docs/SystemLibrary.html

   1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
   2                       "http://www.w3.org/TR/html4/strict.dtd">
   3 <html>
   4 <head>
   5   <title>System Library</title>
   6   <link rel="stylesheet" href="llvm.css" type="text/css">
   7 </head>
   8 <body>
   9
  10 <h1>System Library</h1>
  11 <ul>
  12   <li><a href="#abstract">Abstract</a></li>
  13   <li><a href="#requirements">Keeping LLVM Portable</a>
  14   <ol>
  15     <li><a href="#headers">Don't Include System Headers</a></li>
  16     <li><a href="#expose">Don't Expose System Headers</a></li>
  17     <li><a href="#c_headers">Allow Standard C Header Files</a></li>
  18     <li><a href="#cpp_headers">Allow Standard C++ Header Files</a></li>
  19     <li><a href="#highlev">High-Level Interface</a></li>
  20     <li><a href="#nofunc">No Exposed Functions</a></li>
  21     <li><a href="#nodata">No Exposed Data</a></li>
  22     <li><a href="#nodupl">No Duplicate Implementations</a></li>
  23     <li><a href="#nounused">No Unused Functionality</a></li>
  24     <li><a href="#virtuals">No Virtual Methods</a></li>
  25     <li><a href="#softerrors">Minimize Soft Errors</a></li>
  26     <li><a href="#throw_spec">No throw() Specifications</a></li>
  27     <li><a href="#organization">Code Organization</a></li>
  28     <li><a href="#semantics">Consistent Semantics</a></li>
  29     <li><a href="#bug">Tracking Bugzilla Bug: 351</a></li>
  30   </ol></li>
  31 </ul>
  32
  33 <div class="doc_author">
  34   <p>Written by <a href="mailto:rspencer@x10sys.com">Reid Spencer</a></p>
  35 </div>
  36
  37
  38 <!-- *********************************************************************** -->
  39 <h2><a name="abstract">Abstract</a></h2>
  40 <div>
  41   <p>This document provides some details on LLVM's System Library, located in
  42   the source at <tt>lib/System</tt> and <tt>include/llvm/System</tt>. The
  43   library's purpose is to shield LLVM from the differences between operating
  44   systems for the few services LLVM needs from the operating system. Much of
  45   LLVM is written using portability features of standard C++. However, in a few
  46   areas, system dependent facilities are needed and the System Library is the
  47   wrapper around those system calls.</p>
  48   <p>By centralizing LLVM's use of operating system interfaces, we make it
  49   possible for the LLVM tool chain and runtime libraries to be more easily
  50   ported to new platforms since (theoretically) only <tt>lib/System</tt> needs
  51   to be ported.  This library also unclutters the rest of LLVM from #ifdef use
  52   and special cases for specific operating systems. Such uses are replaced
  53   with simple calls to the interfaces provided in <tt>include/llvm/System</tt>.
  54   </p>
  55   <p>Note that the System Library is not intended to be a complete operating
  56   system wrapper (such as the Adaptive Communications Environment (ACE) or
  57   Apache Portable Runtime (APR)), but only provides the functionality necessary
  58   to support LLVM.
  59   <p>The System Library was written by Reid Spencer who formulated the
  60   design based on similar work originating from the eXtensible Programming
  61   System (XPS). Several people helped with the effort; especially,
  62   Jeff Cohen and Henrik Bach on the Win32 port.</p>
  63 </div>
  64
  65 <!-- *********************************************************************** -->
  66 <h2>
  67   <a name="requirements">Keeping LLVM Portable</a>
  68 </h2>
  69 <div>
  70   <p>In order to keep LLVM portable, LLVM developers should adhere to a set of
  71   portability rules associated with the System Library. Adherence to these rules
  72   should help the System Library achieve its goal of shielding LLVM from the
  73   variations in operating system interfaces and doing so efficiently.  The
  74   following sections define the rules needed to fulfill this objective.</p>
  75
  76 <!-- ======================================================================= -->
  77 <h3><a name="headers">Don't Include System Headers</a></h3>
  78 <div>
  79   <p>Except in <tt>lib/System</tt>, no LLVM source code should directly
  80   <tt>#include</tt> a system header. Care has been taken to remove all such
  81   <tt>#includes</tt> from LLVM while <tt>lib/System</tt> was being
  82   developed.  Specifically this means that header files like "unistd.h",
  83   "windows.h", "stdio.h", and "string.h" are forbidden to be included by LLVM
  84   source code outside the implementation of <tt>lib/System</tt>.</p>
  85   <p>To obtain system-dependent functionality, existing interfaces to the system
  86   found in <tt>include/llvm/System</tt> should be used. If an appropriate
  87   interface is not available, it should be added to <tt>include/llvm/System</tt>
  88   and implemented in <tt>lib/System</tt> for all supported platforms.</p>
  89 </div>
  90
  91 <!-- ======================================================================= -->
  92 <h3><a name="expose">Don't Expose System Headers</a></h3>
  93 <div>
  94   <p>The System Library must shield LLVM from <em>all</em> system headers. To
  95   obtain system level functionality, LLVM source must
  96   <tt>#include "llvm/System/Thing.h"</tt> and nothing else. This means that
  97   <tt>Thing.h</tt> cannot expose any system header files. This protects LLVM
  98   from accidentally using system specific functionality and only allows it
  99   via the <tt>lib/System</tt> interface.</p>
 100 </div>
 101
 102 <!-- ======================================================================= -->
 103 <h3><a name="c_headers">Use Standard C Headers</a></h3>
 104 <div>
 105   <p>The <em>standard</em> C headers (the ones beginning with "c") are allowed
 106   to be exposed through the <tt>lib/System</tt> interface. These headers and
 107   the things they declare are considered to be platform agnostic. LLVM source
 108   files may include them directly or obtain their inclusion through
 109   <tt>lib/System</tt> interfaces.</p>
 110 </div>
 111
 112 <!-- ======================================================================= -->
 113 <h3><a name="cpp_headers">Use Standard C++ Headers</a></h3>
 114 <div>
 115   <p>The <em>standard</em> C++ headers from the standard C++ library and
 116   standard template library may be exposed through the <tt>lib/System</tt>
 117   interface. These headers and the things they declare are considered to be
 118   platform agnostic. LLVM source files may include them or obtain their
 119   inclusion through lib/System interfaces.</p>
 120 </div>
 121
 122 <!-- ======================================================================= -->
 123 <h3><a name="highlev">High Level Interface</a></h3>
 124 <div>
 125   <p>The entry points specified in the interface of lib/System must be aimed at
 126   completing some reasonably high level task needed by LLVM. We do not want to
 127   simply wrap each operating system call. It would be preferable to wrap several
 128   operating system calls that are always used in conjunction with one another by
 129   LLVM.</p>
 130   <p>For example, consider what is needed to execute a program, wait for it to
 131   complete, and return its result code. On Unix, this involves the following
 132   operating system calls: <tt>getenv, fork, execve,</tt> and <tt>wait</tt>. The
 133   correct thing for lib/System to provide is a function, say
 134   <tt>ExecuteProgramAndWait</tt>, that implements the functionality completely.
 135   what we don't want is wrappers for the operating system calls involved.</p>
 136   <p>There must <em>not</em> be a one-to-one relationship between operating
 137   system calls and the System library's interface. Any such interface function
 138   will be suspicious.</p>
 139 </div>
 140
 141 <!-- ======================================================================= -->
 142 <h3><a name="nounused">No Unused Functionality</a></h3>
 143 <div>
 144   <p>There must be no functionality specified in the interface of lib/System
 145   that isn't actually used by LLVM. We're not writing a general purpose
 146   operating system wrapper here, just enough to satisfy LLVM's needs. And, LLVM
 147   doesn't need much. This design goal aims to keep the lib/System interface
 148   small and understandable which should foster its actual use and adoption.</p>
 149 </div>
 150
 151 <!-- ======================================================================= -->
 152 <h3><a name="nodupl">No Duplicate Implementations</a></h3>
 153 <div>
 154   <p>The implementation of a function for a given platform must be written
 155   exactly once. This implies that it must be possible to apply a function's
 156   implementation to multiple operating systems if those operating systems can
 157   share the same implementation. This rule applies to the set of operating
 158   systems supported for a given class of operating system (e.g. Unix, Win32).
 159   </p>
 160 </div>
 161
 162 <!-- ======================================================================= -->
 163 <h3><a name="virtuals">No Virtual Methods</a></h3>
 164 <div>
 165   <p>The System Library interfaces can be called quite frequently by LLVM. In
 166   order to make those calls as efficient as possible, we discourage the use of
 167   virtual methods. There is no need to use inheritance for implementation
 168   differences, it just adds complexity. The <tt>#include</tt> mechanism works
 169   just fine.</p>
 170 </div>
 171
 172 <!-- ======================================================================= -->
 173 <h3><a name="nofunc">No Exposed Functions</a></h3>
 174 <div>
 175   <p>Any functions defined by system libraries (i.e. not defined by lib/System)
 176   must not be exposed through the lib/System interface, even if the header file
 177   for that function is not exposed. This prevents inadvertent use of system
 178   specific functionality.</p>
 179   <p>For example, the <tt>stat</tt> system call is notorious for having
 180   variations in the data it provides. <tt>lib/System</tt> must not declare
 181   <tt>stat</tt> nor allow it to be declared. Instead it should provide its own
 182   interface to discovering information about files and directories. Those
 183   interfaces may be implemented in terms of <tt>stat</tt> but that is strictly
 184   an implementation detail. The interface provided by the System Library must
 185   be implemented on all platforms (even those without <tt>stat</tt>).</p>
 186 </div>
 187
 188 <!-- ======================================================================= -->
 189 <h3><a name="nodata">No Exposed Data</a></h3>
 190 <div>
 191   <p>Any data defined by system libraries (i.e. not defined by lib/System) must
 192   not be exposed through the lib/System interface, even if the header file for
 193   that function is not exposed. As with functions, this prevents inadvertent use
 194   of data that might not exist on all platforms.</p>
 195 </div>
 196
 197 <!-- ======================================================================= -->
 198 <h3><a name="softerrors">Minimize Soft Errors</a></h3>
 199 <div>
 200   <p>Operating system interfaces will generally provide error results for every
 201   little thing that could go wrong. In almost all cases, you can divide these
 202   error results into two groups: normal/good/soft and abnormal/bad/hard. That
 203   is, some of the errors are simply information like "file not found",
 204   "insufficient privileges", etc. while other errors are much harder like
 205   "out of space", "bad disk sector", or "system call interrupted". We'll call
 206   the first group "<i>soft</i>" errors and the second group "<i>hard</i>"
 207   errors.<p>
 208   <p>lib/System must always attempt to minimize soft errors.
 209   This is a design requirement because the
 210   minimization of soft errors can affect the granularity and the nature of the
 211   interface. In general, if you find that you're wanting to throw soft errors,
 212   you must review the granularity of the interface because it is likely you're
 213   trying to implement something that is too low level. The rule of thumb is to
 214   provide interface functions that <em>can't</em> fail, except when faced with
 215   hard errors.</p>
 216   <p>For a trivial example, suppose we wanted to add an "OpenFileForWriting"
 217   function. For many operating systems, if the file doesn't exist, attempting
 218   to open the file will produce an error.  However, lib/System should not
 219   simply throw that error if it occurs because its a soft error. The problem
 220   is that the interface function, OpenFileForWriting is too low level. It should
 221   be OpenOrCreateFileForWriting. In the case of the soft "doesn't exist" error,
 222   this function would just create it and then open it for writing.</p>
 223   <p>This design principle needs to be maintained in lib/System because it
 224   avoids the propagation of soft error handling throughout the rest of LLVM.
 225   Hard errors will generally just cause a termination for an LLVM tool so don't
 226   be bashful about throwing them.</p>
 227   <p>Rules of thumb:</p>
 228   <ol>
 229     <li>Don't throw soft errors, only hard errors.</li>
 230     <li>If you're tempted to throw a soft error, re-think the interface.</li>
 231     <li>Handle internally the most common normal/good/soft error conditions
 232     so the rest of LLVM doesn't have to.</li>
 233   </ol>
 234 </div>
 235
 236 <!-- ======================================================================= -->
 237 <h3><a name="throw_spec">No throw Specifications</a></h3>
 238 <div>
 239   <p>None of the lib/System interface functions may be declared with C++
 240   <tt>throw()</tt> specifications on them. This requirement makes sure that the
 241   compiler does not insert additional exception handling code into the interface
 242   functions. This is a performance consideration: lib/System functions are at
 243   the bottom of many call chains and as such can be frequently called. We
 244   need them to be as efficient as possible.  However, no routines in the
 245   system library should actually throw exceptions.</p>
 246 </div>
 247
 248 <!-- ======================================================================= -->
 249 <h3><a name="organization">Code Organization</a></h3>
 250 <div>
 251   <p>Implementations of the System Library interface are separated by their
 252   general class of operating system. Currently only Unix and Win32 classes are
 253   defined but more could be added for other operating system classifications.
 254   To distinguish which implementation to compile, the code in lib/System uses
 255   the LLVM_ON_UNIX and LLVM_ON_WIN32 #defines provided via configure through the
 256   llvm/Config/config.h file. Each source file in lib/System, after implementing
 257   the generic (operating system independent) functionality needs to include the
 258   correct implementation using a set of <tt>#if defined(LLVM_ON_XYZ)</tt>
 259   directives. For example, if we had lib/System/File.cpp, we'd expect to see in
 260   that file:</p>
 261   <pre><tt>
 262   #if defined(LLVM_ON_UNIX)
 263   #include "Unix/File.cpp"
 264   #endif
 265   #if defined(LLVM_ON_WIN32)
 266   #include "Win32/File.cpp"
 267   #endif
 268   </tt></pre>
 269   <p>The implementation in lib/System/Unix/File.cpp should handle all Unix
 270   variants. The implementation in lib/System/Win32/File.cpp should handle all
 271   Win32 variants.  What this does is quickly differentiate the basic class of
 272   operating system that will provide the implementation. The specific details
 273   for a given platform must still be determined through the use of
 274   <tt>#ifdef</tt>.</p>
 275 </div>
 276
 277 <!-- ======================================================================= -->
 278 <h3><a name="semantics">Consistent Semantics</a></h3>
 279 <div>
 280   <p>The implementation of a lib/System interface can vary drastically between
 281   platforms. That's okay as long as the end result of the interface function
 282   is the same. For example, a function to create a directory is pretty straight
 283   forward on all operating system. System V IPC on the other hand isn't even
 284   supported on all platforms. Instead of "supporting" System V IPC, lib/System
 285   should provide an interface to the basic concept of inter-process
 286   communications. The implementations might use System V IPC if that was
 287   available or named pipes, or whatever gets the job done effectively for a
 288   given operating system.  In all cases, the interface and the implementation
 289   must be semantically consistent. </p>
 290 </div>
 291
 292 <!-- ======================================================================= -->
 293 <h3><a name="bug">Bug 351</a></h3>
 294 <div>
 295   <p>See <a href="http://llvm.org/PR351">bug 351</a>
 296   for further details on the progress of this work</p>
 297 </div>
 298
 299 </div>
 300
 301 <!-- *********************************************************************** -->
 302
 303 <hr>
 304 <address>
 305   <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
 306   src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
 307   <a href="http://validator.w3.org/check/referer"><img
 308   src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
 309
 310   <a href="mailto:rspencer@x10sys.com">Reid Spencer</a><br>
 311   <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
 312   Last modified: $Date$
 313 </address>
 314 </body>
 315 </html>