pointer, so passing <constant>NULL</constant> to a
     varargs function where a pointer is expected might not even work
     correctly if <code>sizeof(NULL) &lt; sizeof(void*)</code>.
     The G++ <constant>__null</constant> extension is defined so that
     <code>sizeof(__null) == sizeof(void*)</code> to avoid this problem.
    </para>

    <para>
     Scott Meyers explains this in more detail in his book
     <link xmlns:xlink="http://www.w3.org/1999/xlink"
      xlink:href="https://www.aristeia.com/books.html"><emphasis>Effective
     Modern C++</emphasis></link> and as a guideline to solve this problem
     recommends to not overload on pointer-vs-integer types to begin with.
    </para>

    <para>
     The C++ 2011 standard added the <constant>nullptr</constant> keyword,
     which is a null pointer constant of a special type,
     <classname>std::nullptr_t</classname>. Values of this type can be
     implicitly converted to <emphasis>any</emphasis> pointer type,
     and cannot convert to integer types or be deduced as an integer type.
     Unless you need to be compatible with C++98/C++03 or C you should prefer
     to use <constant>nullptr</constant>  instead of <constant>NULL</constant>.
    </para>
  </section>

</section>

<section xml:id="std.support.memory" xreflabel="Dynamic Memory"><info><title>Dynamic Memory</title></info>
  <?dbhtml filename="dynamic_memory.html"?>

  <para>
    In C++98 there are six flavors each of <function>operator new</function>
    and <function>operator delete</function>, so make certain that you're
    using the right ones.
    Here are quickie descriptions of <function>operator new</function>:
  </para>
  <variablelist>
    <varlistentry>
      <term><code>void* operator new(std::size_t);</code></term>
      <listitem>
	Single object form.
        Throws <classname>std::bad_alloc</classname> on error.
        This is what most people are used to using.
      </listitem>
    </varlistentry>
    <varlistentry>
      <term><code>void* operator new(std::size_t, std::nothrow_t) noexcept;</code></term>
      <listitem>
	Single object <quote>nothrow</quote> form.
        Calls <code>operator new(std::size_t)</code> but if that throws,
        returns a null pointer instead.
      </listitem>
    </varlistentry>
    <varlistentry>
      <term><code>void* operator new[](std::size_t);</code></term>
      <listitem>
	Array <function>new</function>.
        Calls <code>operator new(std::size_t)</code> and so
	throws <classname>std::bad_alloc</classname> on error.
      </listitem>
    </varlistentry>
    <varlistentry>
      <term><code>void* operator new[](std::size_t, std::nothrow_t) noexcept;</code></term>
      <listitem>
	Array <quote>nothrow</quote> <function>new</function>.
        Calls <code>operator new[](std::size_t)</code> but if that throws,
        returns a null pointer instead.
      </listitem>
    </varlistentry>
    <varlistentry>
      <term><code>void* operator new(std::size_t, void*) noexcept;</code></term>
      <listitem>
	Non-allocating, <quote>placement</quote> single-object <function>new</function>,
        which does nothing except return its argument.
        This function cannot be replaced.
      </listitem>
    </varlistentry>
    <varlistentry>
      <term><code>void* operator new[](std::size_t, void*) noexcept;</code></term>
      <listitem>
	Non-allocating, <quote>placement</quote> array <function>new</function>,
        which also does nothing except return its argument.
        This function cannot be replaced.
      </listitem>
    </varlistentry>
   </variablelist>
   <para>
     They are distinguished by the arguments that you pass to them, like
     any other overloaded function.  The six flavors of
     <function>operator delete</function>
     are distinguished the same way, but none of them are allowed to throw
     an exception under any circumstances anyhow.  (The overloads match up
     with the ones above, for completeness' sake.)
   </para>
   <para>
     The C++ 2014 revision of the standard added two additional overloads of
     <function>operator delete</function> for <quote>sized deallocation</quote>,
     allowing the compiler to provide the size of the storage being freed.
   </para>
   <para>
     The C++ 2017 standard added even more overloads of both
     <function>operator new</function> and <function>operator delete</function>
     for allocating and deallocating storage for overaligned types.
     These overloads correspond to each of the allocating forms of
     <function>operator new</function> and <function>operator delete</function>
     but with an additional parameter of type <type>std::align_val_t</type>.
     These new overloads are not interchangeable with the versions without
     an aligment parameter, so if memory was allocated by an overload of
     <function>operator new</function> taking an alignment parameter,
     then it must be decallocated by the corresponding overload of
     <function>operator delete</function> that takes an alignment parameter.
   </para>
   <para>
     Apart from the non-allocating forms, the default versions of the array
     and nothrow <function>operator new</function> functions will all result
     in a call to either <function>operator new(std::size_t)</function> or
     <function>operator new(std::size_t, std::align_val_t)</function>,
     and similarly the default versions of the array and nothrow
     <function>operator delete</function> functions will result in a call to
     either <function>operator delete(void*)</function> or
     <function>operator delete(void*, std::align_val_t)</function>
     (or the sized versions of those).
   </para>
   <para>
     Apart from the non-allocating forms, any of these functions can be
     replaced by defining a function with the same signature in your program.
     Replacement versions must preserve certain guarantees, such as memory
     obtained from a nothrow <function>operator new</function> being free-able
     by the normal (non-nothrow) <function>operator delete</function>,
     and the sized and unsized forms of <function>operator delete</function>
     being interchangeable (because it's unspecified whether
     the compiler calls the sized delete instead of the normal one).
     The simplest way to meet the guarantees is to only replace the ordinary
     <function>operator new(size_t)</function> and
     <function>operator delete(void*)</function> and
     <function>operator delete(void*, std::size_t)</function>
     functions, and the replaced versions will be used by all of
     <function>operator new(size_t, nothrow_t)</function>,
     <function>operator new[](size_t)</function> and
     <function>operator new[](size_t, nothrow_t)</function>
     and the corresponding <function>operator delete</function> functions.
     To support types with extended alignment you may also need to replace
     <function>operator new(size_t, align_val_t)</function> and
     <function>operator delete(void*, align_val_t)</function>
     <function>operator delete(void*, size_t, align_val_t)</function>
     (which will then be used by the nothrow and array forms for
     extended alignments).
     If you do need to replace other forms (e.g. to define the nothrow
     <function>operator new</function> to allocate memory directly, so it
     works with exceptions disabled) then make sure the memory it allocates
     can still be freed by the non-nothrow forms of
     <function>operator delete</function>.
   </para>
   <para>
     If the default versions of <function>operator new(std::size_t)</function>
     and <function>operator new(size_t, std::align_val_t)</function>
     can't allocate the memory requested, they usually throw an exception
     object of type <classname>std::bad_alloc</classname> (or some class
     derived from that). However, the program can influence that behavior
     by registering a <quote>new-handler</quote>, because what
     <function>operator new</function> actually does is something like:
   </para>
<programlisting>
    while (true)
    {
      if (void* p = /* try to allocate memory */)
        return p;
      else if (std::new_handler h = std::get_new_handler ())
        h ();
      else
        throw bad_alloc{};
    }
</programlisting>
   <para>
     This means you can influence what happens on allocation failure by
     writing your own new-handler and then registering it with
     <function>std::set_new_handler</function>:
   </para>
<programlisting>
   typedef void (*PFV)();

   static char*  safety;
   static PFV    old_handler;

   void my_new_handler ()
   {
       delete[] safety;
       safety = nullptr;
       popup_window ("Dude, you are running low on heap memory.  You"
		     " should, like, close some windows, or something."
		     " The next time you run out, we're gonna burn!");
       set_new_handler (old_handler);
       return;
   }

   int main ()
   {
       safety = new char[500000];
       old_handler = set_new_handler (&amp;my_new_handler);
       ...
   }
</programlisting>

   <section xml:id="std.support.memory.notes" xreflabel="Dynamic Memory Notes"><info><title>Additional Notes</title></info>

   <para>
     Remember that it is perfectly okay to <function>delete</function> a
     null pointer!  Nothing happens, by definition.  That is not the
     same thing as deleting a pointer twice.
   </para>
   <para>
     <classname>std::bad_alloc</classname> is derived from the base
     <classname>std::exception</classname> class,
     see <xref linkend="std.diagnostics.exceptions"/>.
   </para>
   </section>

</section>

<section xml:id="std.support.termination" xreflabel="Termination"><info><title>Termination</title></info>
  <?dbhtml filename="termination.html"?>

  <section xml:id="support.termination.handlers" xreflabel="Termination Handlers"><info><title>Termination Handlers</title></info>

    <para>
      Not many changes here to
      <filename class="headerfile">&lt;cstdlib&gt;</filename>.
      You should note that the
      <function>abort()</function> function does not call the
      destructors of automatic nor static objects, so if you're
      depending on those to do cleanup, it isn't going to happen.
      (The functions registered with <function>atexit()</function>
      don't get called either, so you can forget about that
      possibility, too.)
    </para>
    <para>
      The good old <function>exit()</function> function can be a bit
      funky, too, until you look closer.  Basically, three points to
      remember are:
    </para>
    <orderedlist inheritnum="ignore" continuation="restarts">
      <listitem>
	<para>
	Static objects are destroyed in reverse order of their creation.
	</para>
      </listitem>
      <listitem>
	<para>
	Functions registered with <function>atexit()</function> are called in
	reverse order of registration, once per registration call.
	(This isn't actually new.)
	</para>
      </listitem>
      <listitem>
	<para>
	The previous two actions are <quote>interleaved,</quote> that is,
	given this pseudocode:
	</para>
<programlisting>
  extern "C or C++" void f1 ();
  extern "C or C++" void f2 ();

  static Thing obj1;
  atexit(f1);
  static Thing obj2;
  atexit(f2);
</programlisting>
	<para>
	then at a call of <function>exit()</function>,
	<varname>f2</varname> will be called, then
	<varname>obj2</varname> will be destroyed, then
	<varname>f1</varname> will be called, and finally
	<varname>obj1</varname> will be destroyed. If
	<varname>f1</varname> or <varname>f2</varname> allow an
	exception to propagate out of them, Bad Things happen.
	</para>
      </listitem>
    </orderedlist>
    <para>
      Note also that <function>atexit()</function> is only required to store 32
      functions, and the compiler/library might already be using some of
      those slots.  If you think you may run out, we recommend using
      the <function>xatexit</function>/<function>xexit</function> combination
      from <literal>libiberty</literal>, which has no such limit.
    </para>
  </section>

  <section xml:id="support.termination.verbose" xreflabel="Verbose Terminate Handler"><info><title>Verbose Terminate Handler</title></info>
  <?dbhtml filename="verbose_termination.html"?>

    <para>
      If you are having difficulty with uncaught exceptions and want a
      little bit of help debugging the causes of the core dumps, you can
      make use of a GNU extension, the verbose terminate handler.
    </para>

    <para>
      The verbose terminate handler is only available for hosted environments
      (see <xref linkend="manual.intro.setup.configure"/>) and will be used
      by default unless the library is built with
      <option>--disable-libstdcxx-verbose</option>
      or with exceptions disabled.
      If you need to enable it explicitly you can do so by calling the
      <function>std::set_terminate</function> function.
    </para>

<programlisting>
#include &lt;exception&gt;

int main()
{
  std::set_terminate(__gnu_cxx::__verbose_terminate_handler);
  ...

  throw <replaceable>anything</replaceable>;
}
</programlisting>

   <para>
     The <function>__verbose_terminate_handler</function> function
     obtains the name of the current exception, attempts to demangle
     it, and prints it to <literal>stderr</literal>.
     If the exception is derived from
     <classname>std::exception</classname> then the output from
     <function>what()</function> will be included.
   </para>

   <para>
     Any replacement termination function is required to kill the
     program without returning; this one calls <function>std::abort</function>.
   </para>

   <para>
     For example:
   </para>

<programlisting>
#include &lt;exception&gt;
#include &lt;stdexcept&gt;

struct argument_error : public std::runtime_error
{
  argument_error(const std::string&amp; s): std::runtime_error(s) { }
};

int main(int argc)
{
  std::set_terminate(__gnu_cxx::__verbose_terminate_handler);
  if (argc &gt; 5)
    throw argument_error("argc is greater than 5!");
  else
    throw argc;
}
</programlisting>

   <para>
     With the verbose terminate handler active, this gives:
   </para>

   <screen>
   <computeroutput>
   % ./a.out
   terminate called after throwing a `int'
   Aborted
   % ./a.out f f f f f f f f f f f
   terminate called after throwing an instance of `argument_error'
   what(): argc is greater than 5!
   Aborted
   </computeroutput>
   </screen>

   <para>
     The 'Aborted' line is printed by the shell after the process exits
     by calling <function>abort()</function>.
   </para>

   <para>
     As this is the default termination handler, nothing need be done to
     use it.  To go back to the previous <quote>silent death</quote>
     method, simply include
     <filename class="headerfile">&lt;exception&gt;</filename> and
     <filename class="headerfile">&lt;cstdlib&gt;</filename>, and call
   </para>

<programlisting>
     std::set_terminate(std::abort);
</programlisting>

   <para>
     After this, all calls to <function>terminate</function> will use
     <function>abort</function> as the terminate handler.
   </para>

   <para>
     Note: the verbose terminate handler will attempt to write to
     <literal>stderr</literal>.  If your application closes
     <literal>stderr</literal> or redirects it to an inappropriate location,
     <function>__verbose_terminate_handler</function> will behave in
     an unspecified manner.
   </para>

  </section>
</section>

</chapter>