SemaChecking.cpp revision ac1303eca6cbe3e623fb5ec6fe7ec184ef4b0dfa
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/Initialization.h"
16#include "clang/Sema/Sema.h"
17#include "clang/Sema/SemaInternal.h"
18#include "clang/Sema/Initialization.h"
19#include "clang/Sema/ScopeInfo.h"
20#include "clang/Analysis/Analyses/FormatString.h"
21#include "clang/AST/ASTContext.h"
22#include "clang/AST/CharUnits.h"
23#include "clang/AST/DeclCXX.h"
24#include "clang/AST/DeclObjC.h"
25#include "clang/AST/ExprCXX.h"
26#include "clang/AST/ExprObjC.h"
27#include "clang/AST/EvaluatedExprVisitor.h"
28#include "clang/AST/DeclObjC.h"
29#include "clang/AST/StmtCXX.h"
30#include "clang/AST/StmtObjC.h"
31#include "clang/Lex/Preprocessor.h"
32#include "llvm/ADT/BitVector.h"
33#include "llvm/ADT/SmallString.h"
34#include "llvm/ADT/STLExtras.h"
35#include "llvm/Support/raw_ostream.h"
36#include "clang/Basic/TargetBuiltins.h"
37#include "clang/Basic/TargetInfo.h"
38#include "clang/Basic/ConvertUTF.h"
39#include <limits>
40using namespace clang;
41using namespace sema;
42
43SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL,
44                                                    unsigned ByteNo) const {
45  return SL->getLocationOfByte(ByteNo, PP.getSourceManager(),
46                               PP.getLangOptions(), PP.getTargetInfo());
47}
48
49/// Checks that a call expression's argument count is the desired number.
50/// This is useful when doing custom type-checking.  Returns true on error.
51static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) {
52  unsigned argCount = call->getNumArgs();
53  if (argCount == desiredArgCount) return false;
54
55  if (argCount < desiredArgCount)
56    return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args)
57        << 0 /*function call*/ << desiredArgCount << argCount
58        << call->getSourceRange();
59
60  // Highlight all the excess arguments.
61  SourceRange range(call->getArg(desiredArgCount)->getLocStart(),
62                    call->getArg(argCount - 1)->getLocEnd());
63
64  return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args)
65    << 0 /*function call*/ << desiredArgCount << argCount
66    << call->getArg(1)->getSourceRange();
67}
68
69/// CheckBuiltinAnnotationString - Checks that string argument to the builtin
70/// annotation is a non wide string literal.
71static bool CheckBuiltinAnnotationString(Sema &S, Expr *Arg) {
72  Arg = Arg->IgnoreParenCasts();
73  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
74  if (!Literal || !Literal->isAscii()) {
75    S.Diag(Arg->getLocStart(), diag::err_builtin_annotation_not_string_constant)
76      << Arg->getSourceRange();
77    return true;
78  }
79  return false;
80}
81
82ExprResult
83Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
84  ExprResult TheCallResult(Owned(TheCall));
85
86  // Find out if any arguments are required to be integer constant expressions.
87  unsigned ICEArguments = 0;
88  ASTContext::GetBuiltinTypeError Error;
89  Context.GetBuiltinType(BuiltinID, Error, &ICEArguments);
90  if (Error != ASTContext::GE_None)
91    ICEArguments = 0;  // Don't diagnose previously diagnosed errors.
92
93  // If any arguments are required to be ICE's, check and diagnose.
94  for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) {
95    // Skip arguments not required to be ICE's.
96    if ((ICEArguments & (1 << ArgNo)) == 0) continue;
97
98    llvm::APSInt Result;
99    if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
100      return true;
101    ICEArguments &= ~(1 << ArgNo);
102  }
103
104  switch (BuiltinID) {
105  case Builtin::BI__builtin___CFStringMakeConstantString:
106    assert(TheCall->getNumArgs() == 1 &&
107           "Wrong # arguments to builtin CFStringMakeConstantString");
108    if (CheckObjCString(TheCall->getArg(0)))
109      return ExprError();
110    break;
111  case Builtin::BI__builtin_stdarg_start:
112  case Builtin::BI__builtin_va_start:
113    if (SemaBuiltinVAStart(TheCall))
114      return ExprError();
115    break;
116  case Builtin::BI__builtin_isgreater:
117  case Builtin::BI__builtin_isgreaterequal:
118  case Builtin::BI__builtin_isless:
119  case Builtin::BI__builtin_islessequal:
120  case Builtin::BI__builtin_islessgreater:
121  case Builtin::BI__builtin_isunordered:
122    if (SemaBuiltinUnorderedCompare(TheCall))
123      return ExprError();
124    break;
125  case Builtin::BI__builtin_fpclassify:
126    if (SemaBuiltinFPClassification(TheCall, 6))
127      return ExprError();
128    break;
129  case Builtin::BI__builtin_isfinite:
130  case Builtin::BI__builtin_isinf:
131  case Builtin::BI__builtin_isinf_sign:
132  case Builtin::BI__builtin_isnan:
133  case Builtin::BI__builtin_isnormal:
134    if (SemaBuiltinFPClassification(TheCall, 1))
135      return ExprError();
136    break;
137  case Builtin::BI__builtin_shufflevector:
138    return SemaBuiltinShuffleVector(TheCall);
139    // TheCall will be freed by the smart pointer here, but that's fine, since
140    // SemaBuiltinShuffleVector guts it, but then doesn't release it.
141  case Builtin::BI__builtin_prefetch:
142    if (SemaBuiltinPrefetch(TheCall))
143      return ExprError();
144    break;
145  case Builtin::BI__builtin_object_size:
146    if (SemaBuiltinObjectSize(TheCall))
147      return ExprError();
148    break;
149  case Builtin::BI__builtin_longjmp:
150    if (SemaBuiltinLongjmp(TheCall))
151      return ExprError();
152    break;
153
154  case Builtin::BI__builtin_classify_type:
155    if (checkArgCount(*this, TheCall, 1)) return true;
156    TheCall->setType(Context.IntTy);
157    break;
158  case Builtin::BI__builtin_constant_p:
159    if (checkArgCount(*this, TheCall, 1)) return true;
160    TheCall->setType(Context.IntTy);
161    break;
162  case Builtin::BI__sync_fetch_and_add:
163  case Builtin::BI__sync_fetch_and_add_1:
164  case Builtin::BI__sync_fetch_and_add_2:
165  case Builtin::BI__sync_fetch_and_add_4:
166  case Builtin::BI__sync_fetch_and_add_8:
167  case Builtin::BI__sync_fetch_and_add_16:
168  case Builtin::BI__sync_fetch_and_sub:
169  case Builtin::BI__sync_fetch_and_sub_1:
170  case Builtin::BI__sync_fetch_and_sub_2:
171  case Builtin::BI__sync_fetch_and_sub_4:
172  case Builtin::BI__sync_fetch_and_sub_8:
173  case Builtin::BI__sync_fetch_and_sub_16:
174  case Builtin::BI__sync_fetch_and_or:
175  case Builtin::BI__sync_fetch_and_or_1:
176  case Builtin::BI__sync_fetch_and_or_2:
177  case Builtin::BI__sync_fetch_and_or_4:
178  case Builtin::BI__sync_fetch_and_or_8:
179  case Builtin::BI__sync_fetch_and_or_16:
180  case Builtin::BI__sync_fetch_and_and:
181  case Builtin::BI__sync_fetch_and_and_1:
182  case Builtin::BI__sync_fetch_and_and_2:
183  case Builtin::BI__sync_fetch_and_and_4:
184  case Builtin::BI__sync_fetch_and_and_8:
185  case Builtin::BI__sync_fetch_and_and_16:
186  case Builtin::BI__sync_fetch_and_xor:
187  case Builtin::BI__sync_fetch_and_xor_1:
188  case Builtin::BI__sync_fetch_and_xor_2:
189  case Builtin::BI__sync_fetch_and_xor_4:
190  case Builtin::BI__sync_fetch_and_xor_8:
191  case Builtin::BI__sync_fetch_and_xor_16:
192  case Builtin::BI__sync_add_and_fetch:
193  case Builtin::BI__sync_add_and_fetch_1:
194  case Builtin::BI__sync_add_and_fetch_2:
195  case Builtin::BI__sync_add_and_fetch_4:
196  case Builtin::BI__sync_add_and_fetch_8:
197  case Builtin::BI__sync_add_and_fetch_16:
198  case Builtin::BI__sync_sub_and_fetch:
199  case Builtin::BI__sync_sub_and_fetch_1:
200  case Builtin::BI__sync_sub_and_fetch_2:
201  case Builtin::BI__sync_sub_and_fetch_4:
202  case Builtin::BI__sync_sub_and_fetch_8:
203  case Builtin::BI__sync_sub_and_fetch_16:
204  case Builtin::BI__sync_and_and_fetch:
205  case Builtin::BI__sync_and_and_fetch_1:
206  case Builtin::BI__sync_and_and_fetch_2:
207  case Builtin::BI__sync_and_and_fetch_4:
208  case Builtin::BI__sync_and_and_fetch_8:
209  case Builtin::BI__sync_and_and_fetch_16:
210  case Builtin::BI__sync_or_and_fetch:
211  case Builtin::BI__sync_or_and_fetch_1:
212  case Builtin::BI__sync_or_and_fetch_2:
213  case Builtin::BI__sync_or_and_fetch_4:
214  case Builtin::BI__sync_or_and_fetch_8:
215  case Builtin::BI__sync_or_and_fetch_16:
216  case Builtin::BI__sync_xor_and_fetch:
217  case Builtin::BI__sync_xor_and_fetch_1:
218  case Builtin::BI__sync_xor_and_fetch_2:
219  case Builtin::BI__sync_xor_and_fetch_4:
220  case Builtin::BI__sync_xor_and_fetch_8:
221  case Builtin::BI__sync_xor_and_fetch_16:
222  case Builtin::BI__sync_val_compare_and_swap:
223  case Builtin::BI__sync_val_compare_and_swap_1:
224  case Builtin::BI__sync_val_compare_and_swap_2:
225  case Builtin::BI__sync_val_compare_and_swap_4:
226  case Builtin::BI__sync_val_compare_and_swap_8:
227  case Builtin::BI__sync_val_compare_and_swap_16:
228  case Builtin::BI__sync_bool_compare_and_swap:
229  case Builtin::BI__sync_bool_compare_and_swap_1:
230  case Builtin::BI__sync_bool_compare_and_swap_2:
231  case Builtin::BI__sync_bool_compare_and_swap_4:
232  case Builtin::BI__sync_bool_compare_and_swap_8:
233  case Builtin::BI__sync_bool_compare_and_swap_16:
234  case Builtin::BI__sync_lock_test_and_set:
235  case Builtin::BI__sync_lock_test_and_set_1:
236  case Builtin::BI__sync_lock_test_and_set_2:
237  case Builtin::BI__sync_lock_test_and_set_4:
238  case Builtin::BI__sync_lock_test_and_set_8:
239  case Builtin::BI__sync_lock_test_and_set_16:
240  case Builtin::BI__sync_lock_release:
241  case Builtin::BI__sync_lock_release_1:
242  case Builtin::BI__sync_lock_release_2:
243  case Builtin::BI__sync_lock_release_4:
244  case Builtin::BI__sync_lock_release_8:
245  case Builtin::BI__sync_lock_release_16:
246  case Builtin::BI__sync_swap:
247  case Builtin::BI__sync_swap_1:
248  case Builtin::BI__sync_swap_2:
249  case Builtin::BI__sync_swap_4:
250  case Builtin::BI__sync_swap_8:
251  case Builtin::BI__sync_swap_16:
252    return SemaBuiltinAtomicOverloaded(move(TheCallResult));
253  case Builtin::BI__atomic_load:
254    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Load);
255  case Builtin::BI__atomic_store:
256    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Store);
257  case Builtin::BI__atomic_init:
258    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Init);
259  case Builtin::BI__atomic_exchange:
260    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xchg);
261  case Builtin::BI__atomic_compare_exchange_strong:
262    return SemaAtomicOpsOverloaded(move(TheCallResult),
263                                   AtomicExpr::CmpXchgStrong);
264  case Builtin::BI__atomic_compare_exchange_weak:
265    return SemaAtomicOpsOverloaded(move(TheCallResult),
266                                   AtomicExpr::CmpXchgWeak);
267  case Builtin::BI__atomic_fetch_add:
268    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Add);
269  case Builtin::BI__atomic_fetch_sub:
270    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Sub);
271  case Builtin::BI__atomic_fetch_and:
272    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::And);
273  case Builtin::BI__atomic_fetch_or:
274    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Or);
275  case Builtin::BI__atomic_fetch_xor:
276    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xor);
277  case Builtin::BI__builtin_annotation:
278    if (CheckBuiltinAnnotationString(*this, TheCall->getArg(1)))
279      return ExprError();
280    break;
281  }
282
283  // Since the target specific builtins for each arch overlap, only check those
284  // of the arch we are compiling for.
285  if (BuiltinID >= Builtin::FirstTSBuiltin) {
286    switch (Context.getTargetInfo().getTriple().getArch()) {
287      case llvm::Triple::arm:
288      case llvm::Triple::thumb:
289        if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall))
290          return ExprError();
291        break;
292      default:
293        break;
294    }
295  }
296
297  return move(TheCallResult);
298}
299
300// Get the valid immediate range for the specified NEON type code.
301static unsigned RFT(unsigned t, bool shift = false) {
302  NeonTypeFlags Type(t);
303  int IsQuad = Type.isQuad();
304  switch (Type.getEltType()) {
305  case NeonTypeFlags::Int8:
306  case NeonTypeFlags::Poly8:
307    return shift ? 7 : (8 << IsQuad) - 1;
308  case NeonTypeFlags::Int16:
309  case NeonTypeFlags::Poly16:
310    return shift ? 15 : (4 << IsQuad) - 1;
311  case NeonTypeFlags::Int32:
312    return shift ? 31 : (2 << IsQuad) - 1;
313  case NeonTypeFlags::Int64:
314    return shift ? 63 : (1 << IsQuad) - 1;
315  case NeonTypeFlags::Float16:
316    assert(!shift && "cannot shift float types!");
317    return (4 << IsQuad) - 1;
318  case NeonTypeFlags::Float32:
319    assert(!shift && "cannot shift float types!");
320    return (2 << IsQuad) - 1;
321  }
322  llvm_unreachable("Invalid NeonTypeFlag!");
323}
324
325/// getNeonEltType - Return the QualType corresponding to the elements of
326/// the vector type specified by the NeonTypeFlags.  This is used to check
327/// the pointer arguments for Neon load/store intrinsics.
328static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) {
329  switch (Flags.getEltType()) {
330  case NeonTypeFlags::Int8:
331    return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy;
332  case NeonTypeFlags::Int16:
333    return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy;
334  case NeonTypeFlags::Int32:
335    return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy;
336  case NeonTypeFlags::Int64:
337    return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy;
338  case NeonTypeFlags::Poly8:
339    return Context.SignedCharTy;
340  case NeonTypeFlags::Poly16:
341    return Context.ShortTy;
342  case NeonTypeFlags::Float16:
343    return Context.UnsignedShortTy;
344  case NeonTypeFlags::Float32:
345    return Context.FloatTy;
346  }
347  llvm_unreachable("Invalid NeonTypeFlag!");
348}
349
350bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
351  llvm::APSInt Result;
352
353  unsigned mask = 0;
354  unsigned TV = 0;
355  int PtrArgNum = -1;
356  bool HasConstPtr = false;
357  switch (BuiltinID) {
358#define GET_NEON_OVERLOAD_CHECK
359#include "clang/Basic/arm_neon.inc"
360#undef GET_NEON_OVERLOAD_CHECK
361  }
362
363  // For NEON intrinsics which are overloaded on vector element type, validate
364  // the immediate which specifies which variant to emit.
365  unsigned ImmArg = TheCall->getNumArgs()-1;
366  if (mask) {
367    if (SemaBuiltinConstantArg(TheCall, ImmArg, Result))
368      return true;
369
370    TV = Result.getLimitedValue(64);
371    if ((TV > 63) || (mask & (1 << TV)) == 0)
372      return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code)
373        << TheCall->getArg(ImmArg)->getSourceRange();
374  }
375
376  if (PtrArgNum >= 0) {
377    // Check that pointer arguments have the specified type.
378    Expr *Arg = TheCall->getArg(PtrArgNum);
379    if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg))
380      Arg = ICE->getSubExpr();
381    ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg);
382    QualType RHSTy = RHS.get()->getType();
383    QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context);
384    if (HasConstPtr)
385      EltTy = EltTy.withConst();
386    QualType LHSTy = Context.getPointerType(EltTy);
387    AssignConvertType ConvTy;
388    ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS);
389    if (RHS.isInvalid())
390      return true;
391    if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy,
392                                 RHS.get(), AA_Assigning))
393      return true;
394  }
395
396  // For NEON intrinsics which take an immediate value as part of the
397  // instruction, range check them here.
398  unsigned i = 0, l = 0, u = 0;
399  switch (BuiltinID) {
400  default: return false;
401  case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break;
402  case ARM::BI__builtin_arm_usat: i = 1; u = 31; break;
403  case ARM::BI__builtin_arm_vcvtr_f:
404  case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break;
405#define GET_NEON_IMMEDIATE_CHECK
406#include "clang/Basic/arm_neon.inc"
407#undef GET_NEON_IMMEDIATE_CHECK
408  };
409
410  // Check that the immediate argument is actually a constant.
411  if (SemaBuiltinConstantArg(TheCall, i, Result))
412    return true;
413
414  // Range check against the upper/lower values for this isntruction.
415  unsigned Val = Result.getZExtValue();
416  if (Val < l || Val > (u + l))
417    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
418      << l << u+l << TheCall->getArg(i)->getSourceRange();
419
420  // FIXME: VFP Intrinsics should error if VFP not present.
421  return false;
422}
423
424/// CheckFunctionCall - Check a direct function call for various correctness
425/// and safety properties not strictly enforced by the C type system.
426bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) {
427  // Get the IdentifierInfo* for the called function.
428  IdentifierInfo *FnInfo = FDecl->getIdentifier();
429
430  // None of the checks below are needed for functions that don't have
431  // simple names (e.g., C++ conversion functions).
432  if (!FnInfo)
433    return false;
434
435  // FIXME: This mechanism should be abstracted to be less fragile and
436  // more efficient. For example, just map function ids to custom
437  // handlers.
438
439  // Printf and scanf checking.
440  for (specific_attr_iterator<FormatAttr>
441         i = FDecl->specific_attr_begin<FormatAttr>(),
442         e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) {
443    CheckFormatArguments(*i, TheCall);
444  }
445
446  for (specific_attr_iterator<NonNullAttr>
447         i = FDecl->specific_attr_begin<NonNullAttr>(),
448         e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) {
449    CheckNonNullArguments(*i, TheCall->getArgs(),
450                          TheCall->getCallee()->getLocStart());
451  }
452
453  unsigned CMId = FDecl->getMemoryFunctionKind();
454  if (CMId == 0)
455    return false;
456
457  // Handle memory setting and copying functions.
458  if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat)
459    CheckStrlcpycatArguments(TheCall, FnInfo);
460  else if (CMId == Builtin::BIstrncat)
461    CheckStrncatArguments(TheCall, FnInfo);
462  else
463    CheckMemaccessArguments(TheCall, CMId, FnInfo);
464
465  return false;
466}
467
468bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac,
469                               Expr **Args, unsigned NumArgs) {
470  for (specific_attr_iterator<FormatAttr>
471       i = Method->specific_attr_begin<FormatAttr>(),
472       e = Method->specific_attr_end<FormatAttr>(); i != e ; ++i) {
473
474    CheckFormatArguments(*i, Args, NumArgs, false, lbrac,
475                         Method->getSourceRange());
476  }
477
478  // diagnose nonnull arguments.
479  for (specific_attr_iterator<NonNullAttr>
480       i = Method->specific_attr_begin<NonNullAttr>(),
481       e = Method->specific_attr_end<NonNullAttr>(); i != e; ++i) {
482    CheckNonNullArguments(*i, Args, lbrac);
483  }
484
485  return false;
486}
487
488bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
489  const VarDecl *V = dyn_cast<VarDecl>(NDecl);
490  if (!V)
491    return false;
492
493  QualType Ty = V->getType();
494  if (!Ty->isBlockPointerType())
495    return false;
496
497  // format string checking.
498  for (specific_attr_iterator<FormatAttr>
499       i = NDecl->specific_attr_begin<FormatAttr>(),
500       e = NDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) {
501    CheckFormatArguments(*i, TheCall);
502  }
503
504  return false;
505}
506
507ExprResult
508Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, AtomicExpr::AtomicOp Op) {
509  CallExpr *TheCall = cast<CallExpr>(TheCallResult.get());
510  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
511
512  // All these operations take one of the following four forms:
513  // T   __atomic_load(_Atomic(T)*, int)                              (loads)
514  // T*  __atomic_add(_Atomic(T*)*, ptrdiff_t, int)         (pointer add/sub)
515  // int __atomic_compare_exchange_strong(_Atomic(T)*, T*, T, int, int)
516  //                                                                (cmpxchg)
517  // T   __atomic_exchange(_Atomic(T)*, T, int)             (everything else)
518  // where T is an appropriate type, and the int paremeterss are for orderings.
519  unsigned NumVals = 1;
520  unsigned NumOrders = 1;
521  if (Op == AtomicExpr::Load) {
522    NumVals = 0;
523  } else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) {
524    NumVals = 2;
525    NumOrders = 2;
526  }
527  if (Op == AtomicExpr::Init)
528    NumOrders = 0;
529
530  if (TheCall->getNumArgs() < NumVals+NumOrders+1) {
531    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
532      << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs()
533      << TheCall->getCallee()->getSourceRange();
534    return ExprError();
535  } else if (TheCall->getNumArgs() > NumVals+NumOrders+1) {
536    Diag(TheCall->getArg(NumVals+NumOrders+1)->getLocStart(),
537         diag::err_typecheck_call_too_many_args)
538      << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs()
539      << TheCall->getCallee()->getSourceRange();
540    return ExprError();
541  }
542
543  // Inspect the first argument of the atomic operation.  This should always be
544  // a pointer to an _Atomic type.
545  Expr *Ptr = TheCall->getArg(0);
546  Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get();
547  const PointerType *pointerType = Ptr->getType()->getAs<PointerType>();
548  if (!pointerType) {
549    Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic)
550      << Ptr->getType() << Ptr->getSourceRange();
551    return ExprError();
552  }
553
554  QualType AtomTy = pointerType->getPointeeType();
555  if (!AtomTy->isAtomicType()) {
556    Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic)
557      << Ptr->getType() << Ptr->getSourceRange();
558    return ExprError();
559  }
560  QualType ValType = AtomTy->getAs<AtomicType>()->getValueType();
561
562  if ((Op == AtomicExpr::Add || Op == AtomicExpr::Sub) &&
563      !ValType->isIntegerType() && !ValType->isPointerType()) {
564    Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr)
565      << Ptr->getType() << Ptr->getSourceRange();
566    return ExprError();
567  }
568
569  if (!ValType->isIntegerType() &&
570      (Op == AtomicExpr::And || Op == AtomicExpr::Or || Op == AtomicExpr::Xor)){
571    Diag(DRE->getLocStart(), diag::err_atomic_op_logical_needs_atomic_int)
572      << Ptr->getType() << Ptr->getSourceRange();
573    return ExprError();
574  }
575
576  switch (ValType.getObjCLifetime()) {
577  case Qualifiers::OCL_None:
578  case Qualifiers::OCL_ExplicitNone:
579    // okay
580    break;
581
582  case Qualifiers::OCL_Weak:
583  case Qualifiers::OCL_Strong:
584  case Qualifiers::OCL_Autoreleasing:
585    Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
586      << ValType << Ptr->getSourceRange();
587    return ExprError();
588  }
589
590  QualType ResultType = ValType;
591  if (Op == AtomicExpr::Store || Op == AtomicExpr::Init)
592    ResultType = Context.VoidTy;
593  else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong)
594    ResultType = Context.BoolTy;
595
596  // The first argument --- the pointer --- has a fixed type; we
597  // deduce the types of the rest of the arguments accordingly.  Walk
598  // the remaining arguments, converting them to the deduced value type.
599  for (unsigned i = 1; i != NumVals+NumOrders+1; ++i) {
600    ExprResult Arg = TheCall->getArg(i);
601    QualType Ty;
602    if (i < NumVals+1) {
603      // The second argument to a cmpxchg is a pointer to the data which will
604      // be exchanged. The second argument to a pointer add/subtract is the
605      // amount to add/subtract, which must be a ptrdiff_t.  The third
606      // argument to a cmpxchg and the second argument in all other cases
607      // is the type of the value.
608      if (i == 1 && (Op == AtomicExpr::CmpXchgWeak ||
609                     Op == AtomicExpr::CmpXchgStrong))
610         Ty = Context.getPointerType(ValType.getUnqualifiedType());
611      else if (!ValType->isIntegerType() &&
612               (Op == AtomicExpr::Add || Op == AtomicExpr::Sub))
613        Ty = Context.getPointerDiffType();
614      else
615        Ty = ValType;
616    } else {
617      // The order(s) are always converted to int.
618      Ty = Context.IntTy;
619    }
620    InitializedEntity Entity =
621        InitializedEntity::InitializeParameter(Context, Ty, false);
622    Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
623    if (Arg.isInvalid())
624      return true;
625    TheCall->setArg(i, Arg.get());
626  }
627
628  SmallVector<Expr*, 5> SubExprs;
629  SubExprs.push_back(Ptr);
630  if (Op == AtomicExpr::Load) {
631    SubExprs.push_back(TheCall->getArg(1)); // Order
632  } else if (Op == AtomicExpr::Init) {
633    SubExprs.push_back(TheCall->getArg(1)); // Val1
634  } else if (Op != AtomicExpr::CmpXchgWeak && Op != AtomicExpr::CmpXchgStrong) {
635    SubExprs.push_back(TheCall->getArg(2)); // Order
636    SubExprs.push_back(TheCall->getArg(1)); // Val1
637  } else {
638    SubExprs.push_back(TheCall->getArg(3)); // Order
639    SubExprs.push_back(TheCall->getArg(1)); // Val1
640    SubExprs.push_back(TheCall->getArg(2)); // Val2
641    SubExprs.push_back(TheCall->getArg(4)); // OrderFail
642  }
643
644  return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(),
645                                        SubExprs.data(), SubExprs.size(),
646                                        ResultType, Op,
647                                        TheCall->getRParenLoc()));
648}
649
650
651/// checkBuiltinArgument - Given a call to a builtin function, perform
652/// normal type-checking on the given argument, updating the call in
653/// place.  This is useful when a builtin function requires custom
654/// type-checking for some of its arguments but not necessarily all of
655/// them.
656///
657/// Returns true on error.
658static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) {
659  FunctionDecl *Fn = E->getDirectCallee();
660  assert(Fn && "builtin call without direct callee!");
661
662  ParmVarDecl *Param = Fn->getParamDecl(ArgIndex);
663  InitializedEntity Entity =
664    InitializedEntity::InitializeParameter(S.Context, Param);
665
666  ExprResult Arg = E->getArg(0);
667  Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg);
668  if (Arg.isInvalid())
669    return true;
670
671  E->setArg(ArgIndex, Arg.take());
672  return false;
673}
674
675/// SemaBuiltinAtomicOverloaded - We have a call to a function like
676/// __sync_fetch_and_add, which is an overloaded function based on the pointer
677/// type of its first argument.  The main ActOnCallExpr routines have already
678/// promoted the types of arguments because all of these calls are prototyped as
679/// void(...).
680///
681/// This function goes through and does final semantic checking for these
682/// builtins,
683ExprResult
684Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
685  CallExpr *TheCall = (CallExpr *)TheCallResult.get();
686  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
687  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
688
689  // Ensure that we have at least one argument to do type inference from.
690  if (TheCall->getNumArgs() < 1) {
691    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
692      << 0 << 1 << TheCall->getNumArgs()
693      << TheCall->getCallee()->getSourceRange();
694    return ExprError();
695  }
696
697  // Inspect the first argument of the atomic builtin.  This should always be
698  // a pointer type, whose element is an integral scalar or pointer type.
699  // Because it is a pointer type, we don't have to worry about any implicit
700  // casts here.
701  // FIXME: We don't allow floating point scalars as input.
702  Expr *FirstArg = TheCall->getArg(0);
703  ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg);
704  if (FirstArgResult.isInvalid())
705    return ExprError();
706  FirstArg = FirstArgResult.take();
707  TheCall->setArg(0, FirstArg);
708
709  const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>();
710  if (!pointerType) {
711    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
712      << FirstArg->getType() << FirstArg->getSourceRange();
713    return ExprError();
714  }
715
716  QualType ValType = pointerType->getPointeeType();
717  if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
718      !ValType->isBlockPointerType()) {
719    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
720      << FirstArg->getType() << FirstArg->getSourceRange();
721    return ExprError();
722  }
723
724  switch (ValType.getObjCLifetime()) {
725  case Qualifiers::OCL_None:
726  case Qualifiers::OCL_ExplicitNone:
727    // okay
728    break;
729
730  case Qualifiers::OCL_Weak:
731  case Qualifiers::OCL_Strong:
732  case Qualifiers::OCL_Autoreleasing:
733    Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
734      << ValType << FirstArg->getSourceRange();
735    return ExprError();
736  }
737
738  // Strip any qualifiers off ValType.
739  ValType = ValType.getUnqualifiedType();
740
741  // The majority of builtins return a value, but a few have special return
742  // types, so allow them to override appropriately below.
743  QualType ResultType = ValType;
744
745  // We need to figure out which concrete builtin this maps onto.  For example,
746  // __sync_fetch_and_add with a 2 byte object turns into
747  // __sync_fetch_and_add_2.
748#define BUILTIN_ROW(x) \
749  { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
750    Builtin::BI##x##_8, Builtin::BI##x##_16 }
751
752  static const unsigned BuiltinIndices[][5] = {
753    BUILTIN_ROW(__sync_fetch_and_add),
754    BUILTIN_ROW(__sync_fetch_and_sub),
755    BUILTIN_ROW(__sync_fetch_and_or),
756    BUILTIN_ROW(__sync_fetch_and_and),
757    BUILTIN_ROW(__sync_fetch_and_xor),
758
759    BUILTIN_ROW(__sync_add_and_fetch),
760    BUILTIN_ROW(__sync_sub_and_fetch),
761    BUILTIN_ROW(__sync_and_and_fetch),
762    BUILTIN_ROW(__sync_or_and_fetch),
763    BUILTIN_ROW(__sync_xor_and_fetch),
764
765    BUILTIN_ROW(__sync_val_compare_and_swap),
766    BUILTIN_ROW(__sync_bool_compare_and_swap),
767    BUILTIN_ROW(__sync_lock_test_and_set),
768    BUILTIN_ROW(__sync_lock_release),
769    BUILTIN_ROW(__sync_swap)
770  };
771#undef BUILTIN_ROW
772
773  // Determine the index of the size.
774  unsigned SizeIndex;
775  switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
776  case 1: SizeIndex = 0; break;
777  case 2: SizeIndex = 1; break;
778  case 4: SizeIndex = 2; break;
779  case 8: SizeIndex = 3; break;
780  case 16: SizeIndex = 4; break;
781  default:
782    Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
783      << FirstArg->getType() << FirstArg->getSourceRange();
784    return ExprError();
785  }
786
787  // Each of these builtins has one pointer argument, followed by some number of
788  // values (0, 1 or 2) followed by a potentially empty varags list of stuff
789  // that we ignore.  Find out which row of BuiltinIndices to read from as well
790  // as the number of fixed args.
791  unsigned BuiltinID = FDecl->getBuiltinID();
792  unsigned BuiltinIndex, NumFixed = 1;
793  switch (BuiltinID) {
794  default: llvm_unreachable("Unknown overloaded atomic builtin!");
795  case Builtin::BI__sync_fetch_and_add:
796  case Builtin::BI__sync_fetch_and_add_1:
797  case Builtin::BI__sync_fetch_and_add_2:
798  case Builtin::BI__sync_fetch_and_add_4:
799  case Builtin::BI__sync_fetch_and_add_8:
800  case Builtin::BI__sync_fetch_and_add_16:
801    BuiltinIndex = 0;
802    break;
803
804  case Builtin::BI__sync_fetch_and_sub:
805  case Builtin::BI__sync_fetch_and_sub_1:
806  case Builtin::BI__sync_fetch_and_sub_2:
807  case Builtin::BI__sync_fetch_and_sub_4:
808  case Builtin::BI__sync_fetch_and_sub_8:
809  case Builtin::BI__sync_fetch_and_sub_16:
810    BuiltinIndex = 1;
811    break;
812
813  case Builtin::BI__sync_fetch_and_or:
814  case Builtin::BI__sync_fetch_and_or_1:
815  case Builtin::BI__sync_fetch_and_or_2:
816  case Builtin::BI__sync_fetch_and_or_4:
817  case Builtin::BI__sync_fetch_and_or_8:
818  case Builtin::BI__sync_fetch_and_or_16:
819    BuiltinIndex = 2;
820    break;
821
822  case Builtin::BI__sync_fetch_and_and:
823  case Builtin::BI__sync_fetch_and_and_1:
824  case Builtin::BI__sync_fetch_and_and_2:
825  case Builtin::BI__sync_fetch_and_and_4:
826  case Builtin::BI__sync_fetch_and_and_8:
827  case Builtin::BI__sync_fetch_and_and_16:
828    BuiltinIndex = 3;
829    break;
830
831  case Builtin::BI__sync_fetch_and_xor:
832  case Builtin::BI__sync_fetch_and_xor_1:
833  case Builtin::BI__sync_fetch_and_xor_2:
834  case Builtin::BI__sync_fetch_and_xor_4:
835  case Builtin::BI__sync_fetch_and_xor_8:
836  case Builtin::BI__sync_fetch_and_xor_16:
837    BuiltinIndex = 4;
838    break;
839
840  case Builtin::BI__sync_add_and_fetch:
841  case Builtin::BI__sync_add_and_fetch_1:
842  case Builtin::BI__sync_add_and_fetch_2:
843  case Builtin::BI__sync_add_and_fetch_4:
844  case Builtin::BI__sync_add_and_fetch_8:
845  case Builtin::BI__sync_add_and_fetch_16:
846    BuiltinIndex = 5;
847    break;
848
849  case Builtin::BI__sync_sub_and_fetch:
850  case Builtin::BI__sync_sub_and_fetch_1:
851  case Builtin::BI__sync_sub_and_fetch_2:
852  case Builtin::BI__sync_sub_and_fetch_4:
853  case Builtin::BI__sync_sub_and_fetch_8:
854  case Builtin::BI__sync_sub_and_fetch_16:
855    BuiltinIndex = 6;
856    break;
857
858  case Builtin::BI__sync_and_and_fetch:
859  case Builtin::BI__sync_and_and_fetch_1:
860  case Builtin::BI__sync_and_and_fetch_2:
861  case Builtin::BI__sync_and_and_fetch_4:
862  case Builtin::BI__sync_and_and_fetch_8:
863  case Builtin::BI__sync_and_and_fetch_16:
864    BuiltinIndex = 7;
865    break;
866
867  case Builtin::BI__sync_or_and_fetch:
868  case Builtin::BI__sync_or_and_fetch_1:
869  case Builtin::BI__sync_or_and_fetch_2:
870  case Builtin::BI__sync_or_and_fetch_4:
871  case Builtin::BI__sync_or_and_fetch_8:
872  case Builtin::BI__sync_or_and_fetch_16:
873    BuiltinIndex = 8;
874    break;
875
876  case Builtin::BI__sync_xor_and_fetch:
877  case Builtin::BI__sync_xor_and_fetch_1:
878  case Builtin::BI__sync_xor_and_fetch_2:
879  case Builtin::BI__sync_xor_and_fetch_4:
880  case Builtin::BI__sync_xor_and_fetch_8:
881  case Builtin::BI__sync_xor_and_fetch_16:
882    BuiltinIndex = 9;
883    break;
884
885  case Builtin::BI__sync_val_compare_and_swap:
886  case Builtin::BI__sync_val_compare_and_swap_1:
887  case Builtin::BI__sync_val_compare_and_swap_2:
888  case Builtin::BI__sync_val_compare_and_swap_4:
889  case Builtin::BI__sync_val_compare_and_swap_8:
890  case Builtin::BI__sync_val_compare_and_swap_16:
891    BuiltinIndex = 10;
892    NumFixed = 2;
893    break;
894
895  case Builtin::BI__sync_bool_compare_and_swap:
896  case Builtin::BI__sync_bool_compare_and_swap_1:
897  case Builtin::BI__sync_bool_compare_and_swap_2:
898  case Builtin::BI__sync_bool_compare_and_swap_4:
899  case Builtin::BI__sync_bool_compare_and_swap_8:
900  case Builtin::BI__sync_bool_compare_and_swap_16:
901    BuiltinIndex = 11;
902    NumFixed = 2;
903    ResultType = Context.BoolTy;
904    break;
905
906  case Builtin::BI__sync_lock_test_and_set:
907  case Builtin::BI__sync_lock_test_and_set_1:
908  case Builtin::BI__sync_lock_test_and_set_2:
909  case Builtin::BI__sync_lock_test_and_set_4:
910  case Builtin::BI__sync_lock_test_and_set_8:
911  case Builtin::BI__sync_lock_test_and_set_16:
912    BuiltinIndex = 12;
913    break;
914
915  case Builtin::BI__sync_lock_release:
916  case Builtin::BI__sync_lock_release_1:
917  case Builtin::BI__sync_lock_release_2:
918  case Builtin::BI__sync_lock_release_4:
919  case Builtin::BI__sync_lock_release_8:
920  case Builtin::BI__sync_lock_release_16:
921    BuiltinIndex = 13;
922    NumFixed = 0;
923    ResultType = Context.VoidTy;
924    break;
925
926  case Builtin::BI__sync_swap:
927  case Builtin::BI__sync_swap_1:
928  case Builtin::BI__sync_swap_2:
929  case Builtin::BI__sync_swap_4:
930  case Builtin::BI__sync_swap_8:
931  case Builtin::BI__sync_swap_16:
932    BuiltinIndex = 14;
933    break;
934  }
935
936  // Now that we know how many fixed arguments we expect, first check that we
937  // have at least that many.
938  if (TheCall->getNumArgs() < 1+NumFixed) {
939    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
940      << 0 << 1+NumFixed << TheCall->getNumArgs()
941      << TheCall->getCallee()->getSourceRange();
942    return ExprError();
943  }
944
945  // Get the decl for the concrete builtin from this, we can tell what the
946  // concrete integer type we should convert to is.
947  unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
948  const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
949  IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
950  FunctionDecl *NewBuiltinDecl =
951    cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
952                                           TUScope, false, DRE->getLocStart()));
953
954  // The first argument --- the pointer --- has a fixed type; we
955  // deduce the types of the rest of the arguments accordingly.  Walk
956  // the remaining arguments, converting them to the deduced value type.
957  for (unsigned i = 0; i != NumFixed; ++i) {
958    ExprResult Arg = TheCall->getArg(i+1);
959
960    // GCC does an implicit conversion to the pointer or integer ValType.  This
961    // can fail in some cases (1i -> int**), check for this error case now.
962    // Initialize the argument.
963    InitializedEntity Entity = InitializedEntity::InitializeParameter(Context,
964                                                   ValType, /*consume*/ false);
965    Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
966    if (Arg.isInvalid())
967      return ExprError();
968
969    // Okay, we have something that *can* be converted to the right type.  Check
970    // to see if there is a potentially weird extension going on here.  This can
971    // happen when you do an atomic operation on something like an char* and
972    // pass in 42.  The 42 gets converted to char.  This is even more strange
973    // for things like 45.123 -> char, etc.
974    // FIXME: Do this check.
975    TheCall->setArg(i+1, Arg.take());
976  }
977
978  ASTContext& Context = this->getASTContext();
979
980  // Create a new DeclRefExpr to refer to the new decl.
981  DeclRefExpr* NewDRE = DeclRefExpr::Create(
982      Context,
983      DRE->getQualifierLoc(),
984      SourceLocation(),
985      NewBuiltinDecl,
986      DRE->getLocation(),
987      NewBuiltinDecl->getType(),
988      DRE->getValueKind());
989
990  // Set the callee in the CallExpr.
991  // FIXME: This leaks the original parens and implicit casts.
992  ExprResult PromotedCall = UsualUnaryConversions(NewDRE);
993  if (PromotedCall.isInvalid())
994    return ExprError();
995  TheCall->setCallee(PromotedCall.take());
996
997  // Change the result type of the call to match the original value type. This
998  // is arbitrary, but the codegen for these builtins ins design to handle it
999  // gracefully.
1000  TheCall->setType(ResultType);
1001
1002  return move(TheCallResult);
1003}
1004
1005/// CheckObjCString - Checks that the argument to the builtin
1006/// CFString constructor is correct
1007/// Note: It might also make sense to do the UTF-16 conversion here (would
1008/// simplify the backend).
1009bool Sema::CheckObjCString(Expr *Arg) {
1010  Arg = Arg->IgnoreParenCasts();
1011  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
1012
1013  if (!Literal || !Literal->isAscii()) {
1014    Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
1015      << Arg->getSourceRange();
1016    return true;
1017  }
1018
1019  if (Literal->containsNonAsciiOrNull()) {
1020    StringRef String = Literal->getString();
1021    unsigned NumBytes = String.size();
1022    SmallVector<UTF16, 128> ToBuf(NumBytes);
1023    const UTF8 *FromPtr = (UTF8 *)String.data();
1024    UTF16 *ToPtr = &ToBuf[0];
1025
1026    ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
1027                                                 &ToPtr, ToPtr + NumBytes,
1028                                                 strictConversion);
1029    // Check for conversion failure.
1030    if (Result != conversionOK)
1031      Diag(Arg->getLocStart(),
1032           diag::warn_cfstring_truncated) << Arg->getSourceRange();
1033  }
1034  return false;
1035}
1036
1037/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
1038/// Emit an error and return true on failure, return false on success.
1039bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
1040  Expr *Fn = TheCall->getCallee();
1041  if (TheCall->getNumArgs() > 2) {
1042    Diag(TheCall->getArg(2)->getLocStart(),
1043         diag::err_typecheck_call_too_many_args)
1044      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1045      << Fn->getSourceRange()
1046      << SourceRange(TheCall->getArg(2)->getLocStart(),
1047                     (*(TheCall->arg_end()-1))->getLocEnd());
1048    return true;
1049  }
1050
1051  if (TheCall->getNumArgs() < 2) {
1052    return Diag(TheCall->getLocEnd(),
1053      diag::err_typecheck_call_too_few_args_at_least)
1054      << 0 /*function call*/ << 2 << TheCall->getNumArgs();
1055  }
1056
1057  // Type-check the first argument normally.
1058  if (checkBuiltinArgument(*this, TheCall, 0))
1059    return true;
1060
1061  // Determine whether the current function is variadic or not.
1062  BlockScopeInfo *CurBlock = getCurBlock();
1063  bool isVariadic;
1064  if (CurBlock)
1065    isVariadic = CurBlock->TheDecl->isVariadic();
1066  else if (FunctionDecl *FD = getCurFunctionDecl())
1067    isVariadic = FD->isVariadic();
1068  else
1069    isVariadic = getCurMethodDecl()->isVariadic();
1070
1071  if (!isVariadic) {
1072    Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
1073    return true;
1074  }
1075
1076  // Verify that the second argument to the builtin is the last argument of the
1077  // current function or method.
1078  bool SecondArgIsLastNamedArgument = false;
1079  const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
1080
1081  if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
1082    if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
1083      // FIXME: This isn't correct for methods (results in bogus warning).
1084      // Get the last formal in the current function.
1085      const ParmVarDecl *LastArg;
1086      if (CurBlock)
1087        LastArg = *(CurBlock->TheDecl->param_end()-1);
1088      else if (FunctionDecl *FD = getCurFunctionDecl())
1089        LastArg = *(FD->param_end()-1);
1090      else
1091        LastArg = *(getCurMethodDecl()->param_end()-1);
1092      SecondArgIsLastNamedArgument = PV == LastArg;
1093    }
1094  }
1095
1096  if (!SecondArgIsLastNamedArgument)
1097    Diag(TheCall->getArg(1)->getLocStart(),
1098         diag::warn_second_parameter_of_va_start_not_last_named_argument);
1099  return false;
1100}
1101
1102/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
1103/// friends.  This is declared to take (...), so we have to check everything.
1104bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
1105  if (TheCall->getNumArgs() < 2)
1106    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1107      << 0 << 2 << TheCall->getNumArgs()/*function call*/;
1108  if (TheCall->getNumArgs() > 2)
1109    return Diag(TheCall->getArg(2)->getLocStart(),
1110                diag::err_typecheck_call_too_many_args)
1111      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1112      << SourceRange(TheCall->getArg(2)->getLocStart(),
1113                     (*(TheCall->arg_end()-1))->getLocEnd());
1114
1115  ExprResult OrigArg0 = TheCall->getArg(0);
1116  ExprResult OrigArg1 = TheCall->getArg(1);
1117
1118  // Do standard promotions between the two arguments, returning their common
1119  // type.
1120  QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
1121  if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
1122    return true;
1123
1124  // Make sure any conversions are pushed back into the call; this is
1125  // type safe since unordered compare builtins are declared as "_Bool
1126  // foo(...)".
1127  TheCall->setArg(0, OrigArg0.get());
1128  TheCall->setArg(1, OrigArg1.get());
1129
1130  if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
1131    return false;
1132
1133  // If the common type isn't a real floating type, then the arguments were
1134  // invalid for this operation.
1135  if (!Res->isRealFloatingType())
1136    return Diag(OrigArg0.get()->getLocStart(),
1137                diag::err_typecheck_call_invalid_ordered_compare)
1138      << OrigArg0.get()->getType() << OrigArg1.get()->getType()
1139      << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
1140
1141  return false;
1142}
1143
1144/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
1145/// __builtin_isnan and friends.  This is declared to take (...), so we have
1146/// to check everything. We expect the last argument to be a floating point
1147/// value.
1148bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
1149  if (TheCall->getNumArgs() < NumArgs)
1150    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1151      << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
1152  if (TheCall->getNumArgs() > NumArgs)
1153    return Diag(TheCall->getArg(NumArgs)->getLocStart(),
1154                diag::err_typecheck_call_too_many_args)
1155      << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
1156      << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
1157                     (*(TheCall->arg_end()-1))->getLocEnd());
1158
1159  Expr *OrigArg = TheCall->getArg(NumArgs-1);
1160
1161  if (OrigArg->isTypeDependent())
1162    return false;
1163
1164  // This operation requires a non-_Complex floating-point number.
1165  if (!OrigArg->getType()->isRealFloatingType())
1166    return Diag(OrigArg->getLocStart(),
1167                diag::err_typecheck_call_invalid_unary_fp)
1168      << OrigArg->getType() << OrigArg->getSourceRange();
1169
1170  // If this is an implicit conversion from float -> double, remove it.
1171  if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
1172    Expr *CastArg = Cast->getSubExpr();
1173    if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
1174      assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
1175             "promotion from float to double is the only expected cast here");
1176      Cast->setSubExpr(0);
1177      TheCall->setArg(NumArgs-1, CastArg);
1178      OrigArg = CastArg;
1179    }
1180  }
1181
1182  return false;
1183}
1184
1185/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
1186// This is declared to take (...), so we have to check everything.
1187ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
1188  if (TheCall->getNumArgs() < 2)
1189    return ExprError(Diag(TheCall->getLocEnd(),
1190                          diag::err_typecheck_call_too_few_args_at_least)
1191      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1192      << TheCall->getSourceRange());
1193
1194  // Determine which of the following types of shufflevector we're checking:
1195  // 1) unary, vector mask: (lhs, mask)
1196  // 2) binary, vector mask: (lhs, rhs, mask)
1197  // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
1198  QualType resType = TheCall->getArg(0)->getType();
1199  unsigned numElements = 0;
1200
1201  if (!TheCall->getArg(0)->isTypeDependent() &&
1202      !TheCall->getArg(1)->isTypeDependent()) {
1203    QualType LHSType = TheCall->getArg(0)->getType();
1204    QualType RHSType = TheCall->getArg(1)->getType();
1205
1206    if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
1207      Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
1208        << SourceRange(TheCall->getArg(0)->getLocStart(),
1209                       TheCall->getArg(1)->getLocEnd());
1210      return ExprError();
1211    }
1212
1213    numElements = LHSType->getAs<VectorType>()->getNumElements();
1214    unsigned numResElements = TheCall->getNumArgs() - 2;
1215
1216    // Check to see if we have a call with 2 vector arguments, the unary shuffle
1217    // with mask.  If so, verify that RHS is an integer vector type with the
1218    // same number of elts as lhs.
1219    if (TheCall->getNumArgs() == 2) {
1220      if (!RHSType->hasIntegerRepresentation() ||
1221          RHSType->getAs<VectorType>()->getNumElements() != numElements)
1222        Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1223          << SourceRange(TheCall->getArg(1)->getLocStart(),
1224                         TheCall->getArg(1)->getLocEnd());
1225      numResElements = numElements;
1226    }
1227    else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
1228      Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1229        << SourceRange(TheCall->getArg(0)->getLocStart(),
1230                       TheCall->getArg(1)->getLocEnd());
1231      return ExprError();
1232    } else if (numElements != numResElements) {
1233      QualType eltType = LHSType->getAs<VectorType>()->getElementType();
1234      resType = Context.getVectorType(eltType, numResElements,
1235                                      VectorType::GenericVector);
1236    }
1237  }
1238
1239  for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
1240    if (TheCall->getArg(i)->isTypeDependent() ||
1241        TheCall->getArg(i)->isValueDependent())
1242      continue;
1243
1244    llvm::APSInt Result(32);
1245    if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
1246      return ExprError(Diag(TheCall->getLocStart(),
1247                  diag::err_shufflevector_nonconstant_argument)
1248                << TheCall->getArg(i)->getSourceRange());
1249
1250    if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
1251      return ExprError(Diag(TheCall->getLocStart(),
1252                  diag::err_shufflevector_argument_too_large)
1253               << TheCall->getArg(i)->getSourceRange());
1254  }
1255
1256  SmallVector<Expr*, 32> exprs;
1257
1258  for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
1259    exprs.push_back(TheCall->getArg(i));
1260    TheCall->setArg(i, 0);
1261  }
1262
1263  return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
1264                                            exprs.size(), resType,
1265                                            TheCall->getCallee()->getLocStart(),
1266                                            TheCall->getRParenLoc()));
1267}
1268
1269/// SemaBuiltinPrefetch - Handle __builtin_prefetch.
1270// This is declared to take (const void*, ...) and can take two
1271// optional constant int args.
1272bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
1273  unsigned NumArgs = TheCall->getNumArgs();
1274
1275  if (NumArgs > 3)
1276    return Diag(TheCall->getLocEnd(),
1277             diag::err_typecheck_call_too_many_args_at_most)
1278             << 0 /*function call*/ << 3 << NumArgs
1279             << TheCall->getSourceRange();
1280
1281  // Argument 0 is checked for us and the remaining arguments must be
1282  // constant integers.
1283  for (unsigned i = 1; i != NumArgs; ++i) {
1284    Expr *Arg = TheCall->getArg(i);
1285
1286    llvm::APSInt Result;
1287    if (SemaBuiltinConstantArg(TheCall, i, Result))
1288      return true;
1289
1290    // FIXME: gcc issues a warning and rewrites these to 0. These
1291    // seems especially odd for the third argument since the default
1292    // is 3.
1293    if (i == 1) {
1294      if (Result.getLimitedValue() > 1)
1295        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1296             << "0" << "1" << Arg->getSourceRange();
1297    } else {
1298      if (Result.getLimitedValue() > 3)
1299        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1300            << "0" << "3" << Arg->getSourceRange();
1301    }
1302  }
1303
1304  return false;
1305}
1306
1307/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
1308/// TheCall is a constant expression.
1309bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
1310                                  llvm::APSInt &Result) {
1311  Expr *Arg = TheCall->getArg(ArgNum);
1312  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
1313  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
1314
1315  if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
1316
1317  if (!Arg->isIntegerConstantExpr(Result, Context))
1318    return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
1319                << FDecl->getDeclName() <<  Arg->getSourceRange();
1320
1321  return false;
1322}
1323
1324/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
1325/// int type). This simply type checks that type is one of the defined
1326/// constants (0-3).
1327// For compatibility check 0-3, llvm only handles 0 and 2.
1328bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
1329  llvm::APSInt Result;
1330
1331  // Check constant-ness first.
1332  if (SemaBuiltinConstantArg(TheCall, 1, Result))
1333    return true;
1334
1335  Expr *Arg = TheCall->getArg(1);
1336  if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
1337    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1338             << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1339  }
1340
1341  return false;
1342}
1343
1344/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
1345/// This checks that val is a constant 1.
1346bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
1347  Expr *Arg = TheCall->getArg(1);
1348  llvm::APSInt Result;
1349
1350  // TODO: This is less than ideal. Overload this to take a value.
1351  if (SemaBuiltinConstantArg(TheCall, 1, Result))
1352    return true;
1353
1354  if (Result != 1)
1355    return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
1356             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1357
1358  return false;
1359}
1360
1361// Handle i > 1 ? "x" : "y", recursively.
1362bool Sema::SemaCheckStringLiteral(const Expr *E, Expr **Args,
1363                                  unsigned NumArgs, bool HasVAListArg,
1364                                  unsigned format_idx, unsigned firstDataArg,
1365                                  FormatStringType Type, bool inFunctionCall) {
1366 tryAgain:
1367  if (E->isTypeDependent() || E->isValueDependent())
1368    return false;
1369
1370  E = E->IgnoreParenCasts();
1371
1372  if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull))
1373    // Technically -Wformat-nonliteral does not warn about this case.
1374    // The behavior of printf and friends in this case is implementation
1375    // dependent.  Ideally if the format string cannot be null then
1376    // it should have a 'nonnull' attribute in the function prototype.
1377    return true;
1378
1379  switch (E->getStmtClass()) {
1380  case Stmt::BinaryConditionalOperatorClass:
1381  case Stmt::ConditionalOperatorClass: {
1382    const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E);
1383    return SemaCheckStringLiteral(C->getTrueExpr(), Args, NumArgs, HasVAListArg,
1384                                  format_idx, firstDataArg, Type,
1385                                  inFunctionCall)
1386       && SemaCheckStringLiteral(C->getFalseExpr(), Args, NumArgs, HasVAListArg,
1387                                 format_idx, firstDataArg, Type,
1388                                 inFunctionCall);
1389  }
1390
1391  case Stmt::ImplicitCastExprClass: {
1392    E = cast<ImplicitCastExpr>(E)->getSubExpr();
1393    goto tryAgain;
1394  }
1395
1396  case Stmt::OpaqueValueExprClass:
1397    if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
1398      E = src;
1399      goto tryAgain;
1400    }
1401    return false;
1402
1403  case Stmt::PredefinedExprClass:
1404    // While __func__, etc., are technically not string literals, they
1405    // cannot contain format specifiers and thus are not a security
1406    // liability.
1407    return true;
1408
1409  case Stmt::DeclRefExprClass: {
1410    const DeclRefExpr *DR = cast<DeclRefExpr>(E);
1411
1412    // As an exception, do not flag errors for variables binding to
1413    // const string literals.
1414    if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
1415      bool isConstant = false;
1416      QualType T = DR->getType();
1417
1418      if (const ArrayType *AT = Context.getAsArrayType(T)) {
1419        isConstant = AT->getElementType().isConstant(Context);
1420      } else if (const PointerType *PT = T->getAs<PointerType>()) {
1421        isConstant = T.isConstant(Context) &&
1422                     PT->getPointeeType().isConstant(Context);
1423      } else if (T->isObjCObjectPointerType()) {
1424        // In ObjC, there is usually no "const ObjectPointer" type,
1425        // so don't check if the pointee type is constant.
1426        isConstant = T.isConstant(Context);
1427      }
1428
1429      if (isConstant) {
1430        if (const Expr *Init = VD->getAnyInitializer())
1431          return SemaCheckStringLiteral(Init, Args, NumArgs,
1432                                        HasVAListArg, format_idx, firstDataArg,
1433                                        Type, /*inFunctionCall*/false);
1434      }
1435
1436      // For vprintf* functions (i.e., HasVAListArg==true), we add a
1437      // special check to see if the format string is a function parameter
1438      // of the function calling the printf function.  If the function
1439      // has an attribute indicating it is a printf-like function, then we
1440      // should suppress warnings concerning non-literals being used in a call
1441      // to a vprintf function.  For example:
1442      //
1443      // void
1444      // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
1445      //      va_list ap;
1446      //      va_start(ap, fmt);
1447      //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
1448      //      ...
1449      //
1450      if (HasVAListArg) {
1451        if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) {
1452          if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) {
1453            int PVIndex = PV->getFunctionScopeIndex() + 1;
1454            for (specific_attr_iterator<FormatAttr>
1455                 i = ND->specific_attr_begin<FormatAttr>(),
1456                 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) {
1457              FormatAttr *PVFormat = *i;
1458              // adjust for implicit parameter
1459              if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND))
1460                if (MD->isInstance())
1461                  ++PVIndex;
1462              // We also check if the formats are compatible.
1463              // We can't pass a 'scanf' string to a 'printf' function.
1464              if (PVIndex == PVFormat->getFormatIdx() &&
1465                  Type == GetFormatStringType(PVFormat))
1466                return true;
1467            }
1468          }
1469        }
1470      }
1471    }
1472
1473    return false;
1474  }
1475
1476  case Stmt::CallExprClass:
1477  case Stmt::CXXMemberCallExprClass: {
1478    const CallExpr *CE = cast<CallExpr>(E);
1479    if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) {
1480      if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) {
1481        unsigned ArgIndex = FA->getFormatIdx();
1482        if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND))
1483          if (MD->isInstance())
1484            --ArgIndex;
1485        const Expr *Arg = CE->getArg(ArgIndex - 1);
1486
1487        return SemaCheckStringLiteral(Arg, Args, NumArgs, HasVAListArg,
1488                                      format_idx, firstDataArg, Type,
1489                                      inFunctionCall);
1490      }
1491    }
1492
1493    return false;
1494  }
1495  case Stmt::ObjCStringLiteralClass:
1496  case Stmt::StringLiteralClass: {
1497    const StringLiteral *StrE = NULL;
1498
1499    if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1500      StrE = ObjCFExpr->getString();
1501    else
1502      StrE = cast<StringLiteral>(E);
1503
1504    if (StrE) {
1505      CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx,
1506                        firstDataArg, Type, inFunctionCall);
1507      return true;
1508    }
1509
1510    return false;
1511  }
1512
1513  default:
1514    return false;
1515  }
1516}
1517
1518void
1519Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1520                            const Expr * const *ExprArgs,
1521                            SourceLocation CallSiteLoc) {
1522  for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1523                                  e = NonNull->args_end();
1524       i != e; ++i) {
1525    const Expr *ArgExpr = ExprArgs[*i];
1526    if (ArgExpr->isNullPointerConstant(Context,
1527                                       Expr::NPC_ValueDependentIsNotNull))
1528      Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1529  }
1530}
1531
1532Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) {
1533  return llvm::StringSwitch<FormatStringType>(Format->getType())
1534  .Case("scanf", FST_Scanf)
1535  .Cases("printf", "printf0", FST_Printf)
1536  .Cases("NSString", "CFString", FST_NSString)
1537  .Case("strftime", FST_Strftime)
1538  .Case("strfmon", FST_Strfmon)
1539  .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf)
1540  .Default(FST_Unknown);
1541}
1542
1543/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar
1544/// functions) for correct use of format strings.
1545void Sema::CheckFormatArguments(const FormatAttr *Format, CallExpr *TheCall) {
1546  bool IsCXXMember = false;
1547  // The way the format attribute works in GCC, the implicit this argument
1548  // of member functions is counted. However, it doesn't appear in our own
1549  // lists, so decrement format_idx in that case.
1550  IsCXXMember = isa<CXXMemberCallExpr>(TheCall);
1551  CheckFormatArguments(Format, TheCall->getArgs(), TheCall->getNumArgs(),
1552                       IsCXXMember, TheCall->getRParenLoc(),
1553                       TheCall->getCallee()->getSourceRange());
1554}
1555
1556void Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args,
1557                                unsigned NumArgs, bool IsCXXMember,
1558                                SourceLocation Loc, SourceRange Range) {
1559  bool HasVAListArg = Format->getFirstArg() == 0;
1560  unsigned format_idx = Format->getFormatIdx() - 1;
1561  unsigned firstDataArg = HasVAListArg ? 0 : Format->getFirstArg() - 1;
1562  if (IsCXXMember) {
1563    if (format_idx == 0)
1564      return;
1565    --format_idx;
1566    if(firstDataArg != 0)
1567      --firstDataArg;
1568  }
1569  CheckFormatArguments(Args, NumArgs, HasVAListArg, format_idx,
1570                       firstDataArg, GetFormatStringType(Format), Loc, Range);
1571}
1572
1573void Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs,
1574                                bool HasVAListArg, unsigned format_idx,
1575                                unsigned firstDataArg, FormatStringType Type,
1576                                SourceLocation Loc, SourceRange Range) {
1577  // CHECK: printf/scanf-like function is called with no format string.
1578  if (format_idx >= NumArgs) {
1579    Diag(Loc, diag::warn_missing_format_string) << Range;
1580    return;
1581  }
1582
1583  const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts();
1584
1585  // CHECK: format string is not a string literal.
1586  //
1587  // Dynamically generated format strings are difficult to
1588  // automatically vet at compile time.  Requiring that format strings
1589  // are string literals: (1) permits the checking of format strings by
1590  // the compiler and thereby (2) can practically remove the source of
1591  // many format string exploits.
1592
1593  // Format string can be either ObjC string (e.g. @"%d") or
1594  // C string (e.g. "%d")
1595  // ObjC string uses the same format specifiers as C string, so we can use
1596  // the same format string checking logic for both ObjC and C strings.
1597  if (SemaCheckStringLiteral(OrigFormatExpr, Args, NumArgs, HasVAListArg,
1598                             format_idx, firstDataArg, Type))
1599    return;  // Literal format string found, check done!
1600
1601  // Strftime is particular as it always uses a single 'time' argument,
1602  // so it is safe to pass a non-literal string.
1603  if (Type == FST_Strftime)
1604    return;
1605
1606  // Do not emit diag when the string param is a macro expansion and the
1607  // format is either NSString or CFString. This is a hack to prevent
1608  // diag when using the NSLocalizedString and CFCopyLocalizedString macros
1609  // which are usually used in place of NS and CF string literals.
1610  if (Type == FST_NSString && Args[format_idx]->getLocStart().isMacroID())
1611    return;
1612
1613  // If there are no arguments specified, warn with -Wformat-security, otherwise
1614  // warn only with -Wformat-nonliteral.
1615  if (NumArgs == format_idx+1)
1616    Diag(Args[format_idx]->getLocStart(),
1617         diag::warn_format_nonliteral_noargs)
1618      << OrigFormatExpr->getSourceRange();
1619  else
1620    Diag(Args[format_idx]->getLocStart(),
1621         diag::warn_format_nonliteral)
1622           << OrigFormatExpr->getSourceRange();
1623}
1624
1625namespace {
1626class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1627protected:
1628  Sema &S;
1629  const StringLiteral *FExpr;
1630  const Expr *OrigFormatExpr;
1631  const unsigned FirstDataArg;
1632  const unsigned NumDataArgs;
1633  const bool IsObjCLiteral;
1634  const char *Beg; // Start of format string.
1635  const bool HasVAListArg;
1636  const Expr * const *Args;
1637  const unsigned NumArgs;
1638  unsigned FormatIdx;
1639  llvm::BitVector CoveredArgs;
1640  bool usesPositionalArgs;
1641  bool atFirstArg;
1642  bool inFunctionCall;
1643public:
1644  CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1645                     const Expr *origFormatExpr, unsigned firstDataArg,
1646                     unsigned numDataArgs, bool isObjCLiteral,
1647                     const char *beg, bool hasVAListArg,
1648                     Expr **args, unsigned numArgs,
1649                     unsigned formatIdx, bool inFunctionCall)
1650    : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1651      FirstDataArg(firstDataArg),
1652      NumDataArgs(numDataArgs),
1653      IsObjCLiteral(isObjCLiteral), Beg(beg),
1654      HasVAListArg(hasVAListArg),
1655      Args(args), NumArgs(numArgs), FormatIdx(formatIdx),
1656      usesPositionalArgs(false), atFirstArg(true),
1657      inFunctionCall(inFunctionCall) {
1658        CoveredArgs.resize(numDataArgs);
1659        CoveredArgs.reset();
1660      }
1661
1662  void DoneProcessing();
1663
1664  void HandleIncompleteSpecifier(const char *startSpecifier,
1665                                 unsigned specifierLen);
1666
1667  virtual void HandleInvalidPosition(const char *startSpecifier,
1668                                     unsigned specifierLen,
1669                                     analyze_format_string::PositionContext p);
1670
1671  virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
1672
1673  void HandleNullChar(const char *nullCharacter);
1674
1675  template <typename Range>
1676  static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall,
1677                                   const Expr *ArgumentExpr,
1678                                   PartialDiagnostic PDiag,
1679                                   SourceLocation StringLoc,
1680                                   bool IsStringLocation, Range StringRange,
1681                                   FixItHint Fixit = FixItHint());
1682
1683protected:
1684  bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
1685                                        const char *startSpec,
1686                                        unsigned specifierLen,
1687                                        const char *csStart, unsigned csLen);
1688
1689  void HandlePositionalNonpositionalArgs(SourceLocation Loc,
1690                                         const char *startSpec,
1691                                         unsigned specifierLen);
1692
1693  SourceRange getFormatStringRange();
1694  CharSourceRange getSpecifierRange(const char *startSpecifier,
1695                                    unsigned specifierLen);
1696  SourceLocation getLocationOfByte(const char *x);
1697
1698  const Expr *getDataArg(unsigned i) const;
1699
1700  bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
1701                    const analyze_format_string::ConversionSpecifier &CS,
1702                    const char *startSpecifier, unsigned specifierLen,
1703                    unsigned argIndex);
1704
1705  template <typename Range>
1706  void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc,
1707                            bool IsStringLocation, Range StringRange,
1708                            FixItHint Fixit = FixItHint());
1709
1710  void CheckPositionalAndNonpositionalArgs(
1711      const analyze_format_string::FormatSpecifier *FS);
1712};
1713}
1714
1715SourceRange CheckFormatHandler::getFormatStringRange() {
1716  return OrigFormatExpr->getSourceRange();
1717}
1718
1719CharSourceRange CheckFormatHandler::
1720getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
1721  SourceLocation Start = getLocationOfByte(startSpecifier);
1722  SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
1723
1724  // Advance the end SourceLocation by one due to half-open ranges.
1725  End = End.getLocWithOffset(1);
1726
1727  return CharSourceRange::getCharRange(Start, End);
1728}
1729
1730SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
1731  return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
1732}
1733
1734void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
1735                                                   unsigned specifierLen){
1736  EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier),
1737                       getLocationOfByte(startSpecifier),
1738                       /*IsStringLocation*/true,
1739                       getSpecifierRange(startSpecifier, specifierLen));
1740}
1741
1742void
1743CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
1744                                     analyze_format_string::PositionContext p) {
1745  EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier)
1746                         << (unsigned) p,
1747                       getLocationOfByte(startPos), /*IsStringLocation*/true,
1748                       getSpecifierRange(startPos, posLen));
1749}
1750
1751void CheckFormatHandler::HandleZeroPosition(const char *startPos,
1752                                            unsigned posLen) {
1753  EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier),
1754                               getLocationOfByte(startPos),
1755                               /*IsStringLocation*/true,
1756                               getSpecifierRange(startPos, posLen));
1757}
1758
1759void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
1760  if (!IsObjCLiteral) {
1761    // The presence of a null character is likely an error.
1762    EmitFormatDiagnostic(
1763      S.PDiag(diag::warn_printf_format_string_contains_null_char),
1764      getLocationOfByte(nullCharacter), /*IsStringLocation*/true,
1765      getFormatStringRange());
1766  }
1767}
1768
1769const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
1770  return Args[FirstDataArg + i];
1771}
1772
1773void CheckFormatHandler::DoneProcessing() {
1774    // Does the number of data arguments exceed the number of
1775    // format conversions in the format string?
1776  if (!HasVAListArg) {
1777      // Find any arguments that weren't covered.
1778    CoveredArgs.flip();
1779    signed notCoveredArg = CoveredArgs.find_first();
1780    if (notCoveredArg >= 0) {
1781      assert((unsigned)notCoveredArg < NumDataArgs);
1782      EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used),
1783                           getDataArg((unsigned) notCoveredArg)->getLocStart(),
1784                           /*IsStringLocation*/false, getFormatStringRange());
1785    }
1786  }
1787}
1788
1789bool
1790CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
1791                                                     SourceLocation Loc,
1792                                                     const char *startSpec,
1793                                                     unsigned specifierLen,
1794                                                     const char *csStart,
1795                                                     unsigned csLen) {
1796
1797  bool keepGoing = true;
1798  if (argIndex < NumDataArgs) {
1799    // Consider the argument coverered, even though the specifier doesn't
1800    // make sense.
1801    CoveredArgs.set(argIndex);
1802  }
1803  else {
1804    // If argIndex exceeds the number of data arguments we
1805    // don't issue a warning because that is just a cascade of warnings (and
1806    // they may have intended '%%' anyway). We don't want to continue processing
1807    // the format string after this point, however, as we will like just get
1808    // gibberish when trying to match arguments.
1809    keepGoing = false;
1810  }
1811
1812  EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion)
1813                         << StringRef(csStart, csLen),
1814                       Loc, /*IsStringLocation*/true,
1815                       getSpecifierRange(startSpec, specifierLen));
1816
1817  return keepGoing;
1818}
1819
1820void
1821CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc,
1822                                                      const char *startSpec,
1823                                                      unsigned specifierLen) {
1824  EmitFormatDiagnostic(
1825    S.PDiag(diag::warn_format_mix_positional_nonpositional_args),
1826    Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen));
1827}
1828
1829bool
1830CheckFormatHandler::CheckNumArgs(
1831  const analyze_format_string::FormatSpecifier &FS,
1832  const analyze_format_string::ConversionSpecifier &CS,
1833  const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
1834
1835  if (argIndex >= NumDataArgs) {
1836    PartialDiagnostic PDiag = FS.usesPositionalArg()
1837      ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args)
1838           << (argIndex+1) << NumDataArgs)
1839      : S.PDiag(diag::warn_printf_insufficient_data_args);
1840    EmitFormatDiagnostic(
1841      PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true,
1842      getSpecifierRange(startSpecifier, specifierLen));
1843    return false;
1844  }
1845  return true;
1846}
1847
1848template<typename Range>
1849void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag,
1850                                              SourceLocation Loc,
1851                                              bool IsStringLocation,
1852                                              Range StringRange,
1853                                              FixItHint FixIt) {
1854  EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag,
1855                       Loc, IsStringLocation, StringRange, FixIt);
1856}
1857
1858/// \brief If the format string is not within the funcion call, emit a note
1859/// so that the function call and string are in diagnostic messages.
1860///
1861/// \param inFunctionCall if true, the format string is within the function
1862/// call and only one diagnostic message will be produced.  Otherwise, an
1863/// extra note will be emitted pointing to location of the format string.
1864///
1865/// \param ArgumentExpr the expression that is passed as the format string
1866/// argument in the function call.  Used for getting locations when two
1867/// diagnostics are emitted.
1868///
1869/// \param PDiag the callee should already have provided any strings for the
1870/// diagnostic message.  This function only adds locations and fixits
1871/// to diagnostics.
1872///
1873/// \param Loc primary location for diagnostic.  If two diagnostics are
1874/// required, one will be at Loc and a new SourceLocation will be created for
1875/// the other one.
1876///
1877/// \param IsStringLocation if true, Loc points to the format string should be
1878/// used for the note.  Otherwise, Loc points to the argument list and will
1879/// be used with PDiag.
1880///
1881/// \param StringRange some or all of the string to highlight.  This is
1882/// templated so it can accept either a CharSourceRange or a SourceRange.
1883///
1884/// \param Fixit optional fix it hint for the format string.
1885template<typename Range>
1886void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall,
1887                                              const Expr *ArgumentExpr,
1888                                              PartialDiagnostic PDiag,
1889                                              SourceLocation Loc,
1890                                              bool IsStringLocation,
1891                                              Range StringRange,
1892                                              FixItHint FixIt) {
1893  if (InFunctionCall)
1894    S.Diag(Loc, PDiag) << StringRange << FixIt;
1895  else {
1896    S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag)
1897      << ArgumentExpr->getSourceRange();
1898    S.Diag(IsStringLocation ? Loc : StringRange.getBegin(),
1899           diag::note_format_string_defined)
1900      << StringRange << FixIt;
1901  }
1902}
1903
1904//===--- CHECK: Printf format string checking ------------------------------===//
1905
1906namespace {
1907class CheckPrintfHandler : public CheckFormatHandler {
1908public:
1909  CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
1910                     const Expr *origFormatExpr, unsigned firstDataArg,
1911                     unsigned numDataArgs, bool isObjCLiteral,
1912                     const char *beg, bool hasVAListArg,
1913                     Expr **Args, unsigned NumArgs,
1914                     unsigned formatIdx, bool inFunctionCall)
1915  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1916                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1917                       Args, NumArgs, formatIdx, inFunctionCall) {}
1918
1919
1920  bool HandleInvalidPrintfConversionSpecifier(
1921                                      const analyze_printf::PrintfSpecifier &FS,
1922                                      const char *startSpecifier,
1923                                      unsigned specifierLen);
1924
1925  bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
1926                             const char *startSpecifier,
1927                             unsigned specifierLen);
1928
1929  bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
1930                    const char *startSpecifier, unsigned specifierLen);
1931  void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
1932                           const analyze_printf::OptionalAmount &Amt,
1933                           unsigned type,
1934                           const char *startSpecifier, unsigned specifierLen);
1935  void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1936                  const analyze_printf::OptionalFlag &flag,
1937                  const char *startSpecifier, unsigned specifierLen);
1938  void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
1939                         const analyze_printf::OptionalFlag &ignoredFlag,
1940                         const analyze_printf::OptionalFlag &flag,
1941                         const char *startSpecifier, unsigned specifierLen);
1942};
1943}
1944
1945bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
1946                                      const analyze_printf::PrintfSpecifier &FS,
1947                                      const char *startSpecifier,
1948                                      unsigned specifierLen) {
1949  const analyze_printf::PrintfConversionSpecifier &CS =
1950    FS.getConversionSpecifier();
1951
1952  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1953                                          getLocationOfByte(CS.getStart()),
1954                                          startSpecifier, specifierLen,
1955                                          CS.getStart(), CS.getLength());
1956}
1957
1958bool CheckPrintfHandler::HandleAmount(
1959                               const analyze_format_string::OptionalAmount &Amt,
1960                               unsigned k, const char *startSpecifier,
1961                               unsigned specifierLen) {
1962
1963  if (Amt.hasDataArgument()) {
1964    if (!HasVAListArg) {
1965      unsigned argIndex = Amt.getArgIndex();
1966      if (argIndex >= NumDataArgs) {
1967        EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg)
1968                               << k,
1969                             getLocationOfByte(Amt.getStart()),
1970                             /*IsStringLocation*/true,
1971                             getSpecifierRange(startSpecifier, specifierLen));
1972        // Don't do any more checking.  We will just emit
1973        // spurious errors.
1974        return false;
1975      }
1976
1977      // Type check the data argument.  It should be an 'int'.
1978      // Although not in conformance with C99, we also allow the argument to be
1979      // an 'unsigned int' as that is a reasonably safe case.  GCC also
1980      // doesn't emit a warning for that case.
1981      CoveredArgs.set(argIndex);
1982      const Expr *Arg = getDataArg(argIndex);
1983      QualType T = Arg->getType();
1984
1985      const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context);
1986      assert(ATR.isValid());
1987
1988      if (!ATR.matchesType(S.Context, T)) {
1989        EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type)
1990                               << k << ATR.getRepresentativeTypeName(S.Context)
1991                               << T << Arg->getSourceRange(),
1992                             getLocationOfByte(Amt.getStart()),
1993                             /*IsStringLocation*/true,
1994                             getSpecifierRange(startSpecifier, specifierLen));
1995        // Don't do any more checking.  We will just emit
1996        // spurious errors.
1997        return false;
1998      }
1999    }
2000  }
2001  return true;
2002}
2003
2004void CheckPrintfHandler::HandleInvalidAmount(
2005                                      const analyze_printf::PrintfSpecifier &FS,
2006                                      const analyze_printf::OptionalAmount &Amt,
2007                                      unsigned type,
2008                                      const char *startSpecifier,
2009                                      unsigned specifierLen) {
2010  const analyze_printf::PrintfConversionSpecifier &CS =
2011    FS.getConversionSpecifier();
2012
2013  FixItHint fixit =
2014    Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant
2015      ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
2016                                 Amt.getConstantLength()))
2017      : FixItHint();
2018
2019  EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount)
2020                         << type << CS.toString(),
2021                       getLocationOfByte(Amt.getStart()),
2022                       /*IsStringLocation*/true,
2023                       getSpecifierRange(startSpecifier, specifierLen),
2024                       fixit);
2025}
2026
2027void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
2028                                    const analyze_printf::OptionalFlag &flag,
2029                                    const char *startSpecifier,
2030                                    unsigned specifierLen) {
2031  // Warn about pointless flag with a fixit removal.
2032  const analyze_printf::PrintfConversionSpecifier &CS =
2033    FS.getConversionSpecifier();
2034  EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag)
2035                         << flag.toString() << CS.toString(),
2036                       getLocationOfByte(flag.getPosition()),
2037                       /*IsStringLocation*/true,
2038                       getSpecifierRange(startSpecifier, specifierLen),
2039                       FixItHint::CreateRemoval(
2040                         getSpecifierRange(flag.getPosition(), 1)));
2041}
2042
2043void CheckPrintfHandler::HandleIgnoredFlag(
2044                                const analyze_printf::PrintfSpecifier &FS,
2045                                const analyze_printf::OptionalFlag &ignoredFlag,
2046                                const analyze_printf::OptionalFlag &flag,
2047                                const char *startSpecifier,
2048                                unsigned specifierLen) {
2049  // Warn about ignored flag with a fixit removal.
2050  EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag)
2051                         << ignoredFlag.toString() << flag.toString(),
2052                       getLocationOfByte(ignoredFlag.getPosition()),
2053                       /*IsStringLocation*/true,
2054                       getSpecifierRange(startSpecifier, specifierLen),
2055                       FixItHint::CreateRemoval(
2056                         getSpecifierRange(ignoredFlag.getPosition(), 1)));
2057}
2058
2059bool
2060CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
2061                                            &FS,
2062                                          const char *startSpecifier,
2063                                          unsigned specifierLen) {
2064
2065  using namespace analyze_format_string;
2066  using namespace analyze_printf;
2067  const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
2068
2069  if (FS.consumesDataArgument()) {
2070    if (atFirstArg) {
2071        atFirstArg = false;
2072        usesPositionalArgs = FS.usesPositionalArg();
2073    }
2074    else if (usesPositionalArgs != FS.usesPositionalArg()) {
2075      HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()),
2076                                        startSpecifier, specifierLen);
2077      return false;
2078    }
2079  }
2080
2081  // First check if the field width, precision, and conversion specifier
2082  // have matching data arguments.
2083  if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
2084                    startSpecifier, specifierLen)) {
2085    return false;
2086  }
2087
2088  if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
2089                    startSpecifier, specifierLen)) {
2090    return false;
2091  }
2092
2093  if (!CS.consumesDataArgument()) {
2094    // FIXME: Technically specifying a precision or field width here
2095    // makes no sense.  Worth issuing a warning at some point.
2096    return true;
2097  }
2098
2099  // Consume the argument.
2100  unsigned argIndex = FS.getArgIndex();
2101  if (argIndex < NumDataArgs) {
2102    // The check to see if the argIndex is valid will come later.
2103    // We set the bit here because we may exit early from this
2104    // function if we encounter some other error.
2105    CoveredArgs.set(argIndex);
2106  }
2107
2108  // Check for using an Objective-C specific conversion specifier
2109  // in a non-ObjC literal.
2110  if (!IsObjCLiteral && CS.isObjCArg()) {
2111    return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
2112                                                  specifierLen);
2113  }
2114
2115  // Check for invalid use of field width
2116  if (!FS.hasValidFieldWidth()) {
2117    HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
2118        startSpecifier, specifierLen);
2119  }
2120
2121  // Check for invalid use of precision
2122  if (!FS.hasValidPrecision()) {
2123    HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
2124        startSpecifier, specifierLen);
2125  }
2126
2127  // Check each flag does not conflict with any other component.
2128  if (!FS.hasValidThousandsGroupingPrefix())
2129    HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
2130  if (!FS.hasValidLeadingZeros())
2131    HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
2132  if (!FS.hasValidPlusPrefix())
2133    HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
2134  if (!FS.hasValidSpacePrefix())
2135    HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
2136  if (!FS.hasValidAlternativeForm())
2137    HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
2138  if (!FS.hasValidLeftJustified())
2139    HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
2140
2141  // Check that flags are not ignored by another flag
2142  if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
2143    HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
2144        startSpecifier, specifierLen);
2145  if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
2146    HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
2147            startSpecifier, specifierLen);
2148
2149  // Check the length modifier is valid with the given conversion specifier.
2150  const LengthModifier &LM = FS.getLengthModifier();
2151  if (!FS.hasValidLengthModifier())
2152    EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length)
2153                           << LM.toString() << CS.toString(),
2154                         getLocationOfByte(LM.getStart()),
2155                         /*IsStringLocation*/true,
2156                         getSpecifierRange(startSpecifier, specifierLen),
2157                         FixItHint::CreateRemoval(
2158                           getSpecifierRange(LM.getStart(),
2159                                             LM.getLength())));
2160
2161  // Are we using '%n'?
2162  if (CS.getKind() == ConversionSpecifier::nArg) {
2163    // Issue a warning about this being a possible security issue.
2164    EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back),
2165                         getLocationOfByte(CS.getStart()),
2166                         /*IsStringLocation*/true,
2167                         getSpecifierRange(startSpecifier, specifierLen));
2168    // Continue checking the other format specifiers.
2169    return true;
2170  }
2171
2172  // The remaining checks depend on the data arguments.
2173  if (HasVAListArg)
2174    return true;
2175
2176  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
2177    return false;
2178
2179  // Now type check the data expression that matches the
2180  // format specifier.
2181  const Expr *Ex = getDataArg(argIndex);
2182  const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context,
2183                                                           IsObjCLiteral);
2184  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
2185    // Check if we didn't match because of an implicit cast from a 'char'
2186    // or 'short' to an 'int'.  This is done because printf is a varargs
2187    // function.
2188    if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex))
2189      if (ICE->getType() == S.Context.IntTy) {
2190        // All further checking is done on the subexpression.
2191        Ex = ICE->getSubExpr();
2192        if (ATR.matchesType(S.Context, Ex->getType()))
2193          return true;
2194      }
2195
2196    // We may be able to offer a FixItHint if it is a supported type.
2197    PrintfSpecifier fixedFS = FS;
2198    bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions(),
2199                                   S.Context, IsObjCLiteral);
2200
2201    if (success) {
2202      // Get the fix string from the fixed format specifier
2203      SmallString<128> buf;
2204      llvm::raw_svector_ostream os(buf);
2205      fixedFS.toString(os);
2206
2207      EmitFormatDiagnostic(
2208        S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2209          << ATR.getRepresentativeTypeName(S.Context) << Ex->getType()
2210          << Ex->getSourceRange(),
2211        getLocationOfByte(CS.getStart()),
2212        /*IsStringLocation*/true,
2213        getSpecifierRange(startSpecifier, specifierLen),
2214        FixItHint::CreateReplacement(
2215          getSpecifierRange(startSpecifier, specifierLen),
2216          os.str()));
2217    }
2218    else {
2219      EmitFormatDiagnostic(
2220        S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2221          << ATR.getRepresentativeTypeName(S.Context) << Ex->getType()
2222          << getSpecifierRange(startSpecifier, specifierLen)
2223          << Ex->getSourceRange(),
2224        getLocationOfByte(CS.getStart()),
2225        true,
2226        getSpecifierRange(startSpecifier, specifierLen));
2227    }
2228  }
2229
2230  return true;
2231}
2232
2233//===--- CHECK: Scanf format string checking ------------------------------===//
2234
2235namespace {
2236class CheckScanfHandler : public CheckFormatHandler {
2237public:
2238  CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
2239                    const Expr *origFormatExpr, unsigned firstDataArg,
2240                    unsigned numDataArgs, bool isObjCLiteral,
2241                    const char *beg, bool hasVAListArg,
2242                    Expr **Args, unsigned NumArgs,
2243                    unsigned formatIdx, bool inFunctionCall)
2244  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
2245                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
2246                       Args, NumArgs, formatIdx, inFunctionCall) {}
2247
2248  bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
2249                            const char *startSpecifier,
2250                            unsigned specifierLen);
2251
2252  bool HandleInvalidScanfConversionSpecifier(
2253          const analyze_scanf::ScanfSpecifier &FS,
2254          const char *startSpecifier,
2255          unsigned specifierLen);
2256
2257  void HandleIncompleteScanList(const char *start, const char *end);
2258};
2259}
2260
2261void CheckScanfHandler::HandleIncompleteScanList(const char *start,
2262                                                 const char *end) {
2263  EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete),
2264                       getLocationOfByte(end), /*IsStringLocation*/true,
2265                       getSpecifierRange(start, end - start));
2266}
2267
2268bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
2269                                        const analyze_scanf::ScanfSpecifier &FS,
2270                                        const char *startSpecifier,
2271                                        unsigned specifierLen) {
2272
2273  const analyze_scanf::ScanfConversionSpecifier &CS =
2274    FS.getConversionSpecifier();
2275
2276  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
2277                                          getLocationOfByte(CS.getStart()),
2278                                          startSpecifier, specifierLen,
2279                                          CS.getStart(), CS.getLength());
2280}
2281
2282bool CheckScanfHandler::HandleScanfSpecifier(
2283                                       const analyze_scanf::ScanfSpecifier &FS,
2284                                       const char *startSpecifier,
2285                                       unsigned specifierLen) {
2286
2287  using namespace analyze_scanf;
2288  using namespace analyze_format_string;
2289
2290  const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
2291
2292  // Handle case where '%' and '*' don't consume an argument.  These shouldn't
2293  // be used to decide if we are using positional arguments consistently.
2294  if (FS.consumesDataArgument()) {
2295    if (atFirstArg) {
2296      atFirstArg = false;
2297      usesPositionalArgs = FS.usesPositionalArg();
2298    }
2299    else if (usesPositionalArgs != FS.usesPositionalArg()) {
2300      HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()),
2301                                        startSpecifier, specifierLen);
2302      return false;
2303    }
2304  }
2305
2306  // Check if the field with is non-zero.
2307  const OptionalAmount &Amt = FS.getFieldWidth();
2308  if (Amt.getHowSpecified() == OptionalAmount::Constant) {
2309    if (Amt.getConstantAmount() == 0) {
2310      const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
2311                                                   Amt.getConstantLength());
2312      EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width),
2313                           getLocationOfByte(Amt.getStart()),
2314                           /*IsStringLocation*/true, R,
2315                           FixItHint::CreateRemoval(R));
2316    }
2317  }
2318
2319  if (!FS.consumesDataArgument()) {
2320    // FIXME: Technically specifying a precision or field width here
2321    // makes no sense.  Worth issuing a warning at some point.
2322    return true;
2323  }
2324
2325  // Consume the argument.
2326  unsigned argIndex = FS.getArgIndex();
2327  if (argIndex < NumDataArgs) {
2328      // The check to see if the argIndex is valid will come later.
2329      // We set the bit here because we may exit early from this
2330      // function if we encounter some other error.
2331    CoveredArgs.set(argIndex);
2332  }
2333
2334  // Check the length modifier is valid with the given conversion specifier.
2335  const LengthModifier &LM = FS.getLengthModifier();
2336  if (!FS.hasValidLengthModifier()) {
2337    const CharSourceRange &R = getSpecifierRange(LM.getStart(), LM.getLength());
2338    EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length)
2339                         << LM.toString() << CS.toString()
2340                         << getSpecifierRange(startSpecifier, specifierLen),
2341                         getLocationOfByte(LM.getStart()),
2342                         /*IsStringLocation*/true, R,
2343                         FixItHint::CreateRemoval(R));
2344  }
2345
2346  // The remaining checks depend on the data arguments.
2347  if (HasVAListArg)
2348    return true;
2349
2350  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
2351    return false;
2352
2353  // Check that the argument type matches the format specifier.
2354  const Expr *Ex = getDataArg(argIndex);
2355  const analyze_scanf::ScanfArgTypeResult &ATR = FS.getArgType(S.Context);
2356  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
2357    ScanfSpecifier fixedFS = FS;
2358    bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions(),
2359                                   S.Context);
2360
2361    if (success) {
2362      // Get the fix string from the fixed format specifier.
2363      SmallString<128> buf;
2364      llvm::raw_svector_ostream os(buf);
2365      fixedFS.toString(os);
2366
2367      EmitFormatDiagnostic(
2368        S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2369          << ATR.getRepresentativeTypeName(S.Context) << Ex->getType()
2370          << Ex->getSourceRange(),
2371        getLocationOfByte(CS.getStart()),
2372        /*IsStringLocation*/true,
2373        getSpecifierRange(startSpecifier, specifierLen),
2374        FixItHint::CreateReplacement(
2375          getSpecifierRange(startSpecifier, specifierLen),
2376          os.str()));
2377    } else {
2378      EmitFormatDiagnostic(
2379        S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2380          << ATR.getRepresentativeTypeName(S.Context) << Ex->getType()
2381          << Ex->getSourceRange(),
2382        getLocationOfByte(CS.getStart()),
2383        /*IsStringLocation*/true,
2384        getSpecifierRange(startSpecifier, specifierLen));
2385    }
2386  }
2387
2388  return true;
2389}
2390
2391void Sema::CheckFormatString(const StringLiteral *FExpr,
2392                             const Expr *OrigFormatExpr,
2393                             Expr **Args, unsigned NumArgs,
2394                             bool HasVAListArg, unsigned format_idx,
2395                             unsigned firstDataArg, FormatStringType Type,
2396                             bool inFunctionCall) {
2397
2398  // CHECK: is the format string a wide literal?
2399  if (!FExpr->isAscii()) {
2400    CheckFormatHandler::EmitFormatDiagnostic(
2401      *this, inFunctionCall, Args[format_idx],
2402      PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(),
2403      /*IsStringLocation*/true, OrigFormatExpr->getSourceRange());
2404    return;
2405  }
2406
2407  // Str - The format string.  NOTE: this is NOT null-terminated!
2408  StringRef StrRef = FExpr->getString();
2409  const char *Str = StrRef.data();
2410  unsigned StrLen = StrRef.size();
2411  const unsigned numDataArgs = NumArgs - firstDataArg;
2412
2413  // CHECK: empty format string?
2414  if (StrLen == 0 && numDataArgs > 0) {
2415    CheckFormatHandler::EmitFormatDiagnostic(
2416      *this, inFunctionCall, Args[format_idx],
2417      PDiag(diag::warn_empty_format_string), FExpr->getLocStart(),
2418      /*IsStringLocation*/true, OrigFormatExpr->getSourceRange());
2419    return;
2420  }
2421
2422  if (Type == FST_Printf || Type == FST_NSString) {
2423    CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
2424                         numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr),
2425                         Str, HasVAListArg, Args, NumArgs, format_idx,
2426                         inFunctionCall);
2427
2428    if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen,
2429                                                  getLangOptions()))
2430      H.DoneProcessing();
2431  } else if (Type == FST_Scanf) {
2432    CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
2433                        numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr),
2434                        Str, HasVAListArg, Args, NumArgs, format_idx,
2435                        inFunctionCall);
2436
2437    if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen,
2438                                                 getLangOptions()))
2439      H.DoneProcessing();
2440  } // TODO: handle other formats
2441}
2442
2443//===--- CHECK: Standard memory functions ---------------------------------===//
2444
2445/// \brief Determine whether the given type is a dynamic class type (e.g.,
2446/// whether it has a vtable).
2447static bool isDynamicClassType(QualType T) {
2448  if (CXXRecordDecl *Record = T->getAsCXXRecordDecl())
2449    if (CXXRecordDecl *Definition = Record->getDefinition())
2450      if (Definition->isDynamicClass())
2451        return true;
2452
2453  return false;
2454}
2455
2456/// \brief If E is a sizeof expression, returns its argument expression,
2457/// otherwise returns NULL.
2458static const Expr *getSizeOfExprArg(const Expr* E) {
2459  if (const UnaryExprOrTypeTraitExpr *SizeOf =
2460      dyn_cast<UnaryExprOrTypeTraitExpr>(E))
2461    if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType())
2462      return SizeOf->getArgumentExpr()->IgnoreParenImpCasts();
2463
2464  return 0;
2465}
2466
2467/// \brief If E is a sizeof expression, returns its argument type.
2468static QualType getSizeOfArgType(const Expr* E) {
2469  if (const UnaryExprOrTypeTraitExpr *SizeOf =
2470      dyn_cast<UnaryExprOrTypeTraitExpr>(E))
2471    if (SizeOf->getKind() == clang::UETT_SizeOf)
2472      return SizeOf->getTypeOfArgument();
2473
2474  return QualType();
2475}
2476
2477/// \brief Check for dangerous or invalid arguments to memset().
2478///
2479/// This issues warnings on known problematic, dangerous or unspecified
2480/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp'
2481/// function calls.
2482///
2483/// \param Call The call expression to diagnose.
2484void Sema::CheckMemaccessArguments(const CallExpr *Call,
2485                                   unsigned BId,
2486                                   IdentifierInfo *FnName) {
2487  assert(BId != 0);
2488
2489  // It is possible to have a non-standard definition of memset.  Validate
2490  // we have enough arguments, and if not, abort further checking.
2491  unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3);
2492  if (Call->getNumArgs() < ExpectedNumArgs)
2493    return;
2494
2495  unsigned LastArg = (BId == Builtin::BImemset ||
2496                      BId == Builtin::BIstrndup ? 1 : 2);
2497  unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2);
2498  const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts();
2499
2500  // We have special checking when the length is a sizeof expression.
2501  QualType SizeOfArgTy = getSizeOfArgType(LenExpr);
2502  const Expr *SizeOfArg = getSizeOfExprArg(LenExpr);
2503  llvm::FoldingSetNodeID SizeOfArgID;
2504
2505  for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) {
2506    const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts();
2507    SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange();
2508
2509    QualType DestTy = Dest->getType();
2510    if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
2511      QualType PointeeTy = DestPtrTy->getPointeeType();
2512
2513      // Never warn about void type pointers. This can be used to suppress
2514      // false positives.
2515      if (PointeeTy->isVoidType())
2516        continue;
2517
2518      // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by
2519      // actually comparing the expressions for equality. Because computing the
2520      // expression IDs can be expensive, we only do this if the diagnostic is
2521      // enabled.
2522      if (SizeOfArg &&
2523          Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess,
2524                                   SizeOfArg->getExprLoc())) {
2525        // We only compute IDs for expressions if the warning is enabled, and
2526        // cache the sizeof arg's ID.
2527        if (SizeOfArgID == llvm::FoldingSetNodeID())
2528          SizeOfArg->Profile(SizeOfArgID, Context, true);
2529        llvm::FoldingSetNodeID DestID;
2530        Dest->Profile(DestID, Context, true);
2531        if (DestID == SizeOfArgID) {
2532          // TODO: For strncpy() and friends, this could suggest sizeof(dst)
2533          //       over sizeof(src) as well.
2534          unsigned ActionIdx = 0; // Default is to suggest dereferencing.
2535          if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest))
2536            if (UnaryOp->getOpcode() == UO_AddrOf)
2537              ActionIdx = 1; // If its an address-of operator, just remove it.
2538          if (Context.getTypeSize(PointeeTy) == Context.getCharWidth())
2539            ActionIdx = 2; // If the pointee's size is sizeof(char),
2540                           // suggest an explicit length.
2541          unsigned DestSrcSelect =
2542            (BId == Builtin::BIstrndup ? 1 : ArgIdx);
2543          DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest,
2544                              PDiag(diag::warn_sizeof_pointer_expr_memaccess)
2545                                << FnName << DestSrcSelect << ActionIdx
2546                                << Dest->getSourceRange()
2547                                << SizeOfArg->getSourceRange());
2548          break;
2549        }
2550      }
2551
2552      // Also check for cases where the sizeof argument is the exact same
2553      // type as the memory argument, and where it points to a user-defined
2554      // record type.
2555      if (SizeOfArgTy != QualType()) {
2556        if (PointeeTy->isRecordType() &&
2557            Context.typesAreCompatible(SizeOfArgTy, DestTy)) {
2558          DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest,
2559                              PDiag(diag::warn_sizeof_pointer_type_memaccess)
2560                                << FnName << SizeOfArgTy << ArgIdx
2561                                << PointeeTy << Dest->getSourceRange()
2562                                << LenExpr->getSourceRange());
2563          break;
2564        }
2565      }
2566
2567      // Always complain about dynamic classes.
2568      if (isDynamicClassType(PointeeTy)) {
2569
2570        unsigned OperationType = 0;
2571        // "overwritten" if we're warning about the destination for any call
2572        // but memcmp; otherwise a verb appropriate to the call.
2573        if (ArgIdx != 0 || BId == Builtin::BImemcmp) {
2574          if (BId == Builtin::BImemcpy)
2575            OperationType = 1;
2576          else if(BId == Builtin::BImemmove)
2577            OperationType = 2;
2578          else if (BId == Builtin::BImemcmp)
2579            OperationType = 3;
2580        }
2581
2582        DiagRuntimeBehavior(
2583          Dest->getExprLoc(), Dest,
2584          PDiag(diag::warn_dyn_class_memaccess)
2585            << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx)
2586            << FnName << PointeeTy
2587            << OperationType
2588            << Call->getCallee()->getSourceRange());
2589      } else if (PointeeTy.hasNonTrivialObjCLifetime() &&
2590               BId != Builtin::BImemset)
2591        DiagRuntimeBehavior(
2592          Dest->getExprLoc(), Dest,
2593          PDiag(diag::warn_arc_object_memaccess)
2594            << ArgIdx << FnName << PointeeTy
2595            << Call->getCallee()->getSourceRange());
2596      else
2597        continue;
2598
2599      DiagRuntimeBehavior(
2600        Dest->getExprLoc(), Dest,
2601        PDiag(diag::note_bad_memaccess_silence)
2602          << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
2603      break;
2604    }
2605  }
2606}
2607
2608// A little helper routine: ignore addition and subtraction of integer literals.
2609// This intentionally does not ignore all integer constant expressions because
2610// we don't want to remove sizeof().
2611static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) {
2612  Ex = Ex->IgnoreParenCasts();
2613
2614  for (;;) {
2615    const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex);
2616    if (!BO || !BO->isAdditiveOp())
2617      break;
2618
2619    const Expr *RHS = BO->getRHS()->IgnoreParenCasts();
2620    const Expr *LHS = BO->getLHS()->IgnoreParenCasts();
2621
2622    if (isa<IntegerLiteral>(RHS))
2623      Ex = LHS;
2624    else if (isa<IntegerLiteral>(LHS))
2625      Ex = RHS;
2626    else
2627      break;
2628  }
2629
2630  return Ex;
2631}
2632
2633// Warn if the user has made the 'size' argument to strlcpy or strlcat
2634// be the size of the source, instead of the destination.
2635void Sema::CheckStrlcpycatArguments(const CallExpr *Call,
2636                                    IdentifierInfo *FnName) {
2637
2638  // Don't crash if the user has the wrong number of arguments
2639  if (Call->getNumArgs() != 3)
2640    return;
2641
2642  const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context);
2643  const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context);
2644  const Expr *CompareWithSrc = NULL;
2645
2646  // Look for 'strlcpy(dst, x, sizeof(x))'
2647  if (const Expr *Ex = getSizeOfExprArg(SizeArg))
2648    CompareWithSrc = Ex;
2649  else {
2650    // Look for 'strlcpy(dst, x, strlen(x))'
2651    if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) {
2652      if (SizeCall->isBuiltinCall() == Builtin::BIstrlen
2653          && SizeCall->getNumArgs() == 1)
2654        CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context);
2655    }
2656  }
2657
2658  if (!CompareWithSrc)
2659    return;
2660
2661  // Determine if the argument to sizeof/strlen is equal to the source
2662  // argument.  In principle there's all kinds of things you could do
2663  // here, for instance creating an == expression and evaluating it with
2664  // EvaluateAsBooleanCondition, but this uses a more direct technique:
2665  const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg);
2666  if (!SrcArgDRE)
2667    return;
2668
2669  const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc);
2670  if (!CompareWithSrcDRE ||
2671      SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl())
2672    return;
2673
2674  const Expr *OriginalSizeArg = Call->getArg(2);
2675  Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size)
2676    << OriginalSizeArg->getSourceRange() << FnName;
2677
2678  // Output a FIXIT hint if the destination is an array (rather than a
2679  // pointer to an array).  This could be enhanced to handle some
2680  // pointers if we know the actual size, like if DstArg is 'array+2'
2681  // we could say 'sizeof(array)-2'.
2682  const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts();
2683  QualType DstArgTy = DstArg->getType();
2684
2685  // Only handle constant-sized or VLAs, but not flexible members.
2686  if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) {
2687    // Only issue the FIXIT for arrays of size > 1.
2688    if (CAT->getSize().getSExtValue() <= 1)
2689      return;
2690  } else if (!DstArgTy->isVariableArrayType()) {
2691    return;
2692  }
2693
2694  SmallString<128> sizeString;
2695  llvm::raw_svector_ostream OS(sizeString);
2696  OS << "sizeof(";
2697  DstArg->printPretty(OS, Context, 0, getPrintingPolicy());
2698  OS << ")";
2699
2700  Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size)
2701    << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(),
2702                                    OS.str());
2703}
2704
2705/// Check if two expressions refer to the same declaration.
2706static bool referToTheSameDecl(const Expr *E1, const Expr *E2) {
2707  if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1))
2708    if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2))
2709      return D1->getDecl() == D2->getDecl();
2710  return false;
2711}
2712
2713static const Expr *getStrlenExprArg(const Expr *E) {
2714  if (const CallExpr *CE = dyn_cast<CallExpr>(E)) {
2715    const FunctionDecl *FD = CE->getDirectCallee();
2716    if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen)
2717      return 0;
2718    return CE->getArg(0)->IgnoreParenCasts();
2719  }
2720  return 0;
2721}
2722
2723// Warn on anti-patterns as the 'size' argument to strncat.
2724// The correct size argument should look like following:
2725//   strncat(dst, src, sizeof(dst) - strlen(dest) - 1);
2726void Sema::CheckStrncatArguments(const CallExpr *CE,
2727                                 IdentifierInfo *FnName) {
2728  // Don't crash if the user has the wrong number of arguments.
2729  if (CE->getNumArgs() < 3)
2730    return;
2731  const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts();
2732  const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts();
2733  const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts();
2734
2735  // Identify common expressions, which are wrongly used as the size argument
2736  // to strncat and may lead to buffer overflows.
2737  unsigned PatternType = 0;
2738  if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) {
2739    // - sizeof(dst)
2740    if (referToTheSameDecl(SizeOfArg, DstArg))
2741      PatternType = 1;
2742    // - sizeof(src)
2743    else if (referToTheSameDecl(SizeOfArg, SrcArg))
2744      PatternType = 2;
2745  } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) {
2746    if (BE->getOpcode() == BO_Sub) {
2747      const Expr *L = BE->getLHS()->IgnoreParenCasts();
2748      const Expr *R = BE->getRHS()->IgnoreParenCasts();
2749      // - sizeof(dst) - strlen(dst)
2750      if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) &&
2751          referToTheSameDecl(DstArg, getStrlenExprArg(R)))
2752        PatternType = 1;
2753      // - sizeof(src) - (anything)
2754      else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L)))
2755        PatternType = 2;
2756    }
2757  }
2758
2759  if (PatternType == 0)
2760    return;
2761
2762  // Generate the diagnostic.
2763  SourceLocation SL = LenArg->getLocStart();
2764  SourceRange SR = LenArg->getSourceRange();
2765  SourceManager &SM  = PP.getSourceManager();
2766
2767  // If the function is defined as a builtin macro, do not show macro expansion.
2768  if (SM.isMacroArgExpansion(SL)) {
2769    SL = SM.getSpellingLoc(SL);
2770    SR = SourceRange(SM.getSpellingLoc(SR.getBegin()),
2771                     SM.getSpellingLoc(SR.getEnd()));
2772  }
2773
2774  if (PatternType == 1)
2775    Diag(SL, diag::warn_strncat_large_size) << SR;
2776  else
2777    Diag(SL, diag::warn_strncat_src_size) << SR;
2778
2779  // Output a FIXIT hint if the destination is an array (rather than a
2780  // pointer to an array).  This could be enhanced to handle some
2781  // pointers if we know the actual size, like if DstArg is 'array+2'
2782  // we could say 'sizeof(array)-2'.
2783  QualType DstArgTy = DstArg->getType();
2784
2785  // Only handle constant-sized or VLAs, but not flexible members.
2786  if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) {
2787    // Only issue the FIXIT for arrays of size > 1.
2788    if (CAT->getSize().getSExtValue() <= 1)
2789      return;
2790  } else if (!DstArgTy->isVariableArrayType()) {
2791    return;
2792  }
2793
2794  SmallString<128> sizeString;
2795  llvm::raw_svector_ostream OS(sizeString);
2796  OS << "sizeof(";
2797  DstArg->printPretty(OS, Context, 0, getPrintingPolicy());
2798  OS << ") - ";
2799  OS << "strlen(";
2800  DstArg->printPretty(OS, Context, 0, getPrintingPolicy());
2801  OS << ") - 1";
2802
2803  Diag(SL, diag::note_strncat_wrong_size)
2804    << FixItHint::CreateReplacement(SR, OS.str());
2805}
2806
2807//===--- CHECK: Return Address of Stack Variable --------------------------===//
2808
2809static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars);
2810static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars);
2811
2812/// CheckReturnStackAddr - Check if a return statement returns the address
2813///   of a stack variable.
2814void
2815Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
2816                           SourceLocation ReturnLoc) {
2817
2818  Expr *stackE = 0;
2819  SmallVector<DeclRefExpr *, 8> refVars;
2820
2821  // Perform checking for returned stack addresses, local blocks,
2822  // label addresses or references to temporaries.
2823  if (lhsType->isPointerType() ||
2824      (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) {
2825    stackE = EvalAddr(RetValExp, refVars);
2826  } else if (lhsType->isReferenceType()) {
2827    stackE = EvalVal(RetValExp, refVars);
2828  }
2829
2830  if (stackE == 0)
2831    return; // Nothing suspicious was found.
2832
2833  SourceLocation diagLoc;
2834  SourceRange diagRange;
2835  if (refVars.empty()) {
2836    diagLoc = stackE->getLocStart();
2837    diagRange = stackE->getSourceRange();
2838  } else {
2839    // We followed through a reference variable. 'stackE' contains the
2840    // problematic expression but we will warn at the return statement pointing
2841    // at the reference variable. We will later display the "trail" of
2842    // reference variables using notes.
2843    diagLoc = refVars[0]->getLocStart();
2844    diagRange = refVars[0]->getSourceRange();
2845  }
2846
2847  if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
2848    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
2849                                             : diag::warn_ret_stack_addr)
2850     << DR->getDecl()->getDeclName() << diagRange;
2851  } else if (isa<BlockExpr>(stackE)) { // local block.
2852    Diag(diagLoc, diag::err_ret_local_block) << diagRange;
2853  } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
2854    Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
2855  } else { // local temporary.
2856    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
2857                                             : diag::warn_ret_local_temp_addr)
2858     << diagRange;
2859  }
2860
2861  // Display the "trail" of reference variables that we followed until we
2862  // found the problematic expression using notes.
2863  for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
2864    VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
2865    // If this var binds to another reference var, show the range of the next
2866    // var, otherwise the var binds to the problematic expression, in which case
2867    // show the range of the expression.
2868    SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
2869                                  : stackE->getSourceRange();
2870    Diag(VD->getLocation(), diag::note_ref_var_local_bind)
2871      << VD->getDeclName() << range;
2872  }
2873}
2874
2875/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
2876///  check if the expression in a return statement evaluates to an address
2877///  to a location on the stack, a local block, an address of a label, or a
2878///  reference to local temporary. The recursion is used to traverse the
2879///  AST of the return expression, with recursion backtracking when we
2880///  encounter a subexpression that (1) clearly does not lead to one of the
2881///  above problematic expressions (2) is something we cannot determine leads to
2882///  a problematic expression based on such local checking.
2883///
2884///  Both EvalAddr and EvalVal follow through reference variables to evaluate
2885///  the expression that they point to. Such variables are added to the
2886///  'refVars' vector so that we know what the reference variable "trail" was.
2887///
2888///  EvalAddr processes expressions that are pointers that are used as
2889///  references (and not L-values).  EvalVal handles all other values.
2890///  At the base case of the recursion is a check for the above problematic
2891///  expressions.
2892///
2893///  This implementation handles:
2894///
2895///   * pointer-to-pointer casts
2896///   * implicit conversions from array references to pointers
2897///   * taking the address of fields
2898///   * arbitrary interplay between "&" and "*" operators
2899///   * pointer arithmetic from an address of a stack variable
2900///   * taking the address of an array element where the array is on the stack
2901static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) {
2902  if (E->isTypeDependent())
2903      return NULL;
2904
2905  // We should only be called for evaluating pointer expressions.
2906  assert((E->getType()->isAnyPointerType() ||
2907          E->getType()->isBlockPointerType() ||
2908          E->getType()->isObjCQualifiedIdType()) &&
2909         "EvalAddr only works on pointers");
2910
2911  E = E->IgnoreParens();
2912
2913  // Our "symbolic interpreter" is just a dispatch off the currently
2914  // viewed AST node.  We then recursively traverse the AST by calling
2915  // EvalAddr and EvalVal appropriately.
2916  switch (E->getStmtClass()) {
2917  case Stmt::DeclRefExprClass: {
2918    DeclRefExpr *DR = cast<DeclRefExpr>(E);
2919
2920    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2921      // If this is a reference variable, follow through to the expression that
2922      // it points to.
2923      if (V->hasLocalStorage() &&
2924          V->getType()->isReferenceType() && V->hasInit()) {
2925        // Add the reference variable to the "trail".
2926        refVars.push_back(DR);
2927        return EvalAddr(V->getInit(), refVars);
2928      }
2929
2930    return NULL;
2931  }
2932
2933  case Stmt::UnaryOperatorClass: {
2934    // The only unary operator that make sense to handle here
2935    // is AddrOf.  All others don't make sense as pointers.
2936    UnaryOperator *U = cast<UnaryOperator>(E);
2937
2938    if (U->getOpcode() == UO_AddrOf)
2939      return EvalVal(U->getSubExpr(), refVars);
2940    else
2941      return NULL;
2942  }
2943
2944  case Stmt::BinaryOperatorClass: {
2945    // Handle pointer arithmetic.  All other binary operators are not valid
2946    // in this context.
2947    BinaryOperator *B = cast<BinaryOperator>(E);
2948    BinaryOperatorKind op = B->getOpcode();
2949
2950    if (op != BO_Add && op != BO_Sub)
2951      return NULL;
2952
2953    Expr *Base = B->getLHS();
2954
2955    // Determine which argument is the real pointer base.  It could be
2956    // the RHS argument instead of the LHS.
2957    if (!Base->getType()->isPointerType()) Base = B->getRHS();
2958
2959    assert (Base->getType()->isPointerType());
2960    return EvalAddr(Base, refVars);
2961  }
2962
2963  // For conditional operators we need to see if either the LHS or RHS are
2964  // valid DeclRefExpr*s.  If one of them is valid, we return it.
2965  case Stmt::ConditionalOperatorClass: {
2966    ConditionalOperator *C = cast<ConditionalOperator>(E);
2967
2968    // Handle the GNU extension for missing LHS.
2969    if (Expr *lhsExpr = C->getLHS()) {
2970    // In C++, we can have a throw-expression, which has 'void' type.
2971      if (!lhsExpr->getType()->isVoidType())
2972        if (Expr* LHS = EvalAddr(lhsExpr, refVars))
2973          return LHS;
2974    }
2975
2976    // In C++, we can have a throw-expression, which has 'void' type.
2977    if (C->getRHS()->getType()->isVoidType())
2978      return NULL;
2979
2980    return EvalAddr(C->getRHS(), refVars);
2981  }
2982
2983  case Stmt::BlockExprClass:
2984    if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
2985      return E; // local block.
2986    return NULL;
2987
2988  case Stmt::AddrLabelExprClass:
2989    return E; // address of label.
2990
2991  case Stmt::ExprWithCleanupsClass:
2992    return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars);
2993
2994  // For casts, we need to handle conversions from arrays to
2995  // pointer values, and pointer-to-pointer conversions.
2996  case Stmt::ImplicitCastExprClass:
2997  case Stmt::CStyleCastExprClass:
2998  case Stmt::CXXFunctionalCastExprClass:
2999  case Stmt::ObjCBridgedCastExprClass: {
3000    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
3001    QualType T = SubExpr->getType();
3002
3003    if (cast<CastExpr>(E)->getCastKind() == CK_CopyAndAutoreleaseBlockObject)
3004      return 0;
3005    else if (SubExpr->getType()->isPointerType() ||
3006             SubExpr->getType()->isBlockPointerType() ||
3007             SubExpr->getType()->isObjCQualifiedIdType())
3008      return EvalAddr(SubExpr, refVars);
3009    else if (T->isArrayType())
3010      return EvalVal(SubExpr, refVars);
3011    else
3012      return 0;
3013  }
3014
3015  // C++ casts.  For dynamic casts, static casts, and const casts, we
3016  // are always converting from a pointer-to-pointer, so we just blow
3017  // through the cast.  In the case the dynamic cast doesn't fail (and
3018  // return NULL), we take the conservative route and report cases
3019  // where we return the address of a stack variable.  For Reinterpre
3020  // FIXME: The comment about is wrong; we're not always converting
3021  // from pointer to pointer. I'm guessing that this code should also
3022  // handle references to objects.
3023  case Stmt::CXXStaticCastExprClass:
3024  case Stmt::CXXDynamicCastExprClass:
3025  case Stmt::CXXConstCastExprClass:
3026  case Stmt::CXXReinterpretCastExprClass: {
3027      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
3028      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
3029        return EvalAddr(S, refVars);
3030      else
3031        return NULL;
3032  }
3033
3034  case Stmt::MaterializeTemporaryExprClass:
3035    if (Expr *Result = EvalAddr(
3036                         cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
3037                                refVars))
3038      return Result;
3039
3040    return E;
3041
3042  // Everything else: we simply don't reason about them.
3043  default:
3044    return NULL;
3045  }
3046}
3047
3048
3049///  EvalVal - This function is complements EvalAddr in the mutual recursion.
3050///   See the comments for EvalAddr for more details.
3051static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) {
3052do {
3053  // We should only be called for evaluating non-pointer expressions, or
3054  // expressions with a pointer type that are not used as references but instead
3055  // are l-values (e.g., DeclRefExpr with a pointer type).
3056
3057  // Our "symbolic interpreter" is just a dispatch off the currently
3058  // viewed AST node.  We then recursively traverse the AST by calling
3059  // EvalAddr and EvalVal appropriately.
3060
3061  E = E->IgnoreParens();
3062  switch (E->getStmtClass()) {
3063  case Stmt::ImplicitCastExprClass: {
3064    ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
3065    if (IE->getValueKind() == VK_LValue) {
3066      E = IE->getSubExpr();
3067      continue;
3068    }
3069    return NULL;
3070  }
3071
3072  case Stmt::ExprWithCleanupsClass:
3073    return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars);
3074
3075  case Stmt::DeclRefExprClass: {
3076    // When we hit a DeclRefExpr we are looking at code that refers to a
3077    // variable's name. If it's not a reference variable we check if it has
3078    // local storage within the function, and if so, return the expression.
3079    DeclRefExpr *DR = cast<DeclRefExpr>(E);
3080
3081    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
3082      if (V->hasLocalStorage()) {
3083        if (!V->getType()->isReferenceType())
3084          return DR;
3085
3086        // Reference variable, follow through to the expression that
3087        // it points to.
3088        if (V->hasInit()) {
3089          // Add the reference variable to the "trail".
3090          refVars.push_back(DR);
3091          return EvalVal(V->getInit(), refVars);
3092        }
3093      }
3094
3095    return NULL;
3096  }
3097
3098  case Stmt::UnaryOperatorClass: {
3099    // The only unary operator that make sense to handle here
3100    // is Deref.  All others don't resolve to a "name."  This includes
3101    // handling all sorts of rvalues passed to a unary operator.
3102    UnaryOperator *U = cast<UnaryOperator>(E);
3103
3104    if (U->getOpcode() == UO_Deref)
3105      return EvalAddr(U->getSubExpr(), refVars);
3106
3107    return NULL;
3108  }
3109
3110  case Stmt::ArraySubscriptExprClass: {
3111    // Array subscripts are potential references to data on the stack.  We
3112    // retrieve the DeclRefExpr* for the array variable if it indeed
3113    // has local storage.
3114    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars);
3115  }
3116
3117  case Stmt::ConditionalOperatorClass: {
3118    // For conditional operators we need to see if either the LHS or RHS are
3119    // non-NULL Expr's.  If one is non-NULL, we return it.
3120    ConditionalOperator *C = cast<ConditionalOperator>(E);
3121
3122    // Handle the GNU extension for missing LHS.
3123    if (Expr *lhsExpr = C->getLHS())
3124      if (Expr *LHS = EvalVal(lhsExpr, refVars))
3125        return LHS;
3126
3127    return EvalVal(C->getRHS(), refVars);
3128  }
3129
3130  // Accesses to members are potential references to data on the stack.
3131  case Stmt::MemberExprClass: {
3132    MemberExpr *M = cast<MemberExpr>(E);
3133
3134    // Check for indirect access.  We only want direct field accesses.
3135    if (M->isArrow())
3136      return NULL;
3137
3138    // Check whether the member type is itself a reference, in which case
3139    // we're not going to refer to the member, but to what the member refers to.
3140    if (M->getMemberDecl()->getType()->isReferenceType())
3141      return NULL;
3142
3143    return EvalVal(M->getBase(), refVars);
3144  }
3145
3146  case Stmt::MaterializeTemporaryExprClass:
3147    if (Expr *Result = EvalVal(
3148                          cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
3149                               refVars))
3150      return Result;
3151
3152    return E;
3153
3154  default:
3155    // Check that we don't return or take the address of a reference to a
3156    // temporary. This is only useful in C++.
3157    if (!E->isTypeDependent() && E->isRValue())
3158      return E;
3159
3160    // Everything else: we simply don't reason about them.
3161    return NULL;
3162  }
3163} while (true);
3164}
3165
3166//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
3167
3168/// Check for comparisons of floating point operands using != and ==.
3169/// Issue a warning if these are no self-comparisons, as they are not likely
3170/// to do what the programmer intended.
3171void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) {
3172  bool EmitWarning = true;
3173
3174  Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts();
3175  Expr* RightExprSansParen = RHS->IgnoreParenImpCasts();
3176
3177  // Special case: check for x == x (which is OK).
3178  // Do not emit warnings for such cases.
3179  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
3180    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
3181      if (DRL->getDecl() == DRR->getDecl())
3182        EmitWarning = false;
3183
3184
3185  // Special case: check for comparisons against literals that can be exactly
3186  //  represented by APFloat.  In such cases, do not emit a warning.  This
3187  //  is a heuristic: often comparison against such literals are used to
3188  //  detect if a value in a variable has not changed.  This clearly can
3189  //  lead to false negatives.
3190  if (EmitWarning) {
3191    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
3192      if (FLL->isExact())
3193        EmitWarning = false;
3194    } else
3195      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
3196        if (FLR->isExact())
3197          EmitWarning = false;
3198    }
3199  }
3200
3201  // Check for comparisons with builtin types.
3202  if (EmitWarning)
3203    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
3204      if (CL->isBuiltinCall())
3205        EmitWarning = false;
3206
3207  if (EmitWarning)
3208    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
3209      if (CR->isBuiltinCall())
3210        EmitWarning = false;
3211
3212  // Emit the diagnostic.
3213  if (EmitWarning)
3214    Diag(Loc, diag::warn_floatingpoint_eq)
3215      << LHS->getSourceRange() << RHS->getSourceRange();
3216}
3217
3218//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
3219//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
3220
3221namespace {
3222
3223/// Structure recording the 'active' range of an integer-valued
3224/// expression.
3225struct IntRange {
3226  /// The number of bits active in the int.
3227  unsigned Width;
3228
3229  /// True if the int is known not to have negative values.
3230  bool NonNegative;
3231
3232  IntRange(unsigned Width, bool NonNegative)
3233    : Width(Width), NonNegative(NonNegative)
3234  {}
3235
3236  /// Returns the range of the bool type.
3237  static IntRange forBoolType() {
3238    return IntRange(1, true);
3239  }
3240
3241  /// Returns the range of an opaque value of the given integral type.
3242  static IntRange forValueOfType(ASTContext &C, QualType T) {
3243    return forValueOfCanonicalType(C,
3244                          T->getCanonicalTypeInternal().getTypePtr());
3245  }
3246
3247  /// Returns the range of an opaque value of a canonical integral type.
3248  static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
3249    assert(T->isCanonicalUnqualified());
3250
3251    if (const VectorType *VT = dyn_cast<VectorType>(T))
3252      T = VT->getElementType().getTypePtr();
3253    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
3254      T = CT->getElementType().getTypePtr();
3255
3256    // For enum types, use the known bit width of the enumerators.
3257    if (const EnumType *ET = dyn_cast<EnumType>(T)) {
3258      EnumDecl *Enum = ET->getDecl();
3259      if (!Enum->isCompleteDefinition())
3260        return IntRange(C.getIntWidth(QualType(T, 0)), false);
3261
3262      unsigned NumPositive = Enum->getNumPositiveBits();
3263      unsigned NumNegative = Enum->getNumNegativeBits();
3264
3265      return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
3266    }
3267
3268    const BuiltinType *BT = cast<BuiltinType>(T);
3269    assert(BT->isInteger());
3270
3271    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
3272  }
3273
3274  /// Returns the "target" range of a canonical integral type, i.e.
3275  /// the range of values expressible in the type.
3276  ///
3277  /// This matches forValueOfCanonicalType except that enums have the
3278  /// full range of their type, not the range of their enumerators.
3279  static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
3280    assert(T->isCanonicalUnqualified());
3281
3282    if (const VectorType *VT = dyn_cast<VectorType>(T))
3283      T = VT->getElementType().getTypePtr();
3284    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
3285      T = CT->getElementType().getTypePtr();
3286    if (const EnumType *ET = dyn_cast<EnumType>(T))
3287      T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr();
3288
3289    const BuiltinType *BT = cast<BuiltinType>(T);
3290    assert(BT->isInteger());
3291
3292    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
3293  }
3294
3295  /// Returns the supremum of two ranges: i.e. their conservative merge.
3296  static IntRange join(IntRange L, IntRange R) {
3297    return IntRange(std::max(L.Width, R.Width),
3298                    L.NonNegative && R.NonNegative);
3299  }
3300
3301  /// Returns the infinum of two ranges: i.e. their aggressive merge.
3302  static IntRange meet(IntRange L, IntRange R) {
3303    return IntRange(std::min(L.Width, R.Width),
3304                    L.NonNegative || R.NonNegative);
3305  }
3306};
3307
3308static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value,
3309                              unsigned MaxWidth) {
3310  if (value.isSigned() && value.isNegative())
3311    return IntRange(value.getMinSignedBits(), false);
3312
3313  if (value.getBitWidth() > MaxWidth)
3314    value = value.trunc(MaxWidth);
3315
3316  // isNonNegative() just checks the sign bit without considering
3317  // signedness.
3318  return IntRange(value.getActiveBits(), true);
3319}
3320
3321static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
3322                              unsigned MaxWidth) {
3323  if (result.isInt())
3324    return GetValueRange(C, result.getInt(), MaxWidth);
3325
3326  if (result.isVector()) {
3327    IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
3328    for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
3329      IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
3330      R = IntRange::join(R, El);
3331    }
3332    return R;
3333  }
3334
3335  if (result.isComplexInt()) {
3336    IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
3337    IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
3338    return IntRange::join(R, I);
3339  }
3340
3341  // This can happen with lossless casts to intptr_t of "based" lvalues.
3342  // Assume it might use arbitrary bits.
3343  // FIXME: The only reason we need to pass the type in here is to get
3344  // the sign right on this one case.  It would be nice if APValue
3345  // preserved this.
3346  assert(result.isLValue() || result.isAddrLabelDiff());
3347  return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType());
3348}
3349
3350/// Pseudo-evaluate the given integer expression, estimating the
3351/// range of values it might take.
3352///
3353/// \param MaxWidth - the width to which the value will be truncated
3354static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
3355  E = E->IgnoreParens();
3356
3357  // Try a full evaluation first.
3358  Expr::EvalResult result;
3359  if (E->EvaluateAsRValue(result, C))
3360    return GetValueRange(C, result.Val, E->getType(), MaxWidth);
3361
3362  // I think we only want to look through implicit casts here; if the
3363  // user has an explicit widening cast, we should treat the value as
3364  // being of the new, wider type.
3365  if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
3366    if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue)
3367      return GetExprRange(C, CE->getSubExpr(), MaxWidth);
3368
3369    IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
3370
3371    bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
3372
3373    // Assume that non-integer casts can span the full range of the type.
3374    if (!isIntegerCast)
3375      return OutputTypeRange;
3376
3377    IntRange SubRange
3378      = GetExprRange(C, CE->getSubExpr(),
3379                     std::min(MaxWidth, OutputTypeRange.Width));
3380
3381    // Bail out if the subexpr's range is as wide as the cast type.
3382    if (SubRange.Width >= OutputTypeRange.Width)
3383      return OutputTypeRange;
3384
3385    // Otherwise, we take the smaller width, and we're non-negative if
3386    // either the output type or the subexpr is.
3387    return IntRange(SubRange.Width,
3388                    SubRange.NonNegative || OutputTypeRange.NonNegative);
3389  }
3390
3391  if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
3392    // If we can fold the condition, just take that operand.
3393    bool CondResult;
3394    if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
3395      return GetExprRange(C, CondResult ? CO->getTrueExpr()
3396                                        : CO->getFalseExpr(),
3397                          MaxWidth);
3398
3399    // Otherwise, conservatively merge.
3400    IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
3401    IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
3402    return IntRange::join(L, R);
3403  }
3404
3405  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
3406    switch (BO->getOpcode()) {
3407
3408    // Boolean-valued operations are single-bit and positive.
3409    case BO_LAnd:
3410    case BO_LOr:
3411    case BO_LT:
3412    case BO_GT:
3413    case BO_LE:
3414    case BO_GE:
3415    case BO_EQ:
3416    case BO_NE:
3417      return IntRange::forBoolType();
3418
3419    // The type of the assignments is the type of the LHS, so the RHS
3420    // is not necessarily the same type.
3421    case BO_MulAssign:
3422    case BO_DivAssign:
3423    case BO_RemAssign:
3424    case BO_AddAssign:
3425    case BO_SubAssign:
3426    case BO_XorAssign:
3427    case BO_OrAssign:
3428      // TODO: bitfields?
3429      return IntRange::forValueOfType(C, E->getType());
3430
3431    // Simple assignments just pass through the RHS, which will have
3432    // been coerced to the LHS type.
3433    case BO_Assign:
3434      // TODO: bitfields?
3435      return GetExprRange(C, BO->getRHS(), MaxWidth);
3436
3437    // Operations with opaque sources are black-listed.
3438    case BO_PtrMemD:
3439    case BO_PtrMemI:
3440      return IntRange::forValueOfType(C, E->getType());
3441
3442    // Bitwise-and uses the *infinum* of the two source ranges.
3443    case BO_And:
3444    case BO_AndAssign:
3445      return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
3446                            GetExprRange(C, BO->getRHS(), MaxWidth));
3447
3448    // Left shift gets black-listed based on a judgement call.
3449    case BO_Shl:
3450      // ...except that we want to treat '1 << (blah)' as logically
3451      // positive.  It's an important idiom.
3452      if (IntegerLiteral *I
3453            = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
3454        if (I->getValue() == 1) {
3455          IntRange R = IntRange::forValueOfType(C, E->getType());
3456          return IntRange(R.Width, /*NonNegative*/ true);
3457        }
3458      }
3459      // fallthrough
3460
3461    case BO_ShlAssign:
3462      return IntRange::forValueOfType(C, E->getType());
3463
3464    // Right shift by a constant can narrow its left argument.
3465    case BO_Shr:
3466    case BO_ShrAssign: {
3467      IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
3468
3469      // If the shift amount is a positive constant, drop the width by
3470      // that much.
3471      llvm::APSInt shift;
3472      if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
3473          shift.isNonNegative()) {
3474        unsigned zext = shift.getZExtValue();
3475        if (zext >= L.Width)
3476          L.Width = (L.NonNegative ? 0 : 1);
3477        else
3478          L.Width -= zext;
3479      }
3480
3481      return L;
3482    }
3483
3484    // Comma acts as its right operand.
3485    case BO_Comma:
3486      return GetExprRange(C, BO->getRHS(), MaxWidth);
3487
3488    // Black-list pointer subtractions.
3489    case BO_Sub:
3490      if (BO->getLHS()->getType()->isPointerType())
3491        return IntRange::forValueOfType(C, E->getType());
3492      break;
3493
3494    // The width of a division result is mostly determined by the size
3495    // of the LHS.
3496    case BO_Div: {
3497      // Don't 'pre-truncate' the operands.
3498      unsigned opWidth = C.getIntWidth(E->getType());
3499      IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
3500
3501      // If the divisor is constant, use that.
3502      llvm::APSInt divisor;
3503      if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) {
3504        unsigned log2 = divisor.logBase2(); // floor(log_2(divisor))
3505        if (log2 >= L.Width)
3506          L.Width = (L.NonNegative ? 0 : 1);
3507        else
3508          L.Width = std::min(L.Width - log2, MaxWidth);
3509        return L;
3510      }
3511
3512      // Otherwise, just use the LHS's width.
3513      IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
3514      return IntRange(L.Width, L.NonNegative && R.NonNegative);
3515    }
3516
3517    // The result of a remainder can't be larger than the result of
3518    // either side.
3519    case BO_Rem: {
3520      // Don't 'pre-truncate' the operands.
3521      unsigned opWidth = C.getIntWidth(E->getType());
3522      IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
3523      IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
3524
3525      IntRange meet = IntRange::meet(L, R);
3526      meet.Width = std::min(meet.Width, MaxWidth);
3527      return meet;
3528    }
3529
3530    // The default behavior is okay for these.
3531    case BO_Mul:
3532    case BO_Add:
3533    case BO_Xor:
3534    case BO_Or:
3535      break;
3536    }
3537
3538    // The default case is to treat the operation as if it were closed
3539    // on the narrowest type that encompasses both operands.
3540    IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
3541    IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
3542    return IntRange::join(L, R);
3543  }
3544
3545  if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
3546    switch (UO->getOpcode()) {
3547    // Boolean-valued operations are white-listed.
3548    case UO_LNot:
3549      return IntRange::forBoolType();
3550
3551    // Operations with opaque sources are black-listed.
3552    case UO_Deref:
3553    case UO_AddrOf: // should be impossible
3554      return IntRange::forValueOfType(C, E->getType());
3555
3556    default:
3557      return GetExprRange(C, UO->getSubExpr(), MaxWidth);
3558    }
3559  }
3560
3561  if (dyn_cast<OffsetOfExpr>(E)) {
3562    IntRange::forValueOfType(C, E->getType());
3563  }
3564
3565  if (FieldDecl *BitField = E->getBitField())
3566    return IntRange(BitField->getBitWidthValue(C),
3567                    BitField->getType()->isUnsignedIntegerOrEnumerationType());
3568
3569  return IntRange::forValueOfType(C, E->getType());
3570}
3571
3572static IntRange GetExprRange(ASTContext &C, Expr *E) {
3573  return GetExprRange(C, E, C.getIntWidth(E->getType()));
3574}
3575
3576/// Checks whether the given value, which currently has the given
3577/// source semantics, has the same value when coerced through the
3578/// target semantics.
3579static bool IsSameFloatAfterCast(const llvm::APFloat &value,
3580                                 const llvm::fltSemantics &Src,
3581                                 const llvm::fltSemantics &Tgt) {
3582  llvm::APFloat truncated = value;
3583
3584  bool ignored;
3585  truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
3586  truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
3587
3588  return truncated.bitwiseIsEqual(value);
3589}
3590
3591/// Checks whether the given value, which currently has the given
3592/// source semantics, has the same value when coerced through the
3593/// target semantics.
3594///
3595/// The value might be a vector of floats (or a complex number).
3596static bool IsSameFloatAfterCast(const APValue &value,
3597                                 const llvm::fltSemantics &Src,
3598                                 const llvm::fltSemantics &Tgt) {
3599  if (value.isFloat())
3600    return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
3601
3602  if (value.isVector()) {
3603    for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
3604      if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
3605        return false;
3606    return true;
3607  }
3608
3609  assert(value.isComplexFloat());
3610  return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
3611          IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
3612}
3613
3614static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
3615
3616static bool IsZero(Sema &S, Expr *E) {
3617  // Suppress cases where we are comparing against an enum constant.
3618  if (const DeclRefExpr *DR =
3619      dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
3620    if (isa<EnumConstantDecl>(DR->getDecl()))
3621      return false;
3622
3623  // Suppress cases where the '0' value is expanded from a macro.
3624  if (E->getLocStart().isMacroID())
3625    return false;
3626
3627  llvm::APSInt Value;
3628  return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
3629}
3630
3631static bool HasEnumType(Expr *E) {
3632  // Strip off implicit integral promotions.
3633  while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
3634    if (ICE->getCastKind() != CK_IntegralCast &&
3635        ICE->getCastKind() != CK_NoOp)
3636      break;
3637    E = ICE->getSubExpr();
3638  }
3639
3640  return E->getType()->isEnumeralType();
3641}
3642
3643static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
3644  BinaryOperatorKind op = E->getOpcode();
3645  if (E->isValueDependent())
3646    return;
3647
3648  if (op == BO_LT && IsZero(S, E->getRHS())) {
3649    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
3650      << "< 0" << "false" << HasEnumType(E->getLHS())
3651      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3652  } else if (op == BO_GE && IsZero(S, E->getRHS())) {
3653    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
3654      << ">= 0" << "true" << HasEnumType(E->getLHS())
3655      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3656  } else if (op == BO_GT && IsZero(S, E->getLHS())) {
3657    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
3658      << "0 >" << "false" << HasEnumType(E->getRHS())
3659      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3660  } else if (op == BO_LE && IsZero(S, E->getLHS())) {
3661    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
3662      << "0 <=" << "true" << HasEnumType(E->getRHS())
3663      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3664  }
3665}
3666
3667/// Analyze the operands of the given comparison.  Implements the
3668/// fallback case from AnalyzeComparison.
3669static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
3670  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
3671  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
3672}
3673
3674/// \brief Implements -Wsign-compare.
3675///
3676/// \param E the binary operator to check for warnings
3677static void AnalyzeComparison(Sema &S, BinaryOperator *E) {
3678  // The type the comparison is being performed in.
3679  QualType T = E->getLHS()->getType();
3680  assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
3681         && "comparison with mismatched types");
3682
3683  // We don't do anything special if this isn't an unsigned integral
3684  // comparison:  we're only interested in integral comparisons, and
3685  // signed comparisons only happen in cases we don't care to warn about.
3686  //
3687  // We also don't care about value-dependent expressions or expressions
3688  // whose result is a constant.
3689  if (!T->hasUnsignedIntegerRepresentation()
3690      || E->isValueDependent() || E->isIntegerConstantExpr(S.Context))
3691    return AnalyzeImpConvsInComparison(S, E);
3692
3693  Expr *LHS = E->getLHS()->IgnoreParenImpCasts();
3694  Expr *RHS = E->getRHS()->IgnoreParenImpCasts();
3695
3696  // Check to see if one of the (unmodified) operands is of different
3697  // signedness.
3698  Expr *signedOperand, *unsignedOperand;
3699  if (LHS->getType()->hasSignedIntegerRepresentation()) {
3700    assert(!RHS->getType()->hasSignedIntegerRepresentation() &&
3701           "unsigned comparison between two signed integer expressions?");
3702    signedOperand = LHS;
3703    unsignedOperand = RHS;
3704  } else if (RHS->getType()->hasSignedIntegerRepresentation()) {
3705    signedOperand = RHS;
3706    unsignedOperand = LHS;
3707  } else {
3708    CheckTrivialUnsignedComparison(S, E);
3709    return AnalyzeImpConvsInComparison(S, E);
3710  }
3711
3712  // Otherwise, calculate the effective range of the signed operand.
3713  IntRange signedRange = GetExprRange(S.Context, signedOperand);
3714
3715  // Go ahead and analyze implicit conversions in the operands.  Note
3716  // that we skip the implicit conversions on both sides.
3717  AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc());
3718  AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc());
3719
3720  // If the signed range is non-negative, -Wsign-compare won't fire,
3721  // but we should still check for comparisons which are always true
3722  // or false.
3723  if (signedRange.NonNegative)
3724    return CheckTrivialUnsignedComparison(S, E);
3725
3726  // For (in)equality comparisons, if the unsigned operand is a
3727  // constant which cannot collide with a overflowed signed operand,
3728  // then reinterpreting the signed operand as unsigned will not
3729  // change the result of the comparison.
3730  if (E->isEqualityOp()) {
3731    unsigned comparisonWidth = S.Context.getIntWidth(T);
3732    IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
3733
3734    // We should never be unable to prove that the unsigned operand is
3735    // non-negative.
3736    assert(unsignedRange.NonNegative && "unsigned range includes negative?");
3737
3738    if (unsignedRange.Width < comparisonWidth)
3739      return;
3740  }
3741
3742  S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison)
3743    << LHS->getType() << RHS->getType()
3744    << LHS->getSourceRange() << RHS->getSourceRange();
3745}
3746
3747/// Analyzes an attempt to assign the given value to a bitfield.
3748///
3749/// Returns true if there was something fishy about the attempt.
3750static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
3751                                      SourceLocation InitLoc) {
3752  assert(Bitfield->isBitField());
3753  if (Bitfield->isInvalidDecl())
3754    return false;
3755
3756  // White-list bool bitfields.
3757  if (Bitfield->getType()->isBooleanType())
3758    return false;
3759
3760  // Ignore value- or type-dependent expressions.
3761  if (Bitfield->getBitWidth()->isValueDependent() ||
3762      Bitfield->getBitWidth()->isTypeDependent() ||
3763      Init->isValueDependent() ||
3764      Init->isTypeDependent())
3765    return false;
3766
3767  Expr *OriginalInit = Init->IgnoreParenImpCasts();
3768
3769  llvm::APSInt Value;
3770  if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects))
3771    return false;
3772
3773  unsigned OriginalWidth = Value.getBitWidth();
3774  unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context);
3775
3776  if (OriginalWidth <= FieldWidth)
3777    return false;
3778
3779  // Compute the value which the bitfield will contain.
3780  llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
3781  TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType());
3782
3783  // Check whether the stored value is equal to the original value.
3784  TruncatedValue = TruncatedValue.extend(OriginalWidth);
3785  if (Value == TruncatedValue)
3786    return false;
3787
3788  // Special-case bitfields of width 1: booleans are naturally 0/1, and
3789  // therefore don't strictly fit into a signed bitfield of width 1.
3790  if (FieldWidth == 1 && Value == 1)
3791    return false;
3792
3793  std::string PrettyValue = Value.toString(10);
3794  std::string PrettyTrunc = TruncatedValue.toString(10);
3795
3796  S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
3797    << PrettyValue << PrettyTrunc << OriginalInit->getType()
3798    << Init->getSourceRange();
3799
3800  return true;
3801}
3802
3803/// Analyze the given simple or compound assignment for warning-worthy
3804/// operations.
3805static void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
3806  // Just recurse on the LHS.
3807  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
3808
3809  // We want to recurse on the RHS as normal unless we're assigning to
3810  // a bitfield.
3811  if (FieldDecl *Bitfield = E->getLHS()->getBitField()) {
3812    if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
3813                                  E->getOperatorLoc())) {
3814      // Recurse, ignoring any implicit conversions on the RHS.
3815      return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
3816                                        E->getOperatorLoc());
3817    }
3818  }
3819
3820  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
3821}
3822
3823/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
3824static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
3825                            SourceLocation CContext, unsigned diag,
3826                            bool pruneControlFlow = false) {
3827  if (pruneControlFlow) {
3828    S.DiagRuntimeBehavior(E->getExprLoc(), E,
3829                          S.PDiag(diag)
3830                            << SourceType << T << E->getSourceRange()
3831                            << SourceRange(CContext));
3832    return;
3833  }
3834  S.Diag(E->getExprLoc(), diag)
3835    << SourceType << T << E->getSourceRange() << SourceRange(CContext);
3836}
3837
3838/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
3839static void DiagnoseImpCast(Sema &S, Expr *E, QualType T,
3840                            SourceLocation CContext, unsigned diag,
3841                            bool pruneControlFlow = false) {
3842  DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow);
3843}
3844
3845/// Diagnose an implicit cast from a literal expression. Does not warn when the
3846/// cast wouldn't lose information.
3847void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
3848                                    SourceLocation CContext) {
3849  // Try to convert the literal exactly to an integer. If we can, don't warn.
3850  bool isExact = false;
3851  const llvm::APFloat &Value = FL->getValue();
3852  llvm::APSInt IntegerValue(S.Context.getIntWidth(T),
3853                            T->hasUnsignedIntegerRepresentation());
3854  if (Value.convertToInteger(IntegerValue,
3855                             llvm::APFloat::rmTowardZero, &isExact)
3856      == llvm::APFloat::opOK && isExact)
3857    return;
3858
3859  S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
3860    << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext);
3861}
3862
3863std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
3864  if (!Range.Width) return "0";
3865
3866  llvm::APSInt ValueInRange = Value;
3867  ValueInRange.setIsSigned(!Range.NonNegative);
3868  ValueInRange = ValueInRange.trunc(Range.Width);
3869  return ValueInRange.toString(10);
3870}
3871
3872void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
3873                             SourceLocation CC, bool *ICContext = 0) {
3874  if (E->isTypeDependent() || E->isValueDependent()) return;
3875
3876  const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
3877  const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
3878  if (Source == Target) return;
3879  if (Target->isDependentType()) return;
3880
3881  // If the conversion context location is invalid don't complain. We also
3882  // don't want to emit a warning if the issue occurs from the expansion of
3883  // a system macro. The problem is that 'getSpellingLoc()' is slow, so we
3884  // delay this check as long as possible. Once we detect we are in that
3885  // scenario, we just return.
3886  if (CC.isInvalid())
3887    return;
3888
3889  // Diagnose implicit casts to bool.
3890  if (Target->isSpecificBuiltinType(BuiltinType::Bool)) {
3891    if (isa<StringLiteral>(E))
3892      // Warn on string literal to bool.  Checks for string literals in logical
3893      // expressions, for instances, assert(0 && "error here"), is prevented
3894      // by a check in AnalyzeImplicitConversions().
3895      return DiagnoseImpCast(S, E, T, CC,
3896                             diag::warn_impcast_string_literal_to_bool);
3897    if (Source->isFunctionType()) {
3898      // Warn on function to bool. Checks free functions and static member
3899      // functions. Weakly imported functions are excluded from the check,
3900      // since it's common to test their value to check whether the linker
3901      // found a definition for them.
3902      ValueDecl *D = 0;
3903      if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) {
3904        D = R->getDecl();
3905      } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) {
3906        D = M->getMemberDecl();
3907      }
3908
3909      if (D && !D->isWeak()) {
3910        if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) {
3911          S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool)
3912            << F << E->getSourceRange() << SourceRange(CC);
3913          S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence)
3914            << FixItHint::CreateInsertion(E->getExprLoc(), "&");
3915          QualType ReturnType;
3916          UnresolvedSet<4> NonTemplateOverloads;
3917          S.isExprCallable(*E, ReturnType, NonTemplateOverloads);
3918          if (!ReturnType.isNull()
3919              && ReturnType->isSpecificBuiltinType(BuiltinType::Bool))
3920            S.Diag(E->getExprLoc(), diag::note_function_to_bool_call)
3921              << FixItHint::CreateInsertion(
3922                 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()");
3923          return;
3924        }
3925      }
3926    }
3927    return; // Other casts to bool are not checked.
3928  }
3929
3930  // Strip vector types.
3931  if (isa<VectorType>(Source)) {
3932    if (!isa<VectorType>(Target)) {
3933      if (S.SourceMgr.isInSystemMacro(CC))
3934        return;
3935      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
3936    }
3937
3938    // If the vector cast is cast between two vectors of the same size, it is
3939    // a bitcast, not a conversion.
3940    if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target))
3941      return;
3942
3943    Source = cast<VectorType>(Source)->getElementType().getTypePtr();
3944    Target = cast<VectorType>(Target)->getElementType().getTypePtr();
3945  }
3946
3947  // Strip complex types.
3948  if (isa<ComplexType>(Source)) {
3949    if (!isa<ComplexType>(Target)) {
3950      if (S.SourceMgr.isInSystemMacro(CC))
3951        return;
3952
3953      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
3954    }
3955
3956    Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
3957    Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
3958  }
3959
3960  const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
3961  const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
3962
3963  // If the source is floating point...
3964  if (SourceBT && SourceBT->isFloatingPoint()) {
3965    // ...and the target is floating point...
3966    if (TargetBT && TargetBT->isFloatingPoint()) {
3967      // ...then warn if we're dropping FP rank.
3968
3969      // Builtin FP kinds are ordered by increasing FP rank.
3970      if (SourceBT->getKind() > TargetBT->getKind()) {
3971        // Don't warn about float constants that are precisely
3972        // representable in the target type.
3973        Expr::EvalResult result;
3974        if (E->EvaluateAsRValue(result, S.Context)) {
3975          // Value might be a float, a float vector, or a float complex.
3976          if (IsSameFloatAfterCast(result.Val,
3977                   S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
3978                   S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
3979            return;
3980        }
3981
3982        if (S.SourceMgr.isInSystemMacro(CC))
3983          return;
3984
3985        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
3986      }
3987      return;
3988    }
3989
3990    // If the target is integral, always warn.
3991    if ((TargetBT && TargetBT->isInteger())) {
3992      if (S.SourceMgr.isInSystemMacro(CC))
3993        return;
3994
3995      Expr *InnerE = E->IgnoreParenImpCasts();
3996      // We also want to warn on, e.g., "int i = -1.234"
3997      if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE))
3998        if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus)
3999          InnerE = UOp->getSubExpr()->IgnoreParenImpCasts();
4000
4001      if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
4002        DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
4003      } else {
4004        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
4005      }
4006    }
4007
4008    return;
4009  }
4010
4011  if (!Source->isIntegerType() || !Target->isIntegerType())
4012    return;
4013
4014  if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)
4015           == Expr::NPCK_GNUNull) && Target->isIntegerType()) {
4016    S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer)
4017        << E->getSourceRange() << clang::SourceRange(CC);
4018    return;
4019  }
4020
4021  IntRange SourceRange = GetExprRange(S.Context, E);
4022  IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
4023
4024  if (SourceRange.Width > TargetRange.Width) {
4025    // If the source is a constant, use a default-on diagnostic.
4026    // TODO: this should happen for bitfield stores, too.
4027    llvm::APSInt Value(32);
4028    if (E->isIntegerConstantExpr(Value, S.Context)) {
4029      if (S.SourceMgr.isInSystemMacro(CC))
4030        return;
4031
4032      std::string PrettySourceValue = Value.toString(10);
4033      std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
4034
4035      S.DiagRuntimeBehavior(E->getExprLoc(), E,
4036        S.PDiag(diag::warn_impcast_integer_precision_constant)
4037            << PrettySourceValue << PrettyTargetValue
4038            << E->getType() << T << E->getSourceRange()
4039            << clang::SourceRange(CC));
4040      return;
4041    }
4042
4043    // People want to build with -Wshorten-64-to-32 and not -Wconversion.
4044    if (S.SourceMgr.isInSystemMacro(CC))
4045      return;
4046
4047    if (SourceRange.Width == 64 && TargetRange.Width == 32)
4048      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32,
4049                             /* pruneControlFlow */ true);
4050    return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
4051  }
4052
4053  if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
4054      (!TargetRange.NonNegative && SourceRange.NonNegative &&
4055       SourceRange.Width == TargetRange.Width)) {
4056
4057    if (S.SourceMgr.isInSystemMacro(CC))
4058      return;
4059
4060    unsigned DiagID = diag::warn_impcast_integer_sign;
4061
4062    // Traditionally, gcc has warned about this under -Wsign-compare.
4063    // We also want to warn about it in -Wconversion.
4064    // So if -Wconversion is off, use a completely identical diagnostic
4065    // in the sign-compare group.
4066    // The conditional-checking code will
4067    if (ICContext) {
4068      DiagID = diag::warn_impcast_integer_sign_conditional;
4069      *ICContext = true;
4070    }
4071
4072    return DiagnoseImpCast(S, E, T, CC, DiagID);
4073  }
4074
4075  // Diagnose conversions between different enumeration types.
4076  // In C, we pretend that the type of an EnumConstantDecl is its enumeration
4077  // type, to give us better diagnostics.
4078  QualType SourceType = E->getType();
4079  if (!S.getLangOptions().CPlusPlus) {
4080    if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
4081      if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
4082        EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
4083        SourceType = S.Context.getTypeDeclType(Enum);
4084        Source = S.Context.getCanonicalType(SourceType).getTypePtr();
4085      }
4086  }
4087
4088  if (const EnumType *SourceEnum = Source->getAs<EnumType>())
4089    if (const EnumType *TargetEnum = Target->getAs<EnumType>())
4090      if ((SourceEnum->getDecl()->getIdentifier() ||
4091           SourceEnum->getDecl()->getTypedefNameForAnonDecl()) &&
4092          (TargetEnum->getDecl()->getIdentifier() ||
4093           TargetEnum->getDecl()->getTypedefNameForAnonDecl()) &&
4094          SourceEnum != TargetEnum) {
4095        if (S.SourceMgr.isInSystemMacro(CC))
4096          return;
4097
4098        return DiagnoseImpCast(S, E, SourceType, T, CC,
4099                               diag::warn_impcast_different_enum_types);
4100      }
4101
4102  return;
4103}
4104
4105void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T);
4106
4107void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
4108                             SourceLocation CC, bool &ICContext) {
4109  E = E->IgnoreParenImpCasts();
4110
4111  if (isa<ConditionalOperator>(E))
4112    return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T);
4113
4114  AnalyzeImplicitConversions(S, E, CC);
4115  if (E->getType() != T)
4116    return CheckImplicitConversion(S, E, T, CC, &ICContext);
4117  return;
4118}
4119
4120void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) {
4121  SourceLocation CC = E->getQuestionLoc();
4122
4123  AnalyzeImplicitConversions(S, E->getCond(), CC);
4124
4125  bool Suspicious = false;
4126  CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
4127  CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
4128
4129  // If -Wconversion would have warned about either of the candidates
4130  // for a signedness conversion to the context type...
4131  if (!Suspicious) return;
4132
4133  // ...but it's currently ignored...
4134  if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
4135                                 CC))
4136    return;
4137
4138  // ...then check whether it would have warned about either of the
4139  // candidates for a signedness conversion to the condition type.
4140  if (E->getType() == T) return;
4141
4142  Suspicious = false;
4143  CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
4144                          E->getType(), CC, &Suspicious);
4145  if (!Suspicious)
4146    CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
4147                            E->getType(), CC, &Suspicious);
4148}
4149
4150/// AnalyzeImplicitConversions - Find and report any interesting
4151/// implicit conversions in the given expression.  There are a couple
4152/// of competing diagnostics here, -Wconversion and -Wsign-compare.
4153void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
4154  QualType T = OrigE->getType();
4155  Expr *E = OrigE->IgnoreParenImpCasts();
4156
4157  if (E->isTypeDependent() || E->isValueDependent())
4158    return;
4159
4160  // For conditional operators, we analyze the arguments as if they
4161  // were being fed directly into the output.
4162  if (isa<ConditionalOperator>(E)) {
4163    ConditionalOperator *CO = cast<ConditionalOperator>(E);
4164    CheckConditionalOperator(S, CO, T);
4165    return;
4166  }
4167
4168  // Go ahead and check any implicit conversions we might have skipped.
4169  // The non-canonical typecheck is just an optimization;
4170  // CheckImplicitConversion will filter out dead implicit conversions.
4171  if (E->getType() != T)
4172    CheckImplicitConversion(S, E, T, CC);
4173
4174  // Now continue drilling into this expression.
4175
4176  // Skip past explicit casts.
4177  if (isa<ExplicitCastExpr>(E)) {
4178    E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
4179    return AnalyzeImplicitConversions(S, E, CC);
4180  }
4181
4182  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
4183    // Do a somewhat different check with comparison operators.
4184    if (BO->isComparisonOp())
4185      return AnalyzeComparison(S, BO);
4186
4187    // And with simple assignments.
4188    if (BO->getOpcode() == BO_Assign)
4189      return AnalyzeAssignment(S, BO);
4190  }
4191
4192  // These break the otherwise-useful invariant below.  Fortunately,
4193  // we don't really need to recurse into them, because any internal
4194  // expressions should have been analyzed already when they were
4195  // built into statements.
4196  if (isa<StmtExpr>(E)) return;
4197
4198  // Don't descend into unevaluated contexts.
4199  if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
4200
4201  // Now just recurse over the expression's children.
4202  CC = E->getExprLoc();
4203  BinaryOperator *BO = dyn_cast<BinaryOperator>(E);
4204  bool IsLogicalOperator = BO && BO->isLogicalOp();
4205  for (Stmt::child_range I = E->children(); I; ++I) {
4206    Expr *ChildExpr = dyn_cast_or_null<Expr>(*I);
4207    if (!ChildExpr)
4208      continue;
4209
4210    if (IsLogicalOperator &&
4211        isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts()))
4212      // Ignore checking string literals that are in logical operators.
4213      continue;
4214    AnalyzeImplicitConversions(S, ChildExpr, CC);
4215  }
4216}
4217
4218} // end anonymous namespace
4219
4220/// Diagnoses "dangerous" implicit conversions within the given
4221/// expression (which is a full expression).  Implements -Wconversion
4222/// and -Wsign-compare.
4223///
4224/// \param CC the "context" location of the implicit conversion, i.e.
4225///   the most location of the syntactic entity requiring the implicit
4226///   conversion
4227void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
4228  // Don't diagnose in unevaluated contexts.
4229  if (ExprEvalContexts.back().Context == Sema::Unevaluated)
4230    return;
4231
4232  // Don't diagnose for value- or type-dependent expressions.
4233  if (E->isTypeDependent() || E->isValueDependent())
4234    return;
4235
4236  // Check for array bounds violations in cases where the check isn't triggered
4237  // elsewhere for other Expr types (like BinaryOperators), e.g. when an
4238  // ArraySubscriptExpr is on the RHS of a variable initialization.
4239  CheckArrayAccess(E);
4240
4241  // This is not the right CC for (e.g.) a variable initialization.
4242  AnalyzeImplicitConversions(*this, E, CC);
4243}
4244
4245void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
4246                                       FieldDecl *BitField,
4247                                       Expr *Init) {
4248  (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
4249}
4250
4251/// CheckParmsForFunctionDef - Check that the parameters of the given
4252/// function are appropriate for the definition of a function. This
4253/// takes care of any checks that cannot be performed on the
4254/// declaration itself, e.g., that the types of each of the function
4255/// parameters are complete.
4256bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
4257                                    bool CheckParameterNames) {
4258  bool HasInvalidParm = false;
4259  for (; P != PEnd; ++P) {
4260    ParmVarDecl *Param = *P;
4261
4262    // C99 6.7.5.3p4: the parameters in a parameter type list in a
4263    // function declarator that is part of a function definition of
4264    // that function shall not have incomplete type.
4265    //
4266    // This is also C++ [dcl.fct]p6.
4267    if (!Param->isInvalidDecl() &&
4268        RequireCompleteType(Param->getLocation(), Param->getType(),
4269                               diag::err_typecheck_decl_incomplete_type)) {
4270      Param->setInvalidDecl();
4271      HasInvalidParm = true;
4272    }
4273
4274    // C99 6.9.1p5: If the declarator includes a parameter type list, the
4275    // declaration of each parameter shall include an identifier.
4276    if (CheckParameterNames &&
4277        Param->getIdentifier() == 0 &&
4278        !Param->isImplicit() &&
4279        !getLangOptions().CPlusPlus)
4280      Diag(Param->getLocation(), diag::err_parameter_name_omitted);
4281
4282    // C99 6.7.5.3p12:
4283    //   If the function declarator is not part of a definition of that
4284    //   function, parameters may have incomplete type and may use the [*]
4285    //   notation in their sequences of declarator specifiers to specify
4286    //   variable length array types.
4287    QualType PType = Param->getOriginalType();
4288    if (const ArrayType *AT = Context.getAsArrayType(PType)) {
4289      if (AT->getSizeModifier() == ArrayType::Star) {
4290        // FIXME: This diagnosic should point the the '[*]' if source-location
4291        // information is added for it.
4292        Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
4293      }
4294    }
4295  }
4296
4297  return HasInvalidParm;
4298}
4299
4300/// CheckCastAlign - Implements -Wcast-align, which warns when a
4301/// pointer cast increases the alignment requirements.
4302void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
4303  // This is actually a lot of work to potentially be doing on every
4304  // cast; don't do it if we're ignoring -Wcast_align (as is the default).
4305  if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
4306                                          TRange.getBegin())
4307        == DiagnosticsEngine::Ignored)
4308    return;
4309
4310  // Ignore dependent types.
4311  if (T->isDependentType() || Op->getType()->isDependentType())
4312    return;
4313
4314  // Require that the destination be a pointer type.
4315  const PointerType *DestPtr = T->getAs<PointerType>();
4316  if (!DestPtr) return;
4317
4318  // If the destination has alignment 1, we're done.
4319  QualType DestPointee = DestPtr->getPointeeType();
4320  if (DestPointee->isIncompleteType()) return;
4321  CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
4322  if (DestAlign.isOne()) return;
4323
4324  // Require that the source be a pointer type.
4325  const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
4326  if (!SrcPtr) return;
4327  QualType SrcPointee = SrcPtr->getPointeeType();
4328
4329  // Whitelist casts from cv void*.  We already implicitly
4330  // whitelisted casts to cv void*, since they have alignment 1.
4331  // Also whitelist casts involving incomplete types, which implicitly
4332  // includes 'void'.
4333  if (SrcPointee->isIncompleteType()) return;
4334
4335  CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
4336  if (SrcAlign >= DestAlign) return;
4337
4338  Diag(TRange.getBegin(), diag::warn_cast_align)
4339    << Op->getType() << T
4340    << static_cast<unsigned>(SrcAlign.getQuantity())
4341    << static_cast<unsigned>(DestAlign.getQuantity())
4342    << TRange << Op->getSourceRange();
4343}
4344
4345static const Type* getElementType(const Expr *BaseExpr) {
4346  const Type* EltType = BaseExpr->getType().getTypePtr();
4347  if (EltType->isAnyPointerType())
4348    return EltType->getPointeeType().getTypePtr();
4349  else if (EltType->isArrayType())
4350    return EltType->getBaseElementTypeUnsafe();
4351  return EltType;
4352}
4353
4354/// \brief Check whether this array fits the idiom of a size-one tail padded
4355/// array member of a struct.
4356///
4357/// We avoid emitting out-of-bounds access warnings for such arrays as they are
4358/// commonly used to emulate flexible arrays in C89 code.
4359static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size,
4360                                    const NamedDecl *ND) {
4361  if (Size != 1 || !ND) return false;
4362
4363  const FieldDecl *FD = dyn_cast<FieldDecl>(ND);
4364  if (!FD) return false;
4365
4366  // Don't consider sizes resulting from macro expansions or template argument
4367  // substitution to form C89 tail-padded arrays.
4368  ConstantArrayTypeLoc TL =
4369    cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc());
4370  const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr());
4371  if (!SizeExpr || SizeExpr->getExprLoc().isMacroID())
4372    return false;
4373
4374  const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext());
4375  if (!RD) return false;
4376  if (RD->isUnion()) return false;
4377  if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) {
4378    if (!CRD->isStandardLayout()) return false;
4379  }
4380
4381  // See if this is the last field decl in the record.
4382  const Decl *D = FD;
4383  while ((D = D->getNextDeclInContext()))
4384    if (isa<FieldDecl>(D))
4385      return false;
4386  return true;
4387}
4388
4389void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr,
4390                            const ArraySubscriptExpr *ASE,
4391                            bool AllowOnePastEnd, bool IndexNegated) {
4392  IndexExpr = IndexExpr->IgnoreParenCasts();
4393  if (IndexExpr->isValueDependent())
4394    return;
4395
4396  const Type *EffectiveType = getElementType(BaseExpr);
4397  BaseExpr = BaseExpr->IgnoreParenCasts();
4398  const ConstantArrayType *ArrayTy =
4399    Context.getAsConstantArrayType(BaseExpr->getType());
4400  if (!ArrayTy)
4401    return;
4402
4403  llvm::APSInt index;
4404  if (!IndexExpr->EvaluateAsInt(index, Context))
4405    return;
4406  if (IndexNegated)
4407    index = -index;
4408
4409  const NamedDecl *ND = NULL;
4410  if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
4411    ND = dyn_cast<NamedDecl>(DRE->getDecl());
4412  if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
4413    ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
4414
4415  if (index.isUnsigned() || !index.isNegative()) {
4416    llvm::APInt size = ArrayTy->getSize();
4417    if (!size.isStrictlyPositive())
4418      return;
4419
4420    const Type* BaseType = getElementType(BaseExpr);
4421    if (BaseType != EffectiveType) {
4422      // Make sure we're comparing apples to apples when comparing index to size
4423      uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType);
4424      uint64_t array_typesize = Context.getTypeSize(BaseType);
4425      // Handle ptrarith_typesize being zero, such as when casting to void*
4426      if (!ptrarith_typesize) ptrarith_typesize = 1;
4427      if (ptrarith_typesize != array_typesize) {
4428        // There's a cast to a different size type involved
4429        uint64_t ratio = array_typesize / ptrarith_typesize;
4430        // TODO: Be smarter about handling cases where array_typesize is not a
4431        // multiple of ptrarith_typesize
4432        if (ptrarith_typesize * ratio == array_typesize)
4433          size *= llvm::APInt(size.getBitWidth(), ratio);
4434      }
4435    }
4436
4437    if (size.getBitWidth() > index.getBitWidth())
4438      index = index.sext(size.getBitWidth());
4439    else if (size.getBitWidth() < index.getBitWidth())
4440      size = size.sext(index.getBitWidth());
4441
4442    // For array subscripting the index must be less than size, but for pointer
4443    // arithmetic also allow the index (offset) to be equal to size since
4444    // computing the next address after the end of the array is legal and
4445    // commonly done e.g. in C++ iterators and range-based for loops.
4446    if (AllowOnePastEnd ? index.sle(size) : index.slt(size))
4447      return;
4448
4449    // Also don't warn for arrays of size 1 which are members of some
4450    // structure. These are often used to approximate flexible arrays in C89
4451    // code.
4452    if (IsTailPaddedMemberArray(*this, size, ND))
4453      return;
4454
4455    // Suppress the warning if the subscript expression (as identified by the
4456    // ']' location) and the index expression are both from macro expansions
4457    // within a system header.
4458    if (ASE) {
4459      SourceLocation RBracketLoc = SourceMgr.getSpellingLoc(
4460          ASE->getRBracketLoc());
4461      if (SourceMgr.isInSystemHeader(RBracketLoc)) {
4462        SourceLocation IndexLoc = SourceMgr.getSpellingLoc(
4463            IndexExpr->getLocStart());
4464        if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc))
4465          return;
4466      }
4467    }
4468
4469    unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds;
4470    if (ASE)
4471      DiagID = diag::warn_array_index_exceeds_bounds;
4472
4473    DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
4474                        PDiag(DiagID) << index.toString(10, true)
4475                          << size.toString(10, true)
4476                          << (unsigned)size.getLimitedValue(~0U)
4477                          << IndexExpr->getSourceRange());
4478  } else {
4479    unsigned DiagID = diag::warn_array_index_precedes_bounds;
4480    if (!ASE) {
4481      DiagID = diag::warn_ptr_arith_precedes_bounds;
4482      if (index.isNegative()) index = -index;
4483    }
4484
4485    DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
4486                        PDiag(DiagID) << index.toString(10, true)
4487                          << IndexExpr->getSourceRange());
4488  }
4489
4490  if (!ND) {
4491    // Try harder to find a NamedDecl to point at in the note.
4492    while (const ArraySubscriptExpr *ASE =
4493           dyn_cast<ArraySubscriptExpr>(BaseExpr))
4494      BaseExpr = ASE->getBase()->IgnoreParenCasts();
4495    if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
4496      ND = dyn_cast<NamedDecl>(DRE->getDecl());
4497    if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
4498      ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
4499  }
4500
4501  if (ND)
4502    DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
4503                        PDiag(diag::note_array_index_out_of_bounds)
4504                          << ND->getDeclName());
4505}
4506
4507void Sema::CheckArrayAccess(const Expr *expr) {
4508  int AllowOnePastEnd = 0;
4509  while (expr) {
4510    expr = expr->IgnoreParenImpCasts();
4511    switch (expr->getStmtClass()) {
4512      case Stmt::ArraySubscriptExprClass: {
4513        const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr);
4514        CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE,
4515                         AllowOnePastEnd > 0);
4516        return;
4517      }
4518      case Stmt::UnaryOperatorClass: {
4519        // Only unwrap the * and & unary operators
4520        const UnaryOperator *UO = cast<UnaryOperator>(expr);
4521        expr = UO->getSubExpr();
4522        switch (UO->getOpcode()) {
4523          case UO_AddrOf:
4524            AllowOnePastEnd++;
4525            break;
4526          case UO_Deref:
4527            AllowOnePastEnd--;
4528            break;
4529          default:
4530            return;
4531        }
4532        break;
4533      }
4534      case Stmt::ConditionalOperatorClass: {
4535        const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
4536        if (const Expr *lhs = cond->getLHS())
4537          CheckArrayAccess(lhs);
4538        if (const Expr *rhs = cond->getRHS())
4539          CheckArrayAccess(rhs);
4540        return;
4541      }
4542      default:
4543        return;
4544    }
4545  }
4546}
4547
4548//===--- CHECK: Objective-C retain cycles ----------------------------------//
4549
4550namespace {
4551  struct RetainCycleOwner {
4552    RetainCycleOwner() : Variable(0), Indirect(false) {}
4553    VarDecl *Variable;
4554    SourceRange Range;
4555    SourceLocation Loc;
4556    bool Indirect;
4557
4558    void setLocsFrom(Expr *e) {
4559      Loc = e->getExprLoc();
4560      Range = e->getSourceRange();
4561    }
4562  };
4563}
4564
4565/// Consider whether capturing the given variable can possibly lead to
4566/// a retain cycle.
4567static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) {
4568  // In ARC, it's captured strongly iff the variable has __strong
4569  // lifetime.  In MRR, it's captured strongly if the variable is
4570  // __block and has an appropriate type.
4571  if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
4572    return false;
4573
4574  owner.Variable = var;
4575  owner.setLocsFrom(ref);
4576  return true;
4577}
4578
4579static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) {
4580  while (true) {
4581    e = e->IgnoreParens();
4582    if (CastExpr *cast = dyn_cast<CastExpr>(e)) {
4583      switch (cast->getCastKind()) {
4584      case CK_BitCast:
4585      case CK_LValueBitCast:
4586      case CK_LValueToRValue:
4587      case CK_ARCReclaimReturnedObject:
4588        e = cast->getSubExpr();
4589        continue;
4590
4591      default:
4592        return false;
4593      }
4594    }
4595
4596    if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) {
4597      ObjCIvarDecl *ivar = ref->getDecl();
4598      if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
4599        return false;
4600
4601      // Try to find a retain cycle in the base.
4602      if (!findRetainCycleOwner(S, ref->getBase(), owner))
4603        return false;
4604
4605      if (ref->isFreeIvar()) owner.setLocsFrom(ref);
4606      owner.Indirect = true;
4607      return true;
4608    }
4609
4610    if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) {
4611      VarDecl *var = dyn_cast<VarDecl>(ref->getDecl());
4612      if (!var) return false;
4613      return considerVariable(var, ref, owner);
4614    }
4615
4616    if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) {
4617      owner.Variable = ref->getDecl();
4618      owner.setLocsFrom(ref);
4619      return true;
4620    }
4621
4622    if (MemberExpr *member = dyn_cast<MemberExpr>(e)) {
4623      if (member->isArrow()) return false;
4624
4625      // Don't count this as an indirect ownership.
4626      e = member->getBase();
4627      continue;
4628    }
4629
4630    if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) {
4631      // Only pay attention to pseudo-objects on property references.
4632      ObjCPropertyRefExpr *pre
4633        = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm()
4634                                              ->IgnoreParens());
4635      if (!pre) return false;
4636      if (pre->isImplicitProperty()) return false;
4637      ObjCPropertyDecl *property = pre->getExplicitProperty();
4638      if (!property->isRetaining() &&
4639          !(property->getPropertyIvarDecl() &&
4640            property->getPropertyIvarDecl()->getType()
4641              .getObjCLifetime() == Qualifiers::OCL_Strong))
4642          return false;
4643
4644      owner.Indirect = true;
4645      if (pre->isSuperReceiver()) {
4646        owner.Variable = S.getCurMethodDecl()->getSelfDecl();
4647        if (!owner.Variable)
4648          return false;
4649        owner.Loc = pre->getLocation();
4650        owner.Range = pre->getSourceRange();
4651        return true;
4652      }
4653      e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase())
4654                              ->getSourceExpr());
4655      continue;
4656    }
4657
4658    // Array ivars?
4659
4660    return false;
4661  }
4662}
4663
4664namespace {
4665  struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> {
4666    FindCaptureVisitor(ASTContext &Context, VarDecl *variable)
4667      : EvaluatedExprVisitor<FindCaptureVisitor>(Context),
4668        Variable(variable), Capturer(0) {}
4669
4670    VarDecl *Variable;
4671    Expr *Capturer;
4672
4673    void VisitDeclRefExpr(DeclRefExpr *ref) {
4674      if (ref->getDecl() == Variable && !Capturer)
4675        Capturer = ref;
4676    }
4677
4678    void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) {
4679      if (ref->getDecl() == Variable && !Capturer)
4680        Capturer = ref;
4681    }
4682
4683    void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) {
4684      if (Capturer) return;
4685      Visit(ref->getBase());
4686      if (Capturer && ref->isFreeIvar())
4687        Capturer = ref;
4688    }
4689
4690    void VisitBlockExpr(BlockExpr *block) {
4691      // Look inside nested blocks
4692      if (block->getBlockDecl()->capturesVariable(Variable))
4693        Visit(block->getBlockDecl()->getBody());
4694    }
4695  };
4696}
4697
4698/// Check whether the given argument is a block which captures a
4699/// variable.
4700static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) {
4701  assert(owner.Variable && owner.Loc.isValid());
4702
4703  e = e->IgnoreParenCasts();
4704  BlockExpr *block = dyn_cast<BlockExpr>(e);
4705  if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable))
4706    return 0;
4707
4708  FindCaptureVisitor visitor(S.Context, owner.Variable);
4709  visitor.Visit(block->getBlockDecl()->getBody());
4710  return visitor.Capturer;
4711}
4712
4713static void diagnoseRetainCycle(Sema &S, Expr *capturer,
4714                                RetainCycleOwner &owner) {
4715  assert(capturer);
4716  assert(owner.Variable && owner.Loc.isValid());
4717
4718  S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle)
4719    << owner.Variable << capturer->getSourceRange();
4720  S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner)
4721    << owner.Indirect << owner.Range;
4722}
4723
4724/// Check for a keyword selector that starts with the word 'add' or
4725/// 'set'.
4726static bool isSetterLikeSelector(Selector sel) {
4727  if (sel.isUnarySelector()) return false;
4728
4729  StringRef str = sel.getNameForSlot(0);
4730  while (!str.empty() && str.front() == '_') str = str.substr(1);
4731  if (str.startswith("set"))
4732    str = str.substr(3);
4733  else if (str.startswith("add")) {
4734    // Specially whitelist 'addOperationWithBlock:'.
4735    if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock"))
4736      return false;
4737    str = str.substr(3);
4738  }
4739  else
4740    return false;
4741
4742  if (str.empty()) return true;
4743  return !islower(str.front());
4744}
4745
4746/// Check a message send to see if it's likely to cause a retain cycle.
4747void Sema::checkRetainCycles(ObjCMessageExpr *msg) {
4748  // Only check instance methods whose selector looks like a setter.
4749  if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector()))
4750    return;
4751
4752  // Try to find a variable that the receiver is strongly owned by.
4753  RetainCycleOwner owner;
4754  if (msg->getReceiverKind() == ObjCMessageExpr::Instance) {
4755    if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner))
4756      return;
4757  } else {
4758    assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance);
4759    owner.Variable = getCurMethodDecl()->getSelfDecl();
4760    owner.Loc = msg->getSuperLoc();
4761    owner.Range = msg->getSuperLoc();
4762  }
4763
4764  // Check whether the receiver is captured by any of the arguments.
4765  for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i)
4766    if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner))
4767      return diagnoseRetainCycle(*this, capturer, owner);
4768}
4769
4770/// Check a property assign to see if it's likely to cause a retain cycle.
4771void Sema::checkRetainCycles(Expr *receiver, Expr *argument) {
4772  RetainCycleOwner owner;
4773  if (!findRetainCycleOwner(*this, receiver, owner))
4774    return;
4775
4776  if (Expr *capturer = findCapturingExpr(*this, argument, owner))
4777    diagnoseRetainCycle(*this, capturer, owner);
4778}
4779
4780bool Sema::checkUnsafeAssigns(SourceLocation Loc,
4781                              QualType LHS, Expr *RHS) {
4782  Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime();
4783  if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone)
4784    return false;
4785  // strip off any implicit cast added to get to the one arc-specific
4786  while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
4787    if (cast->getCastKind() == CK_ARCConsumeObject) {
4788      Diag(Loc, diag::warn_arc_retained_assign)
4789        << (LT == Qualifiers::OCL_ExplicitNone)
4790        << RHS->getSourceRange();
4791      return true;
4792    }
4793    RHS = cast->getSubExpr();
4794  }
4795  return false;
4796}
4797
4798void Sema::checkUnsafeExprAssigns(SourceLocation Loc,
4799                              Expr *LHS, Expr *RHS) {
4800  QualType LHSType;
4801  // PropertyRef on LHS type need be directly obtained from
4802  // its declaration as it has a PsuedoType.
4803  ObjCPropertyRefExpr *PRE
4804    = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens());
4805  if (PRE && !PRE->isImplicitProperty()) {
4806    const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
4807    if (PD)
4808      LHSType = PD->getType();
4809  }
4810
4811  if (LHSType.isNull())
4812    LHSType = LHS->getType();
4813  if (checkUnsafeAssigns(Loc, LHSType, RHS))
4814    return;
4815  Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime();
4816  // FIXME. Check for other life times.
4817  if (LT != Qualifiers::OCL_None)
4818    return;
4819
4820  if (PRE) {
4821    if (PRE->isImplicitProperty())
4822      return;
4823    const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
4824    if (!PD)
4825      return;
4826
4827    unsigned Attributes = PD->getPropertyAttributes();
4828    if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) {
4829      // when 'assign' attribute was not explicitly specified
4830      // by user, ignore it and rely on property type itself
4831      // for lifetime info.
4832      unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten();
4833      if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) &&
4834          LHSType->isObjCRetainableType())
4835        return;
4836
4837      while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
4838        if (cast->getCastKind() == CK_ARCConsumeObject) {
4839          Diag(Loc, diag::warn_arc_retained_property_assign)
4840          << RHS->getSourceRange();
4841          return;
4842        }
4843        RHS = cast->getSubExpr();
4844      }
4845    }
4846  }
4847}
4848
4849//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===//
4850
4851namespace {
4852bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr,
4853                                 SourceLocation StmtLoc,
4854                                 const NullStmt *Body) {
4855  // Do not warn if the body is a macro that expands to nothing, e.g:
4856  //
4857  // #define CALL(x)
4858  // if (condition)
4859  //   CALL(0);
4860  //
4861  if (Body->hasLeadingEmptyMacro())
4862    return false;
4863
4864  // Get line numbers of statement and body.
4865  bool StmtLineInvalid;
4866  unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc,
4867                                                      &StmtLineInvalid);
4868  if (StmtLineInvalid)
4869    return false;
4870
4871  bool BodyLineInvalid;
4872  unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(),
4873                                                      &BodyLineInvalid);
4874  if (BodyLineInvalid)
4875    return false;
4876
4877  // Warn if null statement and body are on the same line.
4878  if (StmtLine != BodyLine)
4879    return false;
4880
4881  return true;
4882}
4883} // Unnamed namespace
4884
4885void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc,
4886                                 const Stmt *Body,
4887                                 unsigned DiagID) {
4888  // Since this is a syntactic check, don't emit diagnostic for template
4889  // instantiations, this just adds noise.
4890  if (CurrentInstantiationScope)
4891    return;
4892
4893  // The body should be a null statement.
4894  const NullStmt *NBody = dyn_cast<NullStmt>(Body);
4895  if (!NBody)
4896    return;
4897
4898  // Do the usual checks.
4899  if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody))
4900    return;
4901
4902  Diag(NBody->getSemiLoc(), DiagID);
4903  Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line);
4904}
4905
4906void Sema::DiagnoseEmptyLoopBody(const Stmt *S,
4907                                 const Stmt *PossibleBody) {
4908  assert(!CurrentInstantiationScope); // Ensured by caller
4909
4910  SourceLocation StmtLoc;
4911  const Stmt *Body;
4912  unsigned DiagID;
4913  if (const ForStmt *FS = dyn_cast<ForStmt>(S)) {
4914    StmtLoc = FS->getRParenLoc();
4915    Body = FS->getBody();
4916    DiagID = diag::warn_empty_for_body;
4917  } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) {
4918    StmtLoc = WS->getCond()->getSourceRange().getEnd();
4919    Body = WS->getBody();
4920    DiagID = diag::warn_empty_while_body;
4921  } else
4922    return; // Neither `for' nor `while'.
4923
4924  // The body should be a null statement.
4925  const NullStmt *NBody = dyn_cast<NullStmt>(Body);
4926  if (!NBody)
4927    return;
4928
4929  // Skip expensive checks if diagnostic is disabled.
4930  if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) ==
4931          DiagnosticsEngine::Ignored)
4932    return;
4933
4934  // Do the usual checks.
4935  if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody))
4936    return;
4937
4938  // `for(...);' and `while(...);' are popular idioms, so in order to keep
4939  // noise level low, emit diagnostics only if for/while is followed by a
4940  // CompoundStmt, e.g.:
4941  //    for (int i = 0; i < n; i++);
4942  //    {
4943  //      a(i);
4944  //    }
4945  // or if for/while is followed by a statement with more indentation
4946  // than for/while itself:
4947  //    for (int i = 0; i < n; i++);
4948  //      a(i);
4949  bool ProbableTypo = isa<CompoundStmt>(PossibleBody);
4950  if (!ProbableTypo) {
4951    bool BodyColInvalid;
4952    unsigned BodyCol = SourceMgr.getPresumedColumnNumber(
4953                             PossibleBody->getLocStart(),
4954                             &BodyColInvalid);
4955    if (BodyColInvalid)
4956      return;
4957
4958    bool StmtColInvalid;
4959    unsigned StmtCol = SourceMgr.getPresumedColumnNumber(
4960                             S->getLocStart(),
4961                             &StmtColInvalid);
4962    if (StmtColInvalid)
4963      return;
4964
4965    if (BodyCol > StmtCol)
4966      ProbableTypo = true;
4967  }
4968
4969  if (ProbableTypo) {
4970    Diag(NBody->getSemiLoc(), DiagID);
4971    Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line);
4972  }
4973}
4974
4975