buildtools.git

<sect1 id="manual.appendix.porting.backwards" xreflabel="backwards">
<?dbhtml filename="backwards.html"?>
 
<sect1info>
  <keywordset>
    <keyword>
      ISO C++
    </keyword>
    <keyword>
      backwards
    </keyword>
  </keywordset>
</sect1info>

<title>Backwards Compatibility</title>

<sect2 id="backwards.first" xreflabel="backwards.first">
<title>First</title>

<para>The first generation GNU C++ library was called libg++.  It was a
separate GNU project, although reliably paired with GCC. Rumors imply
that it had a working relationship with at least two kinds of
dinosaur.
</para>

<para>Some background: libg++ was designed and created when there was no
ISO standard to provide guidance.  Classes like linked lists are now
provided for by <classname>list&lt;T&gt;</classname> and do not need to be
created by <function>genclass</function>.  (For that matter, templates exist
now and are well-supported, whereas genclass (mostly) predates them.)
</para>

<para>There are other classes in libg++ that are not specified in the
ISO Standard (e.g., statistical analysis).  While there are a lot of
really useful things that are used by a lot of people, the Standards
Committee couldn't include everything, and so a lot of those
<quote>obvious</quote> classes didn't get included.
</para>

<para>Known Issues include many of the limitations of its immediate ancestor.</para> 

<para>Portability notes and known implementation limitations are as follows.</para>

<sect3>
  <title>No <code>ios_base</code></title>

<para> At least some older implementations don't have <code>std::ios_base</code>, so you should use <code>std::ios::badbit</code>, <code>std::ios::failbit</code> and <code>std::ios::eofbit</code> and <code>std::ios::goodbit</code>.
</para>
</sect3>

<sect3>
<title>No <code>cout</code> in <code>ostream.h</code>, no <code>cin</code> in <code>istream.h</code></title>

<para>
	In earlier versions of the standard,
	<filename class="headerfile">fstream.h</filename>,
	<filename class="headerfile">ostream.h</filename>
	and <filename class="headerfile">istream.h</filename>
	used to define
	<code>cout</code>, <code>cin</code> and so on. ISO C++ specifies that one needs to include
	<filename class="headerfile">iostream</filename>
	explicitly to get the required definitions.
 </para>
<para> Some include adjustment may be required.</para>

<para>This project is no longer maintained or supported, and the sources
archived. For the desperate,
the <ulink url="http://gcc.gnu.org/extensions.html">GCC extensions
page</ulink> describes where to find the last libg++ source. The code is
considered replaced and rewritten.
</para>
</sect3>
</sect2>

<sect2 id="backwards.second" xreflabel="backwards.second">
<title>Second</title>

<para> 
  The second generation GNU C++ library was called libstdc++, or
  libstdc++-v2. It spans the time between libg++ and pre-ISO C++
  standardization and is usually associated with the following GCC
  releases: egcs 1.x, gcc 2.95, and gcc 2.96.
</para>

<para> 
  The STL portions of this library are based on SGI/HP STL release 3.11.
</para>

<para>
  This project is no longer maintained or supported, and the sources
  archived.  The code is considered replaced and rewritten.
</para>

<para>
  Portability notes and known implementation limitations are as follows.
</para>

<sect3>
  <title>Namespace <code>std::</code> not supported</title>

  <para>
    Some care is required to support C++ compiler and or library
    implementation that do not have the standard library in
    <code>namespace std</code>.
  </para>

  <para>
    The following sections list some possible solutions to support compilers
    that cannot ignore <code>std::</code>-qualified names.
  </para>
  
  <para> 
    First, see if the compiler has a flag for this. Namespace
    back-portability-issues are generally not a problem for g++
    compilers that do not have libstdc++ in <code>std::</code>, as the
    compilers use <code>-fno-honor-std</code> (ignore
    <code>std::</code>, <code>:: = std::</code>) by default. That is,
    the responsibility for enabling or disabling <code>std::</code> is
    on the user; the maintainer does not have to care about it. This
    probably applies to some other compilers as well.
  </para>

  <para>
    Second, experiment with a variety of pre-processor tricks.
  </para>

  <para> 
    By defining <code>std</code> as a macro, fully-qualified namespace
    calls become global. Volia.
  </para>
	      
<programlisting>
#ifdef WICKEDLY_OLD_COMPILER
# define std
#endif
</programlisting>

  <para>
    Thanks to Juergen Heinzl who posted this solution on gnu.gcc.help.
  </para>

  <para>
    Another pre-processor based approach is to define a macro
    <code>NAMESPACE_STD</code>, which is defined to either
    <quote> </quote> or <quote>std</quote> based on a compile-type
    test. On GNU systems, this can be done with autotools by means of
    an autoconf test (see below) for <code>HAVE_NAMESPACE_STD</code>,
    then using that to set a value for the <code>NAMESPACE_STD</code>
    macro.  At that point, one is able to use
    <code>NAMESPACE_STD::string</code>, which will evaluate to
    <code>std::string</code> or <code>::string</code> (i.e., in the
    global namespace on systems that do not put <code>string</code> in
    <code>std::</code>).
  </para>

<programlisting>
dnl @synopsis AC_CXX_NAMESPACE_STD
dnl
dnl If the compiler supports namespace std, define
dnl HAVE_NAMESPACE_STD.
dnl
dnl @category Cxx
dnl @author Todd Veldhuizen
dnl @author Luc Maisonobe &lt;luc@spaceroots.org&gt;
dnl @version 2004-02-04
dnl @license AllPermissive
AC_DEFUN([AC_CXX_NAMESPACE_STD], [
  AC_CACHE_CHECK(if g++ supports namespace std,
  ac_cv_cxx_have_std_namespace,
  [AC_LANG_SAVE
  AC_LANG_CPLUSPLUS
  AC_TRY_COMPILE([#include &lt;iostream&gt; 
                  std::istream&amp; is = std::cin;],,
  ac_cv_cxx_have_std_namespace=yes, ac_cv_cxx_have_std_namespace=no)
  AC_LANG_RESTORE
  ])
  if test "$ac_cv_cxx_have_std_namespace" = yes; then
    AC_DEFINE(HAVE_NAMESPACE_STD,,[Define if g++ supports namespace std. ])
  fi
])
</programlisting>
</sect3>

<sect3>
<title>Illegal iterator usage</title>
<para>
  The following illustrate implementation-allowed illegal iterator
  use, and then correct use.
</para>

<itemizedlist> 
  <listitem>
    <para>
      you cannot do <code>ostream::operator&lt;&lt;(iterator)</code>
      to print the address of the iterator =&gt; use
      <code>operator&lt;&lt; &amp;*iterator</code> instead
    </para>
  </listitem>
  <listitem>
    <para>
      you cannot clear an iterator's reference (<code>iterator =
      0</code>) =&gt; use <code>iterator = iterator_type();</code>
    </para>
  </listitem>
  <listitem>
    <para>
      <code>if (iterator)</code> won't work any more =&gt; use
      <code>if (iterator != iterator_type())</code> 
    </para>
  </listitem>
</itemizedlist>
</sect3>

<sect3>
  <title><code>isspace</code> from <filename class="headerfile">cctype</filename> is a macro
  </title>

  <para> 
    Glibc 2.0.x and 2.1.x define <filename
    class="headerfile">ctype.h</filename> functionality as macros
    (isspace, isalpha etc.).
  </para>

  <para>
    This implementations of libstdc++, however, keep these functions
    as macros, and so it is not back-portable to use fully qualified
    names. For example:
  </para>

<programlisting> 
#include &lt;cctype&gt; 
int main() { std::isspace('X'); } 
</programlisting>

<para>
  Results in something like this:
</para>

<programlisting> 
std:: (__ctype_b[(int) ( ( 'X' ) )] &amp; (unsigned short int) _ISspace ) ; 
</programlisting>

<para> 
  A solution is to modify a header-file so that the compiler tells
  <filename class="headerfile">ctype.h</filename> to define functions
  instead of macros:
</para>

<programlisting>
// This keeps isalnum, et al from being propagated as macros. 
#if __linux__
# define __NO_CTYPE 1
#endif
</programlisting>

<para>
  Then, include <filename class="headerfile">ctype.h</filename>
</para>

<para>
  Another problem arises if you put a <code>using namespace
  std;</code> declaration at the top, and include <filename
  class="headerfile">ctype.h</filename>. This will result in
  ambiguities between the definitions in the global namespace
  (<filename class="headerfile">ctype.h</filename>) and the
  definitions in namespace <code>std::</code>
  (<code>&lt;cctype&gt;</code>).
</para>
</sect3>

<sect3>
<title>No <code>vector::at</code>, <code>deque::at</code>, <code>string::at</code></title>

<para>
  One solution is to add an autoconf-test for this:
</para>

<programlisting>
AC_MSG_CHECKING(for container::at)
AC_TRY_COMPILE(
[
#include &lt;vector&gt;
#include &lt;deque&gt;
#include &lt;string&gt;
	
using namespace std;
],
[
deque&lt;int&gt; test_deque(3);
test_deque.at(2);
vector&lt;int&gt; test_vector(2);
test_vector.at(1);
string test_string(<quote>test_string</quote>);
test_string.at(3);
],
[AC_MSG_RESULT(yes)
AC_DEFINE(HAVE_CONTAINER_AT)],
[AC_MSG_RESULT(no)])
</programlisting>

<para>
  If you are using other (non-GNU) compilers it might be a good idea
  to check for <code>string::at</code> separately.
</para>

</sect3>

<sect3>
<title>No <code>std::char_traits&lt;char&gt;::eof</code></title>

<para>
  Use some kind of autoconf test, plus this:
</para>      

<programlisting>
#ifdef HAVE_CHAR_TRAITS
#define CPP_EOF std::char_traits&lt;char&gt;::eof()
#else
#define CPP_EOF EOF
#endif
</programlisting>

</sect3>

<sect3>
<title>No <code>string::clear</code></title>

<para>
  There are two functions for deleting the contents of a string:
  <code>clear</code> and <code>erase</code> (the latter returns the
  string).
</para>
      
<programlisting>
void 
clear() { _M_mutate(0, this-&gt;size(), 0); }
</programlisting>

<programlisting>
basic_string&amp; 
erase(size_type __pos = 0, size_type __n = npos)
{ 
  return this-&gt;replace(_M_check(__pos), _M_fold(__pos, __n),
                          _M_data(), _M_data()); 
}
</programlisting>

<para>
  Unfortunately, <code>clear</code> is not implemented in this
  version, so you should use <code>erase</code> (which is probably
  faster than <code>operator=(charT*)</code>).
</para>
</sect3>

<sect3>
<title>
  Removal of <code>ostream::form</code> and <code>istream::scan</code>
  extensions
</title>

<para> 
  These are no longer supported. Please use stringstreams instead.
</para>
</sect3>

<sect3>
<title>No <code>basic_stringbuf</code>, <code>basic_stringstream</code></title>

<para>
  Although the ISO standard <code>i/ostringstream</code>-classes are
  provided, (<filename class="headerfile">sstream</filename>), for
  compatibility with older implementations the pre-ISO
  <code>i/ostrstream</code> (<filename
  class="headerfile">strstream</filename>) interface is also provided,
  with these caveats:
</para>

<itemizedlist>
  <listitem>
    <para> 
      <code>strstream</code> is considered to be deprecated
    </para>
  </listitem>
  <listitem>
    <para> 
      <code>strstream</code> is limited to <code>char</code>
    </para>
  </listitem>
  <listitem>
    <para> 
      with <code>ostringstream</code> you don't have to take care of
      terminating the string or freeing its memory 
    </para>
  </listitem>
  <listitem>
    <para> 
      <code>istringstream</code> can be re-filled (clear();
      str(input);) 
    </para>
  </listitem>
</itemizedlist>

<para>
  You can then use output-stringstreams like this:
</para>
       
<programlisting>
#ifdef HAVE_SSTREAM
# include &lt;sstream&gt;
#else
# include &lt;strstream&gt;
#endif

#ifdef HAVE_SSTREAM
  std::ostringstream oss;
#else
  std::ostrstream oss;
#endif

oss &lt;&lt; <quote>Name=</quote> &lt;&lt; m_name &lt;&lt; <quote>, number=</quote> &lt;&lt; m_number &lt;&lt; std::endl;
...
#ifndef HAVE_SSTREAM
  oss &lt;&lt; std::ends; // terminate the char*-string
#endif

// str() returns char* for ostrstream and a string for ostringstream
// this also causes ostrstream to think that the buffer's memory
// is yours
m_label.set_text(oss.str());
#ifndef HAVE_SSTREAM
  // let the ostrstream take care of freeing the memory
  oss.freeze(false);
#endif
</programlisting>

<para>
      Input-stringstreams can be used similarly:
</para>
      
<programlisting>
std::string input;
...
#ifdef HAVE_SSTREAM
std::istringstream iss(input);
#else
std::istrstream iss(input.c_str());
#endif

int i;
iss &gt;&gt; i; 
</programlisting>

<para> One (the only?) restriction is that an istrstream cannot be re-filled:
</para>
      
<programlisting>
std::istringstream iss(numerator);
iss &gt;&gt; m_num;
// this is not possible with istrstream
iss.clear();
iss.str(denominator);
iss &gt;&gt; m_den;
</programlisting>
 
<para>
If you don't care about speed, you can put these conversions in
      a template-function:
</para>      
<programlisting>
template &lt;class X&gt;
void fromString(const string&amp; input, X&amp; any)
{
#ifdef HAVE_SSTREAM
std::istringstream iss(input);
#else
std::istrstream iss(input.c_str());
#endif
X temp;
iss &gt;&gt; temp;
if (iss.fail())
throw runtime_error(..)
any = temp;
}
</programlisting>

<para> 
  Another example of using stringstreams is in <link
  linkend="strings.string.shrink">this howto</link>.
</para>

<para> There is additional information in the libstdc++-v2 info files, in
particular <quote>info iostream</quote>.
</para>
</sect3>

<sect3>
  <title>Little or no wide character support</title>
  <para>
    Classes <classname>wstring</classname> and
    <classname>char_traits&lt;wchar_t&gt;</classname> are
    not supported.
  </para>
</sect3>

<sect3>
  <title>No templatized iostreams</title>
  <para>
    Classes <classname>wfilebuf</classname> and
    <classname>wstringstream</classname> are not supported.
  </para>
</sect3>

<sect3>
<title>Thread safety issues</title>

  <para>
    Earlier GCC releases had a somewhat different approach to
    threading configuration and proper compilation.  Before GCC 3.0,
    configuration of the threading model was dictated by compiler
    command-line options and macros (both of which were somewhat
    thread-implementation and port-specific).  There were no
    guarantees related to being able to link code compiled with one
    set of options and macro setting with another set.
  </para>

  <para>
    For GCC 3.0, configuration of the threading model used with
    libraries and user-code is performed when GCC is configured and
    built using the --enable-threads and --disable-threads options.
    The ABI is stable for symbol name-mangling and limited functional
    compatibility exists between code compiled under different
    threading models.
  </para>

   <para>
     The libstdc++ library has been designed so that it can be used in
     multithreaded applications (with libstdc++-v2 this was only true
     of the STL parts.)  The first problem is finding a
     <emphasis>fast</emphasis> method of implementation portable to
     all platforms.  Due to historical reasons, some of the library is
     written against per-CPU-architecture spinlocks and other parts
     against the gthr.h abstraction layer which is provided by gcc.  A
     minor problem that pops up every so often is different
     interpretations of what &quot;thread-safe&quot; means for a
     library (not a general program).  We currently use the <ulink
     url="http://www.sgi.com/tech/stl/thread_safety.html">same
     definition that SGI</ulink> uses for their STL subset.  However,
     the exception for read-only containers only applies to the STL
     components. This definition is widely-used and something similar
     will be used in the next version of the C++ standard library.
   </para>

   <para>
     Here is a small link farm to threads (no pun) in the mail
     archives that discuss the threading problem.  Each link is to the
     first relevant message in the thread; from there you can use
     &quot;Thread Next&quot; to move down the thread.  This farm is in
     latest-to-oldest order.
   </para>

      <itemizedlist>
        <listitem>
	  <para>
	    Our threading expert Loren gives a breakdown of <ulink
	    url="http://gcc.gnu.org/ml/libstdc++/2001-10/msg00024.html">the
	    six situations involving threads</ulink> for the 3.0
	    release series.
	  </para>
      </listitem>
        <listitem>
	  <para>
	    <ulink url="http://gcc.gnu.org/ml/libstdc++/2001-05/msg00384.html">
        This message</ulink> inspired a recent updating of issues with
        threading and the SGI STL library.  It also contains some
        example POSIX-multithreaded STL code.
	  </para>
	</listitem>
      </itemizedlist>

   <para> 
     (A large selection of links to older messages has been removed;
     many of the messages from 1999 were lost in a disk crash, and the
     few people with access to the backup tapes have been too swamped
     with work to restore them.  Many of the points have been
     superseded anyhow.)
   </para>
</sect3>

</sect2>

<sect2 id="backwards.third" xreflabel="backwards.third">
<title>Third</title>

<para> The third generation GNU C++ library is called libstdc++, or
libstdc++-v3.
</para>

      <para>The subset commonly known as the Standard Template Library
         (chapters 23 through 25, mostly) is adapted from the final release
         of the SGI STL (version 3.3), with extensive changes.
      </para>

      <para>A more formal description of the V3 goals can be found in the
         official <ulink url="../17_intro/DESIGN">design document</ulink>.
      </para>

<para>Portability notes and known implementation limitations are as follows.</para>

<sect3>
<title>Pre-ISO headers moved to backwards or removed</title>

<para> The pre-ISO C++ headers
      (<code>iostream.h</code>, <code>defalloc.h</code> etc.) are
      available, unlike previous libstdc++ versions, but inclusion
      generates a warning that you are using deprecated headers.
</para>

    <para>This compatibility layer is constructed by including the
    standard C++ headers, and injecting any items in
    <code>std::</code> into the global namespace.
   </para>
   <para>For those of you new to ISO C++ (welcome, time travelers!), no,
      that isn't a typo. Yes, the headers really have new names.
      Marshall Cline's C++ FAQ Lite has a good explanation in <ulink url="http://www.parashift.com/c++-faq-lite/coding-standards.html#faq-27.4">item
      [27.4]</ulink>.
   </para>

<para> Some include adjustment may be required. What follows is an
autoconf test that defines <code>PRE_STDCXX_HEADERS</code> when they
exist.</para>

<programlisting>
# AC_HEADER_PRE_STDCXX
AC_DEFUN([AC_HEADER_PRE_STDCXX], [
  AC_CACHE_CHECK(for pre-ISO C++ include files,
  ac_cv_cxx_pre_stdcxx,
  [AC_LANG_SAVE
  AC_LANG_CPLUSPLUS
  ac_save_CXXFLAGS="$CXXFLAGS"
  CXXFLAGS="$CXXFLAGS -Wno-deprecated"	

  # Omit defalloc.h, as compilation with newer compilers is problematic.
  AC_TRY_COMPILE([
  #include &lt;new.h&gt;
  #include &lt;iterator.h&gt;
  #include &lt;alloc.h&gt;
  #include &lt;set.h&gt;
  #include &lt;hashtable.h&gt;
  #include &lt;hash_set.h&gt;
  #include &lt;fstream.h&gt;
  #include &lt;tempbuf.h&gt;
  #include &lt;istream.h&gt;
  #include &lt;bvector.h&gt;
  #include &lt;stack.h&gt;
  #include &lt;rope.h&gt;
  #include &lt;complex.h&gt;
  #include &lt;ostream.h&gt;
  #include &lt;heap.h&gt;
  #include &lt;iostream.h&gt;
  #include &lt;function.h&gt;
  #include &lt;multimap.h&gt;
  #include &lt;pair.h&gt;
  #include &lt;stream.h&gt;
  #include &lt;iomanip.h&gt;
  #include &lt;slist.h&gt;
  #include &lt;tree.h&gt;
  #include &lt;vector.h&gt;
  #include &lt;deque.h&gt;
  #include &lt;multiset.h&gt;
  #include &lt;list.h&gt;
  #include &lt;map.h&gt;
  #include &lt;algobase.h&gt;
  #include &lt;hash_map.h&gt;
  #include &lt;algo.h&gt;
  #include &lt;queue.h&gt;
  #include &lt;streambuf.h&gt;
  ],,
  ac_cv_cxx_pre_stdcxx=yes, ac_cv_cxx_pre_stdcxx=no)
  CXXFLAGS="$ac_save_CXXFLAGS"
  AC_LANG_RESTORE
  ])
  if test "$ac_cv_cxx_pre_stdcxx" = yes; then
    AC_DEFINE(PRE_STDCXX_HEADERS,,[Define if pre-ISO C++ header files are present. ])
  fi
])
</programlisting>

<para>Porting between pre-ISO headers and ISO headers is simple: headers
like <filename class="headerfile">vector.h</filename> can be replaced with <filename class="headerfile">vector</filename> and a using
directive <code>using namespace std;</code> can be put at the global
scope. This should be enough to get this code compiling, assuming the
other usage is correct.
</para>
</sect3>

<sect3>
<title>Extension headers hash_map, hash_set moved to ext or backwards</title>

      <para>At this time most of the features of the SGI STL extension have been
         replaced by standardized libraries.
         In particular, the unordered_map and unordered_set containers of TR1
         are suitable replacement for the non-standard hash_map and hash_set
         containers in the SGI STL. 
      </para>
<para> Header files <filename class="headerfile">hash_map</filename> and <filename class="headerfile">hash_set</filename> moved
to <filename class="headerfile">ext/hash_map</filename> and  <filename class="headerfile">ext/hash_set</filename>,
respectively. At the same time, all types in these files are enclosed
in <code>namespace __gnu_cxx</code>. Later versions move deprecate
these files, and suggest using TR1's  <filename class="headerfile">unordered_map</filename>
and  <filename class="headerfile">unordered_set</filename> instead.
</para>

      <para>The extensions are no longer in the global or <code>std</code>
         namespaces, instead they are declared in the <code>__gnu_cxx</code>
         namespace. For maximum portability, consider defining a namespace
         alias to use to talk about extensions, e.g.:
      </para>
      <programlisting>
      #ifdef __GNUC__
      #if __GNUC__ &lt; 3
        #include &lt;hash_map.h&gt;
        namespace extension { using ::hash_map; }; // inherit globals
      #else
        #include &lt;backward/hash_map&gt;
        #if __GNUC__ == 3 &amp;&amp; __GNUC_MINOR__ == 0
          namespace extension = std;               // GCC 3.0
        #else
          namespace extension = ::__gnu_cxx;       // GCC 3.1 and later
        #endif
      #endif
      #else      // ...  there are other compilers, right?
        namespace extension = std;
      #endif

      extension::hash_map&lt;int,int&gt; my_map; 
      </programlisting>
      <para>This is a bit cleaner than defining typedefs for all the
         instantiations you might need.
      </para>
 

<para>The following autoconf tests check for working HP/SGI hash containers.
</para>

<programlisting>
# AC_HEADER_EXT_HASH_MAP
AC_DEFUN([AC_HEADER_EXT_HASH_MAP], [
  AC_CACHE_CHECK(for ext/hash_map,
  ac_cv_cxx_ext_hash_map,
  [AC_LANG_SAVE
  AC_LANG_CPLUSPLUS
  ac_save_CXXFLAGS="$CXXFLAGS"
  CXXFLAGS="$CXXFLAGS -Werror"	
  AC_TRY_COMPILE([#include &lt;ext/hash_map&gt;], [using __gnu_cxx::hash_map;],
  ac_cv_cxx_ext_hash_map=yes, ac_cv_cxx_ext_hash_map=no)
  CXXFLAGS="$ac_save_CXXFLAGS"
  AC_LANG_RESTORE
  ])
  if test "$ac_cv_cxx_ext_hash_map" = yes; then
    AC_DEFINE(HAVE_EXT_HASH_MAP,,[Define if ext/hash_map is present. ])
  fi
])
</programlisting>

<programlisting>
# AC_HEADER_EXT_HASH_SET
AC_DEFUN([AC_HEADER_EXT_HASH_SET], [
  AC_CACHE_CHECK(for ext/hash_set,
  ac_cv_cxx_ext_hash_set,
  [AC_LANG_SAVE
  AC_LANG_CPLUSPLUS
  ac_save_CXXFLAGS="$CXXFLAGS"
  CXXFLAGS="$CXXFLAGS -Werror"	
  AC_TRY_COMPILE([#include &lt;ext/hash_set&gt;], [using __gnu_cxx::hash_set;],
  ac_cv_cxx_ext_hash_set=yes, ac_cv_cxx_ext_hash_set=no)
  CXXFLAGS="$ac_save_CXXFLAGS"
  AC_LANG_RESTORE
  ])
  if test "$ac_cv_cxx_ext_hash_set" = yes; then
    AC_DEFINE(HAVE_EXT_HASH_SET,,[Define if ext/hash_set is present. ])
  fi
])
</programlisting>
</sect3>

<sect3>
<title>No <code>ios::nocreate/ios::noreplace</code>.
</title>

<para> The existence of <code>ios::nocreate</code> being used for
input-streams has been confirmed, most probably because the author
thought it would be more correct to specify nocreate explicitly.  So
it can be left out for input-streams.
</para>

<para>For output streams, <quote>nocreate</quote> is probably the default,
unless you specify <code>std::ios::trunc</code> ? To be safe, you can
open the file for reading, check if it has been opened, and then
decide whether you want to create/replace or not. To my knowledge,
even older implementations support <code>app</code>, <code>ate</code>
and <code>trunc</code> (except for <code>app</code> ?).
</para>
</sect3>

<sect3>
<title>
No <code>stream::attach(int fd)</code>
</title>

<para>
      Phil Edwards writes: It was considered and rejected for the ISO
      standard.  Not all environments use file descriptors.  Of those
      that do, not all of them use integers to represent them.
    </para>  

<para>
      For a portable solution (among systems which use
      file descriptors), you need to implement a subclass of
      <code>std::streambuf</code> (or
      <code>std::basic_streambuf&lt;..&gt;</code>) which opens a file
      given a descriptor, and then pass an instance of this to the
      stream-constructor. 
    </para>

<para>
      An extension is available that implements this.
      <filename class="headerfile">ext/stdio_filebuf.h</filename> contains a derived class called
      <ulink url="http://gcc.gnu.org/onlinedocs/libstdc++/latest-doxygen/class____gnu__cxx_1_1stdio__filebuf.html"><code>__gnu_cxx::stdio_filebuf</code></ulink>.
      This class can be constructed from a C <code>FILE*</code> or a file
      descriptor, and provides the <code>fd()</code> function.
    </para>

<para>
 For another example of this, refer to
      <ulink url="http://www.josuttis.com/cppcode/fdstream.html">fdstream example</ulink> 
      by Nicolai Josuttis.
</para>
</sect3>

<sect3>
<title>
Support for C++98 dialect.
</title>

<para>Check for complete library coverage of the C++1998/2003 standard.
</para>

<programlisting>
# AC_HEADER_STDCXX_98
AC_DEFUN([AC_HEADER_STDCXX_98], [
  AC_CACHE_CHECK(for ISO C++ 98 include files,
  ac_cv_cxx_stdcxx_98,
  [AC_LANG_SAVE
  AC_LANG_CPLUSPLUS
  AC_TRY_COMPILE([
    #include &lt;cassert&gt;
    #include &lt;cctype&gt;
    #include &lt;cerrno&gt;
    #include &lt;cfloat&gt;
    #include &lt;ciso646&gt;
    #include &lt;climits&gt;
    #include &lt;clocale&gt;
    #include &lt;cmath&gt;
    #include &lt;csetjmp&gt;
    #include &lt;csignal&gt;
    #include &lt;cstdarg&gt;
    #include &lt;cstddef&gt;
    #include &lt;cstdio&gt;
    #include &lt;cstdlib&gt;
    #include &lt;cstring&gt;
    #include &lt;ctime&gt;

    #include &lt;algorithm&gt;
    #include &lt;bitset&gt;
    #include &lt;complex&gt;
    #include &lt;deque&gt;
    #include &lt;exception&gt;
    #include &lt;fstream&gt;
    #include &lt;functional&gt;
    #include &lt;iomanip&gt;
    #include &lt;ios&gt;
    #include &lt;iosfwd&gt;
    #include &lt;iostream&gt;
    #include &lt;istream&gt;
    #include &lt;iterator&gt;
    #include &lt;limits&gt;
    #include &lt;list&gt;
    #include &lt;locale&gt;
    #include &lt;map&gt;
    #include &lt;memory&gt;
    #include &lt;new&gt;
    #include &lt;numeric&gt;
    #include &lt;ostream&gt;
    #include &lt;queue&gt;
    #include &lt;set&gt;
    #include &lt;sstream&gt;
    #include &lt;stack&gt;
    #include &lt;stdexcept&gt;
    #include &lt;streambuf&gt;
    #include &lt;string&gt;
    #include &lt;typeinfo&gt;
    #include &lt;utility&gt;
    #include &lt;valarray&gt;
    #include &lt;vector&gt;
  ],,
  ac_cv_cxx_stdcxx_98=yes, ac_cv_cxx_stdcxx_98=no)
  AC_LANG_RESTORE
  ])
  if test "$ac_cv_cxx_stdcxx_98" = yes; then
    AC_DEFINE(STDCXX_98_HEADERS,,[Define if ISO C++ 1998 header files are present. ])
  fi
])
</programlisting>
</sect3>

<sect3>
<title>
Support for C++TR1 dialect.
</title>

<para>Check for library coverage of the TR1 standard.
</para>

<programlisting>
# AC_HEADER_STDCXX_TR1
AC_DEFUN([AC_HEADER_STDCXX_TR1], [
  AC_CACHE_CHECK(for ISO C++ TR1 include files,
  ac_cv_cxx_stdcxx_tr1,
  [AC_LANG_SAVE
  AC_LANG_CPLUSPLUS
  AC_TRY_COMPILE([
  #include &lt;tr1/array&gt;
  #include &lt;tr1/ccomplex&gt;
  #include &lt;tr1/cctype&gt;
  #include &lt;tr1/cfenv&gt;
  #include &lt;tr1/cfloat&gt;
  #include &lt;tr1/cinttypes&gt;
  #include &lt;tr1/climits&gt;
  #include &lt;tr1/cmath&gt;
  #include &lt;tr1/complex&gt;
  #include &lt;tr1/cstdarg&gt;
  #include &lt;tr1/cstdbool&gt;
  #include &lt;tr1/cstdint&gt;
  #include &lt;tr1/cstdio&gt;
  #include &lt;tr1/cstdlib&gt;
  #include &lt;tr1/ctgmath&gt;
  #include &lt;tr1/ctime&gt;
  #include &lt;tr1/cwchar&gt;
  #include &lt;tr1/cwctype&gt;
  #include &lt;tr1/functional&gt;
  #include &lt;tr1/memory&gt;
  #include &lt;tr1/random&gt;
  #include &lt;tr1/regex&gt;
  #include &lt;tr1/tuple&gt;
  #include &lt;tr1/type_traits&gt;
  #include &lt;tr1/unordered_set&gt;
  #include &lt;tr1/unordered_map&gt;
  #include &lt;tr1/utility&gt;
  ],,
  ac_cv_cxx_stdcxx_tr1=yes, ac_cv_cxx_stdcxx_tr1=no)
  AC_LANG_RESTORE
  ])
  if test "$ac_cv_cxx_stdcxx_tr1" = yes; then
    AC_DEFINE(STDCXX_TR1_HEADERS,,[Define if ISO C++ TR1 header files are present. ])
  fi
])
</programlisting>

<para>An alternative is to check just for specific TR1 includes, such as &lt;unordered_map&gt; and &lt;unordered_set&gt;.
</para>

<programlisting>
# AC_HEADER_TR1_UNORDERED_MAP
AC_DEFUN([AC_HEADER_TR1_UNORDERED_MAP], [
  AC_CACHE_CHECK(for tr1/unordered_map,
  ac_cv_cxx_tr1_unordered_map,
  [AC_LANG_SAVE
  AC_LANG_CPLUSPLUS
  AC_TRY_COMPILE([#include &lt;tr1/unordered_map&gt;], [using std::tr1::unordered_map;],
  ac_cv_cxx_tr1_unordered_map=yes, ac_cv_cxx_tr1_unordered_map=no)
  AC_LANG_RESTORE
  ])
  if test "$ac_cv_cxx_tr1_unordered_map" = yes; then
    AC_DEFINE(HAVE_TR1_UNORDERED_MAP,,[Define if tr1/unordered_map is present. ])
  fi
])
</programlisting>

<programlisting>
# AC_HEADER_TR1_UNORDERED_SET
AC_DEFUN([AC_HEADER_TR1_UNORDERED_SET], [
  AC_CACHE_CHECK(for tr1/unordered_set,
  ac_cv_cxx_tr1_unordered_set,
  [AC_LANG_SAVE
  AC_LANG_CPLUSPLUS
  AC_TRY_COMPILE([#include &lt;tr1/unordered_set&gt;], [using std::tr1::unordered_set;],
  ac_cv_cxx_tr1_unordered_set=yes, ac_cv_cxx_tr1_unordered_set=no)
  AC_LANG_RESTORE
  ])
  if test "$ac_cv_cxx_tr1_unordered_set" = yes; then
    AC_DEFINE(HAVE_TR1_UNORDERED_SET,,[Define if tr1/unordered_set is present. ])
  fi
])
</programlisting>
</sect3>


<sect3>
<title>
Support for C++0x dialect.
</title>

<para>Check for baseline language coverage in the compiler for the C++0xstandard.
</para>

<programlisting>
# AC_COMPILE_STDCXX_OX
AC_DEFUN([AC_COMPILE_STDCXX_0X], [
  AC_CACHE_CHECK(if g++ supports C++0x features without additional flags,
  ac_cv_cxx_compile_cxx0x_native,
  [AC_LANG_SAVE
  AC_LANG_CPLUSPLUS
  AC_TRY_COMPILE([
  template &lt;typename T&gt;
    struct check 
    {
      static_assert(sizeof(int) &lt;= sizeof(T), "not big enough");
    };

    typedef check&lt;check&lt;bool&gt;&gt; right_angle_brackets;

    int a;
    decltype(a) b;

    typedef check&lt;int&gt; check_type;
    check_type c;
    check_type&amp;&amp; cr = c;],,
  ac_cv_cxx_compile_cxx0x_native=yes, ac_cv_cxx_compile_cxx0x_native=no)
  AC_LANG_RESTORE
  ])

  AC_CACHE_CHECK(if g++ supports C++0x features with -std=c++0x,
  ac_cv_cxx_compile_cxx0x_cxx,
  [AC_LANG_SAVE
  AC_LANG_CPLUSPLUS
  ac_save_CXXFLAGS="$CXXFLAGS"
  CXXFLAGS="$CXXFLAGS -std=c++0x"	
  AC_TRY_COMPILE([
  template &lt;typename T&gt;
    struct check 
    {
      static_assert(sizeof(int) &lt;= sizeof(T), "not big enough");
    };

    typedef check&lt;check&lt;bool&gt;&gt; right_angle_brackets;

    int a;
    decltype(a) b;

    typedef check&lt;int&gt; check_type;
    check_type c;
    check_type&amp;&amp; cr = c;],,
  ac_cv_cxx_compile_cxx0x_cxx=yes, ac_cv_cxx_compile_cxx0x_cxx=no)
  CXXFLAGS="$ac_save_CXXFLAGS"
  AC_LANG_RESTORE
  ])

  AC_CACHE_CHECK(if g++ supports C++0x features with -std=gnu++0x,
  ac_cv_cxx_compile_cxx0x_gxx,
  [AC_LANG_SAVE
  AC_LANG_CPLUSPLUS
  ac_save_CXXFLAGS="$CXXFLAGS"
  CXXFLAGS="$CXXFLAGS -std=gnu++0x"	
  AC_TRY_COMPILE([
  template &lt;typename T&gt;
    struct check 
    {
      static_assert(sizeof(int) &lt;= sizeof(T), "not big enough");
    };

    typedef check&lt;check&lt;bool&gt;&gt; right_angle_brackets;

    int a;
    decltype(a) b;

    typedef check&lt;int&gt; check_type;
    check_type c;
    check_type&amp;&amp; cr = c;],,
  ac_cv_cxx_compile_cxx0x_gxx=yes, ac_cv_cxx_compile_cxx0x_gxx=no)
  CXXFLAGS="$ac_save_CXXFLAGS"
  AC_LANG_RESTORE
  ])

  if test "$ac_cv_cxx_compile_cxx0x_native" = yes || 
     test "$ac_cv_cxx_compile_cxx0x_cxx" = yes || 
     test "$ac_cv_cxx_compile_cxx0x_gxx" = yes; then
    AC_DEFINE(HAVE_STDCXX_0X,,[Define if g++ supports C++0x features. ])
  fi
])
</programlisting>


<para>Check for library coverage of the C++0xstandard.
</para>

<programlisting>
# AC_HEADER_STDCXX_0X
AC_DEFUN([AC_HEADER_STDCXX_0X], [
  AC_CACHE_CHECK(for ISO C++ 0x include files,
  ac_cv_cxx_stdcxx_0x,
  [AC_REQUIRE([AC_COMPILE_STDCXX_0X])
  AC_LANG_SAVE
  AC_LANG_CPLUSPLUS
  ac_save_CXXFLAGS="$CXXFLAGS"
  CXXFLAGS="$CXXFLAGS -std=gnu++0x"	

  AC_TRY_COMPILE([
    #include &lt;cassert&gt;
    #include &lt;ccomplex&gt;
    #include &lt;cctype&gt;
    #include &lt;cerrno&gt;
    #include &lt;cfenv&gt;
    #include &lt;cfloat&gt;
    #include &lt;cinttypes&gt;
    #include &lt;ciso646&gt;
    #include &lt;climits&gt;
    #include &lt;clocale&gt;
    #include &lt;cmath&gt;
    #include &lt;csetjmp&gt;
    #include &lt;csignal&gt;
    #include &lt;cstdarg&gt;
    #include &lt;cstdbool&gt;
    #include &lt;cstddef&gt;
    #include &lt;cstdint&gt;
    #include &lt;cstdio&gt;
    #include &lt;cstdlib&gt;
    #include &lt;cstring&gt;
    #include &lt;ctgmath&gt;
    #include &lt;ctime&gt;
    #include &lt;cwchar&gt;
    #include &lt;cwctype&gt;

    #include &lt;algorithm&gt;
    #include &lt;array&gt;
    #include &lt;bitset&gt;
    #include &lt;complex&gt;
    #include &lt;deque&gt;
    #include &lt;exception&gt;
    #include &lt;fstream&gt;
    #include &lt;functional&gt;
    #include &lt;iomanip&gt;
    #include &lt;ios&gt;
    #include &lt;iosfwd&gt;
    #include &lt;iostream&gt;
    #include &lt;istream&gt;
    #include &lt;iterator&gt;
    #include &lt;limits&gt;
    #include &lt;list&gt;
    #include &lt;locale&gt;
    #include &lt;map&gt;
    #include &lt;memory&gt;
    #include &lt;new&gt;
    #include &lt;numeric&gt;
    #include &lt;ostream&gt;
    #include &lt;queue&gt;
    #include &lt;random&gt;
    #include &lt;regex&gt;
    #include &lt;set&gt;
    #include &lt;sstream&gt;
    #include &lt;stack&gt;
    #include &lt;stdexcept&gt;
    #include &lt;streambuf&gt;
    #include &lt;string&gt;
    #include &lt;tuple&gt;
    #include &lt;typeinfo&gt;
    #include &lt;type_traits&gt;
    #include &lt;unordered_map&gt;
    #include &lt;unordered_set&gt;
    #include &lt;utility&gt;
    #include &lt;valarray&gt;
    #include &lt;vector&gt;
  ],,
  ac_cv_cxx_stdcxx_0x=yes, ac_cv_cxx_stdcxx_0x=no)
  AC_LANG_RESTORE
  CXXFLAGS="$ac_save_CXXFLAGS"
  ])
  if test "$ac_cv_cxx_stdcxx_0x" = yes; then
    AC_DEFINE(STDCXX_0X_HEADERS,,[Define if ISO C++ 0x header files are present. ])
  fi
])
</programlisting>

<para>As is the case for TR1 support, these autoconf macros can be made for a finer-grained, per-header-file check. For &lt;unordered_map&gt;
</para>

<programlisting>
# AC_HEADER_UNORDERED_MAP
AC_DEFUN([AC_HEADER_UNORDERED_MAP], [
  AC_CACHE_CHECK(for unordered_map,
  ac_cv_cxx_unordered_map,
  [AC_REQUIRE([AC_COMPILE_STDCXX_0X])
  AC_LANG_SAVE
  AC_LANG_CPLUSPLUS
  ac_save_CXXFLAGS="$CXXFLAGS"
  CXXFLAGS="$CXXFLAGS -std=gnu++0x"	
  AC_TRY_COMPILE([#include &lt;unordered_map&gt;], [using std::unordered_map;],
  ac_cv_cxx_unordered_map=yes, ac_cv_cxx_unordered_map=no)
  CXXFLAGS="$ac_save_CXXFLAGS"
  AC_LANG_RESTORE
  ])
  if test "$ac_cv_cxx_unordered_map" = yes; then
    AC_DEFINE(HAVE_UNORDERED_MAP,,[Define if unordered_map is present. ])
  fi
])
</programlisting>

<programlisting>
# AC_HEADER_UNORDERED_SET
AC_DEFUN([AC_HEADER_UNORDERED_SET], [
  AC_CACHE_CHECK(for unordered_set,
  ac_cv_cxx_unordered_set,
  [AC_REQUIRE([AC_COMPILE_STDCXX_0X])
  AC_LANG_SAVE
  AC_LANG_CPLUSPLUS
  ac_save_CXXFLAGS="$CXXFLAGS"
  CXXFLAGS="$CXXFLAGS -std=gnu++0x"	
  AC_TRY_COMPILE([#include &lt;unordered_set&gt;], [using std::unordered_set;],
  ac_cv_cxx_unordered_set=yes, ac_cv_cxx_unordered_set=no)
  CXXFLAGS="$ac_save_CXXFLAGS"
  AC_LANG_RESTORE
  ])
  if test "$ac_cv_cxx_unordered_set" = yes; then
    AC_DEFINE(HAVE_UNORDERED_SET,,[Define if unordered_set is present. ])
  fi
])
</programlisting>
</sect3>

<sect3>
<title>
  Container::iterator_type is not necessarily Container::value_type*
</title>

<para>
  This is a change in behavior from the previous version. Now, most
  <type>iterator_type</type> typedefs in container classes are POD
  objects, not <type>value_type</type> pointers.
</para>
</sect3>

</sect2>

<bibliography id="backwards.biblio" xreflabel="backwards.biblio">
<title>Bibliography</title>

  <biblioentry>
    <abbrev>
      kegel41
    </abbrev>

    <title>
      Migrating to GCC 4.1
    </title>

    <author>
      <firstname>Dan</firstname>
      <surname>Kegel</surname>
    </author>

    <biblioid>
      <ulink url="http://www.kegel.com/gcc/gcc4.html">
      </ulink>
    </biblioid>
  </biblioentry> 


  <biblioentry>
    <abbrev>
      kegel41
    </abbrev>

    <title>
      Building the Whole Debian Archive with GCC 4.1: A Summary
    </title>

    <author>
      <firstname>Martin</firstname>
      <surname>Michlmayr</surname>
    </author>

    <biblioid>
      <ulink url="http://lists.debian.org/debian-gcc/2006/03/msg00405.html">
      </ulink>
    </biblioid>
  </biblioentry> 


  <biblioentry>
    <abbrev>
      lbl32
    </abbrev>

    <title>
      Migration guide for GCC-3.2
    </title>

    <biblioid>
      <ulink url="http://annwm.lbl.gov/~leggett/Atlas/gcc-3.2.html">
      </ulink>
    </biblioid>
  </biblioentry> 

</bibliography>

</sect1>