| blob | 813015ecf659b775964c4a50c16f91c33d24fa70 |
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>
73 <ul>
75 </ul>
76 </li>
77 </ul>
80 <!-- ======================================================================= -->
82 <!-- ======================================================================= -->
84 <p>This document describes some of the more important APIs and internal design
85 decisions made in the Clang C front-end. The purpose of this document is to
86 both capture some of this high level information and also describe some of the
87 design decisions behind it. This is meant for people interested in hacking on
88 Clang, not for end-users. The description below is categorized by
89 libraries, and does not describe any of the clients of the libraries.</p>
91 <!-- ======================================================================= -->
93 <!-- ======================================================================= -->
95 <p>The LLVM libsystem library provides the basic Clang system abstraction layer,
96 which is used for file system access. The LLVM libsupport library provides many
97 underlying libraries and <a
99 including command line option
100 processing and various containers.</p>
102 <!-- ======================================================================= -->
104 <!-- ======================================================================= -->
106 <p>This library certainly needs a better name. The 'basic' library contains a
107 number of low-level utilities for tracking and manipulating source buffers,
108 locations within the source buffers, diagnostics, tokens, target abstraction,
109 and information about the subset of the language being compiled for.</p>
111 <p>Part of this infrastructure is specific to C (such as the TargetInfo class),
112 other parts could be reused for other non-C-based languages (SourceLocation,
113 SourceManager, Diagnostics, FileManager). When and if there is future demand
114 we can figure out if it makes sense to introduce a new library, move the general
115 classes somewhere else, or introduce some other solution.</p>
120 <!-- ======================================================================= -->
122 <!-- ======================================================================= -->
124 <p>The Clang Diagnostics subsystem is an important part of how the compiler
125 communicates with the human. Diagnostics are the warnings and errors produced
126 when the code is incorrect or dubious. In Clang, each diagnostic produced has
127 (at the minimum) a unique ID, an English translation associated with it, a <a
130 of arguments to the dianostic (which fill in "%0"'s in the string) as well as a
131 number of source ranges that related to the diagnostic.</p>
133 <p>In this section, we'll be giving examples produced by the Clang command line
135 different ways</a> depending on how the DiagnosticClient interface is
136 implemented. A representative example of a diagnostic is:</p>
138 <pre>
142 </pre>
144 <p>In this example, you can see the English translation, the severity (error),
145 you can see the source location (the caret ("^") and file/line/column info),
147 float"). You'll have to believe me that there is a unique ID backing the
148 diagnostic :).</p>
150 <p>Getting all of this to happen has several steps and involves many moving
151 pieces, this section describes them and talks about best practices when adding
152 a new diagnostic.</p>
154 <!-- ============================= -->
156 <!-- ============================= -->
159 clang/Basic/Diagnostic*Kinds.td</tt> files, depending on what library will
160 be using it. From this file, tblgen generates the unique ID of the diagnostic,
161 the severity of the diagnostic and the English translation + format string.</p>
163 <p>There is little sanity with the naming of the unique ID's right now. Some
164 start with err_, warn_, ext_ to encode the severity into the name. Since the
165 enum is referenced in the C++ code that produces the diagnostic, it is somewhat
166 useful for it to be reasonably short.</p>
171 acceptable under any circumstances. When an error is emitted, the AST for the
173 severities are used for extensions to the language that Clang accepts. This
174 means that Clang fully understands and can represent them in the AST, but we
175 produce diagnostics to tell the user their code is non-portable. The difference
176 is that the former are ignored by default, and the later warn by default. The
179 level is used to staple more information onto previous diagnostics.</p>
184 subsystem based on various configuration options. Clang internally supports a
185 fully fine grained mapping mechanism that allows you to map almost any
186 diagnostic to the output level that you want. The only diagnostics that cannot
190 for example).</p>
192 <p>Diagnostic mappings are used in many ways. For example, if the user
196 </p>
198 <p>
200 considered so severe that error recovery won't be able to recover sensibly from
201 them (thus spewing a ton of bogus errors). One example of this class of error
202 are failure to #include a file.
203 </p>
205 <!-- ================= -->
207 <!-- ================= -->
209 <p>The format string for the diagnostic is very simple, but it has some power.
210 It takes the form of a string in English with markers that indicate where and
211 how arguments to the diagnostic are inserted and formatted. For example, here
212 are some simple format strings:</p>
214 <pre>
215 "binary integer literals are an extension"
216 "format string contains '\\0' within the string body"
217 "more '<b>%%</b>' conversions than data arguments"
218 "invalid operands to binary expression (<b>%0</b> and <b>%1</b>)"
219 "overloaded '<b>%0</b>' must be a <b>%select{unary|binary|unary or binary}2</b> operator"
220 " (has <b>%1</b> parameter<b>%s1</b>)"
221 </pre>
223 <p>These examples show some important points of format strings. You can use any
224 plain ASCII character in the diagnostic string except "%" without a problem,
225 but these are C strings, so you have to use and be aware of all the C escape
226 sequences (as in the second example). If you want to produce a "%" in the
227 output, use the "%%" escape sequence, like the third diagnostic. Finally,
228 Clang uses the "%...[digit]" sequences to specify where and how arguments to
229 the diagnostic are formatted.</p>
231 <p>Arguments to the diagnostic are numbered according to how they are specified
234 to your diagnostic, you are doing something wrong :). Unlike printf, there
235 is no requirement that arguments to the diagnostic end up in the output in
236 the same order as they are specified, you could have a format string with
238 percent and digit are formatting instructions. If there are no instructions,
239 the argument is just turned into a string and substituted in.</p>
243 <ul>
246 printed, and forces you to think about the important point you are conveying
247 with the diagnostic.</li>
248 <li>Take advantage of location information. The user will be able to see the
249 line and location of the caret, so you don't need to tell them that the
251 <li>Do not capitalize the diagnostic string, and do not end it with a
252 period.</li>
253 <li>If you need to quote something in the diagnostic string, use single
254 quotes.</li>
255 </ul>
257 <p>Diagnostics should never take random English strings as arguments: you
261 other languages (because they'll get random English words in their otherwise
262 localized diagnostic). The exceptions to this are C/C++ language keywords
266 including variable names, types, labels, etc. The 'select' format can be
267 used to achieve this sort of thing in a localizable way, see below.</p>
269 <!-- ==================================== -->
271 <!-- ==================================== -->
273 <p>Arguments to diagnostics are fully typed internally, and come from a couple
274 different classes: integers, types, names, and random strings. Depending on
275 the class of the argument, it can be optionally formatted in different ways.
276 This gives the DiagnosticClient information about what the argument means
277 without requiring it to use a specific presentation (consider this MVC for
278 Clang :).</p>
280 <p>Here are the different diagnostic argument formats currently supported by
281 Clang:</p>
283 <table>
288 useful when producing English diagnostics. When the integer is 1, it prints
290 simple grammatical forms to be to be handled correctly, and eliminates the
298 related diagnostics together into one common one, without requiring the
299 difference to be specified as an English string argument. Instead of
300 specifying the string, the diagnostic gets an integer argument and the
301 format string selects the numbered option. In this case, the "%2" value
304 allows other language translations to substitute reasonable words (or entire
305 phrases) based on the semantics of the diagnostic instead of having to do
306 things textually.</p>
314 It is designed to handle even the requirements of languages with very
315 complex plural forms, as many Baltic languages have. The argument consists
316 of a series of expression/form pairs, separated by ':', where the first form
317 whose expression evaluates to true is the result of the modifier.</p>
318 <p>An expression can be empty, in which case it is always true. See the
319 example at the top. Otherwise, it is a series of one or more numeric
320 conditions, separated by ','. If any condition matches, the expression
321 matches. Each numeric condition can take one of three forms.</p>
322 <ul>
323 <li>number: A simple decimal number matches if the argument is the same
325 <li>range: A range in square brackets matches if the argument is within
326 the range. Then range is inclusive on both ends. Example:
328 <li>modulo: A modulo operator is followed by a number, and
329 equals sign and either a number or a range. The tests are the
330 same as for plain
331 numbers and ranges, but the argument is taken modulo the number first.
334 </ul>
335 <p>The parser is very unforgiving. A syntax error, even whitespace, will
336 abort, as will a failure to match the argument against any
337 expression.</p></td></tr>
345 are not supported.</p>
352 DeclarationName corresponds to an Objective-C class method selector. As
353 such, it prints the selector with a leading '+'.</p></td></tr>
359 DeclarationName corresponds to an Objective-C instance method selector. As
360 such, it prints the selector with a leading '-'.</p></td></tr>
365 <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>
367 </table>
369 <p>It is really easy to add format specifiers to the Clang diagnostics system,
370 but they should be discussed before they are added. If you are creating a lot
371 of repetitive diagnostics and/or have an idea for a useful formatter, please
372 bring it up on the cfe-dev mailing list.</p>
374 <!-- ===================================================== -->
376 <!-- ===================================================== -->
378 <p>Now that you've created the diagnostic in the DiagnosticKinds.td file, you
379 need to write the code that detects the condition in question and emits the
380 new diagnostic. Various components of Clang (e.g. the preprocessor, Sema,
381 etc) provide a helper function named "Diag". It creates a diagnostic and
382 accepts the arguments, ranges, and other information that goes along with
383 it.</p>
387 <pre>
388 if (various things that are bad)
389 Diag(Loc, diag::err_typecheck_invalid_operands)
392 </pre>
396 (which matches the name from DiagnosticKinds.td). If the diagnostic takes
397 arguments, they are specified with the << operator: the first argument
403 SourceRanges are also specified with the << operator, but do not have a
404 specific ordering requirement.</p>
406 <p>As you can see, adding and producing a diagnostic is pretty straightforward.
407 The hard part is deciding exactly what you need to say to help the user, picking
408 a suitable wording, and providing the information needed to format it correctly.
409 The good news is that the call site that issues a diagnostic should be
410 completely independent of how the diagnostic is formatted and in what language
411 it is rendered.
412 </p>
414 <!-- ==================================================== -->
416 <!-- ==================================================== -->
418 <p>In some cases, the front end emits diagnostics when it is clear
419 that some small change to the source code would fix the problem. For
420 example, a missing semicolon at the end of a statement or a use of
421 deprecated syntax that is easily rewritten into a more modern form.
422 Clang tries very hard to emit the diagnostic and recover gracefully
423 in these and other cases.</p>
425 <p>However, for these cases where the fix is obvious, the diagnostic
426 can be annotated with a code
427 modification "hint" that describes how to change the code referenced
428 by the diagnostic to fix the problem. For example, it might add the
429 missing semicolon at the end of the statement or rewrite the use of a
430 deprecated construct into something more palatable. Here is one such
431 example C++ front end, where we warn about the right-shift operator
434 <pre>
435 test.cpp:3:7: warning: use of right-shift operator ('>>') in template argument will require parentheses in C++0x
437 ^
438 ( )
439 </pre>
441 <p>Here, the code modification hint is suggesting that parentheses be
442 added, and showing exactly where those parentheses would be inserted
443 into the source code. The code modification hints themselves describe
444 what changes to make to the source code in an abstract manner, which
445 the text diagnostic printer renders as a line of "insertions" below
447 clients</a> might choose to render the code differently (e.g., as
448 markup inline) or even give the user the ability to automatically fix
449 the problem.</p>
451 <p>All code modification hints are described by the
453 attached to the diagnostic using the << operator in the same way
454 that highlighted source ranges and arguments are passed to the
455 diagnostic. Code modification hints can be created with one of three
456 constructors:</p>
458 <dl>
465 should be removed.</dd>
470 </dl>
472 <!-- ============================================================= -->
474 <!-- ============================================================= -->
476 <p>Once code generates a diagnostic with all of the arguments and the rest of
477 the relevant information, Clang needs to know what to do with it. As previously
478 mentioned, the diagnostic machinery goes through some filtering to map a
479 severity onto a diagnostic level, then (assuming the diagnostic is not mapped to
480 "<tt>Ignore</tt>") it invokes an object that implements the DiagnosticClient
481 interface with the information.</p>
483 <p>It is possible to implement this interface in many different ways. For
484 example, the normal Clang DiagnosticClient (named 'TextDiagnosticPrinter') turns
485 the arguments into strings (according to the various formatting rules), prints
486 out the file/line/column information and the string, then prints out the line of
487 code, the source ranges, and the caret. However, this behavior isn't required.
488 </p>
490 <p>Another implementation of the DiagnosticClient interface is the
491 'TextDiagnosticBuffer' class, which is used when Clang is in -verify mode.
492 Instead of formatting and printing out the diagnostics, this implementation just
493 captures and remembers the diagnostics as they fly by. Then -verify compares
494 the list of produced diagnostics to the list of expected ones. If they disagree,
495 it prints out its own output.
496 </p>
498 <p>There are many other possible implementations of this interface, and this is
499 why we prefer diagnostics to pass down rich structured information in arguments.
500 For example, an HTML output might want declaration names be linkified to where
501 they come from in the source. Another example is that a GUI might let you click
502 on typedefs to expand them. This application would want to pass significantly
503 more information about types through to the GUI than a simple flat string. The
504 interface allows this to happen.</p>
506 <!-- ====================================================== -->
508 <!-- ====================================================== -->
511 can translate to the relevant code page if needed. Each translation completely
512 replaces the format string for the diagnostic.</p>
515 <!-- ======================================================================= -->
517 <!-- ======================================================================= -->
519 <p>Strangely enough, the SourceLocation class represents a location within the
520 source code of the program. Important design points include:</p>
522 <ol>
523 <li>sizeof(SourceLocation) must be extremely small, as these are embedded into
525 <li>SourceLocation must be a simple value object that can be efficiently
526 copied.</li>
527 <li>We should be able to represent a source location for any byte of any input
528 file. This includes in the middle of tokens, in whitespace, in trigraphs,
529 etc.</li>
530 <li>A SourceLocation must encode the current #include stack that was active when
531 the location was processed. For example, if the location corresponds to a
532 token, it should contain the set of #includes active when the token was
533 lexed. This allows us to print the #include stack for a diagnostic.</li>
534 <li>SourceLocation must be able to describe macro expansions, capturing both
535 the ultimate instantiation point and the source of the original character
536 data.</li>
537 </ol>
539 <p>In practice, the SourceLocation works together with the SourceManager class
540 to encode two pieces of information about a location: its spelling location
541 and its instantiation location. For most tokens, these will be the same.
542 However, for a macro expansion (or tokens that came from a _Pragma directive)
543 these will describe the location of the characters corresponding to the token
544 and the location where the token was used (i.e. the macro instantiation point
545 or the location of the _Pragma itself).</p>
547 <p>The Clang front-end inherently depends on the location of a token being
548 tracked correctly. If it is ever incorrect, the front-end may get confused and
549 die. The reason for this is that the notion of the 'spelling' of a Token in
550 Clang depends on being able to find the original input characters for the token.
554 <!-- ======================================================================= -->
556 <!-- ======================================================================= -->
557 <!-- mostly taken from
558 http://lists.cs.uiuc.edu/pipermail/cfe-dev/2010-August/010595.html -->
560 <p>Clang represents most source ranges by [first, last], where first and last
561 each point to the beginning of their respective tokens. For example
562 consider the SourceRange of the following statement:</p>
563 <pre>
564 x = foo + bar;
565 ^first ^last
566 </pre>
568 <p>To map from this representation to a character-based
569 representation, the 'last' location needs to be adjusted to point to
570 (or past) the end of that token with either
573 where character-level source ranges information is needed we use
577 <!-- ======================================================================= -->
579 <!-- ======================================================================= -->
584 <!-- ======================================================================= -->
586 <!-- ======================================================================= -->
588 <p>Clang supports two implementations of precompiled headers. The
589 default implementation, precompiled headers (<a
591 of Clang's internal data structures, encoded with the <a
595 serialized representation of the tokens encountered when
596 preprocessing a header (and anything that header includes).</p>
599 <!-- ======================================================================= -->
601 <!-- ======================================================================= -->
603 <p>The Frontend library contains functionality useful for building
604 tools on top of the clang libraries, for example several methods for
605 outputting diagnostics.</p>
607 <!-- ======================================================================= -->
609 <!-- ======================================================================= -->
611 <p>The Lexer library contains several tightly-connected classes that are involved
612 with the nasty process of lexing and preprocessing C source code. The main
613 interface to this library for outside clients is the large <a
615 It contains the various pieces of state that are required to coherently read
616 tokens out of a translation unit.</p>
618 <p>The core interface to the Preprocessor object (once it is set up) is the
620 the preprocessor stream. There are two types of token providers that the
621 preprocessor is capable of reading from: a buffer lexer (provided by the <a
626 <!-- ======================================================================= -->
628 <!-- ======================================================================= -->
630 <p>The Token class is used to represent a single lexed token. Tokens are
631 intended to be used by the lexer/preprocess and parser libraries, but are not
632 intended to live beyond them (for example, they should not live in the ASTs).<p>
634 <p>Tokens most often live on the stack (or some other location that is efficient
635 to access) as the parser is running, but occasionally do get buffered up. For
636 example, macro definitions are stored as a series of tokens, and the C++
637 front-end periodically needs to buffer tokens up for tentative parsing and
638 various pieces of look-ahead. As such, the size of a Token matter. On a 32-bit
642 Tokens</a>" and normal tokens. Normal tokens are those returned by the lexer,
643 annotation tokens represent semantic information and are produced by the parser,
644 replacing normal tokens in the token stream. Normal tokens contain the
645 following information:</p>
647 <ul>
649 token.</li>
652 SourceBuffer. For tokens that include them, this length includes trigraphs and
653 escaped newlines which are ignored by later phases of the compiler. By pointing
654 into the original source buffer, it is always possible to get the original
655 spelling of a token completely accurately.</li>
658 identifier lookup was enabled when the token was lexed (e.g. the lexer was not
659 reading in 'raw' mode) this contains a pointer to the unique hash value for the
660 identifier. Because the lookup happens before keyword identification, this
661 field is set even for language keywords like 'for'.</li>
667 that some tokens can be spelled multiple ways. For example, C++ supports
669 "&&" operator. In these cases, the kind value is set to
671 consider both forms. For something that cares about which form is used (e.g.
672 the preprocessor 'stringize' operator) the spelling indicates the original
673 form.</li>
676 lexer/preprocessor system on a per-token basis:
678 <ol>
680 source line.</li>
682 before the token or transitively before the token as it was expanded
683 through a macro. The definition of this flag is very closely defined by
684 the stringizing requirements of the preprocessor.</li>
686 represent identifier tokens which have macro expansion disabled. This
687 prevents them from being considered as candidates for macro expansion ever
688 in the future.</li>
690 token includes a trigraph or escaped newline. Since this is uncommon,
691 many pieces of code can fast-path on tokens that did not need cleaning.
692 </p>
693 </ol>
694 </li>
695 </ul>
697 <p>One interesting (and somewhat unusual) aspect of normal tokens is that they
698 don't contain any semantic information about the lexed value. For example, if
699 the token was a pp-number token, we do not represent the value of the number
700 that was lexed (this is left for later pieces of code to decide). Additionally,
701 the lexer library has no notion of typedef names vs variable names: both are
702 returned as identifiers, and the parser is left to decide whether a specific
703 identifier is a typedef or a variable (tracking this requires scope information
704 among other things). The parser can do this translation by replacing tokens
707 <!-- ======================================================================= -->
709 <!-- ======================================================================= -->
711 <p>Annotation Tokens are tokens that are synthesized by the parser and injected
712 into the preprocessor's token stream (replacing existing tokens) to record
713 semantic information found by the parser. For example, if "foo" is found to be
716 makes it easy to handle qualified type names (e.g. "foo::bar::baz<42>::t")
718 reparse does not need to redo semantic analysis to determine whether a token
719 sequence is a variable, type, template, etc.</p>
721 <p>Annotation Tokens are created by the parser and reinjected into the parser's
722 token stream (when backtracking is enabled). Because they can only exist in
723 tokens that the preprocessor-proper is done with, it doesn't need to keep around
724 flags like "start of line" that the preprocessor uses to do its job.
725 Additionally, an annotation token may "cover" a sequence of preprocessor tokens
727 of an annotation token are different than the fields for a normal token (but
728 they are multiplexed into the normal Token fields):</p>
730 <ul>
732 token indicates the first token replaced by the annotation token. In the example
736 last token replaced with the annotation token. In the example above, it would
740 that the parser gets from Sema. The parser merely preserves the
741 information for Sema to later interpret based on the annotation token
742 kind.</li>
745 is. See below for the different valid kinds.</li>
746 </ul>
750 <ol>
752 resolved typename token that is potentially qualified. The
754 Sema::getTypeName(), possibly with source location information
755 attached.</li>
758 scope specifier, such as "A::B::". This corresponds to the grammar
761 the Sema::ActOnCXXGlobalScopeSpecifier and
762 Sema::ActOnCXXNestedNameSpecifier callbacks.</li>
766 of a template. The AnnotationValue pointer is a pointer to a malloc'd
767 TemplateIdAnnotation object. Depending on the context, a parsed
768 template-id that names a type might become a typename annotation token
769 (if all we care about is the named type, e.g., because it occurs in a
770 type specifier) or might remain a template-id token (if we want to
771 retain more source location information or produce a new type, e.g.,
772 in a declaration of a class template specialization). template-id
773 annotation tokens that refer to a type can be "upgraded" to typename
774 annotation tokens by the parser.</li>
776 </ol>
778 <p>As mentioned above, annotation tokens are not returned by the preprocessor,
779 they are formed on demand by the parser. This means that the parser has to be
780 aware of cases where an annotation could occur and form it where appropriate.
781 This is somewhat similar to how the parser handles Translation Phase 6 of C99:
782 String Concatenation (see C99 5.1.1.2). In the case of string concatenation,
783 the preprocessor just returns distinct tok::string_literal and
784 tok::wide_string_literal tokens and the parser eats a sequence of them wherever
785 the grammar indicates that a string literal can occur.</p>
787 <p>In order to do this, whenever the parser expects a tok::identifier or
788 tok::coloncolon, it should call the TryAnnotateTypeOrScopeToken or
789 TryAnnotateCXXScopeToken methods to form the annotation token. These methods
790 will maximally form the specified annotation tokens and replace the current
791 token with them, if applicable. If the current tokens is not valid for an
792 annotation token, it will remain an identifier or :: token.</p>
796 <!-- ======================================================================= -->
798 <!-- ======================================================================= -->
800 <p>The Lexer class provides the mechanics of lexing tokens out of a source
801 buffer and deciding what they mean. The Lexer is complicated by the fact that
802 it operates on raw buffers that have not had spelling eliminated (this is a
803 necessity to get decent performance), but this is countered with careful coding
804 as well as standard performance techniques (for example, the comment handling
805 code is vectorized on X86 and PowerPC hosts).</p>
809 <ul>
810 <li>The lexer can operate in 'raw' mode. This mode has several features that
811 make it possible to quickly lex the file (e.g. it stops identifier lookup,
812 doesn't specially handle preprocessor tokens, handles EOF differently, etc).
813 This mode is used for lexing within an "<tt>#if 0</tt>" block, for
814 example.</li>
815 <li>The lexer can capture and return comments as tokens. This is required to
816 support the -C preprocessor mode, which passes comments through, and is
817 used by the diagnostic checker to identifier expect-error annotations.</li>
818 <li>The lexer can be in ParsingFilename mode, which happens when preprocessing
819 after reading a #include directive. This mode changes the parsing of '<'
820 to return an "angled string" instead of a bunch of tokens for each thing
821 within the filename.</li>
823 ParsingPreprocessorDirective mode is entered. This changes the parser to
824 return EOM at a newline.</li>
825 <li>The Lexer uses a LangOptions object to know whether trigraphs are enabled,
826 whether C++ or ObjC keywords are recognized, etc.</li>
827 </ul>
829 <p>In addition to these modes, the lexer keeps track of a couple of other
830 features that are local to a lexed buffer, which change as the buffer is
831 lexed:</p>
833 <ul>
834 <li>The Lexer uses BufferPtr to keep track of the current character being
835 lexed.</li>
836 <li>The Lexer uses IsAtStartOfLine to keep track of whether the next lexed token
838 <li>The Lexer keeps track of the current #if directives that are active (which
839 can be nested).</li>
841 MultipleIncludeOpt</a> object, which is used to
842 detect whether the buffer uses the standard "<tt>#ifndef XX</tt> /
843 <tt>#define XX</tt>" idiom to prevent multiple inclusion. If a buffer does,
844 subsequent includes can be ignored if the XX macro is defined.</li>
845 </ul>
847 <!-- ======================================================================= -->
849 <!-- ======================================================================= -->
851 <p>The TokenLexer class is a token provider that returns tokens from a list
852 of tokens that came from somewhere else. It typically used for two things: 1)
853 returning tokens from a macro definition as it is being expanded 2) returning
854 tokens from an arbitrary buffer of tokens. The later use is used by _Pragma and
855 will most likely be used to handle unbounded look-ahead for the C++ parser.</p>
857 <!-- ======================================================================= -->
859 <!-- ======================================================================= -->
861 <p>The MultipleIncludeOpt class implements a really simple little state machine
862 that is used to detect the standard "<tt>#ifndef XX</tt> / <tt>#define XX</tt>"
863 idiom that people typically use to prevent multiple inclusion of headers. If a
864 buffer uses this idiom and is subsequently #include'd, the preprocessor can
865 simply check to see whether the guarding condition is defined or not. If so,
866 the preprocessor can completely ignore the include of the header.</p>
870 <!-- ======================================================================= -->
872 <!-- ======================================================================= -->
874 <!-- ======================================================================= -->
876 <!-- ======================================================================= -->
878 <!-- ======================================================================= -->
880 <!-- ======================================================================= -->
882 <p>The Type class (and its subclasses) are an important part of the AST. Types
883 are accessed through the ASTContext class, which implicitly creates and uniques
884 them as they are needed. Types have a couple of non-obvious features: 1) they
885 do not capture type qualifiers like const or volatile (See
887 information. Once created, types are immutable (unlike decls).</p>
889 <p>Typedefs in C make semantic analysis a bit more complex than it would
890 be without them. The issue is that we want to capture typedef information
891 and represent it in the AST perfectly, but the semantics of operations need to
894 <code>
895 void func() {<br>
903 }<br>
904 </code>
906 <p>The code above is illegal, and thus we expect there to be diagnostics emitted
907 on the annotated lines. In this example, we expect to get:</p>
909 <pre>
911 *X; // error
914 **Y; // error
917 **Z; // error
919 </pre>
921 <p>While this example is somewhat silly, it illustrates the point: we want to
922 retain typedef information where possible, so that we can emit errors about
924 Doing this requires properly keeping typedef information (for example, the type
927 to retain this information, the type of these expressions is an instance of the
928 TypedefType class, which indicates that the type of these expressions is a
929 typedef for foo.
930 </p>
932 <p>Representing types like this is great for diagnostics, because the
933 user-specified type is always immediately available. There are two problems
934 with this: first, various semantic checks need to make judgements about the
936 efficient way to query whether two types are structurally identical to each
937 other, ignoring typedefs. The solution to both of these problems is the idea of
938 canonical types.</p>
940 <!-- =============== -->
942 <!-- =============== -->
944 <p>Every instance of the Type class contains a canonical type pointer. For
946 "<tt>int**</tt>"), the type just points to itself. For types that have a
949 structurally equivalent type without any typedefs (e.g. "<tt>int</tt>",
952 <p>This design provides a constant time operation (dereferencing the canonical
953 type pointer) that gives us access to the structure of types. For example,
955 their canonical type pointers and doing a pointer comparison (they both point
958 <p>Canonical types and typedef types bring up some complexities that must be
959 carefully managed. Specifically, the "isa/cast/dyncast" operators generally
960 shouldn't be used in code that is inspecting the AST. For example, when type
961 checking the indirection operator (unary '*' on a pointer), the type checker
962 must verify that the operand has a pointer type. It would not be correct to
963 check that with "<tt>isa<PointerType>(SubExpr->getType())</tt>",
964 because this predicate would fail if the subexpression had a typedef type.</p>
966 <p>The solution to this problem are a set of helper methods on Type, used to
967 check their properties. In this case, it would be correct to use
968 "<tt>SubExpr->getType()->isPointerType()</tt>" to do the check. This
970 true any time the type is structurally a pointer type. The only hard part here
973 <p>The second problem we face is how to get access to the pointer type once we
974 know it exists. To continue the example, the result type of the indirection
975 operator is the pointee type of the subexpression. In order to determine the
976 type, we need to get the instance of PointerType that best captures the typedef
977 information in the program. If the type of the expression is literally a
978 PointerType, we can return that, otherwise we have to dig through the
979 typedefs to find the pointer type. For example, if the subexpression had type
980 "<tt>foo*</tt>", we could return that type as the result. If the subexpression
983 a getAsPointerType() method that checks whether the type is structurally a
984 PointerType and, if so, returns the best one. If not, it returns a null
985 pointer.</p>
987 <p>This structure is somewhat mystical, but after meditating on it, it will
988 make sense to you :).</p>
990 <!-- ======================================================================= -->
992 <!-- ======================================================================= -->
994 <p>The QualType class is designed as a trivial value class that is
995 small, passed by-value and is efficient to query. The idea of
996 QualType is that it stores the type qualifiers (const, volatile,
997 restrict, plus some extended qualifiers required by language
998 extensions) separately from the types themselves. QualType is
1001 <p>By storing the type qualifiers as bits in the conceptual pair, it is
1002 extremely efficient to get the set of qualifiers on a QualType (just return the
1003 field of the pair), add a type qualifier (which is a trivial constant-time
1004 operation that sets a bit), and remove one or more type qualifiers (just return
1005 a QualType with the bitfield set to empty).</p>
1007 <p>Further, because the bits are stored outside of the type itself, we do not
1008 need to create duplicates of types with different sets of qualifiers (i.e. there
1010 both point to the same heap allocated "int" type). This reduces the heap size
1011 used to represent bits and also means we do not have to consider qualifiers when
1014 <p>In practice, the two most common type qualifiers (const and
1015 restrict) are stored in the low bits of the pointer to the Type
1016 object, together with a flag indicating whether extended qualifiers
1017 are present (which must be heap-allocated). This means that QualType
1018 is exactly the same size as a pointer.</p>
1020 <!-- ======================================================================= -->
1022 <!-- ======================================================================= -->
1025 declaration in Clang. Declarations in the C family of languages can
1026 take several different forms. Most declarations are named by
1029 names can also name class constructors ("<code>Class</code>"
1032 and conversion functions ("<code>operator void const *</code>"). In
1033 Objective-C, declaration names can refer to the names of Objective-C
1034 methods, which involve the method name and the parameters,
1036 "<code>setWidth:height:</code>". Since all of these kinds of
1037 entities - variables, functions, Objective-C methods, C++
1038 constructors, destructors, and operators - are represented as
1041 represent any kind of name.</p>
1043 <p>Given
1046 stores. There are 8 options (all of the names are inside
1048 <dl>
1050 <dd>The name is a simple
1053 identifier. Note that C++ overloaded operators (e.g.,
1054 "<code>operator+</code>") are represented as special kinds of
1056 function to determine whether an identifier is an overloaded
1057 operator name.</dd>
1059 <dt>ObjCZeroArgSelector, ObjCOneArgSelector,
1060 ObjCMultiArgSelector</dt>
1061 <dd>The name is an Objective-C selector, which can be retrieved as a
1064 kinds for Objective-C reflect an optimization within
1066 one-argument selectors are stored as a
1068 very little space, since zero- and one-argument selectors are far
1069 more common than multi-argument selectors (which use a different
1070 structure).</dd>
1073 <dd>The name is a C++ constructor
1076 construct. The type is always the canonical type, since all
1077 constructors for a given type have the same name.</dd>
1080 <dd>The name is a C++ destructor
1083 named. This type is always a canonical type.</dd>
1086 <dd>The name is a C++ conversion function. Conversion functions are
1087 named according to the type they convert to, e.g., "<code>operator void
1089 the type that this conversion function converts to. This type is
1090 always a canonical type.</dd>
1093 <dd>The name is a C++ overloaded operator name. Overloaded operators
1094 are named according to their spelling, e.g.,
1097 retrieve the overloaded operator (a value of
1099 </dl>
1102 compare. They require only a single pointer's worth of storage in
1103 the common cases (identifiers, zero-
1104 and one-argument Objective-C selectors) and use dense, uniqued
1105 storage for the other kinds of
1108 comparison, can be ordered
1111 normal identifiers but an unspecified ordering for other kinds of
1116 ways depending on what kind of name the instance will store. Normal
1120 destructors, conversion functions, and overloaded operators can be retrieved from
1126 C++ special function names.</p>
1128 <!-- ======================================================================= -->
1130 <!-- ======================================================================= -->
1132 context</i>, such as a translation unit, namespace, class, or
1133 function. Declaration contexts in Clang are represented by
1135 declaration-context AST nodes
1136 (<code>TranslationUnitDecl</code>, <code>NamespaceDecl</code>, <code>RecordDecl</code>, <code>FunctionDecl</code>,
1138 several facilities common to each declaration context:</p>
1139 <dl>
1142 stored within a declaration context. The source-centric view
1143 accurately represents the program source code as written, including
1144 multiple declarations of entities where present (see the
1146 Overloads</a>), while the semantics-centric view represents the
1147 program semantics. The two views are kept synchronized by semantic
1148 analysis while the ASTs are being constructed.</dd>
1151 <dd>Every declaration context can contain some number of
1152 declarations. For example, a C++ class (represented
1154 fields, nested types, and so on. All of these declarations will be
1156 over the declarations via
1159 the source-centric view of declarations in the context.</dd>
1163 lookup for names within that declaration context. For example,
1167 based on a lazily-constructed array (for declaration contexts
1168 with a small number of declarations) or hash table (for
1169 declaration contexts with more declarations). The lookup
1170 operation provides the semantics-centric view of the declarations
1171 in the context.</dd>
1175 were declared within its declaration context, and is responsible
1176 for the management of their memory as well as their
1177 (de-)serialization.</dd>
1178 </dl>
1180 <p>All declarations are stored within a declaration context, and one
1181 can query
1182 information about the context in which each declaration lives. One
1187 Contexts</a> for more information about how to interpret this
1188 context information.</p>
1191 <p>Within a translation unit, it is common for an entity to be
1192 declared several times. For example, we might declare a function "f"
1193 and then later re-declare it as part of an inlined definition:</p>
1195 <pre>
1196 void f(int x, int y, int z = 1);
1198 inline void f(int x, int y, int z) { /* ... */ }
1199 </pre>
1202 semantics-centric views of a declaration context. In the
1203 source-centric view, all redeclarations will be present, in the
1204 order they occurred in the source code, making
1205 this view suitable for clients that wish to see the structure of
1206 the source code. In the semantics-centric view, only the most recent "f"
1207 will be found by the lookup, since it effectively replaces the first
1210 <p>In the semantics-centric view, overloading of functions is
1211 represented explicitly. For example, given two declarations of a
1213 <pre>
1214 void g();
1215 void g(int);
1216 </pre>
1219 declarations of "g". Clients that perform semantic analysis on a
1220 program that is not concerned with the actual source code will
1221 primarily use this semantics-centric view.</p>
1224 <p>Each declaration has two potentially different
1226 the source-centric view of the declaration context, and
1228 semantics-centric view. The lexical context is accessible
1230 semantic context is accessible
1233 contexts are identical. For example:</p>
1235 <pre>
1236 class X {
1237 public:
1238 void f(int x);
1239 };
1240 </pre>
1247 <pre>
1248 void X::f(int x = 17) { /* ... */ }
1249 </pre>
1251 <p>This definition of has different lexical and semantic
1252 contexts. The lexical context corresponds to the declaration
1253 context in which the actual declaration occurred in the source
1256 the declarations provided by
1258 translation unit.</p>
1265 information about the default argument).</p>
1268 <p>In C and C++, there are several contexts in which names that are
1269 logically declared inside another declaration will actually "leak"
1270 out into the enclosing scope from the perspective of name
1271 lookup. The most obvious instance of this behavior is in
1272 enumeration types, e.g.,</p>
1273 <pre>
1274 enum Color {
1275 Red,
1276 Green,
1277 Blue
1278 };
1279 </pre>
1282 context that contains the
1289 without qualifying the name, e.g.,</p>
1291 <pre>
1292 Color c = Red;
1293 </pre>
1295 <p>There are other entities in C++ that provide similar behavior. For
1296 example, linkage specifications that use curly braces:</p>
1298 <pre>
1299 extern "C" {
1300 void f(int);
1301 void g(int);
1302 }
1303 // f and g are visible here
1304 </pre>
1306 <p>For source-level accuracy, we treat the linkage specification and
1307 enumeration type as a
1308 declaration context in which its enclosed declarations ("Red",
1310 are declared. However, these declarations are visible outside of the
1311 scope of the declaration context.</p>
1313 <p>These language features (and several others, described below) have
1314 roughly the same set of
1315 requirements: declarations are declared within a particular lexical
1316 context, but the declarations are also found via name lookup in
1317 scopes enclosing the declaration itself. This feature is implemented
1320 declarations are visible in the nearest enclosing non-transparent
1321 declaration context. This means that the lexical context of the
1322 declaration (e.g., an enumerator) will be the
1324 context, but the declaration will be visible in every outer context
1325 up to and including the first non-transparent declaration context (since
1326 transparent declaration contexts can be nested).</p>
1329 <ul>
1331 <pre>
1332 enum Color {
1333 Red,
1334 Green,
1335 Blue
1336 };
1337 // Red, Green, and Blue are in scope
1338 </pre></li>
1339 <li>C++ linkage specifications:
1340 <pre>
1341 extern "C" {
1342 void f(int);
1343 void g(int);
1344 }
1345 // f and g are in scope
1346 </pre></li>
1347 <li>Anonymous unions and structs:
1348 <pre>
1349 struct LookupTable {
1350 bool IsVector;
1351 union {
1354 };
1355 };
1357 LookupTable LT;
1358 LT.Vector = 0; // Okay: finds Vector inside the unnamed union
1359 </pre>
1360 </li>
1362 <pre>
1363 namespace mylib {
1364 inline namespace debug {
1365 class X;
1366 }
1367 }
1368 mylib::X *xp; // okay: mylib::X refers to mylib::debug::X
1369 </pre>
1370 </li>
1371 </ul>
1375 <p>C++ namespaces have the interesting--and, so far, unique--property that
1376 the namespace can be defined multiple times, and the declarations
1377 provided by each namespace definition are effectively merged (from
1378 the semantic point of view). For example, the following two code
1379 snippets are semantically indistinguishable:</p>
1380 <pre>
1381 // Snippet #1:
1382 namespace N {
1383 void f();
1384 }
1385 namespace N {
1386 void f(int);
1387 }
1389 // Snippet #2:
1390 namespace N {
1391 void f();
1392 void f(int);
1393 }
1394 </pre>
1396 <p>In Clang's representation, the source-centric view of declaration
1398 nodes in Snippet #1, each of which is a declaration context that
1399 contains a single declaration of "f". However, the semantics-centric
1405 contexts internally. The
1409 the lookup table used for the semantics-centric view. Given the
1410 primary context, one can follow the chain
1413 these functions are used internally within the lookup and insertion
1415 clients can ignore them.</p>
1417 <!-- ======================================================================= -->
1419 <!-- ======================================================================= -->
1426 which includes simple expressions. Control-flow graphs are especially
1427 useful for performing
1431 <!-- ============ -->
1433 <!-- ============ -->
1438 to statements in the AST). The ordering of statements within a block
1439 indicates unconditional flow of control from one statement to the
1441 is represented using edges between basic blocks. The statements
1445 <p>
1448 CFG is also uniquely numbered (accessible
1450 based on the ordering the blocks were created, but no assumptions
1453 the number of basic blocks in the CFG).</p>
1455 <!-- ===================== -->
1457 <!-- ===================== -->
1463 block contains any statements, and they serve the role of providing a
1464 clear entrance and exit for a body of code such as a function body.
1465 The presence of these empty blocks greatly simplifies the
1466 implementation of many analyses built on top of CFGs.
1468 <!-- ===================================================== -->
1470 <!-- ===================================================== -->
1472 <p>Conditional control-flow (such as those induced by if-statements
1474 Because different C language constructs can induce control-flow,
1477 the statement that caused the control-flow, and is used to identify
1478 the nature of the conditional control-flow between blocks. For
1479 example, in the case of an if-statement, the terminator refers to
1481 branch.</p>
1485 <code>
1486 int foo(int x) {<br>
1488 <br>
1494 <br>
1496 }
1497 </code>
1499 <p>After invoking the parser+semantic analyzer on this code fragment,
1503 body by single call to a static class method:</p>
1505 <code>
1508 </code>
1512 longer needed.</p>
1514 <p>Along with providing an interface to iterate over
1516 that are useful for debugging and visualizing CFGs. For example, the
1517 method
1519 standard error. This is especially useful when one is using a
1520 debugger such as gdb. For example, here is the output
1523 <code>
1527 <br>
1534 <br>
1539 <br>
1545 <br>
1550 <br>
1554 </code>
1556 <p>For each block, the pretty-printed output displays for each block
1559 that have control-flow that have incoming control-flow from the given
1560 block). We can also clearly see the special entry and exit blocks at
1561 the beginning and end of the pretty-printed output. For the entry
1562 block (block B5), the number of predecessor blocks is 0, while for the
1565 <p>The most interesting block here is B4, whose outgoing control-flow
1566 represents the branching caused by the sole if-statement
1570 evaluation of the condition of the if-statement, which occurs before
1576 refer to proper C statements.</p>
1579 object in the AST. The pretty-printer outputs <b><tt>if
1580 [B4.2]</tt></b> because the condition expression of the if-statement
1581 has an actual place in the basic block, and thus the terminator is
1582 essentially
1584 block B4 (i.e., B4.2). In this manner, conditions for control-flow
1585 (which also includes conditions for loops and switch statements) are
1586 hoisted into the actual basic block.</p>
1588 <!-- ===================== -->
1589 <!-- <h4>Implicit Control-Flow</h4> -->
1590 <!-- ===================== -->
1592 <!--
1593 <p>A key design principle of the <tt>CFG</tt> class was to not require
1594 any transformations to the AST in order to represent control-flow.
1595 Thus the <tt>CFG</tt> does not perform any "lowering" of the
1596 statements in an AST: loops are not transformed into guarded gotos,
1597 short-circuit operations are not converted to a set of if-statements,
1598 and so on.</p>
1599 -->
1602 <!-- ======================================================================= -->
1604 <!-- ======================================================================= -->
1606 <p>There are several places where constants and constant folding matter a lot to
1607 the Clang front-end. First, in general, we prefer the AST to retain the source
1608 code as close to how the user wrote it as possible. This means that if they
1609 wrote "5+4", we want to keep the addition and two constants in the AST, we don't
1610 want to fold to "9". This means that constant folding in various ways turns
1611 into a tree walk that needs to handle the various cases.</p>
1613 <p>However, there are places in both C and C++ that require constants to be
1614 folded. For example, the C standard defines what an "integer constant
1615 expression" (i-c-e) is with very precise and specific requirements. The
1616 language then requires i-c-e's in a lot of places (for example, the size of a
1617 bitfield, the value for a case statement, etc). For these, we have to be able
1618 to constant fold the constants, to do semantic checks (e.g. verify bitfield size
1619 is non-negative and that case statements aren't duplicated). We aim for Clang
1620 to be very pedantic about this, diagnosing cases when the code does not use an
1621 i-c-e where one is required, but accepting the code unless running with
1624 <p>Things get a little bit more tricky when it comes to compatibility with
1625 real-world source code. Specifically, GCC has historically accepted a huge
1626 superset of expressions as i-c-e's, and a lot of real world code depends on this
1627 unfortuate accident of history (including, e.g., the glibc system headers). GCC
1628 accepts anything its "fold" optimizer is capable of reducing to an integer
1629 constant, which means that the definition of what it accepts changes as its
1630 optimizer does. One example is that GCC accepts things like "case X-X:" even
1633 <p>Another issue are how constants interact with the extensions we support, such
1634 as __builtin_constant_p, __builtin_inf, __extension__ and many others. C99
1635 obviously does not specify the semantics of any of these extensions, and the
1636 definition of i-c-e does not include them. However, these extensions are often
1637 used in real code, and we have to have a way to reason about them.</p>
1639 <p>Finally, this is not just a problem for semantic analysis. The code
1640 generator and other clients have to be able to fold constants (e.g. to
1641 initialize global variables) and has to handle a superset of what C99 allows.
1642 Further, these clients can benefit from extended information. For example, we
1643 know that "foo()||1" always evaluates to true, but we can't replace the
1644 expression with true because it has side effects.</p>
1646 <!-- ======================= -->
1648 <!-- ======================= -->
1650 <p>After trying several different approaches, we've finally converged on a
1651 design (Note, at the time of this writing, not all of this has been implemented,
1652 consider this a design goal!). Our basic approach is to define a single
1655 type (integer, fp, complex, or pointer) this method returns the following
1656 information:</p>
1658 <ul>
1659 <li>Whether the expression is an integer constant expression, a general
1660 constant that was folded but has no side effects, a general constant that
1661 was folded but that does have side effects, or an uncomputable/unfoldable
1662 value.
1663 </li>
1664 <li>If the expression was computable in any way, this method returns the APValue
1665 for the result of the expression.</li>
1666 <li>If the expression is not evaluatable at all, 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 ERROR type.</li>
1670 <li>If the expression is not an integer constant expression, this method returns
1671 information on one of the problems with the expression. This includes a
1672 SourceLocation for where the problem is, and a diagnostic ID that explains
1673 the problem. The diagnostic should be have EXTENSION type.</li>
1674 </ul>
1676 <p>This information gives various clients the flexibility that they want, and we
1677 will eventually have some helper methods for various extensions. For example,
1679 calls Evaluate on the expression. If the expression is not foldable, the error
1680 is emitted, and it would return true. If the expression is not an i-c-e, the
1681 EXTENSION diagnostic is emitted. Finally it would return false to indicate that
1682 the AST is ok.</p>
1684 <p>Other clients can use the information in other ways, for example, codegen can
1685 just use expressions that are foldable in any way.</p>
1687 <!-- ========== -->
1689 <!-- ========== -->
1691 <p>This section describes how some of the various extensions Clang supports
1692 interacts with constant evaluation:</p>
1694 <ul>
1696 any evaluatable subexpression to be accepted as an integer constant
1697 expression.</li>
1699 constant expression) if the operand is any evaluatable constant. As a
1701 parenthesized) condition of a conditional operator expression ("?:"), only
1702 the true side of the conditional operator is considered, and it is evaluated
1703 with full constant folding.</li>
1705 integer constant expression, but we accept any constant as an "extension of
1706 an extension". This only evaluates one operand depending on which way the
1707 condition evaluates.</li>
1709 constant expression.</li>
1711 floating-point literal.</li>
1713 general constant expressions.</li>
1714 <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>
1715 </ul>
1718 <!-- ======================================================================= -->
1720 <!-- ======================================================================= -->
1722 <!-- ======================================================================= -->
1724 <!-- ======================================================================= -->
1726 <p>To add an attribute, you'll have to add it to the list of attributes, add it
1727 to the parsing phase, and look for it in the AST scan.
1729 has a good example of adding a warning attribute.</p>
1731 <p>(Beware that this hasn't been reviewed/fixed by the people who designed the
1732 attributes system yet.)</p>
1734 <h4><a
1735 href="http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/Attr.td?view=markup">include/clang/Basic/Attr.td</a></h4>
1739 to subsequent declarations of the same name.</p>
1746 string for a namespace will allow users to write just the spelling
1750 can appertain (roughly, attach).</p>
1759 href="http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Sema/AttributeList.h?view=markup">include/clang/Sema/AttributeList.h</a>
1765 in <a
1766 href="http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Sema/AttributeList.cpp?view=markup">lib/Sema/AttributeList.cpp</a>
1767 for each spelling of your attribute. Less common attributes should come toward
1768 the end of that list.</p>
1771 href="http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Sema/SemaDeclAttr.cpp?view=markup">lib/Sema/SemaDeclAttr.cpp</a>,
1776 in <a
1777 href="http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/DiagnosticGroups.td?view=markup">include/clang/Basic/DiagnosticGroups.td</a>
1781 href="http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/DiagnosticSemaKinds.td?view=markup">DiagnosticSemaKinds.td</a></p>
1785 <p>Find an appropriate place in Clang to do whatever your attribute needs to do.
1789 document to describe your new attribute.</p>
1791 </div>
1792 </body>
1793 </html>