SemaChecking.cpp revision 22d4fed17df02f5fa07165a69228b0b0f1a73a28
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 "clang/Sema/Sema.h"
16#include "clang/Sema/SemaInternal.h"
17#include "clang/Sema/ScopeInfo.h"
18#include "clang/Analysis/Analyses/FormatString.h"
19#include "clang/AST/ASTContext.h"
20#include "clang/AST/CharUnits.h"
21#include "clang/AST/DeclCXX.h"
22#include "clang/AST/DeclObjC.h"
23#include "clang/AST/ExprCXX.h"
24#include "clang/AST/ExprObjC.h"
25#include "clang/AST/EvaluatedExprVisitor.h"
26#include "clang/AST/DeclObjC.h"
27#include "clang/AST/StmtCXX.h"
28#include "clang/AST/StmtObjC.h"
29#include "clang/Lex/Preprocessor.h"
30#include "llvm/ADT/BitVector.h"
31#include "llvm/ADT/STLExtras.h"
32#include "llvm/Support/raw_ostream.h"
33#include "clang/Basic/TargetBuiltins.h"
34#include "clang/Basic/TargetInfo.h"
35#include "clang/Basic/ConvertUTF.h"
36#include <limits>
37using namespace clang;
38using namespace sema;
39
40SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL,
41                                                    unsigned ByteNo) const {
42  return SL->getLocationOfByte(ByteNo, PP.getSourceManager(),
43                               PP.getLangOptions(), PP.getTargetInfo());
44}
45
46
47/// CheckablePrintfAttr - does a function call have a "printf" attribute
48/// and arguments that merit checking?
49bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) {
50  if (Format->getType() == "printf") return true;
51  if (Format->getType() == "printf0") {
52    // printf0 allows null "format" string; if so don't check format/args
53    unsigned format_idx = Format->getFormatIdx() - 1;
54    // Does the index refer to the implicit object argument?
55    if (isa<CXXMemberCallExpr>(TheCall)) {
56      if (format_idx == 0)
57        return false;
58      --format_idx;
59    }
60    if (format_idx < TheCall->getNumArgs()) {
61      Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts();
62      if (!Format->isNullPointerConstant(Context,
63                                         Expr::NPC_ValueDependentIsNull))
64        return true;
65    }
66  }
67  return false;
68}
69
70/// Checks that a call expression's argument count is the desired number.
71/// This is useful when doing custom type-checking.  Returns true on error.
72static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) {
73  unsigned argCount = call->getNumArgs();
74  if (argCount == desiredArgCount) return false;
75
76  if (argCount < desiredArgCount)
77    return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args)
78        << 0 /*function call*/ << desiredArgCount << argCount
79        << call->getSourceRange();
80
81  // Highlight all the excess arguments.
82  SourceRange range(call->getArg(desiredArgCount)->getLocStart(),
83                    call->getArg(argCount - 1)->getLocEnd());
84
85  return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args)
86    << 0 /*function call*/ << desiredArgCount << argCount
87    << call->getArg(1)->getSourceRange();
88}
89
90ExprResult
91Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
92  ExprResult TheCallResult(Owned(TheCall));
93
94  // Find out if any arguments are required to be integer constant expressions.
95  unsigned ICEArguments = 0;
96  ASTContext::GetBuiltinTypeError Error;
97  Context.GetBuiltinType(BuiltinID, Error, &ICEArguments);
98  if (Error != ASTContext::GE_None)
99    ICEArguments = 0;  // Don't diagnose previously diagnosed errors.
100
101  // If any arguments are required to be ICE's, check and diagnose.
102  for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) {
103    // Skip arguments not required to be ICE's.
104    if ((ICEArguments & (1 << ArgNo)) == 0) continue;
105
106    llvm::APSInt Result;
107    if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
108      return true;
109    ICEArguments &= ~(1 << ArgNo);
110  }
111
112  switch (BuiltinID) {
113  case Builtin::BI__builtin___CFStringMakeConstantString:
114    assert(TheCall->getNumArgs() == 1 &&
115           "Wrong # arguments to builtin CFStringMakeConstantString");
116    if (CheckObjCString(TheCall->getArg(0)))
117      return ExprError();
118    break;
119  case Builtin::BI__builtin_stdarg_start:
120  case Builtin::BI__builtin_va_start:
121    if (SemaBuiltinVAStart(TheCall))
122      return ExprError();
123    break;
124  case Builtin::BI__builtin_isgreater:
125  case Builtin::BI__builtin_isgreaterequal:
126  case Builtin::BI__builtin_isless:
127  case Builtin::BI__builtin_islessequal:
128  case Builtin::BI__builtin_islessgreater:
129  case Builtin::BI__builtin_isunordered:
130    if (SemaBuiltinUnorderedCompare(TheCall))
131      return ExprError();
132    break;
133  case Builtin::BI__builtin_fpclassify:
134    if (SemaBuiltinFPClassification(TheCall, 6))
135      return ExprError();
136    break;
137  case Builtin::BI__builtin_isfinite:
138  case Builtin::BI__builtin_isinf:
139  case Builtin::BI__builtin_isinf_sign:
140  case Builtin::BI__builtin_isnan:
141  case Builtin::BI__builtin_isnormal:
142    if (SemaBuiltinFPClassification(TheCall, 1))
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
162  case Builtin::BI__builtin_classify_type:
163    if (checkArgCount(*this, TheCall, 1)) return true;
164    TheCall->setType(Context.IntTy);
165    break;
166  case Builtin::BI__builtin_constant_p:
167    if (checkArgCount(*this, TheCall, 1)) return true;
168    TheCall->setType(Context.IntTy);
169    break;
170  case Builtin::BI__sync_fetch_and_add:
171  case Builtin::BI__sync_fetch_and_sub:
172  case Builtin::BI__sync_fetch_and_or:
173  case Builtin::BI__sync_fetch_and_and:
174  case Builtin::BI__sync_fetch_and_xor:
175  case Builtin::BI__sync_add_and_fetch:
176  case Builtin::BI__sync_sub_and_fetch:
177  case Builtin::BI__sync_and_and_fetch:
178  case Builtin::BI__sync_or_and_fetch:
179  case Builtin::BI__sync_xor_and_fetch:
180  case Builtin::BI__sync_val_compare_and_swap:
181  case Builtin::BI__sync_bool_compare_and_swap:
182  case Builtin::BI__sync_lock_test_and_set:
183  case Builtin::BI__sync_lock_release:
184  case Builtin::BI__sync_swap:
185    return SemaBuiltinAtomicOverloaded(move(TheCallResult));
186  }
187
188  // Since the target specific builtins for each arch overlap, only check those
189  // of the arch we are compiling for.
190  if (BuiltinID >= Builtin::FirstTSBuiltin) {
191    switch (Context.Target.getTriple().getArch()) {
192      case llvm::Triple::arm:
193      case llvm::Triple::thumb:
194        if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall))
195          return ExprError();
196        break;
197      default:
198        break;
199    }
200  }
201
202  return move(TheCallResult);
203}
204
205// Get the valid immediate range for the specified NEON type code.
206static unsigned RFT(unsigned t, bool shift = false) {
207  bool quad = t & 0x10;
208
209  switch (t & 0x7) {
210    case 0: // i8
211      return shift ? 7 : (8 << (int)quad) - 1;
212    case 1: // i16
213      return shift ? 15 : (4 << (int)quad) - 1;
214    case 2: // i32
215      return shift ? 31 : (2 << (int)quad) - 1;
216    case 3: // i64
217      return shift ? 63 : (1 << (int)quad) - 1;
218    case 4: // f32
219      assert(!shift && "cannot shift float types!");
220      return (2 << (int)quad) - 1;
221    case 5: // poly8
222      return shift ? 7 : (8 << (int)quad) - 1;
223    case 6: // poly16
224      return shift ? 15 : (4 << (int)quad) - 1;
225    case 7: // float16
226      assert(!shift && "cannot shift float types!");
227      return (4 << (int)quad) - 1;
228  }
229  return 0;
230}
231
232bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
233  llvm::APSInt Result;
234
235  unsigned mask = 0;
236  unsigned TV = 0;
237  switch (BuiltinID) {
238#define GET_NEON_OVERLOAD_CHECK
239#include "clang/Basic/arm_neon.inc"
240#undef GET_NEON_OVERLOAD_CHECK
241  }
242
243  // For NEON intrinsics which are overloaded on vector element type, validate
244  // the immediate which specifies which variant to emit.
245  if (mask) {
246    unsigned ArgNo = TheCall->getNumArgs()-1;
247    if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
248      return true;
249
250    TV = Result.getLimitedValue(32);
251    if ((TV > 31) || (mask & (1 << TV)) == 0)
252      return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code)
253        << TheCall->getArg(ArgNo)->getSourceRange();
254  }
255
256  // For NEON intrinsics which take an immediate value as part of the
257  // instruction, range check them here.
258  unsigned i = 0, l = 0, u = 0;
259  switch (BuiltinID) {
260  default: return false;
261  case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break;
262  case ARM::BI__builtin_arm_usat: i = 1; u = 31; break;
263  case ARM::BI__builtin_arm_vcvtr_f:
264  case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break;
265#define GET_NEON_IMMEDIATE_CHECK
266#include "clang/Basic/arm_neon.inc"
267#undef GET_NEON_IMMEDIATE_CHECK
268  };
269
270  // Check that the immediate argument is actually a constant.
271  if (SemaBuiltinConstantArg(TheCall, i, Result))
272    return true;
273
274  // Range check against the upper/lower values for this isntruction.
275  unsigned Val = Result.getZExtValue();
276  if (Val < l || Val > (u + l))
277    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
278      << l << u+l << TheCall->getArg(i)->getSourceRange();
279
280  // FIXME: VFP Intrinsics should error if VFP not present.
281  return false;
282}
283
284/// CheckFunctionCall - Check a direct function call for various correctness
285/// and safety properties not strictly enforced by the C type system.
286bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) {
287  // Get the IdentifierInfo* for the called function.
288  IdentifierInfo *FnInfo = FDecl->getIdentifier();
289
290  // None of the checks below are needed for functions that don't have
291  // simple names (e.g., C++ conversion functions).
292  if (!FnInfo)
293    return false;
294
295  // FIXME: This mechanism should be abstracted to be less fragile and
296  // more efficient. For example, just map function ids to custom
297  // handlers.
298
299  // Printf and scanf checking.
300  for (specific_attr_iterator<FormatAttr>
301         i = FDecl->specific_attr_begin<FormatAttr>(),
302         e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) {
303
304    const FormatAttr *Format = *i;
305    const bool b = Format->getType() == "scanf";
306    if (b || CheckablePrintfAttr(Format, TheCall)) {
307      bool HasVAListArg = Format->getFirstArg() == 0;
308      CheckPrintfScanfArguments(TheCall, HasVAListArg,
309                                Format->getFormatIdx() - 1,
310                                HasVAListArg ? 0 : Format->getFirstArg() - 1,
311                                !b);
312    }
313  }
314
315  for (specific_attr_iterator<NonNullAttr>
316         i = FDecl->specific_attr_begin<NonNullAttr>(),
317         e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) {
318    CheckNonNullArguments(*i, TheCall->getArgs(),
319                          TheCall->getCallee()->getLocStart());
320  }
321
322  // Memset/memcpy/memmove/memcmp handling
323  int CMF = -1;
324  switch (FDecl->getBuiltinID()) {
325  case Builtin::BI__builtin_memset:
326  case Builtin::BI__builtin___memset_chk:
327  case Builtin::BImemset:
328    CMF = CMF_Memset;
329    break;
330
331  case Builtin::BI__builtin_memcpy:
332  case Builtin::BI__builtin___memcpy_chk:
333  case Builtin::BImemcpy:
334    CMF = CMF_Memcpy;
335    break;
336
337  case Builtin::BI__builtin_memmove:
338  case Builtin::BI__builtin___memmove_chk:
339  case Builtin::BImemmove:
340    CMF = CMF_Memmove;
341    break;
342
343  case Builtin::BI__builtin_memcmp:
344    CMF = CMF_Memcmp;
345    break;
346
347  default:
348    if (FDecl->getLinkage() == ExternalLinkage &&
349        (!getLangOptions().CPlusPlus || FDecl->isExternC())) {
350      if (FnInfo->isStr("memset"))
351        CMF = CMF_Memset;
352      else if (FnInfo->isStr("memcpy"))
353        CMF = CMF_Memcpy;
354      else if (FnInfo->isStr("memmove"))
355        CMF = CMF_Memmove;
356      else if (FnInfo->isStr("memcmp"))
357        CMF = CMF_Memcmp;
358    }
359    break;
360  }
361
362  if (CMF != -1)
363    CheckMemaccessArguments(TheCall, CheckedMemoryFunction(CMF), FnInfo);
364
365  return false;
366}
367
368bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
369  // Printf checking.
370  const FormatAttr *Format = NDecl->getAttr<FormatAttr>();
371  if (!Format)
372    return false;
373
374  const VarDecl *V = dyn_cast<VarDecl>(NDecl);
375  if (!V)
376    return false;
377
378  QualType Ty = V->getType();
379  if (!Ty->isBlockPointerType())
380    return false;
381
382  const bool b = Format->getType() == "scanf";
383  if (!b && !CheckablePrintfAttr(Format, TheCall))
384    return false;
385
386  bool HasVAListArg = Format->getFirstArg() == 0;
387  CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
388                            HasVAListArg ? 0 : Format->getFirstArg() - 1, !b);
389
390  return false;
391}
392
393/// SemaBuiltinAtomicOverloaded - We have a call to a function like
394/// __sync_fetch_and_add, which is an overloaded function based on the pointer
395/// type of its first argument.  The main ActOnCallExpr routines have already
396/// promoted the types of arguments because all of these calls are prototyped as
397/// void(...).
398///
399/// This function goes through and does final semantic checking for these
400/// builtins,
401ExprResult
402Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
403  CallExpr *TheCall = (CallExpr *)TheCallResult.get();
404  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
405  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
406
407  // Ensure that we have at least one argument to do type inference from.
408  if (TheCall->getNumArgs() < 1) {
409    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
410      << 0 << 1 << TheCall->getNumArgs()
411      << TheCall->getCallee()->getSourceRange();
412    return ExprError();
413  }
414
415  // Inspect the first argument of the atomic builtin.  This should always be
416  // a pointer type, whose element is an integral scalar or pointer type.
417  // Because it is a pointer type, we don't have to worry about any implicit
418  // casts here.
419  // FIXME: We don't allow floating point scalars as input.
420  Expr *FirstArg = TheCall->getArg(0);
421  const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>();
422  if (!pointerType) {
423    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
424      << FirstArg->getType() << FirstArg->getSourceRange();
425    return ExprError();
426  }
427
428  QualType ValType = pointerType->getPointeeType();
429  if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
430      !ValType->isBlockPointerType()) {
431    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
432      << FirstArg->getType() << FirstArg->getSourceRange();
433    return ExprError();
434  }
435
436  switch (ValType.getObjCLifetime()) {
437  case Qualifiers::OCL_None:
438  case Qualifiers::OCL_ExplicitNone:
439    // okay
440    break;
441
442  case Qualifiers::OCL_Weak:
443  case Qualifiers::OCL_Strong:
444  case Qualifiers::OCL_Autoreleasing:
445    Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
446      << ValType << FirstArg->getSourceRange();
447    return ExprError();
448  }
449
450  // The majority of builtins return a value, but a few have special return
451  // types, so allow them to override appropriately below.
452  QualType ResultType = ValType;
453
454  // We need to figure out which concrete builtin this maps onto.  For example,
455  // __sync_fetch_and_add with a 2 byte object turns into
456  // __sync_fetch_and_add_2.
457#define BUILTIN_ROW(x) \
458  { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
459    Builtin::BI##x##_8, Builtin::BI##x##_16 }
460
461  static const unsigned BuiltinIndices[][5] = {
462    BUILTIN_ROW(__sync_fetch_and_add),
463    BUILTIN_ROW(__sync_fetch_and_sub),
464    BUILTIN_ROW(__sync_fetch_and_or),
465    BUILTIN_ROW(__sync_fetch_and_and),
466    BUILTIN_ROW(__sync_fetch_and_xor),
467
468    BUILTIN_ROW(__sync_add_and_fetch),
469    BUILTIN_ROW(__sync_sub_and_fetch),
470    BUILTIN_ROW(__sync_and_and_fetch),
471    BUILTIN_ROW(__sync_or_and_fetch),
472    BUILTIN_ROW(__sync_xor_and_fetch),
473
474    BUILTIN_ROW(__sync_val_compare_and_swap),
475    BUILTIN_ROW(__sync_bool_compare_and_swap),
476    BUILTIN_ROW(__sync_lock_test_and_set),
477    BUILTIN_ROW(__sync_lock_release),
478    BUILTIN_ROW(__sync_swap)
479  };
480#undef BUILTIN_ROW
481
482  // Determine the index of the size.
483  unsigned SizeIndex;
484  switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
485  case 1: SizeIndex = 0; break;
486  case 2: SizeIndex = 1; break;
487  case 4: SizeIndex = 2; break;
488  case 8: SizeIndex = 3; break;
489  case 16: SizeIndex = 4; break;
490  default:
491    Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
492      << FirstArg->getType() << FirstArg->getSourceRange();
493    return ExprError();
494  }
495
496  // Each of these builtins has one pointer argument, followed by some number of
497  // values (0, 1 or 2) followed by a potentially empty varags list of stuff
498  // that we ignore.  Find out which row of BuiltinIndices to read from as well
499  // as the number of fixed args.
500  unsigned BuiltinID = FDecl->getBuiltinID();
501  unsigned BuiltinIndex, NumFixed = 1;
502  switch (BuiltinID) {
503  default: assert(0 && "Unknown overloaded atomic builtin!");
504  case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break;
505  case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break;
506  case Builtin::BI__sync_fetch_and_or:  BuiltinIndex = 2; break;
507  case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break;
508  case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break;
509
510  case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break;
511  case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break;
512  case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break;
513  case Builtin::BI__sync_or_and_fetch:  BuiltinIndex = 8; break;
514  case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break;
515
516  case Builtin::BI__sync_val_compare_and_swap:
517    BuiltinIndex = 10;
518    NumFixed = 2;
519    break;
520  case Builtin::BI__sync_bool_compare_and_swap:
521    BuiltinIndex = 11;
522    NumFixed = 2;
523    ResultType = Context.BoolTy;
524    break;
525  case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break;
526  case Builtin::BI__sync_lock_release:
527    BuiltinIndex = 13;
528    NumFixed = 0;
529    ResultType = Context.VoidTy;
530    break;
531  case Builtin::BI__sync_swap: BuiltinIndex = 14; break;
532  }
533
534  // Now that we know how many fixed arguments we expect, first check that we
535  // have at least that many.
536  if (TheCall->getNumArgs() < 1+NumFixed) {
537    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
538      << 0 << 1+NumFixed << TheCall->getNumArgs()
539      << TheCall->getCallee()->getSourceRange();
540    return ExprError();
541  }
542
543  // Get the decl for the concrete builtin from this, we can tell what the
544  // concrete integer type we should convert to is.
545  unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
546  const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
547  IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
548  FunctionDecl *NewBuiltinDecl =
549    cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
550                                           TUScope, false, DRE->getLocStart()));
551
552  // The first argument --- the pointer --- has a fixed type; we
553  // deduce the types of the rest of the arguments accordingly.  Walk
554  // the remaining arguments, converting them to the deduced value type.
555  for (unsigned i = 0; i != NumFixed; ++i) {
556    ExprResult Arg = TheCall->getArg(i+1);
557
558    // If the argument is an implicit cast, then there was a promotion due to
559    // "...", just remove it now.
560    if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) {
561      Arg = ICE->getSubExpr();
562      ICE->setSubExpr(0);
563      TheCall->setArg(i+1, Arg.get());
564    }
565
566    // GCC does an implicit conversion to the pointer or integer ValType.  This
567    // can fail in some cases (1i -> int**), check for this error case now.
568    CastKind Kind = CK_Invalid;
569    ExprValueKind VK = VK_RValue;
570    CXXCastPath BasePath;
571    Arg = CheckCastTypes(Arg.get()->getLocStart(), Arg.get()->getSourceRange(),
572                         ValType, Arg.take(), Kind, VK, BasePath);
573    if (Arg.isInvalid())
574      return ExprError();
575
576    // Okay, we have something that *can* be converted to the right type.  Check
577    // to see if there is a potentially weird extension going on here.  This can
578    // happen when you do an atomic operation on something like an char* and
579    // pass in 42.  The 42 gets converted to char.  This is even more strange
580    // for things like 45.123 -> char, etc.
581    // FIXME: Do this check.
582    Arg = ImpCastExprToType(Arg.take(), ValType, Kind, VK, &BasePath);
583    TheCall->setArg(i+1, Arg.get());
584  }
585
586  // Switch the DeclRefExpr to refer to the new decl.
587  DRE->setDecl(NewBuiltinDecl);
588  DRE->setType(NewBuiltinDecl->getType());
589
590  // Set the callee in the CallExpr.
591  // FIXME: This leaks the original parens and implicit casts.
592  ExprResult PromotedCall = UsualUnaryConversions(DRE);
593  if (PromotedCall.isInvalid())
594    return ExprError();
595  TheCall->setCallee(PromotedCall.take());
596
597  // Change the result type of the call to match the original value type. This
598  // is arbitrary, but the codegen for these builtins ins design to handle it
599  // gracefully.
600  TheCall->setType(ResultType);
601
602  return move(TheCallResult);
603}
604
605
606/// CheckObjCString - Checks that the argument to the builtin
607/// CFString constructor is correct
608/// Note: It might also make sense to do the UTF-16 conversion here (would
609/// simplify the backend).
610bool Sema::CheckObjCString(Expr *Arg) {
611  Arg = Arg->IgnoreParenCasts();
612  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
613
614  if (!Literal || !Literal->isAscii()) {
615    Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
616      << Arg->getSourceRange();
617    return true;
618  }
619
620  if (Literal->containsNonAsciiOrNull()) {
621    StringRef String = Literal->getString();
622    unsigned NumBytes = String.size();
623    SmallVector<UTF16, 128> ToBuf(NumBytes);
624    const UTF8 *FromPtr = (UTF8 *)String.data();
625    UTF16 *ToPtr = &ToBuf[0];
626
627    ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
628                                                 &ToPtr, ToPtr + NumBytes,
629                                                 strictConversion);
630    // Check for conversion failure.
631    if (Result != conversionOK)
632      Diag(Arg->getLocStart(),
633           diag::warn_cfstring_truncated) << Arg->getSourceRange();
634  }
635  return false;
636}
637
638/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
639/// Emit an error and return true on failure, return false on success.
640bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
641  Expr *Fn = TheCall->getCallee();
642  if (TheCall->getNumArgs() > 2) {
643    Diag(TheCall->getArg(2)->getLocStart(),
644         diag::err_typecheck_call_too_many_args)
645      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
646      << Fn->getSourceRange()
647      << SourceRange(TheCall->getArg(2)->getLocStart(),
648                     (*(TheCall->arg_end()-1))->getLocEnd());
649    return true;
650  }
651
652  if (TheCall->getNumArgs() < 2) {
653    return Diag(TheCall->getLocEnd(),
654      diag::err_typecheck_call_too_few_args_at_least)
655      << 0 /*function call*/ << 2 << TheCall->getNumArgs();
656  }
657
658  // Determine whether the current function is variadic or not.
659  BlockScopeInfo *CurBlock = getCurBlock();
660  bool isVariadic;
661  if (CurBlock)
662    isVariadic = CurBlock->TheDecl->isVariadic();
663  else if (FunctionDecl *FD = getCurFunctionDecl())
664    isVariadic = FD->isVariadic();
665  else
666    isVariadic = getCurMethodDecl()->isVariadic();
667
668  if (!isVariadic) {
669    Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
670    return true;
671  }
672
673  // Verify that the second argument to the builtin is the last argument of the
674  // current function or method.
675  bool SecondArgIsLastNamedArgument = false;
676  const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
677
678  if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
679    if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
680      // FIXME: This isn't correct for methods (results in bogus warning).
681      // Get the last formal in the current function.
682      const ParmVarDecl *LastArg;
683      if (CurBlock)
684        LastArg = *(CurBlock->TheDecl->param_end()-1);
685      else if (FunctionDecl *FD = getCurFunctionDecl())
686        LastArg = *(FD->param_end()-1);
687      else
688        LastArg = *(getCurMethodDecl()->param_end()-1);
689      SecondArgIsLastNamedArgument = PV == LastArg;
690    }
691  }
692
693  if (!SecondArgIsLastNamedArgument)
694    Diag(TheCall->getArg(1)->getLocStart(),
695         diag::warn_second_parameter_of_va_start_not_last_named_argument);
696  return false;
697}
698
699/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
700/// friends.  This is declared to take (...), so we have to check everything.
701bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
702  if (TheCall->getNumArgs() < 2)
703    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
704      << 0 << 2 << TheCall->getNumArgs()/*function call*/;
705  if (TheCall->getNumArgs() > 2)
706    return Diag(TheCall->getArg(2)->getLocStart(),
707                diag::err_typecheck_call_too_many_args)
708      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
709      << SourceRange(TheCall->getArg(2)->getLocStart(),
710                     (*(TheCall->arg_end()-1))->getLocEnd());
711
712  ExprResult OrigArg0 = TheCall->getArg(0);
713  ExprResult OrigArg1 = TheCall->getArg(1);
714
715  // Do standard promotions between the two arguments, returning their common
716  // type.
717  QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
718  if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
719    return true;
720
721  // Make sure any conversions are pushed back into the call; this is
722  // type safe since unordered compare builtins are declared as "_Bool
723  // foo(...)".
724  TheCall->setArg(0, OrigArg0.get());
725  TheCall->setArg(1, OrigArg1.get());
726
727  if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
728    return false;
729
730  // If the common type isn't a real floating type, then the arguments were
731  // invalid for this operation.
732  if (!Res->isRealFloatingType())
733    return Diag(OrigArg0.get()->getLocStart(),
734                diag::err_typecheck_call_invalid_ordered_compare)
735      << OrigArg0.get()->getType() << OrigArg1.get()->getType()
736      << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
737
738  return false;
739}
740
741/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
742/// __builtin_isnan and friends.  This is declared to take (...), so we have
743/// to check everything. We expect the last argument to be a floating point
744/// value.
745bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
746  if (TheCall->getNumArgs() < NumArgs)
747    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
748      << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
749  if (TheCall->getNumArgs() > NumArgs)
750    return Diag(TheCall->getArg(NumArgs)->getLocStart(),
751                diag::err_typecheck_call_too_many_args)
752      << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
753      << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
754                     (*(TheCall->arg_end()-1))->getLocEnd());
755
756  Expr *OrigArg = TheCall->getArg(NumArgs-1);
757
758  if (OrigArg->isTypeDependent())
759    return false;
760
761  // This operation requires a non-_Complex floating-point number.
762  if (!OrigArg->getType()->isRealFloatingType())
763    return Diag(OrigArg->getLocStart(),
764                diag::err_typecheck_call_invalid_unary_fp)
765      << OrigArg->getType() << OrigArg->getSourceRange();
766
767  // If this is an implicit conversion from float -> double, remove it.
768  if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
769    Expr *CastArg = Cast->getSubExpr();
770    if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
771      assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
772             "promotion from float to double is the only expected cast here");
773      Cast->setSubExpr(0);
774      TheCall->setArg(NumArgs-1, CastArg);
775      OrigArg = CastArg;
776    }
777  }
778
779  return false;
780}
781
782/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
783// This is declared to take (...), so we have to check everything.
784ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
785  if (TheCall->getNumArgs() < 2)
786    return ExprError(Diag(TheCall->getLocEnd(),
787                          diag::err_typecheck_call_too_few_args_at_least)
788      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
789      << TheCall->getSourceRange());
790
791  // Determine which of the following types of shufflevector we're checking:
792  // 1) unary, vector mask: (lhs, mask)
793  // 2) binary, vector mask: (lhs, rhs, mask)
794  // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
795  QualType resType = TheCall->getArg(0)->getType();
796  unsigned numElements = 0;
797
798  if (!TheCall->getArg(0)->isTypeDependent() &&
799      !TheCall->getArg(1)->isTypeDependent()) {
800    QualType LHSType = TheCall->getArg(0)->getType();
801    QualType RHSType = TheCall->getArg(1)->getType();
802
803    if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
804      Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
805        << SourceRange(TheCall->getArg(0)->getLocStart(),
806                       TheCall->getArg(1)->getLocEnd());
807      return ExprError();
808    }
809
810    numElements = LHSType->getAs<VectorType>()->getNumElements();
811    unsigned numResElements = TheCall->getNumArgs() - 2;
812
813    // Check to see if we have a call with 2 vector arguments, the unary shuffle
814    // with mask.  If so, verify that RHS is an integer vector type with the
815    // same number of elts as lhs.
816    if (TheCall->getNumArgs() == 2) {
817      if (!RHSType->hasIntegerRepresentation() ||
818          RHSType->getAs<VectorType>()->getNumElements() != numElements)
819        Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
820          << SourceRange(TheCall->getArg(1)->getLocStart(),
821                         TheCall->getArg(1)->getLocEnd());
822      numResElements = numElements;
823    }
824    else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
825      Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
826        << SourceRange(TheCall->getArg(0)->getLocStart(),
827                       TheCall->getArg(1)->getLocEnd());
828      return ExprError();
829    } else if (numElements != numResElements) {
830      QualType eltType = LHSType->getAs<VectorType>()->getElementType();
831      resType = Context.getVectorType(eltType, numResElements,
832                                      VectorType::GenericVector);
833    }
834  }
835
836  for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
837    if (TheCall->getArg(i)->isTypeDependent() ||
838        TheCall->getArg(i)->isValueDependent())
839      continue;
840
841    llvm::APSInt Result(32);
842    if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
843      return ExprError(Diag(TheCall->getLocStart(),
844                  diag::err_shufflevector_nonconstant_argument)
845                << TheCall->getArg(i)->getSourceRange());
846
847    if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
848      return ExprError(Diag(TheCall->getLocStart(),
849                  diag::err_shufflevector_argument_too_large)
850               << TheCall->getArg(i)->getSourceRange());
851  }
852
853  SmallVector<Expr*, 32> exprs;
854
855  for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
856    exprs.push_back(TheCall->getArg(i));
857    TheCall->setArg(i, 0);
858  }
859
860  return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
861                                            exprs.size(), resType,
862                                            TheCall->getCallee()->getLocStart(),
863                                            TheCall->getRParenLoc()));
864}
865
866/// SemaBuiltinPrefetch - Handle __builtin_prefetch.
867// This is declared to take (const void*, ...) and can take two
868// optional constant int args.
869bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
870  unsigned NumArgs = TheCall->getNumArgs();
871
872  if (NumArgs > 3)
873    return Diag(TheCall->getLocEnd(),
874             diag::err_typecheck_call_too_many_args_at_most)
875             << 0 /*function call*/ << 3 << NumArgs
876             << TheCall->getSourceRange();
877
878  // Argument 0 is checked for us and the remaining arguments must be
879  // constant integers.
880  for (unsigned i = 1; i != NumArgs; ++i) {
881    Expr *Arg = TheCall->getArg(i);
882
883    llvm::APSInt Result;
884    if (SemaBuiltinConstantArg(TheCall, i, Result))
885      return true;
886
887    // FIXME: gcc issues a warning and rewrites these to 0. These
888    // seems especially odd for the third argument since the default
889    // is 3.
890    if (i == 1) {
891      if (Result.getLimitedValue() > 1)
892        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
893             << "0" << "1" << Arg->getSourceRange();
894    } else {
895      if (Result.getLimitedValue() > 3)
896        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
897            << "0" << "3" << Arg->getSourceRange();
898    }
899  }
900
901  return false;
902}
903
904/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
905/// TheCall is a constant expression.
906bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
907                                  llvm::APSInt &Result) {
908  Expr *Arg = TheCall->getArg(ArgNum);
909  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
910  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
911
912  if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
913
914  if (!Arg->isIntegerConstantExpr(Result, Context))
915    return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
916                << FDecl->getDeclName() <<  Arg->getSourceRange();
917
918  return false;
919}
920
921/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
922/// int type). This simply type checks that type is one of the defined
923/// constants (0-3).
924// For compatibility check 0-3, llvm only handles 0 and 2.
925bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
926  llvm::APSInt Result;
927
928  // Check constant-ness first.
929  if (SemaBuiltinConstantArg(TheCall, 1, Result))
930    return true;
931
932  Expr *Arg = TheCall->getArg(1);
933  if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
934    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
935             << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
936  }
937
938  return false;
939}
940
941/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
942/// This checks that val is a constant 1.
943bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
944  Expr *Arg = TheCall->getArg(1);
945  llvm::APSInt Result;
946
947  // TODO: This is less than ideal. Overload this to take a value.
948  if (SemaBuiltinConstantArg(TheCall, 1, Result))
949    return true;
950
951  if (Result != 1)
952    return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
953             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
954
955  return false;
956}
957
958// Handle i > 1 ? "x" : "y", recursively.
959bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall,
960                                  bool HasVAListArg,
961                                  unsigned format_idx, unsigned firstDataArg,
962                                  bool isPrintf) {
963 tryAgain:
964  if (E->isTypeDependent() || E->isValueDependent())
965    return false;
966
967  E = E->IgnoreParens();
968
969  switch (E->getStmtClass()) {
970  case Stmt::BinaryConditionalOperatorClass:
971  case Stmt::ConditionalOperatorClass: {
972    const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E);
973    return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg,
974                                  format_idx, firstDataArg, isPrintf)
975        && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg,
976                                  format_idx, firstDataArg, isPrintf);
977  }
978
979  case Stmt::IntegerLiteralClass:
980    // Technically -Wformat-nonliteral does not warn about this case.
981    // The behavior of printf and friends in this case is implementation
982    // dependent.  Ideally if the format string cannot be null then
983    // it should have a 'nonnull' attribute in the function prototype.
984    return true;
985
986  case Stmt::ImplicitCastExprClass: {
987    E = cast<ImplicitCastExpr>(E)->getSubExpr();
988    goto tryAgain;
989  }
990
991  case Stmt::OpaqueValueExprClass:
992    if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
993      E = src;
994      goto tryAgain;
995    }
996    return false;
997
998  case Stmt::PredefinedExprClass:
999    // While __func__, etc., are technically not string literals, they
1000    // cannot contain format specifiers and thus are not a security
1001    // liability.
1002    return true;
1003
1004  case Stmt::DeclRefExprClass: {
1005    const DeclRefExpr *DR = cast<DeclRefExpr>(E);
1006
1007    // As an exception, do not flag errors for variables binding to
1008    // const string literals.
1009    if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
1010      bool isConstant = false;
1011      QualType T = DR->getType();
1012
1013      if (const ArrayType *AT = Context.getAsArrayType(T)) {
1014        isConstant = AT->getElementType().isConstant(Context);
1015      } else if (const PointerType *PT = T->getAs<PointerType>()) {
1016        isConstant = T.isConstant(Context) &&
1017                     PT->getPointeeType().isConstant(Context);
1018      }
1019
1020      if (isConstant) {
1021        if (const Expr *Init = VD->getAnyInitializer())
1022          return SemaCheckStringLiteral(Init, TheCall,
1023                                        HasVAListArg, format_idx, firstDataArg,
1024                                        isPrintf);
1025      }
1026
1027      // For vprintf* functions (i.e., HasVAListArg==true), we add a
1028      // special check to see if the format string is a function parameter
1029      // of the function calling the printf function.  If the function
1030      // has an attribute indicating it is a printf-like function, then we
1031      // should suppress warnings concerning non-literals being used in a call
1032      // to a vprintf function.  For example:
1033      //
1034      // void
1035      // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
1036      //      va_list ap;
1037      //      va_start(ap, fmt);
1038      //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
1039      //      ...
1040      //
1041      //
1042      //  FIXME: We don't have full attribute support yet, so just check to see
1043      //    if the argument is a DeclRefExpr that references a parameter.  We'll
1044      //    add proper support for checking the attribute later.
1045      if (HasVAListArg)
1046        if (isa<ParmVarDecl>(VD))
1047          return true;
1048    }
1049
1050    return false;
1051  }
1052
1053  case Stmt::CallExprClass: {
1054    const CallExpr *CE = cast<CallExpr>(E);
1055    if (const ImplicitCastExpr *ICE
1056          = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
1057      if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
1058        if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
1059          if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
1060            unsigned ArgIndex = FA->getFormatIdx();
1061            const Expr *Arg = CE->getArg(ArgIndex - 1);
1062
1063            return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg,
1064                                          format_idx, firstDataArg, isPrintf);
1065          }
1066        }
1067      }
1068    }
1069
1070    return false;
1071  }
1072  case Stmt::ObjCStringLiteralClass:
1073  case Stmt::StringLiteralClass: {
1074    const StringLiteral *StrE = NULL;
1075
1076    if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1077      StrE = ObjCFExpr->getString();
1078    else
1079      StrE = cast<StringLiteral>(E);
1080
1081    if (StrE) {
1082      CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx,
1083                        firstDataArg, isPrintf);
1084      return true;
1085    }
1086
1087    return false;
1088  }
1089
1090  default:
1091    return false;
1092  }
1093}
1094
1095void
1096Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1097                            const Expr * const *ExprArgs,
1098                            SourceLocation CallSiteLoc) {
1099  for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1100                                  e = NonNull->args_end();
1101       i != e; ++i) {
1102    const Expr *ArgExpr = ExprArgs[*i];
1103    if (ArgExpr->isNullPointerConstant(Context,
1104                                       Expr::NPC_ValueDependentIsNotNull))
1105      Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1106  }
1107}
1108
1109/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar
1110/// functions) for correct use of format strings.
1111void
1112Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg,
1113                                unsigned format_idx, unsigned firstDataArg,
1114                                bool isPrintf) {
1115
1116  const Expr *Fn = TheCall->getCallee();
1117
1118  // The way the format attribute works in GCC, the implicit this argument
1119  // of member functions is counted. However, it doesn't appear in our own
1120  // lists, so decrement format_idx in that case.
1121  if (isa<CXXMemberCallExpr>(TheCall)) {
1122    const CXXMethodDecl *method_decl =
1123      dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl());
1124    if (method_decl && method_decl->isInstance()) {
1125      // Catch a format attribute mistakenly referring to the object argument.
1126      if (format_idx == 0)
1127        return;
1128      --format_idx;
1129      if(firstDataArg != 0)
1130        --firstDataArg;
1131    }
1132  }
1133
1134  // CHECK: printf/scanf-like function is called with no format string.
1135  if (format_idx >= TheCall->getNumArgs()) {
1136    Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string)
1137      << Fn->getSourceRange();
1138    return;
1139  }
1140
1141  const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts();
1142
1143  // CHECK: format string is not a string literal.
1144  //
1145  // Dynamically generated format strings are difficult to
1146  // automatically vet at compile time.  Requiring that format strings
1147  // are string literals: (1) permits the checking of format strings by
1148  // the compiler and thereby (2) can practically remove the source of
1149  // many format string exploits.
1150
1151  // Format string can be either ObjC string (e.g. @"%d") or
1152  // C string (e.g. "%d")
1153  // ObjC string uses the same format specifiers as C string, so we can use
1154  // the same format string checking logic for both ObjC and C strings.
1155  if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx,
1156                             firstDataArg, isPrintf))
1157    return;  // Literal format string found, check done!
1158
1159  // If there are no arguments specified, warn with -Wformat-security, otherwise
1160  // warn only with -Wformat-nonliteral.
1161  if (TheCall->getNumArgs() == format_idx+1)
1162    Diag(TheCall->getArg(format_idx)->getLocStart(),
1163         diag::warn_format_nonliteral_noargs)
1164      << OrigFormatExpr->getSourceRange();
1165  else
1166    Diag(TheCall->getArg(format_idx)->getLocStart(),
1167         diag::warn_format_nonliteral)
1168           << OrigFormatExpr->getSourceRange();
1169}
1170
1171namespace {
1172class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1173protected:
1174  Sema &S;
1175  const StringLiteral *FExpr;
1176  const Expr *OrigFormatExpr;
1177  const unsigned FirstDataArg;
1178  const unsigned NumDataArgs;
1179  const bool IsObjCLiteral;
1180  const char *Beg; // Start of format string.
1181  const bool HasVAListArg;
1182  const CallExpr *TheCall;
1183  unsigned FormatIdx;
1184  llvm::BitVector CoveredArgs;
1185  bool usesPositionalArgs;
1186  bool atFirstArg;
1187public:
1188  CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1189                     const Expr *origFormatExpr, unsigned firstDataArg,
1190                     unsigned numDataArgs, bool isObjCLiteral,
1191                     const char *beg, bool hasVAListArg,
1192                     const CallExpr *theCall, unsigned formatIdx)
1193    : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1194      FirstDataArg(firstDataArg),
1195      NumDataArgs(numDataArgs),
1196      IsObjCLiteral(isObjCLiteral), Beg(beg),
1197      HasVAListArg(hasVAListArg),
1198      TheCall(theCall), FormatIdx(formatIdx),
1199      usesPositionalArgs(false), atFirstArg(true) {
1200        CoveredArgs.resize(numDataArgs);
1201        CoveredArgs.reset();
1202      }
1203
1204  void DoneProcessing();
1205
1206  void HandleIncompleteSpecifier(const char *startSpecifier,
1207                                 unsigned specifierLen);
1208
1209  virtual void HandleInvalidPosition(const char *startSpecifier,
1210                                     unsigned specifierLen,
1211                                     analyze_format_string::PositionContext p);
1212
1213  virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
1214
1215  void HandleNullChar(const char *nullCharacter);
1216
1217protected:
1218  bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
1219                                        const char *startSpec,
1220                                        unsigned specifierLen,
1221                                        const char *csStart, unsigned csLen);
1222
1223  SourceRange getFormatStringRange();
1224  CharSourceRange getSpecifierRange(const char *startSpecifier,
1225                                    unsigned specifierLen);
1226  SourceLocation getLocationOfByte(const char *x);
1227
1228  const Expr *getDataArg(unsigned i) const;
1229
1230  bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
1231                    const analyze_format_string::ConversionSpecifier &CS,
1232                    const char *startSpecifier, unsigned specifierLen,
1233                    unsigned argIndex);
1234};
1235}
1236
1237SourceRange CheckFormatHandler::getFormatStringRange() {
1238  return OrigFormatExpr->getSourceRange();
1239}
1240
1241CharSourceRange CheckFormatHandler::
1242getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
1243  SourceLocation Start = getLocationOfByte(startSpecifier);
1244  SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
1245
1246  // Advance the end SourceLocation by one due to half-open ranges.
1247  End = End.getFileLocWithOffset(1);
1248
1249  return CharSourceRange::getCharRange(Start, End);
1250}
1251
1252SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
1253  return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
1254}
1255
1256void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
1257                                                   unsigned specifierLen){
1258  SourceLocation Loc = getLocationOfByte(startSpecifier);
1259  S.Diag(Loc, diag::warn_printf_incomplete_specifier)
1260    << getSpecifierRange(startSpecifier, specifierLen);
1261}
1262
1263void
1264CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
1265                                     analyze_format_string::PositionContext p) {
1266  SourceLocation Loc = getLocationOfByte(startPos);
1267  S.Diag(Loc, diag::warn_format_invalid_positional_specifier)
1268    << (unsigned) p << getSpecifierRange(startPos, posLen);
1269}
1270
1271void CheckFormatHandler::HandleZeroPosition(const char *startPos,
1272                                            unsigned posLen) {
1273  SourceLocation Loc = getLocationOfByte(startPos);
1274  S.Diag(Loc, diag::warn_format_zero_positional_specifier)
1275    << getSpecifierRange(startPos, posLen);
1276}
1277
1278void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
1279  if (!IsObjCLiteral) {
1280    // The presence of a null character is likely an error.
1281    S.Diag(getLocationOfByte(nullCharacter),
1282           diag::warn_printf_format_string_contains_null_char)
1283      << getFormatStringRange();
1284  }
1285}
1286
1287const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
1288  return TheCall->getArg(FirstDataArg + i);
1289}
1290
1291void CheckFormatHandler::DoneProcessing() {
1292    // Does the number of data arguments exceed the number of
1293    // format conversions in the format string?
1294  if (!HasVAListArg) {
1295      // Find any arguments that weren't covered.
1296    CoveredArgs.flip();
1297    signed notCoveredArg = CoveredArgs.find_first();
1298    if (notCoveredArg >= 0) {
1299      assert((unsigned)notCoveredArg < NumDataArgs);
1300      S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(),
1301             diag::warn_printf_data_arg_not_used)
1302      << getFormatStringRange();
1303    }
1304  }
1305}
1306
1307bool
1308CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
1309                                                     SourceLocation Loc,
1310                                                     const char *startSpec,
1311                                                     unsigned specifierLen,
1312                                                     const char *csStart,
1313                                                     unsigned csLen) {
1314
1315  bool keepGoing = true;
1316  if (argIndex < NumDataArgs) {
1317    // Consider the argument coverered, even though the specifier doesn't
1318    // make sense.
1319    CoveredArgs.set(argIndex);
1320  }
1321  else {
1322    // If argIndex exceeds the number of data arguments we
1323    // don't issue a warning because that is just a cascade of warnings (and
1324    // they may have intended '%%' anyway). We don't want to continue processing
1325    // the format string after this point, however, as we will like just get
1326    // gibberish when trying to match arguments.
1327    keepGoing = false;
1328  }
1329
1330  S.Diag(Loc, diag::warn_format_invalid_conversion)
1331    << StringRef(csStart, csLen)
1332    << getSpecifierRange(startSpec, specifierLen);
1333
1334  return keepGoing;
1335}
1336
1337bool
1338CheckFormatHandler::CheckNumArgs(
1339  const analyze_format_string::FormatSpecifier &FS,
1340  const analyze_format_string::ConversionSpecifier &CS,
1341  const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
1342
1343  if (argIndex >= NumDataArgs) {
1344    if (FS.usesPositionalArg())  {
1345      S.Diag(getLocationOfByte(CS.getStart()),
1346             diag::warn_printf_positional_arg_exceeds_data_args)
1347      << (argIndex+1) << NumDataArgs
1348      << getSpecifierRange(startSpecifier, specifierLen);
1349    }
1350    else {
1351      S.Diag(getLocationOfByte(CS.getStart()),
1352             diag::warn_printf_insufficient_data_args)
1353      << getSpecifierRange(startSpecifier, specifierLen);
1354    }
1355
1356    return false;
1357  }
1358  return true;
1359}
1360
1361//===--- CHECK: Printf format string checking ------------------------------===//
1362
1363namespace {
1364class CheckPrintfHandler : public CheckFormatHandler {
1365public:
1366  CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
1367                     const Expr *origFormatExpr, unsigned firstDataArg,
1368                     unsigned numDataArgs, bool isObjCLiteral,
1369                     const char *beg, bool hasVAListArg,
1370                     const CallExpr *theCall, unsigned formatIdx)
1371  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1372                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1373                       theCall, formatIdx) {}
1374
1375
1376  bool HandleInvalidPrintfConversionSpecifier(
1377                                      const analyze_printf::PrintfSpecifier &FS,
1378                                      const char *startSpecifier,
1379                                      unsigned specifierLen);
1380
1381  bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
1382                             const char *startSpecifier,
1383                             unsigned specifierLen);
1384
1385  bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
1386                    const char *startSpecifier, unsigned specifierLen);
1387  void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
1388                           const analyze_printf::OptionalAmount &Amt,
1389                           unsigned type,
1390                           const char *startSpecifier, unsigned specifierLen);
1391  void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1392                  const analyze_printf::OptionalFlag &flag,
1393                  const char *startSpecifier, unsigned specifierLen);
1394  void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
1395                         const analyze_printf::OptionalFlag &ignoredFlag,
1396                         const analyze_printf::OptionalFlag &flag,
1397                         const char *startSpecifier, unsigned specifierLen);
1398};
1399}
1400
1401bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
1402                                      const analyze_printf::PrintfSpecifier &FS,
1403                                      const char *startSpecifier,
1404                                      unsigned specifierLen) {
1405  const analyze_printf::PrintfConversionSpecifier &CS =
1406    FS.getConversionSpecifier();
1407
1408  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1409                                          getLocationOfByte(CS.getStart()),
1410                                          startSpecifier, specifierLen,
1411                                          CS.getStart(), CS.getLength());
1412}
1413
1414bool CheckPrintfHandler::HandleAmount(
1415                               const analyze_format_string::OptionalAmount &Amt,
1416                               unsigned k, const char *startSpecifier,
1417                               unsigned specifierLen) {
1418
1419  if (Amt.hasDataArgument()) {
1420    if (!HasVAListArg) {
1421      unsigned argIndex = Amt.getArgIndex();
1422      if (argIndex >= NumDataArgs) {
1423        S.Diag(getLocationOfByte(Amt.getStart()),
1424               diag::warn_printf_asterisk_missing_arg)
1425          << k << getSpecifierRange(startSpecifier, specifierLen);
1426        // Don't do any more checking.  We will just emit
1427        // spurious errors.
1428        return false;
1429      }
1430
1431      // Type check the data argument.  It should be an 'int'.
1432      // Although not in conformance with C99, we also allow the argument to be
1433      // an 'unsigned int' as that is a reasonably safe case.  GCC also
1434      // doesn't emit a warning for that case.
1435      CoveredArgs.set(argIndex);
1436      const Expr *Arg = getDataArg(argIndex);
1437      QualType T = Arg->getType();
1438
1439      const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context);
1440      assert(ATR.isValid());
1441
1442      if (!ATR.matchesType(S.Context, T)) {
1443        S.Diag(getLocationOfByte(Amt.getStart()),
1444               diag::warn_printf_asterisk_wrong_type)
1445          << k
1446          << ATR.getRepresentativeType(S.Context) << T
1447          << getSpecifierRange(startSpecifier, specifierLen)
1448          << Arg->getSourceRange();
1449        // Don't do any more checking.  We will just emit
1450        // spurious errors.
1451        return false;
1452      }
1453    }
1454  }
1455  return true;
1456}
1457
1458void CheckPrintfHandler::HandleInvalidAmount(
1459                                      const analyze_printf::PrintfSpecifier &FS,
1460                                      const analyze_printf::OptionalAmount &Amt,
1461                                      unsigned type,
1462                                      const char *startSpecifier,
1463                                      unsigned specifierLen) {
1464  const analyze_printf::PrintfConversionSpecifier &CS =
1465    FS.getConversionSpecifier();
1466  switch (Amt.getHowSpecified()) {
1467  case analyze_printf::OptionalAmount::Constant:
1468    S.Diag(getLocationOfByte(Amt.getStart()),
1469        diag::warn_printf_nonsensical_optional_amount)
1470      << type
1471      << CS.toString()
1472      << getSpecifierRange(startSpecifier, specifierLen)
1473      << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
1474          Amt.getConstantLength()));
1475    break;
1476
1477  default:
1478    S.Diag(getLocationOfByte(Amt.getStart()),
1479        diag::warn_printf_nonsensical_optional_amount)
1480      << type
1481      << CS.toString()
1482      << getSpecifierRange(startSpecifier, specifierLen);
1483    break;
1484  }
1485}
1486
1487void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1488                                    const analyze_printf::OptionalFlag &flag,
1489                                    const char *startSpecifier,
1490                                    unsigned specifierLen) {
1491  // Warn about pointless flag with a fixit removal.
1492  const analyze_printf::PrintfConversionSpecifier &CS =
1493    FS.getConversionSpecifier();
1494  S.Diag(getLocationOfByte(flag.getPosition()),
1495      diag::warn_printf_nonsensical_flag)
1496    << flag.toString() << CS.toString()
1497    << getSpecifierRange(startSpecifier, specifierLen)
1498    << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1));
1499}
1500
1501void CheckPrintfHandler::HandleIgnoredFlag(
1502                                const analyze_printf::PrintfSpecifier &FS,
1503                                const analyze_printf::OptionalFlag &ignoredFlag,
1504                                const analyze_printf::OptionalFlag &flag,
1505                                const char *startSpecifier,
1506                                unsigned specifierLen) {
1507  // Warn about ignored flag with a fixit removal.
1508  S.Diag(getLocationOfByte(ignoredFlag.getPosition()),
1509      diag::warn_printf_ignored_flag)
1510    << ignoredFlag.toString() << flag.toString()
1511    << getSpecifierRange(startSpecifier, specifierLen)
1512    << FixItHint::CreateRemoval(getSpecifierRange(
1513        ignoredFlag.getPosition(), 1));
1514}
1515
1516bool
1517CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
1518                                            &FS,
1519                                          const char *startSpecifier,
1520                                          unsigned specifierLen) {
1521
1522  using namespace analyze_format_string;
1523  using namespace analyze_printf;
1524  const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
1525
1526  if (FS.consumesDataArgument()) {
1527    if (atFirstArg) {
1528        atFirstArg = false;
1529        usesPositionalArgs = FS.usesPositionalArg();
1530    }
1531    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1532      // Cannot mix-and-match positional and non-positional arguments.
1533      S.Diag(getLocationOfByte(CS.getStart()),
1534             diag::warn_format_mix_positional_nonpositional_args)
1535        << getSpecifierRange(startSpecifier, specifierLen);
1536      return false;
1537    }
1538  }
1539
1540  // First check if the field width, precision, and conversion specifier
1541  // have matching data arguments.
1542  if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
1543                    startSpecifier, specifierLen)) {
1544    return false;
1545  }
1546
1547  if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
1548                    startSpecifier, specifierLen)) {
1549    return false;
1550  }
1551
1552  if (!CS.consumesDataArgument()) {
1553    // FIXME: Technically specifying a precision or field width here
1554    // makes no sense.  Worth issuing a warning at some point.
1555    return true;
1556  }
1557
1558  // Consume the argument.
1559  unsigned argIndex = FS.getArgIndex();
1560  if (argIndex < NumDataArgs) {
1561    // The check to see if the argIndex is valid will come later.
1562    // We set the bit here because we may exit early from this
1563    // function if we encounter some other error.
1564    CoveredArgs.set(argIndex);
1565  }
1566
1567  // Check for using an Objective-C specific conversion specifier
1568  // in a non-ObjC literal.
1569  if (!IsObjCLiteral && CS.isObjCArg()) {
1570    return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
1571                                                  specifierLen);
1572  }
1573
1574  // Check for invalid use of field width
1575  if (!FS.hasValidFieldWidth()) {
1576    HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
1577        startSpecifier, specifierLen);
1578  }
1579
1580  // Check for invalid use of precision
1581  if (!FS.hasValidPrecision()) {
1582    HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
1583        startSpecifier, specifierLen);
1584  }
1585
1586  // Check each flag does not conflict with any other component.
1587  if (!FS.hasValidThousandsGroupingPrefix())
1588    HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
1589  if (!FS.hasValidLeadingZeros())
1590    HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
1591  if (!FS.hasValidPlusPrefix())
1592    HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
1593  if (!FS.hasValidSpacePrefix())
1594    HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
1595  if (!FS.hasValidAlternativeForm())
1596    HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
1597  if (!FS.hasValidLeftJustified())
1598    HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
1599
1600  // Check that flags are not ignored by another flag
1601  if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
1602    HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
1603        startSpecifier, specifierLen);
1604  if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
1605    HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
1606            startSpecifier, specifierLen);
1607
1608  // Check the length modifier is valid with the given conversion specifier.
1609  const LengthModifier &LM = FS.getLengthModifier();
1610  if (!FS.hasValidLengthModifier())
1611    S.Diag(getLocationOfByte(LM.getStart()),
1612        diag::warn_format_nonsensical_length)
1613      << LM.toString() << CS.toString()
1614      << getSpecifierRange(startSpecifier, specifierLen)
1615      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1616          LM.getLength()));
1617
1618  // Are we using '%n'?
1619  if (CS.getKind() == ConversionSpecifier::nArg) {
1620    // Issue a warning about this being a possible security issue.
1621    S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back)
1622      << getSpecifierRange(startSpecifier, specifierLen);
1623    // Continue checking the other format specifiers.
1624    return true;
1625  }
1626
1627  // The remaining checks depend on the data arguments.
1628  if (HasVAListArg)
1629    return true;
1630
1631  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1632    return false;
1633
1634  // Now type check the data expression that matches the
1635  // format specifier.
1636  const Expr *Ex = getDataArg(argIndex);
1637  const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context);
1638  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
1639    // Check if we didn't match because of an implicit cast from a 'char'
1640    // or 'short' to an 'int'.  This is done because printf is a varargs
1641    // function.
1642    if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex))
1643      if (ICE->getType() == S.Context.IntTy) {
1644        // All further checking is done on the subexpression.
1645        Ex = ICE->getSubExpr();
1646        if (ATR.matchesType(S.Context, Ex->getType()))
1647          return true;
1648      }
1649
1650    // We may be able to offer a FixItHint if it is a supported type.
1651    PrintfSpecifier fixedFS = FS;
1652    bool success = fixedFS.fixType(Ex->getType());
1653
1654    if (success) {
1655      // Get the fix string from the fixed format specifier
1656      llvm::SmallString<128> buf;
1657      llvm::raw_svector_ostream os(buf);
1658      fixedFS.toString(os);
1659
1660      // FIXME: getRepresentativeType() perhaps should return a string
1661      // instead of a QualType to better handle when the representative
1662      // type is 'wint_t' (which is defined in the system headers).
1663      S.Diag(getLocationOfByte(CS.getStart()),
1664          diag::warn_printf_conversion_argument_type_mismatch)
1665        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1666        << getSpecifierRange(startSpecifier, specifierLen)
1667        << Ex->getSourceRange()
1668        << FixItHint::CreateReplacement(
1669            getSpecifierRange(startSpecifier, specifierLen),
1670            os.str());
1671    }
1672    else {
1673      S.Diag(getLocationOfByte(CS.getStart()),
1674             diag::warn_printf_conversion_argument_type_mismatch)
1675        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1676        << getSpecifierRange(startSpecifier, specifierLen)
1677        << Ex->getSourceRange();
1678    }
1679  }
1680
1681  return true;
1682}
1683
1684//===--- CHECK: Scanf format string checking ------------------------------===//
1685
1686namespace {
1687class CheckScanfHandler : public CheckFormatHandler {
1688public:
1689  CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
1690                    const Expr *origFormatExpr, unsigned firstDataArg,
1691                    unsigned numDataArgs, bool isObjCLiteral,
1692                    const char *beg, bool hasVAListArg,
1693                    const CallExpr *theCall, unsigned formatIdx)
1694  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1695                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1696                       theCall, formatIdx) {}
1697
1698  bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
1699                            const char *startSpecifier,
1700                            unsigned specifierLen);
1701
1702  bool HandleInvalidScanfConversionSpecifier(
1703          const analyze_scanf::ScanfSpecifier &FS,
1704          const char *startSpecifier,
1705          unsigned specifierLen);
1706
1707  void HandleIncompleteScanList(const char *start, const char *end);
1708};
1709}
1710
1711void CheckScanfHandler::HandleIncompleteScanList(const char *start,
1712                                                 const char *end) {
1713  S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete)
1714    << getSpecifierRange(start, end - start);
1715}
1716
1717bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
1718                                        const analyze_scanf::ScanfSpecifier &FS,
1719                                        const char *startSpecifier,
1720                                        unsigned specifierLen) {
1721
1722  const analyze_scanf::ScanfConversionSpecifier &CS =
1723    FS.getConversionSpecifier();
1724
1725  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1726                                          getLocationOfByte(CS.getStart()),
1727                                          startSpecifier, specifierLen,
1728                                          CS.getStart(), CS.getLength());
1729}
1730
1731bool CheckScanfHandler::HandleScanfSpecifier(
1732                                       const analyze_scanf::ScanfSpecifier &FS,
1733                                       const char *startSpecifier,
1734                                       unsigned specifierLen) {
1735
1736  using namespace analyze_scanf;
1737  using namespace analyze_format_string;
1738
1739  const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
1740
1741  // Handle case where '%' and '*' don't consume an argument.  These shouldn't
1742  // be used to decide if we are using positional arguments consistently.
1743  if (FS.consumesDataArgument()) {
1744    if (atFirstArg) {
1745      atFirstArg = false;
1746      usesPositionalArgs = FS.usesPositionalArg();
1747    }
1748    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1749      // Cannot mix-and-match positional and non-positional arguments.
1750      S.Diag(getLocationOfByte(CS.getStart()),
1751             diag::warn_format_mix_positional_nonpositional_args)
1752        << getSpecifierRange(startSpecifier, specifierLen);
1753      return false;
1754    }
1755  }
1756
1757  // Check if the field with is non-zero.
1758  const OptionalAmount &Amt = FS.getFieldWidth();
1759  if (Amt.getHowSpecified() == OptionalAmount::Constant) {
1760    if (Amt.getConstantAmount() == 0) {
1761      const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
1762                                                   Amt.getConstantLength());
1763      S.Diag(getLocationOfByte(Amt.getStart()),
1764             diag::warn_scanf_nonzero_width)
1765        << R << FixItHint::CreateRemoval(R);
1766    }
1767  }
1768
1769  if (!FS.consumesDataArgument()) {
1770    // FIXME: Technically specifying a precision or field width here
1771    // makes no sense.  Worth issuing a warning at some point.
1772    return true;
1773  }
1774
1775  // Consume the argument.
1776  unsigned argIndex = FS.getArgIndex();
1777  if (argIndex < NumDataArgs) {
1778      // The check to see if the argIndex is valid will come later.
1779      // We set the bit here because we may exit early from this
1780      // function if we encounter some other error.
1781    CoveredArgs.set(argIndex);
1782  }
1783
1784  // Check the length modifier is valid with the given conversion specifier.
1785  const LengthModifier &LM = FS.getLengthModifier();
1786  if (!FS.hasValidLengthModifier()) {
1787    S.Diag(getLocationOfByte(LM.getStart()),
1788           diag::warn_format_nonsensical_length)
1789      << LM.toString() << CS.toString()
1790      << getSpecifierRange(startSpecifier, specifierLen)
1791      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1792                                                    LM.getLength()));
1793  }
1794
1795  // The remaining checks depend on the data arguments.
1796  if (HasVAListArg)
1797    return true;
1798
1799  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1800    return false;
1801
1802  // FIXME: Check that the argument type matches the format specifier.
1803
1804  return true;
1805}
1806
1807void Sema::CheckFormatString(const StringLiteral *FExpr,
1808                             const Expr *OrigFormatExpr,
1809                             const CallExpr *TheCall, bool HasVAListArg,
1810                             unsigned format_idx, unsigned firstDataArg,
1811                             bool isPrintf) {
1812
1813  // CHECK: is the format string a wide literal?
1814  if (!FExpr->isAscii()) {
1815    Diag(FExpr->getLocStart(),
1816         diag::warn_format_string_is_wide_literal)
1817    << OrigFormatExpr->getSourceRange();
1818    return;
1819  }
1820
1821  // Str - The format string.  NOTE: this is NOT null-terminated!
1822  StringRef StrRef = FExpr->getString();
1823  const char *Str = StrRef.data();
1824  unsigned StrLen = StrRef.size();
1825
1826  // CHECK: empty format string?
1827  if (StrLen == 0) {
1828    Diag(FExpr->getLocStart(), diag::warn_empty_format_string)
1829    << OrigFormatExpr->getSourceRange();
1830    return;
1831  }
1832
1833  if (isPrintf) {
1834    CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1835                         TheCall->getNumArgs() - firstDataArg,
1836                         isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1837                         HasVAListArg, TheCall, format_idx);
1838
1839    if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen))
1840      H.DoneProcessing();
1841  }
1842  else {
1843    CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1844                        TheCall->getNumArgs() - firstDataArg,
1845                        isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1846                        HasVAListArg, TheCall, format_idx);
1847
1848    if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen))
1849      H.DoneProcessing();
1850  }
1851}
1852
1853//===--- CHECK: Standard memory functions ---------------------------------===//
1854
1855/// \brief Determine whether the given type is a dynamic class type (e.g.,
1856/// whether it has a vtable).
1857static bool isDynamicClassType(QualType T) {
1858  if (CXXRecordDecl *Record = T->getAsCXXRecordDecl())
1859    if (CXXRecordDecl *Definition = Record->getDefinition())
1860      if (Definition->isDynamicClass())
1861        return true;
1862
1863  return false;
1864}
1865
1866/// \brief If E is a sizeof expression, returns its argument expression,
1867/// otherwise returns NULL.
1868static const Expr *getSizeOfExprArg(const Expr* E) {
1869  if (const UnaryExprOrTypeTraitExpr *SizeOf =
1870      dyn_cast<UnaryExprOrTypeTraitExpr>(E))
1871    if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType())
1872      return SizeOf->getArgumentExpr()->IgnoreParenImpCasts();
1873
1874  return 0;
1875}
1876
1877/// \brief If E is a sizeof expression, returns its argument type.
1878static QualType getSizeOfArgType(const Expr* E) {
1879  if (const UnaryExprOrTypeTraitExpr *SizeOf =
1880      dyn_cast<UnaryExprOrTypeTraitExpr>(E))
1881    if (SizeOf->getKind() == clang::UETT_SizeOf)
1882      return SizeOf->getTypeOfArgument();
1883
1884  return QualType();
1885}
1886
1887/// \brief Check for dangerous or invalid arguments to memset().
1888///
1889/// This issues warnings on known problematic, dangerous or unspecified
1890/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp'
1891/// function calls.
1892///
1893/// \param Call The call expression to diagnose.
1894void Sema::CheckMemaccessArguments(const CallExpr *Call,
1895                                   CheckedMemoryFunction CMF,
1896                                   IdentifierInfo *FnName) {
1897  // It is possible to have a non-standard definition of memset.  Validate
1898  // we have enough arguments, and if not, abort further checking.
1899  if (Call->getNumArgs() < 3)
1900    return;
1901
1902  unsigned LastArg = (CMF == CMF_Memset? 1 : 2);
1903  const Expr *LenExpr = Call->getArg(2)->IgnoreParenImpCasts();
1904
1905  // We have special checking when the length is a sizeof expression.
1906  QualType SizeOfArgTy = getSizeOfArgType(LenExpr);
1907  const Expr *SizeOfArg = getSizeOfExprArg(LenExpr);
1908  llvm::FoldingSetNodeID SizeOfArgID;
1909
1910  for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) {
1911    const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts();
1912    SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange();
1913
1914    QualType DestTy = Dest->getType();
1915    if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
1916      QualType PointeeTy = DestPtrTy->getPointeeType();
1917
1918      // Never warn about void type pointers. This can be used to suppress
1919      // false positives.
1920      if (PointeeTy->isVoidType())
1921        continue;
1922
1923      // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by
1924      // actually comparing the expressions for equality. Because computing the
1925      // expression IDs can be expensive, we only do this if the diagnostic is
1926      // enabled.
1927      if (SizeOfArg &&
1928          Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess,
1929                                   SizeOfArg->getExprLoc())) {
1930        // We only compute IDs for expressions if the warning is enabled, and
1931        // cache the sizeof arg's ID.
1932        if (SizeOfArgID == llvm::FoldingSetNodeID())
1933          SizeOfArg->Profile(SizeOfArgID, Context, true);
1934        llvm::FoldingSetNodeID DestID;
1935        Dest->Profile(DestID, Context, true);
1936        if (DestID == SizeOfArgID) {
1937          unsigned ActionIdx = 0; // Default is to suggest dereferencing.
1938          if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest))
1939            if (UnaryOp->getOpcode() == UO_AddrOf)
1940              ActionIdx = 1; // If its an address-of operator, just remove it.
1941          if (Context.getTypeSize(PointeeTy) == Context.getCharWidth())
1942            ActionIdx = 2; // If the pointee's size is sizeof(char),
1943                           // suggest an explicit length.
1944          DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest,
1945                              PDiag(diag::warn_sizeof_pointer_expr_memaccess)
1946                                << FnName << ArgIdx << ActionIdx
1947                                << Dest->getSourceRange()
1948                                << SizeOfArg->getSourceRange());
1949          break;
1950        }
1951      }
1952
1953      // Also check for cases where the sizeof argument is the exact same
1954      // type as the memory argument, and where it points to a user-defined
1955      // record type.
1956      if (SizeOfArgTy != QualType()) {
1957        if (PointeeTy->isRecordType() &&
1958            Context.typesAreCompatible(SizeOfArgTy, DestTy)) {
1959          DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest,
1960                              PDiag(diag::warn_sizeof_pointer_type_memaccess)
1961                                << FnName << SizeOfArgTy << ArgIdx
1962                                << PointeeTy << Dest->getSourceRange()
1963                                << LenExpr->getSourceRange());
1964          break;
1965        }
1966      }
1967
1968      unsigned DiagID;
1969
1970      // Always complain about dynamic classes.
1971      if (isDynamicClassType(PointeeTy))
1972        DiagID = diag::warn_dyn_class_memaccess;
1973      else if (PointeeTy.hasNonTrivialObjCLifetime() && CMF != CMF_Memset)
1974        DiagID = diag::warn_arc_object_memaccess;
1975      else
1976        continue;
1977
1978      DiagRuntimeBehavior(
1979        Dest->getExprLoc(), Dest,
1980        PDiag(DiagID)
1981          << ArgIdx << FnName << PointeeTy
1982          << Call->getCallee()->getSourceRange());
1983
1984      DiagRuntimeBehavior(
1985        Dest->getExprLoc(), Dest,
1986        PDiag(diag::note_bad_memaccess_silence)
1987          << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
1988      break;
1989    }
1990  }
1991}
1992
1993//===--- CHECK: Return Address of Stack Variable --------------------------===//
1994
1995static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars);
1996static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars);
1997
1998/// CheckReturnStackAddr - Check if a return statement returns the address
1999///   of a stack variable.
2000void
2001Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
2002                           SourceLocation ReturnLoc) {
2003
2004  Expr *stackE = 0;
2005  SmallVector<DeclRefExpr *, 8> refVars;
2006
2007  // Perform checking for returned stack addresses, local blocks,
2008  // label addresses or references to temporaries.
2009  if (lhsType->isPointerType() ||
2010      (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) {
2011    stackE = EvalAddr(RetValExp, refVars);
2012  } else if (lhsType->isReferenceType()) {
2013    stackE = EvalVal(RetValExp, refVars);
2014  }
2015
2016  if (stackE == 0)
2017    return; // Nothing suspicious was found.
2018
2019  SourceLocation diagLoc;
2020  SourceRange diagRange;
2021  if (refVars.empty()) {
2022    diagLoc = stackE->getLocStart();
2023    diagRange = stackE->getSourceRange();
2024  } else {
2025    // We followed through a reference variable. 'stackE' contains the
2026    // problematic expression but we will warn at the return statement pointing
2027    // at the reference variable. We will later display the "trail" of
2028    // reference variables using notes.
2029    diagLoc = refVars[0]->getLocStart();
2030    diagRange = refVars[0]->getSourceRange();
2031  }
2032
2033  if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
2034    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
2035                                             : diag::warn_ret_stack_addr)
2036     << DR->getDecl()->getDeclName() << diagRange;
2037  } else if (isa<BlockExpr>(stackE)) { // local block.
2038    Diag(diagLoc, diag::err_ret_local_block) << diagRange;
2039  } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
2040    Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
2041  } else { // local temporary.
2042    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
2043                                             : diag::warn_ret_local_temp_addr)
2044     << diagRange;
2045  }
2046
2047  // Display the "trail" of reference variables that we followed until we
2048  // found the problematic expression using notes.
2049  for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
2050    VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
2051    // If this var binds to another reference var, show the range of the next
2052    // var, otherwise the var binds to the problematic expression, in which case
2053    // show the range of the expression.
2054    SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
2055                                  : stackE->getSourceRange();
2056    Diag(VD->getLocation(), diag::note_ref_var_local_bind)
2057      << VD->getDeclName() << range;
2058  }
2059}
2060
2061/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
2062///  check if the expression in a return statement evaluates to an address
2063///  to a location on the stack, a local block, an address of a label, or a
2064///  reference to local temporary. The recursion is used to traverse the
2065///  AST of the return expression, with recursion backtracking when we
2066///  encounter a subexpression that (1) clearly does not lead to one of the
2067///  above problematic expressions (2) is something we cannot determine leads to
2068///  a problematic expression based on such local checking.
2069///
2070///  Both EvalAddr and EvalVal follow through reference variables to evaluate
2071///  the expression that they point to. Such variables are added to the
2072///  'refVars' vector so that we know what the reference variable "trail" was.
2073///
2074///  EvalAddr processes expressions that are pointers that are used as
2075///  references (and not L-values).  EvalVal handles all other values.
2076///  At the base case of the recursion is a check for the above problematic
2077///  expressions.
2078///
2079///  This implementation handles:
2080///
2081///   * pointer-to-pointer casts
2082///   * implicit conversions from array references to pointers
2083///   * taking the address of fields
2084///   * arbitrary interplay between "&" and "*" operators
2085///   * pointer arithmetic from an address of a stack variable
2086///   * taking the address of an array element where the array is on the stack
2087static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) {
2088  if (E->isTypeDependent())
2089      return NULL;
2090
2091  // We should only be called for evaluating pointer expressions.
2092  assert((E->getType()->isAnyPointerType() ||
2093          E->getType()->isBlockPointerType() ||
2094          E->getType()->isObjCQualifiedIdType()) &&
2095         "EvalAddr only works on pointers");
2096
2097  E = E->IgnoreParens();
2098
2099  // Our "symbolic interpreter" is just a dispatch off the currently
2100  // viewed AST node.  We then recursively traverse the AST by calling
2101  // EvalAddr and EvalVal appropriately.
2102  switch (E->getStmtClass()) {
2103  case Stmt::DeclRefExprClass: {
2104    DeclRefExpr *DR = cast<DeclRefExpr>(E);
2105
2106    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2107      // If this is a reference variable, follow through to the expression that
2108      // it points to.
2109      if (V->hasLocalStorage() &&
2110          V->getType()->isReferenceType() && V->hasInit()) {
2111        // Add the reference variable to the "trail".
2112        refVars.push_back(DR);
2113        return EvalAddr(V->getInit(), refVars);
2114      }
2115
2116    return NULL;
2117  }
2118
2119  case Stmt::UnaryOperatorClass: {
2120    // The only unary operator that make sense to handle here
2121    // is AddrOf.  All others don't make sense as pointers.
2122    UnaryOperator *U = cast<UnaryOperator>(E);
2123
2124    if (U->getOpcode() == UO_AddrOf)
2125      return EvalVal(U->getSubExpr(), refVars);
2126    else
2127      return NULL;
2128  }
2129
2130  case Stmt::BinaryOperatorClass: {
2131    // Handle pointer arithmetic.  All other binary operators are not valid
2132    // in this context.
2133    BinaryOperator *B = cast<BinaryOperator>(E);
2134    BinaryOperatorKind op = B->getOpcode();
2135
2136    if (op != BO_Add && op != BO_Sub)
2137      return NULL;
2138
2139    Expr *Base = B->getLHS();
2140
2141    // Determine which argument is the real pointer base.  It could be
2142    // the RHS argument instead of the LHS.
2143    if (!Base->getType()->isPointerType()) Base = B->getRHS();
2144
2145    assert (Base->getType()->isPointerType());
2146    return EvalAddr(Base, refVars);
2147  }
2148
2149  // For conditional operators we need to see if either the LHS or RHS are
2150  // valid DeclRefExpr*s.  If one of them is valid, we return it.
2151  case Stmt::ConditionalOperatorClass: {
2152    ConditionalOperator *C = cast<ConditionalOperator>(E);
2153
2154    // Handle the GNU extension for missing LHS.
2155    if (Expr *lhsExpr = C->getLHS()) {
2156    // In C++, we can have a throw-expression, which has 'void' type.
2157      if (!lhsExpr->getType()->isVoidType())
2158        if (Expr* LHS = EvalAddr(lhsExpr, refVars))
2159          return LHS;
2160    }
2161
2162    // In C++, we can have a throw-expression, which has 'void' type.
2163    if (C->getRHS()->getType()->isVoidType())
2164      return NULL;
2165
2166    return EvalAddr(C->getRHS(), refVars);
2167  }
2168
2169  case Stmt::BlockExprClass:
2170    if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
2171      return E; // local block.
2172    return NULL;
2173
2174  case Stmt::AddrLabelExprClass:
2175    return E; // address of label.
2176
2177  // For casts, we need to handle conversions from arrays to
2178  // pointer values, and pointer-to-pointer conversions.
2179  case Stmt::ImplicitCastExprClass:
2180  case Stmt::CStyleCastExprClass:
2181  case Stmt::CXXFunctionalCastExprClass:
2182  case Stmt::ObjCBridgedCastExprClass: {
2183    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
2184    QualType T = SubExpr->getType();
2185
2186    if (SubExpr->getType()->isPointerType() ||
2187        SubExpr->getType()->isBlockPointerType() ||
2188        SubExpr->getType()->isObjCQualifiedIdType())
2189      return EvalAddr(SubExpr, refVars);
2190    else if (T->isArrayType())
2191      return EvalVal(SubExpr, refVars);
2192    else
2193      return 0;
2194  }
2195
2196  // C++ casts.  For dynamic casts, static casts, and const casts, we
2197  // are always converting from a pointer-to-pointer, so we just blow
2198  // through the cast.  In the case the dynamic cast doesn't fail (and
2199  // return NULL), we take the conservative route and report cases
2200  // where we return the address of a stack variable.  For Reinterpre
2201  // FIXME: The comment about is wrong; we're not always converting
2202  // from pointer to pointer. I'm guessing that this code should also
2203  // handle references to objects.
2204  case Stmt::CXXStaticCastExprClass:
2205  case Stmt::CXXDynamicCastExprClass:
2206  case Stmt::CXXConstCastExprClass:
2207  case Stmt::CXXReinterpretCastExprClass: {
2208      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
2209      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
2210        return EvalAddr(S, refVars);
2211      else
2212        return NULL;
2213  }
2214
2215  case Stmt::MaterializeTemporaryExprClass:
2216    if (Expr *Result = EvalAddr(
2217                         cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
2218                                refVars))
2219      return Result;
2220
2221    return E;
2222
2223  // Everything else: we simply don't reason about them.
2224  default:
2225    return NULL;
2226  }
2227}
2228
2229
2230///  EvalVal - This function is complements EvalAddr in the mutual recursion.
2231///   See the comments for EvalAddr for more details.
2232static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) {
2233do {
2234  // We should only be called for evaluating non-pointer expressions, or
2235  // expressions with a pointer type that are not used as references but instead
2236  // are l-values (e.g., DeclRefExpr with a pointer type).
2237
2238  // Our "symbolic interpreter" is just a dispatch off the currently
2239  // viewed AST node.  We then recursively traverse the AST by calling
2240  // EvalAddr and EvalVal appropriately.
2241
2242  E = E->IgnoreParens();
2243  switch (E->getStmtClass()) {
2244  case Stmt::ImplicitCastExprClass: {
2245    ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
2246    if (IE->getValueKind() == VK_LValue) {
2247      E = IE->getSubExpr();
2248      continue;
2249    }
2250    return NULL;
2251  }
2252
2253  case Stmt::DeclRefExprClass: {
2254    // When we hit a DeclRefExpr we are looking at code that refers to a
2255    // variable's name. If it's not a reference variable we check if it has
2256    // local storage within the function, and if so, return the expression.
2257    DeclRefExpr *DR = cast<DeclRefExpr>(E);
2258
2259    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2260      if (V->hasLocalStorage()) {
2261        if (!V->getType()->isReferenceType())
2262          return DR;
2263
2264        // Reference variable, follow through to the expression that
2265        // it points to.
2266        if (V->hasInit()) {
2267          // Add the reference variable to the "trail".
2268          refVars.push_back(DR);
2269          return EvalVal(V->getInit(), refVars);
2270        }
2271      }
2272
2273    return NULL;
2274  }
2275
2276  case Stmt::UnaryOperatorClass: {
2277    // The only unary operator that make sense to handle here
2278    // is Deref.  All others don't resolve to a "name."  This includes
2279    // handling all sorts of rvalues passed to a unary operator.
2280    UnaryOperator *U = cast<UnaryOperator>(E);
2281
2282    if (U->getOpcode() == UO_Deref)
2283      return EvalAddr(U->getSubExpr(), refVars);
2284
2285    return NULL;
2286  }
2287
2288  case Stmt::ArraySubscriptExprClass: {
2289    // Array subscripts are potential references to data on the stack.  We
2290    // retrieve the DeclRefExpr* for the array variable if it indeed
2291    // has local storage.
2292    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars);
2293  }
2294
2295  case Stmt::ConditionalOperatorClass: {
2296    // For conditional operators we need to see if either the LHS or RHS are
2297    // non-NULL Expr's.  If one is non-NULL, we return it.
2298    ConditionalOperator *C = cast<ConditionalOperator>(E);
2299
2300    // Handle the GNU extension for missing LHS.
2301    if (Expr *lhsExpr = C->getLHS())
2302      if (Expr *LHS = EvalVal(lhsExpr, refVars))
2303        return LHS;
2304
2305    return EvalVal(C->getRHS(), refVars);
2306  }
2307
2308  // Accesses to members are potential references to data on the stack.
2309  case Stmt::MemberExprClass: {
2310    MemberExpr *M = cast<MemberExpr>(E);
2311
2312    // Check for indirect access.  We only want direct field accesses.
2313    if (M->isArrow())
2314      return NULL;
2315
2316    // Check whether the member type is itself a reference, in which case
2317    // we're not going to refer to the member, but to what the member refers to.
2318    if (M->getMemberDecl()->getType()->isReferenceType())
2319      return NULL;
2320
2321    return EvalVal(M->getBase(), refVars);
2322  }
2323
2324  case Stmt::MaterializeTemporaryExprClass:
2325    if (Expr *Result = EvalVal(
2326                          cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
2327                               refVars))
2328      return Result;
2329
2330    return E;
2331
2332  default:
2333    // Check that we don't return or take the address of a reference to a
2334    // temporary. This is only useful in C++.
2335    if (!E->isTypeDependent() && E->isRValue())
2336      return E;
2337
2338    // Everything else: we simply don't reason about them.
2339    return NULL;
2340  }
2341} while (true);
2342}
2343
2344//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
2345
2346/// Check for comparisons of floating point operands using != and ==.
2347/// Issue a warning if these are no self-comparisons, as they are not likely
2348/// to do what the programmer intended.
2349void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) {
2350  bool EmitWarning = true;
2351
2352  Expr* LeftExprSansParen = lex->IgnoreParenImpCasts();
2353  Expr* RightExprSansParen = rex->IgnoreParenImpCasts();
2354
2355  // Special case: check for x == x (which is OK).
2356  // Do not emit warnings for such cases.
2357  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
2358    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
2359      if (DRL->getDecl() == DRR->getDecl())
2360        EmitWarning = false;
2361
2362
2363  // Special case: check for comparisons against literals that can be exactly
2364  //  represented by APFloat.  In such cases, do not emit a warning.  This
2365  //  is a heuristic: often comparison against such literals are used to
2366  //  detect if a value in a variable has not changed.  This clearly can
2367  //  lead to false negatives.
2368  if (EmitWarning) {
2369    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
2370      if (FLL->isExact())
2371        EmitWarning = false;
2372    } else
2373      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
2374        if (FLR->isExact())
2375          EmitWarning = false;
2376    }
2377  }
2378
2379  // Check for comparisons with builtin types.
2380  if (EmitWarning)
2381    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
2382      if (CL->isBuiltinCall(Context))
2383        EmitWarning = false;
2384
2385  if (EmitWarning)
2386    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
2387      if (CR->isBuiltinCall(Context))
2388        EmitWarning = false;
2389
2390  // Emit the diagnostic.
2391  if (EmitWarning)
2392    Diag(loc, diag::warn_floatingpoint_eq)
2393      << lex->getSourceRange() << rex->getSourceRange();
2394}
2395
2396//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
2397//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
2398
2399namespace {
2400
2401/// Structure recording the 'active' range of an integer-valued
2402/// expression.
2403struct IntRange {
2404  /// The number of bits active in the int.
2405  unsigned Width;
2406
2407  /// True if the int is known not to have negative values.
2408  bool NonNegative;
2409
2410  IntRange(unsigned Width, bool NonNegative)
2411    : Width(Width), NonNegative(NonNegative)
2412  {}
2413
2414  /// Returns the range of the bool type.
2415  static IntRange forBoolType() {
2416    return IntRange(1, true);
2417  }
2418
2419  /// Returns the range of an opaque value of the given integral type.
2420  static IntRange forValueOfType(ASTContext &C, QualType T) {
2421    return forValueOfCanonicalType(C,
2422                          T->getCanonicalTypeInternal().getTypePtr());
2423  }
2424
2425  /// Returns the range of an opaque value of a canonical integral type.
2426  static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
2427    assert(T->isCanonicalUnqualified());
2428
2429    if (const VectorType *VT = dyn_cast<VectorType>(T))
2430      T = VT->getElementType().getTypePtr();
2431    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2432      T = CT->getElementType().getTypePtr();
2433
2434    // For enum types, use the known bit width of the enumerators.
2435    if (const EnumType *ET = dyn_cast<EnumType>(T)) {
2436      EnumDecl *Enum = ET->getDecl();
2437      if (!Enum->isDefinition())
2438        return IntRange(C.getIntWidth(QualType(T, 0)), false);
2439
2440      unsigned NumPositive = Enum->getNumPositiveBits();
2441      unsigned NumNegative = Enum->getNumNegativeBits();
2442
2443      return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
2444    }
2445
2446    const BuiltinType *BT = cast<BuiltinType>(T);
2447    assert(BT->isInteger());
2448
2449    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2450  }
2451
2452  /// Returns the "target" range of a canonical integral type, i.e.
2453  /// the range of values expressible in the type.
2454  ///
2455  /// This matches forValueOfCanonicalType except that enums have the
2456  /// full range of their type, not the range of their enumerators.
2457  static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
2458    assert(T->isCanonicalUnqualified());
2459
2460    if (const VectorType *VT = dyn_cast<VectorType>(T))
2461      T = VT->getElementType().getTypePtr();
2462    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2463      T = CT->getElementType().getTypePtr();
2464    if (const EnumType *ET = dyn_cast<EnumType>(T))
2465      T = ET->getDecl()->getIntegerType().getTypePtr();
2466
2467    const BuiltinType *BT = cast<BuiltinType>(T);
2468    assert(BT->isInteger());
2469
2470    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2471  }
2472
2473  /// Returns the supremum of two ranges: i.e. their conservative merge.
2474  static IntRange join(IntRange L, IntRange R) {
2475    return IntRange(std::max(L.Width, R.Width),
2476                    L.NonNegative && R.NonNegative);
2477  }
2478
2479  /// Returns the infinum of two ranges: i.e. their aggressive merge.
2480  static IntRange meet(IntRange L, IntRange R) {
2481    return IntRange(std::min(L.Width, R.Width),
2482                    L.NonNegative || R.NonNegative);
2483  }
2484};
2485
2486IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) {
2487  if (value.isSigned() && value.isNegative())
2488    return IntRange(value.getMinSignedBits(), false);
2489
2490  if (value.getBitWidth() > MaxWidth)
2491    value = value.trunc(MaxWidth);
2492
2493  // isNonNegative() just checks the sign bit without considering
2494  // signedness.
2495  return IntRange(value.getActiveBits(), true);
2496}
2497
2498IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
2499                       unsigned MaxWidth) {
2500  if (result.isInt())
2501    return GetValueRange(C, result.getInt(), MaxWidth);
2502
2503  if (result.isVector()) {
2504    IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
2505    for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
2506      IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
2507      R = IntRange::join(R, El);
2508    }
2509    return R;
2510  }
2511
2512  if (result.isComplexInt()) {
2513    IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
2514    IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
2515    return IntRange::join(R, I);
2516  }
2517
2518  // This can happen with lossless casts to intptr_t of "based" lvalues.
2519  // Assume it might use arbitrary bits.
2520  // FIXME: The only reason we need to pass the type in here is to get
2521  // the sign right on this one case.  It would be nice if APValue
2522  // preserved this.
2523  assert(result.isLValue());
2524  return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType());
2525}
2526
2527/// Pseudo-evaluate the given integer expression, estimating the
2528/// range of values it might take.
2529///
2530/// \param MaxWidth - the width to which the value will be truncated
2531IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
2532  E = E->IgnoreParens();
2533
2534  // Try a full evaluation first.
2535  Expr::EvalResult result;
2536  if (E->Evaluate(result, C))
2537    return GetValueRange(C, result.Val, E->getType(), MaxWidth);
2538
2539  // I think we only want to look through implicit casts here; if the
2540  // user has an explicit widening cast, we should treat the value as
2541  // being of the new, wider type.
2542  if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
2543    if (CE->getCastKind() == CK_NoOp)
2544      return GetExprRange(C, CE->getSubExpr(), MaxWidth);
2545
2546    IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
2547
2548    bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
2549
2550    // Assume that non-integer casts can span the full range of the type.
2551    if (!isIntegerCast)
2552      return OutputTypeRange;
2553
2554    IntRange SubRange
2555      = GetExprRange(C, CE->getSubExpr(),
2556                     std::min(MaxWidth, OutputTypeRange.Width));
2557
2558    // Bail out if the subexpr's range is as wide as the cast type.
2559    if (SubRange.Width >= OutputTypeRange.Width)
2560      return OutputTypeRange;
2561
2562    // Otherwise, we take the smaller width, and we're non-negative if
2563    // either the output type or the subexpr is.
2564    return IntRange(SubRange.Width,
2565                    SubRange.NonNegative || OutputTypeRange.NonNegative);
2566  }
2567
2568  if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
2569    // If we can fold the condition, just take that operand.
2570    bool CondResult;
2571    if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
2572      return GetExprRange(C, CondResult ? CO->getTrueExpr()
2573                                        : CO->getFalseExpr(),
2574                          MaxWidth);
2575
2576    // Otherwise, conservatively merge.
2577    IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
2578    IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
2579    return IntRange::join(L, R);
2580  }
2581
2582  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
2583    switch (BO->getOpcode()) {
2584
2585    // Boolean-valued operations are single-bit and positive.
2586    case BO_LAnd:
2587    case BO_LOr:
2588    case BO_LT:
2589    case BO_GT:
2590    case BO_LE:
2591    case BO_GE:
2592    case BO_EQ:
2593    case BO_NE:
2594      return IntRange::forBoolType();
2595
2596    // The type of the assignments is the type of the LHS, so the RHS
2597    // is not necessarily the same type.
2598    case BO_MulAssign:
2599    case BO_DivAssign:
2600    case BO_RemAssign:
2601    case BO_AddAssign:
2602    case BO_SubAssign:
2603    case BO_XorAssign:
2604    case BO_OrAssign:
2605      // TODO: bitfields?
2606      return IntRange::forValueOfType(C, E->getType());
2607
2608    // Simple assignments just pass through the RHS, which will have
2609    // been coerced to the LHS type.
2610    case BO_Assign:
2611      // TODO: bitfields?
2612      return GetExprRange(C, BO->getRHS(), MaxWidth);
2613
2614    // Operations with opaque sources are black-listed.
2615    case BO_PtrMemD:
2616    case BO_PtrMemI:
2617      return IntRange::forValueOfType(C, E->getType());
2618
2619    // Bitwise-and uses the *infinum* of the two source ranges.
2620    case BO_And:
2621    case BO_AndAssign:
2622      return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
2623                            GetExprRange(C, BO->getRHS(), MaxWidth));
2624
2625    // Left shift gets black-listed based on a judgement call.
2626    case BO_Shl:
2627      // ...except that we want to treat '1 << (blah)' as logically
2628      // positive.  It's an important idiom.
2629      if (IntegerLiteral *I
2630            = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
2631        if (I->getValue() == 1) {
2632          IntRange R = IntRange::forValueOfType(C, E->getType());
2633          return IntRange(R.Width, /*NonNegative*/ true);
2634        }
2635      }
2636      // fallthrough
2637
2638    case BO_ShlAssign:
2639      return IntRange::forValueOfType(C, E->getType());
2640
2641    // Right shift by a constant can narrow its left argument.
2642    case BO_Shr:
2643    case BO_ShrAssign: {
2644      IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2645
2646      // If the shift amount is a positive constant, drop the width by
2647      // that much.
2648      llvm::APSInt shift;
2649      if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
2650          shift.isNonNegative()) {
2651        unsigned zext = shift.getZExtValue();
2652        if (zext >= L.Width)
2653          L.Width = (L.NonNegative ? 0 : 1);
2654        else
2655          L.Width -= zext;
2656      }
2657
2658      return L;
2659    }
2660
2661    // Comma acts as its right operand.
2662    case BO_Comma:
2663      return GetExprRange(C, BO->getRHS(), MaxWidth);
2664
2665    // Black-list pointer subtractions.
2666    case BO_Sub:
2667      if (BO->getLHS()->getType()->isPointerType())
2668        return IntRange::forValueOfType(C, E->getType());
2669      break;
2670
2671    // The width of a division result is mostly determined by the size
2672    // of the LHS.
2673    case BO_Div: {
2674      // Don't 'pre-truncate' the operands.
2675      unsigned opWidth = C.getIntWidth(E->getType());
2676      IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
2677
2678      // If the divisor is constant, use that.
2679      llvm::APSInt divisor;
2680      if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) {
2681        unsigned log2 = divisor.logBase2(); // floor(log_2(divisor))
2682        if (log2 >= L.Width)
2683          L.Width = (L.NonNegative ? 0 : 1);
2684        else
2685          L.Width = std::min(L.Width - log2, MaxWidth);
2686        return L;
2687      }
2688
2689      // Otherwise, just use the LHS's width.
2690      IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
2691      return IntRange(L.Width, L.NonNegative && R.NonNegative);
2692    }
2693
2694    // The result of a remainder can't be larger than the result of
2695    // either side.
2696    case BO_Rem: {
2697      // Don't 'pre-truncate' the operands.
2698      unsigned opWidth = C.getIntWidth(E->getType());
2699      IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
2700      IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
2701
2702      IntRange meet = IntRange::meet(L, R);
2703      meet.Width = std::min(meet.Width, MaxWidth);
2704      return meet;
2705    }
2706
2707    // The default behavior is okay for these.
2708    case BO_Mul:
2709    case BO_Add:
2710    case BO_Xor:
2711    case BO_Or:
2712      break;
2713    }
2714
2715    // The default case is to treat the operation as if it were closed
2716    // on the narrowest type that encompasses both operands.
2717    IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2718    IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
2719    return IntRange::join(L, R);
2720  }
2721
2722  if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
2723    switch (UO->getOpcode()) {
2724    // Boolean-valued operations are white-listed.
2725    case UO_LNot:
2726      return IntRange::forBoolType();
2727
2728    // Operations with opaque sources are black-listed.
2729    case UO_Deref:
2730    case UO_AddrOf: // should be impossible
2731      return IntRange::forValueOfType(C, E->getType());
2732
2733    default:
2734      return GetExprRange(C, UO->getSubExpr(), MaxWidth);
2735    }
2736  }
2737
2738  if (dyn_cast<OffsetOfExpr>(E)) {
2739    IntRange::forValueOfType(C, E->getType());
2740  }
2741
2742  FieldDecl *BitField = E->getBitField();
2743  if (BitField) {
2744    llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C);
2745    unsigned BitWidth = BitWidthAP.getZExtValue();
2746
2747    return IntRange(BitWidth,
2748                    BitField->getType()->isUnsignedIntegerOrEnumerationType());
2749  }
2750
2751  return IntRange::forValueOfType(C, E->getType());
2752}
2753
2754IntRange GetExprRange(ASTContext &C, Expr *E) {
2755  return GetExprRange(C, E, C.getIntWidth(E->getType()));
2756}
2757
2758/// Checks whether the given value, which currently has the given
2759/// source semantics, has the same value when coerced through the
2760/// target semantics.
2761bool IsSameFloatAfterCast(const llvm::APFloat &value,
2762                          const llvm::fltSemantics &Src,
2763                          const llvm::fltSemantics &Tgt) {
2764  llvm::APFloat truncated = value;
2765
2766  bool ignored;
2767  truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
2768  truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
2769
2770  return truncated.bitwiseIsEqual(value);
2771}
2772
2773/// Checks whether the given value, which currently has the given
2774/// source semantics, has the same value when coerced through the
2775/// target semantics.
2776///
2777/// The value might be a vector of floats (or a complex number).
2778bool IsSameFloatAfterCast(const APValue &value,
2779                          const llvm::fltSemantics &Src,
2780                          const llvm::fltSemantics &Tgt) {
2781  if (value.isFloat())
2782    return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
2783
2784  if (value.isVector()) {
2785    for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
2786      if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
2787        return false;
2788    return true;
2789  }
2790
2791  assert(value.isComplexFloat());
2792  return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
2793          IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
2794}
2795
2796void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
2797
2798static bool IsZero(Sema &S, Expr *E) {
2799  // Suppress cases where we are comparing against an enum constant.
2800  if (const DeclRefExpr *DR =
2801      dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
2802    if (isa<EnumConstantDecl>(DR->getDecl()))
2803      return false;
2804
2805  // Suppress cases where the '0' value is expanded from a macro.
2806  if (E->getLocStart().isMacroID())
2807    return false;
2808
2809  llvm::APSInt Value;
2810  return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
2811}
2812
2813static bool HasEnumType(Expr *E) {
2814  // Strip off implicit integral promotions.
2815  while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
2816    if (ICE->getCastKind() != CK_IntegralCast &&
2817        ICE->getCastKind() != CK_NoOp)
2818      break;
2819    E = ICE->getSubExpr();
2820  }
2821
2822  return E->getType()->isEnumeralType();
2823}
2824
2825void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
2826  BinaryOperatorKind op = E->getOpcode();
2827  if (E->isValueDependent())
2828    return;
2829
2830  if (op == BO_LT && IsZero(S, E->getRHS())) {
2831    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2832      << "< 0" << "false" << HasEnumType(E->getLHS())
2833      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2834  } else if (op == BO_GE && IsZero(S, E->getRHS())) {
2835    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2836      << ">= 0" << "true" << HasEnumType(E->getLHS())
2837      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2838  } else if (op == BO_GT && IsZero(S, E->getLHS())) {
2839    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2840      << "0 >" << "false" << HasEnumType(E->getRHS())
2841      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2842  } else if (op == BO_LE && IsZero(S, E->getLHS())) {
2843    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2844      << "0 <=" << "true" << HasEnumType(E->getRHS())
2845      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2846  }
2847}
2848
2849/// Analyze the operands of the given comparison.  Implements the
2850/// fallback case from AnalyzeComparison.
2851void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
2852  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
2853  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
2854}
2855
2856/// \brief Implements -Wsign-compare.
2857///
2858/// \param lex the left-hand expression
2859/// \param rex the right-hand expression
2860/// \param OpLoc the location of the joining operator
2861/// \param BinOpc binary opcode or 0
2862void AnalyzeComparison(Sema &S, BinaryOperator *E) {
2863  // The type the comparison is being performed in.
2864  QualType T = E->getLHS()->getType();
2865  assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
2866         && "comparison with mismatched types");
2867
2868  // We don't do anything special if this isn't an unsigned integral
2869  // comparison:  we're only interested in integral comparisons, and
2870  // signed comparisons only happen in cases we don't care to warn about.
2871  //
2872  // We also don't care about value-dependent expressions or expressions
2873  // whose result is a constant.
2874  if (!T->hasUnsignedIntegerRepresentation()
2875      || E->isValueDependent() || E->isIntegerConstantExpr(S.Context))
2876    return AnalyzeImpConvsInComparison(S, E);
2877
2878  Expr *lex = E->getLHS()->IgnoreParenImpCasts();
2879  Expr *rex = E->getRHS()->IgnoreParenImpCasts();
2880
2881  // Check to see if one of the (unmodified) operands is of different
2882  // signedness.
2883  Expr *signedOperand, *unsignedOperand;
2884  if (lex->getType()->hasSignedIntegerRepresentation()) {
2885    assert(!rex->getType()->hasSignedIntegerRepresentation() &&
2886           "unsigned comparison between two signed integer expressions?");
2887    signedOperand = lex;
2888    unsignedOperand = rex;
2889  } else if (rex->getType()->hasSignedIntegerRepresentation()) {
2890    signedOperand = rex;
2891    unsignedOperand = lex;
2892  } else {
2893    CheckTrivialUnsignedComparison(S, E);
2894    return AnalyzeImpConvsInComparison(S, E);
2895  }
2896
2897  // Otherwise, calculate the effective range of the signed operand.
2898  IntRange signedRange = GetExprRange(S.Context, signedOperand);
2899
2900  // Go ahead and analyze implicit conversions in the operands.  Note
2901  // that we skip the implicit conversions on both sides.
2902  AnalyzeImplicitConversions(S, lex, E->getOperatorLoc());
2903  AnalyzeImplicitConversions(S, rex, E->getOperatorLoc());
2904
2905  // If the signed range is non-negative, -Wsign-compare won't fire,
2906  // but we should still check for comparisons which are always true
2907  // or false.
2908  if (signedRange.NonNegative)
2909    return CheckTrivialUnsignedComparison(S, E);
2910
2911  // For (in)equality comparisons, if the unsigned operand is a
2912  // constant which cannot collide with a overflowed signed operand,
2913  // then reinterpreting the signed operand as unsigned will not
2914  // change the result of the comparison.
2915  if (E->isEqualityOp()) {
2916    unsigned comparisonWidth = S.Context.getIntWidth(T);
2917    IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
2918
2919    // We should never be unable to prove that the unsigned operand is
2920    // non-negative.
2921    assert(unsignedRange.NonNegative && "unsigned range includes negative?");
2922
2923    if (unsignedRange.Width < comparisonWidth)
2924      return;
2925  }
2926
2927  S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison)
2928    << lex->getType() << rex->getType()
2929    << lex->getSourceRange() << rex->getSourceRange();
2930}
2931
2932/// Analyzes an attempt to assign the given value to a bitfield.
2933///
2934/// Returns true if there was something fishy about the attempt.
2935bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
2936                               SourceLocation InitLoc) {
2937  assert(Bitfield->isBitField());
2938  if (Bitfield->isInvalidDecl())
2939    return false;
2940
2941  // White-list bool bitfields.
2942  if (Bitfield->getType()->isBooleanType())
2943    return false;
2944
2945  // Ignore value- or type-dependent expressions.
2946  if (Bitfield->getBitWidth()->isValueDependent() ||
2947      Bitfield->getBitWidth()->isTypeDependent() ||
2948      Init->isValueDependent() ||
2949      Init->isTypeDependent())
2950    return false;
2951
2952  Expr *OriginalInit = Init->IgnoreParenImpCasts();
2953
2954  llvm::APSInt Width(32);
2955  Expr::EvalResult InitValue;
2956  if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) ||
2957      !OriginalInit->Evaluate(InitValue, S.Context) ||
2958      !InitValue.Val.isInt())
2959    return false;
2960
2961  const llvm::APSInt &Value = InitValue.Val.getInt();
2962  unsigned OriginalWidth = Value.getBitWidth();
2963  unsigned FieldWidth = Width.getZExtValue();
2964
2965  if (OriginalWidth <= FieldWidth)
2966    return false;
2967
2968  llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
2969
2970  // It's fairly common to write values into signed bitfields
2971  // that, if sign-extended, would end up becoming a different
2972  // value.  We don't want to warn about that.
2973  if (Value.isSigned() && Value.isNegative())
2974    TruncatedValue = TruncatedValue.sext(OriginalWidth);
2975  else
2976    TruncatedValue = TruncatedValue.zext(OriginalWidth);
2977
2978  if (Value == TruncatedValue)
2979    return false;
2980
2981  std::string PrettyValue = Value.toString(10);
2982  std::string PrettyTrunc = TruncatedValue.toString(10);
2983
2984  S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
2985    << PrettyValue << PrettyTrunc << OriginalInit->getType()
2986    << Init->getSourceRange();
2987
2988  return true;
2989}
2990
2991/// Analyze the given simple or compound assignment for warning-worthy
2992/// operations.
2993void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
2994  // Just recurse on the LHS.
2995  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
2996
2997  // We want to recurse on the RHS as normal unless we're assigning to
2998  // a bitfield.
2999  if (FieldDecl *Bitfield = E->getLHS()->getBitField()) {
3000    if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
3001                                  E->getOperatorLoc())) {
3002      // Recurse, ignoring any implicit conversions on the RHS.
3003      return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
3004                                        E->getOperatorLoc());
3005    }
3006  }
3007
3008  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
3009}
3010
3011/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
3012void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
3013                     SourceLocation CContext, unsigned diag) {
3014  S.Diag(E->getExprLoc(), diag)
3015    << SourceType << T << E->getSourceRange() << SourceRange(CContext);
3016}
3017
3018/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
3019void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext,
3020                     unsigned diag) {
3021  DiagnoseImpCast(S, E, E->getType(), T, CContext, diag);
3022}
3023
3024/// Diagnose an implicit cast from a literal expression. Also attemps to supply
3025/// fixit hints when the cast wouldn't lose information to simply write the
3026/// expression with the expected type.
3027void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
3028                                    SourceLocation CContext) {
3029  // Emit the primary warning first, then try to emit a fixit hint note if
3030  // reasonable.
3031  S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
3032    << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext);
3033
3034  const llvm::APFloat &Value = FL->getValue();
3035
3036  // Don't attempt to fix PPC double double literals.
3037  if (&Value.getSemantics() == &llvm::APFloat::PPCDoubleDouble)
3038    return;
3039
3040  // Try to convert this exactly to an integer.
3041  bool isExact = false;
3042  llvm::APSInt IntegerValue(S.Context.getIntWidth(T),
3043                            T->hasUnsignedIntegerRepresentation());
3044  if (Value.convertToInteger(IntegerValue,
3045                             llvm::APFloat::rmTowardZero, &isExact)
3046      != llvm::APFloat::opOK || !isExact)
3047    return;
3048
3049  std::string LiteralValue = IntegerValue.toString(10);
3050  S.Diag(FL->getExprLoc(), diag::note_fix_integral_float_as_integer)
3051    << FixItHint::CreateReplacement(FL->getSourceRange(), LiteralValue);
3052}
3053
3054std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
3055  if (!Range.Width) return "0";
3056
3057  llvm::APSInt ValueInRange = Value;
3058  ValueInRange.setIsSigned(!Range.NonNegative);
3059  ValueInRange = ValueInRange.trunc(Range.Width);
3060  return ValueInRange.toString(10);
3061}
3062
3063static bool isFromSystemMacro(Sema &S, SourceLocation loc) {
3064  SourceManager &smgr = S.Context.getSourceManager();
3065  return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc));
3066}
3067
3068void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
3069                             SourceLocation CC, bool *ICContext = 0) {
3070  if (E->isTypeDependent() || E->isValueDependent()) return;
3071
3072  const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
3073  const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
3074  if (Source == Target) return;
3075  if (Target->isDependentType()) return;
3076
3077  // If the conversion context location is invalid don't complain. We also
3078  // don't want to emit a warning if the issue occurs from the expansion of
3079  // a system macro. The problem is that 'getSpellingLoc()' is slow, so we
3080  // delay this check as long as possible. Once we detect we are in that
3081  // scenario, we just return.
3082  if (CC.isInvalid())
3083    return;
3084
3085  // Never diagnose implicit casts to bool.
3086  if (Target->isSpecificBuiltinType(BuiltinType::Bool))
3087    return;
3088
3089  // Strip vector types.
3090  if (isa<VectorType>(Source)) {
3091    if (!isa<VectorType>(Target)) {
3092      if (isFromSystemMacro(S, CC))
3093        return;
3094      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
3095    }
3096
3097    // If the vector cast is cast between two vectors of the same size, it is
3098    // a bitcast, not a conversion.
3099    if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target))
3100      return;
3101
3102    Source = cast<VectorType>(Source)->getElementType().getTypePtr();
3103    Target = cast<VectorType>(Target)->getElementType().getTypePtr();
3104  }
3105
3106  // Strip complex types.
3107  if (isa<ComplexType>(Source)) {
3108    if (!isa<ComplexType>(Target)) {
3109      if (isFromSystemMacro(S, CC))
3110        return;
3111
3112      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
3113    }
3114
3115    Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
3116    Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
3117  }
3118
3119  const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
3120  const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
3121
3122  // If the source is floating point...
3123  if (SourceBT && SourceBT->isFloatingPoint()) {
3124    // ...and the target is floating point...
3125    if (TargetBT && TargetBT->isFloatingPoint()) {
3126      // ...then warn if we're dropping FP rank.
3127
3128      // Builtin FP kinds are ordered by increasing FP rank.
3129      if (SourceBT->getKind() > TargetBT->getKind()) {
3130        // Don't warn about float constants that are precisely
3131        // representable in the target type.
3132        Expr::EvalResult result;
3133        if (E->Evaluate(result, S.Context)) {
3134          // Value might be a float, a float vector, or a float complex.
3135          if (IsSameFloatAfterCast(result.Val,
3136                   S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
3137                   S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
3138            return;
3139        }
3140
3141        if (isFromSystemMacro(S, CC))
3142          return;
3143
3144        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
3145      }
3146      return;
3147    }
3148
3149    // If the target is integral, always warn.
3150    if ((TargetBT && TargetBT->isInteger())) {
3151      if (isFromSystemMacro(S, CC))
3152        return;
3153
3154      Expr *InnerE = E->IgnoreParenImpCasts();
3155      if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
3156        DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
3157      } else {
3158        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
3159      }
3160    }
3161
3162    return;
3163  }
3164
3165  if (!Source->isIntegerType() || !Target->isIntegerType())
3166    return;
3167
3168  if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)
3169           == Expr::NPCK_GNUNull) && Target->isIntegerType()) {
3170    S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer)
3171        << E->getSourceRange() << clang::SourceRange(CC);
3172    return;
3173  }
3174
3175  IntRange SourceRange = GetExprRange(S.Context, E);
3176  IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
3177
3178  if (SourceRange.Width > TargetRange.Width) {
3179    // If the source is a constant, use a default-on diagnostic.
3180    // TODO: this should happen for bitfield stores, too.
3181    llvm::APSInt Value(32);
3182    if (E->isIntegerConstantExpr(Value, S.Context)) {
3183      if (isFromSystemMacro(S, CC))
3184        return;
3185
3186      std::string PrettySourceValue = Value.toString(10);
3187      std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
3188
3189      S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant)
3190        << PrettySourceValue << PrettyTargetValue
3191        << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC);
3192      return;
3193    }
3194
3195    // People want to build with -Wshorten-64-to-32 and not -Wconversion.
3196    if (isFromSystemMacro(S, CC))
3197      return;
3198
3199    if (SourceRange.Width == 64 && TargetRange.Width == 32)
3200      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32);
3201    return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
3202  }
3203
3204  if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
3205      (!TargetRange.NonNegative && SourceRange.NonNegative &&
3206       SourceRange.Width == TargetRange.Width)) {
3207
3208    if (isFromSystemMacro(S, CC))
3209      return;
3210
3211    unsigned DiagID = diag::warn_impcast_integer_sign;
3212
3213    // Traditionally, gcc has warned about this under -Wsign-compare.
3214    // We also want to warn about it in -Wconversion.
3215    // So if -Wconversion is off, use a completely identical diagnostic
3216    // in the sign-compare group.
3217    // The conditional-checking code will
3218    if (ICContext) {
3219      DiagID = diag::warn_impcast_integer_sign_conditional;
3220      *ICContext = true;
3221    }
3222
3223    return DiagnoseImpCast(S, E, T, CC, DiagID);
3224  }
3225
3226  // Diagnose conversions between different enumeration types.
3227  // In C, we pretend that the type of an EnumConstantDecl is its enumeration
3228  // type, to give us better diagnostics.
3229  QualType SourceType = E->getType();
3230  if (!S.getLangOptions().CPlusPlus) {
3231    if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
3232      if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
3233        EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
3234        SourceType = S.Context.getTypeDeclType(Enum);
3235        Source = S.Context.getCanonicalType(SourceType).getTypePtr();
3236      }
3237  }
3238
3239  if (const EnumType *SourceEnum = Source->getAs<EnumType>())
3240    if (const EnumType *TargetEnum = Target->getAs<EnumType>())
3241      if ((SourceEnum->getDecl()->getIdentifier() ||
3242           SourceEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3243          (TargetEnum->getDecl()->getIdentifier() ||
3244           TargetEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3245          SourceEnum != TargetEnum) {
3246        if (isFromSystemMacro(S, CC))
3247          return;
3248
3249        return DiagnoseImpCast(S, E, SourceType, T, CC,
3250                               diag::warn_impcast_different_enum_types);
3251      }
3252
3253  return;
3254}
3255
3256void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T);
3257
3258void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
3259                             SourceLocation CC, bool &ICContext) {
3260  E = E->IgnoreParenImpCasts();
3261
3262  if (isa<ConditionalOperator>(E))
3263    return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T);
3264
3265  AnalyzeImplicitConversions(S, E, CC);
3266  if (E->getType() != T)
3267    return CheckImplicitConversion(S, E, T, CC, &ICContext);
3268  return;
3269}
3270
3271void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) {
3272  SourceLocation CC = E->getQuestionLoc();
3273
3274  AnalyzeImplicitConversions(S, E->getCond(), CC);
3275
3276  bool Suspicious = false;
3277  CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
3278  CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
3279
3280  // If -Wconversion would have warned about either of the candidates
3281  // for a signedness conversion to the context type...
3282  if (!Suspicious) return;
3283
3284  // ...but it's currently ignored...
3285  if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
3286                                 CC))
3287    return;
3288
3289  // ...then check whether it would have warned about either of the
3290  // candidates for a signedness conversion to the condition type.
3291  if (E->getType() == T) return;
3292
3293  Suspicious = false;
3294  CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
3295                          E->getType(), CC, &Suspicious);
3296  if (!Suspicious)
3297    CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
3298                            E->getType(), CC, &Suspicious);
3299}
3300
3301/// AnalyzeImplicitConversions - Find and report any interesting
3302/// implicit conversions in the given expression.  There are a couple
3303/// of competing diagnostics here, -Wconversion and -Wsign-compare.
3304void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
3305  QualType T = OrigE->getType();
3306  Expr *E = OrigE->IgnoreParenImpCasts();
3307
3308  // For conditional operators, we analyze the arguments as if they
3309  // were being fed directly into the output.
3310  if (isa<ConditionalOperator>(E)) {
3311    ConditionalOperator *CO = cast<ConditionalOperator>(E);
3312    CheckConditionalOperator(S, CO, T);
3313    return;
3314  }
3315
3316  // Go ahead and check any implicit conversions we might have skipped.
3317  // The non-canonical typecheck is just an optimization;
3318  // CheckImplicitConversion will filter out dead implicit conversions.
3319  if (E->getType() != T)
3320    CheckImplicitConversion(S, E, T, CC);
3321
3322  // Now continue drilling into this expression.
3323
3324  // Skip past explicit casts.
3325  if (isa<ExplicitCastExpr>(E)) {
3326    E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
3327    return AnalyzeImplicitConversions(S, E, CC);
3328  }
3329
3330  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
3331    // Do a somewhat different check with comparison operators.
3332    if (BO->isComparisonOp())
3333      return AnalyzeComparison(S, BO);
3334
3335    // And with assignments and compound assignments.
3336    if (BO->isAssignmentOp())
3337      return AnalyzeAssignment(S, BO);
3338  }
3339
3340  // These break the otherwise-useful invariant below.  Fortunately,
3341  // we don't really need to recurse into them, because any internal
3342  // expressions should have been analyzed already when they were
3343  // built into statements.
3344  if (isa<StmtExpr>(E)) return;
3345
3346  // Don't descend into unevaluated contexts.
3347  if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
3348
3349  // Now just recurse over the expression's children.
3350  CC = E->getExprLoc();
3351  for (Stmt::child_range I = E->children(); I; ++I)
3352    AnalyzeImplicitConversions(S, cast<Expr>(*I), CC);
3353}
3354
3355} // end anonymous namespace
3356
3357/// Diagnoses "dangerous" implicit conversions within the given
3358/// expression (which is a full expression).  Implements -Wconversion
3359/// and -Wsign-compare.
3360///
3361/// \param CC the "context" location of the implicit conversion, i.e.
3362///   the most location of the syntactic entity requiring the implicit
3363///   conversion
3364void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
3365  // Don't diagnose in unevaluated contexts.
3366  if (ExprEvalContexts.back().Context == Sema::Unevaluated)
3367    return;
3368
3369  // Don't diagnose for value- or type-dependent expressions.
3370  if (E->isTypeDependent() || E->isValueDependent())
3371    return;
3372
3373  // Check for array bounds violations in cases where the check isn't triggered
3374  // elsewhere for other Expr types (like BinaryOperators), e.g. when an
3375  // ArraySubscriptExpr is on the RHS of a variable initialization.
3376  CheckArrayAccess(E);
3377
3378  // This is not the right CC for (e.g.) a variable initialization.
3379  AnalyzeImplicitConversions(*this, E, CC);
3380}
3381
3382void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
3383                                       FieldDecl *BitField,
3384                                       Expr *Init) {
3385  (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
3386}
3387
3388/// CheckParmsForFunctionDef - Check that the parameters of the given
3389/// function are appropriate for the definition of a function. This
3390/// takes care of any checks that cannot be performed on the
3391/// declaration itself, e.g., that the types of each of the function
3392/// parameters are complete.
3393bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
3394                                    bool CheckParameterNames) {
3395  bool HasInvalidParm = false;
3396  for (; P != PEnd; ++P) {
3397    ParmVarDecl *Param = *P;
3398
3399    // C99 6.7.5.3p4: the parameters in a parameter type list in a
3400    // function declarator that is part of a function definition of
3401    // that function shall not have incomplete type.
3402    //
3403    // This is also C++ [dcl.fct]p6.
3404    if (!Param->isInvalidDecl() &&
3405        RequireCompleteType(Param->getLocation(), Param->getType(),
3406                               diag::err_typecheck_decl_incomplete_type)) {
3407      Param->setInvalidDecl();
3408      HasInvalidParm = true;
3409    }
3410
3411    // C99 6.9.1p5: If the declarator includes a parameter type list, the
3412    // declaration of each parameter shall include an identifier.
3413    if (CheckParameterNames &&
3414        Param->getIdentifier() == 0 &&
3415        !Param->isImplicit() &&
3416        !getLangOptions().CPlusPlus)
3417      Diag(Param->getLocation(), diag::err_parameter_name_omitted);
3418
3419    // C99 6.7.5.3p12:
3420    //   If the function declarator is not part of a definition of that
3421    //   function, parameters may have incomplete type and may use the [*]
3422    //   notation in their sequences of declarator specifiers to specify
3423    //   variable length array types.
3424    QualType PType = Param->getOriginalType();
3425    if (const ArrayType *AT = Context.getAsArrayType(PType)) {
3426      if (AT->getSizeModifier() == ArrayType::Star) {
3427        // FIXME: This diagnosic should point the the '[*]' if source-location
3428        // information is added for it.
3429        Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
3430      }
3431    }
3432  }
3433
3434  return HasInvalidParm;
3435}
3436
3437/// CheckCastAlign - Implements -Wcast-align, which warns when a
3438/// pointer cast increases the alignment requirements.
3439void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
3440  // This is actually a lot of work to potentially be doing on every
3441  // cast; don't do it if we're ignoring -Wcast_align (as is the default).
3442  if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
3443                                          TRange.getBegin())
3444        == Diagnostic::Ignored)
3445    return;
3446
3447  // Ignore dependent types.
3448  if (T->isDependentType() || Op->getType()->isDependentType())
3449    return;
3450
3451  // Require that the destination be a pointer type.
3452  const PointerType *DestPtr = T->getAs<PointerType>();
3453  if (!DestPtr) return;
3454
3455  // If the destination has alignment 1, we're done.
3456  QualType DestPointee = DestPtr->getPointeeType();
3457  if (DestPointee->isIncompleteType()) return;
3458  CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
3459  if (DestAlign.isOne()) return;
3460
3461  // Require that the source be a pointer type.
3462  const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
3463  if (!SrcPtr) return;
3464  QualType SrcPointee = SrcPtr->getPointeeType();
3465
3466  // Whitelist casts from cv void*.  We already implicitly
3467  // whitelisted casts to cv void*, since they have alignment 1.
3468  // Also whitelist casts involving incomplete types, which implicitly
3469  // includes 'void'.
3470  if (SrcPointee->isIncompleteType()) return;
3471
3472  CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
3473  if (SrcAlign >= DestAlign) return;
3474
3475  Diag(TRange.getBegin(), diag::warn_cast_align)
3476    << Op->getType() << T
3477    << static_cast<unsigned>(SrcAlign.getQuantity())
3478    << static_cast<unsigned>(DestAlign.getQuantity())
3479    << TRange << Op->getSourceRange();
3480}
3481
3482static const Type* getElementType(const Expr *BaseExpr) {
3483  const Type* EltType = BaseExpr->getType().getTypePtr();
3484  if (EltType->isAnyPointerType())
3485    return EltType->getPointeeType().getTypePtr();
3486  else if (EltType->isArrayType())
3487    return EltType->getBaseElementTypeUnsafe();
3488  return EltType;
3489}
3490
3491/// \brief Check whether this array fits the idiom of a size-one tail padded
3492/// array member of a struct.
3493///
3494/// We avoid emitting out-of-bounds access warnings for such arrays as they are
3495/// commonly used to emulate flexible arrays in C89 code.
3496static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size,
3497                                    const NamedDecl *ND) {
3498  if (Size != 1 || !ND) return false;
3499
3500  const FieldDecl *FD = dyn_cast<FieldDecl>(ND);
3501  if (!FD) return false;
3502
3503  // Don't consider sizes resulting from macro expansions or template argument
3504  // substitution to form C89 tail-padded arrays.
3505  ConstantArrayTypeLoc TL =
3506    cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc());
3507  const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr());
3508  if (!SizeExpr || SizeExpr->getExprLoc().isMacroID())
3509    return false;
3510
3511  const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext());
3512  if (!RD || !RD->isStruct())
3513    return false;
3514
3515  // See if this is the last field decl in the record.
3516  const Decl *D = FD;
3517  while ((D = D->getNextDeclInContext()))
3518    if (isa<FieldDecl>(D))
3519      return false;
3520  return true;
3521}
3522
3523void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr,
3524                            bool isSubscript, bool AllowOnePastEnd) {
3525  const Type* EffectiveType = getElementType(BaseExpr);
3526  BaseExpr = BaseExpr->IgnoreParenCasts();
3527  IndexExpr = IndexExpr->IgnoreParenCasts();
3528
3529  const ConstantArrayType *ArrayTy =
3530    Context.getAsConstantArrayType(BaseExpr->getType());
3531  if (!ArrayTy)
3532    return;
3533
3534  if (IndexExpr->isValueDependent())
3535    return;
3536  llvm::APSInt index;
3537  if (!IndexExpr->isIntegerConstantExpr(index, Context))
3538    return;
3539
3540  const NamedDecl *ND = NULL;
3541  if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
3542    ND = dyn_cast<NamedDecl>(DRE->getDecl());
3543  if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
3544    ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
3545
3546  if (index.isUnsigned() || !index.isNegative()) {
3547    llvm::APInt size = ArrayTy->getSize();
3548    if (!size.isStrictlyPositive())
3549      return;
3550
3551    const Type* BaseType = getElementType(BaseExpr);
3552    if (!isSubscript && BaseType != EffectiveType) {
3553      // Make sure we're comparing apples to apples when comparing index to size
3554      uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType);
3555      uint64_t array_typesize = Context.getTypeSize(BaseType);
3556      if (ptrarith_typesize != array_typesize) {
3557        // There's a cast to a different size type involved
3558        uint64_t ratio = array_typesize / ptrarith_typesize;
3559        // TODO: Be smarter about handling cases where array_typesize is not a
3560        // multiple of ptrarith_typesize
3561        if (ptrarith_typesize * ratio == array_typesize)
3562          size *= llvm::APInt(size.getBitWidth(), ratio);
3563      }
3564    }
3565
3566    if (size.getBitWidth() > index.getBitWidth())
3567      index = index.sext(size.getBitWidth());
3568    else if (size.getBitWidth() < index.getBitWidth())
3569      size = size.sext(index.getBitWidth());
3570
3571    // For array subscripting the index must be less than size, but for pointer
3572    // arithmetic also allow the index (offset) to be equal to size since
3573    // computing the next address after the end of the array is legal and
3574    // commonly done e.g. in C++ iterators and range-based for loops.
3575    if (AllowOnePastEnd ? index.sle(size) : index.slt(size))
3576      return;
3577
3578    // Also don't warn for arrays of size 1 which are members of some
3579    // structure. These are often used to approximate flexible arrays in C89
3580    // code.
3581    if (IsTailPaddedMemberArray(*this, size, ND))
3582      return;
3583
3584    unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds;
3585    if (isSubscript)
3586      DiagID = diag::warn_array_index_exceeds_bounds;
3587
3588    DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
3589                        PDiag(DiagID) << index.toString(10, true)
3590                          << size.toString(10, true)
3591                          << (unsigned)size.getLimitedValue(~0U)
3592                          << IndexExpr->getSourceRange());
3593  } else {
3594    unsigned DiagID = diag::warn_array_index_precedes_bounds;
3595    if (!isSubscript) {
3596      DiagID = diag::warn_ptr_arith_precedes_bounds;
3597      if (index.isNegative()) index = -index;
3598    }
3599
3600    DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
3601                        PDiag(DiagID) << index.toString(10, true)
3602                          << IndexExpr->getSourceRange());
3603  }
3604
3605  if (ND)
3606    DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
3607                        PDiag(diag::note_array_index_out_of_bounds)
3608                          << ND->getDeclName());
3609}
3610
3611void Sema::CheckArrayAccess(const Expr *expr) {
3612  int AllowOnePastEnd = 0;
3613  while (expr) {
3614    expr = expr->IgnoreParenImpCasts();
3615    switch (expr->getStmtClass()) {
3616      case Stmt::ArraySubscriptExprClass: {
3617        const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr);
3618        CheckArrayAccess(ASE->getBase(), ASE->getIdx(), true,
3619                         AllowOnePastEnd > 0);
3620        return;
3621      }
3622      case Stmt::UnaryOperatorClass: {
3623        // Only unwrap the * and & unary operators
3624        const UnaryOperator *UO = cast<UnaryOperator>(expr);
3625        expr = UO->getSubExpr();
3626        switch (UO->getOpcode()) {
3627          case UO_AddrOf:
3628            AllowOnePastEnd++;
3629            break;
3630          case UO_Deref:
3631            AllowOnePastEnd--;
3632            break;
3633          default:
3634            return;
3635        }
3636        break;
3637      }
3638      case Stmt::ConditionalOperatorClass: {
3639        const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
3640        if (const Expr *lhs = cond->getLHS())
3641          CheckArrayAccess(lhs);
3642        if (const Expr *rhs = cond->getRHS())
3643          CheckArrayAccess(rhs);
3644        return;
3645      }
3646      default:
3647        return;
3648    }
3649  }
3650}
3651
3652//===--- CHECK: Objective-C retain cycles ----------------------------------//
3653
3654namespace {
3655  struct RetainCycleOwner {
3656    RetainCycleOwner() : Variable(0), Indirect(false) {}
3657    VarDecl *Variable;
3658    SourceRange Range;
3659    SourceLocation Loc;
3660    bool Indirect;
3661
3662    void setLocsFrom(Expr *e) {
3663      Loc = e->getExprLoc();
3664      Range = e->getSourceRange();
3665    }
3666  };
3667}
3668
3669/// Consider whether capturing the given variable can possibly lead to
3670/// a retain cycle.
3671static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) {
3672  // In ARC, it's captured strongly iff the variable has __strong
3673  // lifetime.  In MRR, it's captured strongly if the variable is
3674  // __block and has an appropriate type.
3675  if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
3676    return false;
3677
3678  owner.Variable = var;
3679  owner.setLocsFrom(ref);
3680  return true;
3681}
3682
3683static bool findRetainCycleOwner(Expr *e, RetainCycleOwner &owner) {
3684  while (true) {
3685    e = e->IgnoreParens();
3686    if (CastExpr *cast = dyn_cast<CastExpr>(e)) {
3687      switch (cast->getCastKind()) {
3688      case CK_BitCast:
3689      case CK_LValueBitCast:
3690      case CK_LValueToRValue:
3691      case CK_ObjCReclaimReturnedObject:
3692        e = cast->getSubExpr();
3693        continue;
3694
3695      case CK_GetObjCProperty: {
3696        // Bail out if this isn't a strong explicit property.
3697        const ObjCPropertyRefExpr *pre = cast->getSubExpr()->getObjCProperty();
3698        if (pre->isImplicitProperty()) return false;
3699        ObjCPropertyDecl *property = pre->getExplicitProperty();
3700        if (!(property->getPropertyAttributes() &
3701              (ObjCPropertyDecl::OBJC_PR_retain |
3702               ObjCPropertyDecl::OBJC_PR_copy |
3703               ObjCPropertyDecl::OBJC_PR_strong)) &&
3704            !(property->getPropertyIvarDecl() &&
3705              property->getPropertyIvarDecl()->getType()
3706                .getObjCLifetime() == Qualifiers::OCL_Strong))
3707          return false;
3708
3709        owner.Indirect = true;
3710        e = const_cast<Expr*>(pre->getBase());
3711        continue;
3712      }
3713
3714      default:
3715        return false;
3716      }
3717    }
3718
3719    if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) {
3720      ObjCIvarDecl *ivar = ref->getDecl();
3721      if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
3722        return false;
3723
3724      // Try to find a retain cycle in the base.
3725      if (!findRetainCycleOwner(ref->getBase(), owner))
3726        return false;
3727
3728      if (ref->isFreeIvar()) owner.setLocsFrom(ref);
3729      owner.Indirect = true;
3730      return true;
3731    }
3732
3733    if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) {
3734      VarDecl *var = dyn_cast<VarDecl>(ref->getDecl());
3735      if (!var) return false;
3736      return considerVariable(var, ref, owner);
3737    }
3738
3739    if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) {
3740      owner.Variable = ref->getDecl();
3741      owner.setLocsFrom(ref);
3742      return true;
3743    }
3744
3745    if (MemberExpr *member = dyn_cast<MemberExpr>(e)) {
3746      if (member->isArrow()) return false;
3747
3748      // Don't count this as an indirect ownership.
3749      e = member->getBase();
3750      continue;
3751    }
3752
3753    // Array ivars?
3754
3755    return false;
3756  }
3757}
3758
3759namespace {
3760  struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> {
3761    FindCaptureVisitor(ASTContext &Context, VarDecl *variable)
3762      : EvaluatedExprVisitor<FindCaptureVisitor>(Context),
3763        Variable(variable), Capturer(0) {}
3764
3765    VarDecl *Variable;
3766    Expr *Capturer;
3767
3768    void VisitDeclRefExpr(DeclRefExpr *ref) {
3769      if (ref->getDecl() == Variable && !Capturer)
3770        Capturer = ref;
3771    }
3772
3773    void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) {
3774      if (ref->getDecl() == Variable && !Capturer)
3775        Capturer = ref;
3776    }
3777
3778    void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) {
3779      if (Capturer) return;
3780      Visit(ref->getBase());
3781      if (Capturer && ref->isFreeIvar())
3782        Capturer = ref;
3783    }
3784
3785    void VisitBlockExpr(BlockExpr *block) {
3786      // Look inside nested blocks
3787      if (block->getBlockDecl()->capturesVariable(Variable))
3788        Visit(block->getBlockDecl()->getBody());
3789    }
3790  };
3791}
3792
3793/// Check whether the given argument is a block which captures a
3794/// variable.
3795static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) {
3796  assert(owner.Variable && owner.Loc.isValid());
3797
3798  e = e->IgnoreParenCasts();
3799  BlockExpr *block = dyn_cast<BlockExpr>(e);
3800  if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable))
3801    return 0;
3802
3803  FindCaptureVisitor visitor(S.Context, owner.Variable);
3804  visitor.Visit(block->getBlockDecl()->getBody());
3805  return visitor.Capturer;
3806}
3807
3808static void diagnoseRetainCycle(Sema &S, Expr *capturer,
3809                                RetainCycleOwner &owner) {
3810  assert(capturer);
3811  assert(owner.Variable && owner.Loc.isValid());
3812
3813  S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle)
3814    << owner.Variable << capturer->getSourceRange();
3815  S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner)
3816    << owner.Indirect << owner.Range;
3817}
3818
3819/// Check for a keyword selector that starts with the word 'add' or
3820/// 'set'.
3821static bool isSetterLikeSelector(Selector sel) {
3822  if (sel.isUnarySelector()) return false;
3823
3824  StringRef str = sel.getNameForSlot(0);
3825  while (!str.empty() && str.front() == '_') str = str.substr(1);
3826  if (str.startswith("set") || str.startswith("add"))
3827    str = str.substr(3);
3828  else
3829    return false;
3830
3831  if (str.empty()) return true;
3832  return !islower(str.front());
3833}
3834
3835/// Check a message send to see if it's likely to cause a retain cycle.
3836void Sema::checkRetainCycles(ObjCMessageExpr *msg) {
3837  // Only check instance methods whose selector looks like a setter.
3838  if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector()))
3839    return;
3840
3841  // Try to find a variable that the receiver is strongly owned by.
3842  RetainCycleOwner owner;
3843  if (msg->getReceiverKind() == ObjCMessageExpr::Instance) {
3844    if (!findRetainCycleOwner(msg->getInstanceReceiver(), owner))
3845      return;
3846  } else {
3847    assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance);
3848    owner.Variable = getCurMethodDecl()->getSelfDecl();
3849    owner.Loc = msg->getSuperLoc();
3850    owner.Range = msg->getSuperLoc();
3851  }
3852
3853  // Check whether the receiver is captured by any of the arguments.
3854  for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i)
3855    if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner))
3856      return diagnoseRetainCycle(*this, capturer, owner);
3857}
3858
3859/// Check a property assign to see if it's likely to cause a retain cycle.
3860void Sema::checkRetainCycles(Expr *receiver, Expr *argument) {
3861  RetainCycleOwner owner;
3862  if (!findRetainCycleOwner(receiver, owner))
3863    return;
3864
3865  if (Expr *capturer = findCapturingExpr(*this, argument, owner))
3866    diagnoseRetainCycle(*this, capturer, owner);
3867}
3868
3869bool Sema::checkUnsafeAssigns(SourceLocation Loc,
3870                              QualType LHS, Expr *RHS) {
3871  Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime();
3872  if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone)
3873    return false;
3874  // strip off any implicit cast added to get to the one arc-specific
3875  while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
3876    if (cast->getCastKind() == CK_ObjCConsumeObject) {
3877      Diag(Loc, diag::warn_arc_retained_assign)
3878        << (LT == Qualifiers::OCL_ExplicitNone)
3879        << RHS->getSourceRange();
3880      return true;
3881    }
3882    RHS = cast->getSubExpr();
3883  }
3884  return false;
3885}
3886
3887void Sema::checkUnsafeExprAssigns(SourceLocation Loc,
3888                              Expr *LHS, Expr *RHS) {
3889  QualType LHSType = LHS->getType();
3890  if (checkUnsafeAssigns(Loc, LHSType, RHS))
3891    return;
3892  Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime();
3893  // FIXME. Check for other life times.
3894  if (LT != Qualifiers::OCL_None)
3895    return;
3896
3897  if (ObjCPropertyRefExpr *PRE = dyn_cast<ObjCPropertyRefExpr>(LHS)) {
3898    if (PRE->isImplicitProperty())
3899      return;
3900    const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
3901    if (!PD)
3902      return;
3903
3904    unsigned Attributes = PD->getPropertyAttributes();
3905    if (Attributes & ObjCPropertyDecl::OBJC_PR_assign)
3906      while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
3907        if (cast->getCastKind() == CK_ObjCConsumeObject) {
3908          Diag(Loc, diag::warn_arc_retained_property_assign)
3909          << RHS->getSourceRange();
3910          return;
3911        }
3912        RHS = cast->getSubExpr();
3913      }
3914  }
3915}
3916