internals.html revision 1.5
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>Porting to New Hardware or Operating Systems</title><meta name="generator" content="DocBook XSL-NS Stylesheets V1.78.1" /><meta name="keywords" content="ISO C++, internals" /><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="appendix_porting.html" title="Appendix��B.�� Porting and Maintenance" /><link rel="prev" href="documentation_hacking.html" title="Writing and Generating Documentation" /><link rel="next" href="test.html" title="Test" /></head><body><div class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="3" align="center">Porting to New Hardware or Operating Systems</th></tr><tr><td width="20%" align="left"><a accesskey="p" href="documentation_hacking.html">Prev</a>��</td><th width="60%" align="center">Appendix��B.�� 3 Porting and Maintenance 4 5</th><td width="20%" align="right">��<a accesskey="n" href="test.html">Next</a></td></tr></table><hr /></div><div class="section"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="appendix.porting.internals"></a>Porting to New Hardware or Operating Systems</h2></div></div></div><p> 6</p><p>This document explains how to port libstdc++ (the GNU C++ library) to 7a new target. 8</p><p>In order to make the GNU C++ library (libstdc++) work with a new 9target, you must edit some configuration files and provide some new 10header files. Unless this is done, libstdc++ will use generic 11settings which may not be correct for your target; even if they are 12correct, they will likely be inefficient. 13 </p><p>Before you get started, make sure that you have a working C library on 14your target. The C library need not precisely comply with any 15particular standard, but should generally conform to the requirements 16imposed by the ANSI/ISO standard. 17 </p><p>In addition, you should try to verify that the C++ compiler generally 18works. It is difficult to test the C++ compiler without a working 19library, but you should at least try some minimal test cases. 20 </p><p>(Note that what we think of as a "target," the library refers to as 21a "host." The comment at the top of <code class="code">configure.ac</code> explains why.) 22 </p><div class="section"><div class="titlepage"><div><div><h3 class="title"><a id="internals.os"></a>Operating System</h3></div></div></div><p>If you are porting to a new operating system (as opposed to a new chip 23using an existing operating system), you will need to create a new 24directory in the <code class="code">config/os</code> hierarchy. For example, the IRIX 25configuration files are all in <code class="code">config/os/irix</code>. There is no set 26way to organize the OS configuration directory. For example, 27<code class="code">config/os/solaris/solaris-2.6</code> and 28<code class="code">config/os/solaris/solaris-2.7</code> are used as configuration 29directories for these two versions of Solaris. On the other hand, both 30Solaris 2.7 and Solaris 2.8 use the <code class="code">config/os/solaris/solaris-2.7</code> 31directory. The important information is that there needs to be a 32directory under <code class="code">config/os</code> to store the files for your operating 33system. 34</p><p>You might have to change the <code class="code">configure.host</code> file to ensure that 35your new directory is activated. Look for the switch statement that sets 36<code class="code">os_include_dir</code>, and add a pattern to handle your operating system 37if the default will not suffice. The switch statement switches on only 38the OS portion of the standard target triplet; e.g., the <code class="code">solaris2.8</code> 39in <code class="code">sparc-sun-solaris2.8</code>. If the new directory is named after the 40OS portion of the triplet (the default), then nothing needs to be changed. 41 </p><p>The first file to create in this directory, should be called 42<code class="code">os_defines.h</code>. This file contains basic macro definitions 43that are required to allow the C++ library to work with your C library. 44 </p><p>Several libstdc++ source files unconditionally define the macro 45<code class="code">_POSIX_SOURCE</code>. On many systems, defining this macro causes 46large portions of the C library header files to be eliminated 47at preprocessing time. Therefore, you may have to <code class="code">#undef</code> this 48macro, or define other macros (like <code class="code">_LARGEFILE_SOURCE</code> or 49<code class="code">__EXTENSIONS__</code>). You won't know what macros to define or 50undefine at this point; you'll have to try compiling the library and 51seeing what goes wrong. If you see errors about calling functions 52that have not been declared, look in your C library headers to see if 53the functions are declared there, and then figure out what macros you 54need to define. You will need to add them to the 55<code class="code">CPLUSPLUS_CPP_SPEC</code> macro in the GCC configuration file for your 56target. It will not work to simply define these macros in 57<code class="code">os_defines.h</code>. 58 </p><p>At this time, there are a few libstdc++-specific macros which may be 59defined: 60 </p><p><code class="code">_GLIBCXX_USE_C99_CHECK</code> may be defined to 1 to check C99 61function declarations (which are not covered by specialization below) 62found in system headers against versions found in the library headers 63derived from the standard. 64 </p><p><code class="code">_GLIBCXX_USE_C99_DYNAMIC</code> may be defined to an expression that 65yields 0 if and only if the system headers are exposing proper support 66for C99 functions (which are not covered by specialization below). If 67defined, it must be 0 while bootstrapping the compiler/rebuilding the 68library. 69 </p><p><code class="code">_GLIBCXX_USE_C99_LONG_LONG_CHECK</code> may be defined to 1 to check 70the set of C99 long long function declarations found in system headers 71against versions found in the library headers derived from the 72standard. 73 74 </p><p><code class="code">_GLIBCXX_USE_C99_LONG_LONG_DYNAMIC</code> may be defined to an 75expression that yields 0 if and only if the system headers are 76exposing proper support for the set of C99 long long functions. If 77defined, it must be 0 while bootstrapping the compiler/rebuilding the 78library. 79 </p><p><code class="code">_GLIBCXX_USE_C99_FP_MACROS_DYNAMIC</code> may be defined to an 80expression that yields 0 if and only if the system headers 81are exposing proper support for the related set of macros. If defined, 82it must be 0 while bootstrapping the compiler/rebuilding the library. 83 </p><p><code class="code">_GLIBCXX_USE_C99_FLOAT_TRANSCENDENTALS_CHECK</code> may be defined 84to 1 to check the related set of function declarations found in system 85headers against versions found in the library headers derived from 86the standard. 87 </p><p><code class="code">_GLIBCXX_USE_C99_FLOAT_TRANSCENDENTALS_DYNAMIC</code> may be defined 88to an expression that yields 0 if and only if the system headers 89are exposing proper support for the related set of functions. If defined, 90it must be 0 while bootstrapping the compiler/rebuilding the library. 91 </p><p>Finally, you should bracket the entire file in an include-guard, like 92this: 93 </p><pre class="programlisting"> 94 95#ifndef _GLIBCXX_OS_DEFINES 96#define _GLIBCXX_OS_DEFINES 97... 98#endif 99</pre><p>We recommend copying an existing <code class="code">os_defines.h</code> to use as a 100starting point. 101 </p></div><div class="section"><div class="titlepage"><div><div><h3 class="title"><a id="internals.cpu"></a>CPU</h3></div></div></div><p>If you are porting to a new chip (as opposed to a new operating system 102running on an existing chip), you will need to create a new directory in the 103<code class="code">config/cpu</code> hierarchy. Much like the <a class="link" href="internals.html#internals.os" title="Operating System">Operating system</a> setup, 104there are no strict rules on how to organize the CPU configuration 105directory, but careful naming choices will allow the configury to find your 106setup files without explicit help. 107</p><p>We recommend that for a target triplet <code class="code"><CPU>-<vendor>-<OS></code>, you 108name your configuration directory <code class="code">config/cpu/<CPU></code>. If you do this, 109the configury will find the directory by itself. Otherwise you will need to 110edit the <code class="code">configure.host</code> file and, in the switch statement that sets 111<code class="code">cpu_include_dir</code>, add a pattern to handle your chip. 112 </p><p>Note that some chip families share a single configuration directory, for 113example, <code class="code">alpha</code>, <code class="code">alphaev5</code>, and <code class="code">alphaev6</code> all use the 114<code class="code">config/cpu/alpha</code> directory, and there is an entry in the 115<code class="code">configure.host</code> switch statement to handle this. 116 </p><p>The <code class="code">cpu_include_dir</code> sets default locations for the files controlling 117<a class="link" href="internals.html#internals.thread_safety" title="Thread Safety">Thread safety</a> and <a class="link" href="internals.html#internals.numeric_limits" title="Numeric Limits">Numeric limits</a>, if the defaults are not 118appropriate for your chip. 119 </p></div><div class="section"><div class="titlepage"><div><div><h3 class="title"><a id="internals.char_types"></a>Character Types</h3></div></div></div><p>The library requires that you provide three header files to implement 120character classification, analogous to that provided by the C libraries 121<code class="code"><ctype.h></code> header. You can model these on the files provided in 122<code class="code">config/os/generic</code>. However, these files will almost 123certainly need some modification. 124</p><p>The first file to write is <code class="code">ctype_base.h</code>. This file provides 125some very basic information about character classification. The libstdc++ 126library assumes that your C library implements <code class="code"><ctype.h></code> by using 127a table (indexed by character code) containing integers, where each of 128these integers is a bit-mask indicating whether the character is 129upper-case, lower-case, alphabetic, etc. The <code class="code">ctype_base.h</code> 130file gives the type of the integer, and the values of the various bit 131masks. You will have to peer at your own <code class="code"><ctype.h></code> to figure out 132how to define the values required by this file. 133 </p><p>The <code class="code">ctype_base.h</code> header file does not need include guards. 134It should contain a single <code class="code">struct</code> definition called 135<code class="code">ctype_base</code>. This <code class="code">struct</code> should contain two type 136declarations, and one enumeration declaration, like this example, taken 137from the IRIX configuration: 138 </p><pre class="programlisting"> 139 struct ctype_base 140 { 141 typedef unsigned int mask; 142 typedef int* __to_type; 143 144 enum 145 { 146 space = _ISspace, 147 print = _ISprint, 148 cntrl = _IScntrl, 149 upper = _ISupper, 150 lower = _ISlower, 151 alpha = _ISalpha, 152 digit = _ISdigit, 153 punct = _ISpunct, 154 xdigit = _ISxdigit, 155 alnum = _ISalnum, 156 graph = _ISgraph 157 }; 158 }; 159</pre><p>The <code class="code">mask</code> type is the type of the elements in the table. If your 160C library uses a table to map lower-case numbers to upper-case numbers, 161and vice versa, you should define <code class="code">__to_type</code> to be the type of the 162elements in that table. If you don't mind taking a minor performance 163penalty, or if your library doesn't implement <code class="code">toupper</code> and 164<code class="code">tolower</code> in this way, you can pick any pointer-to-integer type, 165but you must still define the type. 166</p><p>The enumeration should give definitions for all the values in the above 167example, using the values from your native <code class="code"><ctype.h></code>. They can 168be given symbolically (as above), or numerically, if you prefer. You do 169not have to include <code class="code"><ctype.h></code> in this header; it will always be 170included before <code class="code">ctype_base.h</code> is included. 171 </p><p>The next file to write is <code class="code">ctype_configure_char.cc</code>. 172The first function that must be written is the <code class="code">ctype<char>::ctype</code> constructor. Here is the IRIX example: 173 </p><pre class="programlisting"> 174ctype<char>::ctype(const mask* __table = 0, bool __del = false, 175 size_t __refs = 0) 176 : _Ctype_nois<char>(__refs), _M_del(__table != 0 && __del), 177 _M_toupper(NULL), 178 _M_tolower(NULL), 179 _M_ctable(NULL), 180 _M_table(!__table 181 ? (const mask*) (__libc_attr._ctype_tbl->_class + 1) 182 : __table) 183 { } 184</pre><p>There are two parts of this that you might choose to alter. The first, 185and most important, is the line involving <code class="code">__libc_attr</code>. That is 186IRIX system-dependent code that gets the base of the table mapping 187character codes to attributes. You need to substitute code that obtains 188the address of this table on your system. If you want to use your 189operating system's tables to map upper-case letters to lower-case, and 190vice versa, you should initialize <code class="code">_M_toupper</code> and 191<code class="code">_M_tolower</code> with those tables, in similar fashion. 192</p><p>Now, you have to write two functions to convert from upper-case to 193lower-case, and vice versa. Here are the IRIX versions: 194 </p><pre class="programlisting"> 195 char 196 ctype<char>::do_toupper(char __c) const 197 { return _toupper(__c); } 198 199 char 200 ctype<char>::do_tolower(char __c) const 201 { return _tolower(__c); } 202</pre><p>Your C library provides equivalents to IRIX's <code class="code">_toupper</code> and 203<code class="code">_tolower</code>. If you initialized <code class="code">_M_toupper</code> and 204<code class="code">_M_tolower</code> above, then you could use those tables instead. 205</p><p>Finally, you have to provide two utility functions that convert strings 206of characters. The versions provided here will always work - but you 207could use specialized routines for greater performance if you have 208machinery to do that on your system: 209 </p><pre class="programlisting"> 210 const char* 211 ctype<char>::do_toupper(char* __low, const char* __high) const 212 { 213 while (__low < __high) 214 { 215 *__low = do_toupper(*__low); 216 ++__low; 217 } 218 return __high; 219 } 220 221 const char* 222 ctype<char>::do_tolower(char* __low, const char* __high) const 223 { 224 while (__low < __high) 225 { 226 *__low = do_tolower(*__low); 227 ++__low; 228 } 229 return __high; 230 } 231</pre><p>You must also provide the <code class="code">ctype_inline.h</code> file, which 232contains a few more functions. On most systems, you can just copy 233<code class="code">config/os/generic/ctype_inline.h</code> and use it on your system. 234 </p><p>In detail, the functions provided test characters for particular 235properties; they are analogous to the functions like <code class="code">isalpha</code> and 236<code class="code">islower</code> provided by the C library. 237 </p><p>The first function is implemented like this on IRIX: 238 </p><pre class="programlisting"> 239 bool 240 ctype<char>:: 241 is(mask __m, char __c) const throw() 242 { return (_M_table)[(unsigned char)(__c)] & __m; } 243</pre><p>The <code class="code">_M_table</code> is the table passed in above, in the constructor. 244This is the table that contains the bitmasks for each character. The 245implementation here should work on all systems. 246</p><p>The next function is: 247 </p><pre class="programlisting"> 248 const char* 249 ctype<char>:: 250 is(const char* __low, const char* __high, mask* __vec) const throw() 251 { 252 while (__low < __high) 253 *__vec++ = (_M_table)[(unsigned char)(*__low++)]; 254 return __high; 255 } 256</pre><p>This function is similar; it copies the masks for all the characters 257from <code class="code">__low</code> up until <code class="code">__high</code> into the vector given by 258<code class="code">__vec</code>. 259</p><p>The last two functions again are entirely generic: 260 </p><pre class="programlisting"> 261 const char* 262 ctype<char>:: 263 scan_is(mask __m, const char* __low, const char* __high) const throw() 264 { 265 while (__low < __high && !this->is(__m, *__low)) 266 ++__low; 267 return __low; 268 } 269 270 const char* 271 ctype<char>:: 272 scan_not(mask __m, const char* __low, const char* __high) const throw() 273 { 274 while (__low < __high && this->is(__m, *__low)) 275 ++__low; 276 return __low; 277 } 278</pre></div><div class="section"><div class="titlepage"><div><div><h3 class="title"><a id="internals.thread_safety"></a>Thread Safety</h3></div></div></div><p>The C++ library string functionality requires a couple of atomic 279operations to provide thread-safety. If you don't take any special 280action, the library will use stub versions of these functions that are 281not thread-safe. They will work fine, unless your applications are 282multi-threaded. 283</p><p>If you want to provide custom, safe, versions of these functions, there 284are two distinct approaches. One is to provide a version for your CPU, 285using assembly language constructs. The other is to use the 286thread-safety primitives in your operating system. In either case, you 287make a file called <code class="code">atomicity.h</code>, and the variable 288<code class="code">ATOMICITYH</code> must point to this file. 289 </p><p>If you are using the assembly-language approach, put this code in 290<code class="code">config/cpu/<chip>/atomicity.h</code>, where chip is the name of 291your processor (see <a class="link" href="internals.html#internals.cpu" title="CPU">CPU</a>). No additional changes are necessary to 292locate the file in this case; <code class="code">ATOMICITYH</code> will be set by default. 293 </p><p>If you are using the operating system thread-safety primitives approach, 294you can also put this code in the same CPU directory, in which case no more 295work is needed to locate the file. For examples of this approach, 296see the <code class="code">atomicity.h</code> file for IRIX or IA64. 297 </p><p>Alternatively, if the primitives are more closely related to the OS 298than they are to the CPU, you can put the <code class="code">atomicity.h</code> file in 299the <a class="link" href="internals.html#internals.os" title="Operating System">Operating system</a> directory instead. In this case, you must 300edit <code class="code">configure.host</code>, and in the switch statement that handles 301operating systems, override the <code class="code">ATOMICITYH</code> variable to point to 302the appropriate <code class="code">os_include_dir</code>. For examples of this approach, 303see the <code class="code">atomicity.h</code> file for AIX. 304 </p><p>With those bits out of the way, you have to actually write 305<code class="code">atomicity.h</code> itself. This file should be wrapped in an 306include guard named <code class="code">_GLIBCXX_ATOMICITY_H</code>. It should define one 307type, and two functions. 308 </p><p>The type is <code class="code">_Atomic_word</code>. Here is the version used on IRIX: 309 </p><pre class="programlisting"> 310typedef long _Atomic_word; 311</pre><p>This type must be a signed integral type supporting atomic operations. 312If you're using the OS approach, use the same type used by your system's 313primitives. Otherwise, use the type for which your CPU provides atomic 314primitives. 315</p><p>Then, you must provide two functions. The bodies of these functions 316must be equivalent to those provided here, but using atomic operations: 317 </p><pre class="programlisting"> 318 static inline _Atomic_word 319 __attribute__ ((__unused__)) 320 __exchange_and_add (_Atomic_word* __mem, int __val) 321 { 322 _Atomic_word __result = *__mem; 323 *__mem += __val; 324 return __result; 325 } 326 327 static inline void 328 __attribute__ ((__unused__)) 329 __atomic_add (_Atomic_word* __mem, int __val) 330 { 331 *__mem += __val; 332 } 333</pre></div><div class="section"><div class="titlepage"><div><div><h3 class="title"><a id="internals.numeric_limits"></a>Numeric Limits</h3></div></div></div><p>The C++ library requires information about the fundamental data types, 334such as the minimum and maximum representable values of each type. 335You can define each of these values individually, but it is usually 336easiest just to indicate how many bits are used in each of the data 337types and let the library do the rest. For information about the 338macros to define, see the top of <code class="code">include/bits/std_limits.h</code>. 339</p><p>If you need to define any macros, you can do so in <code class="code">os_defines.h</code>. 340However, if all operating systems for your CPU are likely to use the 341same values, you can provide a CPU-specific file instead so that you 342do not have to provide the same definitions for each operating system. 343To take that approach, create a new file called <code class="code">cpu_limits.h</code> in 344your CPU configuration directory (see <a class="link" href="internals.html#internals.cpu" title="CPU">CPU</a>). 345 </p></div><div class="section"><div class="titlepage"><div><div><h3 class="title"><a id="internals.libtool"></a>Libtool</h3></div></div></div><p>The C++ library is compiled, archived and linked with libtool. 346Explaining the full workings of libtool is beyond the scope of this 347document, but there are a few, particular bits that are necessary for 348porting. 349</p><p>Some parts of the libstdc++ library are compiled with the libtool 350<code class="code">--tags CXX</code> option (the C++ definitions for libtool). Therefore, 351<code class="code">ltcf-cxx.sh</code> in the top-level directory needs to have the correct 352logic to compile and archive objects equivalent to the C version of libtool, 353<code class="code">ltcf-c.sh</code>. Some libtool targets have definitions for C but not 354for C++, or C++ definitions which have not been kept up to date. 355 </p><p>The C++ run-time library contains initialization code that needs to be 356run as the library is loaded. Often, that requires linking in special 357object files when the C++ library is built as a shared library, or 358taking other system-specific actions. 359 </p><p>The libstdc++ library is linked with the C version of libtool, even 360though it is a C++ library. Therefore, the C version of libtool needs to 361ensure that the run-time library initializers are run. The usual way to 362do this is to build the library using <code class="code">gcc -shared</code>. 363 </p><p>If you need to change how the library is linked, look at 364<code class="code">ltcf-c.sh</code> in the top-level directory. Find the switch statement 365that sets <code class="code">archive_cmds</code>. Here, adjust the setting for your 366operating system. 367 </p></div></div><div class="navfooter"><hr /><table width="100%" summary="Navigation footer"><tr><td width="40%" align="left"><a accesskey="p" href="documentation_hacking.html">Prev</a>��</td><td width="20%" align="center"><a accesskey="u" href="appendix_porting.html">Up</a></td><td width="40%" align="right">��<a accesskey="n" href="test.html">Next</a></td></tr><tr><td width="40%" align="left" valign="top">Writing and Generating Documentation��</td><td width="20%" align="center"><a accesskey="h" href="../index.html">Home</a></td><td width="40%" align="right" valign="top">��Test</td></tr></table></div></body></html>