| blob | 0720329cbd82ccd91397674f76bcb0d467e598b4 |
1 <html>
2 <head>
7 td {
9 }
10 </style>
11 </head>
12 <body>
14 <!--#include virtual="../menu.html.incl"-->
20 <ul>
24 <ul>
27 classes</a></li>
29 </ul>
30 </li>
32 <ul>
33 </ul>
34 </li>
37 <ul>
38 </ul>
39 </li>
41 <ul>
47 </ul>
48 </li>
50 <ul>
51 </ul>
52 </li>
54 <ul>
59 <ul>
62 Contexts</a></li>
65 </ul>
66 </li>
69 </ul>
70 </li>
72 </ul>
75 <!-- ======================================================================= -->
77 <!-- ======================================================================= -->
79 <p>This document describes some of the more important APIs and internal design
80 decisions made in the Clang C front-end. The purpose of this document is to
81 both capture some of this high level information and also describe some of the
82 design decisions behind it. This is meant for people interested in hacking on
83 Clang, not for end-users. The description below is categorized by
84 libraries, and does not describe any of the clients of the libraries.</p>
86 <!-- ======================================================================= -->
88 <!-- ======================================================================= -->
90 <p>The LLVM libsystem library provides the basic Clang system abstraction layer,
91 which is used for file system access. The LLVM libsupport library provides many
92 underlying libraries and <a
94 including command line option
95 processing and various containers.</p>
97 <!-- ======================================================================= -->
99 <!-- ======================================================================= -->
101 <p>This library certainly needs a better name. The 'basic' library contains a
102 number of low-level utilities for tracking and manipulating source buffers,
103 locations within the source buffers, diagnostics, tokens, target abstraction,
104 and information about the subset of the language being compiled for.</p>
106 <p>Part of this infrastructure is specific to C (such as the TargetInfo class),
107 other parts could be reused for other non-C-based languages (SourceLocation,
108 SourceManager, Diagnostics, FileManager). When and if there is future demand
109 we can figure out if it makes sense to introduce a new library, move the general
110 classes somewhere else, or introduce some other solution.</p>
115 <!-- ======================================================================= -->
117 <!-- ======================================================================= -->
119 <p>The Clang Diagnostics subsystem is an important part of how the compiler
120 communicates with the human. Diagnostics are the warnings and errors produced
121 when the code is incorrect or dubious. In Clang, each diagnostic produced has
122 (at the minimum) a unique ID, an English translation associated with it, a <a
125 of arguments to the dianostic (which fill in "%0"'s in the string) as well as a
126 number of source ranges that related to the diagnostic.</p>
128 <p>In this section, we'll be giving examples produced by the Clang command line
130 different ways</a> depending on how the DiagnosticClient interface is
131 implemented. A representative example of a diagnostic is:</p>
133 <pre>
137 </pre>
139 <p>In this example, you can see the English translation, the severity (error),
140 you can see the source location (the caret ("^") and file/line/column info),
142 float"). You'll have to believe me that there is a unique ID backing the
143 diagnostic :).</p>
145 <p>Getting all of this to happen has several steps and involves many moving
146 pieces, this section describes them and talks about best practices when adding
147 a new diagnostic.</p>
149 <!-- ============================== -->
151 <!-- ============================== -->
154 clang/Basic/Diagnostic*Kinds.def</tt> files, depending on what library will
155 be using it. This file encodes the unique ID of the
156 diagnostic (as an enum, the first argument), the severity of the diagnostic
157 (second argument) and the English translation + format string.</p>
159 <p>There is little sanity with the naming of the unique ID's right now. Some
160 start with err_, warn_, ext_ to encode the severity into the name. Since the
161 enum is referenced in the C++ code that produces the diagnostic, it is somewhat
162 useful for it to be reasonably short.</p>
167 acceptable under any circumstances. When an error is emitted, the AST for the
169 severities are used for extensions to the language that Clang accepts. This
170 means that Clang fully understands and can represent them in the AST, but we
171 produce diagnostics to tell the user their code is non-portable. The difference
172 is that the former are ignored by default, and the later warn by default. The
175 level is used to staple more information onto previous diagnostics.</p>
180 subsystem based on various configuration options. Clang internally supports a
181 fully fine grained mapping mechanism that allows you to map almost any
182 diagnostic to the output level that you want. The only diagnostics that cannot
186 for example).</p>
188 <p>Diagnostic mappings are used in many ways. For example, if the user
192 </p>
194 <p>
196 considered so severe that error recovery won't be able to recover sensibly from
197 them (thus spewing a ton of bogus errors). One example of this class of error
198 are failure to #include a file.
199 </p>
201 <!-- ================= -->
203 <!-- ================= -->
205 <p>The format string for the diagnostic is very simple, but it has some power.
206 It takes the form of a string in English with markers that indicate where and
207 how arguments to the diagnostic are inserted and formatted. For example, here
208 are some simple format strings:</p>
210 <pre>
211 "binary integer literals are an extension"
212 "format string contains '\\0' within the string body"
213 "more '<b>%%</b>' conversions than data arguments"
214 "invalid operands to binary expression (<b>%0</b> and <b>%1</b>)"
215 "overloaded '<b>%0</b>' must be a <b>%select{unary|binary|unary or binary}2</b> operator"
216 " (has <b>%1</b> parameter<b>%s1</b>)"
217 </pre>
219 <p>These examples show some important points of format strings. You can use any
220 plain ASCII character in the diagnostic string except "%" without a problem,
221 but these are C strings, so you have to use and be aware of all the C escape
222 sequences (as in the second example). If you want to produce a "%" in the
223 output, use the "%%" escape sequence, like the third diagnostic. Finally,
224 Clang uses the "%...[digit]" sequences to specify where and how arguments to
225 the diagnostic are formatted.</p>
227 <p>Arguments to the diagnostic are numbered according to how they are specified
230 to your diagnostic, you are doing something wrong :). Unlike printf, there
231 is no requirement that arguments to the diagnostic end up in the output in
232 the same order as they are specified, you could have a format string with
234 percent and digit are formatting instructions. If there are no instructions,
235 the argument is just turned into a string and substituted in.</p>
239 <ul>
242 printed, and forces you to think about the important point you are conveying
243 with the diagnostic.</li>
244 <li>Take advantage of location information. The user will be able to see the
245 line and location of the caret, so you don't need to tell them that the
247 <li>Do not capitalize the diagnostic string, and do not end it with a
248 period.</li>
249 <li>If you need to quote something in the diagnostic string, use single
250 quotes.</li>
251 </ul>
253 <p>Diagnostics should never take random English strings as arguments: you
257 other languages (because they'll get random English words in their otherwise
258 localized diagnostic). The exceptions to this are C/C++ language keywords
262 including variable names, types, labels, etc. The 'select' format can be
263 used to achieve this sort of thing in a localizable way, see below.</p>
265 <!-- ==================================== -->
267 <!-- ==================================== -->
269 <p>Arguments to diagnostics are fully typed internally, and come from a couple
270 different classes: integers, types, names, and random strings. Depending on
271 the class of the argument, it can be optionally formatted in different ways.
272 This gives the DiagnosticClient information about what the argument means
273 without requiring it to use a specific presentation (consider this MVC for
274 Clang :).</p>
276 <p>Here are the different diagnostic argument formats currently supported by
277 Clang:</p>
279 <table>
284 useful when producing English diagnostics. When the integer is 1, it prints
286 simple grammatical forms to be to be handled correctly, and eliminates the
294 related diagnostics together into one common one, without requiring the
295 difference to be specified as an English string argument. Instead of
296 specifying the string, the diagnostic gets an integer argument and the
297 format string selects the numbered option. In this case, the "%2" value
300 allows other language translations to substitute reasonable words (or entire
301 phrases) based on the semantics of the diagnostic instead of having to do
302 things textually.</p>
310 It is designed to handle even the requirements of languages with very
311 complex plural forms, as many Baltic languages have. The argument consists
312 of a series of expression/form pairs, separated by ':', where the first form
313 whose expression evaluates to true is the result of the modifier.</p>
314 <p>An expression can be empty, in which case it is always true. See the
315 example at the top. Otherwise, it is a series of one or more numeric
316 conditions, separated by ','. If any condition matches, the expression
317 matches. Each numeric condition can take one of three forms.</p>
318 <ul>
319 <li>number: A simple decimal number matches if the argument is the same
321 <li>range: A range in square brackets matches if the argument is within
322 the range. Then range is inclusive on both ends. Example:
324 <li>modulo: A modulo operator is followed by a number, and
325 equals sign and either a number or a range. The tests are the
326 same as for plain
327 numbers and ranges, but the argument is taken modulo the number first.
330 </ul>
331 <p>The parser is very unforgiving. A syntax error, even whitespace, will
332 abort, as will a failure to match the argument against any
333 expression.</p></td></tr>
341 are not supported.</p>
348 DeclarationName corresponds to an Objective-C class method selector. As
349 such, it prints the selector with a leading '+'.</p></td></tr>
355 DeclarationName corresponds to an Objective-C instance method selector. As
356 such, it prints the selector with a leading '-'.</p></td></tr>
361 <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>
363 </table>
365 <p>It is really easy to add format specifiers to the Clang diagnostics system,
366 but they should be discussed before they are added. If you are creating a lot
367 of repetitive diagnostics and/or have an idea for a useful formatter, please
368 bring it up on the cfe-dev mailing list.</p>
370 <!-- ===================================================== -->
372 <!-- ===================================================== -->
374 <p>Now that you've created the diagnostic in the DiagnosticKinds.def file, you
375 need to write the code that detects the condition in question and emits the
376 new diagnostic. Various components of Clang (e.g. the preprocessor, Sema,
377 etc) provide a helper function named "Diag". It creates a diagnostic and
378 accepts the arguments, ranges, and other information that goes along with
379 it.</p>
383 <pre>
384 if (various things that are bad)
385 Diag(Loc, diag::err_typecheck_invalid_operands)
388 </pre>
392 (which matches the name from DiagnosticKinds.def). If the diagnostic takes
393 arguments, they are specified with the << operator: the first argument
399 SourceRanges are also specified with the << operator, but do not have a
400 specific ordering requirement.</p>
402 <p>As you can see, adding and producing a diagnostic is pretty straightforward.
403 The hard part is deciding exactly what you need to say to help the user, picking
404 a suitable wording, and providing the information needed to format it correctly.
405 The good news is that the call site that issues a diagnostic should be
406 completely independent of how the diagnostic is formatted and in what language
407 it is rendered.
408 </p>
410 <!-- ==================================================== -->
412 <!-- ==================================================== -->
414 <p>In some cases, the front end emits diagnostics when it is clear
415 that some small change to the source code would fix the problem. For
416 example, a missing semicolon at the end of a statement or a use of
417 deprecated syntax that is easily rewritten into a more modern form.
418 Clang tries very hard to emit the diagnostic and recover gracefully
419 in these and other cases.</p>
421 <p>However, for these cases where the fix is obvious, the diagnostic
422 can be annotated with a code
423 modification "hint" that describes how to change the code referenced
424 by the diagnostic to fix the problem. For example, it might add the
425 missing semicolon at the end of the statement or rewrite the use of a
426 deprecated construct into something more palatable. Here is one such
427 example C++ front end, where we warn about the right-shift operator
430 <pre>
431 test.cpp:3:7: warning: use of right-shift operator ('>>') in template argument will require parentheses in C++0x
433 ^
434 ( )
435 </pre>
437 <p>Here, the code modification hint is suggesting that parentheses be
438 added, and showing exactly where those parentheses would be inserted
439 into the source code. The code modification hints themselves describe
440 what changes to make to the source code in an abstract manner, which
441 the text diagnostic printer renders as a line of "insertions" below
443 clients</a> might choose to render the code differently (e.g., as
444 markup inline) or even give the user the ability to automatically fix
445 the problem.</p>
447 <p>All code modification hints are described by the
449 attached to the diagnostic using the << operator in the same way
450 that highlighted source ranges and arguments are passed to the
451 diagnostic. Code modification hints can be created with one of three
452 constructors:</p>
454 <dl>
461 should be removed.</dd>
466 </dl>
468 <!-- ============================================================= -->
470 <!-- ============================================================= -->
472 <p>Once code generates a diagnostic with all of the arguments and the rest of
473 the relevant information, Clang needs to know what to do with it. As previously
474 mentioned, the diagnostic machinery goes through some filtering to map a
475 severity onto a diagnostic level, then (assuming the diagnostic is not mapped to
476 "<tt>Ignore</tt>") it invokes an object that implements the DiagnosticClient
477 interface with the information.</p>
479 <p>It is possible to implement this interface in many different ways. For
480 example, the normal Clang DiagnosticClient (named 'TextDiagnosticPrinter') turns
481 the arguments into strings (according to the various formatting rules), prints
482 out the file/line/column information and the string, then prints out the line of
483 code, the source ranges, and the caret. However, this behavior isn't required.
484 </p>
486 <p>Another implementation of the DiagnosticClient interface is the
487 'TextDiagnosticBuffer' class, which is used when Clang is in -verify mode.
488 Instead of formatting and printing out the diagnostics, this implementation just
489 captures and remembers the diagnostics as they fly by. Then -verify compares
490 the list of produced diagnostics to the list of expected ones. If they disagree,
491 it prints out its own output.
492 </p>
494 <p>There are many other possible implementations of this interface, and this is
495 why we prefer diagnostics to pass down rich structured information in arguments.
496 For example, an HTML output might want declaration names be linkified to where
497 they come from in the source. Another example is that a GUI might let you click
498 on typedefs to expand them. This application would want to pass significantly
499 more information about types through to the GUI than a simple flat string. The
500 interface allows this to happen.</p>
502 <!-- ====================================================== -->
504 <!-- ====================================================== -->
507 can translate to the relevant code page if needed. Each translation completely
508 replaces the format string for the diagnostic.</p>
511 <!-- ======================================================================= -->
513 <!-- ======================================================================= -->
515 <p>Strangely enough, the SourceLocation class represents a location within the
516 source code of the program. Important design points include:</p>
518 <ol>
519 <li>sizeof(SourceLocation) must be extremely small, as these are embedded into
521 <li>SourceLocation must be a simple value object that can be efficiently
522 copied.</li>
523 <li>We should be able to represent a source location for any byte of any input
524 file. This includes in the middle of tokens, in whitespace, in trigraphs,
525 etc.</li>
526 <li>A SourceLocation must encode the current #include stack that was active when
527 the location was processed. For example, if the location corresponds to a
528 token, it should contain the set of #includes active when the token was
529 lexed. This allows us to print the #include stack for a diagnostic.</li>
530 <li>SourceLocation must be able to describe macro expansions, capturing both
531 the ultimate instantiation point and the source of the original character
532 data.</li>
533 </ol>
535 <p>In practice, the SourceLocation works together with the SourceManager class
536 to encode two pieces of information about a location: its spelling location
537 and its instantiation location. For most tokens, these will be the same.
538 However, for a macro expansion (or tokens that came from a _Pragma directive)
539 these will describe the location of the characters corresponding to the token
540 and the location where the token was used (i.e. the macro instantiation point
541 or the location of the _Pragma itself).</p>
543 <p>The Clang front-end inherently depends on the location of a token being
544 tracked correctly. If it is ever incorrect, the front-end may get confused and
545 die. The reason for this is that the notion of the 'spelling' of a Token in
546 Clang depends on being able to find the original input characters for the token.
550 <!-- ======================================================================= -->
552 <!-- ======================================================================= -->
553 <!-- mostly taken from
554 http://lists.cs.uiuc.edu/pipermail/cfe-dev/2010-August/010595.html -->
556 <p>Clang represents most source ranges by [first, last], where first and last
557 each point to the beginning of their respective tokens. For example
558 consider the SourceRange of the following statement:</p>
559 <pre>
560 x = foo + bar;
561 ^first ^last
562 </pre>
564 <p>To map from this representation to a character-based
565 representation, the 'last' location needs to be adjusted to point to
566 (or past) the end of that token with either
569 where character-level source ranges information is needed we use
573 <!-- ======================================================================= -->
575 <!-- ======================================================================= -->
580 <!-- ======================================================================= -->
582 <!-- ======================================================================= -->
584 <p>Clang supports two implementations of precompiled headers. The
585 default implementation, precompiled headers (<a
587 of Clang's internal data structures, encoded with the <a
591 serialized representation of the tokens encountered when
592 preprocessing a header (and anything that header includes).</p>
595 <!-- ======================================================================= -->
597 <!-- ======================================================================= -->
599 <p>The Frontend library contains functionality useful for building
600 tools on top of the clang libraries, for example several methods for
601 outputting diagnostics.</p>
603 <!-- ======================================================================= -->
605 <!-- ======================================================================= -->
607 <p>The Lexer library contains several tightly-connected classes that are involved
608 with the nasty process of lexing and preprocessing C source code. The main
609 interface to this library for outside clients is the large <a
611 It contains the various pieces of state that are required to coherently read
612 tokens out of a translation unit.</p>
614 <p>The core interface to the Preprocessor object (once it is set up) is the
616 the preprocessor stream. There are two types of token providers that the
617 preprocessor is capable of reading from: a buffer lexer (provided by the <a
622 <!-- ======================================================================= -->
624 <!-- ======================================================================= -->
626 <p>The Token class is used to represent a single lexed token. Tokens are
627 intended to be used by the lexer/preprocess and parser libraries, but are not
628 intended to live beyond them (for example, they should not live in the ASTs).<p>
630 <p>Tokens most often live on the stack (or some other location that is efficient
631 to access) as the parser is running, but occasionally do get buffered up. For
632 example, macro definitions are stored as a series of tokens, and the C++
633 front-end periodically needs to buffer tokens up for tentative parsing and
634 various pieces of look-ahead. As such, the size of a Token matter. On a 32-bit
638 Tokens</a>" and normal tokens. Normal tokens are those returned by the lexer,
639 annotation tokens represent semantic information and are produced by the parser,
640 replacing normal tokens in the token stream. Normal tokens contain the
641 following information:</p>
643 <ul>
645 token.</li>
648 SourceBuffer. For tokens that include them, this length includes trigraphs and
649 escaped newlines which are ignored by later phases of the compiler. By pointing
650 into the original source buffer, it is always possible to get the original
651 spelling of a token completely accurately.</li>
654 identifier lookup was enabled when the token was lexed (e.g. the lexer was not
655 reading in 'raw' mode) this contains a pointer to the unique hash value for the
656 identifier. Because the lookup happens before keyword identification, this
657 field is set even for language keywords like 'for'.</li>
663 that some tokens can be spelled multiple ways. For example, C++ supports
665 "&&" operator. In these cases, the kind value is set to
667 consider both forms. For something that cares about which form is used (e.g.
668 the preprocessor 'stringize' operator) the spelling indicates the original
669 form.</li>
672 lexer/preprocessor system on a per-token basis:
674 <ol>
676 source line.</li>
678 before the token or transitively before the token as it was expanded
679 through a macro. The definition of this flag is very closely defined by
680 the stringizing requirements of the preprocessor.</li>
682 represent identifier tokens which have macro expansion disabled. This
683 prevents them from being considered as candidates for macro expansion ever
684 in the future.</li>
686 token includes a trigraph or escaped newline. Since this is uncommon,
687 many pieces of code can fast-path on tokens that did not need cleaning.
688 </p>
689 </ol>
690 </li>
691 </ul>
693 <p>One interesting (and somewhat unusual) aspect of normal tokens is that they
694 don't contain any semantic information about the lexed value. For example, if
695 the token was a pp-number token, we do not represent the value of the number
696 that was lexed (this is left for later pieces of code to decide). Additionally,
697 the lexer library has no notion of typedef names vs variable names: both are
698 returned as identifiers, and the parser is left to decide whether a specific
699 identifier is a typedef or a variable (tracking this requires scope information
700 among other things). The parser can do this translation by replacing tokens
703 <!-- ======================================================================= -->
705 <!-- ======================================================================= -->
707 <p>Annotation Tokens are tokens that are synthesized by the parser and injected
708 into the preprocessor's token stream (replacing existing tokens) to record
709 semantic information found by the parser. For example, if "foo" is found to be
712 makes it easy to handle qualified type names (e.g. "foo::bar::baz<42>::t")
714 reparse does not need to redo semantic analysis to determine whether a token
715 sequence is a variable, type, template, etc.</p>
717 <p>Annotation Tokens are created by the parser and reinjected into the parser's
718 token stream (when backtracking is enabled). Because they can only exist in
719 tokens that the preprocessor-proper is done with, it doesn't need to keep around
720 flags like "start of line" that the preprocessor uses to do its job.
721 Additionally, an annotation token may "cover" a sequence of preprocessor tokens
723 of an annotation token are different than the fields for a normal token (but
724 they are multiplexed into the normal Token fields):</p>
726 <ul>
728 token indicates the first token replaced by the annotation token. In the example
732 last token replaced with the annotation token. In the example above, it would
736 that the parser gets from Sema. The parser merely preserves the
737 information for Sema to later interpret based on the annotation token
738 kind.</li>
741 is. See below for the different valid kinds.</li>
742 </ul>
746 <ol>
748 resolved typename token that is potentially qualified. The
750 Sema::getTypeName(), possibly with source location information
751 attached.</li>
754 scope specifier, such as "A::B::". This corresponds to the grammar
757 the Sema::ActOnCXXGlobalScopeSpecifier and
758 Sema::ActOnCXXNestedNameSpecifier callbacks.</li>
762 of a template. The AnnotationValue pointer is a pointer to a malloc'd
763 TemplateIdAnnotation object. Depending on the context, a parsed
764 template-id that names a type might become a typename annotation token
765 (if all we care about is the named type, e.g., because it occurs in a
766 type specifier) or might remain a template-id token (if we want to
767 retain more source location information or produce a new type, e.g.,
768 in a declaration of a class template specialization). template-id
769 annotation tokens that refer to a type can be "upgraded" to typename
770 annotation tokens by the parser.</li>
772 </ol>
774 <p>As mentioned above, annotation tokens are not returned by the preprocessor,
775 they are formed on demand by the parser. This means that the parser has to be
776 aware of cases where an annotation could occur and form it where appropriate.
777 This is somewhat similar to how the parser handles Translation Phase 6 of C99:
778 String Concatenation (see C99 5.1.1.2). In the case of string concatenation,
779 the preprocessor just returns distinct tok::string_literal and
780 tok::wide_string_literal tokens and the parser eats a sequence of them wherever
781 the grammar indicates that a string literal can occur.</p>
783 <p>In order to do this, whenever the parser expects a tok::identifier or
784 tok::coloncolon, it should call the TryAnnotateTypeOrScopeToken or
785 TryAnnotateCXXScopeToken methods to form the annotation token. These methods
786 will maximally form the specified annotation tokens and replace the current
787 token with them, if applicable. If the current tokens is not valid for an
788 annotation token, it will remain an identifier or :: token.</p>
792 <!-- ======================================================================= -->
794 <!-- ======================================================================= -->
796 <p>The Lexer class provides the mechanics of lexing tokens out of a source
797 buffer and deciding what they mean. The Lexer is complicated by the fact that
798 it operates on raw buffers that have not had spelling eliminated (this is a
799 necessity to get decent performance), but this is countered with careful coding
800 as well as standard performance techniques (for example, the comment handling
801 code is vectorized on X86 and PowerPC hosts).</p>
805 <ul>
806 <li>The lexer can operate in 'raw' mode. This mode has several features that
807 make it possible to quickly lex the file (e.g. it stops identifier lookup,
808 doesn't specially handle preprocessor tokens, handles EOF differently, etc).
809 This mode is used for lexing within an "<tt>#if 0</tt>" block, for
810 example.</li>
811 <li>The lexer can capture and return comments as tokens. This is required to
812 support the -C preprocessor mode, which passes comments through, and is
813 used by the diagnostic checker to identifier expect-error annotations.</li>
814 <li>The lexer can be in ParsingFilename mode, which happens when preprocessing
815 after reading a #include directive. This mode changes the parsing of '<'
816 to return an "angled string" instead of a bunch of tokens for each thing
817 within the filename.</li>
819 ParsingPreprocessorDirective mode is entered. This changes the parser to
820 return EOM at a newline.</li>
821 <li>The Lexer uses a LangOptions object to know whether trigraphs are enabled,
822 whether C++ or ObjC keywords are recognized, etc.</li>
823 </ul>
825 <p>In addition to these modes, the lexer keeps track of a couple of other
826 features that are local to a lexed buffer, which change as the buffer is
827 lexed:</p>
829 <ul>
830 <li>The Lexer uses BufferPtr to keep track of the current character being
831 lexed.</li>
832 <li>The Lexer uses IsAtStartOfLine to keep track of whether the next lexed token
834 <li>The Lexer keeps track of the current #if directives that are active (which
835 can be nested).</li>
837 MultipleIncludeOpt</a> object, which is used to
838 detect whether the buffer uses the standard "<tt>#ifndef XX</tt> /
839 <tt>#define XX</tt>" idiom to prevent multiple inclusion. If a buffer does,
840 subsequent includes can be ignored if the XX macro is defined.</li>
841 </ul>
843 <!-- ======================================================================= -->
845 <!-- ======================================================================= -->
847 <p>The TokenLexer class is a token provider that returns tokens from a list
848 of tokens that came from somewhere else. It typically used for two things: 1)
849 returning tokens from a macro definition as it is being expanded 2) returning
850 tokens from an arbitrary buffer of tokens. The later use is used by _Pragma and
851 will most likely be used to handle unbounded look-ahead for the C++ parser.</p>
853 <!-- ======================================================================= -->
855 <!-- ======================================================================= -->
857 <p>The MultipleIncludeOpt class implements a really simple little state machine
858 that is used to detect the standard "<tt>#ifndef XX</tt> / <tt>#define XX</tt>"
859 idiom that people typically use to prevent multiple inclusion of headers. If a
860 buffer uses this idiom and is subsequently #include'd, the preprocessor can
861 simply check to see whether the guarding condition is defined or not. If so,
862 the preprocessor can completely ignore the include of the header.</p>
866 <!-- ======================================================================= -->
868 <!-- ======================================================================= -->
870 <!-- ======================================================================= -->
872 <!-- ======================================================================= -->
874 <!-- ======================================================================= -->
876 <!-- ======================================================================= -->
878 <p>The Type class (and its subclasses) are an important part of the AST. Types
879 are accessed through the ASTContext class, which implicitly creates and uniques
880 them as they are needed. Types have a couple of non-obvious features: 1) they
881 do not capture type qualifiers like const or volatile (See
883 information. Once created, types are immutable (unlike decls).</p>
885 <p>Typedefs in C make semantic analysis a bit more complex than it would
886 be without them. The issue is that we want to capture typedef information
887 and represent it in the AST perfectly, but the semantics of operations need to
890 <code>
891 void func() {<br>
899 }<br>
900 </code>
902 <p>The code above is illegal, and thus we expect there to be diagnostics emitted
903 on the annotated lines. In this example, we expect to get:</p>
905 <pre>
907 *X; // error
910 **Y; // error
913 **Z; // error
915 </pre>
917 <p>While this example is somewhat silly, it illustrates the point: we want to
918 retain typedef information where possible, so that we can emit errors about
920 Doing this requires properly keeping typedef information (for example, the type
923 to retain this information, the type of these expressions is an instance of the
924 TypedefType class, which indicates that the type of these expressions is a
925 typedef for foo.
926 </p>
928 <p>Representing types like this is great for diagnostics, because the
929 user-specified type is always immediately available. There are two problems
930 with this: first, various semantic checks need to make judgements about the
932 efficient way to query whether two types are structurally identical to each
933 other, ignoring typedefs. The solution to both of these problems is the idea of
934 canonical types.</p>
936 <!-- =============== -->
938 <!-- =============== -->
940 <p>Every instance of the Type class contains a canonical type pointer. For
942 "<tt>int**</tt>"), the type just points to itself. For types that have a
945 structurally equivalent type without any typedefs (e.g. "<tt>int</tt>",
948 <p>This design provides a constant time operation (dereferencing the canonical
949 type pointer) that gives us access to the structure of types. For example,
951 their canonical type pointers and doing a pointer comparison (they both point
954 <p>Canonical types and typedef types bring up some complexities that must be
955 carefully managed. Specifically, the "isa/cast/dyncast" operators generally
956 shouldn't be used in code that is inspecting the AST. For example, when type
957 checking the indirection operator (unary '*' on a pointer), the type checker
958 must verify that the operand has a pointer type. It would not be correct to
959 check that with "<tt>isa<PointerType>(SubExpr->getType())</tt>",
960 because this predicate would fail if the subexpression had a typedef type.</p>
962 <p>The solution to this problem are a set of helper methods on Type, used to
963 check their properties. In this case, it would be correct to use
964 "<tt>SubExpr->getType()->isPointerType()</tt>" to do the check. This
966 true any time the type is structurally a pointer type. The only hard part here
969 <p>The second problem we face is how to get access to the pointer type once we
970 know it exists. To continue the example, the result type of the indirection
971 operator is the pointee type of the subexpression. In order to determine the
972 type, we need to get the instance of PointerType that best captures the typedef
973 information in the program. If the type of the expression is literally a
974 PointerType, we can return that, otherwise we have to dig through the
975 typedefs to find the pointer type. For example, if the subexpression had type
976 "<tt>foo*</tt>", we could return that type as the result. If the subexpression
979 a getAsPointerType() method that checks whether the type is structurally a
980 PointerType and, if so, returns the best one. If not, it returns a null
981 pointer.</p>
983 <p>This structure is somewhat mystical, but after meditating on it, it will
984 make sense to you :).</p>
986 <!-- ======================================================================= -->
988 <!-- ======================================================================= -->
990 <p>The QualType class is designed as a trivial value class that is
991 small, passed by-value and is efficient to query. The idea of
992 QualType is that it stores the type qualifiers (const, volatile,
993 restrict, plus some extended qualifiers required by language
994 extensions) separately from the types themselves. QualType is
997 <p>By storing the type qualifiers as bits in the conceptual pair, it is
998 extremely efficient to get the set of qualifiers on a QualType (just return the
999 field of the pair), add a type qualifier (which is a trivial constant-time
1000 operation that sets a bit), and remove one or more type qualifiers (just return
1001 a QualType with the bitfield set to empty).</p>
1003 <p>Further, because the bits are stored outside of the type itself, we do not
1004 need to create duplicates of types with different sets of qualifiers (i.e. there
1006 both point to the same heap allocated "int" type). This reduces the heap size
1007 used to represent bits and also means we do not have to consider qualifiers when
1010 <p>In practice, the two most common type qualifiers (const and
1011 restrict) are stored in the low bits of the pointer to the Type
1012 object, together with a flag indicating whether extended qualifiers
1013 are present (which must be heap-allocated). This means that QualType
1014 is exactly the same size as a pointer.</p>
1016 <!-- ======================================================================= -->
1018 <!-- ======================================================================= -->
1021 declaration in Clang. Declarations in the C family of languages can
1022 take several different forms. Most declarations are named by
1025 names can also name class constructors ("<code>Class</code>"
1028 and conversion functions ("<code>operator void const *</code>"). In
1029 Objective-C, declaration names can refer to the names of Objective-C
1030 methods, which involve the method name and the parameters,
1032 "<code>setWidth:height:</code>". Since all of these kinds of
1033 entities - variables, functions, Objective-C methods, C++
1034 constructors, destructors, and operators - are represented as
1037 represent any kind of name.</p>
1039 <p>Given
1042 stores. There are 8 options (all of the names are inside
1044 <dl>
1046 <dd>The name is a simple
1049 identifier. Note that C++ overloaded operators (e.g.,
1050 "<code>operator+</code>") are represented as special kinds of
1052 function to determine whether an identifier is an overloaded
1053 operator name.</dd>
1055 <dt>ObjCZeroArgSelector, ObjCOneArgSelector,
1056 ObjCMultiArgSelector</dt>
1057 <dd>The name is an Objective-C selector, which can be retrieved as a
1060 kinds for Objective-C reflect an optimization within
1062 one-argument selectors are stored as a
1064 very little space, since zero- and one-argument selectors are far
1065 more common than multi-argument selectors (which use a different
1066 structure).</dd>
1069 <dd>The name is a C++ constructor
1072 construct. The type is always the canonical type, since all
1073 constructors for a given type have the same name.</dd>
1076 <dd>The name is a C++ destructor
1079 named. This type is always a canonical type.</dd>
1082 <dd>The name is a C++ conversion function. Conversion functions are
1083 named according to the type they convert to, e.g., "<code>operator void
1085 the type that this conversion function converts to. This type is
1086 always a canonical type.</dd>
1089 <dd>The name is a C++ overloaded operator name. Overloaded operators
1090 are named according to their spelling, e.g.,
1093 retrieve the overloaded operator (a value of
1095 </dl>
1098 compare. They require only a single pointer's worth of storage in
1099 the common cases (identifiers, zero-
1100 and one-argument Objective-C selectors) and use dense, uniqued
1101 storage for the other kinds of
1104 comparison, can be ordered
1107 normal identifiers but an unspecified ordering for other kinds of
1112 ways depending on what kind of name the instance will store. Normal
1116 destructors, conversion functions, and overloaded operators can be retrieved from
1122 C++ special function names.</p>
1124 <!-- ======================================================================= -->
1126 <!-- ======================================================================= -->
1128 context</i>, such as a translation unit, namespace, class, or
1129 function. Declaration contexts in Clang are represented by
1131 declaration-context AST nodes
1132 (<code>TranslationUnitDecl</code>, <code>NamespaceDecl</code>, <code>RecordDecl</code>, <code>FunctionDecl</code>,
1134 several facilities common to each declaration context:</p>
1135 <dl>
1138 stored within a declaration context. The source-centric view
1139 accurately represents the program source code as written, including
1140 multiple declarations of entities where present (see the
1142 Overloads</a>), while the semantics-centric view represents the
1143 program semantics. The two views are kept synchronized by semantic
1144 analysis while the ASTs are being constructed.</dd>
1147 <dd>Every declaration context can contain some number of
1148 declarations. For example, a C++ class (represented
1150 fields, nested types, and so on. All of these declarations will be
1152 over the declarations via
1155 the source-centric view of declarations in the context.</dd>
1159 lookup for names within that declaration context. For example,
1163 based on a lazily-constructed array (for declaration contexts
1164 with a small number of declarations) or hash table (for
1165 declaration contexts with more declarations). The lookup
1166 operation provides the semantics-centric view of the declarations
1167 in the context.</dd>
1171 were declared within its declaration context, and is responsible
1172 for the management of their memory as well as their
1173 (de-)serialization.</dd>
1174 </dl>
1176 <p>All declarations are stored within a declaration context, and one
1177 can query
1178 information about the context in which each declaration lives. One
1183 Contexts</a> for more information about how to interpret this
1184 context information.</p>
1187 <p>Within a translation unit, it is common for an entity to be
1188 declared several times. For example, we might declare a function "f"
1189 and then later re-declare it as part of an inlined definition:</p>
1191 <pre>
1192 void f(int x, int y, int z = 1);
1194 inline void f(int x, int y, int z) { /* ... */ }
1195 </pre>
1198 semantics-centric views of a declaration context. In the
1199 source-centric view, all redeclarations will be present, in the
1200 order they occurred in the source code, making
1201 this view suitable for clients that wish to see the structure of
1202 the source code. In the semantics-centric view, only the most recent "f"
1203 will be found by the lookup, since it effectively replaces the first
1206 <p>In the semantics-centric view, overloading of functions is
1207 represented explicitly. For example, given two declarations of a
1209 <pre>
1210 void g();
1211 void g(int);
1212 </pre>
1215 declarations of "g". Clients that perform semantic analysis on a
1216 program that is not concerned with the actual source code will
1217 primarily use this semantics-centric view.</p>
1220 <p>Each declaration has two potentially different
1222 the source-centric view of the declaration context, and
1224 semantics-centric view. The lexical context is accessible
1226 semantic context is accessible
1229 contexts are identical. For example:</p>
1231 <pre>
1232 class X {
1233 public:
1234 void f(int x);
1235 };
1236 </pre>
1243 <pre>
1244 void X::f(int x = 17) { /* ... */ }
1245 </pre>
1247 <p>This definition of has different lexical and semantic
1248 contexts. The lexical context corresponds to the declaration
1249 context in which the actual declaration occurred in the source
1252 the declarations provided by
1254 translation unit.</p>
1261 information about the default argument).</p>
1264 <p>In C and C++, there are several contexts in which names that are
1265 logically declared inside another declaration will actually "leak"
1266 out into the enclosing scope from the perspective of name
1267 lookup. The most obvious instance of this behavior is in
1268 enumeration types, e.g.,</p>
1269 <pre>
1270 enum Color {
1271 Red,
1272 Green,
1273 Blue
1274 };
1275 </pre>
1278 context that contains the
1285 without qualifying the name, e.g.,</p>
1287 <pre>
1288 Color c = Red;
1289 </pre>
1291 <p>There are other entities in C++ that provide similar behavior. For
1292 example, linkage specifications that use curly braces:</p>
1294 <pre>
1295 extern "C" {
1296 void f(int);
1297 void g(int);
1298 }
1299 // f and g are visible here
1300 </pre>
1302 <p>For source-level accuracy, we treat the linkage specification and
1303 enumeration type as a
1304 declaration context in which its enclosed declarations ("Red",
1306 are declared. However, these declarations are visible outside of the
1307 scope of the declaration context.</p>
1309 <p>These language features (and several others, described below) have
1310 roughly the same set of
1311 requirements: declarations are declared within a particular lexical
1312 context, but the declarations are also found via name lookup in
1313 scopes enclosing the declaration itself. This feature is implemented
1316 declarations are visible in the nearest enclosing non-transparent
1317 declaration context. This means that the lexical context of the
1318 declaration (e.g., an enumerator) will be the
1320 context, but the declaration will be visible in every outer context
1321 up to and including the first non-transparent declaration context (since
1322 transparent declaration contexts can be nested).</p>
1325 <ul>
1327 <pre>
1328 enum Color {
1329 Red,
1330 Green,
1331 Blue
1332 };
1333 // Red, Green, and Blue are in scope
1334 </pre></li>
1335 <li>C++ linkage specifications:
1336 <pre>
1337 extern "C" {
1338 void f(int);
1339 void g(int);
1340 }
1341 // f and g are in scope
1342 </pre></li>
1343 <li>Anonymous unions and structs:
1344 <pre>
1345 struct LookupTable {
1346 bool IsVector;
1347 union {
1350 };
1351 };
1353 LookupTable LT;
1354 LT.Vector = 0; // Okay: finds Vector inside the unnamed union
1355 </pre>
1356 </li>
1358 <pre>
1359 namespace mylib {
1360 inline namespace debug {
1361 class X;
1362 }
1363 }
1364 mylib::X *xp; // okay: mylib::X refers to mylib::debug::X
1365 </pre>
1366 </li>
1367 </ul>
1371 <p>C++ namespaces have the interesting--and, so far, unique--property that
1372 the namespace can be defined multiple times, and the declarations
1373 provided by each namespace definition are effectively merged (from
1374 the semantic point of view). For example, the following two code
1375 snippets are semantically indistinguishable:</p>
1376 <pre>
1377 // Snippet #1:
1378 namespace N {
1379 void f();
1380 }
1381 namespace N {
1382 void f(int);
1383 }
1385 // Snippet #2:
1386 namespace N {
1387 void f();
1388 void f(int);
1389 }
1390 </pre>
1392 <p>In Clang's representation, the source-centric view of declaration
1394 nodes in Snippet #1, each of which is a declaration context that
1395 contains a single declaration of "f". However, the semantics-centric
1401 contexts internally. The
1405 the lookup table used for the semantics-centric view. Given the
1406 primary context, one can follow the chain
1409 these functions are used internally within the lookup and insertion
1411 clients can ignore them.</p>
1413 <!-- ======================================================================= -->
1415 <!-- ======================================================================= -->
1422 which includes simple expressions. Control-flow graphs are especially
1423 useful for performing
1427 <!-- ============ -->
1429 <!-- ============ -->
1434 to statements in the AST). The ordering of statements within a block
1435 indicates unconditional flow of control from one statement to the
1437 is represented using edges between basic blocks. The statements
1441 <p>
1444 CFG is also uniquely numbered (accessible
1446 based on the ordering the blocks were created, but no assumptions
1449 the number of basic blocks in the CFG).</p>
1451 <!-- ===================== -->
1453 <!-- ===================== -->
1459 block contains any statements, and they serve the role of providing a
1460 clear entrance and exit for a body of code such as a function body.
1461 The presence of these empty blocks greatly simplifies the
1462 implementation of many analyses built on top of CFGs.
1464 <!-- ===================================================== -->
1466 <!-- ===================================================== -->
1468 <p>Conditional control-flow (such as those induced by if-statements
1470 Because different C language constructs can induce control-flow,
1473 the statement that caused the control-flow, and is used to identify
1474 the nature of the conditional control-flow between blocks. For
1475 example, in the case of an if-statement, the terminator refers to
1477 branch.</p>
1481 <code>
1482 int foo(int x) {<br>
1484 <br>
1490 <br>
1492 }
1493 </code>
1495 <p>After invoking the parser+semantic analyzer on this code fragment,
1499 body by single call to a static class method:</p>
1501 <code>
1504 </code>
1508 longer needed.</p>
1510 <p>Along with providing an interface to iterate over
1512 that are useful for debugging and visualizing CFGs. For example, the
1513 method
1515 standard error. This is especially useful when one is using a
1516 debugger such as gdb. For example, here is the output
1519 <code>
1523 <br>
1530 <br>
1535 <br>
1541 <br>
1546 <br>
1550 </code>
1552 <p>For each block, the pretty-printed output displays for each block
1555 that have control-flow that have incoming control-flow from the given
1556 block). We can also clearly see the special entry and exit blocks at
1557 the beginning and end of the pretty-printed output. For the entry
1558 block (block B5), the number of predecessor blocks is 0, while for the
1561 <p>The most interesting block here is B4, whose outgoing control-flow
1562 represents the branching caused by the sole if-statement
1566 evaluation of the condition of the if-statement, which occurs before
1572 refer to proper C statements.</p>
1575 object in the AST. The pretty-printer outputs <b><tt>if
1576 [B4.2]</tt></b> because the condition expression of the if-statement
1577 has an actual place in the basic block, and thus the terminator is
1578 essentially
1580 block B4 (i.e., B4.2). In this manner, conditions for control-flow
1581 (which also includes conditions for loops and switch statements) are
1582 hoisted into the actual basic block.</p>
1584 <!-- ===================== -->
1585 <!-- <h4>Implicit Control-Flow</h4> -->
1586 <!-- ===================== -->
1588 <!--
1589 <p>A key design principle of the <tt>CFG</tt> class was to not require
1590 any transformations to the AST in order to represent control-flow.
1591 Thus the <tt>CFG</tt> does not perform any "lowering" of the
1592 statements in an AST: loops are not transformed into guarded gotos,
1593 short-circuit operations are not converted to a set of if-statements,
1594 and so on.</p>
1595 -->
1598 <!-- ======================================================================= -->
1600 <!-- ======================================================================= -->
1602 <p>There are several places where constants and constant folding matter a lot to
1603 the Clang front-end. First, in general, we prefer the AST to retain the source
1604 code as close to how the user wrote it as possible. This means that if they
1605 wrote "5+4", we want to keep the addition and two constants in the AST, we don't
1606 want to fold to "9". This means that constant folding in various ways turns
1607 into a tree walk that needs to handle the various cases.</p>
1609 <p>However, there are places in both C and C++ that require constants to be
1610 folded. For example, the C standard defines what an "integer constant
1611 expression" (i-c-e) is with very precise and specific requirements. The
1612 language then requires i-c-e's in a lot of places (for example, the size of a
1613 bitfield, the value for a case statement, etc). For these, we have to be able
1614 to constant fold the constants, to do semantic checks (e.g. verify bitfield size
1615 is non-negative and that case statements aren't duplicated). We aim for Clang
1616 to be very pedantic about this, diagnosing cases when the code does not use an
1617 i-c-e where one is required, but accepting the code unless running with
1620 <p>Things get a little bit more tricky when it comes to compatibility with
1621 real-world source code. Specifically, GCC has historically accepted a huge
1622 superset of expressions as i-c-e's, and a lot of real world code depends on this
1623 unfortuate accident of history (including, e.g., the glibc system headers). GCC
1624 accepts anything its "fold" optimizer is capable of reducing to an integer
1625 constant, which means that the definition of what it accepts changes as its
1626 optimizer does. One example is that GCC accepts things like "case X-X:" even
1629 <p>Another issue are how constants interact with the extensions we support, such
1630 as __builtin_constant_p, __builtin_inf, __extension__ and many others. C99
1631 obviously does not specify the semantics of any of these extensions, and the
1632 definition of i-c-e does not include them. However, these extensions are often
1633 used in real code, and we have to have a way to reason about them.</p>
1635 <p>Finally, this is not just a problem for semantic analysis. The code
1636 generator and other clients have to be able to fold constants (e.g. to
1637 initialize global variables) and has to handle a superset of what C99 allows.
1638 Further, these clients can benefit from extended information. For example, we
1639 know that "foo()||1" always evaluates to true, but we can't replace the
1640 expression with true because it has side effects.</p>
1642 <!-- ======================= -->
1644 <!-- ======================= -->
1646 <p>After trying several different approaches, we've finally converged on a
1647 design (Note, at the time of this writing, not all of this has been implemented,
1648 consider this a design goal!). Our basic approach is to define a single
1651 type (integer, fp, complex, or pointer) this method returns the following
1652 information:</p>
1654 <ul>
1655 <li>Whether the expression is an integer constant expression, a general
1656 constant that was folded but has no side effects, a general constant that
1657 was folded but that does have side effects, or an uncomputable/unfoldable
1658 value.
1659 </li>
1660 <li>If the expression was computable in any way, this method returns the APValue
1661 for the result of the expression.</li>
1662 <li>If the expression is not evaluatable at all, this method returns
1663 information on one of the problems with the expression. This includes a
1664 SourceLocation for where the problem is, and a diagnostic ID that explains
1665 the problem. The diagnostic should be have ERROR type.</li>
1666 <li>If the expression is not an integer constant expression, this method returns
1667 information on one of the problems with the expression. This includes a
1668 SourceLocation for where the problem is, and a diagnostic ID that explains
1669 the problem. The diagnostic should be have EXTENSION type.</li>
1670 </ul>
1672 <p>This information gives various clients the flexibility that they want, and we
1673 will eventually have some helper methods for various extensions. For example,
1675 calls Evaluate on the expression. If the expression is not foldable, the error
1676 is emitted, and it would return true. If the expression is not an i-c-e, the
1677 EXTENSION diagnostic is emitted. Finally it would return false to indicate that
1678 the AST is ok.</p>
1680 <p>Other clients can use the information in other ways, for example, codegen can
1681 just use expressions that are foldable in any way.</p>
1683 <!-- ========== -->
1685 <!-- ========== -->
1687 <p>This section describes how some of the various extensions Clang supports
1688 interacts with constant evaluation:</p>
1690 <ul>
1692 any evaluatable subexpression to be accepted as an integer constant
1693 expression.</li>
1695 constant expression) if the operand is any evaluatable constant. As a
1697 parenthesized) condition of a conditional operator expression ("?:"), only
1698 the true side of the conditional operator is considered, and it is evaluated
1699 with full constant folding.</li>
1701 integer constant expression, but we accept any constant as an "extension of
1702 an extension". This only evaluates one operand depending on which way the
1703 condition evaluates.</li>
1705 constant expression.</li>
1707 floating-point literal.</li>
1709 general constant expressions.</li>
1710 <li><b><tt>__builtin_strlen</tt></b> and <b><tt>strlen</tt></b>: These are constant folded as integer constant expressions if the argument is a string literal.</li>
1711 </ul>
1716 </div>
1717 </body>
1718 </html>