xref: /netbsd-src/external/gpl3/gcc.old/dist/libstdc++-v3/doc/html/manual/using_concurrency.html (revision 8feb0f0b7eaff0608f8350bbfa3098827b4bb91b)
136ac495dSmrg<?xml version="1.0" encoding="UTF-8" standalone="no"?>
236ac495dSmrg<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"><html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>Concurrency</title><meta name="generator" content="DocBook XSL Stylesheets Vsnapshot" /><meta name="keywords" content="ISO C++, library" /><meta name="keywords" content="ISO C++, runtime, library" /><link rel="home" href="../index.html" title="The GNU C++ Library" /><link rel="up" href="using.html" title="Chapter 3. Using" /><link rel="prev" href="using_dynamic_or_shared.html" title="Linking" /><link rel="next" href="using_exceptions.html" title="Exceptions" /></head><body><div class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="3" align="center">Concurrency</th></tr><tr><td width="20%" align="left"><a accesskey="p" href="using_dynamic_or_shared.html">Prev</a> </td><th width="60%" align="center">Chapter 3. Using</th><td width="20%" align="right"> <a accesskey="n" href="using_exceptions.html">Next</a></td></tr></table><hr /></div><div class="section"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="manual.intro.using.concurrency"></a>Concurrency</h2></div></div></div><p>This section discusses issues surrounding the proper compilation
336ac495dSmrg      of multithreaded applications which use the Standard C++
436ac495dSmrg      library.  This information is GCC-specific since the C++
536ac495dSmrg      standard does not address matters of multithreaded applications.
636ac495dSmrg   </p><div class="section"><div class="titlepage"><div><div><h3 class="title"><a id="manual.intro.using.concurrency.prereq"></a>Prerequisites</h3></div></div></div><p>All normal disclaimers aside, multithreaded C++ application are
736ac495dSmrg      only supported when libstdc++ and all user code was built with
836ac495dSmrg      compilers which report (via <code class="code"> gcc/g++ -v </code>) the same thread
936ac495dSmrg      model and that model is not <span class="emphasis"><em>single</em></span>.  As long as your
1036ac495dSmrg      final application is actually single-threaded, then it should be
1136ac495dSmrg      safe to mix user code built with a thread model of
1236ac495dSmrg      <span class="emphasis"><em>single</em></span> with a libstdc++ and other C++ libraries built
1336ac495dSmrg      with another thread model useful on the platform.  Other mixes
1436ac495dSmrg      may or may not work but are not considered supported.  (Thus, if
1536ac495dSmrg      you distribute a shared C++ library in binary form only, it may
1636ac495dSmrg      be best to compile it with a GCC configured with
1736ac495dSmrg      --enable-threads for maximal interchangeability and usefulness
1836ac495dSmrg      with a user population that may have built GCC with either
1936ac495dSmrg      --enable-threads or --disable-threads.)
2036ac495dSmrg   </p><p>When you link a multithreaded application, you will probably
2136ac495dSmrg      need to add a library or flag to g++.  This is a very
2236ac495dSmrg      non-standardized area of GCC across ports.  Some ports support a
2336ac495dSmrg      special flag (the spelling isn't even standardized yet) to add
2436ac495dSmrg      all required macros to a compilation (if any such flags are
2536ac495dSmrg      required then you must provide the flag for all compilations not
2636ac495dSmrg      just linking) and link-library additions and/or replacements at
2736ac495dSmrg      link time.  The documentation is weak.  On several targets (including
2836ac495dSmrg      GNU/Linux, Solaris and various BSDs) -pthread is honored.
2936ac495dSmrg      Some other ports use other switches.
3036ac495dSmrg      This is not well documented anywhere other than
3136ac495dSmrg      in "gcc -dumpspecs" (look at the 'lib' and 'cpp' entries).
3236ac495dSmrg   </p><p>
3336ac495dSmrg     Some uses of <code class="classname">std::atomic</code> also require linking
3436ac495dSmrg     to <code class="filename">libatomic</code>.
3536ac495dSmrg   </p></div><div class="section"><div class="titlepage"><div><div><h3 class="title"><a id="manual.intro.using.concurrency.thread_safety"></a>Thread Safety</h3></div></div></div><p>
3636ac495dSmrgIn the terms of the 2011 C++ standard a thread-safe program is one which
3736ac495dSmrgdoes not perform any conflicting non-atomic operations on memory locations
3836ac495dSmrgand so does not contain any data races.
3936ac495dSmrgThe standard places requirements on the library to ensure that no data
4036ac495dSmrgraces are caused by the library itself or by programs which use the
4136ac495dSmrglibrary correctly (as described below).
4236ac495dSmrgThe C++11 memory model and library requirements are a more formal version
43a2dc1f3fSmrgof the <a class="link" href="https://web.archive.org/web/20171225062613/http://www.sgi.com/tech/stl/thread_safety.html" target="_top">SGI STL</a> definition of thread safety, which the library used
4436ac495dSmrgprior to the 2011 standard.
4536ac495dSmrg</p><p>The library strives to be thread-safe when all of the following
4636ac495dSmrg	 conditions are met:
4736ac495dSmrg      </p><div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; "><li class="listitem"><p>The system's libc is itself thread-safe,
4836ac495dSmrg       </p></li><li class="listitem"><p>
4936ac495dSmrg	   The compiler in use reports a thread model other than
5036ac495dSmrg	   'single'. This can be tested via output from <code class="code">gcc
5136ac495dSmrg	   -v</code>. Multi-thread capable versions of gcc output
5236ac495dSmrg	   something like this:
5336ac495dSmrg	 </p><pre class="programlisting">
5436ac495dSmrg%gcc -v
5536ac495dSmrgUsing built-in specs.
5636ac495dSmrg...
5736ac495dSmrgThread model: posix
5836ac495dSmrggcc version 4.1.2 20070925 (Red Hat 4.1.2-33)
5936ac495dSmrg</pre><p>Look for "Thread model" lines that aren't equal to "single."</p></li><li class="listitem"><p>
6036ac495dSmrg	 Requisite command-line flags are used for atomic operations
6136ac495dSmrg	 and threading. Examples of this include <code class="code">-pthread</code>
6236ac495dSmrg	 and <code class="code">-march=native</code>, although specifics vary
6336ac495dSmrg	 depending on the host environment. See
6436ac495dSmrg	 <a class="link" href="using.html#manual.intro.using.flags" title="Command Options">Command Options</a> and
6536ac495dSmrg	 <a class="link" href="http://gcc.gnu.org/onlinedocs/gcc/Option-Summary.html" target="_top">Machine
6636ac495dSmrg	 Dependent Options</a>.
6736ac495dSmrg       </p></li><li class="listitem"><p>
6836ac495dSmrg	   An implementation of the
6936ac495dSmrg	   <code class="filename">atomicity.h</code> functions
7036ac495dSmrg	   exists for the architecture in question. See the
7136ac495dSmrg	   <a class="link" href="internals.html#internals.thread_safety" title="Thread Safety">internals
7236ac495dSmrg	   documentation</a> for more details.
7336ac495dSmrg       </p></li></ul></div><p>The user code must guard against concurrent function calls which
7436ac495dSmrg         access any particular library object's state when one or more of
7536ac495dSmrg         those accesses modifies the state. An object will be modified by
7636ac495dSmrg         invoking a non-const member function on it or passing it as a
7736ac495dSmrg         non-const argument to a library function. An object will not be
7836ac495dSmrg         modified by invoking a const member function on it or passing it to
7936ac495dSmrg         a function as a pointer- or reference-to-const.
8036ac495dSmrg         Typically, the application
8136ac495dSmrg         programmer may infer what object locks must be held based on the
8236ac495dSmrg         objects referenced in a function call and whether the objects are
8336ac495dSmrg         accessed as const or non-const.  Without getting
8436ac495dSmrg	 into great detail, here is an example which requires user-level
8536ac495dSmrg	 locks:
8636ac495dSmrg      </p><pre class="programlisting">
8736ac495dSmrg     library_class_a shared_object_a;
8836ac495dSmrg
8936ac495dSmrg     void thread_main () {
9036ac495dSmrg       library_class_b *object_b = new library_class_b;
9136ac495dSmrg       shared_object_a.add_b (object_b);   // must hold lock for shared_object_a
9236ac495dSmrg       shared_object_a.mutate ();          // must hold lock for shared_object_a
9336ac495dSmrg     }
9436ac495dSmrg
9536ac495dSmrg     // Multiple copies of thread_main() are started in independent threads.</pre><p>Under the assumption that object_a and object_b are never exposed to
9636ac495dSmrg	 another thread, here is an example that does not require any
9736ac495dSmrg	 user-level locks:
9836ac495dSmrg      </p><pre class="programlisting">
9936ac495dSmrg     void thread_main () {
10036ac495dSmrg       library_class_a object_a;
10136ac495dSmrg       library_class_b *object_b = new library_class_b;
10236ac495dSmrg       object_a.add_b (object_b);
10336ac495dSmrg       object_a.mutate ();
10436ac495dSmrg     } </pre><p>All library types are safe to use in a multithreaded program
10536ac495dSmrg         if objects are not shared between threads or as
10636ac495dSmrg	 long each thread carefully locks out access by any other
10736ac495dSmrg	 thread while it modifies any object visible to another thread.
10836ac495dSmrg	 Unless otherwise documented, the only exceptions to these rules
10936ac495dSmrg         are atomic operations on the types in
11036ac495dSmrg         <code class="filename">&lt;atomic&gt;</code>
11136ac495dSmrg         and lock/unlock operations on the standard mutex types in
11236ac495dSmrg         <code class="filename">&lt;mutex&gt;</code>. These
11336ac495dSmrg         atomic operations allow concurrent accesses to the same object
11436ac495dSmrg         without introducing data races.
11536ac495dSmrg      </p><p>The following member functions of standard containers can be
11636ac495dSmrg         considered to be const for the purposes of avoiding data races:
11736ac495dSmrg         <code class="code">begin</code>, <code class="code">end</code>, <code class="code">rbegin</code>, <code class="code">rend</code>,
11836ac495dSmrg         <code class="code">front</code>, <code class="code">back</code>, <code class="code">data</code>,
11936ac495dSmrg         <code class="code">find</code>, <code class="code">lower_bound</code>, <code class="code">upper_bound</code>,
12036ac495dSmrg         <code class="code">equal_range</code>, <code class="code">at</code>
12136ac495dSmrg         and, except in associative or unordered associative containers,
12236ac495dSmrg         <code class="code">operator[]</code>. In other words, although they are non-const
12336ac495dSmrg         so that they can return mutable iterators, those member functions
12436ac495dSmrg         will not modify the container.
12536ac495dSmrg         Accessing an iterator might cause a non-modifying access to
12636ac495dSmrg         the container the iterator refers to (for example incrementing a
12736ac495dSmrg         list iterator must access the pointers between nodes, which are part
12836ac495dSmrg         of the container and so conflict with other accesses to the container).
12936ac495dSmrg      </p><p>Programs which follow the rules above will not encounter data
13036ac495dSmrg         races in library code, even when using library types which share
13136ac495dSmrg         state between distinct objects.  In the example below the
13236ac495dSmrg         <code class="code">shared_ptr</code> objects share a reference count, but
13336ac495dSmrg         because the code does not perform any non-const operations on the
13436ac495dSmrg         globally-visible object, the library ensures that the reference
13536ac495dSmrg         count updates are atomic and do not introduce data races:
13636ac495dSmrg      </p><pre class="programlisting">
13736ac495dSmrg    std::shared_ptr&lt;int&gt; global_sp;
13836ac495dSmrg
13936ac495dSmrg    void thread_main() {
14036ac495dSmrg      auto local_sp = global_sp;  // OK, copy constructor's parameter is reference-to-const
14136ac495dSmrg
14236ac495dSmrg      int i = *global_sp;         // OK, operator* is const
14336ac495dSmrg      int j = *local_sp;          // OK, does not operate on global_sp
14436ac495dSmrg
14536ac495dSmrg      // *global_sp = 2;          // NOT OK, modifies int visible to other threads
14636ac495dSmrg      // *local_sp = 2;           // NOT OK, modifies int visible to other threads
14736ac495dSmrg
14836ac495dSmrg      // global_sp.reset();       // NOT OK, reset is non-const
14936ac495dSmrg      local_sp.reset();           // OK, does not operate on global_sp
15036ac495dSmrg    }
15136ac495dSmrg
15236ac495dSmrg    int main() {
15336ac495dSmrg      global_sp.reset(new int(1));
15436ac495dSmrg      std::thread t1(thread_main);
15536ac495dSmrg      std::thread t2(thread_main);
15636ac495dSmrg      t1.join();
15736ac495dSmrg      t2.join();
15836ac495dSmrg    }
15936ac495dSmrg      </pre><p>For further details of the C++11 memory model see Hans-J. Boehm's
16036ac495dSmrg      <a class="link" href="https://www.hboehm.info/c++mm/" target="_top">Threads
161*8feb0f0bSmrg      and memory model for C++</a> pages, particularly the <a class="link" href="https://www.hboehm.info/c++mm/threadsintro.html" target="_top">introduction</a>
16236ac495dSmrg      and <a class="link" href="https://www.hboehm.info/c++mm/user-faq.html" target="_top">FAQ</a>.
16336ac495dSmrg      </p></div><div class="section"><div class="titlepage"><div><div><h3 class="title"><a id="manual.intro.using.concurrency.atomics"></a>Atomics</h3></div></div></div><p>
16436ac495dSmrg    </p></div><div class="section"><div class="titlepage"><div><div><h3 class="title"><a id="manual.intro.using.concurrency.io"></a>IO</h3></div></div></div><p>This gets a bit tricky.  Please read carefully, and bear with me.
16536ac495dSmrg   </p><div class="section"><div class="titlepage"><div><div><h4 class="title"><a id="concurrency.io.structure"></a>Structure</h4></div></div></div><p>A wrapper
16636ac495dSmrg      type called <code class="code">__basic_file</code> provides our abstraction layer
16736ac495dSmrg      for the <code class="code">std::filebuf</code> classes.  Nearly all decisions dealing
16836ac495dSmrg      with actual input and output must be made in <code class="code">__basic_file</code>.
16936ac495dSmrg   </p><p>A generic locking mechanism is somewhat in place at the filebuf layer,
17036ac495dSmrg      but is not used in the current code.  Providing locking at any higher
17136ac495dSmrg      level is akin to providing locking within containers, and is not done
17236ac495dSmrg      for the same reasons (see the links above).
17336ac495dSmrg   </p></div><div class="section"><div class="titlepage"><div><div><h4 class="title"><a id="concurrency.io.defaults"></a>Defaults</h4></div></div></div><p>The __basic_file type is simply a collection of small wrappers around
17436ac495dSmrg      the C stdio layer (again, see the link under Structure).  We do no
17536ac495dSmrg      locking ourselves, but simply pass through to calls to <code class="code">fopen</code>,
17636ac495dSmrg      <code class="code">fwrite</code>, and so forth.
17736ac495dSmrg   </p><p>So, for 3.0, the question of "is multithreading safe for I/O"
17836ac495dSmrg      must be answered with, "is your platform's C library threadsafe
17936ac495dSmrg      for I/O?"  Some are by default, some are not; many offer multiple
18036ac495dSmrg      implementations of the C library with varying tradeoffs of threadsafety
18136ac495dSmrg      and efficiency.  You, the programmer, are always required to take care
18236ac495dSmrg      with multiple threads.
18336ac495dSmrg   </p><p>(As an example, the POSIX standard requires that C stdio
18436ac495dSmrg       <code class="code">FILE*</code> operations are atomic.  POSIX-conforming C libraries
18536ac495dSmrg       (e.g, on Solaris and GNU/Linux) have an internal mutex to serialize
18636ac495dSmrg       operations on <code class="code">FILE*</code>s.
18736ac495dSmrg       However, you still need to not do stupid things like calling
18836ac495dSmrg       <code class="code">fclose(fs)</code> in one thread followed by an access of
18936ac495dSmrg       <code class="code">fs</code> in another.)
19036ac495dSmrg   </p><p>So, if your platform's C library is threadsafe, then your
19136ac495dSmrg      <code class="code">fstream</code> I/O operations will be threadsafe at the lowest
19236ac495dSmrg      level.  For higher-level operations, such as manipulating the data
19336ac495dSmrg      contained in the stream formatting classes (e.g., setting up callbacks
19436ac495dSmrg      inside an <code class="code">std::ofstream</code>), you need to guard such accesses
19536ac495dSmrg      like any other critical shared resource.
19636ac495dSmrg   </p></div><div class="section"><div class="titlepage"><div><div><h4 class="title"><a id="concurrency.io.future"></a>Future</h4></div></div></div><p> A
19736ac495dSmrg      second choice may be available for I/O implementations:  libio.  This is
19836ac495dSmrg      disabled by default, and in fact will not currently work due to other
19936ac495dSmrg      issues.  It will be revisited, however.
20036ac495dSmrg   </p><p>The libio code is a subset of the guts of the GNU libc (glibc) I/O
20136ac495dSmrg      implementation.  When libio is in use, the <code class="code">__basic_file</code>
20236ac495dSmrg      type is basically derived from FILE.  (The real situation is more
20336ac495dSmrg      complex than that... it's derived from an internal type used to
20436ac495dSmrg      implement FILE.  See libio/libioP.h to see scary things done with
20536ac495dSmrg      vtbls.)  The result is that there is no "layer" of C stdio
20636ac495dSmrg      to go through; the filebuf makes calls directly into the same
20736ac495dSmrg      functions used to implement <code class="code">fread</code>, <code class="code">fwrite</code>,
20836ac495dSmrg      and so forth, using internal data structures.  (And when I say
20936ac495dSmrg      "makes calls directly," I mean the function is literally
21036ac495dSmrg      replaced by a jump into an internal function.  Fast but frightening.
21136ac495dSmrg      *grin*)
21236ac495dSmrg   </p><p>Also, the libio internal locks are used.  This requires pulling in
21336ac495dSmrg      large chunks of glibc, such as a pthreads implementation, and is one
21436ac495dSmrg      of the issues preventing widespread use of libio as the libstdc++
21536ac495dSmrg      cstdio implementation.
21636ac495dSmrg   </p><p>But we plan to make this work, at least as an option if not a future
21736ac495dSmrg      default.  Platforms running a copy of glibc with a recent-enough
21836ac495dSmrg      version will see calls from libstdc++ directly into the glibc already
21936ac495dSmrg      installed.  For other platforms, a copy of the libio subsection will
22036ac495dSmrg      be built and included in libstdc++.
22136ac495dSmrg   </p></div><div class="section"><div class="titlepage"><div><div><h4 class="title"><a id="concurrency.io.alt"></a>Alternatives</h4></div></div></div><p>Don't forget that other cstdio implementations are possible.  You could
22236ac495dSmrg      easily write one to perform your own forms of locking, to solve your
22336ac495dSmrg      "interesting" problems.
22436ac495dSmrg   </p></div></div><div class="section"><div class="titlepage"><div><div><h3 class="title"><a id="manual.intro.using.concurrency.containers"></a>Containers</h3></div></div></div><p>This section discusses issues surrounding the design of
22536ac495dSmrg      multithreaded applications which use Standard C++ containers.
22636ac495dSmrg      All information in this section is current as of the gcc 3.0
22736ac495dSmrg      release and all later point releases.  Although earlier gcc
22836ac495dSmrg      releases had a different approach to threading configuration and
22936ac495dSmrg      proper compilation, the basic code design rules presented here
23036ac495dSmrg      were similar.  For information on all other aspects of
23136ac495dSmrg      multithreading as it relates to libstdc++, including details on
23236ac495dSmrg      the proper compilation of threaded code (and compatibility between
23336ac495dSmrg      threaded and non-threaded code), see Chapter 17.
23436ac495dSmrg   </p><p>Two excellent pages to read when working with the Standard C++
23536ac495dSmrg      containers and threads are
236a2dc1f3fSmrg      <a class="link" href="https://web.archive.org/web/20171225062613/http://www.sgi.com/tech/stl/thread_safety.html" target="_top">SGI's
237a2dc1f3fSmrg      https://web.archive.org/web/20171225062613/http://www.sgi.com/tech/stl/thread_safety.html</a> and
238a2dc1f3fSmrg      <a class="link" href="https://web.archive.org/web/20171225062613/http://www.sgi.com/tech/stl/Allocators.html" target="_top">SGI's
239a2dc1f3fSmrg      https://web.archive.org/web/20171225062613/http://www.sgi.com/tech/stl/Allocators.html</a>.
24036ac495dSmrg   </p><p><span class="emphasis"><em>However, please ignore all discussions about the user-level
24136ac495dSmrg      configuration of the lock implementation inside the STL
24236ac495dSmrg      container-memory allocator on those pages.  For the sake of this
24336ac495dSmrg      discussion, libstdc++ configures the SGI STL implementation,
24436ac495dSmrg      not you.  This is quite different from how gcc pre-3.0 worked.
24536ac495dSmrg      In particular, past advice was for people using g++ to
24636ac495dSmrg      explicitly define _PTHREADS or other macros or port-specific
24736ac495dSmrg      compilation options on the command line to get a thread-safe
24836ac495dSmrg      STL.  This is no longer required for any port and should no
24936ac495dSmrg      longer be done unless you really know what you are doing and
25036ac495dSmrg      assume all responsibility.</em></span>
25136ac495dSmrg   </p><p>Since the container implementation of libstdc++ uses the SGI
25236ac495dSmrg      code, we use the same definition of thread safety as SGI when
25336ac495dSmrg      discussing design.  A key point that beginners may miss is the
25436ac495dSmrg      fourth major paragraph of the first page mentioned above
25536ac495dSmrg      (<span class="emphasis"><em>For most clients...</em></span>), which points out that
25636ac495dSmrg      locking must nearly always be done outside the container, by
25736ac495dSmrg      client code (that'd be you, not us).  There is a notable
25836ac495dSmrg      exceptions to this rule.  Allocators called while a container or
25936ac495dSmrg      element is constructed uses an internal lock obtained and
26036ac495dSmrg      released solely within libstdc++ code (in fact, this is the
26136ac495dSmrg      reason STL requires any knowledge of the thread configuration).
26236ac495dSmrg   </p><p>For implementing a container which does its own locking, it is
26336ac495dSmrg      trivial to provide a wrapper class which obtains the lock (as
26436ac495dSmrg      SGI suggests), performs the container operation, and then
26536ac495dSmrg      releases the lock.  This could be templatized <span class="emphasis"><em>to a certain
26636ac495dSmrg      extent</em></span>, on the underlying container and/or a locking
26736ac495dSmrg      mechanism.  Trying to provide a catch-all general template
26836ac495dSmrg      solution would probably be more trouble than it's worth.
26936ac495dSmrg   </p><p>The library implementation may be configured to use the
27036ac495dSmrg      high-speed caching memory allocator, which complicates thread
27136ac495dSmrg      safety issues. For all details about how to globally override
27236ac495dSmrg      this at application run-time
27336ac495dSmrg      see <a class="link" href="using_macros.html" title="Macros">here</a>. Also
27436ac495dSmrg      useful are details
27536ac495dSmrg      on <a class="link" href="memory.html#std.util.memory.allocator" title="Allocators">allocator</a>
27636ac495dSmrg      options and capabilities.
27736ac495dSmrg   </p></div></div><div class="navfooter"><hr /><table width="100%" summary="Navigation footer"><tr><td width="40%" align="left"><a accesskey="p" href="using_dynamic_or_shared.html">Prev</a> </td><td width="20%" align="center"><a accesskey="u" href="using.html">Up</a></td><td width="40%" align="right"> <a accesskey="n" href="using_exceptions.html">Next</a></td></tr><tr><td width="40%" align="left" valign="top">Linking </td><td width="20%" align="center"><a accesskey="h" href="../index.html">Home</a></td><td width="40%" align="right" valign="top"> Exceptions</td></tr></table></div></body></html>