Gitiles
Code Review Sign In
gerrit-public.fairphone.software / platform / external / clang / 27d9e9f440acb28894e6a91e317433ee20ac4944 / . / docs / InternalsManual.html
blob: 0eaf722be31f66ad0dda326b1fad5e2377da4f2f [file] [log] [blame]
Ted Kremenek17a295d2008-06-11 06:19:49 +0000[diff] [blame]1<html>
2<head>
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]3<title>"Clang" CFE Internals Manual</title>
Ted Kremenek17a295d2008-06-11 06:19:49 +0000[diff] [blame]4<link type="text/css" rel="stylesheet" href="../menu.css" />
5<link type="text/css" rel="stylesheet" href="../content.css" />
Sebastian Redl68168562008-11-22 22:16:45 +0000[diff] [blame]6<style type="text/css">
7td {
8 vertical-align: top;
9}
10</style>
Ted Kremenek17a295d2008-06-11 06:19:49 +0000[diff] [blame]11</head>
12<body>
13
14<!--#include virtual="../menu.html.incl"-->
15
16<div id="content">
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]17
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]18<h1>"Clang" CFE Internals Manual</h1>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]19
20<ul>
21<li><a href="#intro">Introduction</a></li>
22<li><a href="#libsystem">LLVM System and Support Libraries</a></li>
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]23<li><a href="#libbasic">The Clang 'Basic' Library</a>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]24 <ul>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]25 <li><a href="#Diagnostics">The Diagnostics Subsystem</a></li>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]26 <li><a href="#SourceLocation">The SourceLocation and SourceManager
27 classes</a></li>
28 </ul>
29</li>
Daniel Dunbar27d9e9f2009-03-30 06:50:01 +0000[diff] [blame^]30<li><a href="#libdriver">The Driver Library</a>
31 <ul>
32 </ul>
33</li>
34<li><a href="#libfrontend">The Frontend Library</a>
35 <ul>
36 </ul>
37</li>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]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>
Chris Lattner3932fe02009-01-06 06:02:08 +0000[diff] [blame]42 <li><a href="#AnnotationToken">Annotation Tokens</a></li>
Chris Lattner79281252008-03-09 02:27:26 +0000[diff] [blame]43 <li><a href="#TokenLexer">The TokenLexer class</a></li>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]44 <li><a href="#MultipleIncludeOpt">The MultipleIncludeOpt class</a></li>
45 </ul>
46</li>
47<li><a href="#libparse">The Parser Library</a>
48 <ul>
49 </ul>
50</li>
51<li><a href="#libast">The AST Library</a>
52 <ul>
53 <li><a href="#Type">The Type class and its subclasses</a></li>
54 <li><a href="#QualType">The QualType class</a></li>
Douglas Gregor2e1cd422008-11-17 14:58:09 +0000[diff] [blame]55 <li><a href="#DeclarationName">Declaration names</a></li>
Douglas Gregor074149e2009-01-05 19:45:36 +0000[diff] [blame]56 <li><a href="#DeclContext">Declaration contexts</a>
57 <ul>
58 <li><a href="#Redeclarations">Redeclarations and Overloads</a></li>
59 <li><a href="#LexicalAndSemanticContexts">Lexical and Semantic
60 Contexts</a></li>
61 <li><a href="#TransparentContexts">Transparent Declaration Contexts</a></li>
62 <li><a href="#MultiDeclContext">Multiply-Defined Declaration Contexts</a></li>
63 </ul>
64 </li>
Ted Kremenek8bc05712007-10-10 23:01:43 +0000[diff] [blame]65 <li><a href="#CFG">The CFG class</a></li>
Chris Lattner7bad1992008-11-16 21:48:07 +0000[diff] [blame]66 <li><a href="#Constants">Constant Folding in the Clang AST</a></li>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]67 </ul>
68</li>
69</ul>
70
71
72<!-- ======================================================================= -->
73<h2 id="intro">Introduction</h2>
74<!-- ======================================================================= -->
75
76<p>This document describes some of the more important APIs and internal design
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]77decisions made in the Clang C front-end. The purpose of this document is to
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]78both capture some of this high level information and also describe some of the
79design decisions behind it. This is meant for people interested in hacking on
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]80Clang, not for end-users. The description below is categorized by
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]81libraries, and does not describe any of the clients of the libraries.</p>
82
83<!-- ======================================================================= -->
84<h2 id="libsystem">LLVM System and Support Libraries</h2>
85<!-- ======================================================================= -->
86
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]87<p>The LLVM libsystem library provides the basic Clang system abstraction layer,
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]88which is used for file system access. The LLVM libsupport library provides many
89underlying libraries and <a
90href="http://llvm.org/docs/ProgrammersManual.html">data-structures</a>,
91 including command line option
92processing and various containers.</p>
93
94<!-- ======================================================================= -->
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]95<h2 id="libbasic">The Clang 'Basic' Library</h2>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]96<!-- ======================================================================= -->
97
98<p>This library certainly needs a better name. The 'basic' library contains a
99number of low-level utilities for tracking and manipulating source buffers,
100locations within the source buffers, diagnostics, tokens, target abstraction,
101and information about the subset of the language being compiled for.</p>
102
103<p>Part of this infrastructure is specific to C (such as the TargetInfo class),
104other parts could be reused for other non-C-based languages (SourceLocation,
105SourceManager, Diagnostics, FileManager). When and if there is future demand
106we can figure out if it makes sense to introduce a new library, move the general
107classes somewhere else, or introduce some other solution.</p>
108
109<p>We describe the roles of these classes in order of their dependencies.</p>
110
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]111
112<!-- ======================================================================= -->
113<h3 id="Diagnostics">The Diagnostics Subsystem</h3>
114<!-- ======================================================================= -->
115
116<p>The Clang Diagnostics subsystem is an important part of how the compiler
117communicates with the human. Diagnostics are the warnings and errors produced
118when the code is incorrect or dubious. In Clang, each diagnostic produced has
119(at the minimum) a unique ID, a <a href="#SourceLocation">SourceLocation</a> to
120"put the caret", an English translation associated with it, and a severity (e.g.
121<tt>WARNING</tt> or <tt>ERROR</tt>). They can also optionally include a number
122of arguments to the dianostic (which fill in "%0"'s in the string) as well as a
123number of source ranges that related to the diagnostic.</p>
124
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]125<p>In this section, we'll be giving examples produced by the Clang command line
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]126driver, but diagnostics can be <a href="#DiagnosticClient">rendered in many
127different ways</a> depending on how the DiagnosticClient interface is
128implemented. A representative example of a diagonstic is:</p>
129
130<pre>
131t.c:38:15: error: invalid operands to binary expression ('int *' and '_Complex float')
132 <font color="darkgreen">P = (P-42) + Gamma*4;</font>
133 <font color="blue">~~~~~~ ^ ~~~~~~~</font>
134</pre>
135
136<p>In this example, you can see the English translation, the severity (error),
137you can see the source location (the caret ("^") and file/line/column info),
138the source ranges "~~~~", arguments to the diagnostic ("int*" and "_Complex
139float"). You'll have to believe me that there is a unique ID backing the
140diagnostic :).</p>
141
142<p>Getting all of this to happen has several steps and involves many moving
143pieces, this section describes them and talks about best practices when adding
144a new diagnostic.</p>
145
146<!-- ============================ -->
147<h4>The DiagnosticKinds.def file</h4>
148<!-- ============================ -->
149
150<p>Diagnostics are created by adding an entry to the <tt><a
151href="http://llvm.org/svn/llvm-project/cfe/trunk/include/clang/Basic/DiagnosticKinds.def"
152>DiagnosticKinds.def</a></tt> file. This file encodes the unique ID of the
153diagnostic (as an enum, the first argument), the severity of the diagnostic
154(second argument) and the English translation + format string.</p>
155
156<p>There is little sanity with the naming of the unique ID's right now. Some
157start with err_, warn_, ext_ to encode the severity into the name. Since the
158enum is referenced in the C++ code that produces the diagnostic, it is somewhat
159useful for it to be reasonably short.</p>
160
161<p>The severity of the diagnostic comes from the set {<tt>NOTE</tt>,
162<tt>WARNING</tt>, <tt>EXTENSION</tt>, <tt>EXTWARN</tt>, <tt>ERROR</tt>}. The
163<tt>ERROR</tt> severity is used for diagnostics indicating the program is never
164acceptable under any circumstances. When an error is emitted, the AST for the
165input code may not be fully built. The <tt>EXTENSION</tt> and <tt>EXTWARN</tt>
166severities are used for extensions to the language that Clang accepts. This
167means that Clang fully understands and can represent them in the AST, but we
168produce diagnostics to tell the user their code is non-portable. The difference
169is that the former are ignored by default, and the later warn by default. The
170<tt>WARNING</tt> severity is used for constructs that are valid in the currently
171selected source language but that are dubious in some way. The <tt>NOTE</tt>
Daniel Dunbar426b8632009-02-17 15:49:03 +0000[diff] [blame]172level is used to staple more information onto previous diagnostics.</p>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]173
174<p>These <em>severities</em> are mapped into a smaller set (the
175Diagnostic::Level enum, {<tt>Ignored</tt>, <tt>Note</tt>, <tt>Warning</tt>,
Chris Lattner0aad2972009-02-05 22:49:08 +0000[diff] [blame]176<tt>Error</tt>, <tt>Fatal</tt> }) of output <em>levels</em> by the diagnostics
Chris Lattnera180fdd2009-02-17 07:07:29 +0000[diff] [blame]177subsystem based on various configuration options. Clang internally supports a
178fully fine grained mapping mechanism that allows you to map almost any
179diagnostic to the output level that you want. The only diagnostics that cannot
180be mapped are <tt>NOTE</tt>s, which always follow the severity of the previously
181emitted diagnostic and <tt>ERROR</tt>s, which can only be mapped to
182<tt>Fatal</tt> (it is not possible to turn an error into a warning,
183for example).</p>
184
185<p>Diagnostic mappings are used in many ways. For example, if the user
186specifies <tt>-pedantic</tt>, <tt>EXTENSION</tt> maps to <tt>Warning</tt>, if
187they specify <tt>-pedantic-errors</tt>, it turns into <tt>Error</tt>. This is
188used to implement options like <tt>-Wunused_macros</tt>, <tt>-Wundef</tt> etc.
189</p>
190
191<p>
192Mapping to <tt>Fatal</tt> should only be used for diagnostics that are
193considered so severe that error recovery won't be able to recover sensibly from
194them (thus spewing a ton of bogus errors). One example of this class of error
195are failure to #include a file.
196</p>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]197
198<!-- ================= -->
199<h4>The Format String</h4>
200<!-- ================= -->
201
202<p>The format string for the diagnostic is very simple, but it has some power.
203It takes the form of a string in English with markers that indicate where and
204how arguments to the diagnostic are inserted and formatted. For example, here
205are some simple format strings:</p>
206
207<pre>
208 "binary integer literals are an extension"
209 "format string contains '\\0' within the string body"
210 "more '<b>%%</b>' conversions than data arguments"
Chris Lattner545b3682008-11-23 20:27:13 +0000[diff] [blame]211 "invalid operands to binary expression (<b>%0</b> and <b>%1</b>)"
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]212 "overloaded '<b>%0</b>' must be a <b>%select{unary|binary|unary or binary}2</b> operator"
213 " (has <b>%1</b> parameter<b>%s1</b>)"
214</pre>
215
216<p>These examples show some important points of format strings. You can use any
217 plain ASCII character in the diagnostic string except "%" without a problem,
218 but these are C strings, so you have to use and be aware of all the C escape
219 sequences (as in the second example). If you want to produce a "%" in the
220 output, use the "%%" escape sequence, like the third diagnostic. Finally,
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]221 Clang uses the "%...[digit]" sequences to specify where and how arguments to
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]222 the diagnostic are formatted.</p>
223
224<p>Arguments to the diagnostic are numbered according to how they are specified
225 by the C++ code that <a href="#producingdiag">produces them</a>, and are
226 referenced by <tt>%0</tt> .. <tt>%9</tt>. If you have more than 10 arguments
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]227 to your diagnostic, you are doing something wrong :). Unlike printf, there
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]228 is no requirement that arguments to the diagnostic end up in the output in
229 the same order as they are specified, you could have a format string with
230 <tt>"%1 %0"</tt> that swaps them, for example. The text in between the
231 percent and digit are formatting instructions. If there are no instructions,
232 the argument is just turned into a string and substituted in.</p>
233
234<p>Here are some "best practices" for writing the English format string:</p>
235
236<ul>
237<li>Keep the string short. It should ideally fit in the 80 column limit of the
238 <tt>DiagnosticKinds.def</tt> file. This avoids the diagnostic wrapping when
239 printed, and forces you to think about the important point you are conveying
240 with the diagnostic.</li>
241<li>Take advantage of location information. The user will be able to see the
242 line and location of the caret, so you don't need to tell them that the
243 problem is with the 4th argument to the function: just point to it.</li>
244<li>Do not capitalize the diagnostic string, and do not end it with a
245 period.</li>
246<li>If you need to quote something in the diagnostic string, use single
247 quotes.</li>
248</ul>
249
250<p>Diagnostics should never take random English strings as arguments: you
251shouldn't use <tt>"you have a problem with %0"</tt> and pass in things like
252<tt>"your argument"</tt> or <tt>"your return value"</tt> as arguments. Doing
253this prevents <a href="translation">translating</a> the Clang diagnostics to
254other languages (because they'll get random English words in their otherwise
255localized diagnostic). The exceptions to this are C/C++ language keywords
256(e.g. auto, const, mutable, etc) and C/C++ operators (<tt>/=</tt>). Note
257that things like "pointer" and "reference" are not keywords. On the other
258hand, you <em>can</em> include anything that comes from the user's source code,
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]259including variable names, types, labels, etc. The 'select' format can be
260used to achieve this sort of thing in a localizable way, see below.</p>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]261
262<!-- ==================================== -->
263<h4>Formatting a Diagnostic Argument</a></h4>
264<!-- ==================================== -->
265
266<p>Arguments to diagnostics are fully typed internally, and come from a couple
267different classes: integers, types, names, and random strings. Depending on
268the class of the argument, it can be optionally formatted in different ways.
269This gives the DiagnosticClient information about what the argument means
270without requiring it to use a specific presentation (consider this MVC for
271Clang :).</p>
272
273<p>Here are the different diagnostic argument formats currently supported by
274Clang:</p>
275
276<table>
277<tr><td colspan="2"><b>"s" format</b></td></tr>
278<tr><td>Example:</td><td><tt>"requires %1 parameter%s1"</tt></td></tr>
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]279<tr><td>Class:</td><td>Integers</td></tr>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]280<tr><td>Description:</td><td>This is a simple formatter for integers that is
281 useful when producing English diagnostics. When the integer is 1, it prints
282 as nothing. When the integer is not 1, it prints as "s". This allows some
Chris Lattner627b7052008-11-23 00:28:33 +0000[diff] [blame]283 simple grammatical forms to be to be handled correctly, and eliminates the
284 need to use gross things like <tt>"requires %1 parameter(s)"</tt>.</td></tr>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]285
286<tr><td colspan="2"><b>"select" format</b></td></tr>
287<tr><td>Example:</td><td><tt>"must be a %select{unary|binary|unary or binary}2
288 operator"</tt></td></tr>
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]289<tr><td>Class:</td><td>Integers</td></tr>
Chris Lattnercc543342008-11-22 23:50:47 +0000[diff] [blame]290<tr><td>Description:</td><td>This format specifier is used to merge multiple
291 related diagnostics together into one common one, without requiring the
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]292 difference to be specified as an English string argument. Instead of
Chris Lattnercc543342008-11-22 23:50:47 +0000[diff] [blame]293 specifying the string, the diagnostic gets an integer argument and the
294 format string selects the numbered option. In this case, the "%2" value
295 must be an integer in the range [0..2]. If it is 0, it prints 'unary', if
296 it is 1 it prints 'binary' if it is 2, it prints 'unary or binary'. This
297 allows other language translations to substitute reasonable words (or entire
298 phrases) based on the semantics of the diagnostic instead of having to do
299 things textually.</td></tr>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]300
301<tr><td colspan="2"><b>"plural" format</b></td></tr>
Sebastian Redl68168562008-11-22 22:16:45 +0000[diff] [blame]302<tr><td>Example:</td><td><tt>"you have %1 %plural{1:mouse|:mice}1 connected to
303 your computer"</tt></td></tr>
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]304<tr><td>Class:</td><td>Integers</td></tr>
Sebastian Redl68168562008-11-22 22:16:45 +0000[diff] [blame]305<tr><td>Description:</td><td><p>This is a formatter for complex plural forms.
306 It is designed to handle even the requirements of languages with very
307 complex plural forms, as many Baltic languages have. The argument consists
308 of a series of expression/form pairs, separated by ':', where the first form
309 whose expression evaluates to true is the result of the modifier.</p>
310 <p>An expression can be empty, in which case it is always true. See the
311 example at the top. Otherwise, it is a series of one or more numeric
312 conditions, separated by ','. If any condition matches, the expression
313 matches. Each numeric condition can take one of three forms.</p>
314 <ul>
315 <li>number: A simple decimal number matches if the argument is the same
Chris Lattner627b7052008-11-23 00:28:33 +0000[diff] [blame]316 as the number. Example: <tt>"%plural{1:mouse|:mice}4"</tt></li>
Sebastian Redl68168562008-11-22 22:16:45 +0000[diff] [blame]317 <li>range: A range in square brackets matches if the argument is within
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]318 the range. Then range is inclusive on both ends. Example:
Chris Lattner627b7052008-11-23 00:28:33 +0000[diff] [blame]319 <tt>"%plural{0:none|1:one|[2,5]:some|:many}2"</tt></li>
320 <li>modulo: A modulo operator is followed by a number, and
321 equals sign and either a number or a range. The tests are the
322 same as for plain
Sebastian Redl68168562008-11-22 22:16:45 +0000[diff] [blame]323 numbers and ranges, but the argument is taken modulo the number first.
Chris Lattner627b7052008-11-23 00:28:33 +0000[diff] [blame]324 Example: <tt>"%plural{%100=0:even hundred|%100=[1,50]:lower half|:everything
325 else}1"</tt></li>
Sebastian Redl68168562008-11-22 22:16:45 +0000[diff] [blame]326 </ul>
327 <p>The parser is very unforgiving. A syntax error, even whitespace, will
328 abort, as will a failure to match the argument against any
329 expression.</p></td></tr>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]330
Chris Lattner077bf5e2008-11-24 03:33:13 +0000[diff] [blame]331<tr><td colspan="2"><b>"objcclass" format</b></td></tr>
332<tr><td>Example:</td><td><tt>"method %objcclass0 not found"</tt></td></tr>
333<tr><td>Class:</td><td>DeclarationName</td></tr>
334<tr><td>Description:</td><td><p>This is a simple formatter that indicates the
335 DeclarationName corresponds to an Objective-C class method selector. As
336 such, it prints the selector with a leading '+'.</p></td></tr>
337
338<tr><td colspan="2"><b>"objcinstance" format</b></td></tr>
339<tr><td>Example:</td><td><tt>"method %objcinstance0 not found"</tt></td></tr>
340<tr><td>Class:</td><td>DeclarationName</td></tr>
341<tr><td>Description:</td><td><p>This is a simple formatter that indicates the
342 DeclarationName corresponds to an Objective-C instance method selector. As
343 such, it prints the selector with a leading '-'.</p></td></tr>
344
Douglas Gregor47b9a1c2009-02-04 17:27:36 +0000[diff] [blame]345<tr><td colspan="2"><b>"q" format</b></td></tr>
346<tr><td>Example:</td><td><tt>"candidate found by name lookup is %q0"</tt></td></tr>
347<tr><td>Class:</td><td>NamedDecl*</td></tr>
348<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>
349
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]350</table>
351
Chris Lattnercc543342008-11-22 23:50:47 +0000[diff] [blame]352<p>It is really easy to add format specifiers to the Clang diagnostics system,
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]353but they should be discussed before they are added. If you are creating a lot
354of repetitive diagnostics and/or have an idea for a useful formatter, please
355bring it up on the cfe-dev mailing list.</p>
Chris Lattnercc543342008-11-22 23:50:47 +0000[diff] [blame]356
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]357<!-- ===================================================== -->
358<h4><a name="#producingdiag">Producing the Diagnostic</a></h4>
359<!-- ===================================================== -->
360
Chris Lattner627b7052008-11-23 00:28:33 +0000[diff] [blame]361<p>Now that you've created the diagnostic in the DiagnosticKinds.def file, you
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]362need to write the code that detects the condition in question and emits the
363new diagnostic. Various components of Clang (e.g. the preprocessor, Sema,
Chris Lattner627b7052008-11-23 00:28:33 +0000[diff] [blame]364etc) provide a helper function named "Diag". It creates a diagnostic and
365accepts the arguments, ranges, and other information that goes along with
366it.</p>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]367
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]368<p>For example, the binary expression error comes from code like this:</p>
Chris Lattner627b7052008-11-23 00:28:33 +0000[diff] [blame]369
370<pre>
371 if (various things that are bad)
372 Diag(Loc, diag::err_typecheck_invalid_operands)
373 &lt;&lt; lex-&gt;getType() &lt;&lt; rex-&gt;getType()
374 &lt;&lt; lex-&gt;getSourceRange() &lt;&lt; rex-&gt;getSourceRange();
375</pre>
376
377<p>This shows that use of the Diag method: they take a location (a <a
378href="#SourceLocation">SourceLocation</a> object) and a diagnostic enum value
379(which matches the name from DiagnosticKinds.def). If the diagnostic takes
380arguments, they are specified with the &lt;&lt; operator: the first argument
381becomes %0, the second becomes %1, etc. The diagnostic interface allows you to
Chris Lattnerfd408ea2008-11-23 00:42:53 +0000[diff] [blame]382specify arguments of many different types, including <tt>int</tt> and
383<tt>unsigned</tt> for integer arguments, <tt>const char*</tt> and
384<tt>std::string</tt> for string arguments, <tt>DeclarationName</tt> and
385<tt>const IdentifierInfo*</tt> for names, <tt>QualType</tt> for types, etc.
386SourceRanges are also specified with the &lt;&lt; operator, but do not have a
387specific ordering requirement.</p>
Chris Lattner627b7052008-11-23 00:28:33 +0000[diff] [blame]388
389<p>As you can see, adding and producing a diagnostic is pretty straightforward.
390The hard part is deciding exactly what you need to say to help the user, picking
391a suitable wording, and providing the information needed to format it correctly.
Chris Lattnerfd408ea2008-11-23 00:42:53 +0000[diff] [blame]392The good news is that the call site that issues a diagnostic should be
393completely independent of how the diagnostic is formatted and in what language
394it is rendered.
Chris Lattner627b7052008-11-23 00:28:33 +0000[diff] [blame]395</p>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]396
Douglas Gregorb2fb6de2009-02-27 17:53:17 +0000[diff] [blame]397<!-- ==================================================== -->
398<h4 id="code-modification-hints">Code Modification Hints</h4>
399<!-- ==================================================== -->
400
401<p>In some cases, the front end emits diagnostics when it is clear
402that some small change to the source code would fix the problem. For
403example, a missing semicolon at the end of a statement or a use of
Chris Lattner34c05332009-02-27 19:31:12 +0000[diff] [blame]404deprecated syntax that is easily rewritten into a more modern form.
405Clang tries very hard to emit the diagnostic and recover gracefully
406in these and other cases.</p>
Douglas Gregorb2fb6de2009-02-27 17:53:17 +0000[diff] [blame]407
Chris Lattner34c05332009-02-27 19:31:12 +0000[diff] [blame]408<p>However, for these cases where the fix is obvious, the diagnostic
409can be annotated with a code
410modification "hint" that describes how to change the code referenced
Douglas Gregorb2fb6de2009-02-27 17:53:17 +0000[diff] [blame]411by the diagnostic to fix the problem. For example, it might add the
412missing semicolon at the end of the statement or rewrite the use of a
413deprecated construct into something more palatable. Here is one such
414example C++ front end, where we warn about the right-shift operator
415changing meaning from C++98 to C++0x:</p>
416
417<pre>
418test.cpp:3:7: warning: use of right-shift operator ('&gt;&gt;') in template argument will require parentheses in C++0x
419A&lt;100 &gt;&gt; 2&gt; *a;
420 ^
421 ( )
422</pre>
423
424<p>Here, the code modification hint is suggesting that parentheses be
425added, and showing exactly where those parentheses would be inserted
426into the source code. The code modification hints themselves describe
427what changes to make to the source code in an abstract manner, which
428the text diagnostic printer renders as a line of "insertions" below
429the caret line. <a href="#DiagnosticClient">Other diagnostic
430clients</a> might choose to render the code differently (e.g., as
431markup inline) or even give the user the ability to automatically fix
432the problem.</p>
433
434<p>All code modification hints are described by the
435<code>CodeModificationHint</code> class, instances of which should be
436attached to the diagnostic using the &lt;&lt; operator in the same way
437that highlighted source ranges and arguments are passed to the
438diagnostic. Code modification hints can be created with one of three
439constructors:</p>
440
441<dl>
442 <dt><code>CodeModificationHint::CreateInsertion(Loc, Code)</code></dt>
443 <dd>Specifies that the given <code>Code</code> (a string) should be inserted
444 before the source location <code>Loc</code>.</dd>
445
446 <dt><code>CodeModificationHint::CreateRemoval(Range)</code></dt>
447 <dd>Specifies that the code in the given source <code>Range</code>
448 should be removed.</dd>
449
450 <dt><code>CodeModificationHint::CreateReplacement(Range, Code)</code></dt>
451 <dd>Specifies that the code in the given source <code>Range</code>
452 should be removed, and replaced with the given <code>Code</code> string.</dd>
453</dl>
454
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]455<!-- ============================================================= -->
456<h4><a name="DiagnosticClient">The DiagnosticClient Interface</a></h4>
457<!-- ============================================================= -->
458
Chris Lattner627b7052008-11-23 00:28:33 +0000[diff] [blame]459<p>Once code generates a diagnostic with all of the arguments and the rest of
460the relevant information, Clang needs to know what to do with it. As previously
461mentioned, the diagnostic machinery goes through some filtering to map a
462severity onto a diagnostic level, then (assuming the diagnostic is not mapped to
463"<tt>Ignore</tt>") it invokes an object that implements the DiagnosticClient
464interface with the information.</p>
465
466<p>It is possible to implement this interface in many different ways. For
467example, the normal Clang DiagnosticClient (named 'TextDiagnosticPrinter') turns
468the arguments into strings (according to the various formatting rules), prints
469out the file/line/column information and the string, then prints out the line of
470code, the source ranges, and the caret. However, this behavior isn't required.
471</p>
472
473<p>Another implementation of the DiagnosticClient interface is the
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]474'TextDiagnosticBuffer' class, which is used when Clang is in -verify mode.
Chris Lattnerfd408ea2008-11-23 00:42:53 +0000[diff] [blame]475Instead of formatting and printing out the diagnostics, this implementation just
476captures and remembers the diagnostics as they fly by. Then -verify compares
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]477the list of produced diagnostics to the list of expected ones. If they disagree,
Chris Lattnerfd408ea2008-11-23 00:42:53 +0000[diff] [blame]478it prints out its own output.
Chris Lattner627b7052008-11-23 00:28:33 +0000[diff] [blame]479</p>
480
Chris Lattnerfd408ea2008-11-23 00:42:53 +0000[diff] [blame]481<p>There are many other possible implementations of this interface, and this is
482why we prefer diagnostics to pass down rich structured information in arguments.
483For example, an HTML output might want declaration names be linkified to where
484they come from in the source. Another example is that a GUI might let you click
485on typedefs to expand them. This application would want to pass significantly
486more information about types through to the GUI than a simple flat string. The
487interface allows this to happen.</p>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]488
489<!-- ====================================================== -->
490<h4><a name="translation">Adding Translations to Clang</a></h4>
491<!-- ====================================================== -->
492
Chris Lattner627b7052008-11-23 00:28:33 +0000[diff] [blame]493<p>Not possible yet! Diagnostic strings should be written in UTF-8, the client
Chris Lattnerfd408ea2008-11-23 00:42:53 +0000[diff] [blame]494can translate to the relevant code page if needed. Each translation completely
495replaces the format string for the diagnostic.</p>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]496
497
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]498<!-- ======================================================================= -->
499<h3 id="SourceLocation">The SourceLocation and SourceManager classes</h3>
500<!-- ======================================================================= -->
501
502<p>Strangely enough, the SourceLocation class represents a location within the
503source code of the program. Important design points include:</p>
504
505<ol>
506<li>sizeof(SourceLocation) must be extremely small, as these are embedded into
507 many AST nodes and are passed around often. Currently it is 32 bits.</li>
508<li>SourceLocation must be a simple value object that can be efficiently
509 copied.</li>
510<li>We should be able to represent a source location for any byte of any input
511 file. This includes in the middle of tokens, in whitespace, in trigraphs,
512 etc.</li>
513<li>A SourceLocation must encode the current #include stack that was active when
514 the location was processed. For example, if the location corresponds to a
515 token, it should contain the set of #includes active when the token was
516 lexed. This allows us to print the #include stack for a diagnostic.</li>
517<li>SourceLocation must be able to describe macro expansions, capturing both
518 the ultimate instantiation point and the source of the original character
519 data.</li>
520</ol>
521
522<p>In practice, the SourceLocation works together with the SourceManager class
Chris Lattner18376dd2009-01-16 07:00:50 +0000[diff] [blame]523to encode two pieces of information about a location: it's spelling location
Chris Lattner88054de2009-01-16 07:15:35 +0000[diff] [blame]524and it's instantiation location. For most tokens, these will be the same. However,
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]525for a macro expansion (or tokens that came from a _Pragma directive) these will
526describe the location of the characters corresponding to the token and the
527location where the token was used (i.e. the macro instantiation point or the
528location of the _Pragma itself).</p>
529
Chris Lattner3fcbb892008-11-23 08:32:53 +0000[diff] [blame]530<p>For efficiency, we only track one level of macro instantiations: if a token was
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]531produced by multiple instantiations, we only track the source and ultimate
532destination. Though we could track the intermediate instantiation points, this
533would require extra bookkeeping and no known client would benefit substantially
534from this.</p>
535
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]536<p>The Clang front-end inherently depends on the location of a token being
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]537tracked correctly. If it is ever incorrect, the front-end may get confused and
538die. The reason for this is that the notion of the 'spelling' of a Token in
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]539Clang depends on being able to find the original input characters for the token.
Chris Lattner18376dd2009-01-16 07:00:50 +0000[diff] [blame]540This concept maps directly to the "spelling location" for the token.</p>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]541
542<!-- ======================================================================= -->
Daniel Dunbar27d9e9f2009-03-30 06:50:01 +0000[diff] [blame^]543<h2 id="libdriver">The Driver Library</h2>
544<!-- ======================================================================= -->
545
546<p>The clang Driver and library are documented <a href="DriverInternals.html">here<a>.<p>
547
548<!-- ======================================================================= -->
549<h2 id="libfrontend">The Frontend Library</h2>
550<!-- ======================================================================= -->
551
552<p>The Frontend library contains functionality useful for building
553tools on top of the clang libraries, for example several methods for
554outputting diagnostics.</p>
555
556<!-- ======================================================================= -->
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]557<h2 id="liblex">The Lexer and Preprocessor Library</h2>
558<!-- ======================================================================= -->
559
560<p>The Lexer library contains several tightly-connected classes that are involved
561with the nasty process of lexing and preprocessing C source code. The main
562interface to this library for outside clients is the large <a
563href="#Preprocessor">Preprocessor</a> class.
564It contains the various pieces of state that are required to coherently read
565tokens out of a translation unit.</p>
566
567<p>The core interface to the Preprocessor object (once it is set up) is the
568Preprocessor::Lex method, which returns the next <a href="#Token">Token</a> from
569the preprocessor stream. There are two types of token providers that the
570preprocessor is capable of reading from: a buffer lexer (provided by the <a
571href="#Lexer">Lexer</a> class) and a buffered token stream (provided by the <a
Chris Lattner79281252008-03-09 02:27:26 +0000[diff] [blame]572href="#TokenLexer">TokenLexer</a> class).
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]573
574
575<!-- ======================================================================= -->
576<h3 id="Token">The Token class</h3>
577<!-- ======================================================================= -->
578
579<p>The Token class is used to represent a single lexed token. Tokens are
580intended to be used by the lexer/preprocess and parser libraries, but are not
581intended to live beyond them (for example, they should not live in the ASTs).<p>
582
583<p>Tokens most often live on the stack (or some other location that is efficient
584to access) as the parser is running, but occasionally do get buffered up. For
585example, macro definitions are stored as a series of tokens, and the C++
Chris Lattner3fcbb892008-11-23 08:32:53 +0000[diff] [blame]586front-end periodically needs to buffer tokens up for tentative parsing and
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]587various pieces of look-ahead. As such, the size of a Token matter. On a 32-bit
588system, sizeof(Token) is currently 16 bytes.</p>
589
Chris Lattner3932fe02009-01-06 06:02:08 +0000[diff] [blame]590<p>Tokens occur in two forms: "<a href="#AnnotationToken">Annotation
591Tokens</a>" and normal tokens. Normal tokens are those returned by the lexer,
592annotation tokens represent semantic information and are produced by the parser,
593replacing normal tokens in the token stream. Normal tokens contain the
594following information:</p>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]595
596<ul>
597<li><b>A SourceLocation</b> - This indicates the location of the start of the
598token.</li>
599
600<li><b>A length</b> - This stores the length of the token as stored in the
601SourceBuffer. For tokens that include them, this length includes trigraphs and
602escaped newlines which are ignored by later phases of the compiler. By pointing
603into the original source buffer, it is always possible to get the original
604spelling of a token completely accurately.</li>
605
606<li><b>IdentifierInfo</b> - If a token takes the form of an identifier, and if
607identifier lookup was enabled when the token was lexed (e.g. the lexer was not
608reading in 'raw' mode) this contains a pointer to the unique hash value for the
609identifier. Because the lookup happens before keyword identification, this
610field is set even for language keywords like 'for'.</li>
611
612<li><b>TokenKind</b> - This indicates the kind of token as classified by the
613lexer. This includes things like <tt>tok::starequal</tt> (for the "*="
614operator), <tt>tok::ampamp</tt> for the "&amp;&amp;" token, and keyword values
615(e.g. <tt>tok::kw_for</tt>) for identifiers that correspond to keywords. Note
616that some tokens can be spelled multiple ways. For example, C++ supports
617"operator keywords", where things like "and" are treated exactly like the
618"&amp;&amp;" operator. In these cases, the kind value is set to
619<tt>tok::ampamp</tt>, which is good for the parser, which doesn't have to
620consider both forms. For something that cares about which form is used (e.g.
621the preprocessor 'stringize' operator) the spelling indicates the original
622form.</li>
623
624<li><b>Flags</b> - There are currently four flags tracked by the
625lexer/preprocessor system on a per-token basis:
626
627 <ol>
628 <li><b>StartOfLine</b> - This was the first token that occurred on its input
629 source line.</li>
630 <li><b>LeadingSpace</b> - There was a space character either immediately
631 before the token or transitively before the token as it was expanded
632 through a macro. The definition of this flag is very closely defined by
633 the stringizing requirements of the preprocessor.</li>
634 <li><b>DisableExpand</b> - This flag is used internally to the preprocessor to
635 represent identifier tokens which have macro expansion disabled. This
636 prevents them from being considered as candidates for macro expansion ever
637 in the future.</li>
638 <li><b>NeedsCleaning</b> - This flag is set if the original spelling for the
639 token includes a trigraph or escaped newline. Since this is uncommon,
640 many pieces of code can fast-path on tokens that did not need cleaning.
641 </p>
642 </ol>
643</li>
644</ul>
645
Chris Lattner3932fe02009-01-06 06:02:08 +0000[diff] [blame]646<p>One interesting (and somewhat unusual) aspect of normal tokens is that they
647don't contain any semantic information about the lexed value. For example, if
648the token was a pp-number token, we do not represent the value of the number
649that was lexed (this is left for later pieces of code to decide). Additionally,
650the lexer library has no notion of typedef names vs variable names: both are
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]651returned as identifiers, and the parser is left to decide whether a specific
652identifier is a typedef or a variable (tracking this requires scope information
Chris Lattner3932fe02009-01-06 06:02:08 +0000[diff] [blame]653among other things). The parser can do this translation by replacing tokens
654returned by the preprocessor with "Annotation Tokens".</p>
655
656<!-- ======================================================================= -->
657<h3 id="AnnotationToken">Annotation Tokens</h3>
658<!-- ======================================================================= -->
659
660<p>Annotation Tokens are tokens that are synthesized by the parser and injected
661into the preprocessor's token stream (replacing existing tokens) to record
662semantic information found by the parser. For example, if "foo" is found to be
663a typedef, the "foo" <tt>tok::identifier</tt> token is replaced with an
664<tt>tok::annot_typename</tt>. This is useful for a couple of reasons: 1) this
665makes it easy to handle qualified type names (e.g. "foo::bar::baz&lt;42&gt;::t")
666in C++ as a single "token" in the parser. 2) if the parser backtracks, the
667reparse does not need to redo semantic analysis to determine whether a token
668sequence is a variable, type, template, etc.</p>
669
670<p>Annotation Tokens are created by the parser and reinjected into the parser's
671token stream (when backtracking is enabled). Because they can only exist in
672tokens that the preprocessor-proper is done with, it doesn't need to keep around
673flags like "start of line" that the preprocessor uses to do its job.
674Additionally, an annotation token may "cover" a sequence of preprocessor tokens
675(e.g. <tt>a::b::c</tt> is five preprocessor tokens). As such, the valid fields
676of an annotation token are different than the fields for a normal token (but
677they are multiplexed into the normal Token fields):</p>
678
679<ul>
680<li><b>SourceLocation "Location"</b> - The SourceLocation for the annotation
681token indicates the first token replaced by the annotation token. In the example
682above, it would be the location of the "a" identifier.</li>
683
684<li><b>SourceLocation "AnnotationEndLoc"</b> - This holds the location of the
685last token replaced with the annotation token. In the example above, it would
686be the location of the "c" identifier.</li>
687
688<li><b>void* "AnnotationValue"</b> - This contains an opaque object that the
689parser gets from Sema through an Actions module, it is passed around and Sema
690intepretes it, based on the type of annotation token.</li>
691
692<li><b>TokenKind "Kind"</b> - This indicates the kind of Annotation token this
693is. See below for the different valid kinds.</li>
694</ul>
695
696<p>Annotation tokens currently come in three kinds:</p>
697
698<ol>
699<li><b>tok::annot_typename</b>: This annotation token represents a
700resolved typename token that is potentially qualified. The AnnotationValue
Steve Naroffb43a50f2009-01-28 19:39:02 +0000[diff] [blame]701field contains a pointer returned by Action::getTypeName(). In the case of the
Chris Lattner3932fe02009-01-06 06:02:08 +0000[diff] [blame]702Sema actions module, this is a <tt>Decl*</tt> for the type.</li>
703
704<li><b>tok::annot_cxxscope</b>: This annotation token represents a C++ scope
705specifier, such as "A::B::". This corresponds to the grammar productions "::"
706and ":: [opt] nested-name-specifier". The AnnotationValue pointer is returned
707by the Action::ActOnCXXGlobalScopeSpecifier and
708Action::ActOnCXXNestedNameSpecifier callbacks. In the case of Sema, this is a
709<tt>DeclContext*</tt>.</li>
710
Douglas Gregor39a8de12009-02-25 19:37:18 +0000[diff] [blame]711<li><b>tok::annot_template_id</b>: This annotation token represents a
712C++ template-id such as "foo&lt;int, 4&gt;", where "foo" is the name
713of a template. The AnnotationValue pointer is a pointer to a malloc'd
714TemplateIdAnnotation object. Depending on the context, a parsed template-id that names a type might become a typename annotation token (if all we care about is the named type, e.g., because it occurs in a type specifier) or might remain a template-id token (if we want to retain more source location information or produce a new type, e.g., in a declaration of a class template specialization). template-id annotation tokens that refer to a type can be "upgraded" to typename annotation tokens by the parser.</li>
Chris Lattner3932fe02009-01-06 06:02:08 +0000[diff] [blame]715
716</ol>
717
Cedric Venetda76b282009-01-06 16:22:54 +0000[diff] [blame]718<p>As mentioned above, annotation tokens are not returned by the preprocessor,
Chris Lattner3932fe02009-01-06 06:02:08 +0000[diff] [blame]719they are formed on demand by the parser. This means that the parser has to be
720aware of cases where an annotation could occur and form it where appropriate.
721This is somewhat similar to how the parser handles Translation Phase 6 of C99:
722String Concatenation (see C99 5.1.1.2). In the case of string concatenation,
723the preprocessor just returns distinct tok::string_literal and
724tok::wide_string_literal tokens and the parser eats a sequence of them wherever
725the grammar indicates that a string literal can occur.</p>
726
727<p>In order to do this, whenever the parser expects a tok::identifier or
728tok::coloncolon, it should call the TryAnnotateTypeOrScopeToken or
729TryAnnotateCXXScopeToken methods to form the annotation token. These methods
730will maximally form the specified annotation tokens and replace the current
731token with them, if applicable. If the current tokens is not valid for an
732annotation token, it will remain an identifier or :: token.</p>
733
734
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]735
736<!-- ======================================================================= -->
737<h3 id="Lexer">The Lexer class</h3>
738<!-- ======================================================================= -->
739
740<p>The Lexer class provides the mechanics of lexing tokens out of a source
741buffer and deciding what they mean. The Lexer is complicated by the fact that
742it operates on raw buffers that have not had spelling eliminated (this is a
743necessity to get decent performance), but this is countered with careful coding
744as well as standard performance techniques (for example, the comment handling
745code is vectorized on X86 and PowerPC hosts).</p>
746
747<p>The lexer has a couple of interesting modal features:</p>
748
749<ul>
750<li>The lexer can operate in 'raw' mode. This mode has several features that
751 make it possible to quickly lex the file (e.g. it stops identifier lookup,
752 doesn't specially handle preprocessor tokens, handles EOF differently, etc).
753 This mode is used for lexing within an "<tt>#if 0</tt>" block, for
754 example.</li>
755<li>The lexer can capture and return comments as tokens. This is required to
756 support the -C preprocessor mode, which passes comments through, and is
757 used by the diagnostic checker to identifier expect-error annotations.</li>
758<li>The lexer can be in ParsingFilename mode, which happens when preprocessing
Chris Lattner84386242007-09-16 19:25:23 +0000[diff] [blame]759 after reading a #include directive. This mode changes the parsing of '&lt;'
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]760 to return an "angled string" instead of a bunch of tokens for each thing
761 within the filename.</li>
762<li>When parsing a preprocessor directive (after "<tt>#</tt>") the
763 ParsingPreprocessorDirective mode is entered. This changes the parser to
764 return EOM at a newline.</li>
765<li>The Lexer uses a LangOptions object to know whether trigraphs are enabled,
766 whether C++ or ObjC keywords are recognized, etc.</li>
767</ul>
768
769<p>In addition to these modes, the lexer keeps track of a couple of other
770 features that are local to a lexed buffer, which change as the buffer is
771 lexed:</p>
772
773<ul>
774<li>The Lexer uses BufferPtr to keep track of the current character being
775 lexed.</li>
776<li>The Lexer uses IsAtStartOfLine to keep track of whether the next lexed token
777 will start with its "start of line" bit set.</li>
778<li>The Lexer keeps track of the current #if directives that are active (which
779 can be nested).</li>
780<li>The Lexer keeps track of an <a href="#MultipleIncludeOpt">
781 MultipleIncludeOpt</a> object, which is used to
782 detect whether the buffer uses the standard "<tt>#ifndef XX</tt> /
783 <tt>#define XX</tt>" idiom to prevent multiple inclusion. If a buffer does,
784 subsequent includes can be ignored if the XX macro is defined.</li>
785</ul>
786
787<!-- ======================================================================= -->
Chris Lattner79281252008-03-09 02:27:26 +0000[diff] [blame]788<h3 id="TokenLexer">The TokenLexer class</h3>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]789<!-- ======================================================================= -->
790
Chris Lattner79281252008-03-09 02:27:26 +0000[diff] [blame]791<p>The TokenLexer class is a token provider that returns tokens from a list
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]792of tokens that came from somewhere else. It typically used for two things: 1)
793returning tokens from a macro definition as it is being expanded 2) returning
794tokens from an arbitrary buffer of tokens. The later use is used by _Pragma and
795will most likely be used to handle unbounded look-ahead for the C++ parser.</p>
796
797<!-- ======================================================================= -->
798<h3 id="MultipleIncludeOpt">The MultipleIncludeOpt class</h3>
799<!-- ======================================================================= -->
800
801<p>The MultipleIncludeOpt class implements a really simple little state machine
802that is used to detect the standard "<tt>#ifndef XX</tt> / <tt>#define XX</tt>"
803idiom that people typically use to prevent multiple inclusion of headers. If a
804buffer uses this idiom and is subsequently #include'd, the preprocessor can
805simply check to see whether the guarding condition is defined or not. If so,
806the preprocessor can completely ignore the include of the header.</p>
807
808
809
810<!-- ======================================================================= -->
811<h2 id="libparse">The Parser Library</h2>
812<!-- ======================================================================= -->
813
814<!-- ======================================================================= -->
815<h2 id="libast">The AST Library</h2>
816<!-- ======================================================================= -->
817
818<!-- ======================================================================= -->
819<h3 id="Type">The Type class and its subclasses</h3>
820<!-- ======================================================================= -->
821
822<p>The Type class (and its subclasses) are an important part of the AST. Types
823are accessed through the ASTContext class, which implicitly creates and uniques
824them as they are needed. Types have a couple of non-obvious features: 1) they
825do not capture type qualifiers like const or volatile (See
826<a href="#QualType">QualType</a>), and 2) they implicitly capture typedef
Chris Lattner8a2bc622007-07-31 06:37:39 +0000[diff] [blame]827information. Once created, types are immutable (unlike decls).</p>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]828
829<p>Typedefs in C make semantic analysis a bit more complex than it would
830be without them. The issue is that we want to capture typedef information
831and represent it in the AST perfectly, but the semantics of operations need to
832"see through" typedefs. For example, consider this code:</p>
833
834<code>
835void func() {<br>
Bill Wendling30d17752007-10-06 01:56:01 +0000[diff] [blame]836&nbsp;&nbsp;typedef int foo;<br>
837&nbsp;&nbsp;foo X, *Y;<br>
838&nbsp;&nbsp;typedef foo* bar;<br>
839&nbsp;&nbsp;bar Z;<br>
840&nbsp;&nbsp;*X; <i>// error</i><br>
841&nbsp;&nbsp;**Y; <i>// error</i><br>
842&nbsp;&nbsp;**Z; <i>// error</i><br>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]843}<br>
844</code>
845
846<p>The code above is illegal, and thus we expect there to be diagnostics emitted
847on the annotated lines. In this example, we expect to get:</p>
848
849<pre>
Chris Lattner8a2bc622007-07-31 06:37:39 +0000[diff] [blame]850<b>test.c:6:1: error: indirection requires pointer operand ('foo' invalid)</b>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]851*X; // error
852<font color="blue">^~</font>
Chris Lattner8a2bc622007-07-31 06:37:39 +0000[diff] [blame]853<b>test.c:7:1: error: indirection requires pointer operand ('foo' invalid)</b>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]854**Y; // error
855<font color="blue">^~~</font>
Chris Lattner8a2bc622007-07-31 06:37:39 +0000[diff] [blame]856<b>test.c:8:1: error: indirection requires pointer operand ('foo' invalid)</b>
857**Z; // error
858<font color="blue">^~~</font>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]859</pre>
860
861<p>While this example is somewhat silly, it illustrates the point: we want to
862retain typedef information where possible, so that we can emit errors about
863"<tt>std::string</tt>" instead of "<tt>std::basic_string&lt;char, std:...</tt>".
864Doing this requires properly keeping typedef information (for example, the type
865of "X" is "foo", not "int"), and requires properly propagating it through the
Chris Lattner8a2bc622007-07-31 06:37:39 +0000[diff] [blame]866various operators (for example, the type of *Y is "foo", not "int"). In order
867to retain this information, the type of these expressions is an instance of the
868TypedefType class, which indicates that the type of these expressions is a
869typedef for foo.
870</p>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]871
Chris Lattner8a2bc622007-07-31 06:37:39 +0000[diff] [blame]872<p>Representing types like this is great for diagnostics, because the
873user-specified type is always immediately available. There are two problems
874with this: first, various semantic checks need to make judgements about the
Chris Lattner33fc68a2007-07-31 18:54:50 +0000[diff] [blame]875<em>actual structure</em> of a type, ignoring typdefs. Second, we need an
876efficient way to query whether two types are structurally identical to each
877other, ignoring typedefs. The solution to both of these problems is the idea of
Chris Lattner8a2bc622007-07-31 06:37:39 +0000[diff] [blame]878canonical types.</p>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]879
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]880<!-- =============== -->
Chris Lattner8a2bc622007-07-31 06:37:39 +0000[diff] [blame]881<h4>Canonical Types</h4>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]882<!-- =============== -->
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]883
Chris Lattner8a2bc622007-07-31 06:37:39 +0000[diff] [blame]884<p>Every instance of the Type class contains a canonical type pointer. For
885simple types with no typedefs involved (e.g. "<tt>int</tt>", "<tt>int*</tt>",
886"<tt>int**</tt>"), the type just points to itself. For types that have a
887typedef somewhere in their structure (e.g. "<tt>foo</tt>", "<tt>foo*</tt>",
888"<tt>foo**</tt>", "<tt>bar</tt>"), the canonical type pointer points to their
889structurally equivalent type without any typedefs (e.g. "<tt>int</tt>",
890"<tt>int*</tt>", "<tt>int**</tt>", and "<tt>int*</tt>" respectively).</p>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]891
Chris Lattner8a2bc622007-07-31 06:37:39 +0000[diff] [blame]892<p>This design provides a constant time operation (dereferencing the canonical
893type pointer) that gives us access to the structure of types. For example,
894we can trivially tell that "bar" and "foo*" are the same type by dereferencing
895their canonical type pointers and doing a pointer comparison (they both point
896to the single "<tt>int*</tt>" type).</p>
897
898<p>Canonical types and typedef types bring up some complexities that must be
899carefully managed. Specifically, the "isa/cast/dyncast" operators generally
900shouldn't be used in code that is inspecting the AST. For example, when type
901checking the indirection operator (unary '*' on a pointer), the type checker
902must verify that the operand has a pointer type. It would not be correct to
903check that with "<tt>isa&lt;PointerType&gt;(SubExpr-&gt;getType())</tt>",
904because this predicate would fail if the subexpression had a typedef type.</p>
905
906<p>The solution to this problem are a set of helper methods on Type, used to
907check their properties. In this case, it would be correct to use
908"<tt>SubExpr-&gt;getType()-&gt;isPointerType()</tt>" to do the check. This
909predicate will return true if the <em>canonical type is a pointer</em>, which is
910true any time the type is structurally a pointer type. The only hard part here
911is remembering not to use the <tt>isa/cast/dyncast</tt> operations.</p>
912
913<p>The second problem we face is how to get access to the pointer type once we
914know it exists. To continue the example, the result type of the indirection
915operator is the pointee type of the subexpression. In order to determine the
916type, we need to get the instance of PointerType that best captures the typedef
917information in the program. If the type of the expression is literally a
918PointerType, we can return that, otherwise we have to dig through the
919typedefs to find the pointer type. For example, if the subexpression had type
920"<tt>foo*</tt>", we could return that type as the result. If the subexpression
921had type "<tt>bar</tt>", we want to return "<tt>foo*</tt>" (note that we do
922<em>not</em> want "<tt>int*</tt>"). In order to provide all of this, Type has
Chris Lattner11406c12007-07-31 16:50:51 +0000[diff] [blame]923a getAsPointerType() method that checks whether the type is structurally a
Chris Lattner8a2bc622007-07-31 06:37:39 +0000[diff] [blame]924PointerType and, if so, returns the best one. If not, it returns a null
925pointer.</p>
926
927<p>This structure is somewhat mystical, but after meditating on it, it will
928make sense to you :).</p>
Chris Lattner86920d32007-07-31 05:42:17 +0000[diff] [blame]929
930<!-- ======================================================================= -->
931<h3 id="QualType">The QualType class</h3>
932<!-- ======================================================================= -->
933
934<p>The QualType class is designed as a trivial value class that is small,
935passed by-value and is efficient to query. The idea of QualType is that it
936stores the type qualifiers (const, volatile, restrict) separately from the types
937themselves: QualType is conceptually a pair of "Type*" and bits for the type
938qualifiers.</p>
939
940<p>By storing the type qualifiers as bits in the conceptual pair, it is
941extremely efficient to get the set of qualifiers on a QualType (just return the
942field of the pair), add a type qualifier (which is a trivial constant-time
943operation that sets a bit), and remove one or more type qualifiers (just return
944a QualType with the bitfield set to empty).</p>
945
946<p>Further, because the bits are stored outside of the type itself, we do not
947need to create duplicates of types with different sets of qualifiers (i.e. there
948is only a single heap allocated "int" type: "const int" and "volatile const int"
949both point to the same heap allocated "int" type). This reduces the heap size
950used to represent bits and also means we do not have to consider qualifiers when
951uniquing types (<a href="#Type">Type</a> does not even contain qualifiers).</p>
952
953<p>In practice, on hosts where it is safe, the 3 type qualifiers are stored in
954the low bit of the pointer to the Type object. This means that QualType is
955exactly the same size as a pointer, and this works fine on any system where
956malloc'd objects are at least 8 byte aligned.</p>
Ted Kremenek8bc05712007-10-10 23:01:43 +0000[diff] [blame]957
958<!-- ======================================================================= -->
Douglas Gregor2e1cd422008-11-17 14:58:09 +0000[diff] [blame]959<h3 id="DeclarationName">Declaration names</h3>
960<!-- ======================================================================= -->
961
962<p>The <tt>DeclarationName</tt> class represents the name of a
963 declaration in Clang. Declarations in the C family of languages can
Chris Lattner3fcbb892008-11-23 08:32:53 +0000[diff] [blame]964 take several different forms. Most declarations are named by
Douglas Gregor2e1cd422008-11-17 14:58:09 +0000[diff] [blame]965 simple identifiers, e.g., "<code>f</code>" and "<code>x</code>" in
966 the function declaration <code>f(int x)</code>. In C++, declaration
967 names can also name class constructors ("<code>Class</code>"
968 in <code>struct Class { Class(); }</code>), class destructors
969 ("<code>~Class</code>"), overloaded operator names ("operator+"),
970 and conversion functions ("<code>operator void const *</code>"). In
971 Objective-C, declaration names can refer to the names of Objective-C
972 methods, which involve the method name and the parameters,
Chris Lattner3fcbb892008-11-23 08:32:53 +0000[diff] [blame]973 collectively called a <i>selector</i>, e.g.,
Douglas Gregor2e1cd422008-11-17 14:58:09 +0000[diff] [blame]974 "<code>setWidth:height:</code>". Since all of these kinds of
Chris Lattner3fcbb892008-11-23 08:32:53 +0000[diff] [blame]975 entities - variables, functions, Objective-C methods, C++
976 constructors, destructors, and operators - are represented as
Douglas Gregor2e1cd422008-11-17 14:58:09 +0000[diff] [blame]977 subclasses of Clang's common <code>NamedDecl</code>
978 class, <code>DeclarationName</code> is designed to efficiently
979 represent any kind of name.</p>
980
981<p>Given
982 a <code>DeclarationName</code> <code>N</code>, <code>N.getNameKind()</code>
Douglas Gregor2def4832008-11-17 20:34:05 +0000[diff] [blame]983 will produce a value that describes what kind of name <code>N</code>
Douglas Gregore94ca9e42008-11-18 14:39:36 +0000[diff] [blame]984 stores. There are 8 options (all of the names are inside
Douglas Gregor2e1cd422008-11-17 14:58:09 +0000[diff] [blame]985 the <code>DeclarationName</code> class)</p>
986<dl>
987 <dt>Identifier</dt>
988 <dd>The name is a simple
989 identifier. Use <code>N.getAsIdentifierInfo()</code> to retrieve the
990 corresponding <code>IdentifierInfo*</code> pointing to the actual
991 identifier. Note that C++ overloaded operators (e.g.,
992 "<code>operator+</code>") are represented as special kinds of
993 identifiers. Use <code>IdentifierInfo</code>'s <code>getOverloadedOperatorID</code>
994 function to determine whether an identifier is an overloaded
995 operator name.</dd>
996
997 <dt>ObjCZeroArgSelector, ObjCOneArgSelector,
998 ObjCMultiArgSelector</dt>
999 <dd>The name is an Objective-C selector, which can be retrieved as a
1000 <code>Selector</code> instance
1001 via <code>N.getObjCSelector()</code>. The three possible name
1002 kinds for Objective-C reflect an optimization within
1003 the <code>DeclarationName</code> class: both zero- and
1004 one-argument selectors are stored as a
1005 masked <code>IdentifierInfo</code> pointer, and therefore require
1006 very little space, since zero- and one-argument selectors are far
1007 more common than multi-argument selectors (which use a different
1008 structure).</dd>
1009
1010 <dt>CXXConstructorName</dt>
1011 <dd>The name is a C++ constructor
1012 name. Use <code>N.getCXXNameType()</code> to retrieve
1013 the <a href="#QualType">type</a> that this constructor is meant to
1014 construct. The type is always the canonical type, since all
1015 constructors for a given type have the same name.</dd>
1016
1017 <dt>CXXDestructorName</dt>
1018 <dd>The name is a C++ destructor
1019 name. Use <code>N.getCXXNameType()</code> to retrieve
1020 the <a href="#QualType">type</a> whose destructor is being
1021 named. This type is always a canonical type.</dd>
1022
1023 <dt>CXXConversionFunctionName</dt>
1024 <dd>The name is a C++ conversion function. Conversion functions are
1025 named according to the type they convert to, e.g., "<code>operator void
1026 const *</code>". Use <code>N.getCXXNameType()</code> to retrieve
1027 the type that this conversion function converts to. This type is
1028 always a canonical type.</dd>
Douglas Gregore94ca9e42008-11-18 14:39:36 +0000[diff] [blame]1029
1030 <dt>CXXOperatorName</dt>
1031 <dd>The name is a C++ overloaded operator name. Overloaded operators
1032 are named according to their spelling, e.g.,
1033 "<code>operator+</code>" or "<code>operator new
1034 []</code>". Use <code>N.getCXXOverloadedOperator()</code> to
1035 retrieve the overloaded operator (a value of
1036 type <code>OverloadedOperatorKind</code>).</dd>
Douglas Gregor2e1cd422008-11-17 14:58:09 +0000[diff] [blame]1037</dl>
1038
1039<p><code>DeclarationName</code>s are cheap to create, copy, and
1040 compare. They require only a single pointer's worth of storage in
Douglas Gregore94ca9e42008-11-18 14:39:36 +0000[diff] [blame]1041 the common cases (identifiers, zero-
Douglas Gregor2e1cd422008-11-17 14:58:09 +0000[diff] [blame]1042 and one-argument Objective-C selectors) and use dense, uniqued
1043 storage for the other kinds of
1044 names. Two <code>DeclarationName</code>s can be compared for
1045 equality (<code>==</code>, <code>!=</code>) using a simple bitwise
1046 comparison, can be ordered
1047 with <code>&lt;</code>, <code>&gt;</code>, <code>&lt;=</code>,
1048 and <code>&gt;=</code> (which provide a lexicographical ordering for
1049 normal identifiers but an unspecified ordering for other kinds of
1050 names), and can be placed into LLVM <code>DenseMap</code>s
1051 and <code>DenseSet</code>s.</p>
1052
1053<p><code>DeclarationName</code> instances can be created in different
1054 ways depending on what kind of name the instance will store. Normal
Douglas Gregore94ca9e42008-11-18 14:39:36 +0000[diff] [blame]1055 identifiers (<code>IdentifierInfo</code> pointers) and Objective-C selectors
Douglas Gregor2e1cd422008-11-17 14:58:09 +0000[diff] [blame]1056 (<code>Selector</code>) can be implicitly converted
1057 to <code>DeclarationName</code>s. Names for C++ constructors,
Douglas Gregore94ca9e42008-11-18 14:39:36 +0000[diff] [blame]1058 destructors, conversion functions, and overloaded operators can be retrieved from
Douglas Gregor2e1cd422008-11-17 14:58:09 +0000[diff] [blame]1059 the <code>DeclarationNameTable</code>, an instance of which is
1060 available as <code>ASTContext::DeclarationNames</code>. The member
1061 functions <code>getCXXConstructorName</code>, <code>getCXXDestructorName</code>,
Douglas Gregore94ca9e42008-11-18 14:39:36 +0000[diff] [blame]1062 <code>getCXXConversionFunctionName</code>, and <code>getCXXOperatorName</code>, respectively,
1063 return <code>DeclarationName</code> instances for the four kinds of
Douglas Gregor2e1cd422008-11-17 14:58:09 +0000[diff] [blame]1064 C++ special function names.</p>
1065
1066<!-- ======================================================================= -->
Douglas Gregor074149e2009-01-05 19:45:36 +0000[diff] [blame]1067<h3 id="DeclContext">Declaration contexts</h3>
1068<!-- ======================================================================= -->
1069<p>Every declaration in a program exists within some <i>declaration
1070 context</i>, such as a translation unit, namespace, class, or
1071 function. Declaration contexts in Clang are represented by
1072 the <code>DeclContext</code> class, from which the various
1073 declaration-context AST nodes
1074 (<code>TranslationUnitDecl</code>, <code>NamespaceDecl</code>, <code>RecordDecl</code>, <code>FunctionDecl</code>,
1075 etc.) will derive. The <code>DeclContext</code> class provides
1076 several facilities common to each declaration context:</p>
1077<dl>
1078 <dt>Source-centric vs. Semantics-centric View of Declarations</dt>
1079 <dd><code>DeclContext</code> provides two views of the declarations
1080 stored within a declaration context. The source-centric view
1081 accurately represents the program source code as written, including
1082 multiple declarations of entities where present (see the
1083 section <a href="#Redeclarations">Redeclarations and
1084 Overloads</a>), while the semantics-centric view represents the
1085 program semantics. The two views are kept synchronized by semantic
1086 analysis while the ASTs are being constructed.</dd>
1087
1088 <dt>Storage of declarations within that context</dt>
1089 <dd>Every declaration context can contain some number of
1090 declarations. For example, a C++ class (represented
1091 by <code>RecordDecl</code>) contains various member functions,
1092 fields, nested types, and so on. All of these declarations will be
1093 stored within the <code>DeclContext</code>, and one can iterate
1094 over the declarations via
1095 [<code>DeclContext::decls_begin()</code>,
1096 <code>DeclContext::decls_end()</code>). This mechanism provides
1097 the source-centric view of declarations in the context.</dd>
1098
1099 <dt>Lookup of declarations within that context</dt>
1100 <dd>The <code>DeclContext</code> structure provides efficient name
1101 lookup for names within that declaration context. For example,
1102 if <code>N</code> is a namespace we can look for the
1103 name <code>N::f</code>
1104 using <code>DeclContext::lookup</code>. The lookup itself is
1105 based on a lazily-constructed array (for declaration contexts
1106 with a small number of declarations) or hash table (for
1107 declaration contexts with more declarations). The lookup
1108 operation provides the semantics-centric view of the declarations
1109 in the context.</dd>
1110
1111 <dt>Ownership of declarations</dt>
1112 <dd>The <code>DeclContext</code> owns all of the declarations that
1113 were declared within its declaration context, and is responsible
1114 for the management of their memory as well as their
1115 (de-)serialization.</dd>
1116</dl>
1117
Douglas Gregor4afa39d2009-01-20 01:17:11 +0000[diff] [blame]1118<p>All declarations are stored within a declaration context, and one
1119 can query
1120 information about the context in which each declaration lives. One
Douglas Gregor074149e2009-01-05 19:45:36 +0000[diff] [blame]1121 can retrieve the <code>DeclContext</code> that contains a
Douglas Gregor4afa39d2009-01-20 01:17:11 +0000[diff] [blame]1122 particular <code>Decl</code>
1123 using <code>Decl::getDeclContext</code>. However, see the
Douglas Gregor074149e2009-01-05 19:45:36 +0000[diff] [blame]1124 section <a href="#LexicalAndSemanticContexts">Lexical and Semantic
1125 Contexts</a> for more information about how to interpret this
1126 context information.</p>
1127
1128<h4 id="Redeclarations">Redeclarations and Overloads</h4>
1129<p>Within a translation unit, it is common for an entity to be
1130declared several times. For example, we might declare a function "f"
1131 and then later re-declare it as part of an inlined definition:</p>
1132
1133<pre>
1134void f(int x, int y, int z = 1);
1135
1136inline void f(int x, int y, int z) { /* ... */ }
1137</pre>
1138
1139<p>The representation of "f" differs in the source-centric and
1140 semantics-centric views of a declaration context. In the
1141 source-centric view, all redeclarations will be present, in the
1142 order they occurred in the source code, making
1143 this view suitable for clients that wish to see the structure of
1144 the source code. In the semantics-centric view, only the most recent "f"
1145 will be found by the lookup, since it effectively replaces the first
1146 declaration of "f".</p>
1147
1148<p>In the semantics-centric view, overloading of functions is
1149 represented explicitly. For example, given two declarations of a
1150 function "g" that are overloaded, e.g.,</p>
1151<pre>
1152void g();
1153void g(int);
1154</pre>
1155<p>the <code>DeclContext::lookup</code> operation will return
1156 an <code>OverloadedFunctionDecl</code> that contains both
1157 declarations of "g". Clients that perform semantic analysis on a
1158 program that is not concerned with the actual source code will
1159 primarily use this semantics-centric view.</p>
1160
1161<h4 id="LexicalAndSemanticContexts">Lexical and Semantic Contexts</h4>
Douglas Gregor4afa39d2009-01-20 01:17:11 +0000[diff] [blame]1162<p>Each declaration has two potentially different
Douglas Gregor074149e2009-01-05 19:45:36 +0000[diff] [blame]1163 declaration contexts: a <i>lexical</i> context, which corresponds to
1164 the source-centric view of the declaration context, and
1165 a <i>semantic</i> context, which corresponds to the
1166 semantics-centric view. The lexical context is accessible
Douglas Gregor4afa39d2009-01-20 01:17:11 +0000[diff] [blame]1167 via <code>Decl::getLexicalDeclContext</code> while the
Douglas Gregor074149e2009-01-05 19:45:36 +0000[diff] [blame]1168 semantic context is accessible
Douglas Gregor4afa39d2009-01-20 01:17:11 +0000[diff] [blame]1169 via <code>Decl::getDeclContext</code>, both of which return
Douglas Gregor074149e2009-01-05 19:45:36 +0000[diff] [blame]1170 <code>DeclContext</code> pointers. For most declarations, the two
1171 contexts are identical. For example:</p>
1172
1173<pre>
1174class X {
1175public:
1176 void f(int x);
1177};
1178</pre>
1179
1180<p>Here, the semantic and lexical contexts of <code>X::f</code> are
1181 the <code>DeclContext</code> associated with the
1182 class <code>X</code> (itself stored as a <code>RecordDecl</code> AST
1183 node). However, we can now define <code>X::f</code> out-of-line:</p>
1184
1185<pre>
1186void X::f(int x = 17) { /* ... */ }
1187</pre>
1188
1189<p>This definition of has different lexical and semantic
1190 contexts. The lexical context corresponds to the declaration
1191 context in which the actual declaration occurred in the source
1192 code, e.g., the translation unit containing <code>X</code>. Thus,
1193 this declaration of <code>X::f</code> can be found by traversing
1194 the declarations provided by
1195 [<code>decls_begin()</code>, <code>decls_end()</code>) in the
1196 translation unit.</p>
1197
1198<p>The semantic context of <code>X::f</code> corresponds to the
1199 class <code>X</code>, since this member function is (semantically) a
1200 member of <code>X</code>. Lookup of the name <code>f</code> into
1201 the <code>DeclContext</code> associated with <code>X</code> will
1202 then return the definition of <code>X::f</code> (including
1203 information about the default argument).</p>
1204
1205<h4 id="TransparentContexts">Transparent Declaration Contexts</h4>
1206<p>In C and C++, there are several contexts in which names that are
1207 logically declared inside another declaration will actually "leak"
1208 out into the enclosing scope from the perspective of name
1209 lookup. The most obvious instance of this behavior is in
1210 enumeration types, e.g.,</p>
1211<pre>
1212enum Color {
1213 Red,
1214 Green,
1215 Blue
1216};
1217</pre>
1218
1219<p>Here, <code>Color</code> is an enumeration, which is a declaration
1220 context that contains the
1221 enumerators <code>Red</code>, <code>Green</code>,
1222 and <code>Blue</code>. Thus, traversing the list of declarations
1223 contained in the enumeration <code>Color</code> will
1224 yield <code>Red</code>, <code>Green</code>,
1225 and <code>Blue</code>. However, outside of the scope
1226 of <code>Color</code> one can name the enumerator <code>Red</code>
1227 without qualifying the name, e.g.,</p>
1228
1229<pre>
1230Color c = Red;
1231</pre>
1232
1233<p>There are other entities in C++ that provide similar behavior. For
1234 example, linkage specifications that use curly braces:</p>
1235
1236<pre>
1237extern "C" {
1238 void f(int);
1239 void g(int);
1240}
1241// f and g are visible here
1242</pre>
1243
1244<p>For source-level accuracy, we treat the linkage specification and
1245 enumeration type as a
1246 declaration context in which its enclosed declarations ("Red",
1247 "Green", and "Blue"; "f" and "g")
1248 are declared. However, these declarations are visible outside of the
1249 scope of the declaration context.</p>
1250
1251<p>These language features (and several others, described below) have
1252 roughly the same set of
1253 requirements: declarations are declared within a particular lexical
1254 context, but the declarations are also found via name lookup in
1255 scopes enclosing the declaration itself. This feature is implemented
1256 via <i>transparent</i> declaration contexts
1257 (see <code>DeclContext::isTransparentContext()</code>), whose
1258 declarations are visible in the nearest enclosing non-transparent
1259 declaration context. This means that the lexical context of the
1260 declaration (e.g., an enumerator) will be the
1261 transparent <code>DeclContext</code> itself, as will the semantic
1262 context, but the declaration will be visible in every outer context
1263 up to and including the first non-transparent declaration context (since
1264 transparent declaration contexts can be nested).</p>
1265
1266<p>The transparent <code>DeclContexts</code> are:</p>
1267<ul>
1268 <li>Enumerations (but not C++0x "scoped enumerations"):
1269 <pre>
1270enum Color {
1271 Red,
1272 Green,
1273 Blue
1274};
1275// Red, Green, and Blue are in scope
1276 </pre></li>
1277 <li>C++ linkage specifications:
1278 <pre>
1279extern "C" {
1280 void f(int);
1281 void g(int);
1282}
1283// f and g are in scope
1284 </pre></li>
1285 <li>Anonymous unions and structs:
1286 <pre>
1287struct LookupTable {
1288 bool IsVector;
1289 union {
1290 std::vector&lt;Item&gt; *Vector;
1291 std::set&lt;Item&gt; *Set;
1292 };
1293};
1294
1295LookupTable LT;
1296LT.Vector = 0; // Okay: finds Vector inside the unnamed union
1297 </pre>
1298 </li>
1299 <li>C++0x inline namespaces:
1300<pre>
1301namespace mylib {
1302 inline namespace debug {
1303 class X;
1304 }
1305}
1306mylib::X *xp; // okay: mylib::X refers to mylib::debug::X
1307</pre>
1308</li>
1309</ul>
1310
1311
1312<h4 id="MultiDeclContext">Multiply-Defined Declaration Contexts</h4>
1313<p>C++ namespaces have the interesting--and, so far, unique--property that
1314the namespace can be defined multiple times, and the declarations
1315provided by each namespace definition are effectively merged (from
1316the semantic point of view). For example, the following two code
1317snippets are semantically indistinguishable:</p>
1318<pre>
1319// Snippet #1:
1320namespace N {
1321 void f();
1322}
1323namespace N {
1324 void f(int);
1325}
1326
1327// Snippet #2:
1328namespace N {
1329 void f();
1330 void f(int);
1331}
1332</pre>
1333
1334<p>In Clang's representation, the source-centric view of declaration
1335 contexts will actually have two separate <code>NamespaceDecl</code>
1336 nodes in Snippet #1, each of which is a declaration context that
1337 contains a single declaration of "f". However, the semantics-centric
1338 view provided by name lookup into the namespace <code>N</code> for
1339 "f" will return an <code>OverloadedFunctionDecl</code> that contains
1340 both declarations of "f".</p>
1341
1342<p><code>DeclContext</code> manages multiply-defined declaration
1343 contexts internally. The
1344 function <code>DeclContext::getPrimaryContext</code> retrieves the
1345 "primary" context for a given <code>DeclContext</code> instance,
1346 which is the <code>DeclContext</code> responsible for maintaining
1347 the lookup table used for the semantics-centric view. Given the
1348 primary context, one can follow the chain
1349 of <code>DeclContext</code> nodes that define additional
1350 declarations via <code>DeclContext::getNextContext</code>. Note that
1351 these functions are used internally within the lookup and insertion
1352 methods of the <code>DeclContext</code>, so the vast majority of
1353 clients can ignore them.</p>
1354
1355<!-- ======================================================================= -->
Ted Kremenek8bc05712007-10-10 23:01:43 +0000[diff] [blame]1356<h3 id="CFG">The <tt>CFG</tt> class</h3>
1357<!-- ======================================================================= -->
1358
1359<p>The <tt>CFG</tt> class is designed to represent a source-level
1360control-flow graph for a single statement (<tt>Stmt*</tt>). Typically
1361instances of <tt>CFG</tt> are constructed for function bodies (usually
1362an instance of <tt>CompoundStmt</tt>), but can also be instantiated to
1363represent the control-flow of any class that subclasses <tt>Stmt</tt>,
1364which includes simple expressions. Control-flow graphs are especially
1365useful for performing
1366<a href="http://en.wikipedia.org/wiki/Data_flow_analysis#Sensitivities">flow-
1367or path-sensitive</a> program analyses on a given function.</p>
1368
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]1369<!-- ============ -->
Ted Kremenek8bc05712007-10-10 23:01:43 +0000[diff] [blame]1370<h4>Basic Blocks</h4>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]1371<!-- ============ -->
Ted Kremenek8bc05712007-10-10 23:01:43 +0000[diff] [blame]1372
1373<p>Concretely, an instance of <tt>CFG</tt> is a collection of basic
1374blocks. Each basic block is an instance of <tt>CFGBlock</tt>, which
1375simply contains an ordered sequence of <tt>Stmt*</tt> (each referring
1376to statements in the AST). The ordering of statements within a block
1377indicates unconditional flow of control from one statement to the
1378next. <a href="#ConditionalControlFlow">Conditional control-flow</a>
1379is represented using edges between basic blocks. The statements
1380within a given <tt>CFGBlock</tt> can be traversed using
1381the <tt>CFGBlock::*iterator</tt> interface.</p>
1382
1383<p>
Ted Kremenek18e17e72007-10-18 22:50:52 +0000[diff] [blame]1384A <tt>CFG</tt> object owns the instances of <tt>CFGBlock</tt> within
Ted Kremenek8bc05712007-10-10 23:01:43 +0000[diff] [blame]1385the control-flow graph it represents. Each <tt>CFGBlock</tt> within a
1386CFG is also uniquely numbered (accessible
1387via <tt>CFGBlock::getBlockID()</tt>). Currently the number is
1388based on the ordering the blocks were created, but no assumptions
1389should be made on how <tt>CFGBlock</tt>s are numbered other than their
1390numbers are unique and that they are numbered from 0..N-1 (where N is
1391the number of basic blocks in the CFG).</p>
1392
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]1393<!-- ===================== -->
Ted Kremenek8bc05712007-10-10 23:01:43 +0000[diff] [blame]1394<h4>Entry and Exit Blocks</h4>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]1395<!-- ===================== -->
Ted Kremenek8bc05712007-10-10 23:01:43 +0000[diff] [blame]1396
1397Each instance of <tt>CFG</tt> contains two special blocks:
1398an <i>entry</i> block (accessible via <tt>CFG::getEntry()</tt>), which
1399has no incoming edges, and an <i>exit</i> block (accessible
1400via <tt>CFG::getExit()</tt>), which has no outgoing edges. Neither
1401block contains any statements, and they serve the role of providing a
1402clear entrance and exit for a body of code such as a function body.
1403The presence of these empty blocks greatly simplifies the
1404implementation of many analyses built on top of CFGs.
1405
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]1406<!-- ===================================================== -->
Ted Kremenek8bc05712007-10-10 23:01:43 +0000[diff] [blame]1407<h4 id ="ConditionalControlFlow">Conditional Control-Flow</h4>
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]1408<!-- ===================================================== -->
Ted Kremenek8bc05712007-10-10 23:01:43 +0000[diff] [blame]1409
1410<p>Conditional control-flow (such as those induced by if-statements
1411and loops) is represented as edges between <tt>CFGBlock</tt>s.
1412Because different C language constructs can induce control-flow,
1413each <tt>CFGBlock</tt> also records an extra <tt>Stmt*</tt> that
1414represents the <i>terminator</i> of the block. A terminator is simply
1415the statement that caused the control-flow, and is used to identify
1416the nature of the conditional control-flow between blocks. For
1417example, in the case of an if-statement, the terminator refers to
1418the <tt>IfStmt</tt> object in the AST that represented the given
1419branch.</p>
1420
1421<p>To illustrate, consider the following code example:</p>
1422
1423<code>
1424int foo(int x) {<br>
1425&nbsp;&nbsp;x = x + 1;<br>
1426<br>
1427&nbsp;&nbsp;if (x > 2) x++;<br>
1428&nbsp;&nbsp;else {<br>
1429&nbsp;&nbsp;&nbsp;&nbsp;x += 2;<br>
1430&nbsp;&nbsp;&nbsp;&nbsp;x *= 2;<br>
1431&nbsp;&nbsp;}<br>
1432<br>
1433&nbsp;&nbsp;return x;<br>
1434}
1435</code>
1436
1437<p>After invoking the parser+semantic analyzer on this code fragment,
1438the AST of the body of <tt>foo</tt> is referenced by a
1439single <tt>Stmt*</tt>. We can then construct an instance
1440of <tt>CFG</tt> representing the control-flow graph of this function
1441body by single call to a static class method:</p>
1442
1443<code>
1444&nbsp;&nbsp;Stmt* FooBody = ...<br>
1445&nbsp;&nbsp;CFG* FooCFG = <b>CFG::buildCFG</b>(FooBody);
1446</code>
1447
1448<p>It is the responsibility of the caller of <tt>CFG::buildCFG</tt>
1449to <tt>delete</tt> the returned <tt>CFG*</tt> when the CFG is no
1450longer needed.</p>
1451
1452<p>Along with providing an interface to iterate over
1453its <tt>CFGBlock</tt>s, the <tt>CFG</tt> class also provides methods
1454that are useful for debugging and visualizing CFGs. For example, the
1455method
1456<tt>CFG::dump()</tt> dumps a pretty-printed version of the CFG to
1457standard error. This is especially useful when one is using a
1458debugger such as gdb. For example, here is the output
1459of <tt>FooCFG->dump()</tt>:</p>
1460
1461<code>
1462&nbsp;[ B5 (ENTRY) ]<br>
1463&nbsp;&nbsp;&nbsp;&nbsp;Predecessors (0):<br>
1464&nbsp;&nbsp;&nbsp;&nbsp;Successors (1): B4<br>
1465<br>
1466&nbsp;[ B4 ]<br>
1467&nbsp;&nbsp;&nbsp;&nbsp;1: x = x + 1<br>
1468&nbsp;&nbsp;&nbsp;&nbsp;2: (x > 2)<br>
1469&nbsp;&nbsp;&nbsp;&nbsp;<b>T: if [B4.2]</b><br>
1470&nbsp;&nbsp;&nbsp;&nbsp;Predecessors (1): B5<br>
1471&nbsp;&nbsp;&nbsp;&nbsp;Successors (2): B3 B2<br>
1472<br>
1473&nbsp;[ B3 ]<br>
1474&nbsp;&nbsp;&nbsp;&nbsp;1: x++<br>
1475&nbsp;&nbsp;&nbsp;&nbsp;Predecessors (1): B4<br>
1476&nbsp;&nbsp;&nbsp;&nbsp;Successors (1): B1<br>
1477<br>
1478&nbsp;[ B2 ]<br>
1479&nbsp;&nbsp;&nbsp;&nbsp;1: x += 2<br>
1480&nbsp;&nbsp;&nbsp;&nbsp;2: x *= 2<br>
1481&nbsp;&nbsp;&nbsp;&nbsp;Predecessors (1): B4<br>
1482&nbsp;&nbsp;&nbsp;&nbsp;Successors (1): B1<br>
1483<br>
1484&nbsp;[ B1 ]<br>
1485&nbsp;&nbsp;&nbsp;&nbsp;1: return x;<br>
1486&nbsp;&nbsp;&nbsp;&nbsp;Predecessors (2): B2 B3<br>
1487&nbsp;&nbsp;&nbsp;&nbsp;Successors (1): B0<br>
1488<br>
1489&nbsp;[ B0 (EXIT) ]<br>
1490&nbsp;&nbsp;&nbsp;&nbsp;Predecessors (1): B1<br>
1491&nbsp;&nbsp;&nbsp;&nbsp;Successors (0):
1492</code>
1493
1494<p>For each block, the pretty-printed output displays for each block
1495the number of <i>predecessor</i> blocks (blocks that have outgoing
1496control-flow to the given block) and <i>successor</i> blocks (blocks
1497that have control-flow that have incoming control-flow from the given
1498block). We can also clearly see the special entry and exit blocks at
1499the beginning and end of the pretty-printed output. For the entry
1500block (block B5), the number of predecessor blocks is 0, while for the
1501exit block (block B0) the number of successor blocks is 0.</p>
1502
1503<p>The most interesting block here is B4, whose outgoing control-flow
1504represents the branching caused by the sole if-statement
1505in <tt>foo</tt>. Of particular interest is the second statement in
1506the block, <b><tt>(x > 2)</tt></b>, and the terminator, printed
1507as <b><tt>if [B4.2]</tt></b>. The second statement represents the
1508evaluation of the condition of the if-statement, which occurs before
1509the actual branching of control-flow. Within the <tt>CFGBlock</tt>
1510for B4, the <tt>Stmt*</tt> for the second statement refers to the
1511actual expression in the AST for <b><tt>(x > 2)</tt></b>. Thus
1512pointers to subclasses of <tt>Expr</tt> can appear in the list of
1513statements in a block, and not just subclasses of <tt>Stmt</tt> that
1514refer to proper C statements.</p>
1515
1516<p>The terminator of block B4 is a pointer to the <tt>IfStmt</tt>
1517object in the AST. The pretty-printer outputs <b><tt>if
1518[B4.2]</tt></b> because the condition expression of the if-statement
1519has an actual place in the basic block, and thus the terminator is
1520essentially
1521<i>referring</i> to the expression that is the second statement of
1522block B4 (i.e., B4.2). In this manner, conditions for control-flow
1523(which also includes conditions for loops and switch statements) are
1524hoisted into the actual basic block.</p>
1525
Chris Lattner62fd2782008-11-22 21:41:31 +0000[diff] [blame]1526<!-- ===================== -->
1527<!-- <h4>Implicit Control-Flow</h4> -->
1528<!-- ===================== -->
Ted Kremenek8bc05712007-10-10 23:01:43 +0000[diff] [blame]1529
1530<!--
1531<p>A key design principle of the <tt>CFG</tt> class was to not require
1532any transformations to the AST in order to represent control-flow.
1533Thus the <tt>CFG</tt> does not perform any "lowering" of the
1534statements in an AST: loops are not transformed into guarded gotos,
1535short-circuit operations are not converted to a set of if-statements,
1536and so on.</p>
1537-->
Ted Kremenek17a295d2008-06-11 06:19:49 +0000[diff] [blame]1538
Chris Lattner7bad1992008-11-16 21:48:07 +0000[diff] [blame]1539
1540<!-- ======================================================================= -->
1541<h3 id="Constants">Constant Folding in the Clang AST</h3>
1542<!-- ======================================================================= -->
1543
1544<p>There are several places where constants and constant folding matter a lot to
1545the Clang front-end. First, in general, we prefer the AST to retain the source
1546code as close to how the user wrote it as possible. This means that if they
1547wrote "5+4", we want to keep the addition and two constants in the AST, we don't
1548want to fold to "9". This means that constant folding in various ways turns
1549into a tree walk that needs to handle the various cases.</p>
1550
1551<p>However, there are places in both C and C++ that require constants to be
1552folded. For example, the C standard defines what an "integer constant
1553expression" (i-c-e) is with very precise and specific requirements. The
1554language then requires i-c-e's in a lot of places (for example, the size of a
1555bitfield, the value for a case statement, etc). For these, we have to be able
1556to constant fold the constants, to do semantic checks (e.g. verify bitfield size
1557is non-negative and that case statements aren't duplicated). We aim for Clang
1558to be very pedantic about this, diagnosing cases when the code does not use an
1559i-c-e where one is required, but accepting the code unless running with
1560<tt>-pedantic-errors</tt>.</p>
1561
1562<p>Things get a little bit more tricky when it comes to compatibility with
1563real-world source code. Specifically, GCC has historically accepted a huge
1564superset of expressions as i-c-e's, and a lot of real world code depends on this
1565unfortuate accident of history (including, e.g., the glibc system headers). GCC
1566accepts anything its "fold" optimizer is capable of reducing to an integer
1567constant, which means that the definition of what it accepts changes as its
1568optimizer does. One example is that GCC accepts things like "case X-X:" even
1569when X is a variable, because it can fold this to 0.</p>
1570
1571<p>Another issue are how constants interact with the extensions we support, such
1572as __builtin_constant_p, __builtin_inf, __extension__ and many others. C99
1573obviously does not specify the semantics of any of these extensions, and the
1574definition of i-c-e does not include them. However, these extensions are often
1575used in real code, and we have to have a way to reason about them.</p>
1576
1577<p>Finally, this is not just a problem for semantic analysis. The code
1578generator and other clients have to be able to fold constants (e.g. to
1579initialize global variables) and has to handle a superset of what C99 allows.
1580Further, these clients can benefit from extended information. For example, we
1581know that "foo()||1" always evaluates to true, but we can't replace the
1582expression with true because it has side effects.</p>
1583
1584<!-- ======================= -->
1585<h4>Implementation Approach</h4>
1586<!-- ======================= -->
1587
1588<p>After trying several different approaches, we've finally converged on a
1589design (Note, at the time of this writing, not all of this has been implemented,
1590consider this a design goal!). Our basic approach is to define a single
1591recursive method evaluation method (<tt>Expr::Evaluate</tt>), which is
1592implemented in <tt>AST/ExprConstant.cpp</tt>. Given an expression with 'scalar'
1593type (integer, fp, complex, or pointer) this method returns the following
1594information:</p>
1595
1596<ul>
1597<li>Whether the expression is an integer constant expression, a general
1598 constant that was folded but has no side effects, a general constant that
1599 was folded but that does have side effects, or an uncomputable/unfoldable
1600 value.
1601</li>
1602<li>If the expression was computable in any way, this method returns the APValue
1603 for the result of the expression.</li>
1604<li>If the expression is not evaluatable at all, this method returns
1605 information on one of the problems with the expression. This includes a
1606 SourceLocation for where the problem is, and a diagnostic ID that explains
1607 the problem. The diagnostic should be have ERROR type.</li>
1608<li>If the expression is not an integer constant expression, this method returns
1609 information on one of the problems with the expression. This includes a
1610 SourceLocation for where the problem is, and a diagnostic ID that explains
1611 the problem. The diagnostic should be have EXTENSION type.</li>
1612</ul>
1613
1614<p>This information gives various clients the flexibility that they want, and we
1615will eventually have some helper methods for various extensions. For example,
1616Sema should have a <tt>Sema::VerifyIntegerConstantExpression</tt> method, which
1617calls Evaluate on the expression. If the expression is not foldable, the error
1618is emitted, and it would return true. If the expression is not an i-c-e, the
1619EXTENSION diagnostic is emitted. Finally it would return false to indicate that
1620the AST is ok.</p>
1621
1622<p>Other clients can use the information in other ways, for example, codegen can
1623just use expressions that are foldable in any way.</p>
1624
1625<!-- ========== -->
1626<h4>Extensions</h4>
1627<!-- ========== -->
1628
Chris Lattner552de0a2008-11-23 08:16:56 +0000[diff] [blame]1629<p>This section describes how some of the various extensions Clang supports
Chris Lattner7bad1992008-11-16 21:48:07 +0000[diff] [blame]1630interacts with constant evaluation:</p>
1631
1632<ul>
1633<li><b><tt>__extension__</tt></b>: The expression form of this extension causes
1634 any evaluatable subexpression to be accepted as an integer constant
1635 expression.</li>
1636<li><b><tt>__builtin_constant_p</tt></b>: This returns true (as a integer
Chris Lattner28daa532008-12-12 06:55:44 +0000[diff] [blame]1637 constant expression) if the operand is any evaluatable constant. As a
1638 special case, if <tt>__builtin_constant_p</tt> is the (potentially
1639 parenthesized) condition of a conditional operator expression ("?:"), only
Chris Lattner42b83dd2008-12-12 18:00:51 +0000[diff] [blame]1640 the true side of the conditional operator is considered, and it is evaluated
1641 with full constant folding.</li>
Chris Lattner7bad1992008-11-16 21:48:07 +0000[diff] [blame]1642<li><b><tt>__builtin_choose_expr</tt></b>: The condition is required to be an
1643 integer constant expression, but we accept any constant as an "extension of
1644 an extension". This only evaluates one operand depending on which way the
1645 condition evaluates.</li>
1646<li><b><tt>__builtin_classify_type</tt></b>: This always returns an integer
1647 constant expression.</li>
1648<li><b><tt>__builtin_inf,nan,..</tt></b>: These are treated just like a
1649 floating-point literal.</li>
1650<li><b><tt>__builtin_abs,copysign,..</tt></b>: These are constant folded as
1651 general constant expressions.</li>
1652</ul>
1653
1654
1655
1656
Ted Kremenek17a295d2008-06-11 06:19:49 +0000[diff] [blame]1657</div>
1658</body>
Douglas Gregor2e1cd422008-11-17 14:58:09 +0000[diff] [blame]1659</html>