SemaChecking.cpp revision 813b70d6db54dbf149cfaa14ec7502142121d86f
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    // Does the index refer to the implicit object argument?
95    if (isa<CXXMemberCallExpr>(TheCall)) {
96      if (format_idx == 0)
97        return false;
98      --format_idx;
99    }
100    if (format_idx < TheCall->getNumArgs()) {
101      Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts();
102      if (!Format->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNull))
103        return true;
104    }
105  }
106  return false;
107}
108
109Action::OwningExprResult
110Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
111  OwningExprResult TheCallResult(Owned(TheCall));
112
113  switch (BuiltinID) {
114  case Builtin::BI__builtin___CFStringMakeConstantString:
115    assert(TheCall->getNumArgs() == 1 &&
116           "Wrong # arguments to builtin CFStringMakeConstantString");
117    if (CheckObjCString(TheCall->getArg(0)))
118      return ExprError();
119    break;
120  case Builtin::BI__builtin_stdarg_start:
121  case Builtin::BI__builtin_va_start:
122    if (SemaBuiltinVAStart(TheCall))
123      return ExprError();
124    break;
125  case Builtin::BI__builtin_isgreater:
126  case Builtin::BI__builtin_isgreaterequal:
127  case Builtin::BI__builtin_isless:
128  case Builtin::BI__builtin_islessequal:
129  case Builtin::BI__builtin_islessgreater:
130  case Builtin::BI__builtin_isunordered:
131    if (SemaBuiltinUnorderedCompare(TheCall))
132      return ExprError();
133    break;
134  case Builtin::BI__builtin_isfinite:
135  case Builtin::BI__builtin_isinf:
136  case Builtin::BI__builtin_isinf_sign:
137  case Builtin::BI__builtin_isnan:
138  case Builtin::BI__builtin_isnormal:
139    if (SemaBuiltinUnaryFP(TheCall))
140      return ExprError();
141    break;
142  case Builtin::BI__builtin_return_address:
143  case Builtin::BI__builtin_frame_address:
144    if (SemaBuiltinStackAddress(TheCall))
145      return ExprError();
146    break;
147  case Builtin::BI__builtin_eh_return_data_regno:
148    if (SemaBuiltinEHReturnDataRegNo(TheCall))
149      return ExprError();
150    break;
151  case Builtin::BI__builtin_shufflevector:
152    return SemaBuiltinShuffleVector(TheCall);
153    // TheCall will be freed by the smart pointer here, but that's fine, since
154    // SemaBuiltinShuffleVector guts it, but then doesn't release it.
155  case Builtin::BI__builtin_prefetch:
156    if (SemaBuiltinPrefetch(TheCall))
157      return ExprError();
158    break;
159  case Builtin::BI__builtin_object_size:
160    if (SemaBuiltinObjectSize(TheCall))
161      return ExprError();
162    break;
163  case Builtin::BI__builtin_longjmp:
164    if (SemaBuiltinLongjmp(TheCall))
165      return ExprError();
166    break;
167  case Builtin::BI__sync_fetch_and_add:
168  case Builtin::BI__sync_fetch_and_sub:
169  case Builtin::BI__sync_fetch_and_or:
170  case Builtin::BI__sync_fetch_and_and:
171  case Builtin::BI__sync_fetch_and_xor:
172  case Builtin::BI__sync_fetch_and_nand:
173  case Builtin::BI__sync_add_and_fetch:
174  case Builtin::BI__sync_sub_and_fetch:
175  case Builtin::BI__sync_and_and_fetch:
176  case Builtin::BI__sync_or_and_fetch:
177  case Builtin::BI__sync_xor_and_fetch:
178  case Builtin::BI__sync_nand_and_fetch:
179  case Builtin::BI__sync_val_compare_and_swap:
180  case Builtin::BI__sync_bool_compare_and_swap:
181  case Builtin::BI__sync_lock_test_and_set:
182  case Builtin::BI__sync_lock_release:
183    if (SemaBuiltinAtomicOverloaded(TheCall))
184      return ExprError();
185    break;
186  }
187
188  return move(TheCallResult);
189}
190
191/// CheckFunctionCall - Check a direct function call for various correctness
192/// and safety properties not strictly enforced by the C type system.
193bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) {
194  // Get the IdentifierInfo* for the called function.
195  IdentifierInfo *FnInfo = FDecl->getIdentifier();
196
197  // None of the checks below are needed for functions that don't have
198  // simple names (e.g., C++ conversion functions).
199  if (!FnInfo)
200    return false;
201
202  // FIXME: This mechanism should be abstracted to be less fragile and
203  // more efficient. For example, just map function ids to custom
204  // handlers.
205
206  // Printf checking.
207  if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) {
208    if (CheckablePrintfAttr(Format, TheCall)) {
209      bool HasVAListArg = Format->getFirstArg() == 0;
210      if (!HasVAListArg) {
211        if (const FunctionProtoType *Proto
212            = FDecl->getType()->getAs<FunctionProtoType>())
213          HasVAListArg = !Proto->isVariadic();
214      }
215      CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
216                           HasVAListArg ? 0 : Format->getFirstArg() - 1);
217    }
218  }
219
220  for (const NonNullAttr *NonNull = FDecl->getAttr<NonNullAttr>(); NonNull;
221       NonNull = NonNull->getNext<NonNullAttr>())
222    CheckNonNullArguments(NonNull, TheCall);
223
224  return false;
225}
226
227bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
228  // Printf checking.
229  const FormatAttr *Format = NDecl->getAttr<FormatAttr>();
230  if (!Format)
231    return false;
232
233  const VarDecl *V = dyn_cast<VarDecl>(NDecl);
234  if (!V)
235    return false;
236
237  QualType Ty = V->getType();
238  if (!Ty->isBlockPointerType())
239    return false;
240
241  if (!CheckablePrintfAttr(Format, TheCall))
242    return false;
243
244  bool HasVAListArg = Format->getFirstArg() == 0;
245  if (!HasVAListArg) {
246    const FunctionType *FT =
247      Ty->getAs<BlockPointerType>()->getPointeeType()->getAs<FunctionType>();
248    if (const FunctionProtoType *Proto = dyn_cast<FunctionProtoType>(FT))
249      HasVAListArg = !Proto->isVariadic();
250  }
251  CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
252                       HasVAListArg ? 0 : Format->getFirstArg() - 1);
253
254  return false;
255}
256
257/// SemaBuiltinAtomicOverloaded - We have a call to a function like
258/// __sync_fetch_and_add, which is an overloaded function based on the pointer
259/// type of its first argument.  The main ActOnCallExpr routines have already
260/// promoted the types of arguments because all of these calls are prototyped as
261/// void(...).
262///
263/// This function goes through and does final semantic checking for these
264/// builtins,
265bool Sema::SemaBuiltinAtomicOverloaded(CallExpr *TheCall) {
266  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
267  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
268
269  // Ensure that we have at least one argument to do type inference from.
270  if (TheCall->getNumArgs() < 1)
271    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
272              << 0 << TheCall->getCallee()->getSourceRange();
273
274  // Inspect the first argument of the atomic builtin.  This should always be
275  // a pointer type, whose element is an integral scalar or pointer type.
276  // Because it is a pointer type, we don't have to worry about any implicit
277  // casts here.
278  Expr *FirstArg = TheCall->getArg(0);
279  if (!FirstArg->getType()->isPointerType())
280    return Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
281             << FirstArg->getType() << FirstArg->getSourceRange();
282
283  QualType ValType = FirstArg->getType()->getAs<PointerType>()->getPointeeType();
284  if (!ValType->isIntegerType() && !ValType->isPointerType() &&
285      !ValType->isBlockPointerType())
286    return Diag(DRE->getLocStart(),
287                diag::err_atomic_builtin_must_be_pointer_intptr)
288             << FirstArg->getType() << FirstArg->getSourceRange();
289
290  // We need to figure out which concrete builtin this maps onto.  For example,
291  // __sync_fetch_and_add with a 2 byte object turns into
292  // __sync_fetch_and_add_2.
293#define BUILTIN_ROW(x) \
294  { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
295    Builtin::BI##x##_8, Builtin::BI##x##_16 }
296
297  static const unsigned BuiltinIndices[][5] = {
298    BUILTIN_ROW(__sync_fetch_and_add),
299    BUILTIN_ROW(__sync_fetch_and_sub),
300    BUILTIN_ROW(__sync_fetch_and_or),
301    BUILTIN_ROW(__sync_fetch_and_and),
302    BUILTIN_ROW(__sync_fetch_and_xor),
303    BUILTIN_ROW(__sync_fetch_and_nand),
304
305    BUILTIN_ROW(__sync_add_and_fetch),
306    BUILTIN_ROW(__sync_sub_and_fetch),
307    BUILTIN_ROW(__sync_and_and_fetch),
308    BUILTIN_ROW(__sync_or_and_fetch),
309    BUILTIN_ROW(__sync_xor_and_fetch),
310    BUILTIN_ROW(__sync_nand_and_fetch),
311
312    BUILTIN_ROW(__sync_val_compare_and_swap),
313    BUILTIN_ROW(__sync_bool_compare_and_swap),
314    BUILTIN_ROW(__sync_lock_test_and_set),
315    BUILTIN_ROW(__sync_lock_release)
316  };
317#undef BUILTIN_ROW
318
319  // Determine the index of the size.
320  unsigned SizeIndex;
321  switch (Context.getTypeSize(ValType)/8) {
322  case 1: SizeIndex = 0; break;
323  case 2: SizeIndex = 1; break;
324  case 4: SizeIndex = 2; break;
325  case 8: SizeIndex = 3; break;
326  case 16: SizeIndex = 4; break;
327  default:
328    return Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
329             << FirstArg->getType() << FirstArg->getSourceRange();
330  }
331
332  // Each of these builtins has one pointer argument, followed by some number of
333  // values (0, 1 or 2) followed by a potentially empty varags list of stuff
334  // that we ignore.  Find out which row of BuiltinIndices to read from as well
335  // as the number of fixed args.
336  unsigned BuiltinID = FDecl->getBuiltinID();
337  unsigned BuiltinIndex, NumFixed = 1;
338  switch (BuiltinID) {
339  default: assert(0 && "Unknown overloaded atomic builtin!");
340  case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break;
341  case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break;
342  case Builtin::BI__sync_fetch_and_or:  BuiltinIndex = 2; break;
343  case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break;
344  case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break;
345  case Builtin::BI__sync_fetch_and_nand:BuiltinIndex = 5; break;
346
347  case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 6; break;
348  case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 7; break;
349  case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 8; break;
350  case Builtin::BI__sync_or_and_fetch:  BuiltinIndex = 9; break;
351  case Builtin::BI__sync_xor_and_fetch: BuiltinIndex =10; break;
352  case Builtin::BI__sync_nand_and_fetch:BuiltinIndex =11; break;
353
354  case Builtin::BI__sync_val_compare_and_swap:
355    BuiltinIndex = 12;
356    NumFixed = 2;
357    break;
358  case Builtin::BI__sync_bool_compare_and_swap:
359    BuiltinIndex = 13;
360    NumFixed = 2;
361    break;
362  case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 14; break;
363  case Builtin::BI__sync_lock_release:
364    BuiltinIndex = 15;
365    NumFixed = 0;
366    break;
367  }
368
369  // Now that we know how many fixed arguments we expect, first check that we
370  // have at least that many.
371  if (TheCall->getNumArgs() < 1+NumFixed)
372    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
373            << 0 << TheCall->getCallee()->getSourceRange();
374
375
376  // Get the decl for the concrete builtin from this, we can tell what the
377  // concrete integer type we should convert to is.
378  unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
379  const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
380  IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
381  FunctionDecl *NewBuiltinDecl =
382    cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
383                                           TUScope, false, DRE->getLocStart()));
384  const FunctionProtoType *BuiltinFT =
385    NewBuiltinDecl->getType()->getAs<FunctionProtoType>();
386  ValType = BuiltinFT->getArgType(0)->getAs<PointerType>()->getPointeeType();
387
388  // If the first type needs to be converted (e.g. void** -> int*), do it now.
389  if (BuiltinFT->getArgType(0) != FirstArg->getType()) {
390    ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), CastExpr::CK_BitCast);
391    TheCall->setArg(0, FirstArg);
392  }
393
394  // Next, walk the valid ones promoting to the right type.
395  for (unsigned i = 0; i != NumFixed; ++i) {
396    Expr *Arg = TheCall->getArg(i+1);
397
398    // If the argument is an implicit cast, then there was a promotion due to
399    // "...", just remove it now.
400    if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) {
401      Arg = ICE->getSubExpr();
402      ICE->setSubExpr(0);
403      ICE->Destroy(Context);
404      TheCall->setArg(i+1, Arg);
405    }
406
407    // GCC does an implicit conversion to the pointer or integer ValType.  This
408    // can fail in some cases (1i -> int**), check for this error case now.
409    CastExpr::CastKind Kind = CastExpr::CK_Unknown;
410    CXXMethodDecl *ConversionDecl = 0;
411    if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind,
412                       ConversionDecl))
413      return true;
414
415    // Okay, we have something that *can* be converted to the right type.  Check
416    // to see if there is a potentially weird extension going on here.  This can
417    // happen when you do an atomic operation on something like an char* and
418    // pass in 42.  The 42 gets converted to char.  This is even more strange
419    // for things like 45.123 -> char, etc.
420    // FIXME: Do this check.
421    ImpCastExprToType(Arg, ValType, Kind, /*isLvalue=*/false);
422    TheCall->setArg(i+1, Arg);
423  }
424
425  // Switch the DeclRefExpr to refer to the new decl.
426  DRE->setDecl(NewBuiltinDecl);
427  DRE->setType(NewBuiltinDecl->getType());
428
429  // Set the callee in the CallExpr.
430  // FIXME: This leaks the original parens and implicit casts.
431  Expr *PromotedCall = DRE;
432  UsualUnaryConversions(PromotedCall);
433  TheCall->setCallee(PromotedCall);
434
435
436  // Change the result type of the call to match the result type of the decl.
437  TheCall->setType(NewBuiltinDecl->getResultType());
438  return false;
439}
440
441
442/// CheckObjCString - Checks that the argument to the builtin
443/// CFString constructor is correct
444/// FIXME: GCC currently emits the following warning:
445/// "warning: input conversion stopped due to an input byte that does not
446///           belong to the input codeset UTF-8"
447/// Note: It might also make sense to do the UTF-16 conversion here (would
448/// simplify the backend).
449bool Sema::CheckObjCString(Expr *Arg) {
450  Arg = Arg->IgnoreParenCasts();
451  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
452
453  if (!Literal || Literal->isWide()) {
454    Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
455      << Arg->getSourceRange();
456    return true;
457  }
458
459  const char *Data = Literal->getStrData();
460  unsigned Length = Literal->getByteLength();
461
462  for (unsigned i = 0; i < Length; ++i) {
463    if (!Data[i]) {
464      Diag(getLocationOfStringLiteralByte(Literal, i),
465           diag::warn_cfstring_literal_contains_nul_character)
466        << Arg->getSourceRange();
467      break;
468    }
469  }
470
471  return false;
472}
473
474/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
475/// Emit an error and return true on failure, return false on success.
476bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
477  Expr *Fn = TheCall->getCallee();
478  if (TheCall->getNumArgs() > 2) {
479    Diag(TheCall->getArg(2)->getLocStart(),
480         diag::err_typecheck_call_too_many_args)
481      << 0 /*function call*/ << Fn->getSourceRange()
482      << SourceRange(TheCall->getArg(2)->getLocStart(),
483                     (*(TheCall->arg_end()-1))->getLocEnd());
484    return true;
485  }
486
487  if (TheCall->getNumArgs() < 2) {
488    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
489      << 0 /*function call*/;
490  }
491
492  // Determine whether the current function is variadic or not.
493  bool isVariadic;
494  if (CurBlock)
495    isVariadic = CurBlock->isVariadic;
496  else if (getCurFunctionDecl()) {
497    if (FunctionProtoType* FTP =
498            dyn_cast<FunctionProtoType>(getCurFunctionDecl()->getType()))
499      isVariadic = FTP->isVariadic();
500    else
501      isVariadic = false;
502  } else {
503    isVariadic = getCurMethodDecl()->isVariadic();
504  }
505
506  if (!isVariadic) {
507    Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
508    return true;
509  }
510
511  // Verify that the second argument to the builtin is the last argument of the
512  // current function or method.
513  bool SecondArgIsLastNamedArgument = false;
514  const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
515
516  if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
517    if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
518      // FIXME: This isn't correct for methods (results in bogus warning).
519      // Get the last formal in the current function.
520      const ParmVarDecl *LastArg;
521      if (CurBlock)
522        LastArg = *(CurBlock->TheDecl->param_end()-1);
523      else if (FunctionDecl *FD = getCurFunctionDecl())
524        LastArg = *(FD->param_end()-1);
525      else
526        LastArg = *(getCurMethodDecl()->param_end()-1);
527      SecondArgIsLastNamedArgument = PV == LastArg;
528    }
529  }
530
531  if (!SecondArgIsLastNamedArgument)
532    Diag(TheCall->getArg(1)->getLocStart(),
533         diag::warn_second_parameter_of_va_start_not_last_named_argument);
534  return false;
535}
536
537/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
538/// friends.  This is declared to take (...), so we have to check everything.
539bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
540  if (TheCall->getNumArgs() < 2)
541    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
542      << 0 /*function call*/;
543  if (TheCall->getNumArgs() > 2)
544    return Diag(TheCall->getArg(2)->getLocStart(),
545                diag::err_typecheck_call_too_many_args)
546      << 0 /*function call*/
547      << SourceRange(TheCall->getArg(2)->getLocStart(),
548                     (*(TheCall->arg_end()-1))->getLocEnd());
549
550  Expr *OrigArg0 = TheCall->getArg(0);
551  Expr *OrigArg1 = TheCall->getArg(1);
552
553  // Do standard promotions between the two arguments, returning their common
554  // type.
555  QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
556
557  // Make sure any conversions are pushed back into the call; this is
558  // type safe since unordered compare builtins are declared as "_Bool
559  // foo(...)".
560  TheCall->setArg(0, OrigArg0);
561  TheCall->setArg(1, OrigArg1);
562
563  if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent())
564    return false;
565
566  // If the common type isn't a real floating type, then the arguments were
567  // invalid for this operation.
568  if (!Res->isRealFloatingType())
569    return Diag(OrigArg0->getLocStart(),
570                diag::err_typecheck_call_invalid_ordered_compare)
571      << OrigArg0->getType() << OrigArg1->getType()
572      << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd());
573
574  return false;
575}
576
577/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isnan and
578/// friends.  This is declared to take (...), so we have to check everything.
579bool Sema::SemaBuiltinUnaryFP(CallExpr *TheCall) {
580  if (TheCall->getNumArgs() < 1)
581    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
582      << 0 /*function call*/;
583  if (TheCall->getNumArgs() > 1)
584    return Diag(TheCall->getArg(1)->getLocStart(),
585                diag::err_typecheck_call_too_many_args)
586      << 0 /*function call*/
587      << SourceRange(TheCall->getArg(1)->getLocStart(),
588                     (*(TheCall->arg_end()-1))->getLocEnd());
589
590  Expr *OrigArg = TheCall->getArg(0);
591
592  if (OrigArg->isTypeDependent())
593    return false;
594
595  // This operation requires a floating-point number
596  if (!OrigArg->getType()->isRealFloatingType())
597    return Diag(OrigArg->getLocStart(),
598                diag::err_typecheck_call_invalid_unary_fp)
599      << OrigArg->getType() << OrigArg->getSourceRange();
600
601  return false;
602}
603
604bool Sema::SemaBuiltinStackAddress(CallExpr *TheCall) {
605  // The signature for these builtins is exact; the only thing we need
606  // to check is that the argument is a constant.
607  SourceLocation Loc;
608  if (!TheCall->getArg(0)->isTypeDependent() &&
609      !TheCall->getArg(0)->isValueDependent() &&
610      !TheCall->getArg(0)->isIntegerConstantExpr(Context, &Loc))
611    return Diag(Loc, diag::err_stack_const_level) << TheCall->getSourceRange();
612
613  return false;
614}
615
616/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
617// This is declared to take (...), so we have to check everything.
618Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
619  if (TheCall->getNumArgs() < 3)
620    return ExprError(Diag(TheCall->getLocEnd(),
621                          diag::err_typecheck_call_too_few_args)
622      << 0 /*function call*/ << TheCall->getSourceRange());
623
624  unsigned numElements = std::numeric_limits<unsigned>::max();
625  if (!TheCall->getArg(0)->isTypeDependent() &&
626      !TheCall->getArg(1)->isTypeDependent()) {
627    QualType FAType = TheCall->getArg(0)->getType();
628    QualType SAType = TheCall->getArg(1)->getType();
629
630    if (!FAType->isVectorType() || !SAType->isVectorType()) {
631      Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
632        << SourceRange(TheCall->getArg(0)->getLocStart(),
633                       TheCall->getArg(1)->getLocEnd());
634      return ExprError();
635    }
636
637    if (!Context.hasSameUnqualifiedType(FAType, SAType)) {
638      Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
639        << SourceRange(TheCall->getArg(0)->getLocStart(),
640                       TheCall->getArg(1)->getLocEnd());
641      return ExprError();
642    }
643
644    numElements = FAType->getAs<VectorType>()->getNumElements();
645    if (TheCall->getNumArgs() != numElements+2) {
646      if (TheCall->getNumArgs() < numElements+2)
647        return ExprError(Diag(TheCall->getLocEnd(),
648                              diag::err_typecheck_call_too_few_args)
649                 << 0 /*function call*/ << TheCall->getSourceRange());
650      return ExprError(Diag(TheCall->getLocEnd(),
651                            diag::err_typecheck_call_too_many_args)
652                 << 0 /*function call*/ << TheCall->getSourceRange());
653    }
654  }
655
656  for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
657    if (TheCall->getArg(i)->isTypeDependent() ||
658        TheCall->getArg(i)->isValueDependent())
659      continue;
660
661    llvm::APSInt Result(32);
662    if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
663      return ExprError(Diag(TheCall->getLocStart(),
664                  diag::err_shufflevector_nonconstant_argument)
665                << TheCall->getArg(i)->getSourceRange());
666
667    if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
668      return ExprError(Diag(TheCall->getLocStart(),
669                  diag::err_shufflevector_argument_too_large)
670               << TheCall->getArg(i)->getSourceRange());
671  }
672
673  llvm::SmallVector<Expr*, 32> exprs;
674
675  for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
676    exprs.push_back(TheCall->getArg(i));
677    TheCall->setArg(i, 0);
678  }
679
680  return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
681                                            exprs.size(), exprs[0]->getType(),
682                                            TheCall->getCallee()->getLocStart(),
683                                            TheCall->getRParenLoc()));
684}
685
686/// SemaBuiltinPrefetch - Handle __builtin_prefetch.
687// This is declared to take (const void*, ...) and can take two
688// optional constant int args.
689bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
690  unsigned NumArgs = TheCall->getNumArgs();
691
692  if (NumArgs > 3)
693    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_many_args)
694             << 0 /*function call*/ << TheCall->getSourceRange();
695
696  // Argument 0 is checked for us and the remaining arguments must be
697  // constant integers.
698  for (unsigned i = 1; i != NumArgs; ++i) {
699    Expr *Arg = TheCall->getArg(i);
700    if (Arg->isTypeDependent())
701      continue;
702
703    if (!Arg->getType()->isIntegralType())
704      return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_arg_type)
705              << Arg->getSourceRange();
706
707    ImpCastExprToType(Arg, Context.IntTy, CastExpr::CK_IntegralCast);
708    TheCall->setArg(i, Arg);
709
710    if (Arg->isValueDependent())
711      continue;
712
713    llvm::APSInt Result;
714    if (!Arg->isIntegerConstantExpr(Result, Context))
715      return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_arg_ice)
716        << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
717
718    // FIXME: gcc issues a warning and rewrites these to 0. These
719    // seems especially odd for the third argument since the default
720    // is 3.
721    if (i == 1) {
722      if (Result.getLimitedValue() > 1)
723        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
724             << "0" << "1" << Arg->getSourceRange();
725    } else {
726      if (Result.getLimitedValue() > 3)
727        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
728            << "0" << "3" << Arg->getSourceRange();
729    }
730  }
731
732  return false;
733}
734
735/// SemaBuiltinEHReturnDataRegNo - Handle __builtin_eh_return_data_regno, the
736/// operand must be an integer constant.
737bool Sema::SemaBuiltinEHReturnDataRegNo(CallExpr *TheCall) {
738  llvm::APSInt Result;
739  if (!TheCall->getArg(0)->isIntegerConstantExpr(Result, Context))
740    return Diag(TheCall->getLocStart(), diag::err_expr_not_ice)
741      << TheCall->getArg(0)->getSourceRange();
742
743  return false;
744}
745
746
747/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
748/// int type). This simply type checks that type is one of the defined
749/// constants (0-3).
750bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
751  Expr *Arg = TheCall->getArg(1);
752  if (Arg->isTypeDependent())
753    return false;
754
755  QualType ArgType = Arg->getType();
756  const BuiltinType *BT = ArgType->getAs<BuiltinType>();
757  llvm::APSInt Result(32);
758  if (!BT || BT->getKind() != BuiltinType::Int)
759    return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument)
760             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
761
762  if (Arg->isValueDependent())
763    return false;
764
765  if (!Arg->isIntegerConstantExpr(Result, Context)) {
766    return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument)
767             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
768  }
769
770  if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
771    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
772             << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
773  }
774
775  return false;
776}
777
778/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
779/// This checks that val is a constant 1.
780bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
781  Expr *Arg = TheCall->getArg(1);
782  if (Arg->isTypeDependent() || Arg->isValueDependent())
783    return false;
784
785  llvm::APSInt Result(32);
786  if (!Arg->isIntegerConstantExpr(Result, Context) || Result != 1)
787    return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
788             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
789
790  return false;
791}
792
793// Handle i > 1 ? "x" : "y", recursivelly
794bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall,
795                                  bool HasVAListArg,
796                                  unsigned format_idx, unsigned firstDataArg) {
797  if (E->isTypeDependent() || E->isValueDependent())
798    return false;
799
800  switch (E->getStmtClass()) {
801  case Stmt::ConditionalOperatorClass: {
802    const ConditionalOperator *C = cast<ConditionalOperator>(E);
803    return SemaCheckStringLiteral(C->getTrueExpr(), TheCall,
804                                  HasVAListArg, format_idx, firstDataArg)
805        && SemaCheckStringLiteral(C->getRHS(), TheCall,
806                                  HasVAListArg, format_idx, firstDataArg);
807  }
808
809  case Stmt::ImplicitCastExprClass: {
810    const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E);
811    return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg,
812                                  format_idx, firstDataArg);
813  }
814
815  case Stmt::ParenExprClass: {
816    const ParenExpr *Expr = cast<ParenExpr>(E);
817    return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg,
818                                  format_idx, firstDataArg);
819  }
820
821  case Stmt::DeclRefExprClass: {
822    const DeclRefExpr *DR = cast<DeclRefExpr>(E);
823
824    // As an exception, do not flag errors for variables binding to
825    // const string literals.
826    if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
827      bool isConstant = false;
828      QualType T = DR->getType();
829
830      if (const ArrayType *AT = Context.getAsArrayType(T)) {
831        isConstant = AT->getElementType().isConstant(Context);
832      } else if (const PointerType *PT = T->getAs<PointerType>()) {
833        isConstant = T.isConstant(Context) &&
834                     PT->getPointeeType().isConstant(Context);
835      }
836
837      if (isConstant) {
838        const VarDecl *Def = 0;
839        if (const Expr *Init = VD->getDefinition(Def))
840          return SemaCheckStringLiteral(Init, TheCall,
841                                        HasVAListArg, format_idx, firstDataArg);
842      }
843
844      // For vprintf* functions (i.e., HasVAListArg==true), we add a
845      // special check to see if the format string is a function parameter
846      // of the function calling the printf function.  If the function
847      // has an attribute indicating it is a printf-like function, then we
848      // should suppress warnings concerning non-literals being used in a call
849      // to a vprintf function.  For example:
850      //
851      // void
852      // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
853      //      va_list ap;
854      //      va_start(ap, fmt);
855      //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
856      //      ...
857      //
858      //
859      //  FIXME: We don't have full attribute support yet, so just check to see
860      //    if the argument is a DeclRefExpr that references a parameter.  We'll
861      //    add proper support for checking the attribute later.
862      if (HasVAListArg)
863        if (isa<ParmVarDecl>(VD))
864          return true;
865    }
866
867    return false;
868  }
869
870  case Stmt::CallExprClass: {
871    const CallExpr *CE = cast<CallExpr>(E);
872    if (const ImplicitCastExpr *ICE
873          = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
874      if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
875        if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
876          if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
877            unsigned ArgIndex = FA->getFormatIdx();
878            const Expr *Arg = CE->getArg(ArgIndex - 1);
879
880            return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg,
881                                          format_idx, firstDataArg);
882          }
883        }
884      }
885    }
886
887    return false;
888  }
889  case Stmt::ObjCStringLiteralClass:
890  case Stmt::StringLiteralClass: {
891    const StringLiteral *StrE = NULL;
892
893    if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
894      StrE = ObjCFExpr->getString();
895    else
896      StrE = cast<StringLiteral>(E);
897
898    if (StrE) {
899      CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx,
900                        firstDataArg);
901      return true;
902    }
903
904    return false;
905  }
906
907  default:
908    return false;
909  }
910}
911
912void
913Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
914                            const CallExpr *TheCall) {
915  for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end();
916       i != e; ++i) {
917    const Expr *ArgExpr = TheCall->getArg(*i);
918    if (ArgExpr->isNullPointerConstant(Context,
919                                       Expr::NPC_ValueDependentIsNotNull))
920      Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg)
921        << ArgExpr->getSourceRange();
922  }
923}
924
925/// CheckPrintfArguments - Check calls to printf (and similar functions) for
926/// correct use of format strings.
927///
928///  HasVAListArg - A predicate indicating whether the printf-like
929///    function is passed an explicit va_arg argument (e.g., vprintf)
930///
931///  format_idx - The index into Args for the format string.
932///
933/// Improper format strings to functions in the printf family can be
934/// the source of bizarre bugs and very serious security holes.  A
935/// good source of information is available in the following paper
936/// (which includes additional references):
937///
938///  FormatGuard: Automatic Protection From printf Format String
939///  Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001.
940///
941/// Functionality implemented:
942///
943///  We can statically check the following properties for string
944///  literal format strings for non v.*printf functions (where the
945///  arguments are passed directly):
946//
947///  (1) Are the number of format conversions equal to the number of
948///      data arguments?
949///
950///  (2) Does each format conversion correctly match the type of the
951///      corresponding data argument?  (TODO)
952///
953/// Moreover, for all printf functions we can:
954///
955///  (3) Check for a missing format string (when not caught by type checking).
956///
957///  (4) Check for no-operation flags; e.g. using "#" with format
958///      conversion 'c'  (TODO)
959///
960///  (5) Check the use of '%n', a major source of security holes.
961///
962///  (6) Check for malformed format conversions that don't specify anything.
963///
964///  (7) Check for empty format strings.  e.g: printf("");
965///
966///  (8) Check that the format string is a wide literal.
967///
968///  (9) Also check the arguments of functions with the __format__ attribute.
969///      (TODO).
970///
971/// All of these checks can be done by parsing the format string.
972///
973/// For now, we ONLY do (1), (3), (5), (6), (7), and (8).
974void
975Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg,
976                           unsigned format_idx, unsigned firstDataArg) {
977  const Expr *Fn = TheCall->getCallee();
978
979  // The way the format attribute works in GCC, the implicit this argument
980  // of member functions is counted. However, it doesn't appear in our own
981  // lists, so decrement format_idx in that case.
982  if (isa<CXXMemberCallExpr>(TheCall)) {
983    // Catch a format attribute mistakenly referring to the object argument.
984    if (format_idx == 0)
985      return;
986    --format_idx;
987    if(firstDataArg != 0)
988      --firstDataArg;
989  }
990
991  // CHECK: printf-like function is called with no format string.
992  if (format_idx >= TheCall->getNumArgs()) {
993    Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string)
994      << Fn->getSourceRange();
995    return;
996  }
997
998  const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts();
999
1000  // CHECK: format string is not a string literal.
1001  //
1002  // Dynamically generated format strings are difficult to
1003  // automatically vet at compile time.  Requiring that format strings
1004  // are string literals: (1) permits the checking of format strings by
1005  // the compiler and thereby (2) can practically remove the source of
1006  // many format string exploits.
1007
1008  // Format string can be either ObjC string (e.g. @"%d") or
1009  // C string (e.g. "%d")
1010  // ObjC string uses the same format specifiers as C string, so we can use
1011  // the same format string checking logic for both ObjC and C strings.
1012  if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx,
1013                             firstDataArg))
1014    return;  // Literal format string found, check done!
1015
1016  // If there are no arguments specified, warn with -Wformat-security, otherwise
1017  // warn only with -Wformat-nonliteral.
1018  if (TheCall->getNumArgs() == format_idx+1)
1019    Diag(TheCall->getArg(format_idx)->getLocStart(),
1020         diag::warn_printf_nonliteral_noargs)
1021      << OrigFormatExpr->getSourceRange();
1022  else
1023    Diag(TheCall->getArg(format_idx)->getLocStart(),
1024         diag::warn_printf_nonliteral)
1025           << OrigFormatExpr->getSourceRange();
1026}
1027
1028void Sema::CheckPrintfString(const StringLiteral *FExpr,
1029                             const Expr *OrigFormatExpr,
1030                             const CallExpr *TheCall, bool HasVAListArg,
1031                             unsigned format_idx, unsigned firstDataArg) {
1032
1033  const ObjCStringLiteral *ObjCFExpr =
1034    dyn_cast<ObjCStringLiteral>(OrigFormatExpr);
1035
1036  // CHECK: is the format string a wide literal?
1037  if (FExpr->isWide()) {
1038    Diag(FExpr->getLocStart(),
1039         diag::warn_printf_format_string_is_wide_literal)
1040      << OrigFormatExpr->getSourceRange();
1041    return;
1042  }
1043
1044  // Str - The format string.  NOTE: this is NOT null-terminated!
1045  const char *Str = FExpr->getStrData();
1046
1047  // CHECK: empty format string?
1048  unsigned StrLen = FExpr->getByteLength();
1049
1050  if (StrLen == 0) {
1051    Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string)
1052      << OrigFormatExpr->getSourceRange();
1053    return;
1054  }
1055
1056  // We process the format string using a binary state machine.  The
1057  // current state is stored in CurrentState.
1058  enum {
1059    state_OrdChr,
1060    state_Conversion
1061  } CurrentState = state_OrdChr;
1062
1063  // numConversions - The number of conversions seen so far.  This is
1064  //  incremented as we traverse the format string.
1065  unsigned numConversions = 0;
1066
1067  // numDataArgs - The number of data arguments after the format
1068  //  string.  This can only be determined for non vprintf-like
1069  //  functions.  For those functions, this value is 1 (the sole
1070  //  va_arg argument).
1071  unsigned numDataArgs = TheCall->getNumArgs()-firstDataArg;
1072
1073  // Inspect the format string.
1074  unsigned StrIdx = 0;
1075
1076  // LastConversionIdx - Index within the format string where we last saw
1077  //  a '%' character that starts a new format conversion.
1078  unsigned LastConversionIdx = 0;
1079
1080  for (; StrIdx < StrLen; ++StrIdx) {
1081
1082    // Is the number of detected conversion conversions greater than
1083    // the number of matching data arguments?  If so, stop.
1084    if (!HasVAListArg && numConversions > numDataArgs) break;
1085
1086    // Handle "\0"
1087    if (Str[StrIdx] == '\0') {
1088      // The string returned by getStrData() is not null-terminated,
1089      // so the presence of a null character is likely an error.
1090      Diag(getLocationOfStringLiteralByte(FExpr, StrIdx),
1091           diag::warn_printf_format_string_contains_null_char)
1092        <<  OrigFormatExpr->getSourceRange();
1093      return;
1094    }
1095
1096    // Ordinary characters (not processing a format conversion).
1097    if (CurrentState == state_OrdChr) {
1098      if (Str[StrIdx] == '%') {
1099        CurrentState = state_Conversion;
1100        LastConversionIdx = StrIdx;
1101      }
1102      continue;
1103    }
1104
1105    // Seen '%'.  Now processing a format conversion.
1106    switch (Str[StrIdx]) {
1107    // Handle dynamic precision or width specifier.
1108    case '*': {
1109      ++numConversions;
1110
1111      if (!HasVAListArg) {
1112        if (numConversions > numDataArgs) {
1113          SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx);
1114
1115          if (Str[StrIdx-1] == '.')
1116            Diag(Loc, diag::warn_printf_asterisk_precision_missing_arg)
1117              << OrigFormatExpr->getSourceRange();
1118          else
1119            Diag(Loc, diag::warn_printf_asterisk_width_missing_arg)
1120              << OrigFormatExpr->getSourceRange();
1121
1122          // Don't do any more checking.  We'll just emit spurious errors.
1123          return;
1124        }
1125
1126        // Perform type checking on width/precision specifier.
1127        const Expr *E = TheCall->getArg(format_idx+numConversions);
1128        if (const BuiltinType *BT = E->getType()->getAs<BuiltinType>())
1129          if (BT->getKind() == BuiltinType::Int)
1130            break;
1131
1132        SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx);
1133
1134        if (Str[StrIdx-1] == '.')
1135          Diag(Loc, diag::warn_printf_asterisk_precision_wrong_type)
1136          << E->getType() << E->getSourceRange();
1137        else
1138          Diag(Loc, diag::warn_printf_asterisk_width_wrong_type)
1139          << E->getType() << E->getSourceRange();
1140
1141        break;
1142      }
1143    }
1144
1145    // Characters which can terminate a format conversion
1146    // (e.g. "%d").  Characters that specify length modifiers or
1147    // other flags are handled by the default case below.
1148    //
1149    // FIXME: additional checks will go into the following cases.
1150    case 'i':
1151    case 'd':
1152    case 'o':
1153    case 'u':
1154    case 'x':
1155    case 'X':
1156    case 'D':
1157    case 'O':
1158    case 'U':
1159    case 'e':
1160    case 'E':
1161    case 'f':
1162    case 'F':
1163    case 'g':
1164    case 'G':
1165    case 'a':
1166    case 'A':
1167    case 'c':
1168    case 'C':
1169    case 'S':
1170    case 's':
1171    case 'p':
1172      ++numConversions;
1173      CurrentState = state_OrdChr;
1174      break;
1175
1176    case 'm':
1177      // FIXME: Warn in situations where this isn't supported!
1178      CurrentState = state_OrdChr;
1179      break;
1180
1181    // CHECK: Are we using "%n"?  Issue a warning.
1182    case 'n': {
1183      ++numConversions;
1184      CurrentState = state_OrdChr;
1185      SourceLocation Loc = getLocationOfStringLiteralByte(FExpr,
1186                                                          LastConversionIdx);
1187
1188      Diag(Loc, diag::warn_printf_write_back)<<OrigFormatExpr->getSourceRange();
1189      break;
1190    }
1191
1192    // Handle "%@"
1193    case '@':
1194      // %@ is allowed in ObjC format strings only.
1195      if (ObjCFExpr != NULL)
1196        CurrentState = state_OrdChr;
1197      else {
1198        // Issue a warning: invalid format conversion.
1199        SourceLocation Loc =
1200          getLocationOfStringLiteralByte(FExpr, LastConversionIdx);
1201
1202        Diag(Loc, diag::warn_printf_invalid_conversion)
1203          <<  std::string(Str+LastConversionIdx,
1204                          Str+std::min(LastConversionIdx+2, StrLen))
1205          << OrigFormatExpr->getSourceRange();
1206      }
1207      ++numConversions;
1208      break;
1209
1210    // Handle "%%"
1211    case '%':
1212      // Sanity check: Was the first "%" character the previous one?
1213      // If not, we will assume that we have a malformed format
1214      // conversion, and that the current "%" character is the start
1215      // of a new conversion.
1216      if (StrIdx - LastConversionIdx == 1)
1217        CurrentState = state_OrdChr;
1218      else {
1219        // Issue a warning: invalid format conversion.
1220        SourceLocation Loc =
1221          getLocationOfStringLiteralByte(FExpr, LastConversionIdx);
1222
1223        Diag(Loc, diag::warn_printf_invalid_conversion)
1224          << std::string(Str+LastConversionIdx, Str+StrIdx)
1225          << OrigFormatExpr->getSourceRange();
1226
1227        // This conversion is broken.  Advance to the next format
1228        // conversion.
1229        LastConversionIdx = StrIdx;
1230        ++numConversions;
1231      }
1232      break;
1233
1234    default:
1235      // This case catches all other characters: flags, widths, etc.
1236      // We should eventually process those as well.
1237      break;
1238    }
1239  }
1240
1241  if (CurrentState == state_Conversion) {
1242    // Issue a warning: invalid format conversion.
1243    SourceLocation Loc =
1244      getLocationOfStringLiteralByte(FExpr, LastConversionIdx);
1245
1246    Diag(Loc, diag::warn_printf_invalid_conversion)
1247      << std::string(Str+LastConversionIdx,
1248                     Str+std::min(LastConversionIdx+2, StrLen))
1249      << OrigFormatExpr->getSourceRange();
1250    return;
1251  }
1252
1253  if (!HasVAListArg) {
1254    // CHECK: Does the number of format conversions exceed the number
1255    //        of data arguments?
1256    if (numConversions > numDataArgs) {
1257      SourceLocation Loc =
1258        getLocationOfStringLiteralByte(FExpr, LastConversionIdx);
1259
1260      Diag(Loc, diag::warn_printf_insufficient_data_args)
1261        << OrigFormatExpr->getSourceRange();
1262    }
1263    // CHECK: Does the number of data arguments exceed the number of
1264    //        format conversions in the format string?
1265    else if (numConversions < numDataArgs)
1266      Diag(TheCall->getArg(format_idx+numConversions+1)->getLocStart(),
1267           diag::warn_printf_too_many_data_args)
1268        << OrigFormatExpr->getSourceRange();
1269  }
1270}
1271
1272//===--- CHECK: Return Address of Stack Variable --------------------------===//
1273
1274static DeclRefExpr* EvalVal(Expr *E);
1275static DeclRefExpr* EvalAddr(Expr* E);
1276
1277/// CheckReturnStackAddr - Check if a return statement returns the address
1278///   of a stack variable.
1279void
1280Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
1281                           SourceLocation ReturnLoc) {
1282
1283  // Perform checking for returned stack addresses.
1284  if (lhsType->isPointerType() || lhsType->isBlockPointerType()) {
1285    if (DeclRefExpr *DR = EvalAddr(RetValExp))
1286      Diag(DR->getLocStart(), diag::warn_ret_stack_addr)
1287       << DR->getDecl()->getDeclName() << RetValExp->getSourceRange();
1288
1289    // Skip over implicit cast expressions when checking for block expressions.
1290    RetValExp = RetValExp->IgnoreParenCasts();
1291
1292    if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp))
1293      if (C->hasBlockDeclRefExprs())
1294        Diag(C->getLocStart(), diag::err_ret_local_block)
1295          << C->getSourceRange();
1296
1297    if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp))
1298      Diag(ALE->getLocStart(), diag::warn_ret_addr_label)
1299        << ALE->getSourceRange();
1300
1301  } else if (lhsType->isReferenceType()) {
1302    // Perform checking for stack values returned by reference.
1303    // Check for a reference to the stack
1304    if (DeclRefExpr *DR = EvalVal(RetValExp))
1305      Diag(DR->getLocStart(), diag::warn_ret_stack_ref)
1306        << DR->getDecl()->getDeclName() << RetValExp->getSourceRange();
1307  }
1308}
1309
1310/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
1311///  check if the expression in a return statement evaluates to an address
1312///  to a location on the stack.  The recursion is used to traverse the
1313///  AST of the return expression, with recursion backtracking when we
1314///  encounter a subexpression that (1) clearly does not lead to the address
1315///  of a stack variable or (2) is something we cannot determine leads to
1316///  the address of a stack variable based on such local checking.
1317///
1318///  EvalAddr processes expressions that are pointers that are used as
1319///  references (and not L-values).  EvalVal handles all other values.
1320///  At the base case of the recursion is a check for a DeclRefExpr* in
1321///  the refers to a stack variable.
1322///
1323///  This implementation handles:
1324///
1325///   * pointer-to-pointer casts
1326///   * implicit conversions from array references to pointers
1327///   * taking the address of fields
1328///   * arbitrary interplay between "&" and "*" operators
1329///   * pointer arithmetic from an address of a stack variable
1330///   * taking the address of an array element where the array is on the stack
1331static DeclRefExpr* EvalAddr(Expr *E) {
1332  // We should only be called for evaluating pointer expressions.
1333  assert((E->getType()->isAnyPointerType() ||
1334          E->getType()->isBlockPointerType() ||
1335          E->getType()->isObjCQualifiedIdType()) &&
1336         "EvalAddr only works on pointers");
1337
1338  // Our "symbolic interpreter" is just a dispatch off the currently
1339  // viewed AST node.  We then recursively traverse the AST by calling
1340  // EvalAddr and EvalVal appropriately.
1341  switch (E->getStmtClass()) {
1342  case Stmt::ParenExprClass:
1343    // Ignore parentheses.
1344    return EvalAddr(cast<ParenExpr>(E)->getSubExpr());
1345
1346  case Stmt::UnaryOperatorClass: {
1347    // The only unary operator that make sense to handle here
1348    // is AddrOf.  All others don't make sense as pointers.
1349    UnaryOperator *U = cast<UnaryOperator>(E);
1350
1351    if (U->getOpcode() == UnaryOperator::AddrOf)
1352      return EvalVal(U->getSubExpr());
1353    else
1354      return NULL;
1355  }
1356
1357  case Stmt::BinaryOperatorClass: {
1358    // Handle pointer arithmetic.  All other binary operators are not valid
1359    // in this context.
1360    BinaryOperator *B = cast<BinaryOperator>(E);
1361    BinaryOperator::Opcode op = B->getOpcode();
1362
1363    if (op != BinaryOperator::Add && op != BinaryOperator::Sub)
1364      return NULL;
1365
1366    Expr *Base = B->getLHS();
1367
1368    // Determine which argument is the real pointer base.  It could be
1369    // the RHS argument instead of the LHS.
1370    if (!Base->getType()->isPointerType()) Base = B->getRHS();
1371
1372    assert (Base->getType()->isPointerType());
1373    return EvalAddr(Base);
1374  }
1375
1376  // For conditional operators we need to see if either the LHS or RHS are
1377  // valid DeclRefExpr*s.  If one of them is valid, we return it.
1378  case Stmt::ConditionalOperatorClass: {
1379    ConditionalOperator *C = cast<ConditionalOperator>(E);
1380
1381    // Handle the GNU extension for missing LHS.
1382    if (Expr *lhsExpr = C->getLHS())
1383      if (DeclRefExpr* LHS = EvalAddr(lhsExpr))
1384        return LHS;
1385
1386     return EvalAddr(C->getRHS());
1387  }
1388
1389  // For casts, we need to handle conversions from arrays to
1390  // pointer values, and pointer-to-pointer conversions.
1391  case Stmt::ImplicitCastExprClass:
1392  case Stmt::CStyleCastExprClass:
1393  case Stmt::CXXFunctionalCastExprClass: {
1394    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
1395    QualType T = SubExpr->getType();
1396
1397    if (SubExpr->getType()->isPointerType() ||
1398        SubExpr->getType()->isBlockPointerType() ||
1399        SubExpr->getType()->isObjCQualifiedIdType())
1400      return EvalAddr(SubExpr);
1401    else if (T->isArrayType())
1402      return EvalVal(SubExpr);
1403    else
1404      return 0;
1405  }
1406
1407  // C++ casts.  For dynamic casts, static casts, and const casts, we
1408  // are always converting from a pointer-to-pointer, so we just blow
1409  // through the cast.  In the case the dynamic cast doesn't fail (and
1410  // return NULL), we take the conservative route and report cases
1411  // where we return the address of a stack variable.  For Reinterpre
1412  // FIXME: The comment about is wrong; we're not always converting
1413  // from pointer to pointer. I'm guessing that this code should also
1414  // handle references to objects.
1415  case Stmt::CXXStaticCastExprClass:
1416  case Stmt::CXXDynamicCastExprClass:
1417  case Stmt::CXXConstCastExprClass:
1418  case Stmt::CXXReinterpretCastExprClass: {
1419      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
1420      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
1421        return EvalAddr(S);
1422      else
1423        return NULL;
1424  }
1425
1426  // Everything else: we simply don't reason about them.
1427  default:
1428    return NULL;
1429  }
1430}
1431
1432
1433///  EvalVal - This function is complements EvalAddr in the mutual recursion.
1434///   See the comments for EvalAddr for more details.
1435static DeclRefExpr* EvalVal(Expr *E) {
1436
1437  // We should only be called for evaluating non-pointer expressions, or
1438  // expressions with a pointer type that are not used as references but instead
1439  // are l-values (e.g., DeclRefExpr with a pointer type).
1440
1441  // Our "symbolic interpreter" is just a dispatch off the currently
1442  // viewed AST node.  We then recursively traverse the AST by calling
1443  // EvalAddr and EvalVal appropriately.
1444  switch (E->getStmtClass()) {
1445  case Stmt::DeclRefExprClass: {
1446    // DeclRefExpr: the base case.  When we hit a DeclRefExpr we are looking
1447    //  at code that refers to a variable's name.  We check if it has local
1448    //  storage within the function, and if so, return the expression.
1449    DeclRefExpr *DR = cast<DeclRefExpr>(E);
1450
1451    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
1452      if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR;
1453
1454    return NULL;
1455  }
1456
1457  case Stmt::ParenExprClass:
1458    // Ignore parentheses.
1459    return EvalVal(cast<ParenExpr>(E)->getSubExpr());
1460
1461  case Stmt::UnaryOperatorClass: {
1462    // The only unary operator that make sense to handle here
1463    // is Deref.  All others don't resolve to a "name."  This includes
1464    // handling all sorts of rvalues passed to a unary operator.
1465    UnaryOperator *U = cast<UnaryOperator>(E);
1466
1467    if (U->getOpcode() == UnaryOperator::Deref)
1468      return EvalAddr(U->getSubExpr());
1469
1470    return NULL;
1471  }
1472
1473  case Stmt::ArraySubscriptExprClass: {
1474    // Array subscripts are potential references to data on the stack.  We
1475    // retrieve the DeclRefExpr* for the array variable if it indeed
1476    // has local storage.
1477    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase());
1478  }
1479
1480  case Stmt::ConditionalOperatorClass: {
1481    // For conditional operators we need to see if either the LHS or RHS are
1482    // non-NULL DeclRefExpr's.  If one is non-NULL, we return it.
1483    ConditionalOperator *C = cast<ConditionalOperator>(E);
1484
1485    // Handle the GNU extension for missing LHS.
1486    if (Expr *lhsExpr = C->getLHS())
1487      if (DeclRefExpr *LHS = EvalVal(lhsExpr))
1488        return LHS;
1489
1490    return EvalVal(C->getRHS());
1491  }
1492
1493  // Accesses to members are potential references to data on the stack.
1494  case Stmt::MemberExprClass: {
1495    MemberExpr *M = cast<MemberExpr>(E);
1496
1497    // Check for indirect access.  We only want direct field accesses.
1498    if (!M->isArrow())
1499      return EvalVal(M->getBase());
1500    else
1501      return NULL;
1502  }
1503
1504  // Everything else: we simply don't reason about them.
1505  default:
1506    return NULL;
1507  }
1508}
1509
1510//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
1511
1512/// Check for comparisons of floating point operands using != and ==.
1513/// Issue a warning if these are no self-comparisons, as they are not likely
1514/// to do what the programmer intended.
1515void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) {
1516  bool EmitWarning = true;
1517
1518  Expr* LeftExprSansParen = lex->IgnoreParens();
1519  Expr* RightExprSansParen = rex->IgnoreParens();
1520
1521  // Special case: check for x == x (which is OK).
1522  // Do not emit warnings for such cases.
1523  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
1524    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
1525      if (DRL->getDecl() == DRR->getDecl())
1526        EmitWarning = false;
1527
1528
1529  // Special case: check for comparisons against literals that can be exactly
1530  //  represented by APFloat.  In such cases, do not emit a warning.  This
1531  //  is a heuristic: often comparison against such literals are used to
1532  //  detect if a value in a variable has not changed.  This clearly can
1533  //  lead to false negatives.
1534  if (EmitWarning) {
1535    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
1536      if (FLL->isExact())
1537        EmitWarning = false;
1538    } else
1539      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
1540        if (FLR->isExact())
1541          EmitWarning = false;
1542    }
1543  }
1544
1545  // Check for comparisons with builtin types.
1546  if (EmitWarning)
1547    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
1548      if (CL->isBuiltinCall(Context))
1549        EmitWarning = false;
1550
1551  if (EmitWarning)
1552    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
1553      if (CR->isBuiltinCall(Context))
1554        EmitWarning = false;
1555
1556  // Emit the diagnostic.
1557  if (EmitWarning)
1558    Diag(loc, diag::warn_floatingpoint_eq)
1559      << lex->getSourceRange() << rex->getSourceRange();
1560}
1561