Pragma.cpp revision 2b2453a7d8fe732561795431f39ceb2b2a832d84
1//===--- Pragma.cpp - Pragma registration and handling --------------------===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file implements the PragmaHandler/PragmaTable interfaces and implements
11// pragma related methods of the Preprocessor class.
12//
13//===----------------------------------------------------------------------===//
14
15#include "clang/Lex/Pragma.h"
16#include "clang/Lex/HeaderSearch.h"
17#include "clang/Lex/LiteralSupport.h"
18#include "clang/Lex/Preprocessor.h"
19#include "clang/Basic/Diagnostic.h"
20#include "clang/Basic/FileManager.h"
21#include "clang/Basic/SourceManager.h"
22using namespace clang;
23
24// Out-of-line destructor to provide a home for the class.
25PragmaHandler::~PragmaHandler() {
26}
27
28//===----------------------------------------------------------------------===//
29// PragmaNamespace Implementation.
30//===----------------------------------------------------------------------===//
31
32
33PragmaNamespace::~PragmaNamespace() {
34  for (unsigned i = 0, e = Handlers.size(); i != e; ++i)
35    delete Handlers[i];
36}
37
38/// FindHandler - Check to see if there is already a handler for the
39/// specified name.  If not, return the handler for the null identifier if it
40/// exists, otherwise return null.  If IgnoreNull is true (the default) then
41/// the null handler isn't returned on failure to match.
42PragmaHandler *PragmaNamespace::FindHandler(const IdentifierInfo *Name,
43                                            bool IgnoreNull) const {
44  PragmaHandler *NullHandler = 0;
45  for (unsigned i = 0, e = Handlers.size(); i != e; ++i) {
46    if (Handlers[i]->getName() == Name)
47      return Handlers[i];
48
49    if (Handlers[i]->getName() == 0)
50      NullHandler = Handlers[i];
51  }
52  return IgnoreNull ? 0 : NullHandler;
53}
54
55void PragmaNamespace::RemovePragmaHandler(PragmaHandler *Handler) {
56  for (unsigned i = 0, e = Handlers.size(); i != e; ++i) {
57    if (Handlers[i] == Handler) {
58      Handlers[i] = Handlers.back();
59      Handlers.pop_back();
60      return;
61    }
62  }
63  assert(0 && "Handler not registered in this namespace");
64}
65
66void PragmaNamespace::HandlePragma(Preprocessor &PP, Token &Tok) {
67  // Read the 'namespace' that the directive is in, e.g. STDC.  Do not macro
68  // expand it, the user can have a STDC #define, that should not affect this.
69  PP.LexUnexpandedToken(Tok);
70
71  // Get the handler for this token.  If there is no handler, ignore the pragma.
72  PragmaHandler *Handler = FindHandler(Tok.getIdentifierInfo(), false);
73  if (Handler == 0) return;
74
75  // Otherwise, pass it down.
76  Handler->HandlePragma(PP, Tok);
77}
78
79//===----------------------------------------------------------------------===//
80// Preprocessor Pragma Directive Handling.
81//===----------------------------------------------------------------------===//
82
83/// HandlePragmaDirective - The "#pragma" directive has been parsed.  Lex the
84/// rest of the pragma, passing it to the registered pragma handlers.
85void Preprocessor::HandlePragmaDirective() {
86  ++NumPragma;
87
88  // Invoke the first level of pragma handlers which reads the namespace id.
89  Token Tok;
90  PragmaHandlers->HandlePragma(*this, Tok);
91
92  // If the pragma handler didn't read the rest of the line, consume it now.
93  if (CurPPLexer->ParsingPreprocessorDirective)
94    DiscardUntilEndOfDirective();
95}
96
97/// Handle_Pragma - Read a _Pragma directive, slice it up, process it, then
98/// return the first token after the directive.  The _Pragma token has just
99/// been read into 'Tok'.
100void Preprocessor::Handle_Pragma(Token &Tok) {
101  // Remember the pragma token location.
102  SourceLocation PragmaLoc = Tok.getLocation();
103
104  // Read the '('.
105  Lex(Tok);
106  if (Tok.isNot(tok::l_paren)) {
107    Diag(PragmaLoc, diag::err__Pragma_malformed);
108    return;
109  }
110
111  // Read the '"..."'.
112  Lex(Tok);
113  if (Tok.isNot(tok::string_literal) && Tok.isNot(tok::wide_string_literal)) {
114    Diag(PragmaLoc, diag::err__Pragma_malformed);
115    return;
116  }
117
118  // Remember the string.
119  std::string StrVal = getSpelling(Tok);
120  SourceLocation StrLoc = Tok.getLocation();
121
122  // Read the ')'.
123  Lex(Tok);
124  if (Tok.isNot(tok::r_paren)) {
125    Diag(PragmaLoc, diag::err__Pragma_malformed);
126    return;
127  }
128
129  // The _Pragma is lexically sound.  Destringize according to C99 6.10.9.1:
130  // "The string literal is destringized by deleting the L prefix, if present,
131  // deleting the leading and trailing double-quotes, replacing each escape
132  // sequence \" by a double-quote, and replacing each escape sequence \\ by a
133  // single backslash."
134  if (StrVal[0] == 'L')  // Remove L prefix.
135    StrVal.erase(StrVal.begin());
136  assert(StrVal[0] == '"' && StrVal[StrVal.size()-1] == '"' &&
137         "Invalid string token!");
138
139  // Remove the front quote, replacing it with a space, so that the pragma
140  // contents appear to have a space before them.
141  StrVal[0] = ' ';
142
143  // Replace the terminating quote with a \n\0.
144  StrVal[StrVal.size()-1] = '\n';
145  StrVal += '\0';
146
147  // Remove escaped quotes and escapes.
148  for (unsigned i = 0, e = StrVal.size(); i != e-1; ++i) {
149    if (StrVal[i] == '\\' &&
150        (StrVal[i+1] == '\\' || StrVal[i+1] == '"')) {
151      // \\ -> '\' and \" -> '"'.
152      StrVal.erase(StrVal.begin()+i);
153      --e;
154    }
155  }
156
157  // Plop the string (including the newline and trailing null) into a buffer
158  // where we can lex it.
159  SourceLocation TokLoc = CreateString(&StrVal[0], StrVal.size(), StrLoc);
160  const char *StrData = SourceMgr.getCharacterData(TokLoc);
161
162  // Make and enter a lexer object so that we lex and expand the tokens just
163  // like any others.
164  Lexer *TL = new Lexer(TokLoc, *this,
165                        StrData, StrData+StrVal.size()-1 /* no null */);
166
167  // Ensure that the lexer thinks it is inside a directive, so that end \n will
168  // return an EOM token.
169  TL->ParsingPreprocessorDirective = true;
170
171  // This lexer really is for _Pragma.
172  TL->Is_PragmaLexer = true;
173
174  EnterSourceFileWithLexer(TL, 0);
175
176  // With everything set up, lex this as a #pragma directive.
177  HandlePragmaDirective();
178
179  // Finally, return whatever came after the pragma directive.
180  return Lex(Tok);
181}
182
183
184
185/// HandlePragmaOnce - Handle #pragma once.  OnceTok is the 'once'.
186///
187void Preprocessor::HandlePragmaOnce(Token &OnceTok) {
188  if (isInPrimaryFile()) {
189    Diag(OnceTok, diag::pp_pragma_once_in_main_file);
190    return;
191  }
192
193  // Get the current file lexer we're looking at.  Ignore _Pragma 'files' etc.
194  // Mark the file as a once-only file now.
195  HeaderInfo.MarkFileIncludeOnce(getCurrentFileLexer()->getFileEntry());
196}
197
198void Preprocessor::HandlePragmaMark() {
199  assert(CurPPLexer && "No current lexer?");
200  if (CurLexer) CurLexer->ReadToEndOfLine();
201  else CurPTHLexer->DiscardToEndOfLine();
202}
203
204
205/// HandlePragmaPoison - Handle #pragma GCC poison.  PoisonTok is the 'poison'.
206///
207void Preprocessor::HandlePragmaPoison(Token &PoisonTok) {
208  Token Tok;
209
210  while (1) {
211    // Read the next token to poison.  While doing this, pretend that we are
212    // skipping while reading the identifier to poison.
213    // This avoids errors on code like:
214    //   #pragma GCC poison X
215    //   #pragma GCC poison X
216    if (CurPPLexer) CurPPLexer->LexingRawMode = true;
217    LexUnexpandedToken(Tok);
218    if (CurPPLexer) CurPPLexer->LexingRawMode = false;
219
220    // If we reached the end of line, we're done.
221    if (Tok.is(tok::eom)) return;
222
223    // Can only poison identifiers.
224    if (Tok.isNot(tok::identifier)) {
225      Diag(Tok, diag::err_pp_invalid_poison);
226      return;
227    }
228
229    // Look up the identifier info for the token.  We disabled identifier lookup
230    // by saying we're skipping contents, so we need to do this manually.
231    IdentifierInfo *II = LookUpIdentifierInfo(Tok);
232
233    // Already poisoned.
234    if (II->isPoisoned()) continue;
235
236    // If this is a macro identifier, emit a warning.
237    if (II->hasMacroDefinition())
238      Diag(Tok, diag::pp_poisoning_existing_macro);
239
240    // Finally, poison it!
241    II->setIsPoisoned();
242  }
243}
244
245/// HandlePragmaSystemHeader - Implement #pragma GCC system_header.  We know
246/// that the whole directive has been parsed.
247void Preprocessor::HandlePragmaSystemHeader(Token &SysHeaderTok) {
248  if (isInPrimaryFile()) {
249    Diag(SysHeaderTok, diag::pp_pragma_sysheader_in_main_file);
250    return;
251  }
252
253  // Get the current file lexer we're looking at.  Ignore _Pragma 'files' etc.
254  PreprocessorLexer *TheLexer = getCurrentFileLexer();
255
256  // Mark the file as a system header.
257  HeaderInfo.MarkFileSystemHeader(TheLexer->getFileEntry());
258
259  // Notify the client, if desired, that we are in a new source file.
260  if (Callbacks)
261    Callbacks->FileChanged(SysHeaderTok.getLocation(),
262                           PPCallbacks::SystemHeaderPragma, SrcMgr::C_System);
263}
264
265/// HandlePragmaDependency - Handle #pragma GCC dependency "foo" blah.
266///
267void Preprocessor::HandlePragmaDependency(Token &DependencyTok) {
268  Token FilenameTok;
269  CurPPLexer->LexIncludeFilename(FilenameTok);
270
271  // If the token kind is EOM, the error has already been diagnosed.
272  if (FilenameTok.is(tok::eom))
273    return;
274
275  // Reserve a buffer to get the spelling.
276  llvm::SmallVector<char, 128> FilenameBuffer;
277  FilenameBuffer.resize(FilenameTok.getLength());
278
279  const char *FilenameStart = &FilenameBuffer[0];
280  unsigned Len = getSpelling(FilenameTok, FilenameStart);
281  const char *FilenameEnd = FilenameStart+Len;
282  bool isAngled = GetIncludeFilenameSpelling(FilenameTok.getLocation(),
283                                             FilenameStart, FilenameEnd);
284  // If GetIncludeFilenameSpelling set the start ptr to null, there was an
285  // error.
286  if (FilenameStart == 0)
287    return;
288
289  // Search include directories for this file.
290  const DirectoryLookup *CurDir;
291  const FileEntry *File = LookupFile(FilenameStart, FilenameEnd,
292                                     isAngled, 0, CurDir);
293  if (File == 0) {
294    Diag(FilenameTok, diag::err_pp_file_not_found)
295      << std::string(FilenameStart, FilenameEnd);
296    return;
297  }
298
299  const FileEntry *CurFile = getCurrentFileLexer()->getFileEntry();
300
301  // If this file is older than the file it depends on, emit a diagnostic.
302  if (CurFile && CurFile->getModificationTime() < File->getModificationTime()) {
303    // Lex tokens at the end of the message and include them in the message.
304    std::string Message;
305    Lex(DependencyTok);
306    while (DependencyTok.isNot(tok::eom)) {
307      Message += getSpelling(DependencyTok) + " ";
308      Lex(DependencyTok);
309    }
310
311    Message.erase(Message.end()-1);
312    Diag(FilenameTok, diag::pp_out_of_date_dependency) << Message;
313  }
314}
315
316/// HandlePragmaComment - Handle the microsoft #pragma comment extension.  The
317/// syntax is:
318///   #pragma comment(linker, "foo")
319/// 'linker' is one of five identifiers: compiler, exestr, lib, linker, user.
320/// "foo" is a string, which is fully macro expanded, and permits string
321/// concatenation, embeded escape characters etc.  See MSDN for more details.
322void Preprocessor::HandlePragmaComment(Token &Tok) {
323  SourceLocation CommentLoc = Tok.getLocation();
324  Lex(Tok);
325  if (Tok.isNot(tok::l_paren)) {
326    Diag(CommentLoc, diag::err_pragma_comment_malformed);
327    return;
328  }
329
330  // Read the identifier.
331  Lex(Tok);
332  if (Tok.isNot(tok::identifier)) {
333    Diag(CommentLoc, diag::err_pragma_comment_malformed);
334    return;
335  }
336
337  // Verify that this is one of the 5 whitelisted options.
338  // FIXME: warn that 'exestr' is deprecated.
339  const IdentifierInfo *II = Tok.getIdentifierInfo();
340  if (!II->isStr("compiler") && !II->isStr("exestr") && !II->isStr("lib") &&
341      !II->isStr("linker") && !II->isStr("user")) {
342    Diag(Tok.getLocation(), diag::err_pragma_comment_unknown_kind);
343    return;
344  }
345
346  // Read the optional string if present.
347  Lex(Tok);
348  std::string ArgumentString;
349  if (Tok.is(tok::comma)) {
350    Lex(Tok); // eat the comma.
351
352    // We need at least one string.
353    if (Tok.getKind() != tok::string_literal) {
354      Diag(Tok.getLocation(), diag::err_pragma_comment_malformed);
355      return;
356    }
357
358    // String concatenation allows multiple strings, which can even come from
359    // macro expansion.
360    // "foo " "bar" "Baz"
361    llvm::SmallVector<Token, 4> StrToks;
362    while (Tok.getKind() == tok::string_literal) {
363      StrToks.push_back(Tok);
364      Lex(Tok);
365    }
366
367    // Concatenate and parse the strings.
368    StringLiteralParser Literal(&StrToks[0], StrToks.size(), *this);
369    assert(!Literal.AnyWide && "Didn't allow wide strings in");
370    if (Literal.hadError)
371      return;
372    if (Literal.Pascal) {
373      Diag(StrToks[0].getLocation(), diag::err_pragma_comment_malformed);
374      return;
375    }
376
377    ArgumentString = std::string(Literal.GetString(),
378                                 Literal.GetString()+Literal.GetStringLength());
379  }
380
381  // FIXME: If the kind is "compiler" warn if the string is present (it is
382  // ignored).
383  // FIXME: 'lib' requires a comment string.
384  // FIXME: 'linker' requires a comment string, and has a specific list of
385  // things that are allowable.
386
387  if (Tok.isNot(tok::r_paren)) {
388    Diag(Tok.getLocation(), diag::err_pragma_comment_malformed);
389    return;
390  }
391  Lex(Tok);  // eat the r_paren.
392
393  if (Tok.isNot(tok::eom)) {
394    Diag(Tok.getLocation(), diag::err_pragma_comment_malformed);
395    return;
396  }
397
398  // If the pragma is lexically sound, notify any interested PPCallbacks.
399  if (Callbacks)
400    Callbacks->PragmaComment(CommentLoc, II, ArgumentString);
401}
402
403
404
405
406/// AddPragmaHandler - Add the specified pragma handler to the preprocessor.
407/// If 'Namespace' is non-null, then it is a token required to exist on the
408/// pragma line before the pragma string starts, e.g. "STDC" or "GCC".
409void Preprocessor::AddPragmaHandler(const char *Namespace,
410                                    PragmaHandler *Handler) {
411  PragmaNamespace *InsertNS = PragmaHandlers;
412
413  // If this is specified to be in a namespace, step down into it.
414  if (Namespace) {
415    IdentifierInfo *NSID = getIdentifierInfo(Namespace);
416
417    // If there is already a pragma handler with the name of this namespace,
418    // we either have an error (directive with the same name as a namespace) or
419    // we already have the namespace to insert into.
420    if (PragmaHandler *Existing = PragmaHandlers->FindHandler(NSID)) {
421      InsertNS = Existing->getIfNamespace();
422      assert(InsertNS != 0 && "Cannot have a pragma namespace and pragma"
423             " handler with the same name!");
424    } else {
425      // Otherwise, this namespace doesn't exist yet, create and insert the
426      // handler for it.
427      InsertNS = new PragmaNamespace(NSID);
428      PragmaHandlers->AddPragma(InsertNS);
429    }
430  }
431
432  // Check to make sure we don't already have a pragma for this identifier.
433  assert(!InsertNS->FindHandler(Handler->getName()) &&
434         "Pragma handler already exists for this identifier!");
435  InsertNS->AddPragma(Handler);
436}
437
438/// RemovePragmaHandler - Remove the specific pragma handler from the
439/// preprocessor. If \arg Namespace is non-null, then it should be the
440/// namespace that \arg Handler was added to. It is an error to remove
441/// a handler that has not been registered.
442void Preprocessor::RemovePragmaHandler(const char *Namespace,
443                                       PragmaHandler *Handler) {
444  PragmaNamespace *NS = PragmaHandlers;
445
446  // If this is specified to be in a namespace, step down into it.
447  if (Namespace) {
448    IdentifierInfo *NSID = getIdentifierInfo(Namespace);
449    PragmaHandler *Existing = PragmaHandlers->FindHandler(NSID);
450    assert(Existing && "Namespace containing handler does not exist!");
451
452    NS = Existing->getIfNamespace();
453    assert(NS && "Invalid namespace, registered as a regular pragma handler!");
454  }
455
456  NS->RemovePragmaHandler(Handler);
457
458  // If this is a non-default namespace and it is now empty, remove
459  // it.
460  if (NS != PragmaHandlers && NS->IsEmpty())
461    PragmaHandlers->RemovePragmaHandler(NS);
462}
463
464namespace {
465/// PragmaOnceHandler - "#pragma once" marks the file as atomically included.
466struct PragmaOnceHandler : public PragmaHandler {
467  PragmaOnceHandler(const IdentifierInfo *OnceID) : PragmaHandler(OnceID) {}
468  virtual void HandlePragma(Preprocessor &PP, Token &OnceTok) {
469    PP.CheckEndOfDirective("#pragma once");
470    PP.HandlePragmaOnce(OnceTok);
471  }
472};
473
474/// PragmaMarkHandler - "#pragma mark ..." is ignored by the compiler, and the
475/// rest of the line is not lexed.
476struct PragmaMarkHandler : public PragmaHandler {
477  PragmaMarkHandler(const IdentifierInfo *MarkID) : PragmaHandler(MarkID) {}
478  virtual void HandlePragma(Preprocessor &PP, Token &MarkTok) {
479    PP.HandlePragmaMark();
480  }
481};
482
483/// PragmaPoisonHandler - "#pragma poison x" marks x as not usable.
484struct PragmaPoisonHandler : public PragmaHandler {
485  PragmaPoisonHandler(const IdentifierInfo *ID) : PragmaHandler(ID) {}
486  virtual void HandlePragma(Preprocessor &PP, Token &PoisonTok) {
487    PP.HandlePragmaPoison(PoisonTok);
488  }
489};
490
491/// PragmaSystemHeaderHandler - "#pragma system_header" marks the current file
492/// as a system header, which silences warnings in it.
493struct PragmaSystemHeaderHandler : public PragmaHandler {
494  PragmaSystemHeaderHandler(const IdentifierInfo *ID) : PragmaHandler(ID) {}
495  virtual void HandlePragma(Preprocessor &PP, Token &SHToken) {
496    PP.HandlePragmaSystemHeader(SHToken);
497    PP.CheckEndOfDirective("#pragma");
498  }
499};
500struct PragmaDependencyHandler : public PragmaHandler {
501  PragmaDependencyHandler(const IdentifierInfo *ID) : PragmaHandler(ID) {}
502  virtual void HandlePragma(Preprocessor &PP, Token &DepToken) {
503    PP.HandlePragmaDependency(DepToken);
504  }
505};
506
507/// PragmaCommentHandler - "#pragma comment ...".
508struct PragmaCommentHandler : public PragmaHandler {
509  PragmaCommentHandler(const IdentifierInfo *ID) : PragmaHandler(ID) {}
510  virtual void HandlePragma(Preprocessor &PP, Token &CommentTok) {
511    PP.HandlePragmaComment(CommentTok);
512  }
513};
514}  // end anonymous namespace
515
516
517/// RegisterBuiltinPragmas - Install the standard preprocessor pragmas:
518/// #pragma GCC poison/system_header/dependency and #pragma once.
519void Preprocessor::RegisterBuiltinPragmas() {
520  AddPragmaHandler(0, new PragmaOnceHandler(getIdentifierInfo("once")));
521  AddPragmaHandler(0, new PragmaMarkHandler(getIdentifierInfo("mark")));
522  AddPragmaHandler("GCC", new PragmaPoisonHandler(getIdentifierInfo("poison")));
523  AddPragmaHandler("GCC", new PragmaSystemHeaderHandler(
524                                          getIdentifierInfo("system_header")));
525  AddPragmaHandler("GCC", new PragmaDependencyHandler(
526                                          getIdentifierInfo("dependency")));
527
528  // MS extensions.
529  if (Features.Microsoft)
530    AddPragmaHandler(0, new PragmaCommentHandler(getIdentifierInfo("comment")));
531}
532