SemaChecking.cpp revision a4923eb7c4b04d360cb2747641a5e92818edf804
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===//
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 extra semantic analysis beyond what is enforced
11//  by the C type system.
12//
13//===----------------------------------------------------------------------===//
14
15#include "Sema.h"
16#include "clang/AST/ASTContext.h"
17#include "clang/AST/DeclObjC.h"
18#include "clang/AST/ExprCXX.h"
19#include "clang/AST/ExprObjC.h"
20#include "clang/Lex/LiteralSupport.h"
21#include "clang/Lex/Preprocessor.h"
22#include <limits>
23using namespace clang;
24
25/// getLocationOfStringLiteralByte - Return a source location that points to the
26/// specified byte of the specified string literal.
27///
28/// Strings are amazingly complex.  They can be formed from multiple tokens and
29/// can have escape sequences in them in addition to the usual trigraph and
30/// escaped newline business.  This routine handles this complexity.
31///
32SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL,
33                                                    unsigned ByteNo) const {
34  assert(!SL->isWide() && "This doesn't work for wide strings yet");
35
36  // Loop over all of the tokens in this string until we find the one that
37  // contains the byte we're looking for.
38  unsigned TokNo = 0;
39  while (1) {
40    assert(TokNo < SL->getNumConcatenated() && "Invalid byte number!");
41    SourceLocation StrTokLoc = SL->getStrTokenLoc(TokNo);
42
43    // Get the spelling of the string so that we can get the data that makes up
44    // the string literal, not the identifier for the macro it is potentially
45    // expanded through.
46    SourceLocation StrTokSpellingLoc = SourceMgr.getSpellingLoc(StrTokLoc);
47
48    // Re-lex the token to get its length and original spelling.
49    std::pair<FileID, unsigned> LocInfo =
50      SourceMgr.getDecomposedLoc(StrTokSpellingLoc);
51    std::pair<const char *,const char *> Buffer =
52      SourceMgr.getBufferData(LocInfo.first);
53    const char *StrData = Buffer.first+LocInfo.second;
54
55    // Create a langops struct and enable trigraphs.  This is sufficient for
56    // relexing tokens.
57    LangOptions LangOpts;
58    LangOpts.Trigraphs = true;
59
60    // Create a lexer starting at the beginning of this token.
61    Lexer TheLexer(StrTokSpellingLoc, LangOpts, Buffer.first, StrData,
62                   Buffer.second);
63    Token TheTok;
64    TheLexer.LexFromRawLexer(TheTok);
65
66    // Use the StringLiteralParser to compute the length of the string in bytes.
67    StringLiteralParser SLP(&TheTok, 1, PP);
68    unsigned TokNumBytes = SLP.GetStringLength();
69
70    // If the byte is in this token, return the location of the byte.
71    if (ByteNo < TokNumBytes ||
72        (ByteNo == TokNumBytes && TokNo == SL->getNumConcatenated())) {
73      unsigned Offset =
74        StringLiteralParser::getOffsetOfStringByte(TheTok, ByteNo, PP);
75
76      // Now that we know the offset of the token in the spelling, use the
77      // preprocessor to get the offset in the original source.
78      return PP.AdvanceToTokenCharacter(StrTokLoc, Offset);
79    }
80
81    // Move to the next string token.
82    ++TokNo;
83    ByteNo -= TokNumBytes;
84  }
85}
86
87/// CheckablePrintfAttr - does a function call have a "printf" attribute
88/// and arguments that merit checking?
89bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) {
90  if (Format->getType() == "printf") return true;
91  if (Format->getType() == "printf0") {
92    // printf0 allows null "format" string; if so don't check format/args
93    unsigned format_idx = Format->getFormatIdx() - 1;
94    if (format_idx < TheCall->getNumArgs()) {
95      Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts();
96      if (!Format->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNull))
97        return true;
98    }
99  }
100  return false;
101}
102
103Action::OwningExprResult
104Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
105  OwningExprResult TheCallResult(Owned(TheCall));
106
107  switch (BuiltinID) {
108  case Builtin::BI__builtin___CFStringMakeConstantString:
109    assert(TheCall->getNumArgs() == 1 &&
110           "Wrong # arguments to builtin CFStringMakeConstantString");
111    if (CheckObjCString(TheCall->getArg(0)))
112      return ExprError();
113    break;
114  case Builtin::BI__builtin_stdarg_start:
115  case Builtin::BI__builtin_va_start:
116    if (SemaBuiltinVAStart(TheCall))
117      return ExprError();
118    break;
119  case Builtin::BI__builtin_isgreater:
120  case Builtin::BI__builtin_isgreaterequal:
121  case Builtin::BI__builtin_isless:
122  case Builtin::BI__builtin_islessequal:
123  case Builtin::BI__builtin_islessgreater:
124  case Builtin::BI__builtin_isunordered:
125    if (SemaBuiltinUnorderedCompare(TheCall))
126      return ExprError();
127    break;
128  case Builtin::BI__builtin_isfinite:
129  case Builtin::BI__builtin_isinf:
130  case Builtin::BI__builtin_isinf_sign:
131  case Builtin::BI__builtin_isnan:
132  case Builtin::BI__builtin_isnormal:
133    if (SemaBuiltinUnaryFP(TheCall))
134      return ExprError();
135    break;
136  case Builtin::BI__builtin_return_address:
137  case Builtin::BI__builtin_frame_address:
138    if (SemaBuiltinStackAddress(TheCall))
139      return ExprError();
140    break;
141  case Builtin::BI__builtin_eh_return_data_regno:
142    if (SemaBuiltinEHReturnDataRegNo(TheCall))
143      return ExprError();
144    break;
145  case Builtin::BI__builtin_shufflevector:
146    return SemaBuiltinShuffleVector(TheCall);
147    // TheCall will be freed by the smart pointer here, but that's fine, since
148    // SemaBuiltinShuffleVector guts it, but then doesn't release it.
149  case Builtin::BI__builtin_prefetch:
150    if (SemaBuiltinPrefetch(TheCall))
151      return ExprError();
152    break;
153  case Builtin::BI__builtin_object_size:
154    if (SemaBuiltinObjectSize(TheCall))
155      return ExprError();
156    break;
157  case Builtin::BI__builtin_longjmp:
158    if (SemaBuiltinLongjmp(TheCall))
159      return ExprError();
160    break;
161  case Builtin::BI__sync_fetch_and_add:
162  case Builtin::BI__sync_fetch_and_sub:
163  case Builtin::BI__sync_fetch_and_or:
164  case Builtin::BI__sync_fetch_and_and:
165  case Builtin::BI__sync_fetch_and_xor:
166  case Builtin::BI__sync_fetch_and_nand:
167  case Builtin::BI__sync_add_and_fetch:
168  case Builtin::BI__sync_sub_and_fetch:
169  case Builtin::BI__sync_and_and_fetch:
170  case Builtin::BI__sync_or_and_fetch:
171  case Builtin::BI__sync_xor_and_fetch:
172  case Builtin::BI__sync_nand_and_fetch:
173  case Builtin::BI__sync_val_compare_and_swap:
174  case Builtin::BI__sync_bool_compare_and_swap:
175  case Builtin::BI__sync_lock_test_and_set:
176  case Builtin::BI__sync_lock_release:
177    if (SemaBuiltinAtomicOverloaded(TheCall))
178      return ExprError();
179    break;
180  }
181
182  return move(TheCallResult);
183}
184
185/// CheckFunctionCall - Check a direct function call for various correctness
186/// and safety properties not strictly enforced by the C type system.
187bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) {
188  // Get the IdentifierInfo* for the called function.
189  IdentifierInfo *FnInfo = FDecl->getIdentifier();
190
191  // None of the checks below are needed for functions that don't have
192  // simple names (e.g., C++ conversion functions).
193  if (!FnInfo)
194    return false;
195
196  // FIXME: This mechanism should be abstracted to be less fragile and
197  // more efficient. For example, just map function ids to custom
198  // handlers.
199
200  // Printf checking.
201  if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) {
202    if (CheckablePrintfAttr(Format, TheCall)) {
203      bool HasVAListArg = Format->getFirstArg() == 0;
204      if (!HasVAListArg) {
205        if (const FunctionProtoType *Proto
206            = FDecl->getType()->getAs<FunctionProtoType>())
207        HasVAListArg = !Proto->isVariadic();
208      }
209      CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
210                           HasVAListArg ? 0 : Format->getFirstArg() - 1);
211    }
212  }
213
214  for (const NonNullAttr *NonNull = FDecl->getAttr<NonNullAttr>(); NonNull;
215       NonNull = NonNull->getNext<NonNullAttr>())
216    CheckNonNullArguments(NonNull, TheCall);
217
218  return false;
219}
220
221bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
222  // Printf checking.
223  const FormatAttr *Format = NDecl->getAttr<FormatAttr>();
224  if (!Format)
225    return false;
226
227  const VarDecl *V = dyn_cast<VarDecl>(NDecl);
228  if (!V)
229    return false;
230
231  QualType Ty = V->getType();
232  if (!Ty->isBlockPointerType())
233    return false;
234
235  if (!CheckablePrintfAttr(Format, TheCall))
236    return false;
237
238  bool HasVAListArg = Format->getFirstArg() == 0;
239  if (!HasVAListArg) {
240    const FunctionType *FT =
241      Ty->getAs<BlockPointerType>()->getPointeeType()->getAs<FunctionType>();
242    if (const FunctionProtoType *Proto = dyn_cast<FunctionProtoType>(FT))
243      HasVAListArg = !Proto->isVariadic();
244  }
245  CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
246                       HasVAListArg ? 0 : Format->getFirstArg() - 1);
247
248  return false;
249}
250
251/// SemaBuiltinAtomicOverloaded - We have a call to a function like
252/// __sync_fetch_and_add, which is an overloaded function based on the pointer
253/// type of its first argument.  The main ActOnCallExpr routines have already
254/// promoted the types of arguments because all of these calls are prototyped as
255/// void(...).
256///
257/// This function goes through and does final semantic checking for these
258/// builtins,
259bool Sema::SemaBuiltinAtomicOverloaded(CallExpr *TheCall) {
260  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
261  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
262
263  // Ensure that we have at least one argument to do type inference from.
264  if (TheCall->getNumArgs() < 1)
265    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
266              << 0 << TheCall->getCallee()->getSourceRange();
267
268  // Inspect the first argument of the atomic builtin.  This should always be
269  // a pointer type, whose element is an integral scalar or pointer type.
270  // Because it is a pointer type, we don't have to worry about any implicit
271  // casts here.
272  Expr *FirstArg = TheCall->getArg(0);
273  if (!FirstArg->getType()->isPointerType())
274    return Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
275             << FirstArg->getType() << FirstArg->getSourceRange();
276
277  QualType ValType = FirstArg->getType()->getAs<PointerType>()->getPointeeType();
278  if (!ValType->isIntegerType() && !ValType->isPointerType() &&
279      !ValType->isBlockPointerType())
280    return Diag(DRE->getLocStart(),
281                diag::err_atomic_builtin_must_be_pointer_intptr)
282             << FirstArg->getType() << FirstArg->getSourceRange();
283
284  // We need to figure out which concrete builtin this maps onto.  For example,
285  // __sync_fetch_and_add with a 2 byte object turns into
286  // __sync_fetch_and_add_2.
287#define BUILTIN_ROW(x) \
288  { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
289    Builtin::BI##x##_8, Builtin::BI##x##_16 }
290
291  static const unsigned BuiltinIndices[][5] = {
292    BUILTIN_ROW(__sync_fetch_and_add),
293    BUILTIN_ROW(__sync_fetch_and_sub),
294    BUILTIN_ROW(__sync_fetch_and_or),
295    BUILTIN_ROW(__sync_fetch_and_and),
296    BUILTIN_ROW(__sync_fetch_and_xor),
297    BUILTIN_ROW(__sync_fetch_and_nand),
298
299    BUILTIN_ROW(__sync_add_and_fetch),
300    BUILTIN_ROW(__sync_sub_and_fetch),
301    BUILTIN_ROW(__sync_and_and_fetch),
302    BUILTIN_ROW(__sync_or_and_fetch),
303    BUILTIN_ROW(__sync_xor_and_fetch),
304    BUILTIN_ROW(__sync_nand_and_fetch),
305
306    BUILTIN_ROW(__sync_val_compare_and_swap),
307    BUILTIN_ROW(__sync_bool_compare_and_swap),
308    BUILTIN_ROW(__sync_lock_test_and_set),
309    BUILTIN_ROW(__sync_lock_release)
310  };
311#undef BUILTIN_ROW
312
313  // Determine the index of the size.
314  unsigned SizeIndex;
315  switch (Context.getTypeSize(ValType)/8) {
316  case 1: SizeIndex = 0; break;
317  case 2: SizeIndex = 1; break;
318  case 4: SizeIndex = 2; break;
319  case 8: SizeIndex = 3; break;
320  case 16: SizeIndex = 4; break;
321  default:
322    return Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
323             << FirstArg->getType() << FirstArg->getSourceRange();
324  }
325
326  // Each of these builtins has one pointer argument, followed by some number of
327  // values (0, 1 or 2) followed by a potentially empty varags list of stuff
328  // that we ignore.  Find out which row of BuiltinIndices to read from as well
329  // as the number of fixed args.
330  unsigned BuiltinID = FDecl->getBuiltinID();
331  unsigned BuiltinIndex, NumFixed = 1;
332  switch (BuiltinID) {
333  default: assert(0 && "Unknown overloaded atomic builtin!");
334  case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break;
335  case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break;
336  case Builtin::BI__sync_fetch_and_or:  BuiltinIndex = 2; break;
337  case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break;
338  case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break;
339  case Builtin::BI__sync_fetch_and_nand:BuiltinIndex = 5; break;
340
341  case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 6; break;
342  case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 7; break;
343  case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 8; break;
344  case Builtin::BI__sync_or_and_fetch:  BuiltinIndex = 9; break;
345  case Builtin::BI__sync_xor_and_fetch: BuiltinIndex =10; break;
346  case Builtin::BI__sync_nand_and_fetch:BuiltinIndex =11; break;
347
348  case Builtin::BI__sync_val_compare_and_swap:
349    BuiltinIndex = 12;
350    NumFixed = 2;
351    break;
352  case Builtin::BI__sync_bool_compare_and_swap:
353    BuiltinIndex = 13;
354    NumFixed = 2;
355    break;
356  case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 14; break;
357  case Builtin::BI__sync_lock_release:
358    BuiltinIndex = 15;
359    NumFixed = 0;
360    break;
361  }
362
363  // Now that we know how many fixed arguments we expect, first check that we
364  // have at least that many.
365  if (TheCall->getNumArgs() < 1+NumFixed)
366    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
367            << 0 << TheCall->getCallee()->getSourceRange();
368
369
370  // Get the decl for the concrete builtin from this, we can tell what the
371  // concrete integer type we should convert to is.
372  unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
373  const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
374  IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
375  FunctionDecl *NewBuiltinDecl =
376    cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
377                                           TUScope, false, DRE->getLocStart()));
378  const FunctionProtoType *BuiltinFT =
379    NewBuiltinDecl->getType()->getAs<FunctionProtoType>();
380  ValType = BuiltinFT->getArgType(0)->getAs<PointerType>()->getPointeeType();
381
382  // If the first type needs to be converted (e.g. void** -> int*), do it now.
383  if (BuiltinFT->getArgType(0) != FirstArg->getType()) {
384    ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), CastExpr::CK_BitCast);
385    TheCall->setArg(0, FirstArg);
386  }
387
388  // Next, walk the valid ones promoting to the right type.
389  for (unsigned i = 0; i != NumFixed; ++i) {
390    Expr *Arg = TheCall->getArg(i+1);
391
392    // If the argument is an implicit cast, then there was a promotion due to
393    // "...", just remove it now.
394    if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) {
395      Arg = ICE->getSubExpr();
396      ICE->setSubExpr(0);
397      ICE->Destroy(Context);
398      TheCall->setArg(i+1, Arg);
399    }
400
401    // GCC does an implicit conversion to the pointer or integer ValType.  This
402    // can fail in some cases (1i -> int**), check for this error case now.
403    CastExpr::CastKind Kind = CastExpr::CK_Unknown;
404    CXXMethodDecl *ConversionDecl = 0;
405    if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind,
406                       ConversionDecl))
407      return true;
408
409    // Okay, we have something that *can* be converted to the right type.  Check
410    // to see if there is a potentially weird extension going on here.  This can
411    // happen when you do an atomic operation on something like an char* and
412    // pass in 42.  The 42 gets converted to char.  This is even more strange
413    // for things like 45.123 -> char, etc.
414    // FIXME: Do this check.
415    ImpCastExprToType(Arg, ValType, Kind, /*isLvalue=*/false);
416    TheCall->setArg(i+1, Arg);
417  }
418
419  // Switch the DeclRefExpr to refer to the new decl.
420  DRE->setDecl(NewBuiltinDecl);
421  DRE->setType(NewBuiltinDecl->getType());
422
423  // Set the callee in the CallExpr.
424  // FIXME: This leaks the original parens and implicit casts.
425  Expr *PromotedCall = DRE;
426  UsualUnaryConversions(PromotedCall);
427  TheCall->setCallee(PromotedCall);
428
429
430  // Change the result type of the call to match the result type of the decl.
431  TheCall->setType(NewBuiltinDecl->getResultType());
432  return false;
433}
434
435
436/// CheckObjCString - Checks that the argument to the builtin
437/// CFString constructor is correct
438/// FIXME: GCC currently emits the following warning:
439/// "warning: input conversion stopped due to an input byte that does not
440///           belong to the input codeset UTF-8"
441/// Note: It might also make sense to do the UTF-16 conversion here (would
442/// simplify the backend).
443bool Sema::CheckObjCString(Expr *Arg) {
444  Arg = Arg->IgnoreParenCasts();
445  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
446
447  if (!Literal || Literal->isWide()) {
448    Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
449      << Arg->getSourceRange();
450    return true;
451  }
452
453  const char *Data = Literal->getStrData();
454  unsigned Length = Literal->getByteLength();
455
456  for (unsigned i = 0; i < Length; ++i) {
457    if (!Data[i]) {
458      Diag(getLocationOfStringLiteralByte(Literal, i),
459           diag::warn_cfstring_literal_contains_nul_character)
460        << Arg->getSourceRange();
461      break;
462    }
463  }
464
465  return false;
466}
467
468/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
469/// Emit an error and return true on failure, return false on success.
470bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
471  Expr *Fn = TheCall->getCallee();
472  if (TheCall->getNumArgs() > 2) {
473    Diag(TheCall->getArg(2)->getLocStart(),
474         diag::err_typecheck_call_too_many_args)
475      << 0 /*function call*/ << Fn->getSourceRange()
476      << SourceRange(TheCall->getArg(2)->getLocStart(),
477                     (*(TheCall->arg_end()-1))->getLocEnd());
478    return true;
479  }
480
481  if (TheCall->getNumArgs() < 2) {
482    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
483      << 0 /*function call*/;
484  }
485
486  // Determine whether the current function is variadic or not.
487  bool isVariadic;
488  if (CurBlock)
489    isVariadic = CurBlock->isVariadic;
490  else if (getCurFunctionDecl()) {
491    if (FunctionProtoType* FTP =
492            dyn_cast<FunctionProtoType>(getCurFunctionDecl()->getType()))
493      isVariadic = FTP->isVariadic();
494    else
495      isVariadic = false;
496  } else {
497    isVariadic = getCurMethodDecl()->isVariadic();
498  }
499
500  if (!isVariadic) {
501    Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
502    return true;
503  }
504
505  // Verify that the second argument to the builtin is the last argument of the
506  // current function or method.
507  bool SecondArgIsLastNamedArgument = false;
508  const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
509
510  if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
511    if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
512      // FIXME: This isn't correct for methods (results in bogus warning).
513      // Get the last formal in the current function.
514      const ParmVarDecl *LastArg;
515      if (CurBlock)
516        LastArg = *(CurBlock->TheDecl->param_end()-1);
517      else if (FunctionDecl *FD = getCurFunctionDecl())
518        LastArg = *(FD->param_end()-1);
519      else
520        LastArg = *(getCurMethodDecl()->param_end()-1);
521      SecondArgIsLastNamedArgument = PV == LastArg;
522    }
523  }
524
525  if (!SecondArgIsLastNamedArgument)
526    Diag(TheCall->getArg(1)->getLocStart(),
527         diag::warn_second_parameter_of_va_start_not_last_named_argument);
528  return false;
529}
530
531/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
532/// friends.  This is declared to take (...), so we have to check everything.
533bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
534  if (TheCall->getNumArgs() < 2)
535    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
536      << 0 /*function call*/;
537  if (TheCall->getNumArgs() > 2)
538    return Diag(TheCall->getArg(2)->getLocStart(),
539                diag::err_typecheck_call_too_many_args)
540      << 0 /*function call*/
541      << SourceRange(TheCall->getArg(2)->getLocStart(),
542                     (*(TheCall->arg_end()-1))->getLocEnd());
543
544  Expr *OrigArg0 = TheCall->getArg(0);
545  Expr *OrigArg1 = TheCall->getArg(1);
546
547  // Do standard promotions between the two arguments, returning their common
548  // type.
549  QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
550
551  // Make sure any conversions are pushed back into the call; this is
552  // type safe since unordered compare builtins are declared as "_Bool
553  // foo(...)".
554  TheCall->setArg(0, OrigArg0);
555  TheCall->setArg(1, OrigArg1);
556
557  if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent())
558    return false;
559
560  // If the common type isn't a real floating type, then the arguments were
561  // invalid for this operation.
562  if (!Res->isRealFloatingType())
563    return Diag(OrigArg0->getLocStart(),
564                diag::err_typecheck_call_invalid_ordered_compare)
565      << OrigArg0->getType() << OrigArg1->getType()
566      << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd());
567
568  return false;
569}
570
571/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isnan and
572/// friends.  This is declared to take (...), so we have to check everything.
573bool Sema::SemaBuiltinUnaryFP(CallExpr *TheCall) {
574  if (TheCall->getNumArgs() < 1)
575    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
576      << 0 /*function call*/;
577  if (TheCall->getNumArgs() > 1)
578    return Diag(TheCall->getArg(1)->getLocStart(),
579                diag::err_typecheck_call_too_many_args)
580      << 0 /*function call*/
581      << SourceRange(TheCall->getArg(1)->getLocStart(),
582                     (*(TheCall->arg_end()-1))->getLocEnd());
583
584  Expr *OrigArg = TheCall->getArg(0);
585
586  if (OrigArg->isTypeDependent())
587    return false;
588
589  // This operation requires a floating-point number
590  if (!OrigArg->getType()->isRealFloatingType())
591    return Diag(OrigArg->getLocStart(),
592                diag::err_typecheck_call_invalid_unary_fp)
593      << OrigArg->getType() << OrigArg->getSourceRange();
594
595  return false;
596}
597
598bool Sema::SemaBuiltinStackAddress(CallExpr *TheCall) {
599  // The signature for these builtins is exact; the only thing we need
600  // to check is that the argument is a constant.
601  SourceLocation Loc;
602  if (!TheCall->getArg(0)->isTypeDependent() &&
603      !TheCall->getArg(0)->isValueDependent() &&
604      !TheCall->getArg(0)->isIntegerConstantExpr(Context, &Loc))
605    return Diag(Loc, diag::err_stack_const_level) << TheCall->getSourceRange();
606
607  return false;
608}
609
610/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
611// This is declared to take (...), so we have to check everything.
612Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
613  if (TheCall->getNumArgs() < 3)
614    return ExprError(Diag(TheCall->getLocEnd(),
615                          diag::err_typecheck_call_too_few_args)
616      << 0 /*function call*/ << TheCall->getSourceRange());
617
618  unsigned numElements = std::numeric_limits<unsigned>::max();
619  if (!TheCall->getArg(0)->isTypeDependent() &&
620      !TheCall->getArg(1)->isTypeDependent()) {
621    QualType FAType = TheCall->getArg(0)->getType();
622    QualType SAType = TheCall->getArg(1)->getType();
623
624    if (!FAType->isVectorType() || !SAType->isVectorType()) {
625      Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
626        << SourceRange(TheCall->getArg(0)->getLocStart(),
627                       TheCall->getArg(1)->getLocEnd());
628      return ExprError();
629    }
630
631    if (!Context.hasSameUnqualifiedType(FAType, SAType)) {
632      Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
633        << SourceRange(TheCall->getArg(0)->getLocStart(),
634                       TheCall->getArg(1)->getLocEnd());
635      return ExprError();
636    }
637
638    numElements = FAType->getAs<VectorType>()->getNumElements();
639    if (TheCall->getNumArgs() != numElements+2) {
640      if (TheCall->getNumArgs() < numElements+2)
641        return ExprError(Diag(TheCall->getLocEnd(),
642                              diag::err_typecheck_call_too_few_args)
643                 << 0 /*function call*/ << TheCall->getSourceRange());
644      return ExprError(Diag(TheCall->getLocEnd(),
645                            diag::err_typecheck_call_too_many_args)
646                 << 0 /*function call*/ << TheCall->getSourceRange());
647    }
648  }
649
650  for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
651    if (TheCall->getArg(i)->isTypeDependent() ||
652        TheCall->getArg(i)->isValueDependent())
653      continue;
654
655    llvm::APSInt Result(32);
656    if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
657      return ExprError(Diag(TheCall->getLocStart(),
658                  diag::err_shufflevector_nonconstant_argument)
659                << TheCall->getArg(i)->getSourceRange());
660
661    if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
662      return ExprError(Diag(TheCall->getLocStart(),
663                  diag::err_shufflevector_argument_too_large)
664               << TheCall->getArg(i)->getSourceRange());
665  }
666
667  llvm::SmallVector<Expr*, 32> exprs;
668
669  for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
670    exprs.push_back(TheCall->getArg(i));
671    TheCall->setArg(i, 0);
672  }
673
674  return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
675                                            exprs.size(), exprs[0]->getType(),
676                                            TheCall->getCallee()->getLocStart(),
677                                            TheCall->getRParenLoc()));
678}
679
680/// SemaBuiltinPrefetch - Handle __builtin_prefetch.
681// This is declared to take (const void*, ...) and can take two
682// optional constant int args.
683bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
684  unsigned NumArgs = TheCall->getNumArgs();
685
686  if (NumArgs > 3)
687    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_many_args)
688             << 0 /*function call*/ << TheCall->getSourceRange();
689
690  // Argument 0 is checked for us and the remaining arguments must be
691  // constant integers.
692  for (unsigned i = 1; i != NumArgs; ++i) {
693    Expr *Arg = TheCall->getArg(i);
694    if (Arg->isTypeDependent())
695      continue;
696
697    QualType RWType = Arg->getType();
698
699    const BuiltinType *BT = RWType->getAs<BuiltinType>();
700    llvm::APSInt Result;
701    if (!BT || BT->getKind() != BuiltinType::Int)
702      return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_argument)
703              << Arg->getSourceRange();
704
705    if (Arg->isValueDependent())
706      continue;
707
708    if (!Arg->isIntegerConstantExpr(Result, Context))
709      return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_argument)
710        << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
711
712    // FIXME: gcc issues a warning and rewrites these to 0. These
713    // seems especially odd for the third argument since the default
714    // is 3.
715    if (i == 1) {
716      if (Result.getSExtValue() < 0 || Result.getSExtValue() > 1)
717        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
718             << "0" << "1" << Arg->getSourceRange();
719    } else {
720      if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3)
721        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
722            << "0" << "3" << Arg->getSourceRange();
723    }
724  }
725
726  return false;
727}
728
729/// SemaBuiltinEHReturnDataRegNo - Handle __builtin_eh_return_data_regno, the
730/// operand must be an integer constant.
731bool Sema::SemaBuiltinEHReturnDataRegNo(CallExpr *TheCall) {
732  llvm::APSInt Result;
733  if (!TheCall->getArg(0)->isIntegerConstantExpr(Result, Context))
734    return Diag(TheCall->getLocStart(), diag::err_expr_not_ice)
735      << TheCall->getArg(0)->getSourceRange();
736
737  return false;
738}
739
740
741/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
742/// int type). This simply type checks that type is one of the defined
743/// constants (0-3).
744bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
745  Expr *Arg = TheCall->getArg(1);
746  if (Arg->isTypeDependent())
747    return false;
748
749  QualType ArgType = Arg->getType();
750  const BuiltinType *BT = ArgType->getAs<BuiltinType>();
751  llvm::APSInt Result(32);
752  if (!BT || BT->getKind() != BuiltinType::Int)
753    return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument)
754             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
755
756  if (Arg->isValueDependent())
757    return false;
758
759  if (!Arg->isIntegerConstantExpr(Result, Context)) {
760    return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument)
761             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
762  }
763
764  if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
765    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
766             << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
767  }
768
769  return false;
770}
771
772/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
773/// This checks that val is a constant 1.
774bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
775  Expr *Arg = TheCall->getArg(1);
776  if (Arg->isTypeDependent() || Arg->isValueDependent())
777    return false;
778
779  llvm::APSInt Result(32);
780  if (!Arg->isIntegerConstantExpr(Result, Context) || Result != 1)
781    return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
782             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
783
784  return false;
785}
786
787// Handle i > 1 ? "x" : "y", recursivelly
788bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall,
789                                  bool HasVAListArg,
790                                  unsigned format_idx, unsigned firstDataArg) {
791  if (E->isTypeDependent() || E->isValueDependent())
792    return false;
793
794  switch (E->getStmtClass()) {
795  case Stmt::ConditionalOperatorClass: {
796    const ConditionalOperator *C = cast<ConditionalOperator>(E);
797    return SemaCheckStringLiteral(C->getLHS(), TheCall,
798                                  HasVAListArg, format_idx, firstDataArg)
799        && SemaCheckStringLiteral(C->getRHS(), TheCall,
800                                  HasVAListArg, format_idx, firstDataArg);
801  }
802
803  case Stmt::ImplicitCastExprClass: {
804    const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E);
805    return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg,
806                                  format_idx, firstDataArg);
807  }
808
809  case Stmt::ParenExprClass: {
810    const ParenExpr *Expr = cast<ParenExpr>(E);
811    return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg,
812                                  format_idx, firstDataArg);
813  }
814
815  case Stmt::DeclRefExprClass: {
816    const DeclRefExpr *DR = cast<DeclRefExpr>(E);
817
818    // As an exception, do not flag errors for variables binding to
819    // const string literals.
820    if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
821      bool isConstant = false;
822      QualType T = DR->getType();
823
824      if (const ArrayType *AT = Context.getAsArrayType(T)) {
825        isConstant = AT->getElementType().isConstant(Context);
826      } else if (const PointerType *PT = T->getAs<PointerType>()) {
827        isConstant = T.isConstant(Context) &&
828                     PT->getPointeeType().isConstant(Context);
829      }
830
831      if (isConstant) {
832        const VarDecl *Def = 0;
833        if (const Expr *Init = VD->getDefinition(Def))
834          return SemaCheckStringLiteral(Init, TheCall,
835                                        HasVAListArg, format_idx, firstDataArg);
836      }
837
838      // For vprintf* functions (i.e., HasVAListArg==true), we add a
839      // special check to see if the format string is a function parameter
840      // of the function calling the printf function.  If the function
841      // has an attribute indicating it is a printf-like function, then we
842      // should suppress warnings concerning non-literals being used in a call
843      // to a vprintf function.  For example:
844      //
845      // void
846      // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
847      //      va_list ap;
848      //      va_start(ap, fmt);
849      //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
850      //      ...
851      //
852      //
853      //  FIXME: We don't have full attribute support yet, so just check to see
854      //    if the argument is a DeclRefExpr that references a parameter.  We'll
855      //    add proper support for checking the attribute later.
856      if (HasVAListArg)
857        if (isa<ParmVarDecl>(VD))
858          return true;
859    }
860
861    return false;
862  }
863
864  case Stmt::CallExprClass: {
865    const CallExpr *CE = cast<CallExpr>(E);
866    if (const ImplicitCastExpr *ICE
867          = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
868      if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
869        if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
870          if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
871            unsigned ArgIndex = FA->getFormatIdx();
872            const Expr *Arg = CE->getArg(ArgIndex - 1);
873
874            return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg,
875                                          format_idx, firstDataArg);
876          }
877        }
878      }
879    }
880
881    return false;
882  }
883  case Stmt::ObjCStringLiteralClass:
884  case Stmt::StringLiteralClass: {
885    const StringLiteral *StrE = NULL;
886
887    if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
888      StrE = ObjCFExpr->getString();
889    else
890      StrE = cast<StringLiteral>(E);
891
892    if (StrE) {
893      CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx,
894                        firstDataArg);
895      return true;
896    }
897
898    return false;
899  }
900
901  default:
902    return false;
903  }
904}
905
906void
907Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
908                            const CallExpr *TheCall) {
909  for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end();
910       i != e; ++i) {
911    const Expr *ArgExpr = TheCall->getArg(*i);
912    if (ArgExpr->isNullPointerConstant(Context,
913                                       Expr::NPC_ValueDependentIsNotNull))
914      Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg)
915        << ArgExpr->getSourceRange();
916  }
917}
918
919/// CheckPrintfArguments - Check calls to printf (and similar functions) for
920/// correct use of format strings.
921///
922///  HasVAListArg - A predicate indicating whether the printf-like
923///    function is passed an explicit va_arg argument (e.g., vprintf)
924///
925///  format_idx - The index into Args for the format string.
926///
927/// Improper format strings to functions in the printf family can be
928/// the source of bizarre bugs and very serious security holes.  A
929/// good source of information is available in the following paper
930/// (which includes additional references):
931///
932///  FormatGuard: Automatic Protection From printf Format String
933///  Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001.
934///
935/// Functionality implemented:
936///
937///  We can statically check the following properties for string
938///  literal format strings for non v.*printf functions (where the
939///  arguments are passed directly):
940//
941///  (1) Are the number of format conversions equal to the number of
942///      data arguments?
943///
944///  (2) Does each format conversion correctly match the type of the
945///      corresponding data argument?  (TODO)
946///
947/// Moreover, for all printf functions we can:
948///
949///  (3) Check for a missing format string (when not caught by type checking).
950///
951///  (4) Check for no-operation flags; e.g. using "#" with format
952///      conversion 'c'  (TODO)
953///
954///  (5) Check the use of '%n', a major source of security holes.
955///
956///  (6) Check for malformed format conversions that don't specify anything.
957///
958///  (7) Check for empty format strings.  e.g: printf("");
959///
960///  (8) Check that the format string is a wide literal.
961///
962///  (9) Also check the arguments of functions with the __format__ attribute.
963///      (TODO).
964///
965/// All of these checks can be done by parsing the format string.
966///
967/// For now, we ONLY do (1), (3), (5), (6), (7), and (8).
968void
969Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg,
970                           unsigned format_idx, unsigned firstDataArg) {
971  const Expr *Fn = TheCall->getCallee();
972
973  // CHECK: printf-like function is called with no format string.
974  if (format_idx >= TheCall->getNumArgs()) {
975    Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string)
976      << Fn->getSourceRange();
977    return;
978  }
979
980  const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts();
981
982  // CHECK: format string is not a string literal.
983  //
984  // Dynamically generated format strings are difficult to
985  // automatically vet at compile time.  Requiring that format strings
986  // are string literals: (1) permits the checking of format strings by
987  // the compiler and thereby (2) can practically remove the source of
988  // many format string exploits.
989
990  // Format string can be either ObjC string (e.g. @"%d") or
991  // C string (e.g. "%d")
992  // ObjC string uses the same format specifiers as C string, so we can use
993  // the same format string checking logic for both ObjC and C strings.
994  if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx,
995                             firstDataArg))
996    return;  // Literal format string found, check done!
997
998  // If there are no arguments specified, warn with -Wformat-security, otherwise
999  // warn only with -Wformat-nonliteral.
1000  if (TheCall->getNumArgs() == format_idx+1)
1001    Diag(TheCall->getArg(format_idx)->getLocStart(),
1002         diag::warn_printf_nonliteral_noargs)
1003      << OrigFormatExpr->getSourceRange();
1004  else
1005    Diag(TheCall->getArg(format_idx)->getLocStart(),
1006         diag::warn_printf_nonliteral)
1007           << OrigFormatExpr->getSourceRange();
1008}
1009
1010void Sema::CheckPrintfString(const StringLiteral *FExpr,
1011                             const Expr *OrigFormatExpr,
1012                             const CallExpr *TheCall, bool HasVAListArg,
1013                             unsigned format_idx, unsigned firstDataArg) {
1014
1015  const ObjCStringLiteral *ObjCFExpr =
1016    dyn_cast<ObjCStringLiteral>(OrigFormatExpr);
1017
1018  // CHECK: is the format string a wide literal?
1019  if (FExpr->isWide()) {
1020    Diag(FExpr->getLocStart(),
1021         diag::warn_printf_format_string_is_wide_literal)
1022      << OrigFormatExpr->getSourceRange();
1023    return;
1024  }
1025
1026  // Str - The format string.  NOTE: this is NOT null-terminated!
1027  const char *Str = FExpr->getStrData();
1028
1029  // CHECK: empty format string?
1030  unsigned StrLen = FExpr->getByteLength();
1031
1032  if (StrLen == 0) {
1033    Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string)
1034      << OrigFormatExpr->getSourceRange();
1035    return;
1036  }
1037
1038  // We process the format string using a binary state machine.  The
1039  // current state is stored in CurrentState.
1040  enum {
1041    state_OrdChr,
1042    state_Conversion
1043  } CurrentState = state_OrdChr;
1044
1045  // numConversions - The number of conversions seen so far.  This is
1046  //  incremented as we traverse the format string.
1047  unsigned numConversions = 0;
1048
1049  // numDataArgs - The number of data arguments after the format
1050  //  string.  This can only be determined for non vprintf-like
1051  //  functions.  For those functions, this value is 1 (the sole
1052  //  va_arg argument).
1053  unsigned numDataArgs = TheCall->getNumArgs()-firstDataArg;
1054
1055  // Inspect the format string.
1056  unsigned StrIdx = 0;
1057
1058  // LastConversionIdx - Index within the format string where we last saw
1059  //  a '%' character that starts a new format conversion.
1060  unsigned LastConversionIdx = 0;
1061
1062  for (; StrIdx < StrLen; ++StrIdx) {
1063
1064    // Is the number of detected conversion conversions greater than
1065    // the number of matching data arguments?  If so, stop.
1066    if (!HasVAListArg && numConversions > numDataArgs) break;
1067
1068    // Handle "\0"
1069    if (Str[StrIdx] == '\0') {
1070      // The string returned by getStrData() is not null-terminated,
1071      // so the presence of a null character is likely an error.
1072      Diag(getLocationOfStringLiteralByte(FExpr, StrIdx),
1073           diag::warn_printf_format_string_contains_null_char)
1074        <<  OrigFormatExpr->getSourceRange();
1075      return;
1076    }
1077
1078    // Ordinary characters (not processing a format conversion).
1079    if (CurrentState == state_OrdChr) {
1080      if (Str[StrIdx] == '%') {
1081        CurrentState = state_Conversion;
1082        LastConversionIdx = StrIdx;
1083      }
1084      continue;
1085    }
1086
1087    // Seen '%'.  Now processing a format conversion.
1088    switch (Str[StrIdx]) {
1089    // Handle dynamic precision or width specifier.
1090    case '*': {
1091      ++numConversions;
1092
1093      if (!HasVAListArg) {
1094        if (numConversions > numDataArgs) {
1095          SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx);
1096
1097          if (Str[StrIdx-1] == '.')
1098            Diag(Loc, diag::warn_printf_asterisk_precision_missing_arg)
1099              << OrigFormatExpr->getSourceRange();
1100          else
1101            Diag(Loc, diag::warn_printf_asterisk_width_missing_arg)
1102              << OrigFormatExpr->getSourceRange();
1103
1104          // Don't do any more checking.  We'll just emit spurious errors.
1105          return;
1106        }
1107
1108        // Perform type checking on width/precision specifier.
1109        const Expr *E = TheCall->getArg(format_idx+numConversions);
1110        if (const BuiltinType *BT = E->getType()->getAs<BuiltinType>())
1111          if (BT->getKind() == BuiltinType::Int)
1112            break;
1113
1114        SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx);
1115
1116        if (Str[StrIdx-1] == '.')
1117          Diag(Loc, diag::warn_printf_asterisk_precision_wrong_type)
1118          << E->getType() << E->getSourceRange();
1119        else
1120          Diag(Loc, diag::warn_printf_asterisk_width_wrong_type)
1121          << E->getType() << E->getSourceRange();
1122
1123        break;
1124      }
1125    }
1126
1127    // Characters which can terminate a format conversion
1128    // (e.g. "%d").  Characters that specify length modifiers or
1129    // other flags are handled by the default case below.
1130    //
1131    // FIXME: additional checks will go into the following cases.
1132    case 'i':
1133    case 'd':
1134    case 'o':
1135    case 'u':
1136    case 'x':
1137    case 'X':
1138    case 'D':
1139    case 'O':
1140    case 'U':
1141    case 'e':
1142    case 'E':
1143    case 'f':
1144    case 'F':
1145    case 'g':
1146    case 'G':
1147    case 'a':
1148    case 'A':
1149    case 'c':
1150    case 'C':
1151    case 'S':
1152    case 's':
1153    case 'p':
1154      ++numConversions;
1155      CurrentState = state_OrdChr;
1156      break;
1157
1158    case 'm':
1159      // FIXME: Warn in situations where this isn't supported!
1160      CurrentState = state_OrdChr;
1161      break;
1162
1163    // CHECK: Are we using "%n"?  Issue a warning.
1164    case 'n': {
1165      ++numConversions;
1166      CurrentState = state_OrdChr;
1167      SourceLocation Loc = getLocationOfStringLiteralByte(FExpr,
1168                                                          LastConversionIdx);
1169
1170      Diag(Loc, diag::warn_printf_write_back)<<OrigFormatExpr->getSourceRange();
1171      break;
1172    }
1173
1174    // Handle "%@"
1175    case '@':
1176      // %@ is allowed in ObjC format strings only.
1177      if (ObjCFExpr != NULL)
1178        CurrentState = state_OrdChr;
1179      else {
1180        // Issue a warning: invalid format conversion.
1181        SourceLocation Loc =
1182          getLocationOfStringLiteralByte(FExpr, LastConversionIdx);
1183
1184        Diag(Loc, diag::warn_printf_invalid_conversion)
1185          <<  std::string(Str+LastConversionIdx,
1186                          Str+std::min(LastConversionIdx+2, StrLen))
1187          << OrigFormatExpr->getSourceRange();
1188      }
1189      ++numConversions;
1190      break;
1191
1192    // Handle "%%"
1193    case '%':
1194      // Sanity check: Was the first "%" character the previous one?
1195      // If not, we will assume that we have a malformed format
1196      // conversion, and that the current "%" character is the start
1197      // of a new conversion.
1198      if (StrIdx - LastConversionIdx == 1)
1199        CurrentState = state_OrdChr;
1200      else {
1201        // Issue a warning: invalid format conversion.
1202        SourceLocation Loc =
1203          getLocationOfStringLiteralByte(FExpr, LastConversionIdx);
1204
1205        Diag(Loc, diag::warn_printf_invalid_conversion)
1206          << std::string(Str+LastConversionIdx, Str+StrIdx)
1207          << OrigFormatExpr->getSourceRange();
1208
1209        // This conversion is broken.  Advance to the next format
1210        // conversion.
1211        LastConversionIdx = StrIdx;
1212        ++numConversions;
1213      }
1214      break;
1215
1216    default:
1217      // This case catches all other characters: flags, widths, etc.
1218      // We should eventually process those as well.
1219      break;
1220    }
1221  }
1222
1223  if (CurrentState == state_Conversion) {
1224    // Issue a warning: invalid format conversion.
1225    SourceLocation Loc =
1226      getLocationOfStringLiteralByte(FExpr, LastConversionIdx);
1227
1228    Diag(Loc, diag::warn_printf_invalid_conversion)
1229      << std::string(Str+LastConversionIdx,
1230                     Str+std::min(LastConversionIdx+2, StrLen))
1231      << OrigFormatExpr->getSourceRange();
1232    return;
1233  }
1234
1235  if (!HasVAListArg) {
1236    // CHECK: Does the number of format conversions exceed the number
1237    //        of data arguments?
1238    if (numConversions > numDataArgs) {
1239      SourceLocation Loc =
1240        getLocationOfStringLiteralByte(FExpr, LastConversionIdx);
1241
1242      Diag(Loc, diag::warn_printf_insufficient_data_args)
1243        << OrigFormatExpr->getSourceRange();
1244    }
1245    // CHECK: Does the number of data arguments exceed the number of
1246    //        format conversions in the format string?
1247    else if (numConversions < numDataArgs)
1248      Diag(TheCall->getArg(format_idx+numConversions+1)->getLocStart(),
1249           diag::warn_printf_too_many_data_args)
1250        << OrigFormatExpr->getSourceRange();
1251  }
1252}
1253
1254//===--- CHECK: Return Address of Stack Variable --------------------------===//
1255
1256static DeclRefExpr* EvalVal(Expr *E);
1257static DeclRefExpr* EvalAddr(Expr* E);
1258
1259/// CheckReturnStackAddr - Check if a return statement returns the address
1260///   of a stack variable.
1261void
1262Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
1263                           SourceLocation ReturnLoc) {
1264
1265  // Perform checking for returned stack addresses.
1266  if (lhsType->isPointerType() || lhsType->isBlockPointerType()) {
1267    if (DeclRefExpr *DR = EvalAddr(RetValExp))
1268      Diag(DR->getLocStart(), diag::warn_ret_stack_addr)
1269       << DR->getDecl()->getDeclName() << RetValExp->getSourceRange();
1270
1271    // Skip over implicit cast expressions when checking for block expressions.
1272    RetValExp = RetValExp->IgnoreParenCasts();
1273
1274    if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp))
1275      if (C->hasBlockDeclRefExprs())
1276        Diag(C->getLocStart(), diag::err_ret_local_block)
1277          << C->getSourceRange();
1278
1279    if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp))
1280      Diag(ALE->getLocStart(), diag::warn_ret_addr_label)
1281        << ALE->getSourceRange();
1282
1283  } else if (lhsType->isReferenceType()) {
1284    // Perform checking for stack values returned by reference.
1285    // Check for a reference to the stack
1286    if (DeclRefExpr *DR = EvalVal(RetValExp))
1287      Diag(DR->getLocStart(), diag::warn_ret_stack_ref)
1288        << DR->getDecl()->getDeclName() << RetValExp->getSourceRange();
1289  }
1290}
1291
1292/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
1293///  check if the expression in a return statement evaluates to an address
1294///  to a location on the stack.  The recursion is used to traverse the
1295///  AST of the return expression, with recursion backtracking when we
1296///  encounter a subexpression that (1) clearly does not lead to the address
1297///  of a stack variable or (2) is something we cannot determine leads to
1298///  the address of a stack variable based on such local checking.
1299///
1300///  EvalAddr processes expressions that are pointers that are used as
1301///  references (and not L-values).  EvalVal handles all other values.
1302///  At the base case of the recursion is a check for a DeclRefExpr* in
1303///  the refers to a stack variable.
1304///
1305///  This implementation handles:
1306///
1307///   * pointer-to-pointer casts
1308///   * implicit conversions from array references to pointers
1309///   * taking the address of fields
1310///   * arbitrary interplay between "&" and "*" operators
1311///   * pointer arithmetic from an address of a stack variable
1312///   * taking the address of an array element where the array is on the stack
1313static DeclRefExpr* EvalAddr(Expr *E) {
1314  // We should only be called for evaluating pointer expressions.
1315  assert((E->getType()->isAnyPointerType() ||
1316          E->getType()->isBlockPointerType() ||
1317          E->getType()->isObjCQualifiedIdType()) &&
1318         "EvalAddr only works on pointers");
1319
1320  // Our "symbolic interpreter" is just a dispatch off the currently
1321  // viewed AST node.  We then recursively traverse the AST by calling
1322  // EvalAddr and EvalVal appropriately.
1323  switch (E->getStmtClass()) {
1324  case Stmt::ParenExprClass:
1325    // Ignore parentheses.
1326    return EvalAddr(cast<ParenExpr>(E)->getSubExpr());
1327
1328  case Stmt::UnaryOperatorClass: {
1329    // The only unary operator that make sense to handle here
1330    // is AddrOf.  All others don't make sense as pointers.
1331    UnaryOperator *U = cast<UnaryOperator>(E);
1332
1333    if (U->getOpcode() == UnaryOperator::AddrOf)
1334      return EvalVal(U->getSubExpr());
1335    else
1336      return NULL;
1337  }
1338
1339  case Stmt::BinaryOperatorClass: {
1340    // Handle pointer arithmetic.  All other binary operators are not valid
1341    // in this context.
1342    BinaryOperator *B = cast<BinaryOperator>(E);
1343    BinaryOperator::Opcode op = B->getOpcode();
1344
1345    if (op != BinaryOperator::Add && op != BinaryOperator::Sub)
1346      return NULL;
1347
1348    Expr *Base = B->getLHS();
1349
1350    // Determine which argument is the real pointer base.  It could be
1351    // the RHS argument instead of the LHS.
1352    if (!Base->getType()->isPointerType()) Base = B->getRHS();
1353
1354    assert (Base->getType()->isPointerType());
1355    return EvalAddr(Base);
1356  }
1357
1358  // For conditional operators we need to see if either the LHS or RHS are
1359  // valid DeclRefExpr*s.  If one of them is valid, we return it.
1360  case Stmt::ConditionalOperatorClass: {
1361    ConditionalOperator *C = cast<ConditionalOperator>(E);
1362
1363    // Handle the GNU extension for missing LHS.
1364    if (Expr *lhsExpr = C->getLHS())
1365      if (DeclRefExpr* LHS = EvalAddr(lhsExpr))
1366        return LHS;
1367
1368     return EvalAddr(C->getRHS());
1369  }
1370
1371  // For casts, we need to handle conversions from arrays to
1372  // pointer values, and pointer-to-pointer conversions.
1373  case Stmt::ImplicitCastExprClass:
1374  case Stmt::CStyleCastExprClass:
1375  case Stmt::CXXFunctionalCastExprClass: {
1376    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
1377    QualType T = SubExpr->getType();
1378
1379    if (SubExpr->getType()->isPointerType() ||
1380        SubExpr->getType()->isBlockPointerType() ||
1381        SubExpr->getType()->isObjCQualifiedIdType())
1382      return EvalAddr(SubExpr);
1383    else if (T->isArrayType())
1384      return EvalVal(SubExpr);
1385    else
1386      return 0;
1387  }
1388
1389  // C++ casts.  For dynamic casts, static casts, and const casts, we
1390  // are always converting from a pointer-to-pointer, so we just blow
1391  // through the cast.  In the case the dynamic cast doesn't fail (and
1392  // return NULL), we take the conservative route and report cases
1393  // where we return the address of a stack variable.  For Reinterpre
1394  // FIXME: The comment about is wrong; we're not always converting
1395  // from pointer to pointer. I'm guessing that this code should also
1396  // handle references to objects.
1397  case Stmt::CXXStaticCastExprClass:
1398  case Stmt::CXXDynamicCastExprClass:
1399  case Stmt::CXXConstCastExprClass:
1400  case Stmt::CXXReinterpretCastExprClass: {
1401      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
1402      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
1403        return EvalAddr(S);
1404      else
1405        return NULL;
1406  }
1407
1408  // Everything else: we simply don't reason about them.
1409  default:
1410    return NULL;
1411  }
1412}
1413
1414
1415///  EvalVal - This function is complements EvalAddr in the mutual recursion.
1416///   See the comments for EvalAddr for more details.
1417static DeclRefExpr* EvalVal(Expr *E) {
1418
1419  // We should only be called for evaluating non-pointer expressions, or
1420  // expressions with a pointer type that are not used as references but instead
1421  // are l-values (e.g., DeclRefExpr with a pointer type).
1422
1423  // Our "symbolic interpreter" is just a dispatch off the currently
1424  // viewed AST node.  We then recursively traverse the AST by calling
1425  // EvalAddr and EvalVal appropriately.
1426  switch (E->getStmtClass()) {
1427  case Stmt::DeclRefExprClass: {
1428    // DeclRefExpr: the base case.  When we hit a DeclRefExpr we are looking
1429    //  at code that refers to a variable's name.  We check if it has local
1430    //  storage within the function, and if so, return the expression.
1431    DeclRefExpr *DR = cast<DeclRefExpr>(E);
1432
1433    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
1434      if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR;
1435
1436    return NULL;
1437  }
1438
1439  case Stmt::ParenExprClass:
1440    // Ignore parentheses.
1441    return EvalVal(cast<ParenExpr>(E)->getSubExpr());
1442
1443  case Stmt::UnaryOperatorClass: {
1444    // The only unary operator that make sense to handle here
1445    // is Deref.  All others don't resolve to a "name."  This includes
1446    // handling all sorts of rvalues passed to a unary operator.
1447    UnaryOperator *U = cast<UnaryOperator>(E);
1448
1449    if (U->getOpcode() == UnaryOperator::Deref)
1450      return EvalAddr(U->getSubExpr());
1451
1452    return NULL;
1453  }
1454
1455  case Stmt::ArraySubscriptExprClass: {
1456    // Array subscripts are potential references to data on the stack.  We
1457    // retrieve the DeclRefExpr* for the array variable if it indeed
1458    // has local storage.
1459    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase());
1460  }
1461
1462  case Stmt::ConditionalOperatorClass: {
1463    // For conditional operators we need to see if either the LHS or RHS are
1464    // non-NULL DeclRefExpr's.  If one is non-NULL, we return it.
1465    ConditionalOperator *C = cast<ConditionalOperator>(E);
1466
1467    // Handle the GNU extension for missing LHS.
1468    if (Expr *lhsExpr = C->getLHS())
1469      if (DeclRefExpr *LHS = EvalVal(lhsExpr))
1470        return LHS;
1471
1472    return EvalVal(C->getRHS());
1473  }
1474
1475  // Accesses to members are potential references to data on the stack.
1476  case Stmt::MemberExprClass: {
1477    MemberExpr *M = cast<MemberExpr>(E);
1478
1479    // Check for indirect access.  We only want direct field accesses.
1480    if (!M->isArrow())
1481      return EvalVal(M->getBase());
1482    else
1483      return NULL;
1484  }
1485
1486  // Everything else: we simply don't reason about them.
1487  default:
1488    return NULL;
1489  }
1490}
1491
1492//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
1493
1494/// Check for comparisons of floating point operands using != and ==.
1495/// Issue a warning if these are no self-comparisons, as they are not likely
1496/// to do what the programmer intended.
1497void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) {
1498  bool EmitWarning = true;
1499
1500  Expr* LeftExprSansParen = lex->IgnoreParens();
1501  Expr* RightExprSansParen = rex->IgnoreParens();
1502
1503  // Special case: check for x == x (which is OK).
1504  // Do not emit warnings for such cases.
1505  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
1506    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
1507      if (DRL->getDecl() == DRR->getDecl())
1508        EmitWarning = false;
1509
1510
1511  // Special case: check for comparisons against literals that can be exactly
1512  //  represented by APFloat.  In such cases, do not emit a warning.  This
1513  //  is a heuristic: often comparison against such literals are used to
1514  //  detect if a value in a variable has not changed.  This clearly can
1515  //  lead to false negatives.
1516  if (EmitWarning) {
1517    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
1518      if (FLL->isExact())
1519        EmitWarning = false;
1520    } else
1521      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
1522        if (FLR->isExact())
1523          EmitWarning = false;
1524    }
1525  }
1526
1527  // Check for comparisons with builtin types.
1528  if (EmitWarning)
1529    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
1530      if (CL->isBuiltinCall(Context))
1531        EmitWarning = false;
1532
1533  if (EmitWarning)
1534    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
1535      if (CR->isBuiltinCall(Context))
1536        EmitWarning = false;
1537
1538  // Emit the diagnostic.
1539  if (EmitWarning)
1540    Diag(loc, diag::warn_floatingpoint_eq)
1541      << lex->getSourceRange() << rex->getSourceRange();
1542}
1543