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>"Clang" CFE Internals Manual</title> 6<link type="text/css" rel="stylesheet" href="../menu.css"> 7<link type="text/css" rel="stylesheet" href="../content.css"> 8<style type="text/css"> 9td { 10 vertical-align: top; 11} 12</style> 13</head> 14<body> 15 16<!--#include virtual="../menu.html.incl"--> 17 18<div id="content"> 19 20<h1>"Clang" CFE Internals Manual</h1> 21 22<ul> 23<li><a href="#intro">Introduction</a></li> 24<li><a href="#libsupport">LLVM Support Library</a></li> 25<li><a href="#libbasic">The Clang 'Basic' Library</a> 26 <ul> 27 <li><a href="#Diagnostics">The Diagnostics Subsystem</a></li> 28 <li><a href="#SourceLocation">The SourceLocation and SourceManager 29 classes</a></li> 30 <li><a href="#SourceRange">SourceRange and CharSourceRange</a></li> 31 </ul> 32</li> 33<li><a href="#libdriver">The Driver Library</a> 34</li> 35<li><a href="#pch">Precompiled Headers</a> 36<li><a href="#libfrontend">The Frontend Library</a> 37</li> 38<li><a href="#liblex">The Lexer and Preprocessor Library</a> 39 <ul> 40 <li><a href="#Token">The Token class</a></li> 41 <li><a href="#Lexer">The Lexer class</a></li> 42 <li><a href="#AnnotationToken">Annotation Tokens</a></li> 43 <li><a href="#TokenLexer">The TokenLexer class</a></li> 44 <li><a href="#MultipleIncludeOpt">The MultipleIncludeOpt class</a></li> 45 </ul> 46</li> 47<li><a href="#libparse">The Parser Library</a> 48</li> 49<li><a href="#libast">The AST Library</a> 50 <ul> 51 <li><a href="#Type">The Type class and its subclasses</a></li> 52 <li><a href="#QualType">The QualType class</a></li> 53 <li><a href="#DeclarationName">Declaration names</a></li> 54 <li><a href="#DeclContext">Declaration contexts</a> 55 <ul> 56 <li><a href="#Redeclarations">Redeclarations and Overloads</a></li> 57 <li><a href="#LexicalAndSemanticContexts">Lexical and Semantic 58 Contexts</a></li> 59 <li><a href="#TransparentContexts">Transparent Declaration Contexts</a></li> 60 <li><a href="#MultiDeclContext">Multiply-Defined Declaration Contexts</a></li> 61 </ul> 62 </li> 63 <li><a href="#CFG">The CFG class</a></li> 64 <li><a href="#Constants">Constant Folding in the Clang AST</a></li> 65 </ul> 66</li> 67<li><a href="#Howtos">Howto guides</a> 68 <ul> 69 <li><a href="#AddingAttributes">How to add an attribute</a></li> 70 <li><a href="#AddingExprStmt">How to add a new expression or statement</a></li> 71 </ul> 72</li> 73</ul> 74 75 76<!-- ======================================================================= --> 77<h2 id="intro">Introduction</h2> 78<!-- ======================================================================= --> 79 80<p>This document describes some of the more important APIs and internal design 81decisions made in the Clang C front-end. The purpose of this document is to 82both capture some of this high level information and also describe some of the 83design decisions behind it. This is meant for people interested in hacking on 84Clang, not for end-users. The description below is categorized by 85libraries, and does not describe any of the clients of the libraries.</p> 86 87<!-- ======================================================================= --> 88<h2 id="libsupport">LLVM Support Library</h2> 89<!-- ======================================================================= --> 90 91<p>The LLVM libsupport library provides many underlying libraries and 92<a href="http://llvm.org/docs/ProgrammersManual.html">data-structures</a>, 93including command line option processing, various containers and a system 94abstraction layer, which is used for file system access.</p> 95 96<!-- ======================================================================= --> 97<h2 id="libbasic">The Clang 'Basic' Library</h2> 98<!-- ======================================================================= --> 99 100<p>This library certainly needs a better name. The 'basic' library contains a 101number of low-level utilities for tracking and manipulating source buffers, 102locations within the source buffers, diagnostics, tokens, target abstraction, 103and information about the subset of the language being compiled for.</p> 104 105<p>Part of this infrastructure is specific to C (such as the TargetInfo class), 106other parts could be reused for other non-C-based languages (SourceLocation, 107SourceManager, Diagnostics, FileManager). When and if there is future demand 108we can figure out if it makes sense to introduce a new library, move the general 109classes somewhere else, or introduce some other solution.</p> 110 111<p>We describe the roles of these classes in order of their dependencies.</p> 112 113 114<!-- ======================================================================= --> 115<h3 id="Diagnostics">The Diagnostics Subsystem</h3> 116<!-- ======================================================================= --> 117 118<p>The Clang Diagnostics subsystem is an important part of how the compiler 119communicates with the human. Diagnostics are the warnings and errors produced 120when the code is incorrect or dubious. In Clang, each diagnostic produced has 121(at the minimum) a unique ID, an English translation associated with it, a <a 122href="#SourceLocation">SourceLocation</a> to "put the caret", and a severity (e.g. 123<tt>WARNING</tt> or <tt>ERROR</tt>). They can also optionally include a number 124of arguments to the dianostic (which fill in "%0"'s in the string) as well as a 125number of source ranges that related to the diagnostic.</p> 126 127<p>In this section, we'll be giving examples produced by the Clang command line 128driver, but diagnostics can be <a href="#DiagnosticClient">rendered in many 129different ways</a> depending on how the DiagnosticClient interface is 130implemented. A representative example of a diagnostic is:</p> 131 132<pre> 133t.c:38:15: error: invalid operands to binary expression ('int *' and '_Complex float') 134 <span style="color:darkgreen">P = (P-42) + Gamma*4;</span> 135 <span style="color:blue">~~~~~~ ^ ~~~~~~~</span> 136</pre> 137 138<p>In this example, you can see the English translation, the severity (error), 139you can see the source location (the caret ("^") and file/line/column info), 140the source ranges "~~~~", arguments to the diagnostic ("int*" and "_Complex 141float"). You'll have to believe me that there is a unique ID backing the 142diagnostic :).</p> 143 144<p>Getting all of this to happen has several steps and involves many moving 145pieces, this section describes them and talks about best practices when adding 146a new diagnostic.</p> 147 148<!-- ============================= --> 149<h4>The Diagnostic*Kinds.td files</h4> 150<!-- ============================= --> 151 152<p>Diagnostics are created by adding an entry to one of the <tt> 153clang/Basic/Diagnostic*Kinds.td</tt> files, depending on what library will 154be using it. From this file, tblgen generates the unique ID of the diagnostic, 155the severity of the diagnostic and the English translation + format string.</p> 156 157<p>There is little sanity with the naming of the unique ID's right now. Some 158start with err_, warn_, ext_ to encode the severity into the name. Since the 159enum is referenced in the C++ code that produces the diagnostic, it is somewhat 160useful for it to be reasonably short.</p> 161 162<p>The severity of the diagnostic comes from the set {<tt>NOTE</tt>, 163<tt>WARNING</tt>, <tt>EXTENSION</tt>, <tt>EXTWARN</tt>, <tt>ERROR</tt>}. The 164<tt>ERROR</tt> severity is used for diagnostics indicating the program is never 165acceptable under any circumstances. When an error is emitted, the AST for the 166input code may not be fully built. The <tt>EXTENSION</tt> and <tt>EXTWARN</tt> 167severities are used for extensions to the language that Clang accepts. This 168means that Clang fully understands and can represent them in the AST, but we 169produce diagnostics to tell the user their code is non-portable. The difference 170is that the former are ignored by default, and the later warn by default. The 171<tt>WARNING</tt> severity is used for constructs that are valid in the currently 172selected source language but that are dubious in some way. The <tt>NOTE</tt> 173level is used to staple more information onto previous diagnostics.</p> 174 175<p>These <em>severities</em> are mapped into a smaller set (the 176Diagnostic::Level enum, {<tt>Ignored</tt>, <tt>Note</tt>, <tt>Warning</tt>, 177<tt>Error</tt>, <tt>Fatal</tt> }) of output <em>levels</em> by the diagnostics 178subsystem based on various configuration options. Clang internally supports a 179fully fine grained mapping mechanism that allows you to map almost any 180diagnostic to the output level that you want. The only diagnostics that cannot 181be mapped are <tt>NOTE</tt>s, which always follow the severity of the previously 182emitted diagnostic and <tt>ERROR</tt>s, which can only be mapped to 183<tt>Fatal</tt> (it is not possible to turn an error into a warning, 184for example).</p> 185 186<p>Diagnostic mappings are used in many ways. For example, if the user 187specifies <tt>-pedantic</tt>, <tt>EXTENSION</tt> maps to <tt>Warning</tt>, if 188they specify <tt>-pedantic-errors</tt>, it turns into <tt>Error</tt>. This is 189used to implement options like <tt>-Wunused_macros</tt>, <tt>-Wundef</tt> etc. 190</p> 191 192<p> 193Mapping to <tt>Fatal</tt> should only be used for diagnostics that are 194considered so severe that error recovery won't be able to recover sensibly from 195them (thus spewing a ton of bogus errors). One example of this class of error 196are failure to #include a file. 197</p> 198 199<!-- ================= --> 200<h4>The Format String</h4> 201<!-- ================= --> 202 203<p>The format string for the diagnostic is very simple, but it has some power. 204It takes the form of a string in English with markers that indicate where and 205how arguments to the diagnostic are inserted and formatted. For example, here 206are some simple format strings:</p> 207 208<pre> 209 "binary integer literals are an extension" 210 "format string contains '\\0' within the string body" 211 "more '<b>%%</b>' conversions than data arguments" 212 "invalid operands to binary expression (<b>%0</b> and <b>%1</b>)" 213 "overloaded '<b>%0</b>' must be a <b>%select{unary|binary|unary or binary}2</b> operator" 214 " (has <b>%1</b> parameter<b>%s1</b>)" 215</pre> 216 217<p>These examples show some important points of format strings. You can use any 218 plain ASCII character in the diagnostic string except "%" without a problem, 219 but these are C strings, so you have to use and be aware of all the C escape 220 sequences (as in the second example). If you want to produce a "%" in the 221 output, use the "%%" escape sequence, like the third diagnostic. Finally, 222 Clang uses the "%...[digit]" sequences to specify where and how arguments to 223 the diagnostic are formatted.</p> 224 225<p>Arguments to the diagnostic are numbered according to how they are specified 226 by the C++ code that <a href="#producingdiag">produces them</a>, and are 227 referenced by <tt>%0</tt> .. <tt>%9</tt>. If you have more than 10 arguments 228 to your diagnostic, you are doing something wrong :). Unlike printf, there 229 is no requirement that arguments to the diagnostic end up in the output in 230 the same order as they are specified, you could have a format string with 231 <tt>"%1 %0"</tt> that swaps them, for example. The text in between the 232 percent and digit are formatting instructions. If there are no instructions, 233 the argument is just turned into a string and substituted in.</p> 234 235<p>Here are some "best practices" for writing the English format string:</p> 236 237<ul> 238<li>Keep the string short. It should ideally fit in the 80 column limit of the 239 <tt>DiagnosticKinds.td</tt> file. This avoids the diagnostic wrapping when 240 printed, and forces you to think about the important point you are conveying 241 with the diagnostic.</li> 242<li>Take advantage of location information. The user will be able to see the 243 line and location of the caret, so you don't need to tell them that the 244 problem is with the 4th argument to the function: just point to it.</li> 245<li>Do not capitalize the diagnostic string, and do not end it with a 246 period.</li> 247<li>If you need to quote something in the diagnostic string, use single 248 quotes.</li> 249</ul> 250 251<p>Diagnostics should never take random English strings as arguments: you 252shouldn't use <tt>"you have a problem with %0"</tt> and pass in things like 253<tt>"your argument"</tt> or <tt>"your return value"</tt> as arguments. Doing 254this prevents <a href="#translation">translating</a> the Clang diagnostics to 255other languages (because they'll get random English words in their otherwise 256localized diagnostic). The exceptions to this are C/C++ language keywords 257(e.g. auto, const, mutable, etc) and C/C++ operators (<tt>/=</tt>). Note 258that things like "pointer" and "reference" are not keywords. On the other 259hand, you <em>can</em> include anything that comes from the user's source code, 260including variable names, types, labels, etc. The 'select' format can be 261used to achieve this sort of thing in a localizable way, see below.</p> 262 263<!-- ==================================== --> 264<h4>Formatting a Diagnostic Argument</h4> 265<!-- ==================================== --> 266 267<p>Arguments to diagnostics are fully typed internally, and come from a couple 268different classes: integers, types, names, and random strings. Depending on 269the class of the argument, it can be optionally formatted in different ways. 270This gives the DiagnosticClient information about what the argument means 271without requiring it to use a specific presentation (consider this MVC for 272Clang :).</p> 273 274<p>Here are the different diagnostic argument formats currently supported by 275Clang:</p> 276 277<table> 278<tr><td colspan="2"><b>"s" format</b></td></tr> 279<tr><td>Example:</td><td><tt>"requires %1 parameter%s1"</tt></td></tr> 280<tr><td>Class:</td><td>Integers</td></tr> 281<tr><td>Description:</td><td>This is a simple formatter for integers that is 282 useful when producing English diagnostics. When the integer is 1, it prints 283 as nothing. When the integer is not 1, it prints as "s". This allows some 284 simple grammatical forms to be to be handled correctly, and eliminates the 285 need to use gross things like <tt>"requires %1 parameter(s)"</tt>.</td></tr> 286 287<tr><td colspan="2"><b>"select" format</b></td></tr> 288<tr><td>Example:</td><td><tt>"must be a %select{unary|binary|unary or binary}2 289 operator"</tt></td></tr> 290<tr><td>Class:</td><td>Integers</td></tr> 291<tr><td>Description:</td><td><p>This format specifier is used to merge multiple 292 related diagnostics together into one common one, without requiring the 293 difference to be specified as an English string argument. Instead of 294 specifying the string, the diagnostic gets an integer argument and the 295 format string selects the numbered option. In this case, the "%2" value 296 must be an integer in the range [0..2]. If it is 0, it prints 'unary', if 297 it is 1 it prints 'binary' if it is 2, it prints 'unary or binary'. This 298 allows other language translations to substitute reasonable words (or entire 299 phrases) based on the semantics of the diagnostic instead of having to do 300 things textually.</p> 301 <p>The selected string does undergo formatting.</p></td></tr> 302 303<tr><td colspan="2"><b>"plural" format</b></td></tr> 304<tr><td>Example:</td><td><tt>"you have %1 %plural{1:mouse|:mice}1 connected to 305 your computer"</tt></td></tr> 306<tr><td>Class:</td><td>Integers</td></tr> 307<tr><td>Description:</td><td><p>This is a formatter for complex plural forms. 308 It is designed to handle even the requirements of languages with very 309 complex plural forms, as many Baltic languages have. The argument consists 310 of a series of expression/form pairs, separated by ':', where the first form 311 whose expression evaluates to true is the result of the modifier.</p> 312 <p>An expression can be empty, in which case it is always true. See the 313 example at the top. Otherwise, it is a series of one or more numeric 314 conditions, separated by ','. If any condition matches, the expression 315 matches. Each numeric condition can take one of three forms.</p> 316 <ul> 317 <li>number: A simple decimal number matches if the argument is the same 318 as the number. Example: <tt>"%plural{1:mouse|:mice}4"</tt></li> 319 <li>range: A range in square brackets matches if the argument is within 320 the range. Then range is inclusive on both ends. Example: 321 <tt>"%plural{0:none|1:one|[2,5]:some|:many}2"</tt></li> 322 <li>modulo: A modulo operator is followed by a number, and 323 equals sign and either a number or a range. The tests are the 324 same as for plain 325 numbers and ranges, but the argument is taken modulo the number first. 326 Example: <tt>"%plural{%100=0:even hundred|%100=[1,50]:lower half|:everything 327 else}1"</tt></li> 328 </ul> 329 <p>The parser is very unforgiving. A syntax error, even whitespace, will 330 abort, as will a failure to match the argument against any 331 expression.</p></td></tr> 332 333<tr><td colspan="2"><b>"ordinal" format</b></td></tr> 334<tr><td>Example:</td><td><tt>"ambiguity in %ordinal0 argument"</tt></td></tr> 335<tr><td>Class:</td><td>Integers</td></tr> 336<tr><td>Description:</td><td><p>This is a formatter which represents the 337 argument number as an ordinal: the value <tt>1</tt> becomes <tt>1st</tt>, 338 <tt>3</tt> becomes <tt>3rd</tt>, and so on. Values less than <tt>1</tt> 339 are not supported.</p> 340 <p>This formatter is currently hard-coded to use English ordinals.</p></td></tr> 341 342<tr><td colspan="2"><b>"objcclass" format</b></td></tr> 343<tr><td>Example:</td><td><tt>"method %objcclass0 not found"</tt></td></tr> 344<tr><td>Class:</td><td>DeclarationName</td></tr> 345<tr><td>Description:</td><td><p>This is a simple formatter that indicates the 346 DeclarationName corresponds to an Objective-C class method selector. As 347 such, it prints the selector with a leading '+'.</p></td></tr> 348 349<tr><td colspan="2"><b>"objcinstance" format</b></td></tr> 350<tr><td>Example:</td><td><tt>"method %objcinstance0 not found"</tt></td></tr> 351<tr><td>Class:</td><td>DeclarationName</td></tr> 352<tr><td>Description:</td><td><p>This is a simple formatter that indicates the 353 DeclarationName corresponds to an Objective-C instance method selector. As 354 such, it prints the selector with a leading '-'.</p></td></tr> 355 356<tr><td colspan="2"><b>"q" format</b></td></tr> 357<tr><td>Example:</td><td><tt>"candidate found by name lookup is %q0"</tt></td></tr> 358<tr><td>Class:</td><td>NamedDecl*</td></tr> 359<tr><td>Description</td><td><p>This formatter indicates that the fully-qualified name of the declaration should be printed, e.g., "std::vector" rather than "vector".</p></td></tr> 360 361<tr><td colspan="2"><b>"diff" format</b></td></tr> 362<tr><td>Example:</td><td><tt>"no known conversion %diff{from | to | }1,2"</tt></td></tr> 363<tr><td>Class:</td><td>QualType</td></tr> 364<tr><td>Description</td><td><p>This formatter takes two QualTypes and attempts to print a template difference between the two. If tree printing is off, the text inside the braces before the pipe is printed, with the formatted text replacing the $. If tree printing is on, the text after the pipe is printed and a type tree is printed after the diagnostic message. 365</p></td></tr> 366 367</table> 368 369<p>It is really easy to add format specifiers to the Clang diagnostics system, 370but they should be discussed before they are added. If you are creating a lot 371of repetitive diagnostics and/or have an idea for a useful formatter, please 372bring it up on the cfe-dev mailing list.</p> 373 374<!-- ===================================================== --> 375<h4 id="producingdiag">Producing the Diagnostic</h4> 376<!-- ===================================================== --> 377 378<p>Now that you've created the diagnostic in the DiagnosticKinds.td file, you 379need to write the code that detects the condition in question and emits the 380new diagnostic. Various components of Clang (e.g. the preprocessor, Sema, 381etc) provide a helper function named "Diag". It creates a diagnostic and 382accepts the arguments, ranges, and other information that goes along with 383it.</p> 384 385<p>For example, the binary expression error comes from code like this:</p> 386 387<pre> 388 if (various things that are bad) 389 Diag(Loc, diag::err_typecheck_invalid_operands) 390 << lex->getType() << rex->getType() 391 << lex->getSourceRange() << rex->getSourceRange(); 392</pre> 393 394<p>This shows that use of the Diag method: they take a location (a <a 395href="#SourceLocation">SourceLocation</a> object) and a diagnostic enum value 396(which matches the name from DiagnosticKinds.td). If the diagnostic takes 397arguments, they are specified with the << operator: the first argument 398becomes %0, the second becomes %1, etc. The diagnostic interface allows you to 399specify arguments of many different types, including <tt>int</tt> and 400<tt>unsigned</tt> for integer arguments, <tt>const char*</tt> and 401<tt>std::string</tt> for string arguments, <tt>DeclarationName</tt> and 402<tt>const IdentifierInfo*</tt> for names, <tt>QualType</tt> for types, etc. 403SourceRanges are also specified with the << operator, but do not have a 404specific ordering requirement.</p> 405 406<p>As you can see, adding and producing a diagnostic is pretty straightforward. 407The hard part is deciding exactly what you need to say to help the user, picking 408a suitable wording, and providing the information needed to format it correctly. 409The good news is that the call site that issues a diagnostic should be 410completely independent of how the diagnostic is formatted and in what language 411it is rendered. 412</p> 413 414<!-- ==================================================== --> 415<h4 id="fix-it-hints">Fix-It Hints</h4> 416<!-- ==================================================== --> 417 418<p>In some cases, the front end emits diagnostics when it is clear 419that some small change to the source code would fix the problem. For 420example, a missing semicolon at the end of a statement or a use of 421deprecated syntax that is easily rewritten into a more modern form. 422Clang tries very hard to emit the diagnostic and recover gracefully 423in these and other cases.</p> 424 425<p>However, for these cases where the fix is obvious, the diagnostic 426can be annotated with a hint (referred to as a "fix-it hint") that 427describes how to change the code referenced by the diagnostic to fix 428the problem. For example, it might add the missing semicolon at the 429end of the statement or rewrite the use of a deprecated construct 430into something more palatable. Here is one such example from the C++ 431front end, where we warn about the right-shift operator changing 432meaning from C++98 to C++11:</p> 433 434<pre> 435test.cpp:3:7: warning: use of right-shift operator ('>>') in template argument will require parentheses in C++11 436A<100 >> 2> *a; 437 ^ 438 ( ) 439</pre> 440 441<p>Here, the fix-it hint is suggesting that parentheses be added, 442and showing exactly where those parentheses would be inserted into the 443source code. The fix-it hints themselves describe what changes to make 444to the source code in an abstract manner, which the text diagnostic 445printer renders as a line of "insertions" below the caret line. <a 446href="#DiagnosticClient">Other diagnostic clients</a> might choose 447to render the code differently (e.g., as markup inline) or even give 448the user the ability to automatically fix the problem.</p> 449 450<p>Fix-it hints on errors and warnings need to obey these rules:</p> 451 452<ul> 453<li>Since they are automatically applied if <code>-Xclang -fixit</code> 454is passed to the driver, they should only be used when it's very likely they 455match the user's intent.</li> 456<li>Clang must recover from errors as if the fix-it had been applied.</li> 457</ul> 458 459<p>If a fix-it can't obey these rules, put the fix-it on a note. Fix-its on 460notes are not applied automatically.</p> 461 462<p>All fix-it hints are described by the <code>FixItHint</code> class, 463instances of which should be attached to the diagnostic using the 464<< operator in the same way that highlighted source ranges and 465arguments are passed to the diagnostic. Fix-it hints can be created 466with one of three constructors:</p> 467 468<dl> 469 <dt><code>FixItHint::CreateInsertion(Loc, Code)</code></dt> 470 <dd>Specifies that the given <code>Code</code> (a string) should be inserted 471 before the source location <code>Loc</code>.</dd> 472 473 <dt><code>FixItHint::CreateRemoval(Range)</code></dt> 474 <dd>Specifies that the code in the given source <code>Range</code> 475 should be removed.</dd> 476 477 <dt><code>FixItHint::CreateReplacement(Range, Code)</code></dt> 478 <dd>Specifies that the code in the given source <code>Range</code> 479 should be removed, and replaced with the given <code>Code</code> string.</dd> 480</dl> 481 482<!-- ============================================================= --> 483<h4><a name="DiagnosticClient">The DiagnosticClient Interface</a></h4> 484<!-- ============================================================= --> 485 486<p>Once code generates a diagnostic with all of the arguments and the rest of 487the relevant information, Clang needs to know what to do with it. As previously 488mentioned, the diagnostic machinery goes through some filtering to map a 489severity onto a diagnostic level, then (assuming the diagnostic is not mapped to 490"<tt>Ignore</tt>") it invokes an object that implements the DiagnosticClient 491interface with the information.</p> 492 493<p>It is possible to implement this interface in many different ways. For 494example, the normal Clang DiagnosticClient (named 'TextDiagnosticPrinter') turns 495the arguments into strings (according to the various formatting rules), prints 496out the file/line/column information and the string, then prints out the line of 497code, the source ranges, and the caret. However, this behavior isn't required. 498</p> 499 500<p>Another implementation of the DiagnosticClient interface is the 501'TextDiagnosticBuffer' class, which is used when Clang is in -verify mode. 502Instead of formatting and printing out the diagnostics, this implementation just 503captures and remembers the diagnostics as they fly by. Then -verify compares 504the list of produced diagnostics to the list of expected ones. If they disagree, 505it prints out its own output. 506</p> 507 508<p>There are many other possible implementations of this interface, and this is 509why we prefer diagnostics to pass down rich structured information in arguments. 510For example, an HTML output might want declaration names be linkified to where 511they come from in the source. Another example is that a GUI might let you click 512on typedefs to expand them. This application would want to pass significantly 513more information about types through to the GUI than a simple flat string. The 514interface allows this to happen.</p> 515 516<!-- ====================================================== --> 517<h4><a name="translation">Adding Translations to Clang</a></h4> 518<!-- ====================================================== --> 519 520<p>Not possible yet! Diagnostic strings should be written in UTF-8, the client 521can translate to the relevant code page if needed. Each translation completely 522replaces the format string for the diagnostic.</p> 523 524 525<!-- ======================================================================= --> 526<h3 id="SourceLocation">The SourceLocation and SourceManager classes</h3> 527<!-- ======================================================================= --> 528 529<p>Strangely enough, the SourceLocation class represents a location within the 530source code of the program. Important design points include:</p> 531 532<ol> 533<li>sizeof(SourceLocation) must be extremely small, as these are embedded into 534 many AST nodes and are passed around often. Currently it is 32 bits.</li> 535<li>SourceLocation must be a simple value object that can be efficiently 536 copied.</li> 537<li>We should be able to represent a source location for any byte of any input 538 file. This includes in the middle of tokens, in whitespace, in trigraphs, 539 etc.</li> 540<li>A SourceLocation must encode the current #include stack that was active when 541 the location was processed. For example, if the location corresponds to a 542 token, it should contain the set of #includes active when the token was 543 lexed. This allows us to print the #include stack for a diagnostic.</li> 544<li>SourceLocation must be able to describe macro expansions, capturing both 545 the ultimate instantiation point and the source of the original character 546 data.</li> 547</ol> 548 549<p>In practice, the SourceLocation works together with the SourceManager class 550to encode two pieces of information about a location: its spelling location 551and its instantiation location. For most tokens, these will be the same. 552However, for a macro expansion (or tokens that came from a _Pragma directive) 553these will describe the location of the characters corresponding to the token 554and the location where the token was used (i.e. the macro instantiation point 555or the location of the _Pragma itself).</p> 556 557<p>The Clang front-end inherently depends on the location of a token being 558tracked correctly. If it is ever incorrect, the front-end may get confused and 559die. The reason for this is that the notion of the 'spelling' of a Token in 560Clang depends on being able to find the original input characters for the token. 561This concept maps directly to the "spelling location" for the token.</p> 562 563 564<!-- ======================================================================= --> 565<h3 id="SourceRange">SourceRange and CharSourceRange</h3> 566<!-- ======================================================================= --> 567<!-- mostly taken from 568 http://lists.cs.uiuc.edu/pipermail/cfe-dev/2010-August/010595.html --> 569 570<p>Clang represents most source ranges by [first, last], where first and last 571each point to the beginning of their respective tokens. For example 572consider the SourceRange of the following statement:</p> 573<pre> 574x = foo + bar; 575^first ^last 576</pre> 577 578<p>To map from this representation to a character-based 579representation, the 'last' location needs to be adjusted to point to 580(or past) the end of that token with either 581<code>Lexer::MeasureTokenLength()</code> or 582<code>Lexer::getLocForEndOfToken()</code>. For the rare cases 583where character-level source ranges information is needed we use 584the <code>CharSourceRange</code> class.</p> 585 586 587<!-- ======================================================================= --> 588<h2 id="libdriver">The Driver Library</h2> 589<!-- ======================================================================= --> 590 591<p>The clang Driver and library are documented <a 592href="DriverInternals.html">here</a>.<p> 593 594<!-- ======================================================================= --> 595<h2 id="pch">Precompiled Headers</h2> 596<!-- ======================================================================= --> 597 598<p>Clang supports two implementations of precompiled headers. The 599 default implementation, precompiled headers (<a 600 href="PCHInternals.html">PCH</a>) uses a serialized representation 601 of Clang's internal data structures, encoded with the <a 602 href="http://llvm.org/docs/BitCodeFormat.html">LLVM bitstream 603 format</a>. Pretokenized headers (<a 604 href="PTHInternals.html">PTH</a>), on the other hand, contain a 605 serialized representation of the tokens encountered when 606 preprocessing a header (and anything that header includes).</p> 607 608 609<!-- ======================================================================= --> 610<h2 id="libfrontend">The Frontend Library</h2> 611<!-- ======================================================================= --> 612 613<p>The Frontend library contains functionality useful for building 614tools on top of the clang libraries, for example several methods for 615outputting diagnostics.</p> 616 617<!-- ======================================================================= --> 618<h2 id="liblex">The Lexer and Preprocessor Library</h2> 619<!-- ======================================================================= --> 620 621<p>The Lexer library contains several tightly-connected classes that are involved 622with the nasty process of lexing and preprocessing C source code. The main 623interface to this library for outside clients is the large <a 624href="#Preprocessor">Preprocessor</a> class. 625It contains the various pieces of state that are required to coherently read 626tokens out of a translation unit.</p> 627 628<p>The core interface to the Preprocessor object (once it is set up) is the 629Preprocessor::Lex method, which returns the next <a href="#Token">Token</a> from 630the preprocessor stream. There are two types of token providers that the 631preprocessor is capable of reading from: a buffer lexer (provided by the <a 632href="#Lexer">Lexer</a> class) and a buffered token stream (provided by the <a 633href="#TokenLexer">TokenLexer</a> class). 634 635 636<!-- ======================================================================= --> 637<h3 id="Token">The Token class</h3> 638<!-- ======================================================================= --> 639 640<p>The Token class is used to represent a single lexed token. Tokens are 641intended to be used by the lexer/preprocess and parser libraries, but are not 642intended to live beyond them (for example, they should not live in the ASTs).<p> 643 644<p>Tokens most often live on the stack (or some other location that is efficient 645to access) as the parser is running, but occasionally do get buffered up. For 646example, macro definitions are stored as a series of tokens, and the C++ 647front-end periodically needs to buffer tokens up for tentative parsing and 648various pieces of look-ahead. As such, the size of a Token matter. On a 32-bit 649system, sizeof(Token) is currently 16 bytes.</p> 650 651<p>Tokens occur in two forms: "<a href="#AnnotationToken">Annotation 652Tokens</a>" and normal tokens. Normal tokens are those returned by the lexer, 653annotation tokens represent semantic information and are produced by the parser, 654replacing normal tokens in the token stream. Normal tokens contain the 655following information:</p> 656 657<ul> 658<li><b>A SourceLocation</b> - This indicates the location of the start of the 659token.</li> 660 661<li><b>A length</b> - This stores the length of the token as stored in the 662SourceBuffer. For tokens that include them, this length includes trigraphs and 663escaped newlines which are ignored by later phases of the compiler. By pointing 664into the original source buffer, it is always possible to get the original 665spelling of a token completely accurately.</li> 666 667<li><b>IdentifierInfo</b> - If a token takes the form of an identifier, and if 668identifier lookup was enabled when the token was lexed (e.g. the lexer was not 669reading in 'raw' mode) this contains a pointer to the unique hash value for the 670identifier. Because the lookup happens before keyword identification, this 671field is set even for language keywords like 'for'.</li> 672 673<li><b>TokenKind</b> - This indicates the kind of token as classified by the 674lexer. This includes things like <tt>tok::starequal</tt> (for the "*=" 675operator), <tt>tok::ampamp</tt> for the "&&" token, and keyword values 676(e.g. <tt>tok::kw_for</tt>) for identifiers that correspond to keywords. Note 677that some tokens can be spelled multiple ways. For example, C++ supports 678"operator keywords", where things like "and" are treated exactly like the 679"&&" operator. In these cases, the kind value is set to 680<tt>tok::ampamp</tt>, which is good for the parser, which doesn't have to 681consider both forms. For something that cares about which form is used (e.g. 682the preprocessor 'stringize' operator) the spelling indicates the original 683form.</li> 684 685<li><b>Flags</b> - There are currently four flags tracked by the 686lexer/preprocessor system on a per-token basis: 687 688 <ol> 689 <li><b>StartOfLine</b> - This was the first token that occurred on its input 690 source line.</li> 691 <li><b>LeadingSpace</b> - There was a space character either immediately 692 before the token or transitively before the token as it was expanded 693 through a macro. The definition of this flag is very closely defined by 694 the stringizing requirements of the preprocessor.</li> 695 <li><b>DisableExpand</b> - This flag is used internally to the preprocessor to 696 represent identifier tokens which have macro expansion disabled. This 697 prevents them from being considered as candidates for macro expansion ever 698 in the future.</li> 699 <li><b>NeedsCleaning</b> - This flag is set if the original spelling for the 700 token includes a trigraph or escaped newline. Since this is uncommon, 701 many pieces of code can fast-path on tokens that did not need cleaning. 702 </ol> 703</li> 704</ul> 705 706<p>One interesting (and somewhat unusual) aspect of normal tokens is that they 707don't contain any semantic information about the lexed value. For example, if 708the token was a pp-number token, we do not represent the value of the number 709that was lexed (this is left for later pieces of code to decide). Additionally, 710the lexer library has no notion of typedef names vs variable names: both are 711returned as identifiers, and the parser is left to decide whether a specific 712identifier is a typedef or a variable (tracking this requires scope information 713among other things). The parser can do this translation by replacing tokens 714returned by the preprocessor with "Annotation Tokens".</p> 715 716<!-- ======================================================================= --> 717<h3 id="AnnotationToken">Annotation Tokens</h3> 718<!-- ======================================================================= --> 719 720<p>Annotation Tokens are tokens that are synthesized by the parser and injected 721into the preprocessor's token stream (replacing existing tokens) to record 722semantic information found by the parser. For example, if "foo" is found to be 723a typedef, the "foo" <tt>tok::identifier</tt> token is replaced with an 724<tt>tok::annot_typename</tt>. This is useful for a couple of reasons: 1) this 725makes it easy to handle qualified type names (e.g. "foo::bar::baz<42>::t") 726in C++ as a single "token" in the parser. 2) if the parser backtracks, the 727reparse does not need to redo semantic analysis to determine whether a token 728sequence is a variable, type, template, etc.</p> 729 730<p>Annotation Tokens are created by the parser and reinjected into the parser's 731token stream (when backtracking is enabled). Because they can only exist in 732tokens that the preprocessor-proper is done with, it doesn't need to keep around 733flags like "start of line" that the preprocessor uses to do its job. 734Additionally, an annotation token may "cover" a sequence of preprocessor tokens 735(e.g. <tt>a::b::c</tt> is five preprocessor tokens). As such, the valid fields 736of an annotation token are different than the fields for a normal token (but 737they are multiplexed into the normal Token fields):</p> 738 739<ul> 740<li><b>SourceLocation "Location"</b> - The SourceLocation for the annotation 741token indicates the first token replaced by the annotation token. In the example 742above, it would be the location of the "a" identifier.</li> 743 744<li><b>SourceLocation "AnnotationEndLoc"</b> - This holds the location of the 745last token replaced with the annotation token. In the example above, it would 746be the location of the "c" identifier.</li> 747 748<li><b>void* "AnnotationValue"</b> - This contains an opaque object 749that the parser gets from Sema. The parser merely preserves the 750information for Sema to later interpret based on the annotation token 751kind.</li> 752 753<li><b>TokenKind "Kind"</b> - This indicates the kind of Annotation token this 754is. See below for the different valid kinds.</li> 755</ul> 756 757<p>Annotation tokens currently come in three kinds:</p> 758 759<ol> 760<li><b>tok::annot_typename</b>: This annotation token represents a 761resolved typename token that is potentially qualified. The 762AnnotationValue field contains the <tt>QualType</tt> returned by 763Sema::getTypeName(), possibly with source location information 764attached.</li> 765 766<li><b>tok::annot_cxxscope</b>: This annotation token represents a C++ 767scope specifier, such as "A::B::". This corresponds to the grammar 768productions "::" and ":: [opt] nested-name-specifier". The 769AnnotationValue pointer is a <tt>NestedNameSpecifier*</tt> returned by 770the Sema::ActOnCXXGlobalScopeSpecifier and 771Sema::ActOnCXXNestedNameSpecifier callbacks.</li> 772 773<li><b>tok::annot_template_id</b>: This annotation token represents a 774C++ template-id such as "foo<int, 4>", where "foo" is the name 775of a template. The AnnotationValue pointer is a pointer to a malloc'd 776TemplateIdAnnotation object. Depending on the context, a parsed 777template-id that names a type might become a typename annotation token 778(if all we care about is the named type, e.g., because it occurs in a 779type specifier) or might remain a template-id token (if we want to 780retain more source location information or produce a new type, e.g., 781in a declaration of a class template specialization). template-id 782annotation tokens that refer to a type can be "upgraded" to typename 783annotation tokens by the parser.</li> 784 785</ol> 786 787<p>As mentioned above, annotation tokens are not returned by the preprocessor, 788they are formed on demand by the parser. This means that the parser has to be 789aware of cases where an annotation could occur and form it where appropriate. 790This is somewhat similar to how the parser handles Translation Phase 6 of C99: 791String Concatenation (see C99 5.1.1.2). In the case of string concatenation, 792the preprocessor just returns distinct tok::string_literal and 793tok::wide_string_literal tokens and the parser eats a sequence of them wherever 794the grammar indicates that a string literal can occur.</p> 795 796<p>In order to do this, whenever the parser expects a tok::identifier or 797tok::coloncolon, it should call the TryAnnotateTypeOrScopeToken or 798TryAnnotateCXXScopeToken methods to form the annotation token. These methods 799will maximally form the specified annotation tokens and replace the current 800token with them, if applicable. If the current tokens is not valid for an 801annotation token, it will remain an identifier or :: token.</p> 802 803 804 805<!-- ======================================================================= --> 806<h3 id="Lexer">The Lexer class</h3> 807<!-- ======================================================================= --> 808 809<p>The Lexer class provides the mechanics of lexing tokens out of a source 810buffer and deciding what they mean. The Lexer is complicated by the fact that 811it operates on raw buffers that have not had spelling eliminated (this is a 812necessity to get decent performance), but this is countered with careful coding 813as well as standard performance techniques (for example, the comment handling 814code is vectorized on X86 and PowerPC hosts).</p> 815 816<p>The lexer has a couple of interesting modal features:</p> 817 818<ul> 819<li>The lexer can operate in 'raw' mode. This mode has several features that 820 make it possible to quickly lex the file (e.g. it stops identifier lookup, 821 doesn't specially handle preprocessor tokens, handles EOF differently, etc). 822 This mode is used for lexing within an "<tt>#if 0</tt>" block, for 823 example.</li> 824<li>The lexer can capture and return comments as tokens. This is required to 825 support the -C preprocessor mode, which passes comments through, and is 826 used by the diagnostic checker to identifier expect-error annotations.</li> 827<li>The lexer can be in ParsingFilename mode, which happens when preprocessing 828 after reading a #include directive. This mode changes the parsing of '<' 829 to return an "angled string" instead of a bunch of tokens for each thing 830 within the filename.</li> 831<li>When parsing a preprocessor directive (after "<tt>#</tt>") the 832 ParsingPreprocessorDirective mode is entered. This changes the parser to 833 return EOD at a newline.</li> 834<li>The Lexer uses a LangOptions object to know whether trigraphs are enabled, 835 whether C++ or ObjC keywords are recognized, etc.</li> 836</ul> 837 838<p>In addition to these modes, the lexer keeps track of a couple of other 839 features that are local to a lexed buffer, which change as the buffer is 840 lexed:</p> 841 842<ul> 843<li>The Lexer uses BufferPtr to keep track of the current character being 844 lexed.</li> 845<li>The Lexer uses IsAtStartOfLine to keep track of whether the next lexed token 846 will start with its "start of line" bit set.</li> 847<li>The Lexer keeps track of the current #if directives that are active (which 848 can be nested).</li> 849<li>The Lexer keeps track of an <a href="#MultipleIncludeOpt"> 850 MultipleIncludeOpt</a> object, which is used to 851 detect whether the buffer uses the standard "<tt>#ifndef XX</tt> / 852 <tt>#define XX</tt>" idiom to prevent multiple inclusion. If a buffer does, 853 subsequent includes can be ignored if the XX macro is defined.</li> 854</ul> 855 856<!-- ======================================================================= --> 857<h3 id="TokenLexer">The TokenLexer class</h3> 858<!-- ======================================================================= --> 859 860<p>The TokenLexer class is a token provider that returns tokens from a list 861of tokens that came from somewhere else. It typically used for two things: 1) 862returning tokens from a macro definition as it is being expanded 2) returning 863tokens from an arbitrary buffer of tokens. The later use is used by _Pragma and 864will most likely be used to handle unbounded look-ahead for the C++ parser.</p> 865 866<!-- ======================================================================= --> 867<h3 id="MultipleIncludeOpt">The MultipleIncludeOpt class</h3> 868<!-- ======================================================================= --> 869 870<p>The MultipleIncludeOpt class implements a really simple little state machine 871that is used to detect the standard "<tt>#ifndef XX</tt> / <tt>#define XX</tt>" 872idiom that people typically use to prevent multiple inclusion of headers. If a 873buffer uses this idiom and is subsequently #include'd, the preprocessor can 874simply check to see whether the guarding condition is defined or not. If so, 875the preprocessor can completely ignore the include of the header.</p> 876 877 878 879<!-- ======================================================================= --> 880<h2 id="libparse">The Parser Library</h2> 881<!-- ======================================================================= --> 882 883<!-- ======================================================================= --> 884<h2 id="libast">The AST Library</h2> 885<!-- ======================================================================= --> 886 887<!-- ======================================================================= --> 888<h3 id="Type">The Type class and its subclasses</h3> 889<!-- ======================================================================= --> 890 891<p>The Type class (and its subclasses) are an important part of the AST. Types 892are accessed through the ASTContext class, which implicitly creates and uniques 893them as they are needed. Types have a couple of non-obvious features: 1) they 894do not capture type qualifiers like const or volatile (See 895<a href="#QualType">QualType</a>), and 2) they implicitly capture typedef 896information. Once created, types are immutable (unlike decls).</p> 897 898<p>Typedefs in C make semantic analysis a bit more complex than it would 899be without them. The issue is that we want to capture typedef information 900and represent it in the AST perfectly, but the semantics of operations need to 901"see through" typedefs. For example, consider this code:</p> 902 903<code> 904void func() {<br> 905 typedef int foo;<br> 906 foo X, *Y;<br> 907 typedef foo* bar;<br> 908 bar Z;<br> 909 *X; <i>// error</i><br> 910 **Y; <i>// error</i><br> 911 **Z; <i>// error</i><br> 912}<br> 913</code> 914 915<p>The code above is illegal, and thus we expect there to be diagnostics emitted 916on the annotated lines. In this example, we expect to get:</p> 917 918<pre> 919<b>test.c:6:1: error: indirection requires pointer operand ('foo' invalid)</b> 920*X; // error 921<span style="color:blue">^~</span> 922<b>test.c:7:1: error: indirection requires pointer operand ('foo' invalid)</b> 923**Y; // error 924<span style="color:blue">^~~</span> 925<b>test.c:8:1: error: indirection requires pointer operand ('foo' invalid)</b> 926**Z; // error 927<span style="color:blue">^~~</span> 928</pre> 929 930<p>While this example is somewhat silly, it illustrates the point: we want to 931retain typedef information where possible, so that we can emit errors about 932"<tt>std::string</tt>" instead of "<tt>std::basic_string<char, std:...</tt>". 933Doing this requires properly keeping typedef information (for example, the type 934of "X" is "foo", not "int"), and requires properly propagating it through the 935various operators (for example, the type of *Y is "foo", not "int"). In order 936to retain this information, the type of these expressions is an instance of the 937TypedefType class, which indicates that the type of these expressions is a 938typedef for foo. 939</p> 940 941<p>Representing types like this is great for diagnostics, because the 942user-specified type is always immediately available. There are two problems 943with this: first, various semantic checks need to make judgements about the 944<em>actual structure</em> of a type, ignoring typedefs. Second, we need an 945efficient way to query whether two types are structurally identical to each 946other, ignoring typedefs. The solution to both of these problems is the idea of 947canonical types.</p> 948 949<!-- =============== --> 950<h4>Canonical Types</h4> 951<!-- =============== --> 952 953<p>Every instance of the Type class contains a canonical type pointer. For 954simple types with no typedefs involved (e.g. "<tt>int</tt>", "<tt>int*</tt>", 955"<tt>int**</tt>"), the type just points to itself. For types that have a 956typedef somewhere in their structure (e.g. "<tt>foo</tt>", "<tt>foo*</tt>", 957"<tt>foo**</tt>", "<tt>bar</tt>"), the canonical type pointer points to their 958structurally equivalent type without any typedefs (e.g. "<tt>int</tt>", 959"<tt>int*</tt>", "<tt>int**</tt>", and "<tt>int*</tt>" respectively).</p> 960 961<p>This design provides a constant time operation (dereferencing the canonical 962type pointer) that gives us access to the structure of types. For example, 963we can trivially tell that "bar" and "foo*" are the same type by dereferencing 964their canonical type pointers and doing a pointer comparison (they both point 965to the single "<tt>int*</tt>" type).</p> 966 967<p>Canonical types and typedef types bring up some complexities that must be 968carefully managed. Specifically, the "isa/cast/dyncast" operators generally 969shouldn't be used in code that is inspecting the AST. For example, when type 970checking the indirection operator (unary '*' on a pointer), the type checker 971must verify that the operand has a pointer type. It would not be correct to 972check that with "<tt>isa<PointerType>(SubExpr->getType())</tt>", 973because this predicate would fail if the subexpression had a typedef type.</p> 974 975<p>The solution to this problem are a set of helper methods on Type, used to 976check their properties. In this case, it would be correct to use 977"<tt>SubExpr->getType()->isPointerType()</tt>" to do the check. This 978predicate will return true if the <em>canonical type is a pointer</em>, which is 979true any time the type is structurally a pointer type. The only hard part here 980is remembering not to use the <tt>isa/cast/dyncast</tt> operations.</p> 981 982<p>The second problem we face is how to get access to the pointer type once we 983know it exists. To continue the example, the result type of the indirection 984operator is the pointee type of the subexpression. In order to determine the 985type, we need to get the instance of PointerType that best captures the typedef 986information in the program. If the type of the expression is literally a 987PointerType, we can return that, otherwise we have to dig through the 988typedefs to find the pointer type. For example, if the subexpression had type 989"<tt>foo*</tt>", we could return that type as the result. If the subexpression 990had type "<tt>bar</tt>", we want to return "<tt>foo*</tt>" (note that we do 991<em>not</em> want "<tt>int*</tt>"). In order to provide all of this, Type has 992a getAsPointerType() method that checks whether the type is structurally a 993PointerType and, if so, returns the best one. If not, it returns a null 994pointer.</p> 995 996<p>This structure is somewhat mystical, but after meditating on it, it will 997make sense to you :).</p> 998 999<!-- ======================================================================= --> 1000<h3 id="QualType">The QualType class</h3> 1001<!-- ======================================================================= --> 1002 1003<p>The QualType class is designed as a trivial value class that is 1004small, passed by-value and is efficient to query. The idea of 1005QualType is that it stores the type qualifiers (const, volatile, 1006restrict, plus some extended qualifiers required by language 1007extensions) separately from the types themselves. QualType is 1008conceptually a pair of "Type*" and the bits for these type qualifiers.</p> 1009 1010<p>By storing the type qualifiers as bits in the conceptual pair, it is 1011extremely efficient to get the set of qualifiers on a QualType (just return the 1012field of the pair), add a type qualifier (which is a trivial constant-time 1013operation that sets a bit), and remove one or more type qualifiers (just return 1014a QualType with the bitfield set to empty).</p> 1015 1016<p>Further, because the bits are stored outside of the type itself, we do not 1017need to create duplicates of types with different sets of qualifiers (i.e. there 1018is only a single heap allocated "int" type: "const int" and "volatile const int" 1019both point to the same heap allocated "int" type). This reduces the heap size 1020used to represent bits and also means we do not have to consider qualifiers when 1021uniquing types (<a href="#Type">Type</a> does not even contain qualifiers).</p> 1022 1023<p>In practice, the two most common type qualifiers (const and 1024restrict) are stored in the low bits of the pointer to the Type 1025object, together with a flag indicating whether extended qualifiers 1026are present (which must be heap-allocated). This means that QualType 1027is exactly the same size as a pointer.</p> 1028 1029<!-- ======================================================================= --> 1030<h3 id="DeclarationName">Declaration names</h3> 1031<!-- ======================================================================= --> 1032 1033<p>The <tt>DeclarationName</tt> class represents the name of a 1034 declaration in Clang. Declarations in the C family of languages can 1035 take several different forms. Most declarations are named by 1036 simple identifiers, e.g., "<code>f</code>" and "<code>x</code>" in 1037 the function declaration <code>f(int x)</code>. In C++, declaration 1038 names can also name class constructors ("<code>Class</code>" 1039 in <code>struct Class { Class(); }</code>), class destructors 1040 ("<code>~Class</code>"), overloaded operator names ("operator+"), 1041 and conversion functions ("<code>operator void const *</code>"). In 1042 Objective-C, declaration names can refer to the names of Objective-C 1043 methods, which involve the method name and the parameters, 1044 collectively called a <i>selector</i>, e.g., 1045 "<code>setWidth:height:</code>". Since all of these kinds of 1046 entities - variables, functions, Objective-C methods, C++ 1047 constructors, destructors, and operators - are represented as 1048 subclasses of Clang's common <code>NamedDecl</code> 1049 class, <code>DeclarationName</code> is designed to efficiently 1050 represent any kind of name.</p> 1051 1052<p>Given 1053 a <code>DeclarationName</code> <code>N</code>, <code>N.getNameKind()</code> 1054 will produce a value that describes what kind of name <code>N</code> 1055 stores. There are 8 options (all of the names are inside 1056 the <code>DeclarationName</code> class)</p> 1057<dl> 1058 <dt>Identifier</dt> 1059 <dd>The name is a simple 1060 identifier. Use <code>N.getAsIdentifierInfo()</code> to retrieve the 1061 corresponding <code>IdentifierInfo*</code> pointing to the actual 1062 identifier. Note that C++ overloaded operators (e.g., 1063 "<code>operator+</code>") are represented as special kinds of 1064 identifiers. Use <code>IdentifierInfo</code>'s <code>getOverloadedOperatorID</code> 1065 function to determine whether an identifier is an overloaded 1066 operator name.</dd> 1067 1068 <dt>ObjCZeroArgSelector, ObjCOneArgSelector, 1069 ObjCMultiArgSelector</dt> 1070 <dd>The name is an Objective-C selector, which can be retrieved as a 1071 <code>Selector</code> instance 1072 via <code>N.getObjCSelector()</code>. The three possible name 1073 kinds for Objective-C reflect an optimization within 1074 the <code>DeclarationName</code> class: both zero- and 1075 one-argument selectors are stored as a 1076 masked <code>IdentifierInfo</code> pointer, and therefore require 1077 very little space, since zero- and one-argument selectors are far 1078 more common than multi-argument selectors (which use a different 1079 structure).</dd> 1080 1081 <dt>CXXConstructorName</dt> 1082 <dd>The name is a C++ constructor 1083 name. Use <code>N.getCXXNameType()</code> to retrieve 1084 the <a href="#QualType">type</a> that this constructor is meant to 1085 construct. The type is always the canonical type, since all 1086 constructors for a given type have the same name.</dd> 1087 1088 <dt>CXXDestructorName</dt> 1089 <dd>The name is a C++ destructor 1090 name. Use <code>N.getCXXNameType()</code> to retrieve 1091 the <a href="#QualType">type</a> whose destructor is being 1092 named. This type is always a canonical type.</dd> 1093 1094 <dt>CXXConversionFunctionName</dt> 1095 <dd>The name is a C++ conversion function. Conversion functions are 1096 named according to the type they convert to, e.g., "<code>operator void 1097 const *</code>". Use <code>N.getCXXNameType()</code> to retrieve 1098 the type that this conversion function converts to. This type is 1099 always a canonical type.</dd> 1100 1101 <dt>CXXOperatorName</dt> 1102 <dd>The name is a C++ overloaded operator name. Overloaded operators 1103 are named according to their spelling, e.g., 1104 "<code>operator+</code>" or "<code>operator new 1105 []</code>". Use <code>N.getCXXOverloadedOperator()</code> to 1106 retrieve the overloaded operator (a value of 1107 type <code>OverloadedOperatorKind</code>).</dd> 1108</dl> 1109 1110<p><code>DeclarationName</code>s are cheap to create, copy, and 1111 compare. They require only a single pointer's worth of storage in 1112 the common cases (identifiers, zero- 1113 and one-argument Objective-C selectors) and use dense, uniqued 1114 storage for the other kinds of 1115 names. Two <code>DeclarationName</code>s can be compared for 1116 equality (<code>==</code>, <code>!=</code>) using a simple bitwise 1117 comparison, can be ordered 1118 with <code><</code>, <code>></code>, <code><=</code>, 1119 and <code>>=</code> (which provide a lexicographical ordering for 1120 normal identifiers but an unspecified ordering for other kinds of 1121 names), and can be placed into LLVM <code>DenseMap</code>s 1122 and <code>DenseSet</code>s.</p> 1123 1124<p><code>DeclarationName</code> instances can be created in different 1125 ways depending on what kind of name the instance will store. Normal 1126 identifiers (<code>IdentifierInfo</code> pointers) and Objective-C selectors 1127 (<code>Selector</code>) can be implicitly converted 1128 to <code>DeclarationName</code>s. Names for C++ constructors, 1129 destructors, conversion functions, and overloaded operators can be retrieved from 1130 the <code>DeclarationNameTable</code>, an instance of which is 1131 available as <code>ASTContext::DeclarationNames</code>. The member 1132 functions <code>getCXXConstructorName</code>, <code>getCXXDestructorName</code>, 1133 <code>getCXXConversionFunctionName</code>, and <code>getCXXOperatorName</code>, respectively, 1134 return <code>DeclarationName</code> instances for the four kinds of 1135 C++ special function names.</p> 1136 1137<!-- ======================================================================= --> 1138<h3 id="DeclContext">Declaration contexts</h3> 1139<!-- ======================================================================= --> 1140<p>Every declaration in a program exists within some <i>declaration 1141 context</i>, such as a translation unit, namespace, class, or 1142 function. Declaration contexts in Clang are represented by 1143 the <code>DeclContext</code> class, from which the various 1144 declaration-context AST nodes 1145 (<code>TranslationUnitDecl</code>, <code>NamespaceDecl</code>, <code>RecordDecl</code>, <code>FunctionDecl</code>, 1146 etc.) will derive. The <code>DeclContext</code> class provides 1147 several facilities common to each declaration context:</p> 1148<dl> 1149 <dt>Source-centric vs. Semantics-centric View of Declarations</dt> 1150 <dd><code>DeclContext</code> provides two views of the declarations 1151 stored within a declaration context. The source-centric view 1152 accurately represents the program source code as written, including 1153 multiple declarations of entities where present (see the 1154 section <a href="#Redeclarations">Redeclarations and 1155 Overloads</a>), while the semantics-centric view represents the 1156 program semantics. The two views are kept synchronized by semantic 1157 analysis while the ASTs are being constructed.</dd> 1158 1159 <dt>Storage of declarations within that context</dt> 1160 <dd>Every declaration context can contain some number of 1161 declarations. For example, a C++ class (represented 1162 by <code>RecordDecl</code>) contains various member functions, 1163 fields, nested types, and so on. All of these declarations will be 1164 stored within the <code>DeclContext</code>, and one can iterate 1165 over the declarations via 1166 [<code>DeclContext::decls_begin()</code>, 1167 <code>DeclContext::decls_end()</code>). This mechanism provides 1168 the source-centric view of declarations in the context.</dd> 1169 1170 <dt>Lookup of declarations within that context</dt> 1171 <dd>The <code>DeclContext</code> structure provides efficient name 1172 lookup for names within that declaration context. For example, 1173 if <code>N</code> is a namespace we can look for the 1174 name <code>N::f</code> 1175 using <code>DeclContext::lookup</code>. The lookup itself is 1176 based on a lazily-constructed array (for declaration contexts 1177 with a small number of declarations) or hash table (for 1178 declaration contexts with more declarations). The lookup 1179 operation provides the semantics-centric view of the declarations 1180 in the context.</dd> 1181 1182 <dt>Ownership of declarations</dt> 1183 <dd>The <code>DeclContext</code> owns all of the declarations that 1184 were declared within its declaration context, and is responsible 1185 for the management of their memory as well as their 1186 (de-)serialization.</dd> 1187</dl> 1188 1189<p>All declarations are stored within a declaration context, and one 1190 can query 1191 information about the context in which each declaration lives. One 1192 can retrieve the <code>DeclContext</code> that contains a 1193 particular <code>Decl</code> 1194 using <code>Decl::getDeclContext</code>. However, see the 1195 section <a href="#LexicalAndSemanticContexts">Lexical and Semantic 1196 Contexts</a> for more information about how to interpret this 1197 context information.</p> 1198 1199<h4 id="Redeclarations">Redeclarations and Overloads</h4> 1200<p>Within a translation unit, it is common for an entity to be 1201declared several times. For example, we might declare a function "f" 1202 and then later re-declare it as part of an inlined definition:</p> 1203 1204<pre> 1205void f(int x, int y, int z = 1); 1206 1207inline void f(int x, int y, int z) { /* ... */ } 1208</pre> 1209 1210<p>The representation of "f" differs in the source-centric and 1211 semantics-centric views of a declaration context. In the 1212 source-centric view, all redeclarations will be present, in the 1213 order they occurred in the source code, making 1214 this view suitable for clients that wish to see the structure of 1215 the source code. In the semantics-centric view, only the most recent "f" 1216 will be found by the lookup, since it effectively replaces the first 1217 declaration of "f".</p> 1218 1219<p>In the semantics-centric view, overloading of functions is 1220 represented explicitly. For example, given two declarations of a 1221 function "g" that are overloaded, e.g.,</p> 1222<pre> 1223void g(); 1224void g(int); 1225</pre> 1226<p>the <code>DeclContext::lookup</code> operation will return 1227 a <code>DeclContext::lookup_result</code> that contains a range of iterators 1228 over declarations of "g". Clients that perform semantic analysis on a 1229 program that is not concerned with the actual source code will 1230 primarily use this semantics-centric view.</p> 1231 1232<h4 id="LexicalAndSemanticContexts">Lexical and Semantic Contexts</h4> 1233<p>Each declaration has two potentially different 1234 declaration contexts: a <i>lexical</i> context, which corresponds to 1235 the source-centric view of the declaration context, and 1236 a <i>semantic</i> context, which corresponds to the 1237 semantics-centric view. The lexical context is accessible 1238 via <code>Decl::getLexicalDeclContext</code> while the 1239 semantic context is accessible 1240 via <code>Decl::getDeclContext</code>, both of which return 1241 <code>DeclContext</code> pointers. For most declarations, the two 1242 contexts are identical. For example:</p> 1243 1244<pre> 1245class X { 1246public: 1247 void f(int x); 1248}; 1249</pre> 1250 1251<p>Here, the semantic and lexical contexts of <code>X::f</code> are 1252 the <code>DeclContext</code> associated with the 1253 class <code>X</code> (itself stored as a <code>RecordDecl</code> AST 1254 node). However, we can now define <code>X::f</code> out-of-line:</p> 1255 1256<pre> 1257void X::f(int x = 17) { /* ... */ } 1258</pre> 1259 1260<p>This definition of has different lexical and semantic 1261 contexts. The lexical context corresponds to the declaration 1262 context in which the actual declaration occurred in the source 1263 code, e.g., the translation unit containing <code>X</code>. Thus, 1264 this declaration of <code>X::f</code> can be found by traversing 1265 the declarations provided by 1266 [<code>decls_begin()</code>, <code>decls_end()</code>) in the 1267 translation unit.</p> 1268 1269<p>The semantic context of <code>X::f</code> corresponds to the 1270 class <code>X</code>, since this member function is (semantically) a 1271 member of <code>X</code>. Lookup of the name <code>f</code> into 1272 the <code>DeclContext</code> associated with <code>X</code> will 1273 then return the definition of <code>X::f</code> (including 1274 information about the default argument).</p> 1275 1276<h4 id="TransparentContexts">Transparent Declaration Contexts</h4> 1277<p>In C and C++, there are several contexts in which names that are 1278 logically declared inside another declaration will actually "leak" 1279 out into the enclosing scope from the perspective of name 1280 lookup. The most obvious instance of this behavior is in 1281 enumeration types, e.g.,</p> 1282<pre> 1283enum Color { 1284 Red, 1285 Green, 1286 Blue 1287}; 1288</pre> 1289 1290<p>Here, <code>Color</code> is an enumeration, which is a declaration 1291 context that contains the 1292 enumerators <code>Red</code>, <code>Green</code>, 1293 and <code>Blue</code>. Thus, traversing the list of declarations 1294 contained in the enumeration <code>Color</code> will 1295 yield <code>Red</code>, <code>Green</code>, 1296 and <code>Blue</code>. However, outside of the scope 1297 of <code>Color</code> one can name the enumerator <code>Red</code> 1298 without qualifying the name, e.g.,</p> 1299 1300<pre> 1301Color c = Red; 1302</pre> 1303 1304<p>There are other entities in C++ that provide similar behavior. For 1305 example, linkage specifications that use curly braces:</p> 1306 1307<pre> 1308extern "C" { 1309 void f(int); 1310 void g(int); 1311} 1312// f and g are visible here 1313</pre> 1314 1315<p>For source-level accuracy, we treat the linkage specification and 1316 enumeration type as a 1317 declaration context in which its enclosed declarations ("Red", 1318 "Green", and "Blue"; "f" and "g") 1319 are declared. However, these declarations are visible outside of the 1320 scope of the declaration context.</p> 1321 1322<p>These language features (and several others, described below) have 1323 roughly the same set of 1324 requirements: declarations are declared within a particular lexical 1325 context, but the declarations are also found via name lookup in 1326 scopes enclosing the declaration itself. This feature is implemented 1327 via <i>transparent</i> declaration contexts 1328 (see <code>DeclContext::isTransparentContext()</code>), whose 1329 declarations are visible in the nearest enclosing non-transparent 1330 declaration context. This means that the lexical context of the 1331 declaration (e.g., an enumerator) will be the 1332 transparent <code>DeclContext</code> itself, as will the semantic 1333 context, but the declaration will be visible in every outer context 1334 up to and including the first non-transparent declaration context (since 1335 transparent declaration contexts can be nested).</p> 1336 1337<p>The transparent <code>DeclContexts</code> are:</p> 1338<ul> 1339 <li>Enumerations (but not C++11 "scoped enumerations"): 1340 <pre> 1341enum Color { 1342 Red, 1343 Green, 1344 Blue 1345}; 1346// Red, Green, and Blue are in scope 1347 </pre></li> 1348 <li>C++ linkage specifications: 1349 <pre> 1350extern "C" { 1351 void f(int); 1352 void g(int); 1353} 1354// f and g are in scope 1355 </pre></li> 1356 <li>Anonymous unions and structs: 1357 <pre> 1358struct LookupTable { 1359 bool IsVector; 1360 union { 1361 std::vector<Item> *Vector; 1362 std::set<Item> *Set; 1363 }; 1364}; 1365 1366LookupTable LT; 1367LT.Vector = 0; // Okay: finds Vector inside the unnamed union 1368 </pre> 1369 </li> 1370 <li>C++11 inline namespaces: 1371<pre> 1372namespace mylib { 1373 inline namespace debug { 1374 class X; 1375 } 1376} 1377mylib::X *xp; // okay: mylib::X refers to mylib::debug::X 1378</pre> 1379</li> 1380</ul> 1381 1382 1383<h4 id="MultiDeclContext">Multiply-Defined Declaration Contexts</h4> 1384<p>C++ namespaces have the interesting--and, so far, unique--property that 1385the namespace can be defined multiple times, and the declarations 1386provided by each namespace definition are effectively merged (from 1387the semantic point of view). For example, the following two code 1388snippets are semantically indistinguishable:</p> 1389<pre> 1390// Snippet #1: 1391namespace N { 1392 void f(); 1393} 1394namespace N { 1395 void f(int); 1396} 1397 1398// Snippet #2: 1399namespace N { 1400 void f(); 1401 void f(int); 1402} 1403</pre> 1404 1405<p>In Clang's representation, the source-centric view of declaration 1406 contexts will actually have two separate <code>NamespaceDecl</code> 1407 nodes in Snippet #1, each of which is a declaration context that 1408 contains a single declaration of "f". However, the semantics-centric 1409 view provided by name lookup into the namespace <code>N</code> for 1410 "f" will return a <code>DeclContext::lookup_result</code> that contains 1411 a range of iterators over declarations of "f".</p> 1412 1413<p><code>DeclContext</code> manages multiply-defined declaration 1414 contexts internally. The 1415 function <code>DeclContext::getPrimaryContext</code> retrieves the 1416 "primary" context for a given <code>DeclContext</code> instance, 1417 which is the <code>DeclContext</code> responsible for maintaining 1418 the lookup table used for the semantics-centric view. Given the 1419 primary context, one can follow the chain 1420 of <code>DeclContext</code> nodes that define additional 1421 declarations via <code>DeclContext::getNextContext</code>. Note that 1422 these functions are used internally within the lookup and insertion 1423 methods of the <code>DeclContext</code>, so the vast majority of 1424 clients can ignore them.</p> 1425 1426<!-- ======================================================================= --> 1427<h3 id="CFG">The <tt>CFG</tt> class</h3> 1428<!-- ======================================================================= --> 1429 1430<p>The <tt>CFG</tt> class is designed to represent a source-level 1431control-flow graph for a single statement (<tt>Stmt*</tt>). Typically 1432instances of <tt>CFG</tt> are constructed for function bodies (usually 1433an instance of <tt>CompoundStmt</tt>), but can also be instantiated to 1434represent the control-flow of any class that subclasses <tt>Stmt</tt>, 1435which includes simple expressions. Control-flow graphs are especially 1436useful for performing 1437<a href="http://en.wikipedia.org/wiki/Data_flow_analysis#Sensitivities">flow- 1438or path-sensitive</a> program analyses on a given function.</p> 1439 1440<!-- ============ --> 1441<h4>Basic Blocks</h4> 1442<!-- ============ --> 1443 1444<p>Concretely, an instance of <tt>CFG</tt> is a collection of basic 1445blocks. Each basic block is an instance of <tt>CFGBlock</tt>, which 1446simply contains an ordered sequence of <tt>Stmt*</tt> (each referring 1447to statements in the AST). The ordering of statements within a block 1448indicates unconditional flow of control from one statement to the 1449next. <a href="#ConditionalControlFlow">Conditional control-flow</a> 1450is represented using edges between basic blocks. The statements 1451within a given <tt>CFGBlock</tt> can be traversed using 1452the <tt>CFGBlock::*iterator</tt> interface.</p> 1453 1454<p> 1455A <tt>CFG</tt> object owns the instances of <tt>CFGBlock</tt> within 1456the control-flow graph it represents. Each <tt>CFGBlock</tt> within a 1457CFG is also uniquely numbered (accessible 1458via <tt>CFGBlock::getBlockID()</tt>). Currently the number is 1459based on the ordering the blocks were created, but no assumptions 1460should be made on how <tt>CFGBlock</tt>s are numbered other than their 1461numbers are unique and that they are numbered from 0..N-1 (where N is 1462the number of basic blocks in the CFG).</p> 1463 1464<!-- ===================== --> 1465<h4>Entry and Exit Blocks</h4> 1466<!-- ===================== --> 1467 1468Each instance of <tt>CFG</tt> contains two special blocks: 1469an <i>entry</i> block (accessible via <tt>CFG::getEntry()</tt>), which 1470has no incoming edges, and an <i>exit</i> block (accessible 1471via <tt>CFG::getExit()</tt>), which has no outgoing edges. Neither 1472block contains any statements, and they serve the role of providing a 1473clear entrance and exit for a body of code such as a function body. 1474The presence of these empty blocks greatly simplifies the 1475implementation of many analyses built on top of CFGs. 1476 1477<!-- ===================================================== --> 1478<h4 id ="ConditionalControlFlow">Conditional Control-Flow</h4> 1479<!-- ===================================================== --> 1480 1481<p>Conditional control-flow (such as those induced by if-statements 1482and loops) is represented as edges between <tt>CFGBlock</tt>s. 1483Because different C language constructs can induce control-flow, 1484each <tt>CFGBlock</tt> also records an extra <tt>Stmt*</tt> that 1485represents the <i>terminator</i> of the block. A terminator is simply 1486the statement that caused the control-flow, and is used to identify 1487the nature of the conditional control-flow between blocks. For 1488example, in the case of an if-statement, the terminator refers to 1489the <tt>IfStmt</tt> object in the AST that represented the given 1490branch.</p> 1491 1492<p>To illustrate, consider the following code example:</p> 1493 1494<code> 1495int foo(int x) {<br> 1496 x = x + 1;<br> 1497<br> 1498 if (x > 2) x++;<br> 1499 else {<br> 1500 x += 2;<br> 1501 x *= 2;<br> 1502 }<br> 1503<br> 1504 return x;<br> 1505} 1506</code> 1507 1508<p>After invoking the parser+semantic analyzer on this code fragment, 1509the AST of the body of <tt>foo</tt> is referenced by a 1510single <tt>Stmt*</tt>. We can then construct an instance 1511of <tt>CFG</tt> representing the control-flow graph of this function 1512body by single call to a static class method:</p> 1513 1514<code> 1515 Stmt* FooBody = ...<br> 1516 CFG* FooCFG = <b>CFG::buildCFG</b>(FooBody); 1517</code> 1518 1519<p>It is the responsibility of the caller of <tt>CFG::buildCFG</tt> 1520to <tt>delete</tt> the returned <tt>CFG*</tt> when the CFG is no 1521longer needed.</p> 1522 1523<p>Along with providing an interface to iterate over 1524its <tt>CFGBlock</tt>s, the <tt>CFG</tt> class also provides methods 1525that are useful for debugging and visualizing CFGs. For example, the 1526method 1527<tt>CFG::dump()</tt> dumps a pretty-printed version of the CFG to 1528standard error. This is especially useful when one is using a 1529debugger such as gdb. For example, here is the output 1530of <tt>FooCFG->dump()</tt>:</p> 1531 1532<code> 1533 [ B5 (ENTRY) ]<br> 1534 Predecessors (0):<br> 1535 Successors (1): B4<br> 1536<br> 1537 [ B4 ]<br> 1538 1: x = x + 1<br> 1539 2: (x > 2)<br> 1540 <b>T: if [B4.2]</b><br> 1541 Predecessors (1): B5<br> 1542 Successors (2): B3 B2<br> 1543<br> 1544 [ B3 ]<br> 1545 1: x++<br> 1546 Predecessors (1): B4<br> 1547 Successors (1): B1<br> 1548<br> 1549 [ B2 ]<br> 1550 1: x += 2<br> 1551 2: x *= 2<br> 1552 Predecessors (1): B4<br> 1553 Successors (1): B1<br> 1554<br> 1555 [ B1 ]<br> 1556 1: return x;<br> 1557 Predecessors (2): B2 B3<br> 1558 Successors (1): B0<br> 1559<br> 1560 [ B0 (EXIT) ]<br> 1561 Predecessors (1): B1<br> 1562 Successors (0): 1563</code> 1564 1565<p>For each block, the pretty-printed output displays for each block 1566the number of <i>predecessor</i> blocks (blocks that have outgoing 1567control-flow to the given block) and <i>successor</i> blocks (blocks 1568that have control-flow that have incoming control-flow from the given 1569block). We can also clearly see the special entry and exit blocks at 1570the beginning and end of the pretty-printed output. For the entry 1571block (block B5), the number of predecessor blocks is 0, while for the 1572exit block (block B0) the number of successor blocks is 0.</p> 1573 1574<p>The most interesting block here is B4, whose outgoing control-flow 1575represents the branching caused by the sole if-statement 1576in <tt>foo</tt>. Of particular interest is the second statement in 1577the block, <b><tt>(x > 2)</tt></b>, and the terminator, printed 1578as <b><tt>if [B4.2]</tt></b>. The second statement represents the 1579evaluation of the condition of the if-statement, which occurs before 1580the actual branching of control-flow. Within the <tt>CFGBlock</tt> 1581for B4, the <tt>Stmt*</tt> for the second statement refers to the 1582actual expression in the AST for <b><tt>(x > 2)</tt></b>. Thus 1583pointers to subclasses of <tt>Expr</tt> can appear in the list of 1584statements in a block, and not just subclasses of <tt>Stmt</tt> that 1585refer to proper C statements.</p> 1586 1587<p>The terminator of block B4 is a pointer to the <tt>IfStmt</tt> 1588object in the AST. The pretty-printer outputs <b><tt>if 1589[B4.2]</tt></b> because the condition expression of the if-statement 1590has an actual place in the basic block, and thus the terminator is 1591essentially 1592<i>referring</i> to the expression that is the second statement of 1593block B4 (i.e., B4.2). In this manner, conditions for control-flow 1594(which also includes conditions for loops and switch statements) are 1595hoisted into the actual basic block.</p> 1596 1597<!-- ===================== --> 1598<!-- <h4>Implicit Control-Flow</h4> --> 1599<!-- ===================== --> 1600 1601<!-- 1602<p>A key design principle of the <tt>CFG</tt> class was to not require 1603any transformations to the AST in order to represent control-flow. 1604Thus the <tt>CFG</tt> does not perform any "lowering" of the 1605statements in an AST: loops are not transformed into guarded gotos, 1606short-circuit operations are not converted to a set of if-statements, 1607and so on.</p> 1608--> 1609 1610 1611<!-- ======================================================================= --> 1612<h3 id="Constants">Constant Folding in the Clang AST</h3> 1613<!-- ======================================================================= --> 1614 1615<p>There are several places where constants and constant folding matter a lot to 1616the Clang front-end. First, in general, we prefer the AST to retain the source 1617code as close to how the user wrote it as possible. This means that if they 1618wrote "5+4", we want to keep the addition and two constants in the AST, we don't 1619want to fold to "9". This means that constant folding in various ways turns 1620into a tree walk that needs to handle the various cases.</p> 1621 1622<p>However, there are places in both C and C++ that require constants to be 1623folded. For example, the C standard defines what an "integer constant 1624expression" (i-c-e) is with very precise and specific requirements. The 1625language then requires i-c-e's in a lot of places (for example, the size of a 1626bitfield, the value for a case statement, etc). For these, we have to be able 1627to constant fold the constants, to do semantic checks (e.g. verify bitfield size 1628is non-negative and that case statements aren't duplicated). We aim for Clang 1629to be very pedantic about this, diagnosing cases when the code does not use an 1630i-c-e where one is required, but accepting the code unless running with 1631<tt>-pedantic-errors</tt>.</p> 1632 1633<p>Things get a little bit more tricky when it comes to compatibility with 1634real-world source code. Specifically, GCC has historically accepted a huge 1635superset of expressions as i-c-e's, and a lot of real world code depends on this 1636unfortuate accident of history (including, e.g., the glibc system headers). GCC 1637accepts anything its "fold" optimizer is capable of reducing to an integer 1638constant, which means that the definition of what it accepts changes as its 1639optimizer does. One example is that GCC accepts things like "case X-X:" even 1640when X is a variable, because it can fold this to 0.</p> 1641 1642<p>Another issue are how constants interact with the extensions we support, such 1643as __builtin_constant_p, __builtin_inf, __extension__ and many others. C99 1644obviously does not specify the semantics of any of these extensions, and the 1645definition of i-c-e does not include them. However, these extensions are often 1646used in real code, and we have to have a way to reason about them.</p> 1647 1648<p>Finally, this is not just a problem for semantic analysis. The code 1649generator and other clients have to be able to fold constants (e.g. to 1650initialize global variables) and has to handle a superset of what C99 allows. 1651Further, these clients can benefit from extended information. For example, we 1652know that "foo()||1" always evaluates to true, but we can't replace the 1653expression with true because it has side effects.</p> 1654 1655<!-- ======================= --> 1656<h4>Implementation Approach</h4> 1657<!-- ======================= --> 1658 1659<p>After trying several different approaches, we've finally converged on a 1660design (Note, at the time of this writing, not all of this has been implemented, 1661consider this a design goal!). Our basic approach is to define a single 1662recursive method evaluation method (<tt>Expr::Evaluate</tt>), which is 1663implemented in <tt>AST/ExprConstant.cpp</tt>. Given an expression with 'scalar' 1664type (integer, fp, complex, or pointer) this method returns the following 1665information:</p> 1666 1667<ul> 1668<li>Whether the expression is an integer constant expression, a general 1669 constant that was folded but has no side effects, a general constant that 1670 was folded but that does have side effects, or an uncomputable/unfoldable 1671 value. 1672</li> 1673<li>If the expression was computable in any way, this method returns the APValue 1674 for the result of the expression.</li> 1675<li>If the expression is not evaluatable at all, this method returns 1676 information on one of the problems with the expression. This includes a 1677 SourceLocation for where the problem is, and a diagnostic ID that explains 1678 the problem. The diagnostic should be have ERROR type.</li> 1679<li>If the expression is not an integer constant expression, this method returns 1680 information on one of the problems with the expression. This includes a 1681 SourceLocation for where the problem is, and a diagnostic ID that explains 1682 the problem. The diagnostic should be have EXTENSION type.</li> 1683</ul> 1684 1685<p>This information gives various clients the flexibility that they want, and we 1686will eventually have some helper methods for various extensions. For example, 1687Sema should have a <tt>Sema::VerifyIntegerConstantExpression</tt> method, which 1688calls Evaluate on the expression. If the expression is not foldable, the error 1689is emitted, and it would return true. If the expression is not an i-c-e, the 1690EXTENSION diagnostic is emitted. Finally it would return false to indicate that 1691the AST is ok.</p> 1692 1693<p>Other clients can use the information in other ways, for example, codegen can 1694just use expressions that are foldable in any way.</p> 1695 1696<!-- ========== --> 1697<h4>Extensions</h4> 1698<!-- ========== --> 1699 1700<p>This section describes how some of the various extensions Clang supports 1701interacts with constant evaluation:</p> 1702 1703<ul> 1704<li><b><tt>__extension__</tt></b>: The expression form of this extension causes 1705 any evaluatable subexpression to be accepted as an integer constant 1706 expression.</li> 1707<li><b><tt>__builtin_constant_p</tt></b>: This returns true (as an integer 1708 constant expression) if the operand evaluates to either a numeric value 1709 (that is, not a pointer cast to integral type) of integral, enumeration, 1710 floating or complex type, or if it evaluates to the address of the first 1711 character of a string literal (possibly cast to some other type). As a 1712 special case, if <tt>__builtin_constant_p</tt> is the (potentially 1713 parenthesized) condition of a conditional operator expression ("?:"), only 1714 the true side of the conditional operator is considered, and it is evaluated 1715 with full constant folding.</li> 1716<li><b><tt>__builtin_choose_expr</tt></b>: The condition is required to be an 1717 integer constant expression, but we accept any constant as an "extension of 1718 an extension". This only evaluates one operand depending on which way the 1719 condition evaluates.</li> 1720<li><b><tt>__builtin_classify_type</tt></b>: This always returns an integer 1721 constant expression.</li> 1722<li><b><tt>__builtin_inf,nan,..</tt></b>: These are treated just like a 1723 floating-point literal.</li> 1724<li><b><tt>__builtin_abs,copysign,..</tt></b>: These are constant folded as 1725 general constant expressions.</li> 1726<li><b><tt>__builtin_strlen</tt></b> and <b><tt>strlen</tt></b>: These are 1727 constant folded as integer constant expressions if the argument is a string 1728 literal.</li> 1729</ul> 1730 1731 1732<!-- ======================================================================= --> 1733<h2 id="Howtos">How to change Clang</h2> 1734<!-- ======================================================================= --> 1735 1736<!-- ======================================================================= --> 1737<h3 id="AddingAttributes">How to add an attribute</h3> 1738<!-- ======================================================================= --> 1739 1740<p>To add an attribute, you'll have to add it to the list of attributes, add it 1741to the parsing phase, and look for it in the AST scan. 1742<a href="http://llvm.org/viewvc/llvm-project?view=rev&revision=124217">r124217</a> 1743has a good example of adding a warning attribute.</p> 1744 1745<p>(Beware that this hasn't been reviewed/fixed by the people who designed the 1746attributes system yet.)</p> 1747 1748<h4><a 1749href="http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/Attr.td?view=markup">include/clang/Basic/Attr.td</a></h4> 1750 1751<p>Each attribute gets a <tt>def</tt> inheriting from <tt>Attr</tt> or one of 1752its subclasses. <tt>InheritableAttr</tt> means that the attribute also applies 1753to subsequent declarations of the same name.</p> 1754 1755<p><tt>Spellings</tt> lists the strings that can appear in 1756<tt>__attribute__((here))</tt> or <tt>[[here]]</tt>. All such strings 1757will be synonymous. If you want to allow the <tt>[[]]</tt> C++11 1758syntax, you have to define a list of <tt>Namespaces</tt>, which will 1759let users write <tt>[[namespace:spelling]]</tt>. Using the empty 1760string for a namespace will allow users to write just the spelling 1761with no "<tt>:</tt>".</p> 1762 1763<p><tt>Subjects</tt> restricts what kinds of AST node to which this attribute 1764can appertain (roughly, attach).</p> 1765 1766<p><tt>Args</tt> names the arguments the attribute takes, in order. If 1767<tt>Args</tt> is <tt>[StringArgument<"Arg1">, IntArgument<"Arg2">]</tt> 1768then <tt>__attribute__((myattribute("Hello", 3)))</tt> will be a valid use.</p> 1769 1770<h4>Boilerplate</h4> 1771 1772<p>Write a new <tt>HandleYourAttr()</tt> function in <a 1773href="http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Sema/SemaDeclAttr.cpp?view=markup">lib/Sema/SemaDeclAttr.cpp</a>, 1774and add a case to the switch in <tt>ProcessNonInheritableDeclAttr()</tt> or 1775<tt>ProcessInheritableDeclAttr()</tt> forwarding to it.</p> 1776 1777<p>If your attribute causes extra warnings to fire, define a <tt>DiagGroup</tt> 1778in <a 1779href="http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/DiagnosticGroups.td?view=markup">include/clang/Basic/DiagnosticGroups.td</a> 1780named after the attribute's <tt>Spelling</tt> with "_"s replaced by "-"s. If 1781you're only defining one diagnostic, you can skip <tt>DiagnosticGroups.td</tt> 1782and use <tt>InGroup<DiagGroup<"your-attribute">></tt> directly in <a 1783href="http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/DiagnosticSemaKinds.td?view=markup">DiagnosticSemaKinds.td</a></p> 1784 1785<h4>The meat of your attribute</h4> 1786 1787<p>Find an appropriate place in Clang to do whatever your attribute needs to do. 1788Check for the attribute's presence using <tt>Decl::getAttr<YourAttr>()</tt>.</p> 1789 1790<p>Update the <a href="LanguageExtensions.html">Clang Language Extensions</a> 1791document to describe your new attribute.</p> 1792 1793<!-- ======================================================================= --> 1794<h3 id="AddingExprStmt">How to add an expression or statement</h3> 1795<!-- ======================================================================= --> 1796 1797<p>Expressions and statements are one of the most fundamental constructs within a 1798compiler, because they interact with many different parts of the AST, 1799semantic analysis, and IR generation. Therefore, adding a new 1800expression or statement kind into Clang requires some care. The following list 1801details the various places in Clang where an expression or statement needs to be 1802introduced, along with patterns to follow to ensure that the new 1803expression or statement works well across all of the C languages. We 1804focus on expressions, but statements are similar.</p> 1805 1806<ol> 1807 <li>Introduce parsing actions into the parser. Recursive-descent 1808 parsing is mostly self-explanatory, but there are a few things that 1809 are worth keeping in mind: 1810 <ul> 1811 <li>Keep as much source location information as possible! You'll 1812 want it later to produce great diagnostics and support Clang's 1813 various features that map between source code and the AST.</li> 1814 <li>Write tests for all of the "bad" parsing cases, to make sure 1815 your recovery is good. If you have matched delimiters (e.g., 1816 parentheses, square brackets, etc.), use 1817 <tt>Parser::BalancedDelimiterTracker</tt> to give nice diagnostics when 1818 things go wrong.</li> 1819 </ul> 1820 </li> 1821 1822 <li>Introduce semantic analysis actions into <tt>Sema</tt>. Semantic 1823 analysis should always involve two functions: an <tt>ActOnXXX</tt> 1824 function that will be called directly from the parser, and a 1825 <tt>BuildXXX</tt> function that performs the actual semantic 1826 analysis and will (eventually!) build the AST node. It's fairly 1827 common for the <tt>ActOnCXX</tt> function to do very little (often 1828 just some minor translation from the parser's representation to 1829 <tt>Sema</tt>'s representation of the same thing), but the separation 1830 is still important: C++ template instantiation, for example, 1831 should always call the <tt>BuildXXX</tt> variant. Several notes on 1832 semantic analysis before we get into construction of the AST: 1833 <ul> 1834 <li>Your expression probably involves some types and some 1835 subexpressions. Make sure to fully check that those types, and the 1836 types of those subexpressions, meet your expectations. Add 1837 implicit conversions where necessary to make sure that all of the 1838 types line up exactly the way you want them. Write extensive tests 1839 to check that you're getting good diagnostics for mistakes and 1840 that you can use various forms of subexpressions with your 1841 expression.</li> 1842 <li>When type-checking a type or subexpression, make sure to first 1843 check whether the type is "dependent" 1844 (<tt>Type::isDependentType()</tt>) or whether a subexpression is 1845 type-dependent (<tt>Expr::isTypeDependent()</tt>). If any of these 1846 return true, then you're inside a template and you can't do much 1847 type-checking now. That's normal, and your AST node (when you get 1848 there) will have to deal with this case. At this point, you can 1849 write tests that use your expression within templates, but don't 1850 try to instantiate the templates.</li> 1851 <li>For each subexpression, be sure to call 1852 <tt>Sema::CheckPlaceholderExpr()</tt> to deal with "weird" 1853 expressions that don't behave well as subexpressions. Then, 1854 determine whether you need to perform 1855 lvalue-to-rvalue conversions 1856 (<tt>Sema::DefaultLvalueConversion</tt>e) or 1857 the usual unary conversions 1858 (<tt>Sema::UsualUnaryConversions</tt>), for places where the 1859 subexpression is producing a value you intend to use.</li> 1860 <li>Your <tt>BuildXXX</tt> function will probably just return 1861 <tt>ExprError()</tt> at this point, since you don't have an AST. 1862 That's perfectly fine, and shouldn't impact your testing.</li> 1863 </ul> 1864 </li> 1865 1866 <li>Introduce an AST node for your new expression. This starts with 1867 declaring the node in <tt>include/Basic/StmtNodes.td</tt> and 1868 creating a new class for your expression in the appropriate 1869 <tt>include/AST/Expr*.h</tt> header. It's best to look at the class 1870 for a similar expression to get ideas, and there are some specific 1871 things to watch for: 1872 <ul> 1873 <li>If you need to allocate memory, use the <tt>ASTContext</tt> 1874 allocator to allocate memory. Never use raw <tt>malloc</tt> or 1875 <tt>new</tt>, and never hold any resources in an AST node, because 1876 the destructor of an AST node is never called.</li> 1877 1878 <li>Make sure that <tt>getSourceRange()</tt> covers the exact 1879 source range of your expression. This is needed for diagnostics 1880 and for IDE support.</li> 1881 1882 <li>Make sure that <tt>children()</tt> visits all of the 1883 subexpressions. This is important for a number of features (e.g., IDE 1884 support, C++ variadic templates). If you have sub-types, you'll 1885 also need to visit those sub-types in the 1886 <tt>RecursiveASTVisitor</tt>.</li> 1887 1888 <li>Add printing support (<tt>StmtPrinter.cpp</tt>) and dumping 1889 support (<tt>StmtDumper.cpp</tt>) for your expression.</li> 1890 1891 <li>Add profiling support (<tt>StmtProfile.cpp</tt>) for your AST 1892 node, noting the distinguishing (non-source location) 1893 characteristics of an instance of your expression. Omitting this 1894 step will lead to hard-to-diagnose failures regarding matching of 1895 template declarations.</li> 1896 </ul> 1897 </li> 1898 1899 <li>Teach semantic analysis to build your AST node! At this point, 1900 you can wire up your <tt>Sema::BuildXXX</tt> function to actually 1901 create your AST. A few things to check at this point: 1902 <ul> 1903 <li>If your expression can construct a new C++ class or return a 1904 new Objective-C object, be sure to update and then call 1905 <tt>Sema::MaybeBindToTemporary</tt> for your just-created AST node 1906 to be sure that the object gets properly destructed. An easy way 1907 to test this is to return a C++ class with a private destructor: 1908 semantic analysis should flag an error here with the attempt to 1909 call the destructor.</li> 1910 <li>Inspect the generated AST by printing it using <tt>clang -cc1 1911 -ast-print</tt>, to make sure you're capturing all of the 1912 important information about how the AST was written.</li> 1913 <li>Inspect the generated AST under <tt>clang -cc1 -ast-dump</tt> 1914 to verify that all of the types in the generated AST line up the 1915 way you want them. Remember that clients of the AST should never 1916 have to "think" to understand what's going on. For example, all 1917 implicit conversions should show up explicitly in the AST.</li> 1918 <li>Write tests that use your expression as a subexpression of 1919 other, well-known expressions. Can you call a function using your 1920 expression as an argument? Can you use the ternary operator?</li> 1921 </ul> 1922 </li> 1923 1924 <li>Teach code generation to create IR to your AST node. This step 1925 is the first (and only) that requires knowledge of LLVM IR. There 1926 are several things to keep in mind: 1927 <ul> 1928 <li>Code generation is separated into scalar/aggregate/complex and 1929 lvalue/rvalue paths, depending on what kind of result your 1930 expression produces. On occasion, this requires some careful 1931 factoring of code to avoid duplication.</li> 1932 1933 <li><tt>CodeGenFunction</tt> contains functions 1934 <tt>ConvertType</tt> and <tt>ConvertTypeForMem</tt> that convert 1935 Clang's types (<tt>clang::Type*</tt> or <tt>clang::QualType</tt>) 1936 to LLVM types. 1937 Use the former for values, and the later for memory locations: 1938 test with the C++ "bool" type to check this. If you find 1939 that you are having to use LLVM bitcasts to make 1940 the subexpressions of your expression have the type that your 1941 expression expects, STOP! Go fix semantic analysis and the AST so 1942 that you don't need these bitcasts.</li> 1943 1944 <li>The <tt>CodeGenFunction</tt> class has a number of helper 1945 functions to make certain operations easy, such as generating code 1946 to produce an lvalue or an rvalue, or to initialize a memory 1947 location with a given value. Prefer to use these functions rather 1948 than directly writing loads and stores, because these functions 1949 take care of some of the tricky details for you (e.g., for 1950 exceptions).</li> 1951 1952 <li>If your expression requires some special behavior in the event 1953 of an exception, look at the <tt>push*Cleanup</tt> functions in 1954 <tt>CodeGenFunction</tt> to introduce a cleanup. You shouldn't 1955 have to deal with exception-handling directly.</li> 1956 1957 <li>Testing is extremely important in IR generation. Use <tt>clang 1958 -cc1 -emit-llvm</tt> and <a 1959 href="http://llvm.org/cmds/FileCheck.html">FileCheck</a> to verify 1960 that you're generating the right IR.</li> 1961 </ul> 1962 </li> 1963 1964 <li>Teach template instantiation how to cope with your AST 1965 node, which requires some fairly simple code: 1966 <ul> 1967 <li>Make sure that your expression's constructor properly 1968 computes the flags for type dependence (i.e., the type your 1969 expression produces can change from one instantiation to the 1970 next), value dependence (i.e., the constant value your expression 1971 produces can change from one instantiation to the next), 1972 instantiation dependence (i.e., a template parameter occurs 1973 anywhere in your expression), and whether your expression contains 1974 a parameter pack (for variadic templates). Often, computing these 1975 flags just means combining the results from the various types and 1976 subexpressions.</li> 1977 1978 <li>Add <tt>TransformXXX</tt> and <tt>RebuildXXX</tt> functions to 1979 the 1980 <tt>TreeTransform</tt> class template in <tt>Sema</tt>. 1981 <tt>TransformXXX</tt> should (recursively) transform all of the 1982 subexpressions and types 1983 within your expression, using <tt>getDerived().TransformYYY</tt>. 1984 If all of the subexpressions and types transform without error, it 1985 will then call the <tt>RebuildXXX</tt> function, which will in 1986 turn call <tt>getSema().BuildXXX</tt> to perform semantic analysis 1987 and build your expression.</li> 1988 1989 <li>To test template instantiation, take those tests you wrote to 1990 make sure that you were type checking with type-dependent 1991 expressions and dependent types (from step #2) and instantiate 1992 those templates with various types, some of which type-check and 1993 some that don't, and test the error messages in each case.</li> 1994 </ul> 1995 </li> 1996 1997 <li>There are some "extras" that make other features work better. 1998 It's worth handling these extras to give your expression complete 1999 integration into Clang: 2000 <ul> 2001 <li>Add code completion support for your expression in 2002 <tt>SemaCodeComplete.cpp</tt>.</li> 2003 2004 <li>If your expression has types in it, or has any "interesting" 2005 features other than subexpressions, extend libclang's 2006 <tt>CursorVisitor</tt> to provide proper visitation for your 2007 expression, enabling various IDE features such as syntax 2008 highlighting, cross-referencing, and so on. The 2009 <tt>c-index-test</tt> helper program can be used to test these 2010 features.</li> 2011 </ul> 2012 </li> 2013</ol> 2014 2015</div> 2016</body> 2017</html> 2018