Blame - docs/InternalsManual.html - platform/external/clang

blob: 6ecc5d6149001af0eb5694631ecb1cbb52740a77 [file] [log] [blame]

Ted Kremenek	17a295d	2008-06-11 06:19:49 +0000	[diff] [blame]	1	<html>
				2	<head>
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	3	<title>"clang" CFE Internals Manual</title>
Ted Kremenek	17a295d	2008-06-11 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" />
				6	</head>
				7	<body>
				8
				9	<!--#include virtual="../menu.html.incl"-->
				10
				11	<div id="content">
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	12
				13	<h1>"clang" CFE Internals Manual</h1>
				14
				15	<ul>
				16	<li><a href="#intro">Introduction</a></li>
				17	<li><a href="#libsystem">LLVM System and Support Libraries</a></li>
				18	<li><a href="#libbasic">The clang 'Basic' Library</a>
				19	<ul>
Chris Lattner	62fd278	2008-11-22 21:41:31 +0000	[diff] [blame^]	20	<li><a href="#Diagnostics">The Diagnostics Subsystem</a></li>
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	21	<li><a href="#SourceLocation">The SourceLocation and SourceManager
				22	classes</a></li>
				23	</ul>
				24	</li>
				25	<li><a href="#liblex">The Lexer and Preprocessor Library</a>
				26	<ul>
				27	<li><a href="#Token">The Token class</a></li>
				28	<li><a href="#Lexer">The Lexer class</a></li>
Chris Lattner	7928125	2008-03-09 02:27:26 +0000	[diff] [blame]	29	<li><a href="#TokenLexer">The TokenLexer class</a></li>
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	30	<li><a href="#MultipleIncludeOpt">The MultipleIncludeOpt class</a></li>
				31	</ul>
				32	</li>
				33	<li><a href="#libparse">The Parser Library</a>
				34	<ul>
				35	</ul>
				36	</li>
				37	<li><a href="#libast">The AST Library</a>
				38	<ul>
				39	<li><a href="#Type">The Type class and its subclasses</a></li>
				40	<li><a href="#QualType">The QualType class</a></li>
Douglas Gregor	2e1cd42	2008-11-17 14:58:09 +0000	[diff] [blame]	41	<li><a href="#DeclarationName">Declaration names</a></li>
Ted Kremenek	8bc0571	2007-10-10 23:01:43 +0000	[diff] [blame]	42	<li><a href="#CFG">The CFG class</a></li>
Chris Lattner	7bad199	2008-11-16 21:48:07 +0000	[diff] [blame]	43	<li><a href="#Constants">Constant Folding in the Clang AST</a></li>
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	44	</ul>
				45	</li>
				46	</ul>
				47
				48
				49	<!-- ======================================================================= -->
				50	<h2 id="intro">Introduction</h2>
				51	<!-- ======================================================================= -->
				52
				53	<p>This document describes some of the more important APIs and internal design
				54	decisions made in the clang C front-end. The purpose of this document is to
				55	both capture some of this high level information and also describe some of the
				56	design decisions behind it. This is meant for people interested in hacking on
				57	clang, not for end-users. The description below is categorized by
				58	libraries, and does not describe any of the clients of the libraries.</p>
				59
				60	<!-- ======================================================================= -->
				61	<h2 id="libsystem">LLVM System and Support Libraries</h2>
				62	<!-- ======================================================================= -->
				63
				64	<p>The LLVM libsystem library provides the basic clang system abstraction layer,
				65	which is used for file system access. The LLVM libsupport library provides many
				66	underlying libraries and <a
				67	href="http://llvm.org/docs/ProgrammersManual.html">data-structures</a>,
				68	including command line option
				69	processing and various containers.</p>
				70
				71	<!-- ======================================================================= -->
				72	<h2 id="libbasic">The clang 'Basic' Library</h2>
				73	<!-- ======================================================================= -->
				74
				75	<p>This library certainly needs a better name. The 'basic' library contains a
				76	number of low-level utilities for tracking and manipulating source buffers,
				77	locations within the source buffers, diagnostics, tokens, target abstraction,
				78	and information about the subset of the language being compiled for.</p>
				79
				80	<p>Part of this infrastructure is specific to C (such as the TargetInfo class),
				81	other parts could be reused for other non-C-based languages (SourceLocation,
				82	SourceManager, Diagnostics, FileManager). When and if there is future demand
				83	we can figure out if it makes sense to introduce a new library, move the general
				84	classes somewhere else, or introduce some other solution.</p>
				85
				86	<p>We describe the roles of these classes in order of their dependencies.</p>
				87
Chris Lattner	62fd278	2008-11-22 21:41:31 +0000	[diff] [blame^]	88
				89	<!-- ======================================================================= -->
				90	<h3 id="Diagnostics">The Diagnostics Subsystem</h3>
				91	<!-- ======================================================================= -->
				92
				93	<p>The Clang Diagnostics subsystem is an important part of how the compiler
				94	communicates with the human. Diagnostics are the warnings and errors produced
				95	when the code is incorrect or dubious. In Clang, each diagnostic produced has
				96	(at the minimum) a unique ID, a <a href="#SourceLocation">SourceLocation</a> to
				97	"put the caret", an English translation associated with it, and a severity (e.g.
				98	<tt>WARNING</tt> or <tt>ERROR</tt>). They can also optionally include a number
				99	of arguments to the dianostic (which fill in "%0"'s in the string) as well as a
				100	number of source ranges that related to the diagnostic.</p>
				101
				102	<p>In this section, we'll be giving examples produced by the clang command line
				103	driver, but diagnostics can be <a href="#DiagnosticClient">rendered in many
				104	different ways</a> depending on how the DiagnosticClient interface is
				105	implemented. A representative example of a diagonstic is:</p>
				106
				107	<pre>
				108	t.c:38:15: error: invalid operands to binary expression ('int *' and '_Complex float')
				109	<font color="darkgreen">P = (P-42) + Gamma*4;</font>
				110	<font color="blue">~~~~~~ ^ ~~~~~~~</font>
				111	</pre>
				112
				113	<p>In this example, you can see the English translation, the severity (error),
				114	you can see the source location (the caret ("^") and file/line/column info),
				115	the source ranges "~~~~", arguments to the diagnostic ("int*" and "_Complex
				116	float"). You'll have to believe me that there is a unique ID backing the
				117	diagnostic :).</p>
				118
				119	<p>Getting all of this to happen has several steps and involves many moving
				120	pieces, this section describes them and talks about best practices when adding
				121	a new diagnostic.</p>
				122
				123	<!-- ============================ -->
				124	<h4>The DiagnosticKinds.def file</h4>
				125	<!-- ============================ -->
				126
				127	<p>Diagnostics are created by adding an entry to the <tt><a
				128	href="http://llvm.org/svn/llvm-project/cfe/trunk/include/clang/Basic/DiagnosticKinds.def"
				129	>DiagnosticKinds.def</a></tt> file. This file encodes the unique ID of the
				130	diagnostic (as an enum, the first argument), the severity of the diagnostic
				131	(second argument) and the English translation + format string.</p>
				132
				133	<p>There is little sanity with the naming of the unique ID's right now. Some
				134	start with err_, warn_, ext_ to encode the severity into the name. Since the
				135	enum is referenced in the C++ code that produces the diagnostic, it is somewhat
				136	useful for it to be reasonably short.</p>
				137
				138	<p>The severity of the diagnostic comes from the set {<tt>NOTE</tt>,
				139	<tt>WARNING</tt>, <tt>EXTENSION</tt>, <tt>EXTWARN</tt>, <tt>ERROR</tt>}. The
				140	<tt>ERROR</tt> severity is used for diagnostics indicating the program is never
				141	acceptable under any circumstances. When an error is emitted, the AST for the
				142	input code may not be fully built. The <tt>EXTENSION</tt> and <tt>EXTWARN</tt>
				143	severities are used for extensions to the language that Clang accepts. This
				144	means that Clang fully understands and can represent them in the AST, but we
				145	produce diagnostics to tell the user their code is non-portable. The difference
				146	is that the former are ignored by default, and the later warn by default. The
				147	<tt>WARNING</tt> severity is used for constructs that are valid in the currently
				148	selected source language but that are dubious in some way. The <tt>NOTE</tt>
				149	level is used to staple more information onto a previous diagnostics.</p>
				150
				151	<p>These <em>severities</em> are mapped into a smaller set (the
				152	Diagnostic::Level enum, {<tt>Ignored</tt>, <tt>Note</tt>, <tt>Warning</tt>,
				153	<tt>Error</tt> }) of output <em>levels</em> by the diagnostics subsystem based
				154	on various configuration options. For example, if the user specifies
				155	<tt>-pedantic</tt>, <tt>EXTENSION</tt> maps to <tt>Warning</tt>, if they specify
				156	<tt>-pedantic-errors</tt>, it turns into <tt>Error</tt>. Clang also internally
				157	supports a fully fine grained mapping mechanism that allows you to map any
				158	diagnostic that doesn't have <tt>ERRROR</tt> severity to any output level that
				159	you want. This is used to implement options like <tt>-Wunused_macros</tt>,
				160	<tt>-Wundef</tt> etc.</p>
				161
				162	<!-- ================= -->
				163	<h4>The Format String</h4>
				164	<!-- ================= -->
				165
				166	<p>The format string for the diagnostic is very simple, but it has some power.
				167	It takes the form of a string in English with markers that indicate where and
				168	how arguments to the diagnostic are inserted and formatted. For example, here
				169	are some simple format strings:</p>
				170
				171	<pre>
				172	"binary integer literals are an extension"
				173	"format string contains '\\0' within the string body"
				174	"more '<b>%%</b>' conversions than data arguments"
				175	"invalid operands to binary expression ('<b>%0</b>' and '<b>%1</b>')"
				176	"overloaded '<b>%0</b>' must be a <b>%select{unary\|binary\|unary or binary}2</b> operator"
				177	" (has <b>%1</b> parameter<b>%s1</b>)"
				178	</pre>
				179
				180	<p>These examples show some important points of format strings. You can use any
				181	plain ASCII character in the diagnostic string except "%" without a problem,
				182	but these are C strings, so you have to use and be aware of all the C escape
				183	sequences (as in the second example). If you want to produce a "%" in the
				184	output, use the "%%" escape sequence, like the third diagnostic. Finally,
				185	clang uses the "%...[digit]" sequences to specify where and how arguments to
				186	the diagnostic are formatted.</p>
				187
				188	<p>Arguments to the diagnostic are numbered according to how they are specified
				189	by the C++ code that <a href="#producingdiag">produces them</a>, and are
				190	referenced by <tt>%0</tt> .. <tt>%9</tt>. If you have more than 10 arguments
				191	to your diagnostic, you are doing something wrong. :). Unlike printf, there
				192	is no requirement that arguments to the diagnostic end up in the output in
				193	the same order as they are specified, you could have a format string with
				194	<tt>"%1 %0"</tt> that swaps them, for example. The text in between the
				195	percent and digit are formatting instructions. If there are no instructions,
				196	the argument is just turned into a string and substituted in.</p>
				197
				198	<p>Here are some "best practices" for writing the English format string:</p>
				199
				200	<ul>
				201	<li>Keep the string short. It should ideally fit in the 80 column limit of the
				202	<tt>DiagnosticKinds.def</tt> file. This avoids the diagnostic wrapping when
				203	printed, and forces you to think about the important point you are conveying
				204	with the diagnostic.</li>
				205	<li>Take advantage of location information. The user will be able to see the
				206	line and location of the caret, so you don't need to tell them that the
				207	problem is with the 4th argument to the function: just point to it.</li>
				208	<li>Do not capitalize the diagnostic string, and do not end it with a
				209	period.</li>
				210	<li>If you need to quote something in the diagnostic string, use single
				211	quotes.</li>
				212	</ul>
				213
				214	<p>Diagnostics should never take random English strings as arguments: you
				215	shouldn't use <tt>"you have a problem with %0"</tt> and pass in things like
				216	<tt>"your argument"</tt> or <tt>"your return value"</tt> as arguments. Doing
				217	this prevents <a href="translation">translating</a> the Clang diagnostics to
				218	other languages (because they'll get random English words in their otherwise
				219	localized diagnostic). The exceptions to this are C/C++ language keywords
				220	(e.g. auto, const, mutable, etc) and C/C++ operators (<tt>/=</tt>). Note
				221	that things like "pointer" and "reference" are not keywords. On the other
				222	hand, you <em>can</em> include anything that comes from the user's source code,
				223	including variable names, types, labels, etc.</p>
				224
				225	<!-- ==================================== -->
				226	<h4>Formatting a Diagnostic Argument</a></h4>
				227	<!-- ==================================== -->
				228
				229	<p>Arguments to diagnostics are fully typed internally, and come from a couple
				230	different classes: integers, types, names, and random strings. Depending on
				231	the class of the argument, it can be optionally formatted in different ways.
				232	This gives the DiagnosticClient information about what the argument means
				233	without requiring it to use a specific presentation (consider this MVC for
				234	Clang :).</p>
				235
				236	<p>Here are the different diagnostic argument formats currently supported by
				237	Clang:</p>
				238
				239	<table>
				240	<tr><td colspan="2"><b>"s" format</b></td></tr>
				241	<tr><td>Example:</td><td><tt>"requires %1 parameter%s1"</tt></td></tr>
				242	<tr><td>Classes:</td><td>Integers</td></tr>
				243	<tr><td>Description:</td><td>This is a simple formatter for integers that is
				244	useful when producing English diagnostics. When the integer is 1, it prints
				245	as nothing. When the integer is not 1, it prints as "s". This allows some
				246	simple grammar to be to be handled correctly, and eliminates the need to use
				247	gross things like <tt>"rewrite %1 parameter(s)"</tt>.</td></tr>
				248
				249	<tr><td colspan="2"><b>"select" format</b></td></tr>
				250	<tr><td>Example:</td><td><tt>"must be a %select{unary\|binary\|unary or binary}2
				251	operator"</tt></td></tr>
				252	<tr><td>Classes:</td><td>Integers</td></tr>
				253	<tr><td>Description:</td><td>...</td></tr>
				254
				255	<tr><td colspan="2"><b>"plural" format</b></td></tr>
				256	<tr><td>Example:</td><td><tt>".."</tt></td></tr>
				257	<tr><td>Classes:</td><td>Integers</td></tr>
				258	<tr><td>Description:</td><td>...</td></tr>
				259
				260
				261	</table>
				262
				263
				264
				265
				266	<!-- ===================================================== -->
				267	<h4><a name="#producingdiag">Producing the Diagnostic</a></h4>
				268	<!-- ===================================================== -->
				269
				270	<p>SemaExpr.cpp example</p>
				271
				272
				273	<!-- ============================================================= -->
				274	<h4><a name="DiagnosticClient">The DiagnosticClient Interface</a></h4>
				275	<!-- ============================================================= -->
				276
				277	<p>Clang command line, buffering, HTMLizing, etc.</p>
				278
				279	<!-- ====================================================== -->
				280	<h4><a name="translation">Adding Translations to Clang</a></h4>
				281	<!-- ====================================================== -->
				282
				283	<p>Not possible yet!</p>
				284
				285
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	286	<!-- ======================================================================= -->
				287	<h3 id="SourceLocation">The SourceLocation and SourceManager classes</h3>
				288	<!-- ======================================================================= -->
				289
				290	<p>Strangely enough, the SourceLocation class represents a location within the
				291	source code of the program. Important design points include:</p>
				292
				293	<ol>
				294	<li>sizeof(SourceLocation) must be extremely small, as these are embedded into
				295	many AST nodes and are passed around often. Currently it is 32 bits.</li>
				296	<li>SourceLocation must be a simple value object that can be efficiently
				297	copied.</li>
				298	<li>We should be able to represent a source location for any byte of any input
				299	file. This includes in the middle of tokens, in whitespace, in trigraphs,
				300	etc.</li>
				301	<li>A SourceLocation must encode the current #include stack that was active when
				302	the location was processed. For example, if the location corresponds to a
				303	token, it should contain the set of #includes active when the token was
				304	lexed. This allows us to print the #include stack for a diagnostic.</li>
				305	<li>SourceLocation must be able to describe macro expansions, capturing both
				306	the ultimate instantiation point and the source of the original character
				307	data.</li>
				308	</ol>
				309
				310	<p>In practice, the SourceLocation works together with the SourceManager class
				311	to encode two pieces of information about a location: it's physical location
				312	and it's virtual location. For most tokens, these will be the same. However,
				313	for a macro expansion (or tokens that came from a _Pragma directive) these will
				314	describe the location of the characters corresponding to the token and the
				315	location where the token was used (i.e. the macro instantiation point or the
				316	location of the _Pragma itself).</p>
				317
				318	<p>For efficiency, we only track one level of macro instantions: if a token was
				319	produced by multiple instantiations, we only track the source and ultimate
				320	destination. Though we could track the intermediate instantiation points, this
				321	would require extra bookkeeping and no known client would benefit substantially
				322	from this.</p>
				323
				324	<p>The clang front-end inherently depends on the location of a token being
				325	tracked correctly. If it is ever incorrect, the front-end may get confused and
				326	die. The reason for this is that the notion of the 'spelling' of a Token in
				327	clang depends on being able to find the original input characters for the token.
				328	This concept maps directly to the "physical" location for the token.</p>
				329
				330	<!-- ======================================================================= -->
				331	<h2 id="liblex">The Lexer and Preprocessor Library</h2>
				332	<!-- ======================================================================= -->
				333
				334	<p>The Lexer library contains several tightly-connected classes that are involved
				335	with the nasty process of lexing and preprocessing C source code. The main
				336	interface to this library for outside clients is the large <a
				337	href="#Preprocessor">Preprocessor</a> class.
				338	It contains the various pieces of state that are required to coherently read
				339	tokens out of a translation unit.</p>
				340
				341	<p>The core interface to the Preprocessor object (once it is set up) is the
				342	Preprocessor::Lex method, which returns the next <a href="#Token">Token</a> from
				343	the preprocessor stream. There are two types of token providers that the
				344	preprocessor is capable of reading from: a buffer lexer (provided by the <a
				345	href="#Lexer">Lexer</a> class) and a buffered token stream (provided by the <a
Chris Lattner	7928125	2008-03-09 02:27:26 +0000	[diff] [blame]	346	href="#TokenLexer">TokenLexer</a> class).
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	347
				348
				349	<!-- ======================================================================= -->
				350	<h3 id="Token">The Token class</h3>
				351	<!-- ======================================================================= -->
				352
				353	<p>The Token class is used to represent a single lexed token. Tokens are
				354	intended to be used by the lexer/preprocess and parser libraries, but are not
				355	intended to live beyond them (for example, they should not live in the ASTs).<p>
				356
				357	<p>Tokens most often live on the stack (or some other location that is efficient
				358	to access) as the parser is running, but occasionally do get buffered up. For
				359	example, macro definitions are stored as a series of tokens, and the C++
				360	front-end will eventually need to buffer tokens up for tentative parsing and
				361	various pieces of look-ahead. As such, the size of a Token matter. On a 32-bit
				362	system, sizeof(Token) is currently 16 bytes.</p>
				363
				364	<p>Tokens contain the following information:</p>
				365
				366	<ul>
				367	<li><b>A SourceLocation</b> - This indicates the location of the start of the
				368	token.</li>
				369
				370	<li><b>A length</b> - This stores the length of the token as stored in the
				371	SourceBuffer. For tokens that include them, this length includes trigraphs and
				372	escaped newlines which are ignored by later phases of the compiler. By pointing
				373	into the original source buffer, it is always possible to get the original
				374	spelling of a token completely accurately.</li>
				375
				376	<li><b>IdentifierInfo</b> - If a token takes the form of an identifier, and if
				377	identifier lookup was enabled when the token was lexed (e.g. the lexer was not
				378	reading in 'raw' mode) this contains a pointer to the unique hash value for the
				379	identifier. Because the lookup happens before keyword identification, this
				380	field is set even for language keywords like 'for'.</li>
				381
				382	<li><b>TokenKind</b> - This indicates the kind of token as classified by the
				383	lexer. This includes things like <tt>tok::starequal</tt> (for the "*="
				384	operator), <tt>tok::ampamp</tt> for the "&&" token, and keyword values
				385	(e.g. <tt>tok::kw_for</tt>) for identifiers that correspond to keywords. Note
				386	that some tokens can be spelled multiple ways. For example, C++ supports
				387	"operator keywords", where things like "and" are treated exactly like the
				388	"&&" operator. In these cases, the kind value is set to
				389	<tt>tok::ampamp</tt>, which is good for the parser, which doesn't have to
				390	consider both forms. For something that cares about which form is used (e.g.
				391	the preprocessor 'stringize' operator) the spelling indicates the original
				392	form.</li>
				393
				394	<li><b>Flags</b> - There are currently four flags tracked by the
				395	lexer/preprocessor system on a per-token basis:
				396
				397	<ol>
				398	<li><b>StartOfLine</b> - This was the first token that occurred on its input
				399	source line.</li>
				400	<li><b>LeadingSpace</b> - There was a space character either immediately
				401	before the token or transitively before the token as it was expanded
				402	through a macro. The definition of this flag is very closely defined by
				403	the stringizing requirements of the preprocessor.</li>
				404	<li><b>DisableExpand</b> - This flag is used internally to the preprocessor to
				405	represent identifier tokens which have macro expansion disabled. This
				406	prevents them from being considered as candidates for macro expansion ever
				407	in the future.</li>
				408	<li><b>NeedsCleaning</b> - This flag is set if the original spelling for the
				409	token includes a trigraph or escaped newline. Since this is uncommon,
				410	many pieces of code can fast-path on tokens that did not need cleaning.
				411	</p>
				412	</ol>
				413	</li>
				414	</ul>
				415
				416	<p>One interesting (and somewhat unusual) aspect of tokens is that they don't
				417	contain any semantic information about the lexed value. For example, if the
				418	token was a pp-number token, we do not represent the value of the number that
				419	was lexed (this is left for later pieces of code to decide). Additionally, the
				420	lexer library has no notion of typedef names vs variable names: both are
				421	returned as identifiers, and the parser is left to decide whether a specific
				422	identifier is a typedef or a variable (tracking this requires scope information
				423	among other things).</p>
				424
				425	<!-- ======================================================================= -->
				426	<h3 id="Lexer">The Lexer class</h3>
				427	<!-- ======================================================================= -->
				428
				429	<p>The Lexer class provides the mechanics of lexing tokens out of a source
				430	buffer and deciding what they mean. The Lexer is complicated by the fact that
				431	it operates on raw buffers that have not had spelling eliminated (this is a
				432	necessity to get decent performance), but this is countered with careful coding
				433	as well as standard performance techniques (for example, the comment handling
				434	code is vectorized on X86 and PowerPC hosts).</p>
				435
				436	<p>The lexer has a couple of interesting modal features:</p>
				437
				438	<ul>
				439	<li>The lexer can operate in 'raw' mode. This mode has several features that
				440	make it possible to quickly lex the file (e.g. it stops identifier lookup,
				441	doesn't specially handle preprocessor tokens, handles EOF differently, etc).
				442	This mode is used for lexing within an "<tt>#if 0</tt>" block, for
				443	example.</li>
				444	<li>The lexer can capture and return comments as tokens. This is required to
				445	support the -C preprocessor mode, which passes comments through, and is
				446	used by the diagnostic checker to identifier expect-error annotations.</li>
				447	<li>The lexer can be in ParsingFilename mode, which happens when preprocessing
Chris Lattner	8438624	2007-09-16 19:25:23 +0000	[diff] [blame]	448	after reading a #include directive. This mode changes the parsing of '<'
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	449	to return an "angled string" instead of a bunch of tokens for each thing
				450	within the filename.</li>
				451	<li>When parsing a preprocessor directive (after "<tt>#</tt>") the
				452	ParsingPreprocessorDirective mode is entered. This changes the parser to
				453	return EOM at a newline.</li>
				454	<li>The Lexer uses a LangOptions object to know whether trigraphs are enabled,
				455	whether C++ or ObjC keywords are recognized, etc.</li>
				456	</ul>
				457
				458	<p>In addition to these modes, the lexer keeps track of a couple of other
				459	features that are local to a lexed buffer, which change as the buffer is
				460	lexed:</p>
				461
				462	<ul>
				463	<li>The Lexer uses BufferPtr to keep track of the current character being
				464	lexed.</li>
				465	<li>The Lexer uses IsAtStartOfLine to keep track of whether the next lexed token
				466	will start with its "start of line" bit set.</li>
				467	<li>The Lexer keeps track of the current #if directives that are active (which
				468	can be nested).</li>
				469	<li>The Lexer keeps track of an <a href="#MultipleIncludeOpt">
				470	MultipleIncludeOpt</a> object, which is used to
				471	detect whether the buffer uses the standard "<tt>#ifndef XX</tt> /
				472	<tt>#define XX</tt>" idiom to prevent multiple inclusion. If a buffer does,
				473	subsequent includes can be ignored if the XX macro is defined.</li>
				474	</ul>
				475
				476	<!-- ======================================================================= -->
Chris Lattner	7928125	2008-03-09 02:27:26 +0000	[diff] [blame]	477	<h3 id="TokenLexer">The TokenLexer class</h3>
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	478	<!-- ======================================================================= -->
				479
Chris Lattner	7928125	2008-03-09 02:27:26 +0000	[diff] [blame]	480	<p>The TokenLexer class is a token provider that returns tokens from a list
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	481	of tokens that came from somewhere else. It typically used for two things: 1)
				482	returning tokens from a macro definition as it is being expanded 2) returning
				483	tokens from an arbitrary buffer of tokens. The later use is used by _Pragma and
				484	will most likely be used to handle unbounded look-ahead for the C++ parser.</p>
				485
				486	<!-- ======================================================================= -->
				487	<h3 id="MultipleIncludeOpt">The MultipleIncludeOpt class</h3>
				488	<!-- ======================================================================= -->
				489
				490	<p>The MultipleIncludeOpt class implements a really simple little state machine
				491	that is used to detect the standard "<tt>#ifndef XX</tt> / <tt>#define XX</tt>"
				492	idiom that people typically use to prevent multiple inclusion of headers. If a
				493	buffer uses this idiom and is subsequently #include'd, the preprocessor can
				494	simply check to see whether the guarding condition is defined or not. If so,
				495	the preprocessor can completely ignore the include of the header.</p>
				496
				497
				498
				499	<!-- ======================================================================= -->
				500	<h2 id="libparse">The Parser Library</h2>
				501	<!-- ======================================================================= -->
				502
				503	<!-- ======================================================================= -->
				504	<h2 id="libast">The AST Library</h2>
				505	<!-- ======================================================================= -->
				506
				507	<!-- ======================================================================= -->
				508	<h3 id="Type">The Type class and its subclasses</h3>
				509	<!-- ======================================================================= -->
				510
				511	<p>The Type class (and its subclasses) are an important part of the AST. Types
				512	are accessed through the ASTContext class, which implicitly creates and uniques
				513	them as they are needed. Types have a couple of non-obvious features: 1) they
				514	do not capture type qualifiers like const or volatile (See
				515	<a href="#QualType">QualType</a>), and 2) they implicitly capture typedef
Chris Lattner	8a2bc62	2007-07-31 06:37:39 +0000	[diff] [blame]	516	information. Once created, types are immutable (unlike decls).</p>
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	517
				518	<p>Typedefs in C make semantic analysis a bit more complex than it would
				519	be without them. The issue is that we want to capture typedef information
				520	and represent it in the AST perfectly, but the semantics of operations need to
				521	"see through" typedefs. For example, consider this code:</p>
				522
				523	<code>
				524	void func() {<br>
Bill Wendling	30d1775	2007-10-06 01:56:01 +0000	[diff] [blame]	525	typedef int foo;<br>
				526	foo X, *Y;<br>
				527	typedef foo* bar;<br>
				528	bar Z;<br>
				529	*X; <i>// error</i><br>
				530	**Y; <i>// error</i><br>
				531	**Z; <i>// error</i><br>
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	532	}<br>
				533	</code>
				534
				535	<p>The code above is illegal, and thus we expect there to be diagnostics emitted
				536	on the annotated lines. In this example, we expect to get:</p>
				537
				538	<pre>
Chris Lattner	8a2bc62	2007-07-31 06:37:39 +0000	[diff] [blame]	539	<b>test.c:6:1: error: indirection requires pointer operand ('foo' invalid)</b>
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	540	*X; // error
				541	<font color="blue">^~</font>
Chris Lattner	8a2bc62	2007-07-31 06:37:39 +0000	[diff] [blame]	542	<b>test.c:7:1: error: indirection requires pointer operand ('foo' invalid)</b>
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	543	**Y; // error
				544	<font color="blue">^~~</font>
Chris Lattner	8a2bc62	2007-07-31 06:37:39 +0000	[diff] [blame]	545	<b>test.c:8:1: error: indirection requires pointer operand ('foo' invalid)</b>
				546	**Z; // error
				547	<font color="blue">^~~</font>
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	548	</pre>
				549
				550	<p>While this example is somewhat silly, it illustrates the point: we want to
				551	retain typedef information where possible, so that we can emit errors about
				552	"<tt>std::string</tt>" instead of "<tt>std::basic_string<char, std:...</tt>".
				553	Doing this requires properly keeping typedef information (for example, the type
				554	of "X" is "foo", not "int"), and requires properly propagating it through the
Chris Lattner	8a2bc62	2007-07-31 06:37:39 +0000	[diff] [blame]	555	various operators (for example, the type of *Y is "foo", not "int"). In order
				556	to retain this information, the type of these expressions is an instance of the
				557	TypedefType class, which indicates that the type of these expressions is a
				558	typedef for foo.
				559	</p>
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	560
Chris Lattner	8a2bc62	2007-07-31 06:37:39 +0000	[diff] [blame]	561	<p>Representing types like this is great for diagnostics, because the
				562	user-specified type is always immediately available. There are two problems
				563	with this: first, various semantic checks need to make judgements about the
Chris Lattner	33fc68a	2007-07-31 18:54:50 +0000	[diff] [blame]	564	<em>actual structure</em> of a type, ignoring typdefs. Second, we need an
				565	efficient way to query whether two types are structurally identical to each
				566	other, ignoring typedefs. The solution to both of these problems is the idea of
Chris Lattner	8a2bc62	2007-07-31 06:37:39 +0000	[diff] [blame]	567	canonical types.</p>
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	568
Chris Lattner	62fd278	2008-11-22 21:41:31 +0000	[diff] [blame^]	569	<!-- =============== -->
Chris Lattner	8a2bc62	2007-07-31 06:37:39 +0000	[diff] [blame]	570	<h4>Canonical Types</h4>
Chris Lattner	62fd278	2008-11-22 21:41:31 +0000	[diff] [blame^]	571	<!-- =============== -->
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	572
Chris Lattner	8a2bc62	2007-07-31 06:37:39 +0000	[diff] [blame]	573	<p>Every instance of the Type class contains a canonical type pointer. For
				574	simple types with no typedefs involved (e.g. "<tt>int</tt>", "<tt>int*</tt>",
				575	"<tt>int**</tt>"), the type just points to itself. For types that have a
				576	typedef somewhere in their structure (e.g. "<tt>foo</tt>", "<tt>foo*</tt>",
				577	"<tt>foo**</tt>", "<tt>bar</tt>"), the canonical type pointer points to their
				578	structurally equivalent type without any typedefs (e.g. "<tt>int</tt>",
				579	"<tt>int</tt>", "<tt>int</tt>", and "<tt>int</tt>" respectively).</p>
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	580
Chris Lattner	8a2bc62	2007-07-31 06:37:39 +0000	[diff] [blame]	581	<p>This design provides a constant time operation (dereferencing the canonical
				582	type pointer) that gives us access to the structure of types. For example,
				583	we can trivially tell that "bar" and "foo*" are the same type by dereferencing
				584	their canonical type pointers and doing a pointer comparison (they both point
				585	to the single "<tt>int*</tt>" type).</p>
				586
				587	<p>Canonical types and typedef types bring up some complexities that must be
				588	carefully managed. Specifically, the "isa/cast/dyncast" operators generally
				589	shouldn't be used in code that is inspecting the AST. For example, when type
				590	checking the indirection operator (unary '*' on a pointer), the type checker
				591	must verify that the operand has a pointer type. It would not be correct to
				592	check that with "<tt>isa<PointerType>(SubExpr->getType())</tt>",
				593	because this predicate would fail if the subexpression had a typedef type.</p>
				594
				595	<p>The solution to this problem are a set of helper methods on Type, used to
				596	check their properties. In this case, it would be correct to use
				597	"<tt>SubExpr->getType()->isPointerType()</tt>" to do the check. This
				598	predicate will return true if the <em>canonical type is a pointer</em>, which is
				599	true any time the type is structurally a pointer type. The only hard part here
				600	is remembering not to use the <tt>isa/cast/dyncast</tt> operations.</p>
				601
				602	<p>The second problem we face is how to get access to the pointer type once we
				603	know it exists. To continue the example, the result type of the indirection
				604	operator is the pointee type of the subexpression. In order to determine the
				605	type, we need to get the instance of PointerType that best captures the typedef
				606	information in the program. If the type of the expression is literally a
				607	PointerType, we can return that, otherwise we have to dig through the
				608	typedefs to find the pointer type. For example, if the subexpression had type
				609	"<tt>foo*</tt>", we could return that type as the result. If the subexpression
				610	had type "<tt>bar</tt>", we want to return "<tt>foo*</tt>" (note that we do
				611	<em>not</em> want "<tt>int*</tt>"). In order to provide all of this, Type has
Chris Lattner	11406c1	2007-07-31 16:50:51 +0000	[diff] [blame]	612	a getAsPointerType() method that checks whether the type is structurally a
Chris Lattner	8a2bc62	2007-07-31 06:37:39 +0000	[diff] [blame]	613	PointerType and, if so, returns the best one. If not, it returns a null
				614	pointer.</p>
				615
				616	<p>This structure is somewhat mystical, but after meditating on it, it will
				617	make sense to you :).</p>
Chris Lattner	86920d3	2007-07-31 05:42:17 +0000	[diff] [blame]	618
				619	<!-- ======================================================================= -->
				620	<h3 id="QualType">The QualType class</h3>
				621	<!-- ======================================================================= -->
				622
				623	<p>The QualType class is designed as a trivial value class that is small,
				624	passed by-value and is efficient to query. The idea of QualType is that it
				625	stores the type qualifiers (const, volatile, restrict) separately from the types
				626	themselves: QualType is conceptually a pair of "Type*" and bits for the type
				627	qualifiers.</p>
				628
				629	<p>By storing the type qualifiers as bits in the conceptual pair, it is
				630	extremely efficient to get the set of qualifiers on a QualType (just return the
				631	field of the pair), add a type qualifier (which is a trivial constant-time
				632	operation that sets a bit), and remove one or more type qualifiers (just return
				633	a QualType with the bitfield set to empty).</p>
				634
				635	<p>Further, because the bits are stored outside of the type itself, we do not
				636	need to create duplicates of types with different sets of qualifiers (i.e. there
				637	is only a single heap allocated "int" type: "const int" and "volatile const int"
				638	both point to the same heap allocated "int" type). This reduces the heap size
				639	used to represent bits and also means we do not have to consider qualifiers when
				640	uniquing types (<a href="#Type">Type</a> does not even contain qualifiers).</p>
				641
				642	<p>In practice, on hosts where it is safe, the 3 type qualifiers are stored in
				643	the low bit of the pointer to the Type object. This means that QualType is
				644	exactly the same size as a pointer, and this works fine on any system where
				645	malloc'd objects are at least 8 byte aligned.</p>
Ted Kremenek	8bc0571	2007-10-10 23:01:43 +0000	[diff] [blame]	646
				647	<!-- ======================================================================= -->
Douglas Gregor	2e1cd42	2008-11-17 14:58:09 +0000	[diff] [blame]	648	<h3 id="DeclarationName">Declaration names</h3>
				649	<!-- ======================================================================= -->
				650
				651	<p>The <tt>DeclarationName</tt> class represents the name of a
				652	declaration in Clang. Declarations in the C family of languages can
				653	take several different forms. Most declarations are named by are
				654	simple identifiers, e.g., "<code>f</code>" and "<code>x</code>" in
				655	the function declaration <code>f(int x)</code>. In C++, declaration
				656	names can also name class constructors ("<code>Class</code>"
				657	in <code>struct Class { Class(); }</code>), class destructors
				658	("<code>~Class</code>"), overloaded operator names ("operator+"),
				659	and conversion functions ("<code>operator void const *</code>"). In
				660	Objective-C, declaration names can refer to the names of Objective-C
				661	methods, which involve the method name and the parameters,
				662	collectively called a <i>selector</i>, e.g..,
				663	"<code>setWidth:height:</code>". Since all of these kinds of
				664	entities--variables, functions, Objective-C methods, C++
				665	constructors, destructors, and operators---are represented as
				666	subclasses of Clang's common <code>NamedDecl</code>
				667	class, <code>DeclarationName</code> is designed to efficiently
				668	represent any kind of name.</p>
				669
				670	<p>Given
				671	a <code>DeclarationName</code> <code>N</code>, <code>N.getNameKind()</code>
Douglas Gregor	2def483	2008-11-17 20:34:05 +0000	[diff] [blame]	672	will produce a value that describes what kind of name <code>N</code>
Douglas Gregor	e94ca9e4	2008-11-18 14:39:36 +0000	[diff] [blame]	673	stores. There are 8 options (all of the names are inside
Douglas Gregor	2e1cd42	2008-11-17 14:58:09 +0000	[diff] [blame]	674	the <code>DeclarationName</code> class)</p>
				675	<dl>
				676	<dt>Identifier</dt>
				677	<dd>The name is a simple
				678	identifier. Use <code>N.getAsIdentifierInfo()</code> to retrieve the
				679	corresponding <code>IdentifierInfo*</code> pointing to the actual
				680	identifier. Note that C++ overloaded operators (e.g.,
				681	"<code>operator+</code>") are represented as special kinds of
				682	identifiers. Use <code>IdentifierInfo</code>'s <code>getOverloadedOperatorID</code>
				683	function to determine whether an identifier is an overloaded
				684	operator name.</dd>
				685
				686	<dt>ObjCZeroArgSelector, ObjCOneArgSelector,
				687	ObjCMultiArgSelector</dt>
				688	<dd>The name is an Objective-C selector, which can be retrieved as a
				689	<code>Selector</code> instance
				690	via <code>N.getObjCSelector()</code>. The three possible name
				691	kinds for Objective-C reflect an optimization within
				692	the <code>DeclarationName</code> class: both zero- and
				693	one-argument selectors are stored as a
				694	masked <code>IdentifierInfo</code> pointer, and therefore require
				695	very little space, since zero- and one-argument selectors are far
				696	more common than multi-argument selectors (which use a different
				697	structure).</dd>
				698
				699	<dt>CXXConstructorName</dt>
				700	<dd>The name is a C++ constructor
				701	name. Use <code>N.getCXXNameType()</code> to retrieve
				702	the <a href="#QualType">type</a> that this constructor is meant to
				703	construct. The type is always the canonical type, since all
				704	constructors for a given type have the same name.</dd>
				705
				706	<dt>CXXDestructorName</dt>
				707	<dd>The name is a C++ destructor
				708	name. Use <code>N.getCXXNameType()</code> to retrieve
				709	the <a href="#QualType">type</a> whose destructor is being
				710	named. This type is always a canonical type.</dd>
				711
				712	<dt>CXXConversionFunctionName</dt>
				713	<dd>The name is a C++ conversion function. Conversion functions are
				714	named according to the type they convert to, e.g., "<code>operator void
				715	const *</code>". Use <code>N.getCXXNameType()</code> to retrieve
				716	the type that this conversion function converts to. This type is
				717	always a canonical type.</dd>
Douglas Gregor	e94ca9e4	2008-11-18 14:39:36 +0000	[diff] [blame]	718
				719	<dt>CXXOperatorName</dt>
				720	<dd>The name is a C++ overloaded operator name. Overloaded operators
				721	are named according to their spelling, e.g.,
				722	"<code>operator+</code>" or "<code>operator new
				723	[]</code>". Use <code>N.getCXXOverloadedOperator()</code> to
				724	retrieve the overloaded operator (a value of
				725	type <code>OverloadedOperatorKind</code>).</dd>
Douglas Gregor	2e1cd42	2008-11-17 14:58:09 +0000	[diff] [blame]	726	</dl>
				727
				728	<p><code>DeclarationName</code>s are cheap to create, copy, and
				729	compare. They require only a single pointer's worth of storage in
Douglas Gregor	e94ca9e4	2008-11-18 14:39:36 +0000	[diff] [blame]	730	the common cases (identifiers, zero-
Douglas Gregor	2e1cd42	2008-11-17 14:58:09 +0000	[diff] [blame]	731	and one-argument Objective-C selectors) and use dense, uniqued
				732	storage for the other kinds of
				733	names. Two <code>DeclarationName</code>s can be compared for
				734	equality (<code>==</code>, <code>!=</code>) using a simple bitwise
				735	comparison, can be ordered
				736	with <code><</code>, <code>></code>, <code><=</code>,
				737	and <code>>=</code> (which provide a lexicographical ordering for
				738	normal identifiers but an unspecified ordering for other kinds of
				739	names), and can be placed into LLVM <code>DenseMap</code>s
				740	and <code>DenseSet</code>s.</p>
				741
				742	<p><code>DeclarationName</code> instances can be created in different
				743	ways depending on what kind of name the instance will store. Normal
Douglas Gregor	e94ca9e4	2008-11-18 14:39:36 +0000	[diff] [blame]	744	identifiers (<code>IdentifierInfo</code> pointers) and Objective-C selectors
Douglas Gregor	2e1cd42	2008-11-17 14:58:09 +0000	[diff] [blame]	745	(<code>Selector</code>) can be implicitly converted
				746	to <code>DeclarationName</code>s. Names for C++ constructors,
Douglas Gregor	e94ca9e4	2008-11-18 14:39:36 +0000	[diff] [blame]	747	destructors, conversion functions, and overloaded operators can be retrieved from
Douglas Gregor	2e1cd42	2008-11-17 14:58:09 +0000	[diff] [blame]	748	the <code>DeclarationNameTable</code>, an instance of which is
				749	available as <code>ASTContext::DeclarationNames</code>. The member
				750	functions <code>getCXXConstructorName</code>, <code>getCXXDestructorName</code>,
Douglas Gregor	e94ca9e4	2008-11-18 14:39:36 +0000	[diff] [blame]	751	<code>getCXXConversionFunctionName</code>, and <code>getCXXOperatorName</code>, respectively,
				752	return <code>DeclarationName</code> instances for the four kinds of
Douglas Gregor	2e1cd42	2008-11-17 14:58:09 +0000	[diff] [blame]	753	C++ special function names.</p>
				754
				755	<!-- ======================================================================= -->
Ted Kremenek	8bc0571	2007-10-10 23:01:43 +0000	[diff] [blame]	756	<h3 id="CFG">The <tt>CFG</tt> class</h3>
				757	<!-- ======================================================================= -->
				758
				759	<p>The <tt>CFG</tt> class is designed to represent a source-level
				760	control-flow graph for a single statement (<tt>Stmt*</tt>). Typically
				761	instances of <tt>CFG</tt> are constructed for function bodies (usually
				762	an instance of <tt>CompoundStmt</tt>), but can also be instantiated to
				763	represent the control-flow of any class that subclasses <tt>Stmt</tt>,
				764	which includes simple expressions. Control-flow graphs are especially
				765	useful for performing
				766	<a href="http://en.wikipedia.org/wiki/Data_flow_analysis#Sensitivities">flow-
				767	or path-sensitive</a> program analyses on a given function.</p>
				768
Chris Lattner	62fd278	2008-11-22 21:41:31 +0000	[diff] [blame^]	769	<!-- ============ -->
Ted Kremenek	8bc0571	2007-10-10 23:01:43 +0000	[diff] [blame]	770	<h4>Basic Blocks</h4>
Chris Lattner	62fd278	2008-11-22 21:41:31 +0000	[diff] [blame^]	771	<!-- ============ -->
Ted Kremenek	8bc0571	2007-10-10 23:01:43 +0000	[diff] [blame]	772
				773	<p>Concretely, an instance of <tt>CFG</tt> is a collection of basic
				774	blocks. Each basic block is an instance of <tt>CFGBlock</tt>, which
				775	simply contains an ordered sequence of <tt>Stmt*</tt> (each referring
				776	to statements in the AST). The ordering of statements within a block
				777	indicates unconditional flow of control from one statement to the
				778	next. <a href="#ConditionalControlFlow">Conditional control-flow</a>
				779	is represented using edges between basic blocks. The statements
				780	within a given <tt>CFGBlock</tt> can be traversed using
				781	the <tt>CFGBlock::*iterator</tt> interface.</p>
				782
				783	<p>
Ted Kremenek	18e17e7	2007-10-18 22:50:52 +0000	[diff] [blame]	784	A <tt>CFG</tt> object owns the instances of <tt>CFGBlock</tt> within
Ted Kremenek	8bc0571	2007-10-10 23:01:43 +0000	[diff] [blame]	785	the control-flow graph it represents. Each <tt>CFGBlock</tt> within a
				786	CFG is also uniquely numbered (accessible
				787	via <tt>CFGBlock::getBlockID()</tt>). Currently the number is
				788	based on the ordering the blocks were created, but no assumptions
				789	should be made on how <tt>CFGBlock</tt>s are numbered other than their
				790	numbers are unique and that they are numbered from 0..N-1 (where N is
				791	the number of basic blocks in the CFG).</p>
				792
Chris Lattner	62fd278	2008-11-22 21:41:31 +0000	[diff] [blame^]	793	<!-- ===================== -->
Ted Kremenek	8bc0571	2007-10-10 23:01:43 +0000	[diff] [blame]	794	<h4>Entry and Exit Blocks</h4>
Chris Lattner	62fd278	2008-11-22 21:41:31 +0000	[diff] [blame^]	795	<!-- ===================== -->
Ted Kremenek	8bc0571	2007-10-10 23:01:43 +0000	[diff] [blame]	796
				797	Each instance of <tt>CFG</tt> contains two special blocks:
				798	an <i>entry</i> block (accessible via <tt>CFG::getEntry()</tt>), which
				799	has no incoming edges, and an <i>exit</i> block (accessible
				800	via <tt>CFG::getExit()</tt>), which has no outgoing edges. Neither
				801	block contains any statements, and they serve the role of providing a
				802	clear entrance and exit for a body of code such as a function body.
				803	The presence of these empty blocks greatly simplifies the
				804	implementation of many analyses built on top of CFGs.
				805
Chris Lattner	62fd278	2008-11-22 21:41:31 +0000	[diff] [blame^]	806	<!-- ===================================================== -->
Ted Kremenek	8bc0571	2007-10-10 23:01:43 +0000	[diff] [blame]	807	<h4 id ="ConditionalControlFlow">Conditional Control-Flow</h4>
Chris Lattner	62fd278	2008-11-22 21:41:31 +0000	[diff] [blame^]	808	<!-- ===================================================== -->
Ted Kremenek	8bc0571	2007-10-10 23:01:43 +0000	[diff] [blame]	809
				810	<p>Conditional control-flow (such as those induced by if-statements
				811	and loops) is represented as edges between <tt>CFGBlock</tt>s.
				812	Because different C language constructs can induce control-flow,
				813	each <tt>CFGBlock</tt> also records an extra <tt>Stmt*</tt> that
				814	represents the <i>terminator</i> of the block. A terminator is simply
				815	the statement that caused the control-flow, and is used to identify
				816	the nature of the conditional control-flow between blocks. For
				817	example, in the case of an if-statement, the terminator refers to
				818	the <tt>IfStmt</tt> object in the AST that represented the given
				819	branch.</p>
				820
				821	<p>To illustrate, consider the following code example:</p>
				822
				823	<code>
				824	int foo(int x) {<br>
				825	x = x + 1;<br>
				826	<br>
				827	if (x > 2) x++;<br>
				828	else {<br>
				829	x += 2;<br>
				830	x *= 2;<br>
				831	}<br>
				832	<br>
				833	return x;<br>
				834	}
				835	</code>
				836
				837	<p>After invoking the parser+semantic analyzer on this code fragment,
				838	the AST of the body of <tt>foo</tt> is referenced by a
				839	single <tt>Stmt*</tt>. We can then construct an instance
				840	of <tt>CFG</tt> representing the control-flow graph of this function
				841	body by single call to a static class method:</p>
				842
				843	<code>
				844	Stmt* FooBody = ...<br>
				845	CFG* FooCFG = <b>CFG::buildCFG</b>(FooBody);
				846	</code>
				847
				848	<p>It is the responsibility of the caller of <tt>CFG::buildCFG</tt>
				849	to <tt>delete</tt> the returned <tt>CFG*</tt> when the CFG is no
				850	longer needed.</p>
				851
				852	<p>Along with providing an interface to iterate over
				853	its <tt>CFGBlock</tt>s, the <tt>CFG</tt> class also provides methods
				854	that are useful for debugging and visualizing CFGs. For example, the
				855	method
				856	<tt>CFG::dump()</tt> dumps a pretty-printed version of the CFG to
				857	standard error. This is especially useful when one is using a
				858	debugger such as gdb. For example, here is the output
				859	of <tt>FooCFG->dump()</tt>:</p>
				860
				861	<code>
				862	[ B5 (ENTRY) ]<br>
				863	Predecessors (0):<br>
				864	Successors (1): B4<br>
				865	<br>
				866	[ B4 ]<br>
				867	1: x = x + 1<br>
				868	2: (x > 2)<br>
				869	<b>T: if [B4.2]</b><br>
				870	Predecessors (1): B5<br>
				871	Successors (2): B3 B2<br>
				872	<br>
				873	[ B3 ]<br>
				874	1: x++<br>
				875	Predecessors (1): B4<br>
				876	Successors (1): B1<br>
				877	<br>
				878	[ B2 ]<br>
				879	1: x += 2<br>
				880	2: x *= 2<br>
				881	Predecessors (1): B4<br>
				882	Successors (1): B1<br>
				883	<br>
				884	[ B1 ]<br>
				885	1: return x;<br>
				886	Predecessors (2): B2 B3<br>
				887	Successors (1): B0<br>
				888	<br>
				889	[ B0 (EXIT) ]<br>
				890	Predecessors (1): B1<br>
				891	Successors (0):
				892	</code>
				893
				894	<p>For each block, the pretty-printed output displays for each block
				895	the number of <i>predecessor</i> blocks (blocks that have outgoing
				896	control-flow to the given block) and <i>successor</i> blocks (blocks
				897	that have control-flow that have incoming control-flow from the given
				898	block). We can also clearly see the special entry and exit blocks at
				899	the beginning and end of the pretty-printed output. For the entry
				900	block (block B5), the number of predecessor blocks is 0, while for the
				901	exit block (block B0) the number of successor blocks is 0.</p>
				902
				903	<p>The most interesting block here is B4, whose outgoing control-flow
				904	represents the branching caused by the sole if-statement
				905	in <tt>foo</tt>. Of particular interest is the second statement in
				906	the block, <b><tt>(x > 2)</tt></b>, and the terminator, printed
				907	as <b><tt>if [B4.2]</tt></b>. The second statement represents the
				908	evaluation of the condition of the if-statement, which occurs before
				909	the actual branching of control-flow. Within the <tt>CFGBlock</tt>
				910	for B4, the <tt>Stmt*</tt> for the second statement refers to the
				911	actual expression in the AST for <b><tt>(x > 2)</tt></b>. Thus
				912	pointers to subclasses of <tt>Expr</tt> can appear in the list of
				913	statements in a block, and not just subclasses of <tt>Stmt</tt> that
				914	refer to proper C statements.</p>
				915
				916	<p>The terminator of block B4 is a pointer to the <tt>IfStmt</tt>
				917	object in the AST. The pretty-printer outputs <b><tt>if
				918	[B4.2]</tt></b> because the condition expression of the if-statement
				919	has an actual place in the basic block, and thus the terminator is
				920	essentially
				921	<i>referring</i> to the expression that is the second statement of
				922	block B4 (i.e., B4.2). In this manner, conditions for control-flow
				923	(which also includes conditions for loops and switch statements) are
				924	hoisted into the actual basic block.</p>
				925
Chris Lattner	62fd278	2008-11-22 21:41:31 +0000	[diff] [blame^]	926	<!-- ===================== -->
				927	<!-- <h4>Implicit Control-Flow</h4> -->
				928	<!-- ===================== -->
Ted Kremenek	8bc0571	2007-10-10 23:01:43 +0000	[diff] [blame]	929
				930	<!--
				931	<p>A key design principle of the <tt>CFG</tt> class was to not require
				932	any transformations to the AST in order to represent control-flow.
				933	Thus the <tt>CFG</tt> does not perform any "lowering" of the
				934	statements in an AST: loops are not transformed into guarded gotos,
				935	short-circuit operations are not converted to a set of if-statements,
				936	and so on.</p>
				937	-->
Ted Kremenek	17a295d	2008-06-11 06:19:49 +0000	[diff] [blame]	938
Chris Lattner	7bad199	2008-11-16 21:48:07 +0000	[diff] [blame]	939
				940	<!-- ======================================================================= -->
				941	<h3 id="Constants">Constant Folding in the Clang AST</h3>
				942	<!-- ======================================================================= -->
				943
				944	<p>There are several places where constants and constant folding matter a lot to
				945	the Clang front-end. First, in general, we prefer the AST to retain the source
				946	code as close to how the user wrote it as possible. This means that if they
				947	wrote "5+4", we want to keep the addition and two constants in the AST, we don't
				948	want to fold to "9". This means that constant folding in various ways turns
				949	into a tree walk that needs to handle the various cases.</p>
				950
				951	<p>However, there are places in both C and C++ that require constants to be
				952	folded. For example, the C standard defines what an "integer constant
				953	expression" (i-c-e) is with very precise and specific requirements. The
				954	language then requires i-c-e's in a lot of places (for example, the size of a
				955	bitfield, the value for a case statement, etc). For these, we have to be able
				956	to constant fold the constants, to do semantic checks (e.g. verify bitfield size
				957	is non-negative and that case statements aren't duplicated). We aim for Clang
				958	to be very pedantic about this, diagnosing cases when the code does not use an
				959	i-c-e where one is required, but accepting the code unless running with
				960	<tt>-pedantic-errors</tt>.</p>
				961
				962	<p>Things get a little bit more tricky when it comes to compatibility with
				963	real-world source code. Specifically, GCC has historically accepted a huge
				964	superset of expressions as i-c-e's, and a lot of real world code depends on this
				965	unfortuate accident of history (including, e.g., the glibc system headers). GCC
				966	accepts anything its "fold" optimizer is capable of reducing to an integer
				967	constant, which means that the definition of what it accepts changes as its
				968	optimizer does. One example is that GCC accepts things like "case X-X:" even
				969	when X is a variable, because it can fold this to 0.</p>
				970
				971	<p>Another issue are how constants interact with the extensions we support, such
				972	as __builtin_constant_p, __builtin_inf, __extension__ and many others. C99
				973	obviously does not specify the semantics of any of these extensions, and the
				974	definition of i-c-e does not include them. However, these extensions are often
				975	used in real code, and we have to have a way to reason about them.</p>
				976
				977	<p>Finally, this is not just a problem for semantic analysis. The code
				978	generator and other clients have to be able to fold constants (e.g. to
				979	initialize global variables) and has to handle a superset of what C99 allows.
				980	Further, these clients can benefit from extended information. For example, we
				981	know that "foo()\|\|1" always evaluates to true, but we can't replace the
				982	expression with true because it has side effects.</p>
				983
				984	<!-- ======================= -->
				985	<h4>Implementation Approach</h4>
				986	<!-- ======================= -->
				987
				988	<p>After trying several different approaches, we've finally converged on a
				989	design (Note, at the time of this writing, not all of this has been implemented,
				990	consider this a design goal!). Our basic approach is to define a single
				991	recursive method evaluation method (<tt>Expr::Evaluate</tt>), which is
				992	implemented in <tt>AST/ExprConstant.cpp</tt>. Given an expression with 'scalar'
				993	type (integer, fp, complex, or pointer) this method returns the following
				994	information:</p>
				995
				996	<ul>
				997	<li>Whether the expression is an integer constant expression, a general
				998	constant that was folded but has no side effects, a general constant that
				999	was folded but that does have side effects, or an uncomputable/unfoldable
				1000	value.
				1001	</li>
				1002	<li>If the expression was computable in any way, this method returns the APValue
				1003	for the result of the expression.</li>
				1004	<li>If the expression is not evaluatable at all, this method returns
				1005	information on one of the problems with the expression. This includes a
				1006	SourceLocation for where the problem is, and a diagnostic ID that explains
				1007	the problem. The diagnostic should be have ERROR type.</li>
				1008	<li>If the expression is not an integer constant expression, this method returns
				1009	information on one of the problems with the expression. This includes a
				1010	SourceLocation for where the problem is, and a diagnostic ID that explains
				1011	the problem. The diagnostic should be have EXTENSION type.</li>
				1012	</ul>
				1013
				1014	<p>This information gives various clients the flexibility that they want, and we
				1015	will eventually have some helper methods for various extensions. For example,
				1016	Sema should have a <tt>Sema::VerifyIntegerConstantExpression</tt> method, which
				1017	calls Evaluate on the expression. If the expression is not foldable, the error
				1018	is emitted, and it would return true. If the expression is not an i-c-e, the
				1019	EXTENSION diagnostic is emitted. Finally it would return false to indicate that
				1020	the AST is ok.</p>
				1021
				1022	<p>Other clients can use the information in other ways, for example, codegen can
				1023	just use expressions that are foldable in any way.</p>
				1024
				1025	<!-- ========== -->
				1026	<h4>Extensions</h4>
				1027	<!-- ========== -->
				1028
				1029	<p>This section describes how some of the various extensions clang supports
				1030	interacts with constant evaluation:</p>
				1031
				1032	<ul>
				1033	<li><b><tt>__extension__</tt></b>: The expression form of this extension causes
				1034	any evaluatable subexpression to be accepted as an integer constant
				1035	expression.</li>
				1036	<li><b><tt>__builtin_constant_p</tt></b>: This returns true (as a integer
				1037	constant expression) if the operand is any evaluatable constant.</li>
				1038	<li><b><tt>__builtin_choose_expr</tt></b>: The condition is required to be an
				1039	integer constant expression, but we accept any constant as an "extension of
				1040	an extension". This only evaluates one operand depending on which way the
				1041	condition evaluates.</li>
				1042	<li><b><tt>__builtin_classify_type</tt></b>: This always returns an integer
				1043	constant expression.</li>
				1044	<li><b><tt>__builtin_inf,nan,..</tt></b>: These are treated just like a
				1045	floating-point literal.</li>
				1046	<li><b><tt>__builtin_abs,copysign,..</tt></b>: These are constant folded as
				1047	general constant expressions.</li>
				1048	</ul>
				1049
				1050
				1051
				1052
Ted Kremenek	17a295d	2008-06-11 06:19:49 +0000	[diff] [blame]	1053	</div>
				1054	</body>
Douglas Gregor	2e1cd42	2008-11-17 14:58:09 +0000	[diff] [blame]	1055	</html>