1<?xml version="1.0" encoding="UTF-8" standalone="no"?> 2<!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-NS Stylesheets V1.78.1" /><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 3 of multithreaded applications which use the Standard C++ 4 library. This information is GCC-specific since the C++ 5 standard does not address matters of multithreaded applications. 6 </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 7 only supported when libstdc++ and all user code was built with 8 compilers which report (via <code class="code"> gcc/g++ -v </code>) the same thread 9 model and that model is not <span class="emphasis"><em>single</em></span>. As long as your 10 final application is actually single-threaded, then it should be 11 safe to mix user code built with a thread model of 12 <span class="emphasis"><em>single</em></span> with a libstdc++ and other C++ libraries built 13 with another thread model useful on the platform. Other mixes 14 may or may not work but are not considered supported. (Thus, if 15 you distribute a shared C++ library in binary form only, it may 16 be best to compile it with a GCC configured with 17 --enable-threads for maximal interchangeability and usefulness 18 with a user population that may have built GCC with either 19 --enable-threads or --disable-threads.) 20 </p><p>When you link a multithreaded application, you will probably 21 need to add a library or flag to g++. This is a very 22 non-standardized area of GCC across ports. Some ports support a 23 special flag (the spelling isn't even standardized yet) to add 24 all required macros to a compilation (if any such flags are 25 required then you must provide the flag for all compilations not 26 just linking) and link-library additions and/or replacements at 27 link time. The documentation is weak. Here is a quick summary 28 to display how ad hoc this is: On Solaris, both -pthreads and 29 -threads (with subtly different meanings) are honored. 30 On GNU/Linux x86, -pthread is honored. On FreeBSD, 31 -pthread is honored. Some other ports use other switches. 32 AFAIK, none of this is properly documented anywhere other than 33 in ``gcc -dumpspecs'' (look at lib and cpp entries). 34 </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> 35In the terms of the 2011 C++ standard a thread-safe program is one which 36does not perform any conflicting non-atomic operations on memory locations 37and so does not contain any data races. 38The standard places requirements on the library to ensure that no data 39races are caused by the library itself or by programs which use the 40library correctly (as described below). 41The C++11 memory model and library requirements are a more formal version 42of the <a class="link" href="http://www.sgi.com/tech/stl/thread_safety.html" target="_top">SGI STL</a> definition of thread safety, which the library used 43prior to the 2011 standard. 44</p><p>The library strives to be thread-safe when all of the following 45 conditions are met: 46 </p><div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; "><li class="listitem"><p>The system's libc is itself thread-safe, 47 </p></li><li class="listitem"><p> 48 The compiler in use reports a thread model other than 49 'single'. This can be tested via output from <code class="code">gcc 50 -v</code>. Multi-thread capable versions of gcc output 51 something like this: 52 </p><pre class="programlisting"> 53%gcc -v 54Using built-in specs. 55... 56Thread model: posix 57gcc version 4.1.2 20070925 (Red Hat 4.1.2-33) 58</pre><p>Look for "Thread model" lines that aren't equal to "single."</p></li><li class="listitem"><p> 59 Requisite command-line flags are used for atomic operations 60 and threading. Examples of this include <code class="code">-pthread</code> 61 and <code class="code">-march=native</code>, although specifics vary 62 depending on the host environment. See <a class="link" href="http://gcc.gnu.org/onlinedocs/gcc/Option-Summary.html" target="_top">Machine 63 Dependent Options</a>. 64 </p></li><li class="listitem"><p> 65 An implementation of atomicity.h functions 66 exists for the architecture in question. See the internals documentation for more <a class="link" href="internals.html#internals.thread_safety" title="Thread Safety">details</a>. 67 </p></li></ul></div><p>The user code must guard against concurrent function calls which 68 access any particular library object's state when one or more of 69 those accesses modifies the state. An object will be modified by 70 invoking a non-const member function on it or passing it as a 71 non-const argument to a library function. An object will not be 72 modified by invoking a const member function on it or passing it to 73 a function as a pointer- or reference-to-const. 74 Typically, the application 75 programmer may infer what object locks must be held based on the 76 objects referenced in a function call and whether the objects are 77 accessed as const or non-const. Without getting 78 into great detail, here is an example which requires user-level 79 locks: 80 </p><pre class="programlisting"> 81 library_class_a shared_object_a; 82 83 void thread_main () { 84 library_class_b *object_b = new library_class_b; 85 shared_object_a.add_b (object_b); // must hold lock for shared_object_a 86 shared_object_a.mutate (); // must hold lock for shared_object_a 87 } 88 89 // 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 90 another thread, here is an example that does not require any 91 user-level locks: 92 </p><pre class="programlisting"> 93 void thread_main () { 94 library_class_a object_a; 95 library_class_b *object_b = new library_class_b; 96 object_a.add_b (object_b); 97 object_a.mutate (); 98 } </pre><p>All library types are safe to use in a multithreaded program 99 if objects are not shared between threads or as 100 long each thread carefully locks out access by any other 101 thread while it modifies any object visible to another thread. 102 Unless otherwise documented, the only exceptions to these rules 103 are atomic operations on the types in 104 <code class="filename"><atomic></code> 105 and lock/unlock operations on the standard mutex types in 106 <code class="filename"><mutex></code>. These 107 atomic operations allow concurrent accesses to the same object 108 without introducing data races. 109 </p><p>The following member functions of standard containers can be 110 considered to be const for the purposes of avoiding data races: 111 <code class="code">begin</code>, <code class="code">end</code>, <code class="code">rbegin</code>, <code class="code">rend</code>, 112 <code class="code">front</code>, <code class="code">back</code>, <code class="code">data</code>, 113 <code class="code">find</code>, <code class="code">lower_bound</code>, <code class="code">upper_bound</code>, 114 <code class="code">equal_range</code>, <code class="code">at</code> 115 and, except in associative or unordered associative containers, 116 <code class="code">operator[]</code>. In other words, although they are non-const 117 so that they can return mutable iterators, those member functions 118 will not modify the container. 119 Accessing an iterator might cause a non-modifying access to 120 the container the iterator refers to (for example incrementing a 121 list iterator must access the pointers between nodes, which are part 122 of the container and so conflict with other accesses to the container). 123 </p><p>Programs which follow the rules above will not encounter data 124 races in library code, even when using library types which share 125 state between distinct objects. In the example below the 126 <code class="code">shared_ptr</code> objects share a reference count, but 127 because the code does not perform any non-const operations on the 128 globally-visible object, the library ensures that the reference 129 count updates are atomic and do not introduce data races: 130 </p><pre class="programlisting"> 131 std::shared_ptr<int> global_sp; 132 133 void thread_main() { 134 auto local_sp = global_sp; // OK, copy constructor's parameter is reference-to-const 135 136 int i = *global_sp; // OK, operator* is const 137 int j = *local_sp; // OK, does not operate on global_sp 138 139 // *global_sp = 2; // NOT OK, modifies int visible to other threads 140 // *local_sp = 2; // NOT OK, modifies int visible to other threads 141 142 // global_sp.reset(); // NOT OK, reset is non-const 143 local_sp.reset(); // OK, does not operate on global_sp 144 } 145 146 int main() { 147 global_sp.reset(new int(1)); 148 std::thread t1(thread_main); 149 std::thread t2(thread_main); 150 t1.join(); 151 t2.join(); 152 } 153 </pre><p>For further details of the C++11 memory model see Hans-J. Boehm's 154 <a class="link" href="http://www.hpl.hp.com/personal/Hans_Boehm/c++mm/user-faq.html" target="_top">Threads 155 and memory model for C++</a> pages, particularly the <a class="link" href="http://www.hpl.hp.com/personal/Hans_Boehm/c++mm/threadsintro.html" target="_top">introduction</a> 156 and <a class="link" href="http://www.hpl.hp.com/personal/Hans_Boehm/c++mm/user-faq.html" target="_top">FAQ</a>. 157 </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> 158 </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. 159 </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 160 type called <code class="code">__basic_file</code> provides our abstraction layer 161 for the <code class="code">std::filebuf</code> classes. Nearly all decisions dealing 162 with actual input and output must be made in <code class="code">__basic_file</code>. 163 </p><p>A generic locking mechanism is somewhat in place at the filebuf layer, 164 but is not used in the current code. Providing locking at any higher 165 level is akin to providing locking within containers, and is not done 166 for the same reasons (see the links above). 167 </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 168 the C stdio layer (again, see the link under Structure). We do no 169 locking ourselves, but simply pass through to calls to <code class="code">fopen</code>, 170 <code class="code">fwrite</code>, and so forth. 171 </p><p>So, for 3.0, the question of "is multithreading safe for I/O" 172 must be answered with, "is your platform's C library threadsafe 173 for I/O?" Some are by default, some are not; many offer multiple 174 implementations of the C library with varying tradeoffs of threadsafety 175 and efficiency. You, the programmer, are always required to take care 176 with multiple threads. 177 </p><p>(As an example, the POSIX standard requires that C stdio FILE* 178 operations are atomic. POSIX-conforming C libraries (e.g, on Solaris 179 and GNU/Linux) have an internal mutex to serialize operations on 180 FILE*s. However, you still need to not do stupid things like calling 181 <code class="code">fclose(fs)</code> in one thread followed by an access of 182 <code class="code">fs</code> in another.) 183 </p><p>So, if your platform's C library is threadsafe, then your 184 <code class="code">fstream</code> I/O operations will be threadsafe at the lowest 185 level. For higher-level operations, such as manipulating the data 186 contained in the stream formatting classes (e.g., setting up callbacks 187 inside an <code class="code">std::ofstream</code>), you need to guard such accesses 188 like any other critical shared resource. 189 </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 190 second choice may be available for I/O implementations: libio. This is 191 disabled by default, and in fact will not currently work due to other 192 issues. It will be revisited, however. 193 </p><p>The libio code is a subset of the guts of the GNU libc (glibc) I/O 194 implementation. When libio is in use, the <code class="code">__basic_file</code> 195 type is basically derived from FILE. (The real situation is more 196 complex than that... it's derived from an internal type used to 197 implement FILE. See libio/libioP.h to see scary things done with 198 vtbls.) The result is that there is no "layer" of C stdio 199 to go through; the filebuf makes calls directly into the same 200 functions used to implement <code class="code">fread</code>, <code class="code">fwrite</code>, 201 and so forth, using internal data structures. (And when I say 202 "makes calls directly," I mean the function is literally 203 replaced by a jump into an internal function. Fast but frightening. 204 *grin*) 205 </p><p>Also, the libio internal locks are used. This requires pulling in 206 large chunks of glibc, such as a pthreads implementation, and is one 207 of the issues preventing widespread use of libio as the libstdc++ 208 cstdio implementation. 209 </p><p>But we plan to make this work, at least as an option if not a future 210 default. Platforms running a copy of glibc with a recent-enough 211 version will see calls from libstdc++ directly into the glibc already 212 installed. For other platforms, a copy of the libio subsection will 213 be built and included in libstdc++. 214 </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 215 easily write one to perform your own forms of locking, to solve your 216 "interesting" problems. 217 </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 218 multithreaded applications which use Standard C++ containers. 219 All information in this section is current as of the gcc 3.0 220 release and all later point releases. Although earlier gcc 221 releases had a different approach to threading configuration and 222 proper compilation, the basic code design rules presented here 223 were similar. For information on all other aspects of 224 multithreading as it relates to libstdc++, including details on 225 the proper compilation of threaded code (and compatibility between 226 threaded and non-threaded code), see Chapter 17. 227 </p><p>Two excellent pages to read when working with the Standard C++ 228 containers and threads are 229 <a class="link" href="http://www.sgi.com/tech/stl/thread_safety.html" target="_top">SGI's 230 http://www.sgi.com/tech/stl/thread_safety.html</a> and 231 <a class="link" href="http://www.sgi.com/tech/stl/Allocators.html" target="_top">SGI's 232 http://www.sgi.com/tech/stl/Allocators.html</a>. 233 </p><p><span class="emphasis"><em>However, please ignore all discussions about the user-level 234 configuration of the lock implementation inside the STL 235 container-memory allocator on those pages. For the sake of this 236 discussion, libstdc++ configures the SGI STL implementation, 237 not you. This is quite different from how gcc pre-3.0 worked. 238 In particular, past advice was for people using g++ to 239 explicitly define _PTHREADS or other macros or port-specific 240 compilation options on the command line to get a thread-safe 241 STL. This is no longer required for any port and should no 242 longer be done unless you really know what you are doing and 243 assume all responsibility.</em></span> 244 </p><p>Since the container implementation of libstdc++ uses the SGI 245 code, we use the same definition of thread safety as SGI when 246 discussing design. A key point that beginners may miss is the 247 fourth major paragraph of the first page mentioned above 248 (<span class="emphasis"><em>For most clients...</em></span>), which points out that 249 locking must nearly always be done outside the container, by 250 client code (that'd be you, not us). There is a notable 251 exceptions to this rule. Allocators called while a container or 252 element is constructed uses an internal lock obtained and 253 released solely within libstdc++ code (in fact, this is the 254 reason STL requires any knowledge of the thread configuration). 255 </p><p>For implementing a container which does its own locking, it is 256 trivial to provide a wrapper class which obtains the lock (as 257 SGI suggests), performs the container operation, and then 258 releases the lock. This could be templatized <span class="emphasis"><em>to a certain 259 extent</em></span>, on the underlying container and/or a locking 260 mechanism. Trying to provide a catch-all general template 261 solution would probably be more trouble than it's worth. 262 </p><p>The library implementation may be configured to use the 263 high-speed caching memory allocator, which complicates thread 264 safety issues. For all details about how to globally override 265 this at application run-time 266 see <a class="link" href="using_macros.html" title="Macros">here</a>. Also 267 useful are details 268 on <a class="link" href="memory.html#std.util.memory.allocator" title="Allocators">allocator</a> 269 options and capabilities. 270 </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>