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/SemaInternal.h"
16 #include "clang/AST/ASTContext.h"
17 #include "clang/AST/CharUnits.h"
18 #include "clang/AST/DeclCXX.h"
19 #include "clang/AST/DeclObjC.h"
20 #include "clang/AST/EvaluatedExprVisitor.h"
21 #include "clang/AST/Expr.h"
22 #include "clang/AST/ExprCXX.h"
23 #include "clang/AST/ExprObjC.h"
24 #include "clang/AST/StmtCXX.h"
25 #include "clang/AST/StmtObjC.h"
26 #include "clang/Analysis/Analyses/FormatString.h"
27 #include "clang/Basic/CharInfo.h"
28 #include "clang/Basic/TargetBuiltins.h"
29 #include "clang/Basic/TargetInfo.h"
30 #include "clang/Lex/Preprocessor.h"
31 #include "clang/Sema/Initialization.h"
32 #include "clang/Sema/Lookup.h"
33 #include "clang/Sema/ScopeInfo.h"
34 #include "clang/Sema/Sema.h"
35 #include "llvm/ADT/BitVector.h"
36 #include "llvm/ADT/STLExtras.h"
37 #include "llvm/ADT/SmallString.h"
38 #include "llvm/Support/ConvertUTF.h"
39 #include "llvm/Support/raw_ostream.h"
40 #include <limits>
41 using namespace clang;
42 using namespace sema;
43 
44 SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL,
45                                                     unsigned ByteNo) const {
46   return SL->getLocationOfByte(ByteNo, PP.getSourceManager(),
47                                PP.getLangOpts(), PP.getTargetInfo());
48 }
49 
50 /// Checks that a call expression's argument count is the desired number.
51 /// This is useful when doing custom type-checking.  Returns true on error.
52 static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) {
53   unsigned argCount = call->getNumArgs();
54   if (argCount == desiredArgCount) return false;
55 
56   if (argCount < desiredArgCount)
57     return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args)
58         << 0 /*function call*/ << desiredArgCount << argCount
59         << call->getSourceRange();
60 
61   // Highlight all the excess arguments.
62   SourceRange range(call->getArg(desiredArgCount)->getLocStart(),
63                     call->getArg(argCount - 1)->getLocEnd());
64 
65   return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args)
66     << 0 /*function call*/ << desiredArgCount << argCount
67     << call->getArg(1)->getSourceRange();
68 }
69 
70 /// Check that the first argument to __builtin_annotation is an integer
71 /// and the second argument is a non-wide string literal.
72 static bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) {
73   if (checkArgCount(S, TheCall, 2))
74     return true;
75 
76   // First argument should be an integer.
77   Expr *ValArg = TheCall->getArg(0);
78   QualType Ty = ValArg->getType();
79   if (!Ty->isIntegerType()) {
80     S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg)
81       << ValArg->getSourceRange();
82     return true;
83   }
84 
85   // Second argument should be a constant string.
86   Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts();
87   StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg);
88   if (!Literal || !Literal->isAscii()) {
89     S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg)
90       << StrArg->getSourceRange();
91     return true;
92   }
93 
94   TheCall->setType(Ty);
95   return false;
96 }
97 
98 ExprResult
99 Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
100   ExprResult TheCallResult(Owned(TheCall));
101 
102   // Find out if any arguments are required to be integer constant expressions.
103   unsigned ICEArguments = 0;
104   ASTContext::GetBuiltinTypeError Error;
105   Context.GetBuiltinType(BuiltinID, Error, &ICEArguments);
106   if (Error != ASTContext::GE_None)
107     ICEArguments = 0;  // Don't diagnose previously diagnosed errors.
108 
109   // If any arguments are required to be ICE's, check and diagnose.
110   for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) {
111     // Skip arguments not required to be ICE's.
112     if ((ICEArguments & (1 << ArgNo)) == 0) continue;
113 
114     llvm::APSInt Result;
115     if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
116       return true;
117     ICEArguments &= ~(1 << ArgNo);
118   }
119 
120   switch (BuiltinID) {
121   case Builtin::BI__builtin___CFStringMakeConstantString:
122     assert(TheCall->getNumArgs() == 1 &&
123            "Wrong # arguments to builtin CFStringMakeConstantString");
124     if (CheckObjCString(TheCall->getArg(0)))
125       return ExprError();
126     break;
127   case Builtin::BI__builtin_stdarg_start:
128   case Builtin::BI__builtin_va_start:
129     if (SemaBuiltinVAStart(TheCall))
130       return ExprError();
131     break;
132   case Builtin::BI__builtin_isgreater:
133   case Builtin::BI__builtin_isgreaterequal:
134   case Builtin::BI__builtin_isless:
135   case Builtin::BI__builtin_islessequal:
136   case Builtin::BI__builtin_islessgreater:
137   case Builtin::BI__builtin_isunordered:
138     if (SemaBuiltinUnorderedCompare(TheCall))
139       return ExprError();
140     break;
141   case Builtin::BI__builtin_fpclassify:
142     if (SemaBuiltinFPClassification(TheCall, 6))
143       return ExprError();
144     break;
145   case Builtin::BI__builtin_isfinite:
146   case Builtin::BI__builtin_isinf:
147   case Builtin::BI__builtin_isinf_sign:
148   case Builtin::BI__builtin_isnan:
149   case Builtin::BI__builtin_isnormal:
150     if (SemaBuiltinFPClassification(TheCall, 1))
151       return ExprError();
152     break;
153   case Builtin::BI__builtin_shufflevector:
154     return SemaBuiltinShuffleVector(TheCall);
155     // TheCall will be freed by the smart pointer here, but that's fine, since
156     // SemaBuiltinShuffleVector guts it, but then doesn't release it.
157   case Builtin::BI__builtin_prefetch:
158     if (SemaBuiltinPrefetch(TheCall))
159       return ExprError();
160     break;
161   case Builtin::BI__builtin_object_size:
162     if (SemaBuiltinObjectSize(TheCall))
163       return ExprError();
164     break;
165   case Builtin::BI__builtin_longjmp:
166     if (SemaBuiltinLongjmp(TheCall))
167       return ExprError();
168     break;
169 
170   case Builtin::BI__builtin_classify_type:
171     if (checkArgCount(*this, TheCall, 1)) return true;
172     TheCall->setType(Context.IntTy);
173     break;
174   case Builtin::BI__builtin_constant_p:
175     if (checkArgCount(*this, TheCall, 1)) return true;
176     TheCall->setType(Context.IntTy);
177     break;
178   case Builtin::BI__sync_fetch_and_add:
179   case Builtin::BI__sync_fetch_and_add_1:
180   case Builtin::BI__sync_fetch_and_add_2:
181   case Builtin::BI__sync_fetch_and_add_4:
182   case Builtin::BI__sync_fetch_and_add_8:
183   case Builtin::BI__sync_fetch_and_add_16:
184   case Builtin::BI__sync_fetch_and_sub:
185   case Builtin::BI__sync_fetch_and_sub_1:
186   case Builtin::BI__sync_fetch_and_sub_2:
187   case Builtin::BI__sync_fetch_and_sub_4:
188   case Builtin::BI__sync_fetch_and_sub_8:
189   case Builtin::BI__sync_fetch_and_sub_16:
190   case Builtin::BI__sync_fetch_and_or:
191   case Builtin::BI__sync_fetch_and_or_1:
192   case Builtin::BI__sync_fetch_and_or_2:
193   case Builtin::BI__sync_fetch_and_or_4:
194   case Builtin::BI__sync_fetch_and_or_8:
195   case Builtin::BI__sync_fetch_and_or_16:
196   case Builtin::BI__sync_fetch_and_and:
197   case Builtin::BI__sync_fetch_and_and_1:
198   case Builtin::BI__sync_fetch_and_and_2:
199   case Builtin::BI__sync_fetch_and_and_4:
200   case Builtin::BI__sync_fetch_and_and_8:
201   case Builtin::BI__sync_fetch_and_and_16:
202   case Builtin::BI__sync_fetch_and_xor:
203   case Builtin::BI__sync_fetch_and_xor_1:
204   case Builtin::BI__sync_fetch_and_xor_2:
205   case Builtin::BI__sync_fetch_and_xor_4:
206   case Builtin::BI__sync_fetch_and_xor_8:
207   case Builtin::BI__sync_fetch_and_xor_16:
208   case Builtin::BI__sync_add_and_fetch:
209   case Builtin::BI__sync_add_and_fetch_1:
210   case Builtin::BI__sync_add_and_fetch_2:
211   case Builtin::BI__sync_add_and_fetch_4:
212   case Builtin::BI__sync_add_and_fetch_8:
213   case Builtin::BI__sync_add_and_fetch_16:
214   case Builtin::BI__sync_sub_and_fetch:
215   case Builtin::BI__sync_sub_and_fetch_1:
216   case Builtin::BI__sync_sub_and_fetch_2:
217   case Builtin::BI__sync_sub_and_fetch_4:
218   case Builtin::BI__sync_sub_and_fetch_8:
219   case Builtin::BI__sync_sub_and_fetch_16:
220   case Builtin::BI__sync_and_and_fetch:
221   case Builtin::BI__sync_and_and_fetch_1:
222   case Builtin::BI__sync_and_and_fetch_2:
223   case Builtin::BI__sync_and_and_fetch_4:
224   case Builtin::BI__sync_and_and_fetch_8:
225   case Builtin::BI__sync_and_and_fetch_16:
226   case Builtin::BI__sync_or_and_fetch:
227   case Builtin::BI__sync_or_and_fetch_1:
228   case Builtin::BI__sync_or_and_fetch_2:
229   case Builtin::BI__sync_or_and_fetch_4:
230   case Builtin::BI__sync_or_and_fetch_8:
231   case Builtin::BI__sync_or_and_fetch_16:
232   case Builtin::BI__sync_xor_and_fetch:
233   case Builtin::BI__sync_xor_and_fetch_1:
234   case Builtin::BI__sync_xor_and_fetch_2:
235   case Builtin::BI__sync_xor_and_fetch_4:
236   case Builtin::BI__sync_xor_and_fetch_8:
237   case Builtin::BI__sync_xor_and_fetch_16:
238   case Builtin::BI__sync_val_compare_and_swap:
239   case Builtin::BI__sync_val_compare_and_swap_1:
240   case Builtin::BI__sync_val_compare_and_swap_2:
241   case Builtin::BI__sync_val_compare_and_swap_4:
242   case Builtin::BI__sync_val_compare_and_swap_8:
243   case Builtin::BI__sync_val_compare_and_swap_16:
244   case Builtin::BI__sync_bool_compare_and_swap:
245   case Builtin::BI__sync_bool_compare_and_swap_1:
246   case Builtin::BI__sync_bool_compare_and_swap_2:
247   case Builtin::BI__sync_bool_compare_and_swap_4:
248   case Builtin::BI__sync_bool_compare_and_swap_8:
249   case Builtin::BI__sync_bool_compare_and_swap_16:
250   case Builtin::BI__sync_lock_test_and_set:
251   case Builtin::BI__sync_lock_test_and_set_1:
252   case Builtin::BI__sync_lock_test_and_set_2:
253   case Builtin::BI__sync_lock_test_and_set_4:
254   case Builtin::BI__sync_lock_test_and_set_8:
255   case Builtin::BI__sync_lock_test_and_set_16:
256   case Builtin::BI__sync_lock_release:
257   case Builtin::BI__sync_lock_release_1:
258   case Builtin::BI__sync_lock_release_2:
259   case Builtin::BI__sync_lock_release_4:
260   case Builtin::BI__sync_lock_release_8:
261   case Builtin::BI__sync_lock_release_16:
262   case Builtin::BI__sync_swap:
263   case Builtin::BI__sync_swap_1:
264   case Builtin::BI__sync_swap_2:
265   case Builtin::BI__sync_swap_4:
266   case Builtin::BI__sync_swap_8:
267   case Builtin::BI__sync_swap_16:
268     return SemaBuiltinAtomicOverloaded(TheCallResult);
269 #define BUILTIN(ID, TYPE, ATTRS)
270 #define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \
271   case Builtin::BI##ID: \
272     return SemaAtomicOpsOverloaded(TheCallResult, AtomicExpr::AO##ID);
273 #include "clang/Basic/Builtins.def"
274   case Builtin::BI__builtin_annotation:
275     if (SemaBuiltinAnnotation(*this, TheCall))
276       return ExprError();
277     break;
278   }
279 
280   // Since the target specific builtins for each arch overlap, only check those
281   // of the arch we are compiling for.
282   if (BuiltinID >= Builtin::FirstTSBuiltin) {
283     switch (Context.getTargetInfo().getTriple().getArch()) {
284       case llvm::Triple::arm:
285       case llvm::Triple::thumb:
286         if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall))
287           return ExprError();
288         break;
289       case llvm::Triple::mips:
290       case llvm::Triple::mipsel:
291       case llvm::Triple::mips64:
292       case llvm::Triple::mips64el:
293         if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall))
294           return ExprError();
295         break;
296       default:
297         break;
298     }
299   }
300 
301   return TheCallResult;
302 }
303 
304 // Get the valid immediate range for the specified NEON type code.
305 static unsigned RFT(unsigned t, bool shift = false) {
306   NeonTypeFlags Type(t);
307   int IsQuad = Type.isQuad();
308   switch (Type.getEltType()) {
309   case NeonTypeFlags::Int8:
310   case NeonTypeFlags::Poly8:
311     return shift ? 7 : (8 << IsQuad) - 1;
312   case NeonTypeFlags::Int16:
313   case NeonTypeFlags::Poly16:
314     return shift ? 15 : (4 << IsQuad) - 1;
315   case NeonTypeFlags::Int32:
316     return shift ? 31 : (2 << IsQuad) - 1;
317   case NeonTypeFlags::Int64:
318     return shift ? 63 : (1 << IsQuad) - 1;
319   case NeonTypeFlags::Float16:
320     assert(!shift && "cannot shift float types!");
321     return (4 << IsQuad) - 1;
322   case NeonTypeFlags::Float32:
323     assert(!shift && "cannot shift float types!");
324     return (2 << IsQuad) - 1;
325   }
326   llvm_unreachable("Invalid NeonTypeFlag!");
327 }
328 
329 /// getNeonEltType - Return the QualType corresponding to the elements of
330 /// the vector type specified by the NeonTypeFlags.  This is used to check
331 /// the pointer arguments for Neon load/store intrinsics.
332 static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) {
333   switch (Flags.getEltType()) {
334   case NeonTypeFlags::Int8:
335     return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy;
336   case NeonTypeFlags::Int16:
337     return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy;
338   case NeonTypeFlags::Int32:
339     return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy;
340   case NeonTypeFlags::Int64:
341     return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy;
342   case NeonTypeFlags::Poly8:
343     return Context.SignedCharTy;
344   case NeonTypeFlags::Poly16:
345     return Context.ShortTy;
346   case NeonTypeFlags::Float16:
347     return Context.UnsignedShortTy;
348   case NeonTypeFlags::Float32:
349     return Context.FloatTy;
350   }
351   llvm_unreachable("Invalid NeonTypeFlag!");
352 }
353 
354 bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
355   llvm::APSInt Result;
356 
357   uint64_t mask = 0;
358   unsigned TV = 0;
359   int PtrArgNum = -1;
360   bool HasConstPtr = false;
361   switch (BuiltinID) {
362 #define GET_NEON_OVERLOAD_CHECK
363 #include "clang/Basic/arm_neon.inc"
364 #undef GET_NEON_OVERLOAD_CHECK
365   }
366 
367   // For NEON intrinsics which are overloaded on vector element type, validate
368   // the immediate which specifies which variant to emit.
369   unsigned ImmArg = TheCall->getNumArgs()-1;
370   if (mask) {
371     if (SemaBuiltinConstantArg(TheCall, ImmArg, Result))
372       return true;
373 
374     TV = Result.getLimitedValue(64);
375     if ((TV > 63) || (mask & (1ULL << TV)) == 0)
376       return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code)
377         << TheCall->getArg(ImmArg)->getSourceRange();
378   }
379 
380   if (PtrArgNum >= 0) {
381     // Check that pointer arguments have the specified type.
382     Expr *Arg = TheCall->getArg(PtrArgNum);
383     if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg))
384       Arg = ICE->getSubExpr();
385     ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg);
386     QualType RHSTy = RHS.get()->getType();
387     QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context);
388     if (HasConstPtr)
389       EltTy = EltTy.withConst();
390     QualType LHSTy = Context.getPointerType(EltTy);
391     AssignConvertType ConvTy;
392     ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS);
393     if (RHS.isInvalid())
394       return true;
395     if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy,
396                                  RHS.get(), AA_Assigning))
397       return true;
398   }
399 
400   // For NEON intrinsics which take an immediate value as part of the
401   // instruction, range check them here.
402   unsigned i = 0, l = 0, u = 0;
403   switch (BuiltinID) {
404   default: return false;
405   case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break;
406   case ARM::BI__builtin_arm_usat: i = 1; u = 31; break;
407   case ARM::BI__builtin_arm_vcvtr_f:
408   case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break;
409 #define GET_NEON_IMMEDIATE_CHECK
410 #include "clang/Basic/arm_neon.inc"
411 #undef GET_NEON_IMMEDIATE_CHECK
412   };
413 
414   // We can't check the value of a dependent argument.
415   if (TheCall->getArg(i)->isTypeDependent() ||
416       TheCall->getArg(i)->isValueDependent())
417     return false;
418 
419   // Check that the immediate argument is actually a constant.
420   if (SemaBuiltinConstantArg(TheCall, i, Result))
421     return true;
422 
423   // Range check against the upper/lower values for this isntruction.
424   unsigned Val = Result.getZExtValue();
425   if (Val < l || Val > (u + l))
426     return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
427       << l << u+l << TheCall->getArg(i)->getSourceRange();
428 
429   // FIXME: VFP Intrinsics should error if VFP not present.
430   return false;
431 }
432 
433 bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
434   unsigned i = 0, l = 0, u = 0;
435   switch (BuiltinID) {
436   default: return false;
437   case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break;
438   case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break;
439   case Mips::BI__builtin_mips_append: i = 2; l = 0; u = 31; break;
440   case Mips::BI__builtin_mips_balign: i = 2; l = 0; u = 3; break;
441   case Mips::BI__builtin_mips_precr_sra_ph_w: i = 2; l = 0; u = 31; break;
442   case Mips::BI__builtin_mips_precr_sra_r_ph_w: i = 2; l = 0; u = 31; break;
443   case Mips::BI__builtin_mips_prepend: i = 2; l = 0; u = 31; break;
444   };
445 
446   // We can't check the value of a dependent argument.
447   if (TheCall->getArg(i)->isTypeDependent() ||
448       TheCall->getArg(i)->isValueDependent())
449     return false;
450 
451   // Check that the immediate argument is actually a constant.
452   llvm::APSInt Result;
453   if (SemaBuiltinConstantArg(TheCall, i, Result))
454     return true;
455 
456   // Range check against the upper/lower values for this instruction.
457   unsigned Val = Result.getZExtValue();
458   if (Val < l || Val > u)
459     return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
460       << l << u << TheCall->getArg(i)->getSourceRange();
461 
462   return false;
463 }
464 
465 /// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo
466 /// parameter with the FormatAttr's correct format_idx and firstDataArg.
467 /// Returns true when the format fits the function and the FormatStringInfo has
468 /// been populated.
469 bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember,
470                                FormatStringInfo *FSI) {
471   FSI->HasVAListArg = Format->getFirstArg() == 0;
472   FSI->FormatIdx = Format->getFormatIdx() - 1;
473   FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1;
474 
475   // The way the format attribute works in GCC, the implicit this argument
476   // of member functions is counted. However, it doesn't appear in our own
477   // lists, so decrement format_idx in that case.
478   if (IsCXXMember) {
479     if(FSI->FormatIdx == 0)
480       return false;
481     --FSI->FormatIdx;
482     if (FSI->FirstDataArg != 0)
483       --FSI->FirstDataArg;
484   }
485   return true;
486 }
487 
488 /// Handles the checks for format strings, non-POD arguments to vararg
489 /// functions, and NULL arguments passed to non-NULL parameters.
490 void Sema::checkCall(NamedDecl *FDecl,
491                      ArrayRef<const Expr *> Args,
492                      unsigned NumProtoArgs,
493                      bool IsMemberFunction,
494                      SourceLocation Loc,
495                      SourceRange Range,
496                      VariadicCallType CallType) {
497   if (CurContext->isDependentContext())
498     return;
499 
500   // Printf and scanf checking.
501   bool HandledFormatString = false;
502   for (specific_attr_iterator<FormatAttr>
503          I = FDecl->specific_attr_begin<FormatAttr>(),
504          E = FDecl->specific_attr_end<FormatAttr>(); I != E ; ++I)
505     if (CheckFormatArguments(*I, Args, IsMemberFunction, CallType, Loc, Range))
506         HandledFormatString = true;
507 
508   // Refuse POD arguments that weren't caught by the format string
509   // checks above.
510   if (!HandledFormatString && CallType != VariadicDoesNotApply)
511     for (unsigned ArgIdx = NumProtoArgs; ArgIdx < Args.size(); ++ArgIdx) {
512       // Args[ArgIdx] can be null in malformed code.
513       if (const Expr *Arg = Args[ArgIdx])
514         variadicArgumentPODCheck(Arg, CallType);
515     }
516 
517   for (specific_attr_iterator<NonNullAttr>
518          I = FDecl->specific_attr_begin<NonNullAttr>(),
519          E = FDecl->specific_attr_end<NonNullAttr>(); I != E; ++I)
520     CheckNonNullArguments(*I, Args.data(), Loc);
521 
522   // Type safety checking.
523   for (specific_attr_iterator<ArgumentWithTypeTagAttr>
524          i = FDecl->specific_attr_begin<ArgumentWithTypeTagAttr>(),
525          e = FDecl->specific_attr_end<ArgumentWithTypeTagAttr>(); i != e; ++i) {
526     CheckArgumentWithTypeTag(*i, Args.data());
527   }
528 }
529 
530 /// CheckConstructorCall - Check a constructor call for correctness and safety
531 /// properties not enforced by the C type system.
532 void Sema::CheckConstructorCall(FunctionDecl *FDecl,
533                                 ArrayRef<const Expr *> Args,
534                                 const FunctionProtoType *Proto,
535                                 SourceLocation Loc) {
536   VariadicCallType CallType =
537     Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply;
538   checkCall(FDecl, Args, Proto->getNumArgs(),
539             /*IsMemberFunction=*/true, Loc, SourceRange(), CallType);
540 }
541 
542 /// CheckFunctionCall - Check a direct function call for various correctness
543 /// and safety properties not strictly enforced by the C type system.
544 bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall,
545                              const FunctionProtoType *Proto) {
546   bool IsMemberOperatorCall = isa<CXXOperatorCallExpr>(TheCall) &&
547                               isa<CXXMethodDecl>(FDecl);
548   bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall) ||
549                           IsMemberOperatorCall;
550   VariadicCallType CallType = getVariadicCallType(FDecl, Proto,
551                                                   TheCall->getCallee());
552   unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0;
553   Expr** Args = TheCall->getArgs();
554   unsigned NumArgs = TheCall->getNumArgs();
555   if (IsMemberOperatorCall) {
556     // If this is a call to a member operator, hide the first argument
557     // from checkCall.
558     // FIXME: Our choice of AST representation here is less than ideal.
559     ++Args;
560     --NumArgs;
561   }
562   checkCall(FDecl, llvm::makeArrayRef<const Expr *>(Args, NumArgs),
563             NumProtoArgs,
564             IsMemberFunction, TheCall->getRParenLoc(),
565             TheCall->getCallee()->getSourceRange(), CallType);
566 
567   IdentifierInfo *FnInfo = FDecl->getIdentifier();
568   // None of the checks below are needed for functions that don't have
569   // simple names (e.g., C++ conversion functions).
570   if (!FnInfo)
571     return false;
572 
573   unsigned CMId = FDecl->getMemoryFunctionKind();
574   if (CMId == 0)
575     return false;
576 
577   // Handle memory setting and copying functions.
578   if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat)
579     CheckStrlcpycatArguments(TheCall, FnInfo);
580   else if (CMId == Builtin::BIstrncat)
581     CheckStrncatArguments(TheCall, FnInfo);
582   else
583     CheckMemaccessArguments(TheCall, CMId, FnInfo);
584 
585   return false;
586 }
587 
588 bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac,
589                                Expr **Args, unsigned NumArgs) {
590   VariadicCallType CallType =
591       Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply;
592 
593   checkCall(Method, llvm::makeArrayRef<const Expr *>(Args, NumArgs),
594             Method->param_size(),
595             /*IsMemberFunction=*/false,
596             lbrac, Method->getSourceRange(), CallType);
597 
598   return false;
599 }
600 
601 bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall,
602                           const FunctionProtoType *Proto) {
603   const VarDecl *V = dyn_cast<VarDecl>(NDecl);
604   if (!V)
605     return false;
606 
607   QualType Ty = V->getType();
608   if (!Ty->isBlockPointerType())
609     return false;
610 
611   VariadicCallType CallType =
612       Proto && Proto->isVariadic() ? VariadicBlock : VariadicDoesNotApply ;
613   unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0;
614 
615   checkCall(NDecl,
616             llvm::makeArrayRef<const Expr *>(TheCall->getArgs(),
617                                              TheCall->getNumArgs()),
618             NumProtoArgs, /*IsMemberFunction=*/false,
619             TheCall->getRParenLoc(),
620             TheCall->getCallee()->getSourceRange(), CallType);
621 
622   return false;
623 }
624 
625 ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult,
626                                          AtomicExpr::AtomicOp Op) {
627   CallExpr *TheCall = cast<CallExpr>(TheCallResult.get());
628   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
629 
630   // All these operations take one of the following forms:
631   enum {
632     // C    __c11_atomic_init(A *, C)
633     Init,
634     // C    __c11_atomic_load(A *, int)
635     Load,
636     // void __atomic_load(A *, CP, int)
637     Copy,
638     // C    __c11_atomic_add(A *, M, int)
639     Arithmetic,
640     // C    __atomic_exchange_n(A *, CP, int)
641     Xchg,
642     // void __atomic_exchange(A *, C *, CP, int)
643     GNUXchg,
644     // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int)
645     C11CmpXchg,
646     // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int)
647     GNUCmpXchg
648   } Form = Init;
649   const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 };
650   const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 };
651   // where:
652   //   C is an appropriate type,
653   //   A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins,
654   //   CP is C for __c11 builtins and GNU _n builtins and is C * otherwise,
655   //   M is C if C is an integer, and ptrdiff_t if C is a pointer, and
656   //   the int parameters are for orderings.
657 
658   assert(AtomicExpr::AO__c11_atomic_init == 0 &&
659          AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load
660          && "need to update code for modified C11 atomics");
661   bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init &&
662                Op <= AtomicExpr::AO__c11_atomic_fetch_xor;
663   bool IsN = Op == AtomicExpr::AO__atomic_load_n ||
664              Op == AtomicExpr::AO__atomic_store_n ||
665              Op == AtomicExpr::AO__atomic_exchange_n ||
666              Op == AtomicExpr::AO__atomic_compare_exchange_n;
667   bool IsAddSub = false;
668 
669   switch (Op) {
670   case AtomicExpr::AO__c11_atomic_init:
671     Form = Init;
672     break;
673 
674   case AtomicExpr::AO__c11_atomic_load:
675   case AtomicExpr::AO__atomic_load_n:
676     Form = Load;
677     break;
678 
679   case AtomicExpr::AO__c11_atomic_store:
680   case AtomicExpr::AO__atomic_load:
681   case AtomicExpr::AO__atomic_store:
682   case AtomicExpr::AO__atomic_store_n:
683     Form = Copy;
684     break;
685 
686   case AtomicExpr::AO__c11_atomic_fetch_add:
687   case AtomicExpr::AO__c11_atomic_fetch_sub:
688   case AtomicExpr::AO__atomic_fetch_add:
689   case AtomicExpr::AO__atomic_fetch_sub:
690   case AtomicExpr::AO__atomic_add_fetch:
691   case AtomicExpr::AO__atomic_sub_fetch:
692     IsAddSub = true;
693     // Fall through.
694   case AtomicExpr::AO__c11_atomic_fetch_and:
695   case AtomicExpr::AO__c11_atomic_fetch_or:
696   case AtomicExpr::AO__c11_atomic_fetch_xor:
697   case AtomicExpr::AO__atomic_fetch_and:
698   case AtomicExpr::AO__atomic_fetch_or:
699   case AtomicExpr::AO__atomic_fetch_xor:
700   case AtomicExpr::AO__atomic_fetch_nand:
701   case AtomicExpr::AO__atomic_and_fetch:
702   case AtomicExpr::AO__atomic_or_fetch:
703   case AtomicExpr::AO__atomic_xor_fetch:
704   case AtomicExpr::AO__atomic_nand_fetch:
705     Form = Arithmetic;
706     break;
707 
708   case AtomicExpr::AO__c11_atomic_exchange:
709   case AtomicExpr::AO__atomic_exchange_n:
710     Form = Xchg;
711     break;
712 
713   case AtomicExpr::AO__atomic_exchange:
714     Form = GNUXchg;
715     break;
716 
717   case AtomicExpr::AO__c11_atomic_compare_exchange_strong:
718   case AtomicExpr::AO__c11_atomic_compare_exchange_weak:
719     Form = C11CmpXchg;
720     break;
721 
722   case AtomicExpr::AO__atomic_compare_exchange:
723   case AtomicExpr::AO__atomic_compare_exchange_n:
724     Form = GNUCmpXchg;
725     break;
726   }
727 
728   // Check we have the right number of arguments.
729   if (TheCall->getNumArgs() < NumArgs[Form]) {
730     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
731       << 0 << NumArgs[Form] << TheCall->getNumArgs()
732       << TheCall->getCallee()->getSourceRange();
733     return ExprError();
734   } else if (TheCall->getNumArgs() > NumArgs[Form]) {
735     Diag(TheCall->getArg(NumArgs[Form])->getLocStart(),
736          diag::err_typecheck_call_too_many_args)
737       << 0 << NumArgs[Form] << TheCall->getNumArgs()
738       << TheCall->getCallee()->getSourceRange();
739     return ExprError();
740   }
741 
742   // Inspect the first argument of the atomic operation.
743   Expr *Ptr = TheCall->getArg(0);
744   Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get();
745   const PointerType *pointerType = Ptr->getType()->getAs<PointerType>();
746   if (!pointerType) {
747     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
748       << Ptr->getType() << Ptr->getSourceRange();
749     return ExprError();
750   }
751 
752   // For a __c11 builtin, this should be a pointer to an _Atomic type.
753   QualType AtomTy = pointerType->getPointeeType(); // 'A'
754   QualType ValType = AtomTy; // 'C'
755   if (IsC11) {
756     if (!AtomTy->isAtomicType()) {
757       Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic)
758         << Ptr->getType() << Ptr->getSourceRange();
759       return ExprError();
760     }
761     if (AtomTy.isConstQualified()) {
762       Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic)
763         << Ptr->getType() << Ptr->getSourceRange();
764       return ExprError();
765     }
766     ValType = AtomTy->getAs<AtomicType>()->getValueType();
767   }
768 
769   // For an arithmetic operation, the implied arithmetic must be well-formed.
770   if (Form == Arithmetic) {
771     // gcc does not enforce these rules for GNU atomics, but we do so for sanity.
772     if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) {
773       Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr)
774         << IsC11 << Ptr->getType() << Ptr->getSourceRange();
775       return ExprError();
776     }
777     if (!IsAddSub && !ValType->isIntegerType()) {
778       Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int)
779         << IsC11 << Ptr->getType() << Ptr->getSourceRange();
780       return ExprError();
781     }
782   } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) {
783     // For __atomic_*_n operations, the value type must be a scalar integral or
784     // pointer type which is 1, 2, 4, 8 or 16 bytes in length.
785     Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr)
786       << IsC11 << Ptr->getType() << Ptr->getSourceRange();
787     return ExprError();
788   }
789 
790   if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) {
791     // For GNU atomics, require a trivially-copyable type. This is not part of
792     // the GNU atomics specification, but we enforce it for sanity.
793     Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy)
794       << Ptr->getType() << Ptr->getSourceRange();
795     return ExprError();
796   }
797 
798   // FIXME: For any builtin other than a load, the ValType must not be
799   // const-qualified.
800 
801   switch (ValType.getObjCLifetime()) {
802   case Qualifiers::OCL_None:
803   case Qualifiers::OCL_ExplicitNone:
804     // okay
805     break;
806 
807   case Qualifiers::OCL_Weak:
808   case Qualifiers::OCL_Strong:
809   case Qualifiers::OCL_Autoreleasing:
810     // FIXME: Can this happen? By this point, ValType should be known
811     // to be trivially copyable.
812     Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
813       << ValType << Ptr->getSourceRange();
814     return ExprError();
815   }
816 
817   QualType ResultType = ValType;
818   if (Form == Copy || Form == GNUXchg || Form == Init)
819     ResultType = Context.VoidTy;
820   else if (Form == C11CmpXchg || Form == GNUCmpXchg)
821     ResultType = Context.BoolTy;
822 
823   // The type of a parameter passed 'by value'. In the GNU atomics, such
824   // arguments are actually passed as pointers.
825   QualType ByValType = ValType; // 'CP'
826   if (!IsC11 && !IsN)
827     ByValType = Ptr->getType();
828 
829   // The first argument --- the pointer --- has a fixed type; we
830   // deduce the types of the rest of the arguments accordingly.  Walk
831   // the remaining arguments, converting them to the deduced value type.
832   for (unsigned i = 1; i != NumArgs[Form]; ++i) {
833     QualType Ty;
834     if (i < NumVals[Form] + 1) {
835       switch (i) {
836       case 1:
837         // The second argument is the non-atomic operand. For arithmetic, this
838         // is always passed by value, and for a compare_exchange it is always
839         // passed by address. For the rest, GNU uses by-address and C11 uses
840         // by-value.
841         assert(Form != Load);
842         if (Form == Init || (Form == Arithmetic && ValType->isIntegerType()))
843           Ty = ValType;
844         else if (Form == Copy || Form == Xchg)
845           Ty = ByValType;
846         else if (Form == Arithmetic)
847           Ty = Context.getPointerDiffType();
848         else
849           Ty = Context.getPointerType(ValType.getUnqualifiedType());
850         break;
851       case 2:
852         // The third argument to compare_exchange / GNU exchange is a
853         // (pointer to a) desired value.
854         Ty = ByValType;
855         break;
856       case 3:
857         // The fourth argument to GNU compare_exchange is a 'weak' flag.
858         Ty = Context.BoolTy;
859         break;
860       }
861     } else {
862       // The order(s) are always converted to int.
863       Ty = Context.IntTy;
864     }
865 
866     InitializedEntity Entity =
867         InitializedEntity::InitializeParameter(Context, Ty, false);
868     ExprResult Arg = TheCall->getArg(i);
869     Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
870     if (Arg.isInvalid())
871       return true;
872     TheCall->setArg(i, Arg.get());
873   }
874 
875   // Permute the arguments into a 'consistent' order.
876   SmallVector<Expr*, 5> SubExprs;
877   SubExprs.push_back(Ptr);
878   switch (Form) {
879   case Init:
880     // Note, AtomicExpr::getVal1() has a special case for this atomic.
881     SubExprs.push_back(TheCall->getArg(1)); // Val1
882     break;
883   case Load:
884     SubExprs.push_back(TheCall->getArg(1)); // Order
885     break;
886   case Copy:
887   case Arithmetic:
888   case Xchg:
889     SubExprs.push_back(TheCall->getArg(2)); // Order
890     SubExprs.push_back(TheCall->getArg(1)); // Val1
891     break;
892   case GNUXchg:
893     // Note, AtomicExpr::getVal2() has a special case for this atomic.
894     SubExprs.push_back(TheCall->getArg(3)); // Order
895     SubExprs.push_back(TheCall->getArg(1)); // Val1
896     SubExprs.push_back(TheCall->getArg(2)); // Val2
897     break;
898   case C11CmpXchg:
899     SubExprs.push_back(TheCall->getArg(3)); // Order
900     SubExprs.push_back(TheCall->getArg(1)); // Val1
901     SubExprs.push_back(TheCall->getArg(4)); // OrderFail
902     SubExprs.push_back(TheCall->getArg(2)); // Val2
903     break;
904   case GNUCmpXchg:
905     SubExprs.push_back(TheCall->getArg(4)); // Order
906     SubExprs.push_back(TheCall->getArg(1)); // Val1
907     SubExprs.push_back(TheCall->getArg(5)); // OrderFail
908     SubExprs.push_back(TheCall->getArg(2)); // Val2
909     SubExprs.push_back(TheCall->getArg(3)); // Weak
910     break;
911   }
912 
913   return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(),
914                                         SubExprs, ResultType, Op,
915                                         TheCall->getRParenLoc()));
916 }
917 
918 
919 /// checkBuiltinArgument - Given a call to a builtin function, perform
920 /// normal type-checking on the given argument, updating the call in
921 /// place.  This is useful when a builtin function requires custom
922 /// type-checking for some of its arguments but not necessarily all of
923 /// them.
924 ///
925 /// Returns true on error.
926 static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) {
927   FunctionDecl *Fn = E->getDirectCallee();
928   assert(Fn && "builtin call without direct callee!");
929 
930   ParmVarDecl *Param = Fn->getParamDecl(ArgIndex);
931   InitializedEntity Entity =
932     InitializedEntity::InitializeParameter(S.Context, Param);
933 
934   ExprResult Arg = E->getArg(0);
935   Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg);
936   if (Arg.isInvalid())
937     return true;
938 
939   E->setArg(ArgIndex, Arg.take());
940   return false;
941 }
942 
943 /// SemaBuiltinAtomicOverloaded - We have a call to a function like
944 /// __sync_fetch_and_add, which is an overloaded function based on the pointer
945 /// type of its first argument.  The main ActOnCallExpr routines have already
946 /// promoted the types of arguments because all of these calls are prototyped as
947 /// void(...).
948 ///
949 /// This function goes through and does final semantic checking for these
950 /// builtins,
951 ExprResult
952 Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
953   CallExpr *TheCall = (CallExpr *)TheCallResult.get();
954   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
955   FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
956 
957   // Ensure that we have at least one argument to do type inference from.
958   if (TheCall->getNumArgs() < 1) {
959     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
960       << 0 << 1 << TheCall->getNumArgs()
961       << TheCall->getCallee()->getSourceRange();
962     return ExprError();
963   }
964 
965   // Inspect the first argument of the atomic builtin.  This should always be
966   // a pointer type, whose element is an integral scalar or pointer type.
967   // Because it is a pointer type, we don't have to worry about any implicit
968   // casts here.
969   // FIXME: We don't allow floating point scalars as input.
970   Expr *FirstArg = TheCall->getArg(0);
971   ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg);
972   if (FirstArgResult.isInvalid())
973     return ExprError();
974   FirstArg = FirstArgResult.take();
975   TheCall->setArg(0, FirstArg);
976 
977   const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>();
978   if (!pointerType) {
979     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
980       << FirstArg->getType() << FirstArg->getSourceRange();
981     return ExprError();
982   }
983 
984   QualType ValType = pointerType->getPointeeType();
985   if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
986       !ValType->isBlockPointerType()) {
987     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
988       << FirstArg->getType() << FirstArg->getSourceRange();
989     return ExprError();
990   }
991 
992   switch (ValType.getObjCLifetime()) {
993   case Qualifiers::OCL_None:
994   case Qualifiers::OCL_ExplicitNone:
995     // okay
996     break;
997 
998   case Qualifiers::OCL_Weak:
999   case Qualifiers::OCL_Strong:
1000   case Qualifiers::OCL_Autoreleasing:
1001     Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
1002       << ValType << FirstArg->getSourceRange();
1003     return ExprError();
1004   }
1005 
1006   // Strip any qualifiers off ValType.
1007   ValType = ValType.getUnqualifiedType();
1008 
1009   // The majority of builtins return a value, but a few have special return
1010   // types, so allow them to override appropriately below.
1011   QualType ResultType = ValType;
1012 
1013   // We need to figure out which concrete builtin this maps onto.  For example,
1014   // __sync_fetch_and_add with a 2 byte object turns into
1015   // __sync_fetch_and_add_2.
1016 #define BUILTIN_ROW(x) \
1017   { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
1018     Builtin::BI##x##_8, Builtin::BI##x##_16 }
1019 
1020   static const unsigned BuiltinIndices[][5] = {
1021     BUILTIN_ROW(__sync_fetch_and_add),
1022     BUILTIN_ROW(__sync_fetch_and_sub),
1023     BUILTIN_ROW(__sync_fetch_and_or),
1024     BUILTIN_ROW(__sync_fetch_and_and),
1025     BUILTIN_ROW(__sync_fetch_and_xor),
1026 
1027     BUILTIN_ROW(__sync_add_and_fetch),
1028     BUILTIN_ROW(__sync_sub_and_fetch),
1029     BUILTIN_ROW(__sync_and_and_fetch),
1030     BUILTIN_ROW(__sync_or_and_fetch),
1031     BUILTIN_ROW(__sync_xor_and_fetch),
1032 
1033     BUILTIN_ROW(__sync_val_compare_and_swap),
1034     BUILTIN_ROW(__sync_bool_compare_and_swap),
1035     BUILTIN_ROW(__sync_lock_test_and_set),
1036     BUILTIN_ROW(__sync_lock_release),
1037     BUILTIN_ROW(__sync_swap)
1038   };
1039 #undef BUILTIN_ROW
1040 
1041   // Determine the index of the size.
1042   unsigned SizeIndex;
1043   switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
1044   case 1: SizeIndex = 0; break;
1045   case 2: SizeIndex = 1; break;
1046   case 4: SizeIndex = 2; break;
1047   case 8: SizeIndex = 3; break;
1048   case 16: SizeIndex = 4; break;
1049   default:
1050     Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
1051       << FirstArg->getType() << FirstArg->getSourceRange();
1052     return ExprError();
1053   }
1054 
1055   // Each of these builtins has one pointer argument, followed by some number of
1056   // values (0, 1 or 2) followed by a potentially empty varags list of stuff
1057   // that we ignore.  Find out which row of BuiltinIndices to read from as well
1058   // as the number of fixed args.
1059   unsigned BuiltinID = FDecl->getBuiltinID();
1060   unsigned BuiltinIndex, NumFixed = 1;
1061   switch (BuiltinID) {
1062   default: llvm_unreachable("Unknown overloaded atomic builtin!");
1063   case Builtin::BI__sync_fetch_and_add:
1064   case Builtin::BI__sync_fetch_and_add_1:
1065   case Builtin::BI__sync_fetch_and_add_2:
1066   case Builtin::BI__sync_fetch_and_add_4:
1067   case Builtin::BI__sync_fetch_and_add_8:
1068   case Builtin::BI__sync_fetch_and_add_16:
1069     BuiltinIndex = 0;
1070     break;
1071 
1072   case Builtin::BI__sync_fetch_and_sub:
1073   case Builtin::BI__sync_fetch_and_sub_1:
1074   case Builtin::BI__sync_fetch_and_sub_2:
1075   case Builtin::BI__sync_fetch_and_sub_4:
1076   case Builtin::BI__sync_fetch_and_sub_8:
1077   case Builtin::BI__sync_fetch_and_sub_16:
1078     BuiltinIndex = 1;
1079     break;
1080 
1081   case Builtin::BI__sync_fetch_and_or:
1082   case Builtin::BI__sync_fetch_and_or_1:
1083   case Builtin::BI__sync_fetch_and_or_2:
1084   case Builtin::BI__sync_fetch_and_or_4:
1085   case Builtin::BI__sync_fetch_and_or_8:
1086   case Builtin::BI__sync_fetch_and_or_16:
1087     BuiltinIndex = 2;
1088     break;
1089 
1090   case Builtin::BI__sync_fetch_and_and:
1091   case Builtin::BI__sync_fetch_and_and_1:
1092   case Builtin::BI__sync_fetch_and_and_2:
1093   case Builtin::BI__sync_fetch_and_and_4:
1094   case Builtin::BI__sync_fetch_and_and_8:
1095   case Builtin::BI__sync_fetch_and_and_16:
1096     BuiltinIndex = 3;
1097     break;
1098 
1099   case Builtin::BI__sync_fetch_and_xor:
1100   case Builtin::BI__sync_fetch_and_xor_1:
1101   case Builtin::BI__sync_fetch_and_xor_2:
1102   case Builtin::BI__sync_fetch_and_xor_4:
1103   case Builtin::BI__sync_fetch_and_xor_8:
1104   case Builtin::BI__sync_fetch_and_xor_16:
1105     BuiltinIndex = 4;
1106     break;
1107 
1108   case Builtin::BI__sync_add_and_fetch:
1109   case Builtin::BI__sync_add_and_fetch_1:
1110   case Builtin::BI__sync_add_and_fetch_2:
1111   case Builtin::BI__sync_add_and_fetch_4:
1112   case Builtin::BI__sync_add_and_fetch_8:
1113   case Builtin::BI__sync_add_and_fetch_16:
1114     BuiltinIndex = 5;
1115     break;
1116 
1117   case Builtin::BI__sync_sub_and_fetch:
1118   case Builtin::BI__sync_sub_and_fetch_1:
1119   case Builtin::BI__sync_sub_and_fetch_2:
1120   case Builtin::BI__sync_sub_and_fetch_4:
1121   case Builtin::BI__sync_sub_and_fetch_8:
1122   case Builtin::BI__sync_sub_and_fetch_16:
1123     BuiltinIndex = 6;
1124     break;
1125 
1126   case Builtin::BI__sync_and_and_fetch:
1127   case Builtin::BI__sync_and_and_fetch_1:
1128   case Builtin::BI__sync_and_and_fetch_2:
1129   case Builtin::BI__sync_and_and_fetch_4:
1130   case Builtin::BI__sync_and_and_fetch_8:
1131   case Builtin::BI__sync_and_and_fetch_16:
1132     BuiltinIndex = 7;
1133     break;
1134 
1135   case Builtin::BI__sync_or_and_fetch:
1136   case Builtin::BI__sync_or_and_fetch_1:
1137   case Builtin::BI__sync_or_and_fetch_2:
1138   case Builtin::BI__sync_or_and_fetch_4:
1139   case Builtin::BI__sync_or_and_fetch_8:
1140   case Builtin::BI__sync_or_and_fetch_16:
1141     BuiltinIndex = 8;
1142     break;
1143 
1144   case Builtin::BI__sync_xor_and_fetch:
1145   case Builtin::BI__sync_xor_and_fetch_1:
1146   case Builtin::BI__sync_xor_and_fetch_2:
1147   case Builtin::BI__sync_xor_and_fetch_4:
1148   case Builtin::BI__sync_xor_and_fetch_8:
1149   case Builtin::BI__sync_xor_and_fetch_16:
1150     BuiltinIndex = 9;
1151     break;
1152 
1153   case Builtin::BI__sync_val_compare_and_swap:
1154   case Builtin::BI__sync_val_compare_and_swap_1:
1155   case Builtin::BI__sync_val_compare_and_swap_2:
1156   case Builtin::BI__sync_val_compare_and_swap_4:
1157   case Builtin::BI__sync_val_compare_and_swap_8:
1158   case Builtin::BI__sync_val_compare_and_swap_16:
1159     BuiltinIndex = 10;
1160     NumFixed = 2;
1161     break;
1162 
1163   case Builtin::BI__sync_bool_compare_and_swap:
1164   case Builtin::BI__sync_bool_compare_and_swap_1:
1165   case Builtin::BI__sync_bool_compare_and_swap_2:
1166   case Builtin::BI__sync_bool_compare_and_swap_4:
1167   case Builtin::BI__sync_bool_compare_and_swap_8:
1168   case Builtin::BI__sync_bool_compare_and_swap_16:
1169     BuiltinIndex = 11;
1170     NumFixed = 2;
1171     ResultType = Context.BoolTy;
1172     break;
1173 
1174   case Builtin::BI__sync_lock_test_and_set:
1175   case Builtin::BI__sync_lock_test_and_set_1:
1176   case Builtin::BI__sync_lock_test_and_set_2:
1177   case Builtin::BI__sync_lock_test_and_set_4:
1178   case Builtin::BI__sync_lock_test_and_set_8:
1179   case Builtin::BI__sync_lock_test_and_set_16:
1180     BuiltinIndex = 12;
1181     break;
1182 
1183   case Builtin::BI__sync_lock_release:
1184   case Builtin::BI__sync_lock_release_1:
1185   case Builtin::BI__sync_lock_release_2:
1186   case Builtin::BI__sync_lock_release_4:
1187   case Builtin::BI__sync_lock_release_8:
1188   case Builtin::BI__sync_lock_release_16:
1189     BuiltinIndex = 13;
1190     NumFixed = 0;
1191     ResultType = Context.VoidTy;
1192     break;
1193 
1194   case Builtin::BI__sync_swap:
1195   case Builtin::BI__sync_swap_1:
1196   case Builtin::BI__sync_swap_2:
1197   case Builtin::BI__sync_swap_4:
1198   case Builtin::BI__sync_swap_8:
1199   case Builtin::BI__sync_swap_16:
1200     BuiltinIndex = 14;
1201     break;
1202   }
1203 
1204   // Now that we know how many fixed arguments we expect, first check that we
1205   // have at least that many.
1206   if (TheCall->getNumArgs() < 1+NumFixed) {
1207     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
1208       << 0 << 1+NumFixed << TheCall->getNumArgs()
1209       << TheCall->getCallee()->getSourceRange();
1210     return ExprError();
1211   }
1212 
1213   // Get the decl for the concrete builtin from this, we can tell what the
1214   // concrete integer type we should convert to is.
1215   unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
1216   const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
1217   FunctionDecl *NewBuiltinDecl;
1218   if (NewBuiltinID == BuiltinID)
1219     NewBuiltinDecl = FDecl;
1220   else {
1221     // Perform builtin lookup to avoid redeclaring it.
1222     DeclarationName DN(&Context.Idents.get(NewBuiltinName));
1223     LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName);
1224     LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true);
1225     assert(Res.getFoundDecl());
1226     NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl());
1227     if (NewBuiltinDecl == 0)
1228       return ExprError();
1229   }
1230 
1231   // The first argument --- the pointer --- has a fixed type; we
1232   // deduce the types of the rest of the arguments accordingly.  Walk
1233   // the remaining arguments, converting them to the deduced value type.
1234   for (unsigned i = 0; i != NumFixed; ++i) {
1235     ExprResult Arg = TheCall->getArg(i+1);
1236 
1237     // GCC does an implicit conversion to the pointer or integer ValType.  This
1238     // can fail in some cases (1i -> int**), check for this error case now.
1239     // Initialize the argument.
1240     InitializedEntity Entity = InitializedEntity::InitializeParameter(Context,
1241                                                    ValType, /*consume*/ false);
1242     Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
1243     if (Arg.isInvalid())
1244       return ExprError();
1245 
1246     // Okay, we have something that *can* be converted to the right type.  Check
1247     // to see if there is a potentially weird extension going on here.  This can
1248     // happen when you do an atomic operation on something like an char* and
1249     // pass in 42.  The 42 gets converted to char.  This is even more strange
1250     // for things like 45.123 -> char, etc.
1251     // FIXME: Do this check.
1252     TheCall->setArg(i+1, Arg.take());
1253   }
1254 
1255   ASTContext& Context = this->getASTContext();
1256 
1257   // Create a new DeclRefExpr to refer to the new decl.
1258   DeclRefExpr* NewDRE = DeclRefExpr::Create(
1259       Context,
1260       DRE->getQualifierLoc(),
1261       SourceLocation(),
1262       NewBuiltinDecl,
1263       /*enclosing*/ false,
1264       DRE->getLocation(),
1265       Context.BuiltinFnTy,
1266       DRE->getValueKind());
1267 
1268   // Set the callee in the CallExpr.
1269   // FIXME: This loses syntactic information.
1270   QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType());
1271   ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy,
1272                                               CK_BuiltinFnToFnPtr);
1273   TheCall->setCallee(PromotedCall.take());
1274 
1275   // Change the result type of the call to match the original value type. This
1276   // is arbitrary, but the codegen for these builtins ins design to handle it
1277   // gracefully.
1278   TheCall->setType(ResultType);
1279 
1280   return TheCallResult;
1281 }
1282 
1283 /// CheckObjCString - Checks that the argument to the builtin
1284 /// CFString constructor is correct
1285 /// Note: It might also make sense to do the UTF-16 conversion here (would
1286 /// simplify the backend).
1287 bool Sema::CheckObjCString(Expr *Arg) {
1288   Arg = Arg->IgnoreParenCasts();
1289   StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
1290 
1291   if (!Literal || !Literal->isAscii()) {
1292     Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
1293       << Arg->getSourceRange();
1294     return true;
1295   }
1296 
1297   if (Literal->containsNonAsciiOrNull()) {
1298     StringRef String = Literal->getString();
1299     unsigned NumBytes = String.size();
1300     SmallVector<UTF16, 128> ToBuf(NumBytes);
1301     const UTF8 *FromPtr = (const UTF8 *)String.data();
1302     UTF16 *ToPtr = &ToBuf[0];
1303 
1304     ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
1305                                                  &ToPtr, ToPtr + NumBytes,
1306                                                  strictConversion);
1307     // Check for conversion failure.
1308     if (Result != conversionOK)
1309       Diag(Arg->getLocStart(),
1310            diag::warn_cfstring_truncated) << Arg->getSourceRange();
1311   }
1312   return false;
1313 }
1314 
1315 /// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
1316 /// Emit an error and return true on failure, return false on success.
1317 bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
1318   Expr *Fn = TheCall->getCallee();
1319   if (TheCall->getNumArgs() > 2) {
1320     Diag(TheCall->getArg(2)->getLocStart(),
1321          diag::err_typecheck_call_too_many_args)
1322       << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1323       << Fn->getSourceRange()
1324       << SourceRange(TheCall->getArg(2)->getLocStart(),
1325                      (*(TheCall->arg_end()-1))->getLocEnd());
1326     return true;
1327   }
1328 
1329   if (TheCall->getNumArgs() < 2) {
1330     return Diag(TheCall->getLocEnd(),
1331       diag::err_typecheck_call_too_few_args_at_least)
1332       << 0 /*function call*/ << 2 << TheCall->getNumArgs();
1333   }
1334 
1335   // Type-check the first argument normally.
1336   if (checkBuiltinArgument(*this, TheCall, 0))
1337     return true;
1338 
1339   // Determine whether the current function is variadic or not.
1340   BlockScopeInfo *CurBlock = getCurBlock();
1341   bool isVariadic;
1342   if (CurBlock)
1343     isVariadic = CurBlock->TheDecl->isVariadic();
1344   else if (FunctionDecl *FD = getCurFunctionDecl())
1345     isVariadic = FD->isVariadic();
1346   else
1347     isVariadic = getCurMethodDecl()->isVariadic();
1348 
1349   if (!isVariadic) {
1350     Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
1351     return true;
1352   }
1353 
1354   // Verify that the second argument to the builtin is the last argument of the
1355   // current function or method.
1356   bool SecondArgIsLastNamedArgument = false;
1357   const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
1358 
1359   if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
1360     if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
1361       // FIXME: This isn't correct for methods (results in bogus warning).
1362       // Get the last formal in the current function.
1363       const ParmVarDecl *LastArg;
1364       if (CurBlock)
1365         LastArg = *(CurBlock->TheDecl->param_end()-1);
1366       else if (FunctionDecl *FD = getCurFunctionDecl())
1367         LastArg = *(FD->param_end()-1);
1368       else
1369         LastArg = *(getCurMethodDecl()->param_end()-1);
1370       SecondArgIsLastNamedArgument = PV == LastArg;
1371     }
1372   }
1373 
1374   if (!SecondArgIsLastNamedArgument)
1375     Diag(TheCall->getArg(1)->getLocStart(),
1376          diag::warn_second_parameter_of_va_start_not_last_named_argument);
1377   return false;
1378 }
1379 
1380 /// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
1381 /// friends.  This is declared to take (...), so we have to check everything.
1382 bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
1383   if (TheCall->getNumArgs() < 2)
1384     return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1385       << 0 << 2 << TheCall->getNumArgs()/*function call*/;
1386   if (TheCall->getNumArgs() > 2)
1387     return Diag(TheCall->getArg(2)->getLocStart(),
1388                 diag::err_typecheck_call_too_many_args)
1389       << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1390       << SourceRange(TheCall->getArg(2)->getLocStart(),
1391                      (*(TheCall->arg_end()-1))->getLocEnd());
1392 
1393   ExprResult OrigArg0 = TheCall->getArg(0);
1394   ExprResult OrigArg1 = TheCall->getArg(1);
1395 
1396   // Do standard promotions between the two arguments, returning their common
1397   // type.
1398   QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
1399   if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
1400     return true;
1401 
1402   // Make sure any conversions are pushed back into the call; this is
1403   // type safe since unordered compare builtins are declared as "_Bool
1404   // foo(...)".
1405   TheCall->setArg(0, OrigArg0.get());
1406   TheCall->setArg(1, OrigArg1.get());
1407 
1408   if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
1409     return false;
1410 
1411   // If the common type isn't a real floating type, then the arguments were
1412   // invalid for this operation.
1413   if (Res.isNull() || !Res->isRealFloatingType())
1414     return Diag(OrigArg0.get()->getLocStart(),
1415                 diag::err_typecheck_call_invalid_ordered_compare)
1416       << OrigArg0.get()->getType() << OrigArg1.get()->getType()
1417       << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
1418 
1419   return false;
1420 }
1421 
1422 /// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
1423 /// __builtin_isnan and friends.  This is declared to take (...), so we have
1424 /// to check everything. We expect the last argument to be a floating point
1425 /// value.
1426 bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
1427   if (TheCall->getNumArgs() < NumArgs)
1428     return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1429       << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
1430   if (TheCall->getNumArgs() > NumArgs)
1431     return Diag(TheCall->getArg(NumArgs)->getLocStart(),
1432                 diag::err_typecheck_call_too_many_args)
1433       << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
1434       << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
1435                      (*(TheCall->arg_end()-1))->getLocEnd());
1436 
1437   Expr *OrigArg = TheCall->getArg(NumArgs-1);
1438 
1439   if (OrigArg->isTypeDependent())
1440     return false;
1441 
1442   // This operation requires a non-_Complex floating-point number.
1443   if (!OrigArg->getType()->isRealFloatingType())
1444     return Diag(OrigArg->getLocStart(),
1445                 diag::err_typecheck_call_invalid_unary_fp)
1446       << OrigArg->getType() << OrigArg->getSourceRange();
1447 
1448   // If this is an implicit conversion from float -> double, remove it.
1449   if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
1450     Expr *CastArg = Cast->getSubExpr();
1451     if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
1452       assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
1453              "promotion from float to double is the only expected cast here");
1454       Cast->setSubExpr(0);
1455       TheCall->setArg(NumArgs-1, CastArg);
1456     }
1457   }
1458 
1459   return false;
1460 }
1461 
1462 /// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
1463 // This is declared to take (...), so we have to check everything.
1464 ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
1465   if (TheCall->getNumArgs() < 2)
1466     return ExprError(Diag(TheCall->getLocEnd(),
1467                           diag::err_typecheck_call_too_few_args_at_least)
1468       << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1469       << TheCall->getSourceRange());
1470 
1471   // Determine which of the following types of shufflevector we're checking:
1472   // 1) unary, vector mask: (lhs, mask)
1473   // 2) binary, vector mask: (lhs, rhs, mask)
1474   // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
1475   QualType resType = TheCall->getArg(0)->getType();
1476   unsigned numElements = 0;
1477 
1478   if (!TheCall->getArg(0)->isTypeDependent() &&
1479       !TheCall->getArg(1)->isTypeDependent()) {
1480     QualType LHSType = TheCall->getArg(0)->getType();
1481     QualType RHSType = TheCall->getArg(1)->getType();
1482 
1483     if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
1484       Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
1485         << SourceRange(TheCall->getArg(0)->getLocStart(),
1486                        TheCall->getArg(1)->getLocEnd());
1487       return ExprError();
1488     }
1489 
1490     numElements = LHSType->getAs<VectorType>()->getNumElements();
1491     unsigned numResElements = TheCall->getNumArgs() - 2;
1492 
1493     // Check to see if we have a call with 2 vector arguments, the unary shuffle
1494     // with mask.  If so, verify that RHS is an integer vector type with the
1495     // same number of elts as lhs.
1496     if (TheCall->getNumArgs() == 2) {
1497       if (!RHSType->hasIntegerRepresentation() ||
1498           RHSType->getAs<VectorType>()->getNumElements() != numElements)
1499         Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1500           << SourceRange(TheCall->getArg(1)->getLocStart(),
1501                          TheCall->getArg(1)->getLocEnd());
1502       numResElements = numElements;
1503     }
1504     else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
1505       Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1506         << SourceRange(TheCall->getArg(0)->getLocStart(),
1507                        TheCall->getArg(1)->getLocEnd());
1508       return ExprError();
1509     } else if (numElements != numResElements) {
1510       QualType eltType = LHSType->getAs<VectorType>()->getElementType();
1511       resType = Context.getVectorType(eltType, numResElements,
1512                                       VectorType::GenericVector);
1513     }
1514   }
1515 
1516   for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
1517     if (TheCall->getArg(i)->isTypeDependent() ||
1518         TheCall->getArg(i)->isValueDependent())
1519       continue;
1520 
1521     llvm::APSInt Result(32);
1522     if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
1523       return ExprError(Diag(TheCall->getLocStart(),
1524                   diag::err_shufflevector_nonconstant_argument)
1525                 << TheCall->getArg(i)->getSourceRange());
1526 
1527     if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
1528       return ExprError(Diag(TheCall->getLocStart(),
1529                   diag::err_shufflevector_argument_too_large)
1530                << TheCall->getArg(i)->getSourceRange());
1531   }
1532 
1533   SmallVector<Expr*, 32> exprs;
1534 
1535   for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
1536     exprs.push_back(TheCall->getArg(i));
1537     TheCall->setArg(i, 0);
1538   }
1539 
1540   return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType,
1541                                             TheCall->getCallee()->getLocStart(),
1542                                             TheCall->getRParenLoc()));
1543 }
1544 
1545 /// SemaBuiltinPrefetch - Handle __builtin_prefetch.
1546 // This is declared to take (const void*, ...) and can take two
1547 // optional constant int args.
1548 bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
1549   unsigned NumArgs = TheCall->getNumArgs();
1550 
1551   if (NumArgs > 3)
1552     return Diag(TheCall->getLocEnd(),
1553              diag::err_typecheck_call_too_many_args_at_most)
1554              << 0 /*function call*/ << 3 << NumArgs
1555              << TheCall->getSourceRange();
1556 
1557   // Argument 0 is checked for us and the remaining arguments must be
1558   // constant integers.
1559   for (unsigned i = 1; i != NumArgs; ++i) {
1560     Expr *Arg = TheCall->getArg(i);
1561 
1562     // We can't check the value of a dependent argument.
1563     if (Arg->isTypeDependent() || Arg->isValueDependent())
1564       continue;
1565 
1566     llvm::APSInt Result;
1567     if (SemaBuiltinConstantArg(TheCall, i, Result))
1568       return true;
1569 
1570     // FIXME: gcc issues a warning and rewrites these to 0. These
1571     // seems especially odd for the third argument since the default
1572     // is 3.
1573     if (i == 1) {
1574       if (Result.getLimitedValue() > 1)
1575         return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1576              << "0" << "1" << Arg->getSourceRange();
1577     } else {
1578       if (Result.getLimitedValue() > 3)
1579         return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1580             << "0" << "3" << Arg->getSourceRange();
1581     }
1582   }
1583 
1584   return false;
1585 }
1586 
1587 /// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
1588 /// TheCall is a constant expression.
1589 bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
1590                                   llvm::APSInt &Result) {
1591   Expr *Arg = TheCall->getArg(ArgNum);
1592   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
1593   FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
1594 
1595   if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
1596 
1597   if (!Arg->isIntegerConstantExpr(Result, Context))
1598     return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
1599                 << FDecl->getDeclName() <<  Arg->getSourceRange();
1600 
1601   return false;
1602 }
1603 
1604 /// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
1605 /// int type). This simply type checks that type is one of the defined
1606 /// constants (0-3).
1607 // For compatibility check 0-3, llvm only handles 0 and 2.
1608 bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
1609   llvm::APSInt Result;
1610 
1611   // We can't check the value of a dependent argument.
1612   if (TheCall->getArg(1)->isTypeDependent() ||
1613       TheCall->getArg(1)->isValueDependent())
1614     return false;
1615 
1616   // Check constant-ness first.
1617   if (SemaBuiltinConstantArg(TheCall, 1, Result))
1618     return true;
1619 
1620   Expr *Arg = TheCall->getArg(1);
1621   if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
1622     return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1623              << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1624   }
1625 
1626   return false;
1627 }
1628 
1629 /// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
1630 /// This checks that val is a constant 1.
1631 bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
1632   Expr *Arg = TheCall->getArg(1);
1633   llvm::APSInt Result;
1634 
1635   // TODO: This is less than ideal. Overload this to take a value.
1636   if (SemaBuiltinConstantArg(TheCall, 1, Result))
1637     return true;
1638 
1639   if (Result != 1)
1640     return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
1641              << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1642 
1643   return false;
1644 }
1645 
1646 // Determine if an expression is a string literal or constant string.
1647 // If this function returns false on the arguments to a function expecting a
1648 // format string, we will usually need to emit a warning.
1649 // True string literals are then checked by CheckFormatString.
1650 Sema::StringLiteralCheckType
1651 Sema::checkFormatStringExpr(const Expr *E, ArrayRef<const Expr *> Args,
1652                             bool HasVAListArg,
1653                             unsigned format_idx, unsigned firstDataArg,
1654                             FormatStringType Type, VariadicCallType CallType,
1655                             bool inFunctionCall) {
1656  tryAgain:
1657   if (E->isTypeDependent() || E->isValueDependent())
1658     return SLCT_NotALiteral;
1659 
1660   E = E->IgnoreParenCasts();
1661 
1662   if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull))
1663     // Technically -Wformat-nonliteral does not warn about this case.
1664     // The behavior of printf and friends in this case is implementation
1665     // dependent.  Ideally if the format string cannot be null then
1666     // it should have a 'nonnull' attribute in the function prototype.
1667     return SLCT_CheckedLiteral;
1668 
1669   switch (E->getStmtClass()) {
1670   case Stmt::BinaryConditionalOperatorClass:
1671   case Stmt::ConditionalOperatorClass: {
1672     // The expression is a literal if both sub-expressions were, and it was
1673     // completely checked only if both sub-expressions were checked.
1674     const AbstractConditionalOperator *C =
1675         cast<AbstractConditionalOperator>(E);
1676     StringLiteralCheckType Left =
1677         checkFormatStringExpr(C->getTrueExpr(), Args,
1678                               HasVAListArg, format_idx, firstDataArg,
1679                               Type, CallType, inFunctionCall);
1680     if (Left == SLCT_NotALiteral)
1681       return SLCT_NotALiteral;
1682     StringLiteralCheckType Right =
1683         checkFormatStringExpr(C->getFalseExpr(), Args,
1684                               HasVAListArg, format_idx, firstDataArg,
1685                               Type, CallType, inFunctionCall);
1686     return Left < Right ? Left : Right;
1687   }
1688 
1689   case Stmt::ImplicitCastExprClass: {
1690     E = cast<ImplicitCastExpr>(E)->getSubExpr();
1691     goto tryAgain;
1692   }
1693 
1694   case Stmt::OpaqueValueExprClass:
1695     if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
1696       E = src;
1697       goto tryAgain;
1698     }
1699     return SLCT_NotALiteral;
1700 
1701   case Stmt::PredefinedExprClass:
1702     // While __func__, etc., are technically not string literals, they
1703     // cannot contain format specifiers and thus are not a security
1704     // liability.
1705     return SLCT_UncheckedLiteral;
1706 
1707   case Stmt::DeclRefExprClass: {
1708     const DeclRefExpr *DR = cast<DeclRefExpr>(E);
1709 
1710     // As an exception, do not flag errors for variables binding to
1711     // const string literals.
1712     if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
1713       bool isConstant = false;
1714       QualType T = DR->getType();
1715 
1716       if (const ArrayType *AT = Context.getAsArrayType(T)) {
1717         isConstant = AT->getElementType().isConstant(Context);
1718       } else if (const PointerType *PT = T->getAs<PointerType>()) {
1719         isConstant = T.isConstant(Context) &&
1720                      PT->getPointeeType().isConstant(Context);
1721       } else if (T->isObjCObjectPointerType()) {
1722         // In ObjC, there is usually no "const ObjectPointer" type,
1723         // so don't check if the pointee type is constant.
1724         isConstant = T.isConstant(Context);
1725       }
1726 
1727       if (isConstant) {
1728         if (const Expr *Init = VD->getAnyInitializer()) {
1729           // Look through initializers like const char c[] = { "foo" }
1730           if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) {
1731             if (InitList->isStringLiteralInit())
1732               Init = InitList->getInit(0)->IgnoreParenImpCasts();
1733           }
1734           return checkFormatStringExpr(Init, Args,
1735                                        HasVAListArg, format_idx,
1736                                        firstDataArg, Type, CallType,
1737                                        /*inFunctionCall*/false);
1738         }
1739       }
1740 
1741       // For vprintf* functions (i.e., HasVAListArg==true), we add a
1742       // special check to see if the format string is a function parameter
1743       // of the function calling the printf function.  If the function
1744       // has an attribute indicating it is a printf-like function, then we
1745       // should suppress warnings concerning non-literals being used in a call
1746       // to a vprintf function.  For example:
1747       //
1748       // void
1749       // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
1750       //      va_list ap;
1751       //      va_start(ap, fmt);
1752       //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
1753       //      ...
1754       //
1755       if (HasVAListArg) {
1756         if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) {
1757           if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) {
1758             int PVIndex = PV->getFunctionScopeIndex() + 1;
1759             for (specific_attr_iterator<FormatAttr>
1760                  i = ND->specific_attr_begin<FormatAttr>(),
1761                  e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) {
1762               FormatAttr *PVFormat = *i;
1763               // adjust for implicit parameter
1764               if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND))
1765                 if (MD->isInstance())
1766                   ++PVIndex;
1767               // We also check if the formats are compatible.
1768               // We can't pass a 'scanf' string to a 'printf' function.
1769               if (PVIndex == PVFormat->getFormatIdx() &&
1770                   Type == GetFormatStringType(PVFormat))
1771                 return SLCT_UncheckedLiteral;
1772             }
1773           }
1774         }
1775       }
1776     }
1777 
1778     return SLCT_NotALiteral;
1779   }
1780 
1781   case Stmt::CallExprClass:
1782   case Stmt::CXXMemberCallExprClass: {
1783     const CallExpr *CE = cast<CallExpr>(E);
1784     if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) {
1785       if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) {
1786         unsigned ArgIndex = FA->getFormatIdx();
1787         if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND))
1788           if (MD->isInstance())
1789             --ArgIndex;
1790         const Expr *Arg = CE->getArg(ArgIndex - 1);
1791 
1792         return checkFormatStringExpr(Arg, Args,
1793                                      HasVAListArg, format_idx, firstDataArg,
1794                                      Type, CallType, inFunctionCall);
1795       } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) {
1796         unsigned BuiltinID = FD->getBuiltinID();
1797         if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString ||
1798             BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) {
1799           const Expr *Arg = CE->getArg(0);
1800           return checkFormatStringExpr(Arg, Args,
1801                                        HasVAListArg, format_idx,
1802                                        firstDataArg, Type, CallType,
1803                                        inFunctionCall);
1804         }
1805       }
1806     }
1807 
1808     return SLCT_NotALiteral;
1809   }
1810   case Stmt::ObjCStringLiteralClass:
1811   case Stmt::StringLiteralClass: {
1812     const StringLiteral *StrE = NULL;
1813 
1814     if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1815       StrE = ObjCFExpr->getString();
1816     else
1817       StrE = cast<StringLiteral>(E);
1818 
1819     if (StrE) {
1820       CheckFormatString(StrE, E, Args, HasVAListArg, format_idx,
1821                         firstDataArg, Type, inFunctionCall, CallType);
1822       return SLCT_CheckedLiteral;
1823     }
1824 
1825     return SLCT_NotALiteral;
1826   }
1827 
1828   default:
1829     return SLCT_NotALiteral;
1830   }
1831 }
1832 
1833 void
1834 Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1835                             const Expr * const *ExprArgs,
1836                             SourceLocation CallSiteLoc) {
1837   for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1838                                   e = NonNull->args_end();
1839        i != e; ++i) {
1840     const Expr *ArgExpr = ExprArgs[*i];
1841 
1842     // As a special case, transparent unions initialized with zero are
1843     // considered null for the purposes of the nonnull attribute.
1844     if (const RecordType *UT = ArgExpr->getType()->getAsUnionType()) {
1845       if (UT->getDecl()->hasAttr<TransparentUnionAttr>())
1846         if (const CompoundLiteralExpr *CLE =
1847             dyn_cast<CompoundLiteralExpr>(ArgExpr))
1848           if (const InitListExpr *ILE =
1849               dyn_cast<InitListExpr>(CLE->getInitializer()))
1850             ArgExpr = ILE->getInit(0);
1851     }
1852 
1853     bool Result;
1854     if (ArgExpr->EvaluateAsBooleanCondition(Result, Context) && !Result)
1855       Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1856   }
1857 }
1858 
1859 Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) {
1860   return llvm::StringSwitch<FormatStringType>(Format->getType())
1861   .Case("scanf", FST_Scanf)
1862   .Cases("printf", "printf0", FST_Printf)
1863   .Cases("NSString", "CFString", FST_NSString)
1864   .Case("strftime", FST_Strftime)
1865   .Case("strfmon", FST_Strfmon)
1866   .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf)
1867   .Default(FST_Unknown);
1868 }
1869 
1870 /// CheckFormatArguments - Check calls to printf and scanf (and similar
1871 /// functions) for correct use of format strings.
1872 /// Returns true if a format string has been fully checked.
1873 bool Sema::CheckFormatArguments(const FormatAttr *Format,
1874                                 ArrayRef<const Expr *> Args,
1875                                 bool IsCXXMember,
1876                                 VariadicCallType CallType,
1877                                 SourceLocation Loc, SourceRange Range) {
1878   FormatStringInfo FSI;
1879   if (getFormatStringInfo(Format, IsCXXMember, &FSI))
1880     return CheckFormatArguments(Args, FSI.HasVAListArg, FSI.FormatIdx,
1881                                 FSI.FirstDataArg, GetFormatStringType(Format),
1882                                 CallType, Loc, Range);
1883   return false;
1884 }
1885 
1886 bool Sema::CheckFormatArguments(ArrayRef<const Expr *> Args,
1887                                 bool HasVAListArg, unsigned format_idx,
1888                                 unsigned firstDataArg, FormatStringType Type,
1889                                 VariadicCallType CallType,
1890                                 SourceLocation Loc, SourceRange Range) {
1891   // CHECK: printf/scanf-like function is called with no format string.
1892   if (format_idx >= Args.size()) {
1893     Diag(Loc, diag::warn_missing_format_string) << Range;
1894     return false;
1895   }
1896 
1897   const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts();
1898 
1899   // CHECK: format string is not a string literal.
1900   //
1901   // Dynamically generated format strings are difficult to
1902   // automatically vet at compile time.  Requiring that format strings
1903   // are string literals: (1) permits the checking of format strings by
1904   // the compiler and thereby (2) can practically remove the source of
1905   // many format string exploits.
1906 
1907   // Format string can be either ObjC string (e.g. @"%d") or
1908   // C string (e.g. "%d")
1909   // ObjC string uses the same format specifiers as C string, so we can use
1910   // the same format string checking logic for both ObjC and C strings.
1911   StringLiteralCheckType CT =
1912       checkFormatStringExpr(OrigFormatExpr, Args, HasVAListArg,
1913                             format_idx, firstDataArg, Type, CallType);
1914   if (CT != SLCT_NotALiteral)
1915     // Literal format string found, check done!
1916     return CT == SLCT_CheckedLiteral;
1917 
1918   // Strftime is particular as it always uses a single 'time' argument,
1919   // so it is safe to pass a non-literal string.
1920   if (Type == FST_Strftime)
1921     return false;
1922 
1923   // Do not emit diag when the string param is a macro expansion and the
1924   // format is either NSString or CFString. This is a hack to prevent
1925   // diag when using the NSLocalizedString and CFCopyLocalizedString macros
1926   // which are usually used in place of NS and CF string literals.
1927   if (Type == FST_NSString &&
1928       SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart()))
1929     return false;
1930 
1931   // If there are no arguments specified, warn with -Wformat-security, otherwise
1932   // warn only with -Wformat-nonliteral.
1933   if (Args.size() == format_idx+1)
1934     Diag(Args[format_idx]->getLocStart(),
1935          diag::warn_format_nonliteral_noargs)
1936       << OrigFormatExpr->getSourceRange();
1937   else
1938     Diag(Args[format_idx]->getLocStart(),
1939          diag::warn_format_nonliteral)
1940            << OrigFormatExpr->getSourceRange();
1941   return false;
1942 }
1943 
1944 namespace {
1945 class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1946 protected:
1947   Sema &S;
1948   const StringLiteral *FExpr;
1949   const Expr *OrigFormatExpr;
1950   const unsigned FirstDataArg;
1951   const unsigned NumDataArgs;
1952   const char *Beg; // Start of format string.
1953   const bool HasVAListArg;
1954   ArrayRef<const Expr *> Args;
1955   unsigned FormatIdx;
1956   llvm::BitVector CoveredArgs;
1957   bool usesPositionalArgs;
1958   bool atFirstArg;
1959   bool inFunctionCall;
1960   Sema::VariadicCallType CallType;
1961 public:
1962   CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1963                      const Expr *origFormatExpr, unsigned firstDataArg,
1964                      unsigned numDataArgs, const char *beg, bool hasVAListArg,
1965                      ArrayRef<const Expr *> Args,
1966                      unsigned formatIdx, bool inFunctionCall,
1967                      Sema::VariadicCallType callType)
1968     : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1969       FirstDataArg(firstDataArg), NumDataArgs(numDataArgs),
1970       Beg(beg), HasVAListArg(hasVAListArg),
1971       Args(Args), FormatIdx(formatIdx),
1972       usesPositionalArgs(false), atFirstArg(true),
1973       inFunctionCall(inFunctionCall), CallType(callType) {
1974         CoveredArgs.resize(numDataArgs);
1975         CoveredArgs.reset();
1976       }
1977 
1978   void DoneProcessing();
1979 
1980   void HandleIncompleteSpecifier(const char *startSpecifier,
1981                                  unsigned specifierLen);
1982 
1983   void HandleInvalidLengthModifier(
1984       const analyze_format_string::FormatSpecifier &FS,
1985       const analyze_format_string::ConversionSpecifier &CS,
1986       const char *startSpecifier, unsigned specifierLen, unsigned DiagID);
1987 
1988   void HandleNonStandardLengthModifier(
1989       const analyze_format_string::FormatSpecifier &FS,
1990       const char *startSpecifier, unsigned specifierLen);
1991 
1992   void HandleNonStandardConversionSpecifier(
1993       const analyze_format_string::ConversionSpecifier &CS,
1994       const char *startSpecifier, unsigned specifierLen);
1995 
1996   virtual void HandlePosition(const char *startPos, unsigned posLen);
1997 
1998   virtual void HandleInvalidPosition(const char *startSpecifier,
1999                                      unsigned specifierLen,
2000                                      analyze_format_string::PositionContext p);
2001 
2002   virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
2003 
2004   void HandleNullChar(const char *nullCharacter);
2005 
2006   template <typename Range>
2007   static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall,
2008                                    const Expr *ArgumentExpr,
2009                                    PartialDiagnostic PDiag,
2010                                    SourceLocation StringLoc,
2011                                    bool IsStringLocation, Range StringRange,
2012                             ArrayRef<FixItHint> Fixit = ArrayRef<FixItHint>());
2013 
2014 protected:
2015   bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
2016                                         const char *startSpec,
2017                                         unsigned specifierLen,
2018                                         const char *csStart, unsigned csLen);
2019 
2020   void HandlePositionalNonpositionalArgs(SourceLocation Loc,
2021                                          const char *startSpec,
2022                                          unsigned specifierLen);
2023 
2024   SourceRange getFormatStringRange();
2025   CharSourceRange getSpecifierRange(const char *startSpecifier,
2026                                     unsigned specifierLen);
2027   SourceLocation getLocationOfByte(const char *x);
2028 
2029   const Expr *getDataArg(unsigned i) const;
2030 
2031   bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
2032                     const analyze_format_string::ConversionSpecifier &CS,
2033                     const char *startSpecifier, unsigned specifierLen,
2034                     unsigned argIndex);
2035 
2036   template <typename Range>
2037   void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc,
2038                             bool IsStringLocation, Range StringRange,
2039                             ArrayRef<FixItHint> Fixit = ArrayRef<FixItHint>());
2040 
2041   void CheckPositionalAndNonpositionalArgs(
2042       const analyze_format_string::FormatSpecifier *FS);
2043 };
2044 }
2045 
2046 SourceRange CheckFormatHandler::getFormatStringRange() {
2047   return OrigFormatExpr->getSourceRange();
2048 }
2049 
2050 CharSourceRange CheckFormatHandler::
2051 getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
2052   SourceLocation Start = getLocationOfByte(startSpecifier);
2053   SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
2054 
2055   // Advance the end SourceLocation by one due to half-open ranges.
2056   End = End.getLocWithOffset(1);
2057 
2058   return CharSourceRange::getCharRange(Start, End);
2059 }
2060 
2061 SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
2062   return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
2063 }
2064 
2065 void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
2066                                                    unsigned specifierLen){
2067   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier),
2068                        getLocationOfByte(startSpecifier),
2069                        /*IsStringLocation*/true,
2070                        getSpecifierRange(startSpecifier, specifierLen));
2071 }
2072 
2073 void CheckFormatHandler::HandleInvalidLengthModifier(
2074     const analyze_format_string::FormatSpecifier &FS,
2075     const analyze_format_string::ConversionSpecifier &CS,
2076     const char *startSpecifier, unsigned specifierLen, unsigned DiagID) {
2077   using namespace analyze_format_string;
2078 
2079   const LengthModifier &LM = FS.getLengthModifier();
2080   CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength());
2081 
2082   // See if we know how to fix this length modifier.
2083   Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier();
2084   if (FixedLM) {
2085     EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(),
2086                          getLocationOfByte(LM.getStart()),
2087                          /*IsStringLocation*/true,
2088                          getSpecifierRange(startSpecifier, specifierLen));
2089 
2090     S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier)
2091       << FixedLM->toString()
2092       << FixItHint::CreateReplacement(LMRange, FixedLM->toString());
2093 
2094   } else {
2095     FixItHint Hint;
2096     if (DiagID == diag::warn_format_nonsensical_length)
2097       Hint = FixItHint::CreateRemoval(LMRange);
2098 
2099     EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(),
2100                          getLocationOfByte(LM.getStart()),
2101                          /*IsStringLocation*/true,
2102                          getSpecifierRange(startSpecifier, specifierLen),
2103                          Hint);
2104   }
2105 }
2106 
2107 void CheckFormatHandler::HandleNonStandardLengthModifier(
2108     const analyze_format_string::FormatSpecifier &FS,
2109     const char *startSpecifier, unsigned specifierLen) {
2110   using namespace analyze_format_string;
2111 
2112   const LengthModifier &LM = FS.getLengthModifier();
2113   CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength());
2114 
2115   // See if we know how to fix this length modifier.
2116   Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier();
2117   if (FixedLM) {
2118     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2119                            << LM.toString() << 0,
2120                          getLocationOfByte(LM.getStart()),
2121                          /*IsStringLocation*/true,
2122                          getSpecifierRange(startSpecifier, specifierLen));
2123 
2124     S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier)
2125       << FixedLM->toString()
2126       << FixItHint::CreateReplacement(LMRange, FixedLM->toString());
2127 
2128   } else {
2129     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2130                            << LM.toString() << 0,
2131                          getLocationOfByte(LM.getStart()),
2132                          /*IsStringLocation*/true,
2133                          getSpecifierRange(startSpecifier, specifierLen));
2134   }
2135 }
2136 
2137 void CheckFormatHandler::HandleNonStandardConversionSpecifier(
2138     const analyze_format_string::ConversionSpecifier &CS,
2139     const char *startSpecifier, unsigned specifierLen) {
2140   using namespace analyze_format_string;
2141 
2142   // See if we know how to fix this conversion specifier.
2143   Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier();
2144   if (FixedCS) {
2145     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2146                           << CS.toString() << /*conversion specifier*/1,
2147                          getLocationOfByte(CS.getStart()),
2148                          /*IsStringLocation*/true,
2149                          getSpecifierRange(startSpecifier, specifierLen));
2150 
2151     CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength());
2152     S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier)
2153       << FixedCS->toString()
2154       << FixItHint::CreateReplacement(CSRange, FixedCS->toString());
2155   } else {
2156     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2157                           << CS.toString() << /*conversion specifier*/1,
2158                          getLocationOfByte(CS.getStart()),
2159                          /*IsStringLocation*/true,
2160                          getSpecifierRange(startSpecifier, specifierLen));
2161   }
2162 }
2163 
2164 void CheckFormatHandler::HandlePosition(const char *startPos,
2165                                         unsigned posLen) {
2166   EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg),
2167                                getLocationOfByte(startPos),
2168                                /*IsStringLocation*/true,
2169                                getSpecifierRange(startPos, posLen));
2170 }
2171 
2172 void
2173 CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
2174                                      analyze_format_string::PositionContext p) {
2175   EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier)
2176                          << (unsigned) p,
2177                        getLocationOfByte(startPos), /*IsStringLocation*/true,
2178                        getSpecifierRange(startPos, posLen));
2179 }
2180 
2181 void CheckFormatHandler::HandleZeroPosition(const char *startPos,
2182                                             unsigned posLen) {
2183   EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier),
2184                                getLocationOfByte(startPos),
2185                                /*IsStringLocation*/true,
2186                                getSpecifierRange(startPos, posLen));
2187 }
2188 
2189 void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
2190   if (!isa<ObjCStringLiteral>(OrigFormatExpr)) {
2191     // The presence of a null character is likely an error.
2192     EmitFormatDiagnostic(
2193       S.PDiag(diag::warn_printf_format_string_contains_null_char),
2194       getLocationOfByte(nullCharacter), /*IsStringLocation*/true,
2195       getFormatStringRange());
2196   }
2197 }
2198 
2199 // Note that this may return NULL if there was an error parsing or building
2200 // one of the argument expressions.
2201 const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
2202   return Args[FirstDataArg + i];
2203 }
2204 
2205 void CheckFormatHandler::DoneProcessing() {
2206     // Does the number of data arguments exceed the number of
2207     // format conversions in the format string?
2208   if (!HasVAListArg) {
2209       // Find any arguments that weren't covered.
2210     CoveredArgs.flip();
2211     signed notCoveredArg = CoveredArgs.find_first();
2212     if (notCoveredArg >= 0) {
2213       assert((unsigned)notCoveredArg < NumDataArgs);
2214       if (const Expr *E = getDataArg((unsigned) notCoveredArg)) {
2215         SourceLocation Loc = E->getLocStart();
2216         if (!S.getSourceManager().isInSystemMacro(Loc)) {
2217           EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used),
2218                                Loc, /*IsStringLocation*/false,
2219                                getFormatStringRange());
2220         }
2221       }
2222     }
2223   }
2224 }
2225 
2226 bool
2227 CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
2228                                                      SourceLocation Loc,
2229                                                      const char *startSpec,
2230                                                      unsigned specifierLen,
2231                                                      const char *csStart,
2232                                                      unsigned csLen) {
2233 
2234   bool keepGoing = true;
2235   if (argIndex < NumDataArgs) {
2236     // Consider the argument coverered, even though the specifier doesn't
2237     // make sense.
2238     CoveredArgs.set(argIndex);
2239   }
2240   else {
2241     // If argIndex exceeds the number of data arguments we
2242     // don't issue a warning because that is just a cascade of warnings (and
2243     // they may have intended '%%' anyway). We don't want to continue processing
2244     // the format string after this point, however, as we will like just get
2245     // gibberish when trying to match arguments.
2246     keepGoing = false;
2247   }
2248 
2249   EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion)
2250                          << StringRef(csStart, csLen),
2251                        Loc, /*IsStringLocation*/true,
2252                        getSpecifierRange(startSpec, specifierLen));
2253 
2254   return keepGoing;
2255 }
2256 
2257 void
2258 CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc,
2259                                                       const char *startSpec,
2260                                                       unsigned specifierLen) {
2261   EmitFormatDiagnostic(
2262     S.PDiag(diag::warn_format_mix_positional_nonpositional_args),
2263     Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen));
2264 }
2265 
2266 bool
2267 CheckFormatHandler::CheckNumArgs(
2268   const analyze_format_string::FormatSpecifier &FS,
2269   const analyze_format_string::ConversionSpecifier &CS,
2270   const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
2271 
2272   if (argIndex >= NumDataArgs) {
2273     PartialDiagnostic PDiag = FS.usesPositionalArg()
2274       ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args)
2275            << (argIndex+1) << NumDataArgs)
2276       : S.PDiag(diag::warn_printf_insufficient_data_args);
2277     EmitFormatDiagnostic(
2278       PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true,
2279       getSpecifierRange(startSpecifier, specifierLen));
2280     return false;
2281   }
2282   return true;
2283 }
2284 
2285 template<typename Range>
2286 void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag,
2287                                               SourceLocation Loc,
2288                                               bool IsStringLocation,
2289                                               Range StringRange,
2290                                               ArrayRef<FixItHint> FixIt) {
2291   EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag,
2292                        Loc, IsStringLocation, StringRange, FixIt);
2293 }
2294 
2295 /// \brief If the format string is not within the funcion call, emit a note
2296 /// so that the function call and string are in diagnostic messages.
2297 ///
2298 /// \param InFunctionCall if true, the format string is within the function
2299 /// call and only one diagnostic message will be produced.  Otherwise, an
2300 /// extra note will be emitted pointing to location of the format string.
2301 ///
2302 /// \param ArgumentExpr the expression that is passed as the format string
2303 /// argument in the function call.  Used for getting locations when two
2304 /// diagnostics are emitted.
2305 ///
2306 /// \param PDiag the callee should already have provided any strings for the
2307 /// diagnostic message.  This function only adds locations and fixits
2308 /// to diagnostics.
2309 ///
2310 /// \param Loc primary location for diagnostic.  If two diagnostics are
2311 /// required, one will be at Loc and a new SourceLocation will be created for
2312 /// the other one.
2313 ///
2314 /// \param IsStringLocation if true, Loc points to the format string should be
2315 /// used for the note.  Otherwise, Loc points to the argument list and will
2316 /// be used with PDiag.
2317 ///
2318 /// \param StringRange some or all of the string to highlight.  This is
2319 /// templated so it can accept either a CharSourceRange or a SourceRange.
2320 ///
2321 /// \param FixIt optional fix it hint for the format string.
2322 template<typename Range>
2323 void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall,
2324                                               const Expr *ArgumentExpr,
2325                                               PartialDiagnostic PDiag,
2326                                               SourceLocation Loc,
2327                                               bool IsStringLocation,
2328                                               Range StringRange,
2329                                               ArrayRef<FixItHint> FixIt) {
2330   if (InFunctionCall) {
2331     const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag);
2332     D << StringRange;
2333     for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end();
2334          I != E; ++I) {
2335       D << *I;
2336     }
2337   } else {
2338     S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag)
2339       << ArgumentExpr->getSourceRange();
2340 
2341     const Sema::SemaDiagnosticBuilder &Note =
2342       S.Diag(IsStringLocation ? Loc : StringRange.getBegin(),
2343              diag::note_format_string_defined);
2344 
2345     Note << StringRange;
2346     for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end();
2347          I != E; ++I) {
2348       Note << *I;
2349     }
2350   }
2351 }
2352 
2353 //===--- CHECK: Printf format string checking ------------------------------===//
2354 
2355 namespace {
2356 class CheckPrintfHandler : public CheckFormatHandler {
2357   bool ObjCContext;
2358 public:
2359   CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
2360                      const Expr *origFormatExpr, unsigned firstDataArg,
2361                      unsigned numDataArgs, bool isObjC,
2362                      const char *beg, bool hasVAListArg,
2363                      ArrayRef<const Expr *> Args,
2364                      unsigned formatIdx, bool inFunctionCall,
2365                      Sema::VariadicCallType CallType)
2366   : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
2367                        numDataArgs, beg, hasVAListArg, Args,
2368                        formatIdx, inFunctionCall, CallType), ObjCContext(isObjC)
2369   {}
2370 
2371 
2372   bool HandleInvalidPrintfConversionSpecifier(
2373                                       const analyze_printf::PrintfSpecifier &FS,
2374                                       const char *startSpecifier,
2375                                       unsigned specifierLen);
2376 
2377   bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
2378                              const char *startSpecifier,
2379                              unsigned specifierLen);
2380   bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS,
2381                        const char *StartSpecifier,
2382                        unsigned SpecifierLen,
2383                        const Expr *E);
2384 
2385   bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
2386                     const char *startSpecifier, unsigned specifierLen);
2387   void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
2388                            const analyze_printf::OptionalAmount &Amt,
2389                            unsigned type,
2390                            const char *startSpecifier, unsigned specifierLen);
2391   void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
2392                   const analyze_printf::OptionalFlag &flag,
2393                   const char *startSpecifier, unsigned specifierLen);
2394   void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
2395                          const analyze_printf::OptionalFlag &ignoredFlag,
2396                          const analyze_printf::OptionalFlag &flag,
2397                          const char *startSpecifier, unsigned specifierLen);
2398   bool checkForCStrMembers(const analyze_printf::ArgType &AT,
2399                            const Expr *E, const CharSourceRange &CSR);
2400 
2401 };
2402 }
2403 
2404 bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
2405                                       const analyze_printf::PrintfSpecifier &FS,
2406                                       const char *startSpecifier,
2407                                       unsigned specifierLen) {
2408   const analyze_printf::PrintfConversionSpecifier &CS =
2409     FS.getConversionSpecifier();
2410 
2411   return HandleInvalidConversionSpecifier(FS.getArgIndex(),
2412                                           getLocationOfByte(CS.getStart()),
2413                                           startSpecifier, specifierLen,
2414                                           CS.getStart(), CS.getLength());
2415 }
2416 
2417 bool CheckPrintfHandler::HandleAmount(
2418                                const analyze_format_string::OptionalAmount &Amt,
2419                                unsigned k, const char *startSpecifier,
2420                                unsigned specifierLen) {
2421 
2422   if (Amt.hasDataArgument()) {
2423     if (!HasVAListArg) {
2424       unsigned argIndex = Amt.getArgIndex();
2425       if (argIndex >= NumDataArgs) {
2426         EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg)
2427                                << k,
2428                              getLocationOfByte(Amt.getStart()),
2429                              /*IsStringLocation*/true,
2430                              getSpecifierRange(startSpecifier, specifierLen));
2431         // Don't do any more checking.  We will just emit
2432         // spurious errors.
2433         return false;
2434       }
2435 
2436       // Type check the data argument.  It should be an 'int'.
2437       // Although not in conformance with C99, we also allow the argument to be
2438       // an 'unsigned int' as that is a reasonably safe case.  GCC also
2439       // doesn't emit a warning for that case.
2440       CoveredArgs.set(argIndex);
2441       const Expr *Arg = getDataArg(argIndex);
2442       if (!Arg)
2443         return false;
2444 
2445       QualType T = Arg->getType();
2446 
2447       const analyze_printf::ArgType &AT = Amt.getArgType(S.Context);
2448       assert(AT.isValid());
2449 
2450       if (!AT.matchesType(S.Context, T)) {
2451         EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type)
2452                                << k << AT.getRepresentativeTypeName(S.Context)
2453                                << T << Arg->getSourceRange(),
2454                              getLocationOfByte(Amt.getStart()),
2455                              /*IsStringLocation*/true,
2456                              getSpecifierRange(startSpecifier, specifierLen));
2457         // Don't do any more checking.  We will just emit
2458         // spurious errors.
2459         return false;
2460       }
2461     }
2462   }
2463   return true;
2464 }
2465 
2466 void CheckPrintfHandler::HandleInvalidAmount(
2467                                       const analyze_printf::PrintfSpecifier &FS,
2468                                       const analyze_printf::OptionalAmount &Amt,
2469                                       unsigned type,
2470                                       const char *startSpecifier,
2471                                       unsigned specifierLen) {
2472   const analyze_printf::PrintfConversionSpecifier &CS =
2473     FS.getConversionSpecifier();
2474 
2475   FixItHint fixit =
2476     Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant
2477       ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
2478                                  Amt.getConstantLength()))
2479       : FixItHint();
2480 
2481   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount)
2482                          << type << CS.toString(),
2483                        getLocationOfByte(Amt.getStart()),
2484                        /*IsStringLocation*/true,
2485                        getSpecifierRange(startSpecifier, specifierLen),
2486                        fixit);
2487 }
2488 
2489 void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
2490                                     const analyze_printf::OptionalFlag &flag,
2491                                     const char *startSpecifier,
2492                                     unsigned specifierLen) {
2493   // Warn about pointless flag with a fixit removal.
2494   const analyze_printf::PrintfConversionSpecifier &CS =
2495     FS.getConversionSpecifier();
2496   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag)
2497                          << flag.toString() << CS.toString(),
2498                        getLocationOfByte(flag.getPosition()),
2499                        /*IsStringLocation*/true,
2500                        getSpecifierRange(startSpecifier, specifierLen),
2501                        FixItHint::CreateRemoval(
2502                          getSpecifierRange(flag.getPosition(), 1)));
2503 }
2504 
2505 void CheckPrintfHandler::HandleIgnoredFlag(
2506                                 const analyze_printf::PrintfSpecifier &FS,
2507                                 const analyze_printf::OptionalFlag &ignoredFlag,
2508                                 const analyze_printf::OptionalFlag &flag,
2509                                 const char *startSpecifier,
2510                                 unsigned specifierLen) {
2511   // Warn about ignored flag with a fixit removal.
2512   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag)
2513                          << ignoredFlag.toString() << flag.toString(),
2514                        getLocationOfByte(ignoredFlag.getPosition()),
2515                        /*IsStringLocation*/true,
2516                        getSpecifierRange(startSpecifier, specifierLen),
2517                        FixItHint::CreateRemoval(
2518                          getSpecifierRange(ignoredFlag.getPosition(), 1)));
2519 }
2520 
2521 // Determines if the specified is a C++ class or struct containing
2522 // a member with the specified name and kind (e.g. a CXXMethodDecl named
2523 // "c_str()").
2524 template<typename MemberKind>
2525 static llvm::SmallPtrSet<MemberKind*, 1>
2526 CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) {
2527   const RecordType *RT = Ty->getAs<RecordType>();
2528   llvm::SmallPtrSet<MemberKind*, 1> Results;
2529 
2530   if (!RT)
2531     return Results;
2532   const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl());
2533   if (!RD)
2534     return Results;
2535 
2536   LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(),
2537                  Sema::LookupMemberName);
2538 
2539   // We just need to include all members of the right kind turned up by the
2540   // filter, at this point.
2541   if (S.LookupQualifiedName(R, RT->getDecl()))
2542     for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) {
2543       NamedDecl *decl = (*I)->getUnderlyingDecl();
2544       if (MemberKind *FK = dyn_cast<MemberKind>(decl))
2545         Results.insert(FK);
2546     }
2547   return Results;
2548 }
2549 
2550 // Check if a (w)string was passed when a (w)char* was needed, and offer a
2551 // better diagnostic if so. AT is assumed to be valid.
2552 // Returns true when a c_str() conversion method is found.
2553 bool CheckPrintfHandler::checkForCStrMembers(
2554     const analyze_printf::ArgType &AT, const Expr *E,
2555     const CharSourceRange &CSR) {
2556   typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet;
2557 
2558   MethodSet Results =
2559       CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType());
2560 
2561   for (MethodSet::iterator MI = Results.begin(), ME = Results.end();
2562        MI != ME; ++MI) {
2563     const CXXMethodDecl *Method = *MI;
2564     if (Method->getNumParams() == 0 &&
2565           AT.matchesType(S.Context, Method->getResultType())) {
2566       // FIXME: Suggest parens if the expression needs them.
2567       SourceLocation EndLoc =
2568           S.getPreprocessor().getLocForEndOfToken(E->getLocEnd());
2569       S.Diag(E->getLocStart(), diag::note_printf_c_str)
2570           << "c_str()"
2571           << FixItHint::CreateInsertion(EndLoc, ".c_str()");
2572       return true;
2573     }
2574   }
2575 
2576   return false;
2577 }
2578 
2579 bool
2580 CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
2581                                             &FS,
2582                                           const char *startSpecifier,
2583                                           unsigned specifierLen) {
2584 
2585   using namespace analyze_format_string;
2586   using namespace analyze_printf;
2587   const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
2588 
2589   if (FS.consumesDataArgument()) {
2590     if (atFirstArg) {
2591         atFirstArg = false;
2592         usesPositionalArgs = FS.usesPositionalArg();
2593     }
2594     else if (usesPositionalArgs != FS.usesPositionalArg()) {
2595       HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()),
2596                                         startSpecifier, specifierLen);
2597       return false;
2598     }
2599   }
2600 
2601   // First check if the field width, precision, and conversion specifier
2602   // have matching data arguments.
2603   if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
2604                     startSpecifier, specifierLen)) {
2605     return false;
2606   }
2607 
2608   if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
2609                     startSpecifier, specifierLen)) {
2610     return false;
2611   }
2612 
2613   if (!CS.consumesDataArgument()) {
2614     // FIXME: Technically specifying a precision or field width here
2615     // makes no sense.  Worth issuing a warning at some point.
2616     return true;
2617   }
2618 
2619   // Consume the argument.
2620   unsigned argIndex = FS.getArgIndex();
2621   if (argIndex < NumDataArgs) {
2622     // The check to see if the argIndex is valid will come later.
2623     // We set the bit here because we may exit early from this
2624     // function if we encounter some other error.
2625     CoveredArgs.set(argIndex);
2626   }
2627 
2628   // Check for using an Objective-C specific conversion specifier
2629   // in a non-ObjC literal.
2630   if (!ObjCContext && CS.isObjCArg()) {
2631     return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
2632                                                   specifierLen);
2633   }
2634 
2635   // Check for invalid use of field width
2636   if (!FS.hasValidFieldWidth()) {
2637     HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
2638         startSpecifier, specifierLen);
2639   }
2640 
2641   // Check for invalid use of precision
2642   if (!FS.hasValidPrecision()) {
2643     HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
2644         startSpecifier, specifierLen);
2645   }
2646 
2647   // Check each flag does not conflict with any other component.
2648   if (!FS.hasValidThousandsGroupingPrefix())
2649     HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
2650   if (!FS.hasValidLeadingZeros())
2651     HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
2652   if (!FS.hasValidPlusPrefix())
2653     HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
2654   if (!FS.hasValidSpacePrefix())
2655     HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
2656   if (!FS.hasValidAlternativeForm())
2657     HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
2658   if (!FS.hasValidLeftJustified())
2659     HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
2660 
2661   // Check that flags are not ignored by another flag
2662   if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
2663     HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
2664         startSpecifier, specifierLen);
2665   if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
2666     HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
2667             startSpecifier, specifierLen);
2668 
2669   // Check the length modifier is valid with the given conversion specifier.
2670   if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo()))
2671     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
2672                                 diag::warn_format_nonsensical_length);
2673   else if (!FS.hasStandardLengthModifier())
2674     HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen);
2675   else if (!FS.hasStandardLengthConversionCombination())
2676     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
2677                                 diag::warn_format_non_standard_conversion_spec);
2678 
2679   if (!FS.hasStandardConversionSpecifier(S.getLangOpts()))
2680     HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen);
2681 
2682   // The remaining checks depend on the data arguments.
2683   if (HasVAListArg)
2684     return true;
2685 
2686   if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
2687     return false;
2688 
2689   const Expr *Arg = getDataArg(argIndex);
2690   if (!Arg)
2691     return true;
2692 
2693   return checkFormatExpr(FS, startSpecifier, specifierLen, Arg);
2694 }
2695 
2696 static bool requiresParensToAddCast(const Expr *E) {
2697   // FIXME: We should have a general way to reason about operator
2698   // precedence and whether parens are actually needed here.
2699   // Take care of a few common cases where they aren't.
2700   const Expr *Inside = E->IgnoreImpCasts();
2701   if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside))
2702     Inside = POE->getSyntacticForm()->IgnoreImpCasts();
2703 
2704   switch (Inside->getStmtClass()) {
2705   case Stmt::ArraySubscriptExprClass:
2706   case Stmt::CallExprClass:
2707   case Stmt::CharacterLiteralClass:
2708   case Stmt::CXXBoolLiteralExprClass:
2709   case Stmt::DeclRefExprClass:
2710   case Stmt::FloatingLiteralClass:
2711   case Stmt::IntegerLiteralClass:
2712   case Stmt::MemberExprClass:
2713   case Stmt::ObjCArrayLiteralClass:
2714   case Stmt::ObjCBoolLiteralExprClass:
2715   case Stmt::ObjCBoxedExprClass:
2716   case Stmt::ObjCDictionaryLiteralClass:
2717   case Stmt::ObjCEncodeExprClass:
2718   case Stmt::ObjCIvarRefExprClass:
2719   case Stmt::ObjCMessageExprClass:
2720   case Stmt::ObjCPropertyRefExprClass:
2721   case Stmt::ObjCStringLiteralClass:
2722   case Stmt::ObjCSubscriptRefExprClass:
2723   case Stmt::ParenExprClass:
2724   case Stmt::StringLiteralClass:
2725   case Stmt::UnaryOperatorClass:
2726     return false;
2727   default:
2728     return true;
2729   }
2730 }
2731 
2732 bool
2733 CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS,
2734                                     const char *StartSpecifier,
2735                                     unsigned SpecifierLen,
2736                                     const Expr *E) {
2737   using namespace analyze_format_string;
2738   using namespace analyze_printf;
2739   // Now type check the data expression that matches the
2740   // format specifier.
2741   const analyze_printf::ArgType &AT = FS.getArgType(S.Context,
2742                                                     ObjCContext);
2743   if (!AT.isValid())
2744     return true;
2745 
2746   QualType ExprTy = E->getType();
2747   while (const TypeOfExprType *TET = dyn_cast<TypeOfExprType>(ExprTy)) {
2748     ExprTy = TET->getUnderlyingExpr()->getType();
2749   }
2750 
2751   if (AT.matchesType(S.Context, ExprTy))
2752     return true;
2753 
2754   // Look through argument promotions for our error message's reported type.
2755   // This includes the integral and floating promotions, but excludes array
2756   // and function pointer decay; seeing that an argument intended to be a
2757   // string has type 'char [6]' is probably more confusing than 'char *'.
2758   if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
2759     if (ICE->getCastKind() == CK_IntegralCast ||
2760         ICE->getCastKind() == CK_FloatingCast) {
2761       E = ICE->getSubExpr();
2762       ExprTy = E->getType();
2763 
2764       // Check if we didn't match because of an implicit cast from a 'char'
2765       // or 'short' to an 'int'.  This is done because printf is a varargs
2766       // function.
2767       if (ICE->getType() == S.Context.IntTy ||
2768           ICE->getType() == S.Context.UnsignedIntTy) {
2769         // All further checking is done on the subexpression.
2770         if (AT.matchesType(S.Context, ExprTy))
2771           return true;
2772       }
2773     }
2774   } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) {
2775     // Special case for 'a', which has type 'int' in C.
2776     // Note, however, that we do /not/ want to treat multibyte constants like
2777     // 'MooV' as characters! This form is deprecated but still exists.
2778     if (ExprTy == S.Context.IntTy)
2779       if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue()))
2780         ExprTy = S.Context.CharTy;
2781   }
2782 
2783   // %C in an Objective-C context prints a unichar, not a wchar_t.
2784   // If the argument is an integer of some kind, believe the %C and suggest
2785   // a cast instead of changing the conversion specifier.
2786   QualType IntendedTy = ExprTy;
2787   if (ObjCContext &&
2788       FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) {
2789     if (ExprTy->isIntegralOrUnscopedEnumerationType() &&
2790         !ExprTy->isCharType()) {
2791       // 'unichar' is defined as a typedef of unsigned short, but we should
2792       // prefer using the typedef if it is visible.
2793       IntendedTy = S.Context.UnsignedShortTy;
2794 
2795       LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(),
2796                           Sema::LookupOrdinaryName);
2797       if (S.LookupName(Result, S.getCurScope())) {
2798         NamedDecl *ND = Result.getFoundDecl();
2799         if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND))
2800           if (TD->getUnderlyingType() == IntendedTy)
2801             IntendedTy = S.Context.getTypedefType(TD);
2802       }
2803     }
2804   }
2805 
2806   // Special-case some of Darwin's platform-independence types by suggesting
2807   // casts to primitive types that are known to be large enough.
2808   bool ShouldNotPrintDirectly = false;
2809   if (S.Context.getTargetInfo().getTriple().isOSDarwin()) {
2810     // Use a 'while' to peel off layers of typedefs.
2811     QualType TyTy = IntendedTy;
2812     while (const TypedefType *UserTy = TyTy->getAs<TypedefType>()) {
2813       StringRef Name = UserTy->getDecl()->getName();
2814       QualType CastTy = llvm::StringSwitch<QualType>(Name)
2815         .Case("NSInteger", S.Context.LongTy)
2816         .Case("NSUInteger", S.Context.UnsignedLongTy)
2817         .Case("SInt32", S.Context.IntTy)
2818         .Case("UInt32", S.Context.UnsignedIntTy)
2819         .Default(QualType());
2820 
2821       if (!CastTy.isNull()) {
2822         ShouldNotPrintDirectly = true;
2823         IntendedTy = CastTy;
2824         break;
2825       }
2826       TyTy = UserTy->desugar();
2827     }
2828   }
2829 
2830   // We may be able to offer a FixItHint if it is a supported type.
2831   PrintfSpecifier fixedFS = FS;
2832   bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(),
2833                                  S.Context, ObjCContext);
2834 
2835   if (success) {
2836     // Get the fix string from the fixed format specifier
2837     SmallString<16> buf;
2838     llvm::raw_svector_ostream os(buf);
2839     fixedFS.toString(os);
2840 
2841     CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen);
2842 
2843     if (IntendedTy == ExprTy) {
2844       // In this case, the specifier is wrong and should be changed to match
2845       // the argument.
2846       EmitFormatDiagnostic(
2847         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2848           << AT.getRepresentativeTypeName(S.Context) << IntendedTy
2849           << E->getSourceRange(),
2850         E->getLocStart(),
2851         /*IsStringLocation*/false,
2852         SpecRange,
2853         FixItHint::CreateReplacement(SpecRange, os.str()));
2854 
2855     } else {
2856       // The canonical type for formatting this value is different from the
2857       // actual type of the expression. (This occurs, for example, with Darwin's
2858       // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but
2859       // should be printed as 'long' for 64-bit compatibility.)
2860       // Rather than emitting a normal format/argument mismatch, we want to
2861       // add a cast to the recommended type (and correct the format string
2862       // if necessary).
2863       SmallString<16> CastBuf;
2864       llvm::raw_svector_ostream CastFix(CastBuf);
2865       CastFix << "(";
2866       IntendedTy.print(CastFix, S.Context.getPrintingPolicy());
2867       CastFix << ")";
2868 
2869       SmallVector<FixItHint,4> Hints;
2870       if (!AT.matchesType(S.Context, IntendedTy))
2871         Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str()));
2872 
2873       if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) {
2874         // If there's already a cast present, just replace it.
2875         SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc());
2876         Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str()));
2877 
2878       } else if (!requiresParensToAddCast(E)) {
2879         // If the expression has high enough precedence,
2880         // just write the C-style cast.
2881         Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(),
2882                                                    CastFix.str()));
2883       } else {
2884         // Otherwise, add parens around the expression as well as the cast.
2885         CastFix << "(";
2886         Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(),
2887                                                    CastFix.str()));
2888 
2889         SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd());
2890         Hints.push_back(FixItHint::CreateInsertion(After, ")"));
2891       }
2892 
2893       if (ShouldNotPrintDirectly) {
2894         // The expression has a type that should not be printed directly.
2895         // We extract the name from the typedef because we don't want to show
2896         // the underlying type in the diagnostic.
2897         StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName();
2898 
2899         EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast)
2900                                << Name << IntendedTy
2901                                << E->getSourceRange(),
2902                              E->getLocStart(), /*IsStringLocation=*/false,
2903                              SpecRange, Hints);
2904       } else {
2905         // In this case, the expression could be printed using a different
2906         // specifier, but we've decided that the specifier is probably correct
2907         // and we should cast instead. Just use the normal warning message.
2908         EmitFormatDiagnostic(
2909           S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2910             << AT.getRepresentativeTypeName(S.Context) << ExprTy
2911             << E->getSourceRange(),
2912           E->getLocStart(), /*IsStringLocation*/false,
2913           SpecRange, Hints);
2914       }
2915     }
2916   } else {
2917     const CharSourceRange &CSR = getSpecifierRange(StartSpecifier,
2918                                                    SpecifierLen);
2919     // Since the warning for passing non-POD types to variadic functions
2920     // was deferred until now, we emit a warning for non-POD
2921     // arguments here.
2922     if (S.isValidVarArgType(ExprTy) == Sema::VAK_Invalid) {
2923       unsigned DiagKind;
2924       if (ExprTy->isObjCObjectType())
2925         DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format;
2926       else
2927         DiagKind = diag::warn_non_pod_vararg_with_format_string;
2928 
2929       EmitFormatDiagnostic(
2930         S.PDiag(DiagKind)
2931           << S.getLangOpts().CPlusPlus11
2932           << ExprTy
2933           << CallType
2934           << AT.getRepresentativeTypeName(S.Context)
2935           << CSR
2936           << E->getSourceRange(),
2937         E->getLocStart(), /*IsStringLocation*/false, CSR);
2938 
2939       checkForCStrMembers(AT, E, CSR);
2940     } else
2941       EmitFormatDiagnostic(
2942         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2943           << AT.getRepresentativeTypeName(S.Context) << ExprTy
2944           << CSR
2945           << E->getSourceRange(),
2946         E->getLocStart(), /*IsStringLocation*/false, CSR);
2947   }
2948 
2949   return true;
2950 }
2951 
2952 //===--- CHECK: Scanf format string checking ------------------------------===//
2953 
2954 namespace {
2955 class CheckScanfHandler : public CheckFormatHandler {
2956 public:
2957   CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
2958                     const Expr *origFormatExpr, unsigned firstDataArg,
2959                     unsigned numDataArgs, const char *beg, bool hasVAListArg,
2960                     ArrayRef<const Expr *> Args,
2961                     unsigned formatIdx, bool inFunctionCall,
2962                     Sema::VariadicCallType CallType)
2963   : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
2964                        numDataArgs, beg, hasVAListArg,
2965                        Args, formatIdx, inFunctionCall, CallType)
2966   {}
2967 
2968   bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
2969                             const char *startSpecifier,
2970                             unsigned specifierLen);
2971 
2972   bool HandleInvalidScanfConversionSpecifier(
2973           const analyze_scanf::ScanfSpecifier &FS,
2974           const char *startSpecifier,
2975           unsigned specifierLen);
2976 
2977   void HandleIncompleteScanList(const char *start, const char *end);
2978 };
2979 }
2980 
2981 void CheckScanfHandler::HandleIncompleteScanList(const char *start,
2982                                                  const char *end) {
2983   EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete),
2984                        getLocationOfByte(end), /*IsStringLocation*/true,
2985                        getSpecifierRange(start, end - start));
2986 }
2987 
2988 bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
2989                                         const analyze_scanf::ScanfSpecifier &FS,
2990                                         const char *startSpecifier,
2991                                         unsigned specifierLen) {
2992 
2993   const analyze_scanf::ScanfConversionSpecifier &CS =
2994     FS.getConversionSpecifier();
2995 
2996   return HandleInvalidConversionSpecifier(FS.getArgIndex(),
2997                                           getLocationOfByte(CS.getStart()),
2998                                           startSpecifier, specifierLen,
2999                                           CS.getStart(), CS.getLength());
3000 }
3001 
3002 bool CheckScanfHandler::HandleScanfSpecifier(
3003                                        const analyze_scanf::ScanfSpecifier &FS,
3004                                        const char *startSpecifier,
3005                                        unsigned specifierLen) {
3006 
3007   using namespace analyze_scanf;
3008   using namespace analyze_format_string;
3009 
3010   const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
3011 
3012   // Handle case where '%' and '*' don't consume an argument.  These shouldn't
3013   // be used to decide if we are using positional arguments consistently.
3014   if (FS.consumesDataArgument()) {
3015     if (atFirstArg) {
3016       atFirstArg = false;
3017       usesPositionalArgs = FS.usesPositionalArg();
3018     }
3019     else if (usesPositionalArgs != FS.usesPositionalArg()) {
3020       HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()),
3021                                         startSpecifier, specifierLen);
3022       return false;
3023     }
3024   }
3025 
3026   // Check if the field with is non-zero.
3027   const OptionalAmount &Amt = FS.getFieldWidth();
3028   if (Amt.getHowSpecified() == OptionalAmount::Constant) {
3029     if (Amt.getConstantAmount() == 0) {
3030       const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
3031                                                    Amt.getConstantLength());
3032       EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width),
3033                            getLocationOfByte(Amt.getStart()),
3034                            /*IsStringLocation*/true, R,
3035                            FixItHint::CreateRemoval(R));
3036     }
3037   }
3038 
3039   if (!FS.consumesDataArgument()) {
3040     // FIXME: Technically specifying a precision or field width here
3041     // makes no sense.  Worth issuing a warning at some point.
3042     return true;
3043   }
3044 
3045   // Consume the argument.
3046   unsigned argIndex = FS.getArgIndex();
3047   if (argIndex < NumDataArgs) {
3048       // The check to see if the argIndex is valid will come later.
3049       // We set the bit here because we may exit early from this
3050       // function if we encounter some other error.
3051     CoveredArgs.set(argIndex);
3052   }
3053 
3054   // Check the length modifier is valid with the given conversion specifier.
3055   if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo()))
3056     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
3057                                 diag::warn_format_nonsensical_length);
3058   else if (!FS.hasStandardLengthModifier())
3059     HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen);
3060   else if (!FS.hasStandardLengthConversionCombination())
3061     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
3062                                 diag::warn_format_non_standard_conversion_spec);
3063 
3064   if (!FS.hasStandardConversionSpecifier(S.getLangOpts()))
3065     HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen);
3066 
3067   // The remaining checks depend on the data arguments.
3068   if (HasVAListArg)
3069     return true;
3070 
3071   if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
3072     return false;
3073 
3074   // Check that the argument type matches the format specifier.
3075   const Expr *Ex = getDataArg(argIndex);
3076   if (!Ex)
3077     return true;
3078 
3079   const analyze_format_string::ArgType &AT = FS.getArgType(S.Context);
3080   if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) {
3081     ScanfSpecifier fixedFS = FS;
3082     bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(),
3083                                    S.Context);
3084 
3085     if (success) {
3086       // Get the fix string from the fixed format specifier.
3087       SmallString<128> buf;
3088       llvm::raw_svector_ostream os(buf);
3089       fixedFS.toString(os);
3090 
3091       EmitFormatDiagnostic(
3092         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
3093           << AT.getRepresentativeTypeName(S.Context) << Ex->getType()
3094           << Ex->getSourceRange(),
3095         Ex->getLocStart(),
3096         /*IsStringLocation*/false,
3097         getSpecifierRange(startSpecifier, specifierLen),
3098         FixItHint::CreateReplacement(
3099           getSpecifierRange(startSpecifier, specifierLen),
3100           os.str()));
3101     } else {
3102       EmitFormatDiagnostic(
3103         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
3104           << AT.getRepresentativeTypeName(S.Context) << Ex->getType()
3105           << Ex->getSourceRange(),
3106         Ex->getLocStart(),
3107         /*IsStringLocation*/false,
3108         getSpecifierRange(startSpecifier, specifierLen));
3109     }
3110   }
3111 
3112   return true;
3113 }
3114 
3115 void Sema::CheckFormatString(const StringLiteral *FExpr,
3116                              const Expr *OrigFormatExpr,
3117                              ArrayRef<const Expr *> Args,
3118                              bool HasVAListArg, unsigned format_idx,
3119                              unsigned firstDataArg, FormatStringType Type,
3120                              bool inFunctionCall, VariadicCallType CallType) {
3121 
3122   // CHECK: is the format string a wide literal?
3123   if (!FExpr->isAscii() && !FExpr->isUTF8()) {
3124     CheckFormatHandler::EmitFormatDiagnostic(
3125       *this, inFunctionCall, Args[format_idx],
3126       PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(),
3127       /*IsStringLocation*/true, OrigFormatExpr->getSourceRange());
3128     return;
3129   }
3130 
3131   // Str - The format string.  NOTE: this is NOT null-terminated!
3132   StringRef StrRef = FExpr->getString();
3133   const char *Str = StrRef.data();
3134   unsigned StrLen = StrRef.size();
3135   const unsigned numDataArgs = Args.size() - firstDataArg;
3136 
3137   // CHECK: empty format string?
3138   if (StrLen == 0 && numDataArgs > 0) {
3139     CheckFormatHandler::EmitFormatDiagnostic(
3140       *this, inFunctionCall, Args[format_idx],
3141       PDiag(diag::warn_empty_format_string), FExpr->getLocStart(),
3142       /*IsStringLocation*/true, OrigFormatExpr->getSourceRange());
3143     return;
3144   }
3145 
3146   if (Type == FST_Printf || Type == FST_NSString) {
3147     CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
3148                          numDataArgs, (Type == FST_NSString),
3149                          Str, HasVAListArg, Args, format_idx,
3150                          inFunctionCall, CallType);
3151 
3152     if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen,
3153                                                   getLangOpts(),
3154                                                   Context.getTargetInfo()))
3155       H.DoneProcessing();
3156   } else if (Type == FST_Scanf) {
3157     CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs,
3158                         Str, HasVAListArg, Args, format_idx,
3159                         inFunctionCall, CallType);
3160 
3161     if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen,
3162                                                  getLangOpts(),
3163                                                  Context.getTargetInfo()))
3164       H.DoneProcessing();
3165   } // TODO: handle other formats
3166 }
3167 
3168 //===--- CHECK: Standard memory functions ---------------------------------===//
3169 
3170 /// \brief Determine whether the given type is a dynamic class type (e.g.,
3171 /// whether it has a vtable).
3172 static bool isDynamicClassType(QualType T) {
3173   if (CXXRecordDecl *Record = T->getAsCXXRecordDecl())
3174     if (CXXRecordDecl *Definition = Record->getDefinition())
3175       if (Definition->isDynamicClass())
3176         return true;
3177 
3178   return false;
3179 }
3180 
3181 /// \brief If E is a sizeof expression, returns its argument expression,
3182 /// otherwise returns NULL.
3183 static const Expr *getSizeOfExprArg(const Expr* E) {
3184   if (const UnaryExprOrTypeTraitExpr *SizeOf =
3185       dyn_cast<UnaryExprOrTypeTraitExpr>(E))
3186     if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType())
3187       return SizeOf->getArgumentExpr()->IgnoreParenImpCasts();
3188 
3189   return 0;
3190 }
3191 
3192 /// \brief If E is a sizeof expression, returns its argument type.
3193 static QualType getSizeOfArgType(const Expr* E) {
3194   if (const UnaryExprOrTypeTraitExpr *SizeOf =
3195       dyn_cast<UnaryExprOrTypeTraitExpr>(E))
3196     if (SizeOf->getKind() == clang::UETT_SizeOf)
3197       return SizeOf->getTypeOfArgument();
3198 
3199   return QualType();
3200 }
3201 
3202 /// \brief Check for dangerous or invalid arguments to memset().
3203 ///
3204 /// This issues warnings on known problematic, dangerous or unspecified
3205 /// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp'
3206 /// function calls.
3207 ///
3208 /// \param Call The call expression to diagnose.
3209 void Sema::CheckMemaccessArguments(const CallExpr *Call,
3210                                    unsigned BId,
3211                                    IdentifierInfo *FnName) {
3212   assert(BId != 0);
3213 
3214   // It is possible to have a non-standard definition of memset.  Validate
3215   // we have enough arguments, and if not, abort further checking.
3216   unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3);
3217   if (Call->getNumArgs() < ExpectedNumArgs)
3218     return;
3219 
3220   unsigned LastArg = (BId == Builtin::BImemset ||
3221                       BId == Builtin::BIstrndup ? 1 : 2);
3222   unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2);
3223   const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts();
3224 
3225   // We have special checking when the length is a sizeof expression.
3226   QualType SizeOfArgTy = getSizeOfArgType(LenExpr);
3227   const Expr *SizeOfArg = getSizeOfExprArg(LenExpr);
3228   llvm::FoldingSetNodeID SizeOfArgID;
3229 
3230   for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) {
3231     const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts();
3232     SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange();
3233 
3234     QualType DestTy = Dest->getType();
3235     if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
3236       QualType PointeeTy = DestPtrTy->getPointeeType();
3237 
3238       // Never warn about void type pointers. This can be used to suppress
3239       // false positives.
3240       if (PointeeTy->isVoidType())
3241         continue;
3242 
3243       // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by
3244       // actually comparing the expressions for equality. Because computing the
3245       // expression IDs can be expensive, we only do this if the diagnostic is
3246       // enabled.
3247       if (SizeOfArg &&
3248           Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess,
3249                                    SizeOfArg->getExprLoc())) {
3250         // We only compute IDs for expressions if the warning is enabled, and
3251         // cache the sizeof arg's ID.
3252         if (SizeOfArgID == llvm::FoldingSetNodeID())
3253           SizeOfArg->Profile(SizeOfArgID, Context, true);
3254         llvm::FoldingSetNodeID DestID;
3255         Dest->Profile(DestID, Context, true);
3256         if (DestID == SizeOfArgID) {
3257           // TODO: For strncpy() and friends, this could suggest sizeof(dst)
3258           //       over sizeof(src) as well.
3259           unsigned ActionIdx = 0; // Default is to suggest dereferencing.
3260           StringRef ReadableName = FnName->getName();
3261 
3262           if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest))
3263             if (UnaryOp->getOpcode() == UO_AddrOf)
3264               ActionIdx = 1; // If its an address-of operator, just remove it.
3265           if (!PointeeTy->isIncompleteType() &&
3266               (Context.getTypeSize(PointeeTy) == Context.getCharWidth()))
3267             ActionIdx = 2; // If the pointee's size is sizeof(char),
3268                            // suggest an explicit length.
3269 
3270           // If the function is defined as a builtin macro, do not show macro
3271           // expansion.
3272           SourceLocation SL = SizeOfArg->getExprLoc();
3273           SourceRange DSR = Dest->getSourceRange();
3274           SourceRange SSR = SizeOfArg->getSourceRange();
3275           SourceManager &SM  = PP.getSourceManager();
3276 
3277           if (SM.isMacroArgExpansion(SL)) {
3278             ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts);
3279             SL = SM.getSpellingLoc(SL);
3280             DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()),
3281                              SM.getSpellingLoc(DSR.getEnd()));
3282             SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()),
3283                              SM.getSpellingLoc(SSR.getEnd()));
3284           }
3285 
3286           DiagRuntimeBehavior(SL, SizeOfArg,
3287                               PDiag(diag::warn_sizeof_pointer_expr_memaccess)
3288                                 << ReadableName
3289                                 << PointeeTy
3290                                 << DestTy
3291                                 << DSR
3292                                 << SSR);
3293           DiagRuntimeBehavior(SL, SizeOfArg,
3294                          PDiag(diag::warn_sizeof_pointer_expr_memaccess_note)
3295                                 << ActionIdx
3296                                 << SSR);
3297 
3298           break;
3299         }
3300       }
3301 
3302       // Also check for cases where the sizeof argument is the exact same
3303       // type as the memory argument, and where it points to a user-defined
3304       // record type.
3305       if (SizeOfArgTy != QualType()) {
3306         if (PointeeTy->isRecordType() &&
3307             Context.typesAreCompatible(SizeOfArgTy, DestTy)) {
3308           DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest,
3309                               PDiag(diag::warn_sizeof_pointer_type_memaccess)
3310                                 << FnName << SizeOfArgTy << ArgIdx
3311                                 << PointeeTy << Dest->getSourceRange()
3312                                 << LenExpr->getSourceRange());
3313           break;
3314         }
3315       }
3316 
3317       // Always complain about dynamic classes.
3318       if (isDynamicClassType(PointeeTy)) {
3319 
3320         unsigned OperationType = 0;
3321         // "overwritten" if we're warning about the destination for any call
3322         // but memcmp; otherwise a verb appropriate to the call.
3323         if (ArgIdx != 0 || BId == Builtin::BImemcmp) {
3324           if (BId == Builtin::BImemcpy)
3325             OperationType = 1;
3326           else if(BId == Builtin::BImemmove)
3327             OperationType = 2;
3328           else if (BId == Builtin::BImemcmp)
3329             OperationType = 3;
3330         }
3331 
3332         DiagRuntimeBehavior(
3333           Dest->getExprLoc(), Dest,
3334           PDiag(diag::warn_dyn_class_memaccess)
3335             << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx)
3336             << FnName << PointeeTy
3337             << OperationType
3338             << Call->getCallee()->getSourceRange());
3339       } else if (PointeeTy.hasNonTrivialObjCLifetime() &&
3340                BId != Builtin::BImemset)
3341         DiagRuntimeBehavior(
3342           Dest->getExprLoc(), Dest,
3343           PDiag(diag::warn_arc_object_memaccess)
3344             << ArgIdx << FnName << PointeeTy
3345             << Call->getCallee()->getSourceRange());
3346       else
3347         continue;
3348 
3349       DiagRuntimeBehavior(
3350         Dest->getExprLoc(), Dest,
3351         PDiag(diag::note_bad_memaccess_silence)
3352           << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
3353       break;
3354     }
3355   }
3356 }
3357 
3358 // A little helper routine: ignore addition and subtraction of integer literals.
3359 // This intentionally does not ignore all integer constant expressions because
3360 // we don't want to remove sizeof().
3361 static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) {
3362   Ex = Ex->IgnoreParenCasts();
3363 
3364   for (;;) {
3365     const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex);
3366     if (!BO || !BO->isAdditiveOp())
3367       break;
3368 
3369     const Expr *RHS = BO->getRHS()->IgnoreParenCasts();
3370     const Expr *LHS = BO->getLHS()->IgnoreParenCasts();
3371 
3372     if (isa<IntegerLiteral>(RHS))
3373       Ex = LHS;
3374     else if (isa<IntegerLiteral>(LHS))
3375       Ex = RHS;
3376     else
3377       break;
3378   }
3379 
3380   return Ex;
3381 }
3382 
3383 static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty,
3384                                                       ASTContext &Context) {
3385   // Only handle constant-sized or VLAs, but not flexible members.
3386   if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) {
3387     // Only issue the FIXIT for arrays of size > 1.
3388     if (CAT->getSize().getSExtValue() <= 1)
3389       return false;
3390   } else if (!Ty->isVariableArrayType()) {
3391     return false;
3392   }
3393   return true;
3394 }
3395 
3396 // Warn if the user has made the 'size' argument to strlcpy or strlcat
3397 // be the size of the source, instead of the destination.
3398 void Sema::CheckStrlcpycatArguments(const CallExpr *Call,
3399                                     IdentifierInfo *FnName) {
3400 
3401   // Don't crash if the user has the wrong number of arguments
3402   if (Call->getNumArgs() != 3)
3403     return;
3404 
3405   const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context);
3406   const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context);
3407   const Expr *CompareWithSrc = NULL;
3408 
3409   // Look for 'strlcpy(dst, x, sizeof(x))'
3410   if (const Expr *Ex = getSizeOfExprArg(SizeArg))
3411     CompareWithSrc = Ex;
3412   else {
3413     // Look for 'strlcpy(dst, x, strlen(x))'
3414     if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) {
3415       if (SizeCall->isBuiltinCall() == Builtin::BIstrlen
3416           && SizeCall->getNumArgs() == 1)
3417         CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context);
3418     }
3419   }
3420 
3421   if (!CompareWithSrc)
3422     return;
3423 
3424   // Determine if the argument to sizeof/strlen is equal to the source
3425   // argument.  In principle there's all kinds of things you could do
3426   // here, for instance creating an == expression and evaluating it with
3427   // EvaluateAsBooleanCondition, but this uses a more direct technique:
3428   const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg);
3429   if (!SrcArgDRE)
3430     return;
3431 
3432   const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc);
3433   if (!CompareWithSrcDRE ||
3434       SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl())
3435     return;
3436 
3437   const Expr *OriginalSizeArg = Call->getArg(2);
3438   Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size)
3439     << OriginalSizeArg->getSourceRange() << FnName;
3440 
3441   // Output a FIXIT hint if the destination is an array (rather than a
3442   // pointer to an array).  This could be enhanced to handle some
3443   // pointers if we know the actual size, like if DstArg is 'array+2'
3444   // we could say 'sizeof(array)-2'.
3445   const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts();
3446   if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context))
3447     return;
3448 
3449   SmallString<128> sizeString;
3450   llvm::raw_svector_ostream OS(sizeString);
3451   OS << "sizeof(";
3452   DstArg->printPretty(OS, 0, getPrintingPolicy());
3453   OS << ")";
3454 
3455   Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size)
3456     << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(),
3457                                     OS.str());
3458 }
3459 
3460 /// Check if two expressions refer to the same declaration.
3461 static bool referToTheSameDecl(const Expr *E1, const Expr *E2) {
3462   if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1))
3463     if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2))
3464       return D1->getDecl() == D2->getDecl();
3465   return false;
3466 }
3467 
3468 static const Expr *getStrlenExprArg(const Expr *E) {
3469   if (const CallExpr *CE = dyn_cast<CallExpr>(E)) {
3470     const FunctionDecl *FD = CE->getDirectCallee();
3471     if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen)
3472       return 0;
3473     return CE->getArg(0)->IgnoreParenCasts();
3474   }
3475   return 0;
3476 }
3477 
3478 // Warn on anti-patterns as the 'size' argument to strncat.
3479 // The correct size argument should look like following:
3480 //   strncat(dst, src, sizeof(dst) - strlen(dest) - 1);
3481 void Sema::CheckStrncatArguments(const CallExpr *CE,
3482                                  IdentifierInfo *FnName) {
3483   // Don't crash if the user has the wrong number of arguments.
3484   if (CE->getNumArgs() < 3)
3485     return;
3486   const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts();
3487   const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts();
3488   const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts();
3489 
3490   // Identify common expressions, which are wrongly used as the size argument
3491   // to strncat and may lead to buffer overflows.
3492   unsigned PatternType = 0;
3493   if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) {
3494     // - sizeof(dst)
3495     if (referToTheSameDecl(SizeOfArg, DstArg))
3496       PatternType = 1;
3497     // - sizeof(src)
3498     else if (referToTheSameDecl(SizeOfArg, SrcArg))
3499       PatternType = 2;
3500   } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) {
3501     if (BE->getOpcode() == BO_Sub) {
3502       const Expr *L = BE->getLHS()->IgnoreParenCasts();
3503       const Expr *R = BE->getRHS()->IgnoreParenCasts();
3504       // - sizeof(dst) - strlen(dst)
3505       if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) &&
3506           referToTheSameDecl(DstArg, getStrlenExprArg(R)))
3507         PatternType = 1;
3508       // - sizeof(src) - (anything)
3509       else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L)))
3510         PatternType = 2;
3511     }
3512   }
3513 
3514   if (PatternType == 0)
3515     return;
3516 
3517   // Generate the diagnostic.
3518   SourceLocation SL = LenArg->getLocStart();
3519   SourceRange SR = LenArg->getSourceRange();
3520   SourceManager &SM  = PP.getSourceManager();
3521 
3522   // If the function is defined as a builtin macro, do not show macro expansion.
3523   if (SM.isMacroArgExpansion(SL)) {
3524     SL = SM.getSpellingLoc(SL);
3525     SR = SourceRange(SM.getSpellingLoc(SR.getBegin()),
3526                      SM.getSpellingLoc(SR.getEnd()));
3527   }
3528 
3529   // Check if the destination is an array (rather than a pointer to an array).
3530   QualType DstTy = DstArg->getType();
3531   bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy,
3532                                                                     Context);
3533   if (!isKnownSizeArray) {
3534     if (PatternType == 1)
3535       Diag(SL, diag::warn_strncat_wrong_size) << SR;
3536     else
3537       Diag(SL, diag::warn_strncat_src_size) << SR;
3538     return;
3539   }
3540 
3541   if (PatternType == 1)
3542     Diag(SL, diag::warn_strncat_large_size) << SR;
3543   else
3544     Diag(SL, diag::warn_strncat_src_size) << SR;
3545 
3546   SmallString<128> sizeString;
3547   llvm::raw_svector_ostream OS(sizeString);
3548   OS << "sizeof(";
3549   DstArg->printPretty(OS, 0, getPrintingPolicy());
3550   OS << ") - ";
3551   OS << "strlen(";
3552   DstArg->printPretty(OS, 0, getPrintingPolicy());
3553   OS << ") - 1";
3554 
3555   Diag(SL, diag::note_strncat_wrong_size)
3556     << FixItHint::CreateReplacement(SR, OS.str());
3557 }
3558 
3559 //===--- CHECK: Return Address of Stack Variable --------------------------===//
3560 
3561 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
3562                      Decl *ParentDecl);
3563 static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars,
3564                       Decl *ParentDecl);
3565 
3566 /// CheckReturnStackAddr - Check if a return statement returns the address
3567 ///   of a stack variable.
3568 void
3569 Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
3570                            SourceLocation ReturnLoc) {
3571 
3572   Expr *stackE = 0;
3573   SmallVector<DeclRefExpr *, 8> refVars;
3574 
3575   // Perform checking for returned stack addresses, local blocks,
3576   // label addresses or references to temporaries.
3577   if (lhsType->isPointerType() ||
3578       (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) {
3579     stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0);
3580   } else if (lhsType->isReferenceType()) {
3581     stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0);
3582   }
3583 
3584   if (stackE == 0)
3585     return; // Nothing suspicious was found.
3586 
3587   SourceLocation diagLoc;
3588   SourceRange diagRange;
3589   if (refVars.empty()) {
3590     diagLoc = stackE->getLocStart();
3591     diagRange = stackE->getSourceRange();
3592   } else {
3593     // We followed through a reference variable. 'stackE' contains the
3594     // problematic expression but we will warn at the return statement pointing
3595     // at the reference variable. We will later display the "trail" of
3596     // reference variables using notes.
3597     diagLoc = refVars[0]->getLocStart();
3598     diagRange = refVars[0]->getSourceRange();
3599   }
3600 
3601   if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
3602     Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
3603                                              : diag::warn_ret_stack_addr)
3604      << DR->getDecl()->getDeclName() << diagRange;
3605   } else if (isa<BlockExpr>(stackE)) { // local block.
3606     Diag(diagLoc, diag::err_ret_local_block) << diagRange;
3607   } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
3608     Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
3609   } else { // local temporary.
3610     Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
3611                                              : diag::warn_ret_local_temp_addr)
3612      << diagRange;
3613   }
3614 
3615   // Display the "trail" of reference variables that we followed until we
3616   // found the problematic expression using notes.
3617   for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
3618     VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
3619     // If this var binds to another reference var, show the range of the next
3620     // var, otherwise the var binds to the problematic expression, in which case
3621     // show the range of the expression.
3622     SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
3623                                   : stackE->getSourceRange();
3624     Diag(VD->getLocation(), diag::note_ref_var_local_bind)
3625       << VD->getDeclName() << range;
3626   }
3627 }
3628 
3629 /// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
3630 ///  check if the expression in a return statement evaluates to an address
3631 ///  to a location on the stack, a local block, an address of a label, or a
3632 ///  reference to local temporary. The recursion is used to traverse the
3633 ///  AST of the return expression, with recursion backtracking when we
3634 ///  encounter a subexpression that (1) clearly does not lead to one of the
3635 ///  above problematic expressions (2) is something we cannot determine leads to
3636 ///  a problematic expression based on such local checking.
3637 ///
3638 ///  Both EvalAddr and EvalVal follow through reference variables to evaluate
3639 ///  the expression that they point to. Such variables are added to the
3640 ///  'refVars' vector so that we know what the reference variable "trail" was.
3641 ///
3642 ///  EvalAddr processes expressions that are pointers that are used as
3643 ///  references (and not L-values).  EvalVal handles all other values.
3644 ///  At the base case of the recursion is a check for the above problematic
3645 ///  expressions.
3646 ///
3647 ///  This implementation handles:
3648 ///
3649 ///   * pointer-to-pointer casts
3650 ///   * implicit conversions from array references to pointers
3651 ///   * taking the address of fields
3652 ///   * arbitrary interplay between "&" and "*" operators
3653 ///   * pointer arithmetic from an address of a stack variable
3654 ///   * taking the address of an array element where the array is on the stack
3655 static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
3656                       Decl *ParentDecl) {
3657   if (E->isTypeDependent())
3658       return NULL;
3659 
3660   // We should only be called for evaluating pointer expressions.
3661   assert((E->getType()->isAnyPointerType() ||
3662           E->getType()->isBlockPointerType() ||
3663           E->getType()->isObjCQualifiedIdType()) &&
3664          "EvalAddr only works on pointers");
3665 
3666   E = E->IgnoreParens();
3667 
3668   // Our "symbolic interpreter" is just a dispatch off the currently
3669   // viewed AST node.  We then recursively traverse the AST by calling
3670   // EvalAddr and EvalVal appropriately.
3671   switch (E->getStmtClass()) {
3672   case Stmt::DeclRefExprClass: {
3673     DeclRefExpr *DR = cast<DeclRefExpr>(E);
3674 
3675     if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
3676       // If this is a reference variable, follow through to the expression that
3677       // it points to.
3678       if (V->hasLocalStorage() &&
3679           V->getType()->isReferenceType() && V->hasInit()) {
3680         // Add the reference variable to the "trail".
3681         refVars.push_back(DR);
3682         return EvalAddr(V->getInit(), refVars, ParentDecl);
3683       }
3684 
3685     return NULL;
3686   }
3687 
3688   case Stmt::UnaryOperatorClass: {
3689     // The only unary operator that make sense to handle here
3690     // is AddrOf.  All others don't make sense as pointers.
3691     UnaryOperator *U = cast<UnaryOperator>(E);
3692 
3693     if (U->getOpcode() == UO_AddrOf)
3694       return EvalVal(U->getSubExpr(), refVars, ParentDecl);
3695     else
3696       return NULL;
3697   }
3698 
3699   case Stmt::BinaryOperatorClass: {
3700     // Handle pointer arithmetic.  All other binary operators are not valid
3701     // in this context.
3702     BinaryOperator *B = cast<BinaryOperator>(E);
3703     BinaryOperatorKind op = B->getOpcode();
3704 
3705     if (op != BO_Add && op != BO_Sub)
3706       return NULL;
3707 
3708     Expr *Base = B->getLHS();
3709 
3710     // Determine which argument is the real pointer base.  It could be
3711     // the RHS argument instead of the LHS.
3712     if (!Base->getType()->isPointerType()) Base = B->getRHS();
3713 
3714     assert (Base->getType()->isPointerType());
3715     return EvalAddr(Base, refVars, ParentDecl);
3716   }
3717 
3718   // For conditional operators we need to see if either the LHS or RHS are
3719   // valid DeclRefExpr*s.  If one of them is valid, we return it.
3720   case Stmt::ConditionalOperatorClass: {
3721     ConditionalOperator *C = cast<ConditionalOperator>(E);
3722 
3723     // Handle the GNU extension for missing LHS.
3724     if (Expr *lhsExpr = C->getLHS()) {
3725     // In C++, we can have a throw-expression, which has 'void' type.
3726       if (!lhsExpr->getType()->isVoidType())
3727         if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl))
3728           return LHS;
3729     }
3730 
3731     // In C++, we can have a throw-expression, which has 'void' type.
3732     if (C->getRHS()->getType()->isVoidType())
3733       return NULL;
3734 
3735     return EvalAddr(C->getRHS(), refVars, ParentDecl);
3736   }
3737 
3738   case Stmt::BlockExprClass:
3739     if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
3740       return E; // local block.
3741     return NULL;
3742 
3743   case Stmt::AddrLabelExprClass:
3744     return E; // address of label.
3745 
3746   case Stmt::ExprWithCleanupsClass:
3747     return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,
3748                     ParentDecl);
3749 
3750   // For casts, we need to handle conversions from arrays to
3751   // pointer values, and pointer-to-pointer conversions.
3752   case Stmt::ImplicitCastExprClass:
3753   case Stmt::CStyleCastExprClass:
3754   case Stmt::CXXFunctionalCastExprClass:
3755   case Stmt::ObjCBridgedCastExprClass:
3756   case Stmt::CXXStaticCastExprClass:
3757   case Stmt::CXXDynamicCastExprClass:
3758   case Stmt::CXXConstCastExprClass:
3759   case Stmt::CXXReinterpretCastExprClass: {
3760     Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
3761     switch (cast<CastExpr>(E)->getCastKind()) {
3762     case CK_BitCast:
3763     case CK_LValueToRValue:
3764     case CK_NoOp:
3765     case CK_BaseToDerived:
3766     case CK_DerivedToBase:
3767     case CK_UncheckedDerivedToBase:
3768     case CK_Dynamic:
3769     case CK_CPointerToObjCPointerCast:
3770     case CK_BlockPointerToObjCPointerCast:
3771     case CK_AnyPointerToBlockPointerCast:
3772       return EvalAddr(SubExpr, refVars, ParentDecl);
3773 
3774     case CK_ArrayToPointerDecay:
3775       return EvalVal(SubExpr, refVars, ParentDecl);
3776 
3777     default:
3778       return 0;
3779     }
3780   }
3781 
3782   case Stmt::MaterializeTemporaryExprClass:
3783     if (Expr *Result = EvalAddr(
3784                          cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
3785                                 refVars, ParentDecl))
3786       return Result;
3787 
3788     return E;
3789 
3790   // Everything else: we simply don't reason about them.
3791   default:
3792     return NULL;
3793   }
3794 }
3795 
3796 
3797 ///  EvalVal - This function is complements EvalAddr in the mutual recursion.
3798 ///   See the comments for EvalAddr for more details.
3799 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
3800                      Decl *ParentDecl) {
3801 do {
3802   // We should only be called for evaluating non-pointer expressions, or
3803   // expressions with a pointer type that are not used as references but instead
3804   // are l-values (e.g., DeclRefExpr with a pointer type).
3805 
3806   // Our "symbolic interpreter" is just a dispatch off the currently
3807   // viewed AST node.  We then recursively traverse the AST by calling
3808   // EvalAddr and EvalVal appropriately.
3809 
3810   E = E->IgnoreParens();
3811   switch (E->getStmtClass()) {
3812   case Stmt::ImplicitCastExprClass: {
3813     ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
3814     if (IE->getValueKind() == VK_LValue) {
3815       E = IE->getSubExpr();
3816       continue;
3817     }
3818     return NULL;
3819   }
3820 
3821   case Stmt::ExprWithCleanupsClass:
3822     return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl);
3823 
3824   case Stmt::DeclRefExprClass: {
3825     // When we hit a DeclRefExpr we are looking at code that refers to a
3826     // variable's name. If it's not a reference variable we check if it has
3827     // local storage within the function, and if so, return the expression.
3828     DeclRefExpr *DR = cast<DeclRefExpr>(E);
3829 
3830     if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) {
3831       // Check if it refers to itself, e.g. "int& i = i;".
3832       if (V == ParentDecl)
3833         return DR;
3834 
3835       if (V->hasLocalStorage()) {
3836         if (!V->getType()->isReferenceType())
3837           return DR;
3838 
3839         // Reference variable, follow through to the expression that
3840         // it points to.
3841         if (V->hasInit()) {
3842           // Add the reference variable to the "trail".
3843           refVars.push_back(DR);
3844           return EvalVal(V->getInit(), refVars, V);
3845         }
3846       }
3847     }
3848 
3849     return NULL;
3850   }
3851 
3852   case Stmt::UnaryOperatorClass: {
3853     // The only unary operator that make sense to handle here
3854     // is Deref.  All others don't resolve to a "name."  This includes
3855     // handling all sorts of rvalues passed to a unary operator.
3856     UnaryOperator *U = cast<UnaryOperator>(E);
3857 
3858     if (U->getOpcode() == UO_Deref)
3859       return EvalAddr(U->getSubExpr(), refVars, ParentDecl);
3860 
3861     return NULL;
3862   }
3863 
3864   case Stmt::ArraySubscriptExprClass: {
3865     // Array subscripts are potential references to data on the stack.  We
3866     // retrieve the DeclRefExpr* for the array variable if it indeed
3867     // has local storage.
3868     return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl);
3869   }
3870 
3871   case Stmt::ConditionalOperatorClass: {
3872     // For conditional operators we need to see if either the LHS or RHS are
3873     // non-NULL Expr's.  If one is non-NULL, we return it.
3874     ConditionalOperator *C = cast<ConditionalOperator>(E);
3875 
3876     // Handle the GNU extension for missing LHS.
3877     if (Expr *lhsExpr = C->getLHS())
3878       if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl))
3879         return LHS;
3880 
3881     return EvalVal(C->getRHS(), refVars, ParentDecl);
3882   }
3883 
3884   // Accesses to members are potential references to data on the stack.
3885   case Stmt::MemberExprClass: {
3886     MemberExpr *M = cast<MemberExpr>(E);
3887 
3888     // Check for indirect access.  We only want direct field accesses.
3889     if (M->isArrow())
3890       return NULL;
3891 
3892     // Check whether the member type is itself a reference, in which case
3893     // we're not going to refer to the member, but to what the member refers to.
3894     if (M->getMemberDecl()->getType()->isReferenceType())
3895       return NULL;
3896 
3897     return EvalVal(M->getBase(), refVars, ParentDecl);
3898   }
3899 
3900   case Stmt::MaterializeTemporaryExprClass:
3901     if (Expr *Result = EvalVal(
3902                           cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
3903                                refVars, ParentDecl))
3904       return Result;
3905 
3906     return E;
3907 
3908   default:
3909     // Check that we don't return or take the address of a reference to a
3910     // temporary. This is only useful in C++.
3911     if (!E->isTypeDependent() && E->isRValue())
3912       return E;
3913 
3914     // Everything else: we simply don't reason about them.
3915     return NULL;
3916   }
3917 } while (true);
3918 }
3919 
3920 //===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
3921 
3922 /// Check for comparisons of floating point operands using != and ==.
3923 /// Issue a warning if these are no self-comparisons, as they are not likely
3924 /// to do what the programmer intended.
3925 void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) {
3926   Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts();
3927   Expr* RightExprSansParen = RHS->IgnoreParenImpCasts();
3928 
3929   // Special case: check for x == x (which is OK).
3930   // Do not emit warnings for such cases.
3931   if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
3932     if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
3933       if (DRL->getDecl() == DRR->getDecl())
3934         return;
3935 
3936 
3937   // Special case: check for comparisons against literals that can be exactly
3938   //  represented by APFloat.  In such cases, do not emit a warning.  This
3939   //  is a heuristic: often comparison against such literals are used to
3940   //  detect if a value in a variable has not changed.  This clearly can
3941   //  lead to false negatives.
3942   if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
3943     if (FLL->isExact())
3944       return;
3945   } else
3946     if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen))
3947       if (FLR->isExact())
3948         return;
3949 
3950   // Check for comparisons with builtin types.
3951   if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
3952     if (CL->isBuiltinCall())
3953       return;
3954 
3955   if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
3956     if (CR->isBuiltinCall())
3957       return;
3958 
3959   // Emit the diagnostic.
3960   Diag(Loc, diag::warn_floatingpoint_eq)
3961     << LHS->getSourceRange() << RHS->getSourceRange();
3962 }
3963 
3964 //===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
3965 //===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
3966 
3967 namespace {
3968 
3969 /// Structure recording the 'active' range of an integer-valued
3970 /// expression.
3971 struct IntRange {
3972   /// The number of bits active in the int.
3973   unsigned Width;
3974 
3975   /// True if the int is known not to have negative values.
3976   bool NonNegative;
3977 
3978   IntRange(unsigned Width, bool NonNegative)
3979     : Width(Width), NonNegative(NonNegative)
3980   {}
3981 
3982   /// Returns the range of the bool type.
3983   static IntRange forBoolType() {
3984     return IntRange(1, true);
3985   }
3986 
3987   /// Returns the range of an opaque value of the given integral type.
3988   static IntRange forValueOfType(ASTContext &C, QualType T) {
3989     return forValueOfCanonicalType(C,
3990                           T->getCanonicalTypeInternal().getTypePtr());
3991   }
3992 
3993   /// Returns the range of an opaque value of a canonical integral type.
3994   static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
3995     assert(T->isCanonicalUnqualified());
3996 
3997     if (const VectorType *VT = dyn_cast<VectorType>(T))
3998       T = VT->getElementType().getTypePtr();
3999     if (const ComplexType *CT = dyn_cast<ComplexType>(T))
4000       T = CT->getElementType().getTypePtr();
4001 
4002     // For enum types, use the known bit width of the enumerators.
4003     if (const EnumType *ET = dyn_cast<EnumType>(T)) {
4004       EnumDecl *Enum = ET->getDecl();
4005       if (!Enum->isCompleteDefinition())
4006         return IntRange(C.getIntWidth(QualType(T, 0)), false);
4007 
4008       unsigned NumPositive = Enum->getNumPositiveBits();
4009       unsigned NumNegative = Enum->getNumNegativeBits();
4010 
4011       if (NumNegative == 0)
4012         return IntRange(NumPositive, true/*NonNegative*/);
4013       else
4014         return IntRange(std::max(NumPositive + 1, NumNegative),
4015                         false/*NonNegative*/);
4016     }
4017 
4018     const BuiltinType *BT = cast<BuiltinType>(T);
4019     assert(BT->isInteger());
4020 
4021     return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
4022   }
4023 
4024   /// Returns the "target" range of a canonical integral type, i.e.
4025   /// the range of values expressible in the type.
4026   ///
4027   /// This matches forValueOfCanonicalType except that enums have the
4028   /// full range of their type, not the range of their enumerators.
4029   static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
4030     assert(T->isCanonicalUnqualified());
4031 
4032     if (const VectorType *VT = dyn_cast<VectorType>(T))
4033       T = VT->getElementType().getTypePtr();
4034     if (const ComplexType *CT = dyn_cast<ComplexType>(T))
4035       T = CT->getElementType().getTypePtr();
4036     if (const EnumType *ET = dyn_cast<EnumType>(T))
4037       T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr();
4038 
4039     const BuiltinType *BT = cast<BuiltinType>(T);
4040     assert(BT->isInteger());
4041 
4042     return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
4043   }
4044 
4045   /// Returns the supremum of two ranges: i.e. their conservative merge.
4046   static IntRange join(IntRange L, IntRange R) {
4047     return IntRange(std::max(L.Width, R.Width),
4048                     L.NonNegative && R.NonNegative);
4049   }
4050 
4051   /// Returns the infinum of two ranges: i.e. their aggressive merge.
4052   static IntRange meet(IntRange L, IntRange R) {
4053     return IntRange(std::min(L.Width, R.Width),
4054                     L.NonNegative || R.NonNegative);
4055   }
4056 };
4057 
4058 static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value,
4059                               unsigned MaxWidth) {
4060   if (value.isSigned() && value.isNegative())
4061     return IntRange(value.getMinSignedBits(), false);
4062 
4063   if (value.getBitWidth() > MaxWidth)
4064     value = value.trunc(MaxWidth);
4065 
4066   // isNonNegative() just checks the sign bit without considering
4067   // signedness.
4068   return IntRange(value.getActiveBits(), true);
4069 }
4070 
4071 static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
4072                               unsigned MaxWidth) {
4073   if (result.isInt())
4074     return GetValueRange(C, result.getInt(), MaxWidth);
4075 
4076   if (result.isVector()) {
4077     IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
4078     for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
4079       IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
4080       R = IntRange::join(R, El);
4081     }
4082     return R;
4083   }
4084 
4085   if (result.isComplexInt()) {
4086     IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
4087     IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
4088     return IntRange::join(R, I);
4089   }
4090 
4091   // This can happen with lossless casts to intptr_t of "based" lvalues.
4092   // Assume it might use arbitrary bits.
4093   // FIXME: The only reason we need to pass the type in here is to get
4094   // the sign right on this one case.  It would be nice if APValue
4095   // preserved this.
4096   assert(result.isLValue() || result.isAddrLabelDiff());
4097   return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType());
4098 }
4099 
4100 /// Pseudo-evaluate the given integer expression, estimating the
4101 /// range of values it might take.
4102 ///
4103 /// \param MaxWidth - the width to which the value will be truncated
4104 static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
4105   E = E->IgnoreParens();
4106 
4107   // Try a full evaluation first.
4108   Expr::EvalResult result;
4109   if (E->EvaluateAsRValue(result, C))
4110     return GetValueRange(C, result.Val, E->getType(), MaxWidth);
4111 
4112   // I think we only want to look through implicit casts here; if the
4113   // user has an explicit widening cast, we should treat the value as
4114   // being of the new, wider type.
4115   if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
4116     if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue)
4117       return GetExprRange(C, CE->getSubExpr(), MaxWidth);
4118 
4119     IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
4120 
4121     bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
4122 
4123     // Assume that non-integer casts can span the full range of the type.
4124     if (!isIntegerCast)
4125       return OutputTypeRange;
4126 
4127     IntRange SubRange
4128       = GetExprRange(C, CE->getSubExpr(),
4129                      std::min(MaxWidth, OutputTypeRange.Width));
4130 
4131     // Bail out if the subexpr's range is as wide as the cast type.
4132     if (SubRange.Width >= OutputTypeRange.Width)
4133       return OutputTypeRange;
4134 
4135     // Otherwise, we take the smaller width, and we're non-negative if
4136     // either the output type or the subexpr is.
4137     return IntRange(SubRange.Width,
4138                     SubRange.NonNegative || OutputTypeRange.NonNegative);
4139   }
4140 
4141   if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
4142     // If we can fold the condition, just take that operand.
4143     bool CondResult;
4144     if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
4145       return GetExprRange(C, CondResult ? CO->getTrueExpr()
4146                                         : CO->getFalseExpr(),
4147                           MaxWidth);
4148 
4149     // Otherwise, conservatively merge.
4150     IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
4151     IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
4152     return IntRange::join(L, R);
4153   }
4154 
4155   if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
4156     switch (BO->getOpcode()) {
4157 
4158     // Boolean-valued operations are single-bit and positive.
4159     case BO_LAnd:
4160     case BO_LOr:
4161     case BO_LT:
4162     case BO_GT:
4163     case BO_LE:
4164     case BO_GE:
4165     case BO_EQ:
4166     case BO_NE:
4167       return IntRange::forBoolType();
4168 
4169     // The type of the assignments is the type of the LHS, so the RHS
4170     // is not necessarily the same type.
4171     case BO_MulAssign:
4172     case BO_DivAssign:
4173     case BO_RemAssign:
4174     case BO_AddAssign:
4175     case BO_SubAssign:
4176     case BO_XorAssign:
4177     case BO_OrAssign:
4178       // TODO: bitfields?
4179       return IntRange::forValueOfType(C, E->getType());
4180 
4181     // Simple assignments just pass through the RHS, which will have
4182     // been coerced to the LHS type.
4183     case BO_Assign:
4184       // TODO: bitfields?
4185       return GetExprRange(C, BO->getRHS(), MaxWidth);
4186 
4187     // Operations with opaque sources are black-listed.
4188     case BO_PtrMemD:
4189     case BO_PtrMemI:
4190       return IntRange::forValueOfType(C, E->getType());
4191 
4192     // Bitwise-and uses the *infinum* of the two source ranges.
4193     case BO_And:
4194     case BO_AndAssign:
4195       return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
4196                             GetExprRange(C, BO->getRHS(), MaxWidth));
4197 
4198     // Left shift gets black-listed based on a judgement call.
4199     case BO_Shl:
4200       // ...except that we want to treat '1 << (blah)' as logically
4201       // positive.  It's an important idiom.
4202       if (IntegerLiteral *I
4203             = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
4204         if (I->getValue() == 1) {
4205           IntRange R = IntRange::forValueOfType(C, E->getType());
4206           return IntRange(R.Width, /*NonNegative*/ true);
4207         }
4208       }
4209       // fallthrough
4210 
4211     case BO_ShlAssign:
4212       return IntRange::forValueOfType(C, E->getType());
4213 
4214     // Right shift by a constant can narrow its left argument.
4215     case BO_Shr:
4216     case BO_ShrAssign: {
4217       IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
4218 
4219       // If the shift amount is a positive constant, drop the width by
4220       // that much.
4221       llvm::APSInt shift;
4222       if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
4223           shift.isNonNegative()) {
4224         unsigned zext = shift.getZExtValue();
4225         if (zext >= L.Width)
4226           L.Width = (L.NonNegative ? 0 : 1);
4227         else
4228           L.Width -= zext;
4229       }
4230 
4231       return L;
4232     }
4233 
4234     // Comma acts as its right operand.
4235     case BO_Comma:
4236       return GetExprRange(C, BO->getRHS(), MaxWidth);
4237 
4238     // Black-list pointer subtractions.
4239     case BO_Sub:
4240       if (BO->getLHS()->getType()->isPointerType())
4241         return IntRange::forValueOfType(C, E->getType());
4242       break;
4243 
4244     // The width of a division result is mostly determined by the size
4245     // of the LHS.
4246     case BO_Div: {
4247       // Don't 'pre-truncate' the operands.
4248       unsigned opWidth = C.getIntWidth(E->getType());
4249       IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
4250 
4251       // If the divisor is constant, use that.
4252       llvm::APSInt divisor;
4253       if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) {
4254         unsigned log2 = divisor.logBase2(); // floor(log_2(divisor))
4255         if (log2 >= L.Width)
4256           L.Width = (L.NonNegative ? 0 : 1);
4257         else
4258           L.Width = std::min(L.Width - log2, MaxWidth);
4259         return L;
4260       }
4261 
4262       // Otherwise, just use the LHS's width.
4263       IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
4264       return IntRange(L.Width, L.NonNegative && R.NonNegative);
4265     }
4266 
4267     // The result of a remainder can't be larger than the result of
4268     // either side.
4269     case BO_Rem: {
4270       // Don't 'pre-truncate' the operands.
4271       unsigned opWidth = C.getIntWidth(E->getType());
4272       IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
4273       IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
4274 
4275       IntRange meet = IntRange::meet(L, R);
4276       meet.Width = std::min(meet.Width, MaxWidth);
4277       return meet;
4278     }
4279 
4280     // The default behavior is okay for these.
4281     case BO_Mul:
4282     case BO_Add:
4283     case BO_Xor:
4284     case BO_Or:
4285       break;
4286     }
4287 
4288     // The default case is to treat the operation as if it were closed
4289     // on the narrowest type that encompasses both operands.
4290     IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
4291     IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
4292     return IntRange::join(L, R);
4293   }
4294 
4295   if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
4296     switch (UO->getOpcode()) {
4297     // Boolean-valued operations are white-listed.
4298     case UO_LNot:
4299       return IntRange::forBoolType();
4300 
4301     // Operations with opaque sources are black-listed.
4302     case UO_Deref:
4303     case UO_AddrOf: // should be impossible
4304       return IntRange::forValueOfType(C, E->getType());
4305 
4306     default:
4307       return GetExprRange(C, UO->getSubExpr(), MaxWidth);
4308     }
4309   }
4310 
4311   if (dyn_cast<OffsetOfExpr>(E)) {
4312     IntRange::forValueOfType(C, E->getType());
4313   }
4314 
4315   if (FieldDecl *BitField = E->getBitField())
4316     return IntRange(BitField->getBitWidthValue(C),
4317                     BitField->getType()->isUnsignedIntegerOrEnumerationType());
4318 
4319   return IntRange::forValueOfType(C, E->getType());
4320 }
4321 
4322 static IntRange GetExprRange(ASTContext &C, Expr *E) {
4323   return GetExprRange(C, E, C.getIntWidth(E->getType()));
4324 }
4325 
4326 /// Checks whether the given value, which currently has the given
4327 /// source semantics, has the same value when coerced through the
4328 /// target semantics.
4329 static bool IsSameFloatAfterCast(const llvm::APFloat &value,
4330                                  const llvm::fltSemantics &Src,
4331                                  const llvm::fltSemantics &Tgt) {
4332   llvm::APFloat truncated = value;
4333 
4334   bool ignored;
4335   truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
4336   truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
4337 
4338   return truncated.bitwiseIsEqual(value);
4339 }
4340 
4341 /// Checks whether the given value, which currently has the given
4342 /// source semantics, has the same value when coerced through the
4343 /// target semantics.
4344 ///
4345 /// The value might be a vector of floats (or a complex number).
4346 static bool IsSameFloatAfterCast(const APValue &value,
4347                                  const llvm::fltSemantics &Src,
4348                                  const llvm::fltSemantics &Tgt) {
4349   if (value.isFloat())
4350     return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
4351 
4352   if (value.isVector()) {
4353     for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
4354       if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
4355         return false;
4356     return true;
4357   }
4358 
4359   assert(value.isComplexFloat());
4360   return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
4361           IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
4362 }
4363 
4364 static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
4365 
4366 static bool IsZero(Sema &S, Expr *E) {
4367   // Suppress cases where we are comparing against an enum constant.
4368   if (const DeclRefExpr *DR =
4369       dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
4370     if (isa<EnumConstantDecl>(DR->getDecl()))
4371       return false;
4372 
4373   // Suppress cases where the '0' value is expanded from a macro.
4374   if (E->getLocStart().isMacroID())
4375     return false;
4376 
4377   llvm::APSInt Value;
4378   return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
4379 }
4380 
4381 static bool HasEnumType(Expr *E) {
4382   // Strip off implicit integral promotions.
4383   while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
4384     if (ICE->getCastKind() != CK_IntegralCast &&
4385         ICE->getCastKind() != CK_NoOp)
4386       break;
4387     E = ICE->getSubExpr();
4388   }
4389 
4390   return E->getType()->isEnumeralType();
4391 }
4392 
4393 static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
4394   BinaryOperatorKind op = E->getOpcode();
4395   if (E->isValueDependent())
4396     return;
4397 
4398   if (op == BO_LT && IsZero(S, E->getRHS())) {
4399     S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
4400       << "< 0" << "false" << HasEnumType(E->getLHS())
4401       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4402   } else if (op == BO_GE && IsZero(S, E->getRHS())) {
4403     S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
4404       << ">= 0" << "true" << HasEnumType(E->getLHS())
4405       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4406   } else if (op == BO_GT && IsZero(S, E->getLHS())) {
4407     S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
4408       << "0 >" << "false" << HasEnumType(E->getRHS())
4409       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4410   } else if (op == BO_LE && IsZero(S, E->getLHS())) {
4411     S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
4412       << "0 <=" << "true" << HasEnumType(E->getRHS())
4413       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4414   }
4415 }
4416 
4417 static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E,
4418                                          Expr *Constant, Expr *Other,
4419                                          llvm::APSInt Value,
4420                                          bool RhsConstant) {
4421   // 0 values are handled later by CheckTrivialUnsignedComparison().
4422   if (Value == 0)
4423     return;
4424 
4425   BinaryOperatorKind op = E->getOpcode();
4426   QualType OtherT = Other->getType();
4427   QualType ConstantT = Constant->getType();
4428   QualType CommonT = E->getLHS()->getType();
4429   if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT))
4430     return;
4431   assert((OtherT->isIntegerType() && ConstantT->isIntegerType())
4432          && "comparison with non-integer type");
4433 
4434   bool ConstantSigned = ConstantT->isSignedIntegerType();
4435   bool CommonSigned = CommonT->isSignedIntegerType();
4436 
4437   bool EqualityOnly = false;
4438 
4439   // TODO: Investigate using GetExprRange() to get tighter bounds on
4440   // on the bit ranges.
4441   IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT);
4442   unsigned OtherWidth = OtherRange.Width;
4443 
4444   if (CommonSigned) {
4445     // The common type is signed, therefore no signed to unsigned conversion.
4446     if (!OtherRange.NonNegative) {
4447       // Check that the constant is representable in type OtherT.
4448       if (ConstantSigned) {
4449         if (OtherWidth >= Value.getMinSignedBits())
4450           return;
4451       } else { // !ConstantSigned
4452         if (OtherWidth >= Value.getActiveBits() + 1)
4453           return;
4454       }
4455     } else { // !OtherSigned
4456       // Check that the constant is representable in type OtherT.
4457       // Negative values are out of range.
4458       if (ConstantSigned) {
4459         if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits())
4460           return;
4461       } else { // !ConstantSigned
4462         if (OtherWidth >= Value.getActiveBits())
4463           return;
4464       }
4465     }
4466   } else {  // !CommonSigned
4467     if (OtherRange.NonNegative) {
4468       if (OtherWidth >= Value.getActiveBits())
4469         return;
4470     } else if (!OtherRange.NonNegative && !ConstantSigned) {
4471       // Check to see if the constant is representable in OtherT.
4472       if (OtherWidth > Value.getActiveBits())
4473         return;
4474       // Check to see if the constant is equivalent to a negative value
4475       // cast to CommonT.
4476       if (S.Context.getIntWidth(ConstantT) == S.Context.getIntWidth(CommonT) &&
4477           Value.isNegative() && Value.getMinSignedBits() <= OtherWidth)
4478         return;
4479       // The constant value rests between values that OtherT can represent after
4480       // conversion.  Relational comparison still works, but equality
4481       // comparisons will be tautological.
4482       EqualityOnly = true;
4483     } else { // OtherSigned && ConstantSigned
4484       assert(0 && "Two signed types converted to unsigned types.");
4485     }
4486   }
4487 
4488   bool PositiveConstant = !ConstantSigned || Value.isNonNegative();
4489 
4490   bool IsTrue = true;
4491   if (op == BO_EQ || op == BO_NE) {
4492     IsTrue = op == BO_NE;
4493   } else if (EqualityOnly) {
4494     return;
4495   } else if (RhsConstant) {
4496     if (op == BO_GT || op == BO_GE)
4497       IsTrue = !PositiveConstant;
4498     else // op == BO_LT || op == BO_LE
4499       IsTrue = PositiveConstant;
4500   } else {
4501     if (op == BO_LT || op == BO_LE)
4502       IsTrue = !PositiveConstant;
4503     else // op == BO_GT || op == BO_GE
4504       IsTrue = PositiveConstant;
4505   }
4506 
4507   // If this is a comparison to an enum constant, include that
4508   // constant in the diagnostic.
4509   const EnumConstantDecl *ED = 0;
4510   if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Constant))
4511     ED = dyn_cast<EnumConstantDecl>(DR->getDecl());
4512 
4513   SmallString<64> PrettySourceValue;
4514   llvm::raw_svector_ostream OS(PrettySourceValue);
4515   if (ED)
4516     OS << '\'' << *ED << "' (" << Value << ")";
4517   else
4518     OS << Value;
4519 
4520   S.Diag(E->getOperatorLoc(), diag::warn_out_of_range_compare)
4521       << OS.str() << OtherT << IsTrue
4522       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4523 }
4524 
4525 /// Analyze the operands of the given comparison.  Implements the
4526 /// fallback case from AnalyzeComparison.
4527 static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
4528   AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
4529   AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
4530 }
4531 
4532 /// \brief Implements -Wsign-compare.
4533 ///
4534 /// \param E the binary operator to check for warnings
4535 static void AnalyzeComparison(Sema &S, BinaryOperator *E) {
4536   // The type the comparison is being performed in.
4537   QualType T = E->getLHS()->getType();
4538   assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
4539          && "comparison with mismatched types");
4540   if (E->isValueDependent())
4541     return AnalyzeImpConvsInComparison(S, E);
4542 
4543   Expr *LHS = E->getLHS()->IgnoreParenImpCasts();
4544   Expr *RHS = E->getRHS()->IgnoreParenImpCasts();
4545 
4546   bool IsComparisonConstant = false;
4547 
4548   // Check whether an integer constant comparison results in a value
4549   // of 'true' or 'false'.
4550   if (T->isIntegralType(S.Context)) {
4551     llvm::APSInt RHSValue;
4552     bool IsRHSIntegralLiteral =
4553       RHS->isIntegerConstantExpr(RHSValue, S.Context);
4554     llvm::APSInt LHSValue;
4555     bool IsLHSIntegralLiteral =
4556       LHS->isIntegerConstantExpr(LHSValue, S.Context);
4557     if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral)
4558         DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true);
4559     else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral)
4560       DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false);
4561     else
4562       IsComparisonConstant =
4563         (IsRHSIntegralLiteral && IsLHSIntegralLiteral);
4564   } else if (!T->hasUnsignedIntegerRepresentation())
4565       IsComparisonConstant = E->isIntegerConstantExpr(S.Context);
4566 
4567   // We don't do anything special if this isn't an unsigned integral
4568   // comparison:  we're only interested in integral comparisons, and
4569   // signed comparisons only happen in cases we don't care to warn about.
4570   //
4571   // We also don't care about value-dependent expressions or expressions
4572   // whose result is a constant.
4573   if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant)
4574     return AnalyzeImpConvsInComparison(S, E);
4575 
4576   // Check to see if one of the (unmodified) operands is of different
4577   // signedness.
4578   Expr *signedOperand, *unsignedOperand;
4579   if (LHS->getType()->hasSignedIntegerRepresentation()) {
4580     assert(!RHS->getType()->hasSignedIntegerRepresentation() &&
4581            "unsigned comparison between two signed integer expressions?");
4582     signedOperand = LHS;
4583     unsignedOperand = RHS;
4584   } else if (RHS->getType()->hasSignedIntegerRepresentation()) {
4585     signedOperand = RHS;
4586     unsignedOperand = LHS;
4587   } else {
4588     CheckTrivialUnsignedComparison(S, E);
4589     return AnalyzeImpConvsInComparison(S, E);
4590   }
4591 
4592   // Otherwise, calculate the effective range of the signed operand.
4593   IntRange signedRange = GetExprRange(S.Context, signedOperand);
4594 
4595   // Go ahead and analyze implicit conversions in the operands.  Note
4596   // that we skip the implicit conversions on both sides.
4597   AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc());
4598   AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc());
4599 
4600   // If the signed range is non-negative, -Wsign-compare won't fire,
4601   // but we should still check for comparisons which are always true
4602   // or false.
4603   if (signedRange.NonNegative)
4604     return CheckTrivialUnsignedComparison(S, E);
4605 
4606   // For (in)equality comparisons, if the unsigned operand is a
4607   // constant which cannot collide with a overflowed signed operand,
4608   // then reinterpreting the signed operand as unsigned will not
4609   // change the result of the comparison.
4610   if (E->isEqualityOp()) {
4611     unsigned comparisonWidth = S.Context.getIntWidth(T);
4612     IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
4613 
4614     // We should never be unable to prove that the unsigned operand is
4615     // non-negative.
4616     assert(unsignedRange.NonNegative && "unsigned range includes negative?");
4617 
4618     if (unsignedRange.Width < comparisonWidth)
4619       return;
4620   }
4621 
4622   S.DiagRuntimeBehavior(E->getOperatorLoc(), E,
4623     S.PDiag(diag::warn_mixed_sign_comparison)
4624       << LHS->getType() << RHS->getType()
4625       << LHS->getSourceRange() << RHS->getSourceRange());
4626 }
4627 
4628 /// Analyzes an attempt to assign the given value to a bitfield.
4629 ///
4630 /// Returns true if there was something fishy about the attempt.
4631 static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
4632                                       SourceLocation InitLoc) {
4633   assert(Bitfield->isBitField());
4634   if (Bitfield->isInvalidDecl())
4635     return false;
4636 
4637   // White-list bool bitfields.
4638   if (Bitfield->getType()->isBooleanType())
4639     return false;
4640 
4641   // Ignore value- or type-dependent expressions.
4642   if (Bitfield->getBitWidth()->isValueDependent() ||
4643       Bitfield->getBitWidth()->isTypeDependent() ||
4644       Init->isValueDependent() ||
4645       Init->isTypeDependent())
4646     return false;
4647 
4648   Expr *OriginalInit = Init->IgnoreParenImpCasts();
4649 
4650   llvm::APSInt Value;
4651   if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects))
4652     return false;
4653 
4654   unsigned OriginalWidth = Value.getBitWidth();
4655   unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context);
4656 
4657   if (OriginalWidth <= FieldWidth)
4658     return false;
4659 
4660   // Compute the value which the bitfield will contain.
4661   llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
4662   TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType());
4663 
4664   // Check whether the stored value is equal to the original value.
4665   TruncatedValue = TruncatedValue.extend(OriginalWidth);
4666   if (llvm::APSInt::isSameValue(Value, TruncatedValue))
4667     return false;
4668 
4669   // Special-case bitfields of width 1: booleans are naturally 0/1, and
4670   // therefore don't strictly fit into a signed bitfield of width 1.
4671   if (FieldWidth == 1 && Value == 1)
4672     return false;
4673 
4674   std::string PrettyValue = Value.toString(10);
4675   std::string PrettyTrunc = TruncatedValue.toString(10);
4676 
4677   S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
4678     << PrettyValue << PrettyTrunc << OriginalInit->getType()
4679     << Init->getSourceRange();
4680 
4681   return true;
4682 }
4683 
4684 /// Analyze the given simple or compound assignment for warning-worthy
4685 /// operations.
4686 static void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
4687   // Just recurse on the LHS.
4688   AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
4689 
4690   // We want to recurse on the RHS as normal unless we're assigning to
4691   // a bitfield.
4692   if (FieldDecl *Bitfield = E->getLHS()->getBitField()) {
4693     if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
4694                                   E->getOperatorLoc())) {
4695       // Recurse, ignoring any implicit conversions on the RHS.
4696       return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
4697                                         E->getOperatorLoc());
4698     }
4699   }
4700 
4701   AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
4702 }
4703 
4704 /// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
4705 static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
4706                             SourceLocation CContext, unsigned diag,
4707                             bool pruneControlFlow = false) {
4708   if (pruneControlFlow) {
4709     S.DiagRuntimeBehavior(E->getExprLoc(), E,
4710                           S.PDiag(diag)
4711                             << SourceType << T << E->getSourceRange()
4712                             << SourceRange(CContext));
4713     return;
4714   }
4715   S.Diag(E->getExprLoc(), diag)
4716     << SourceType << T << E->getSourceRange() << SourceRange(CContext);
4717 }
4718 
4719 /// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
4720 static void DiagnoseImpCast(Sema &S, Expr *E, QualType T,
4721                             SourceLocation CContext, unsigned diag,
4722                             bool pruneControlFlow = false) {
4723   DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow);
4724 }
4725 
4726 /// Diagnose an implicit cast from a literal expression. Does not warn when the
4727 /// cast wouldn't lose information.
4728 void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
4729                                     SourceLocation CContext) {
4730   // Try to convert the literal exactly to an integer. If we can, don't warn.
4731   bool isExact = false;
4732   const llvm::APFloat &Value = FL->getValue();
4733   llvm::APSInt IntegerValue(S.Context.getIntWidth(T),
4734                             T->hasUnsignedIntegerRepresentation());
4735   if (Value.convertToInteger(IntegerValue,
4736                              llvm::APFloat::rmTowardZero, &isExact)
4737       == llvm::APFloat::opOK && isExact)
4738     return;
4739 
4740   SmallString<16> PrettySourceValue;
4741   Value.toString(PrettySourceValue);
4742   SmallString<16> PrettyTargetValue;
4743   if (T->isSpecificBuiltinType(BuiltinType::Bool))
4744     PrettyTargetValue = IntegerValue == 0 ? "false" : "true";
4745   else
4746     IntegerValue.toString(PrettyTargetValue);
4747 
4748   S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
4749     << FL->getType() << T.getUnqualifiedType() << PrettySourceValue
4750     << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext);
4751 }
4752 
4753 std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
4754   if (!Range.Width) return "0";
4755 
4756   llvm::APSInt ValueInRange = Value;
4757   ValueInRange.setIsSigned(!Range.NonNegative);
4758   ValueInRange = ValueInRange.trunc(Range.Width);
4759   return ValueInRange.toString(10);
4760 }
4761 
4762 static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) {
4763   if (!isa<ImplicitCastExpr>(Ex))
4764     return false;
4765 
4766   Expr *InnerE = Ex->IgnoreParenImpCasts();
4767   const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr();
4768   const Type *Source =
4769     S.Context.getCanonicalType(InnerE->getType()).getTypePtr();
4770   if (Target->isDependentType())
4771     return false;
4772 
4773   const BuiltinType *FloatCandidateBT =
4774     dyn_cast<BuiltinType>(ToBool ? Source : Target);
4775   const Type *BoolCandidateType = ToBool ? Target : Source;
4776 
4777   return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) &&
4778           FloatCandidateBT && (FloatCandidateBT->isFloatingPoint()));
4779 }
4780 
4781 void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall,
4782                                       SourceLocation CC) {
4783   unsigned NumArgs = TheCall->getNumArgs();
4784   for (unsigned i = 0; i < NumArgs; ++i) {
4785     Expr *CurrA = TheCall->getArg(i);
4786     if (!IsImplicitBoolFloatConversion(S, CurrA, true))
4787       continue;
4788 
4789     bool IsSwapped = ((i > 0) &&
4790         IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false));
4791     IsSwapped |= ((i < (NumArgs - 1)) &&
4792         IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false));
4793     if (IsSwapped) {
4794       // Warn on this floating-point to bool conversion.
4795       DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(),
4796                       CurrA->getType(), CC,
4797                       diag::warn_impcast_floating_point_to_bool);
4798     }
4799   }
4800 }
4801 
4802 void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
4803                              SourceLocation CC, bool *ICContext = 0) {
4804   if (E->isTypeDependent() || E->isValueDependent()) return;
4805 
4806   const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
4807   const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
4808   if (Source == Target) return;
4809   if (Target->isDependentType()) return;
4810 
4811   // If the conversion context location is invalid don't complain. We also
4812   // don't want to emit a warning if the issue occurs from the expansion of
4813   // a system macro. The problem is that 'getSpellingLoc()' is slow, so we
4814   // delay this check as long as possible. Once we detect we are in that
4815   // scenario, we just return.
4816   if (CC.isInvalid())
4817     return;
4818 
4819   // Diagnose implicit casts to bool.
4820   if (Target->isSpecificBuiltinType(BuiltinType::Bool)) {
4821     if (isa<StringLiteral>(E))
4822       // Warn on string literal to bool.  Checks for string literals in logical
4823       // expressions, for instances, assert(0 && "error here"), is prevented
4824       // by a check in AnalyzeImplicitConversions().
4825       return DiagnoseImpCast(S, E, T, CC,
4826                              diag::warn_impcast_string_literal_to_bool);
4827     if (Source->isFunctionType()) {
4828       // Warn on function to bool. Checks free functions and static member
4829       // functions. Weakly imported functions are excluded from the check,
4830       // since it's common to test their value to check whether the linker
4831       // found a definition for them.
4832       ValueDecl *D = 0;
4833       if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) {
4834         D = R->getDecl();
4835       } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) {
4836         D = M->getMemberDecl();
4837       }
4838 
4839       if (D && !D->isWeak()) {
4840         if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) {
4841           S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool)
4842             << F << E->getSourceRange() << SourceRange(CC);
4843           S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence)
4844             << FixItHint::CreateInsertion(E->getExprLoc(), "&");
4845           QualType ReturnType;
4846           UnresolvedSet<4> NonTemplateOverloads;
4847           S.isExprCallable(*E, ReturnType, NonTemplateOverloads);
4848           if (!ReturnType.isNull()
4849               && ReturnType->isSpecificBuiltinType(BuiltinType::Bool))
4850             S.Diag(E->getExprLoc(), diag::note_function_to_bool_call)
4851               << FixItHint::CreateInsertion(
4852                  S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()");
4853           return;
4854         }
4855       }
4856     }
4857   }
4858 
4859   // Strip vector types.
4860   if (isa<VectorType>(Source)) {
4861     if (!isa<VectorType>(Target)) {
4862       if (S.SourceMgr.isInSystemMacro(CC))
4863         return;
4864       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
4865     }
4866 
4867     // If the vector cast is cast between two vectors of the same size, it is
4868     // a bitcast, not a conversion.
4869     if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target))
4870       return;
4871 
4872     Source = cast<VectorType>(Source)->getElementType().getTypePtr();
4873     Target = cast<VectorType>(Target)->getElementType().getTypePtr();
4874   }
4875 
4876   // Strip complex types.
4877   if (isa<ComplexType>(Source)) {
4878     if (!isa<ComplexType>(Target)) {
4879       if (S.SourceMgr.isInSystemMacro(CC))
4880         return;
4881 
4882       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
4883     }
4884 
4885     Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
4886     Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
4887   }
4888 
4889   const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
4890   const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
4891 
4892   // If the source is floating point...
4893   if (SourceBT && SourceBT->isFloatingPoint()) {
4894     // ...and the target is floating point...
4895     if (TargetBT && TargetBT->isFloatingPoint()) {
4896       // ...then warn if we're dropping FP rank.
4897 
4898       // Builtin FP kinds are ordered by increasing FP rank.
4899       if (SourceBT->getKind() > TargetBT->getKind()) {
4900         // Don't warn about float constants that are precisely
4901         // representable in the target type.
4902         Expr::EvalResult result;
4903         if (E->EvaluateAsRValue(result, S.Context)) {
4904           // Value might be a float, a float vector, or a float complex.
4905           if (IsSameFloatAfterCast(result.Val,
4906                    S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
4907                    S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
4908             return;
4909         }
4910 
4911         if (S.SourceMgr.isInSystemMacro(CC))
4912           return;
4913 
4914         DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
4915       }
4916       return;
4917     }
4918 
4919     // If the target is integral, always warn.
4920     if (TargetBT && TargetBT->isInteger()) {
4921       if (S.SourceMgr.isInSystemMacro(CC))
4922         return;
4923 
4924       Expr *InnerE = E->IgnoreParenImpCasts();
4925       // We also want to warn on, e.g., "int i = -1.234"
4926       if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE))
4927         if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus)
4928           InnerE = UOp->getSubExpr()->IgnoreParenImpCasts();
4929 
4930       if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
4931         DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
4932       } else {
4933         DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
4934       }
4935     }
4936 
4937     // If the target is bool, warn if expr is a function or method call.
4938     if (Target->isSpecificBuiltinType(BuiltinType::Bool) &&
4939         isa<CallExpr>(E)) {
4940       // Check last argument of function call to see if it is an
4941       // implicit cast from a type matching the type the result
4942       // is being cast to.
4943       CallExpr *CEx = cast<CallExpr>(E);
4944       unsigned NumArgs = CEx->getNumArgs();
4945       if (NumArgs > 0) {
4946         Expr *LastA = CEx->getArg(NumArgs - 1);
4947         Expr *InnerE = LastA->IgnoreParenImpCasts();
4948         const Type *InnerType =
4949           S.Context.getCanonicalType(InnerE->getType()).getTypePtr();
4950         if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) {
4951           // Warn on this floating-point to bool conversion
4952           DiagnoseImpCast(S, E, T, CC,
4953                           diag::warn_impcast_floating_point_to_bool);
4954         }
4955       }
4956     }
4957     return;
4958   }
4959 
4960   if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)
4961            == Expr::NPCK_GNUNull) && !Target->isAnyPointerType()
4962       && !Target->isBlockPointerType() && !Target->isMemberPointerType()
4963       && Target->isScalarType() && !Target->isNullPtrType()) {
4964     SourceLocation Loc = E->getSourceRange().getBegin();
4965     if (Loc.isMacroID())
4966       Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first;
4967     if (!Loc.isMacroID() || CC.isMacroID())
4968       S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer)
4969           << T << clang::SourceRange(CC)
4970           << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T));
4971   }
4972 
4973   if (!Source->isIntegerType() || !Target->isIntegerType())
4974     return;
4975 
4976   // TODO: remove this early return once the false positives for constant->bool
4977   // in templates, macros, etc, are reduced or removed.
4978   if (Target->isSpecificBuiltinType(BuiltinType::Bool))
4979     return;
4980 
4981   IntRange SourceRange = GetExprRange(S.Context, E);
4982   IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
4983 
4984   if (SourceRange.Width > TargetRange.Width) {
4985     // If the source is a constant, use a default-on diagnostic.
4986     // TODO: this should happen for bitfield stores, too.
4987     llvm::APSInt Value(32);
4988     if (E->isIntegerConstantExpr(Value, S.Context)) {
4989       if (S.SourceMgr.isInSystemMacro(CC))
4990         return;
4991 
4992       std::string PrettySourceValue = Value.toString(10);
4993       std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
4994 
4995       S.DiagRuntimeBehavior(E->getExprLoc(), E,
4996         S.PDiag(diag::warn_impcast_integer_precision_constant)
4997             << PrettySourceValue << PrettyTargetValue
4998             << E->getType() << T << E->getSourceRange()
4999             << clang::SourceRange(CC));
5000       return;
5001     }
5002 
5003     // People want to build with -Wshorten-64-to-32 and not -Wconversion.
5004     if (S.SourceMgr.isInSystemMacro(CC))
5005       return;
5006 
5007     if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64)
5008       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32,
5009                              /* pruneControlFlow */ true);
5010     return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
5011   }
5012 
5013   if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
5014       (!TargetRange.NonNegative && SourceRange.NonNegative &&
5015        SourceRange.Width == TargetRange.Width)) {
5016 
5017     if (S.SourceMgr.isInSystemMacro(CC))
5018       return;
5019 
5020     unsigned DiagID = diag::warn_impcast_integer_sign;
5021 
5022     // Traditionally, gcc has warned about this under -Wsign-compare.
5023     // We also want to warn about it in -Wconversion.
5024     // So if -Wconversion is off, use a completely identical diagnostic
5025     // in the sign-compare group.
5026     // The conditional-checking code will
5027     if (ICContext) {
5028       DiagID = diag::warn_impcast_integer_sign_conditional;
5029       *ICContext = true;
5030     }
5031 
5032     return DiagnoseImpCast(S, E, T, CC, DiagID);
5033   }
5034 
5035   // Diagnose conversions between different enumeration types.
5036   // In C, we pretend that the type of an EnumConstantDecl is its enumeration
5037   // type, to give us better diagnostics.
5038   QualType SourceType = E->getType();
5039   if (!S.getLangOpts().CPlusPlus) {
5040     if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
5041       if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
5042         EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
5043         SourceType = S.Context.getTypeDeclType(Enum);
5044         Source = S.Context.getCanonicalType(SourceType).getTypePtr();
5045       }
5046   }
5047 
5048   if (const EnumType *SourceEnum = Source->getAs<EnumType>())
5049     if (const EnumType *TargetEnum = Target->getAs<EnumType>())
5050       if (SourceEnum->getDecl()->hasNameForLinkage() &&
5051           TargetEnum->getDecl()->hasNameForLinkage() &&
5052           SourceEnum != TargetEnum) {
5053         if (S.SourceMgr.isInSystemMacro(CC))
5054           return;
5055 
5056         return DiagnoseImpCast(S, E, SourceType, T, CC,
5057                                diag::warn_impcast_different_enum_types);
5058       }
5059 
5060   return;
5061 }
5062 
5063 void CheckConditionalOperator(Sema &S, ConditionalOperator *E,
5064                               SourceLocation CC, QualType T);
5065 
5066 void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
5067                              SourceLocation CC, bool &ICContext) {
5068   E = E->IgnoreParenImpCasts();
5069 
5070   if (isa<ConditionalOperator>(E))
5071     return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T);
5072 
5073   AnalyzeImplicitConversions(S, E, CC);
5074   if (E->getType() != T)
5075     return CheckImplicitConversion(S, E, T, CC, &ICContext);
5076   return;
5077 }
5078 
5079 void CheckConditionalOperator(Sema &S, ConditionalOperator *E,
5080                               SourceLocation CC, QualType T) {
5081   AnalyzeImplicitConversions(S, E->getCond(), CC);
5082 
5083   bool Suspicious = false;
5084   CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
5085   CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
5086 
5087   // If -Wconversion would have warned about either of the candidates
5088   // for a signedness conversion to the context type...
5089   if (!Suspicious) return;
5090 
5091   // ...but it's currently ignored...
5092   if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
5093                                  CC))
5094     return;
5095 
5096   // ...then check whether it would have warned about either of the
5097   // candidates for a signedness conversion to the condition type.
5098   if (E->getType() == T) return;
5099 
5100   Suspicious = false;
5101   CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
5102                           E->getType(), CC, &Suspicious);
5103   if (!Suspicious)
5104     CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
5105                             E->getType(), CC, &Suspicious);
5106 }
5107 
5108 /// AnalyzeImplicitConversions - Find and report any interesting
5109 /// implicit conversions in the given expression.  There are a couple
5110 /// of competing diagnostics here, -Wconversion and -Wsign-compare.
5111 void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
5112   QualType T = OrigE->getType();
5113   Expr *E = OrigE->IgnoreParenImpCasts();
5114 
5115   if (E->isTypeDependent() || E->isValueDependent())
5116     return;
5117 
5118   // For conditional operators, we analyze the arguments as if they
5119   // were being fed directly into the output.
5120   if (isa<ConditionalOperator>(E)) {
5121     ConditionalOperator *CO = cast<ConditionalOperator>(E);
5122     CheckConditionalOperator(S, CO, CC, T);
5123     return;
5124   }
5125 
5126   // Check implicit argument conversions for function calls.
5127   if (CallExpr *Call = dyn_cast<CallExpr>(E))
5128     CheckImplicitArgumentConversions(S, Call, CC);
5129 
5130   // Go ahead and check any implicit conversions we might have skipped.
5131   // The non-canonical typecheck is just an optimization;
5132   // CheckImplicitConversion will filter out dead implicit conversions.
5133   if (E->getType() != T)
5134     CheckImplicitConversion(S, E, T, CC);
5135 
5136   // Now continue drilling into this expression.
5137 
5138   // Skip past explicit casts.
5139   if (isa<ExplicitCastExpr>(E)) {
5140     E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
5141     return AnalyzeImplicitConversions(S, E, CC);
5142   }
5143 
5144   if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
5145     // Do a somewhat different check with comparison operators.
5146     if (BO->isComparisonOp())
5147       return AnalyzeComparison(S, BO);
5148 
5149     // And with simple assignments.
5150     if (BO->getOpcode() == BO_Assign)
5151       return AnalyzeAssignment(S, BO);
5152   }
5153 
5154   // These break the otherwise-useful invariant below.  Fortunately,
5155   // we don't really need to recurse into them, because any internal
5156   // expressions should have been analyzed already when they were
5157   // built into statements.
5158   if (isa<StmtExpr>(E)) return;
5159 
5160   // Don't descend into unevaluated contexts.
5161   if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
5162 
5163   // Now just recurse over the expression's children.
5164   CC = E->getExprLoc();
5165   BinaryOperator *BO = dyn_cast<BinaryOperator>(E);
5166   bool IsLogicalOperator = BO && BO->isLogicalOp();
5167   for (Stmt::child_range I = E->children(); I; ++I) {
5168     Expr *ChildExpr = dyn_cast_or_null<Expr>(*I);
5169     if (!ChildExpr)
5170       continue;
5171 
5172     if (IsLogicalOperator &&
5173         isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts()))
5174       // Ignore checking string literals that are in logical operators.
5175       continue;
5176     AnalyzeImplicitConversions(S, ChildExpr, CC);
5177   }
5178 }
5179 
5180 } // end anonymous namespace
5181 
5182 /// Diagnoses "dangerous" implicit conversions within the given
5183 /// expression (which is a full expression).  Implements -Wconversion
5184 /// and -Wsign-compare.
5185 ///
5186 /// \param CC the "context" location of the implicit conversion, i.e.
5187 ///   the most location of the syntactic entity requiring the implicit
5188 ///   conversion
5189 void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
5190   // Don't diagnose in unevaluated contexts.
5191   if (isUnevaluatedContext())
5192     return;
5193 
5194   // Don't diagnose for value- or type-dependent expressions.
5195   if (E->isTypeDependent() || E->isValueDependent())
5196     return;
5197 
5198   // Check for array bounds violations in cases where the check isn't triggered
5199   // elsewhere for other Expr types (like BinaryOperators), e.g. when an
5200   // ArraySubscriptExpr is on the RHS of a variable initialization.
5201   CheckArrayAccess(E);
5202 
5203   // This is not the right CC for (e.g.) a variable initialization.
5204   AnalyzeImplicitConversions(*this, E, CC);
5205 }
5206 
5207 /// Diagnose when expression is an integer constant expression and its evaluation
5208 /// results in integer overflow
5209 void Sema::CheckForIntOverflow (Expr *E) {
5210   if (isa<BinaryOperator>(E->IgnoreParens())) {
5211     llvm::SmallVector<PartialDiagnosticAt, 4> Diags;
5212     E->EvaluateForOverflow(Context, &Diags);
5213   }
5214 }
5215 
5216 namespace {
5217 /// \brief Visitor for expressions which looks for unsequenced operations on the
5218 /// same object.
5219 class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> {
5220   /// \brief A tree of sequenced regions within an expression. Two regions are
5221   /// unsequenced if one is an ancestor or a descendent of the other. When we
5222   /// finish processing an expression with sequencing, such as a comma
5223   /// expression, we fold its tree nodes into its parent, since they are
5224   /// unsequenced with respect to nodes we will visit later.
5225   class SequenceTree {
5226     struct Value {
5227       explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {}
5228       unsigned Parent : 31;
5229       bool Merged : 1;
5230     };
5231     llvm::SmallVector<Value, 8> Values;
5232 
5233   public:
5234     /// \brief A region within an expression which may be sequenced with respect
5235     /// to some other region.
5236     class Seq {
5237       explicit Seq(unsigned N) : Index(N) {}
5238       unsigned Index;
5239       friend class SequenceTree;
5240     public:
5241       Seq() : Index(0) {}
5242     };
5243 
5244     SequenceTree() { Values.push_back(Value(0)); }
5245     Seq root() const { return Seq(0); }
5246 
5247     /// \brief Create a new sequence of operations, which is an unsequenced
5248     /// subset of \p Parent. This sequence of operations is sequenced with
5249     /// respect to other children of \p Parent.
5250     Seq allocate(Seq Parent) {
5251       Values.push_back(Value(Parent.Index));
5252       return Seq(Values.size() - 1);
5253     }
5254 
5255     /// \brief Merge a sequence of operations into its parent.
5256     void merge(Seq S) {
5257       Values[S.Index].Merged = true;
5258     }
5259 
5260     /// \brief Determine whether two operations are unsequenced. This operation
5261     /// is asymmetric: \p Cur should be the more recent sequence, and \p Old
5262     /// should have been merged into its parent as appropriate.
5263     bool isUnsequenced(Seq Cur, Seq Old) {
5264       unsigned C = representative(Cur.Index);
5265       unsigned Target = representative(Old.Index);
5266       while (C >= Target) {
5267         if (C == Target)
5268           return true;
5269         C = Values[C].Parent;
5270       }
5271       return false;
5272     }
5273 
5274   private:
5275     /// \brief Pick a representative for a sequence.
5276     unsigned representative(unsigned K) {
5277       if (Values[K].Merged)
5278         // Perform path compression as we go.
5279         return Values[K].Parent = representative(Values[K].Parent);
5280       return K;
5281     }
5282   };
5283 
5284   /// An object for which we can track unsequenced uses.
5285   typedef NamedDecl *Object;
5286 
5287   /// Different flavors of object usage which we track. We only track the
5288   /// least-sequenced usage of each kind.
5289   enum UsageKind {
5290     /// A read of an object. Multiple unsequenced reads are OK.
5291     UK_Use,
5292     /// A modification of an object which is sequenced before the value
5293     /// computation of the expression, such as ++n.
5294     UK_ModAsValue,
5295     /// A modification of an object which is not sequenced before the value
5296     /// computation of the expression, such as n++.
5297     UK_ModAsSideEffect,
5298 
5299     UK_Count = UK_ModAsSideEffect + 1
5300   };
5301 
5302   struct Usage {
5303     Usage() : Use(0), Seq() {}
5304     Expr *Use;
5305     SequenceTree::Seq Seq;
5306   };
5307 
5308   struct UsageInfo {
5309     UsageInfo() : Diagnosed(false) {}
5310     Usage Uses[UK_Count];
5311     /// Have we issued a diagnostic for this variable already?
5312     bool Diagnosed;
5313   };
5314   typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap;
5315 
5316   Sema &SemaRef;
5317   /// Sequenced regions within the expression.
5318   SequenceTree Tree;
5319   /// Declaration modifications and references which we have seen.
5320   UsageInfoMap UsageMap;
5321   /// The region we are currently within.
5322   SequenceTree::Seq Region;
5323   /// Filled in with declarations which were modified as a side-effect
5324   /// (that is, post-increment operations).
5325   llvm::SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect;
5326   /// Expressions to check later. We defer checking these to reduce
5327   /// stack usage.
5328   llvm::SmallVectorImpl<Expr*> &WorkList;
5329 
5330   /// RAII object wrapping the visitation of a sequenced subexpression of an
5331   /// expression. At the end of this process, the side-effects of the evaluation
5332   /// become sequenced with respect to the value computation of the result, so
5333   /// we downgrade any UK_ModAsSideEffect within the evaluation to
5334   /// UK_ModAsValue.
5335   struct SequencedSubexpression {
5336     SequencedSubexpression(SequenceChecker &Self)
5337       : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) {
5338       Self.ModAsSideEffect = &ModAsSideEffect;
5339     }
5340     ~SequencedSubexpression() {
5341       for (unsigned I = 0, E = ModAsSideEffect.size(); I != E; ++I) {
5342         UsageInfo &U = Self.UsageMap[ModAsSideEffect[I].first];
5343         U.Uses[UK_ModAsSideEffect] = ModAsSideEffect[I].second;
5344         Self.addUsage(U, ModAsSideEffect[I].first,
5345                       ModAsSideEffect[I].second.Use, UK_ModAsValue);
5346       }
5347       Self.ModAsSideEffect = OldModAsSideEffect;
5348     }
5349 
5350     SequenceChecker &Self;
5351     llvm::SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect;
5352     llvm::SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect;
5353   };
5354 
5355   /// \brief Find the object which is produced by the specified expression,
5356   /// if any.
5357   Object getObject(Expr *E, bool Mod) const {
5358     E = E->IgnoreParenCasts();
5359     if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
5360       if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec))
5361         return getObject(UO->getSubExpr(), Mod);
5362     } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
5363       if (BO->getOpcode() == BO_Comma)
5364         return getObject(BO->getRHS(), Mod);
5365       if (Mod && BO->isAssignmentOp())
5366         return getObject(BO->getLHS(), Mod);
5367     } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) {
5368       // FIXME: Check for more interesting cases, like "x.n = ++x.n".
5369       if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts()))
5370         return ME->getMemberDecl();
5371     } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
5372       // FIXME: If this is a reference, map through to its value.
5373       return DRE->getDecl();
5374     return 0;
5375   }
5376 
5377   /// \brief Note that an object was modified or used by an expression.
5378   void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) {
5379     Usage &U = UI.Uses[UK];
5380     if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) {
5381       if (UK == UK_ModAsSideEffect && ModAsSideEffect)
5382         ModAsSideEffect->push_back(std::make_pair(O, U));
5383       U.Use = Ref;
5384       U.Seq = Region;
5385     }
5386   }
5387   /// \brief Check whether a modification or use conflicts with a prior usage.
5388   void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind,
5389                   bool IsModMod) {
5390     if (UI.Diagnosed)
5391       return;
5392 
5393     const Usage &U = UI.Uses[OtherKind];
5394     if (!U.Use || !Tree.isUnsequenced(Region, U.Seq))
5395       return;
5396 
5397     Expr *Mod = U.Use;
5398     Expr *ModOrUse = Ref;
5399     if (OtherKind == UK_Use)
5400       std::swap(Mod, ModOrUse);
5401 
5402     SemaRef.Diag(Mod->getExprLoc(),
5403                  IsModMod ? diag::warn_unsequenced_mod_mod
5404                           : diag::warn_unsequenced_mod_use)
5405       << O << SourceRange(ModOrUse->getExprLoc());
5406     UI.Diagnosed = true;
5407   }
5408 
5409   void notePreUse(Object O, Expr *Use) {
5410     UsageInfo &U = UsageMap[O];
5411     // Uses conflict with other modifications.
5412     checkUsage(O, U, Use, UK_ModAsValue, false);
5413   }
5414   void notePostUse(Object O, Expr *Use) {
5415     UsageInfo &U = UsageMap[O];
5416     checkUsage(O, U, Use, UK_ModAsSideEffect, false);
5417     addUsage(U, O, Use, UK_Use);
5418   }
5419 
5420   void notePreMod(Object O, Expr *Mod) {
5421     UsageInfo &U = UsageMap[O];
5422     // Modifications conflict with other modifications and with uses.
5423     checkUsage(O, U, Mod, UK_ModAsValue, true);
5424     checkUsage(O, U, Mod, UK_Use, false);
5425   }
5426   void notePostMod(Object O, Expr *Use, UsageKind UK) {
5427     UsageInfo &U = UsageMap[O];
5428     checkUsage(O, U, Use, UK_ModAsSideEffect, true);
5429     addUsage(U, O, Use, UK);
5430   }
5431 
5432 public:
5433   SequenceChecker(Sema &S, Expr *E,
5434                   llvm::SmallVectorImpl<Expr*> &WorkList)
5435     : EvaluatedExprVisitor<SequenceChecker>(S.Context), SemaRef(S),
5436       Region(Tree.root()), ModAsSideEffect(0), WorkList(WorkList) {
5437     Visit(E);
5438   }
5439 
5440   void VisitStmt(Stmt *S) {
5441     // Skip all statements which aren't expressions for now.
5442   }
5443 
5444   void VisitExpr(Expr *E) {
5445     // By default, just recurse to evaluated subexpressions.
5446     EvaluatedExprVisitor<SequenceChecker>::VisitStmt(E);
5447   }
5448 
5449   void VisitCastExpr(CastExpr *E) {
5450     Object O = Object();
5451     if (E->getCastKind() == CK_LValueToRValue)
5452       O = getObject(E->getSubExpr(), false);
5453 
5454     if (O)
5455       notePreUse(O, E);
5456     VisitExpr(E);
5457     if (O)
5458       notePostUse(O, E);
5459   }
5460 
5461   void VisitBinComma(BinaryOperator *BO) {
5462     // C++11 [expr.comma]p1:
5463     //   Every value computation and side effect associated with the left
5464     //   expression is sequenced before every value computation and side
5465     //   effect associated with the right expression.
5466     SequenceTree::Seq LHS = Tree.allocate(Region);
5467     SequenceTree::Seq RHS = Tree.allocate(Region);
5468     SequenceTree::Seq OldRegion = Region;
5469 
5470     {
5471       SequencedSubexpression SeqLHS(*this);
5472       Region = LHS;
5473       Visit(BO->getLHS());
5474     }
5475 
5476     Region = RHS;
5477     Visit(BO->getRHS());
5478 
5479     Region = OldRegion;
5480 
5481     // Forget that LHS and RHS are sequenced. They are both unsequenced
5482     // with respect to other stuff.
5483     Tree.merge(LHS);
5484     Tree.merge(RHS);
5485   }
5486 
5487   void VisitBinAssign(BinaryOperator *BO) {
5488     // The modification is sequenced after the value computation of the LHS
5489     // and RHS, so check it before inspecting the operands and update the
5490     // map afterwards.
5491     Object O = getObject(BO->getLHS(), true);
5492     if (!O)
5493       return VisitExpr(BO);
5494 
5495     notePreMod(O, BO);
5496 
5497     // C++11 [expr.ass]p7:
5498     //   E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated
5499     //   only once.
5500     //
5501     // Therefore, for a compound assignment operator, O is considered used
5502     // everywhere except within the evaluation of E1 itself.
5503     if (isa<CompoundAssignOperator>(BO))
5504       notePreUse(O, BO);
5505 
5506     Visit(BO->getLHS());
5507 
5508     if (isa<CompoundAssignOperator>(BO))
5509       notePostUse(O, BO);
5510 
5511     Visit(BO->getRHS());
5512 
5513     notePostMod(O, BO, UK_ModAsValue);
5514   }
5515   void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) {
5516     VisitBinAssign(CAO);
5517   }
5518 
5519   void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); }
5520   void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); }
5521   void VisitUnaryPreIncDec(UnaryOperator *UO) {
5522     Object O = getObject(UO->getSubExpr(), true);
5523     if (!O)
5524       return VisitExpr(UO);
5525 
5526     notePreMod(O, UO);
5527     Visit(UO->getSubExpr());
5528     notePostMod(O, UO, UK_ModAsValue);
5529   }
5530 
5531   void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); }
5532   void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); }
5533   void VisitUnaryPostIncDec(UnaryOperator *UO) {
5534     Object O = getObject(UO->getSubExpr(), true);
5535     if (!O)
5536       return VisitExpr(UO);
5537 
5538     notePreMod(O, UO);
5539     Visit(UO->getSubExpr());
5540     notePostMod(O, UO, UK_ModAsSideEffect);
5541   }
5542 
5543   /// Don't visit the RHS of '&&' or '||' if it might not be evaluated.
5544   void VisitBinLOr(BinaryOperator *BO) {
5545     // The side-effects of the LHS of an '&&' are sequenced before the
5546     // value computation of the RHS, and hence before the value computation
5547     // of the '&&' itself, unless the LHS evaluates to zero. We treat them
5548     // as if they were unconditionally sequenced.
5549     {
5550       SequencedSubexpression Sequenced(*this);
5551       Visit(BO->getLHS());
5552     }
5553 
5554     bool Result;
5555     if (!BO->getLHS()->isValueDependent() &&
5556         BO->getLHS()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) {
5557       if (!Result)
5558         Visit(BO->getRHS());
5559     } else {
5560       // Check for unsequenced operations in the RHS, treating it as an
5561       // entirely separate evaluation.
5562       //
5563       // FIXME: If there are operations in the RHS which are unsequenced
5564       // with respect to operations outside the RHS, and those operations
5565       // are unconditionally evaluated, diagnose them.
5566       WorkList.push_back(BO->getRHS());
5567     }
5568   }
5569   void VisitBinLAnd(BinaryOperator *BO) {
5570     {
5571       SequencedSubexpression Sequenced(*this);
5572       Visit(BO->getLHS());
5573     }
5574 
5575     bool Result;
5576     if (!BO->getLHS()->isValueDependent() &&
5577         BO->getLHS()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) {
5578       if (Result)
5579         Visit(BO->getRHS());
5580     } else {
5581       WorkList.push_back(BO->getRHS());
5582     }
5583   }
5584 
5585   // Only visit the condition, unless we can be sure which subexpression will
5586   // be chosen.
5587   void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) {
5588     SequencedSubexpression Sequenced(*this);
5589     Visit(CO->getCond());
5590 
5591     bool Result;
5592     if (!CO->getCond()->isValueDependent() &&
5593         CO->getCond()->EvaluateAsBooleanCondition(Result, SemaRef.Context))
5594       Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr());
5595     else {
5596       WorkList.push_back(CO->getTrueExpr());
5597       WorkList.push_back(CO->getFalseExpr());
5598     }
5599   }
5600 
5601   void VisitCXXConstructExpr(CXXConstructExpr *CCE) {
5602     if (!CCE->isListInitialization())
5603       return VisitExpr(CCE);
5604 
5605     // In C++11, list initializations are sequenced.
5606     llvm::SmallVector<SequenceTree::Seq, 32> Elts;
5607     SequenceTree::Seq Parent = Region;
5608     for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(),
5609                                         E = CCE->arg_end();
5610          I != E; ++I) {
5611       Region = Tree.allocate(Parent);
5612       Elts.push_back(Region);
5613       Visit(*I);
5614     }
5615 
5616     // Forget that the initializers are sequenced.
5617     Region = Parent;
5618     for (unsigned I = 0; I < Elts.size(); ++I)
5619       Tree.merge(Elts[I]);
5620   }
5621 
5622   void VisitInitListExpr(InitListExpr *ILE) {
5623     if (!SemaRef.getLangOpts().CPlusPlus11)
5624       return VisitExpr(ILE);
5625 
5626     // In C++11, list initializations are sequenced.
5627     llvm::SmallVector<SequenceTree::Seq, 32> Elts;
5628     SequenceTree::Seq Parent = Region;
5629     for (unsigned I = 0; I < ILE->getNumInits(); ++I) {
5630       Expr *E = ILE->getInit(I);
5631       if (!E) continue;
5632       Region = Tree.allocate(Parent);
5633       Elts.push_back(Region);
5634       Visit(E);
5635     }
5636 
5637     // Forget that the initializers are sequenced.
5638     Region = Parent;
5639     for (unsigned I = 0; I < Elts.size(); ++I)
5640       Tree.merge(Elts[I]);
5641   }
5642 };
5643 }
5644 
5645 void Sema::CheckUnsequencedOperations(Expr *E) {
5646   llvm::SmallVector<Expr*, 8> WorkList;
5647   WorkList.push_back(E);
5648   while (!WorkList.empty()) {
5649     Expr *Item = WorkList.back();
5650     WorkList.pop_back();
5651     SequenceChecker(*this, Item, WorkList);
5652   }
5653 }
5654 
5655 void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc,
5656                               bool IsConstexpr) {
5657   CheckImplicitConversions(E, CheckLoc);
5658   CheckUnsequencedOperations(E);
5659   if (!IsConstexpr && !E->isValueDependent())
5660     CheckForIntOverflow(E);
5661 }
5662 
5663 void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
5664                                        FieldDecl *BitField,
5665                                        Expr *Init) {
5666   (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
5667 }
5668 
5669 /// CheckParmsForFunctionDef - Check that the parameters of the given
5670 /// function are appropriate for the definition of a function. This
5671 /// takes care of any checks that cannot be performed on the
5672 /// declaration itself, e.g., that the types of each of the function
5673 /// parameters are complete.
5674 bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
5675                                     bool CheckParameterNames) {
5676   bool HasInvalidParm = false;
5677   for (; P != PEnd; ++P) {
5678     ParmVarDecl *Param = *P;
5679 
5680     // C99 6.7.5.3p4: the parameters in a parameter type list in a
5681     // function declarator that is part of a function definition of
5682     // that function shall not have incomplete type.
5683     //
5684     // This is also C++ [dcl.fct]p6.
5685     if (!Param->isInvalidDecl() &&
5686         RequireCompleteType(Param->getLocation(), Param->getType(),
5687                             diag::err_typecheck_decl_incomplete_type)) {
5688       Param->setInvalidDecl();
5689       HasInvalidParm = true;
5690     }
5691 
5692     // C99 6.9.1p5: If the declarator includes a parameter type list, the
5693     // declaration of each parameter shall include an identifier.
5694     if (CheckParameterNames &&
5695         Param->getIdentifier() == 0 &&
5696         !Param->isImplicit() &&
5697         !getLangOpts().CPlusPlus)
5698       Diag(Param->getLocation(), diag::err_parameter_name_omitted);
5699 
5700     // C99 6.7.5.3p12:
5701     //   If the function declarator is not part of a definition of that
5702     //   function, parameters may have incomplete type and may use the [*]
5703     //   notation in their sequences of declarator specifiers to specify
5704     //   variable length array types.
5705     QualType PType = Param->getOriginalType();
5706     if (const ArrayType *AT = Context.getAsArrayType(PType)) {
5707       if (AT->getSizeModifier() == ArrayType::Star) {
5708         // FIXME: This diagnostic should point the '[*]' if source-location
5709         // information is added for it.
5710         Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
5711       }
5712     }
5713   }
5714 
5715   return HasInvalidParm;
5716 }
5717 
5718 /// CheckCastAlign - Implements -Wcast-align, which warns when a
5719 /// pointer cast increases the alignment requirements.
5720 void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
5721   // This is actually a lot of work to potentially be doing on every
5722   // cast; don't do it if we're ignoring -Wcast_align (as is the default).
5723   if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
5724                                           TRange.getBegin())
5725         == DiagnosticsEngine::Ignored)
5726     return;
5727 
5728   // Ignore dependent types.
5729   if (T->isDependentType() || Op->getType()->isDependentType())
5730     return;
5731 
5732   // Require that the destination be a pointer type.
5733   const PointerType *DestPtr = T->getAs<PointerType>();
5734   if (!DestPtr) return;
5735 
5736   // If the destination has alignment 1, we're done.
5737   QualType DestPointee = DestPtr->getPointeeType();
5738   if (DestPointee->isIncompleteType()) return;
5739   CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
5740   if (DestAlign.isOne()) return;
5741 
5742   // Require that the source be a pointer type.
5743   const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
5744   if (!SrcPtr) return;
5745   QualType SrcPointee = SrcPtr->getPointeeType();
5746 
5747   // Whitelist casts from cv void*.  We already implicitly
5748   // whitelisted casts to cv void*, since they have alignment 1.
5749   // Also whitelist casts involving incomplete types, which implicitly
5750   // includes 'void'.
5751   if (SrcPointee->isIncompleteType()) return;
5752 
5753   CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
5754   if (SrcAlign >= DestAlign) return;
5755 
5756   Diag(TRange.getBegin(), diag::warn_cast_align)
5757     << Op->getType() << T
5758     << static_cast<unsigned>(SrcAlign.getQuantity())
5759     << static_cast<unsigned>(DestAlign.getQuantity())
5760     << TRange << Op->getSourceRange();
5761 }
5762 
5763 static const Type* getElementType(const Expr *BaseExpr) {
5764   const Type* EltType = BaseExpr->getType().getTypePtr();
5765   if (EltType->isAnyPointerType())
5766     return EltType->getPointeeType().getTypePtr();
5767   else if (EltType->isArrayType())
5768     return EltType->getBaseElementTypeUnsafe();
5769   return EltType;
5770 }
5771 
5772 /// \brief Check whether this array fits the idiom of a size-one tail padded
5773 /// array member of a struct.
5774 ///
5775 /// We avoid emitting out-of-bounds access warnings for such arrays as they are
5776 /// commonly used to emulate flexible arrays in C89 code.
5777 static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size,
5778                                     const NamedDecl *ND) {
5779   if (Size != 1 || !ND) return false;
5780 
5781   const FieldDecl *FD = dyn_cast<FieldDecl>(ND);
5782   if (!FD) return false;
5783 
5784   // Don't consider sizes resulting from macro expansions or template argument
5785   // substitution to form C89 tail-padded arrays.
5786 
5787   TypeSourceInfo *TInfo = FD->getTypeSourceInfo();
5788   while (TInfo) {
5789     TypeLoc TL = TInfo->getTypeLoc();
5790     // Look through typedefs.
5791     if (TypedefTypeLoc TTL = TL.getAs<TypedefTypeLoc>()) {
5792       const TypedefNameDecl *TDL = TTL.getTypedefNameDecl();
5793       TInfo = TDL->getTypeSourceInfo();
5794       continue;
5795     }
5796     if (ConstantArrayTypeLoc CTL = TL.getAs<ConstantArrayTypeLoc>()) {
5797       const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr());
5798       if (!SizeExpr || SizeExpr->getExprLoc().isMacroID())
5799         return false;
5800     }
5801     break;
5802   }
5803 
5804   const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext());
5805   if (!RD) return false;
5806   if (RD->isUnion()) return false;
5807   if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) {
5808     if (!CRD->isStandardLayout()) return false;
5809   }
5810 
5811   // See if this is the last field decl in the record.
5812   const Decl *D = FD;
5813   while ((D = D->getNextDeclInContext()))
5814     if (isa<FieldDecl>(D))
5815       return false;
5816   return true;
5817 }
5818 
5819 void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr,
5820                             const ArraySubscriptExpr *ASE,
5821                             bool AllowOnePastEnd, bool IndexNegated) {
5822   IndexExpr = IndexExpr->IgnoreParenImpCasts();
5823   if (IndexExpr->isValueDependent())
5824     return;
5825 
5826   const Type *EffectiveType = getElementType(BaseExpr);
5827   BaseExpr = BaseExpr->IgnoreParenCasts();
5828   const ConstantArrayType *ArrayTy =
5829     Context.getAsConstantArrayType(BaseExpr->getType());
5830   if (!ArrayTy)
5831     return;
5832 
5833   llvm::APSInt index;
5834   if (!IndexExpr->EvaluateAsInt(index, Context))
5835     return;
5836   if (IndexNegated)
5837     index = -index;
5838 
5839   const NamedDecl *ND = NULL;
5840   if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
5841     ND = dyn_cast<NamedDecl>(DRE->getDecl());
5842   if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
5843     ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
5844 
5845   if (index.isUnsigned() || !index.isNegative()) {
5846     llvm::APInt size = ArrayTy->getSize();
5847     if (!size.isStrictlyPositive())
5848       return;
5849 
5850     const Type* BaseType = getElementType(BaseExpr);
5851     if (BaseType != EffectiveType) {
5852       // Make sure we're comparing apples to apples when comparing index to size
5853       uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType);
5854       uint64_t array_typesize = Context.getTypeSize(BaseType);
5855       // Handle ptrarith_typesize being zero, such as when casting to void*
5856       if (!ptrarith_typesize) ptrarith_typesize = 1;
5857       if (ptrarith_typesize != array_typesize) {
5858         // There's a cast to a different size type involved
5859         uint64_t ratio = array_typesize / ptrarith_typesize;
5860         // TODO: Be smarter about handling cases where array_typesize is not a
5861         // multiple of ptrarith_typesize
5862         if (ptrarith_typesize * ratio == array_typesize)
5863           size *= llvm::APInt(size.getBitWidth(), ratio);
5864       }
5865     }
5866 
5867     if (size.getBitWidth() > index.getBitWidth())
5868       index = index.zext(size.getBitWidth());
5869     else if (size.getBitWidth() < index.getBitWidth())
5870       size = size.zext(index.getBitWidth());
5871 
5872     // For array subscripting the index must be less than size, but for pointer
5873     // arithmetic also allow the index (offset) to be equal to size since
5874     // computing the next address after the end of the array is legal and
5875     // commonly done e.g. in C++ iterators and range-based for loops.
5876     if (AllowOnePastEnd ? index.ule(size) : index.ult(size))
5877       return;
5878 
5879     // Also don't warn for arrays of size 1 which are members of some
5880     // structure. These are often used to approximate flexible arrays in C89
5881     // code.
5882     if (IsTailPaddedMemberArray(*this, size, ND))
5883       return;
5884 
5885     // Suppress the warning if the subscript expression (as identified by the
5886     // ']' location) and the index expression are both from macro expansions
5887     // within a system header.
5888     if (ASE) {
5889       SourceLocation RBracketLoc = SourceMgr.getSpellingLoc(
5890           ASE->getRBracketLoc());
5891       if (SourceMgr.isInSystemHeader(RBracketLoc)) {
5892         SourceLocation IndexLoc = SourceMgr.getSpellingLoc(
5893             IndexExpr->getLocStart());
5894         if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc))
5895           return;
5896       }
5897     }
5898 
5899     unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds;
5900     if (ASE)
5901       DiagID = diag::warn_array_index_exceeds_bounds;
5902 
5903     DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
5904                         PDiag(DiagID) << index.toString(10, true)
5905                           << size.toString(10, true)
5906                           << (unsigned)size.getLimitedValue(~0U)
5907                           << IndexExpr->getSourceRange());
5908   } else {
5909     unsigned DiagID = diag::warn_array_index_precedes_bounds;
5910     if (!ASE) {
5911       DiagID = diag::warn_ptr_arith_precedes_bounds;
5912       if (index.isNegative()) index = -index;
5913     }
5914 
5915     DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
5916                         PDiag(DiagID) << index.toString(10, true)
5917                           << IndexExpr->getSourceRange());
5918   }
5919 
5920   if (!ND) {
5921     // Try harder to find a NamedDecl to point at in the note.
5922     while (const ArraySubscriptExpr *ASE =
5923            dyn_cast<ArraySubscriptExpr>(BaseExpr))
5924       BaseExpr = ASE->getBase()->IgnoreParenCasts();
5925     if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
5926       ND = dyn_cast<NamedDecl>(DRE->getDecl());
5927     if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
5928       ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
5929   }
5930 
5931   if (ND)
5932     DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
5933                         PDiag(diag::note_array_index_out_of_bounds)
5934                           << ND->getDeclName());
5935 }
5936 
5937 void Sema::CheckArrayAccess(const Expr *expr) {
5938   int AllowOnePastEnd = 0;
5939   while (expr) {
5940     expr = expr->IgnoreParenImpCasts();
5941     switch (expr->getStmtClass()) {
5942       case Stmt::ArraySubscriptExprClass: {
5943         const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr);
5944         CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE,
5945                          AllowOnePastEnd > 0);
5946         return;
5947       }
5948       case Stmt::UnaryOperatorClass: {
5949         // Only unwrap the * and & unary operators
5950         const UnaryOperator *UO = cast<UnaryOperator>(expr);
5951         expr = UO->getSubExpr();
5952         switch (UO->getOpcode()) {
5953           case UO_AddrOf:
5954             AllowOnePastEnd++;
5955             break;
5956           case UO_Deref:
5957             AllowOnePastEnd--;
5958             break;
5959           default:
5960             return;
5961         }
5962         break;
5963       }
5964       case Stmt::ConditionalOperatorClass: {
5965         const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
5966         if (const Expr *lhs = cond->getLHS())
5967           CheckArrayAccess(lhs);
5968         if (const Expr *rhs = cond->getRHS())
5969           CheckArrayAccess(rhs);
5970         return;
5971       }
5972       default:
5973         return;
5974     }
5975   }
5976 }
5977 
5978 //===--- CHECK: Objective-C retain cycles ----------------------------------//
5979 
5980 namespace {
5981   struct RetainCycleOwner {
5982     RetainCycleOwner() : Variable(0), Indirect(false) {}
5983     VarDecl *Variable;
5984     SourceRange Range;
5985     SourceLocation Loc;
5986     bool Indirect;
5987 
5988     void setLocsFrom(Expr *e) {
5989       Loc = e->getExprLoc();
5990       Range = e->getSourceRange();
5991     }
5992   };
5993 }
5994 
5995 /// Consider whether capturing the given variable can possibly lead to
5996 /// a retain cycle.
5997 static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) {
5998   // In ARC, it's captured strongly iff the variable has __strong
5999   // lifetime.  In MRR, it's captured strongly if the variable is
6000   // __block and has an appropriate type.
6001   if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
6002     return false;
6003 
6004   owner.Variable = var;
6005   if (ref)
6006     owner.setLocsFrom(ref);
6007   return true;
6008 }
6009 
6010 static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) {
6011   while (true) {
6012     e = e->IgnoreParens();
6013     if (CastExpr *cast = dyn_cast<CastExpr>(e)) {
6014       switch (cast->getCastKind()) {
6015       case CK_BitCast:
6016       case CK_LValueBitCast:
6017       case CK_LValueToRValue:
6018       case CK_ARCReclaimReturnedObject:
6019         e = cast->getSubExpr();
6020         continue;
6021 
6022       default:
6023         return false;
6024       }
6025     }
6026 
6027     if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) {
6028       ObjCIvarDecl *ivar = ref->getDecl();
6029       if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
6030         return false;
6031 
6032       // Try to find a retain cycle in the base.
6033       if (!findRetainCycleOwner(S, ref->getBase(), owner))
6034         return false;
6035 
6036       if (ref->isFreeIvar()) owner.setLocsFrom(ref);
6037       owner.Indirect = true;
6038       return true;
6039     }
6040 
6041     if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) {
6042       VarDecl *var = dyn_cast<VarDecl>(ref->getDecl());
6043       if (!var) return false;
6044       return considerVariable(var, ref, owner);
6045     }
6046 
6047     if (MemberExpr *member = dyn_cast<MemberExpr>(e)) {
6048       if (member->isArrow()) return false;
6049 
6050       // Don't count this as an indirect ownership.
6051       e = member->getBase();
6052       continue;
6053     }
6054 
6055     if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) {
6056       // Only pay attention to pseudo-objects on property references.
6057       ObjCPropertyRefExpr *pre
6058         = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm()
6059                                               ->IgnoreParens());
6060       if (!pre) return false;
6061       if (pre->isImplicitProperty()) return false;
6062       ObjCPropertyDecl *property = pre->getExplicitProperty();
6063       if (!property->isRetaining() &&
6064           !(property->getPropertyIvarDecl() &&
6065             property->getPropertyIvarDecl()->getType()
6066               .getObjCLifetime() == Qualifiers::OCL_Strong))
6067           return false;
6068 
6069       owner.Indirect = true;
6070       if (pre->isSuperReceiver()) {
6071         owner.Variable = S.getCurMethodDecl()->getSelfDecl();
6072         if (!owner.Variable)
6073           return false;
6074         owner.Loc = pre->getLocation();
6075         owner.Range = pre->getSourceRange();
6076         return true;
6077       }
6078       e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase())
6079                               ->getSourceExpr());
6080       continue;
6081     }
6082 
6083     // Array ivars?
6084 
6085     return false;
6086   }
6087 }
6088 
6089 namespace {
6090   struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> {
6091     FindCaptureVisitor(ASTContext &Context, VarDecl *variable)
6092       : EvaluatedExprVisitor<FindCaptureVisitor>(Context),
6093         Variable(variable), Capturer(0) {}
6094 
6095     VarDecl *Variable;
6096     Expr *Capturer;
6097 
6098     void VisitDeclRefExpr(DeclRefExpr *ref) {
6099       if (ref->getDecl() == Variable && !Capturer)
6100         Capturer = ref;
6101     }
6102 
6103     void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) {
6104       if (Capturer) return;
6105       Visit(ref->getBase());
6106       if (Capturer && ref->isFreeIvar())
6107         Capturer = ref;
6108     }
6109 
6110     void VisitBlockExpr(BlockExpr *block) {
6111       // Look inside nested blocks
6112       if (block->getBlockDecl()->capturesVariable(Variable))
6113         Visit(block->getBlockDecl()->getBody());
6114     }
6115 
6116     void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) {
6117       if (Capturer) return;
6118       if (OVE->getSourceExpr())
6119         Visit(OVE->getSourceExpr());
6120     }
6121   };
6122 }
6123 
6124 /// Check whether the given argument is a block which captures a
6125 /// variable.
6126 static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) {
6127   assert(owner.Variable && owner.Loc.isValid());
6128 
6129   e = e->IgnoreParenCasts();
6130 
6131   // Look through [^{...} copy] and Block_copy(^{...}).
6132   if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) {
6133     Selector Cmd = ME->getSelector();
6134     if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") {
6135       e = ME->getInstanceReceiver();
6136       if (!e)
6137         return 0;
6138       e = e->IgnoreParenCasts();
6139     }
6140   } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) {
6141     if (CE->getNumArgs() == 1) {
6142       FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl());
6143       if (Fn) {
6144         const IdentifierInfo *FnI = Fn->getIdentifier();
6145         if (FnI && FnI->isStr("_Block_copy")) {
6146           e = CE->getArg(0)->IgnoreParenCasts();
6147         }
6148       }
6149     }
6150   }
6151 
6152   BlockExpr *block = dyn_cast<BlockExpr>(e);
6153   if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable))
6154     return 0;
6155 
6156   FindCaptureVisitor visitor(S.Context, owner.Variable);
6157   visitor.Visit(block->getBlockDecl()->getBody());
6158   return visitor.Capturer;
6159 }
6160 
6161 static void diagnoseRetainCycle(Sema &S, Expr *capturer,
6162                                 RetainCycleOwner &owner) {
6163   assert(capturer);
6164   assert(owner.Variable && owner.Loc.isValid());
6165 
6166   S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle)
6167     << owner.Variable << capturer->getSourceRange();
6168   S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner)
6169     << owner.Indirect << owner.Range;
6170 }
6171 
6172 /// Check for a keyword selector that starts with the word 'add' or
6173 /// 'set'.
6174 static bool isSetterLikeSelector(Selector sel) {
6175   if (sel.isUnarySelector()) return false;
6176 
6177   StringRef str = sel.getNameForSlot(0);
6178   while (!str.empty() && str.front() == '_') str = str.substr(1);
6179   if (str.startswith("set"))
6180     str = str.substr(3);
6181   else if (str.startswith("add")) {
6182     // Specially whitelist 'addOperationWithBlock:'.
6183     if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock"))
6184       return false;
6185     str = str.substr(3);
6186   }
6187   else
6188     return false;
6189 
6190   if (str.empty()) return true;
6191   return !isLowercase(str.front());
6192 }
6193 
6194 /// Check a message send to see if it's likely to cause a retain cycle.
6195 void Sema::checkRetainCycles(ObjCMessageExpr *msg) {
6196   // Only check instance methods whose selector looks like a setter.
6197   if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector()))
6198     return;
6199 
6200   // Try to find a variable that the receiver is strongly owned by.
6201   RetainCycleOwner owner;
6202   if (msg->getReceiverKind() == ObjCMessageExpr::Instance) {
6203     if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner))
6204       return;
6205   } else {
6206     assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance);
6207     owner.Variable = getCurMethodDecl()->getSelfDecl();
6208     owner.Loc = msg->getSuperLoc();
6209     owner.Range = msg->getSuperLoc();
6210   }
6211 
6212   // Check whether the receiver is captured by any of the arguments.
6213   for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i)
6214     if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner))
6215       return diagnoseRetainCycle(*this, capturer, owner);
6216 }
6217 
6218 /// Check a property assign to see if it's likely to cause a retain cycle.
6219 void Sema::checkRetainCycles(Expr *receiver, Expr *argument) {
6220   RetainCycleOwner owner;
6221   if (!findRetainCycleOwner(*this, receiver, owner))
6222     return;
6223 
6224   if (Expr *capturer = findCapturingExpr(*this, argument, owner))
6225     diagnoseRetainCycle(*this, capturer, owner);
6226 }
6227 
6228 void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) {
6229   RetainCycleOwner Owner;
6230   if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner))
6231     return;
6232 
6233   // Because we don't have an expression for the variable, we have to set the
6234   // location explicitly here.
6235   Owner.Loc = Var->getLocation();
6236   Owner.Range = Var->getSourceRange();
6237 
6238   if (Expr *Capturer = findCapturingExpr(*this, Init, Owner))
6239     diagnoseRetainCycle(*this, Capturer, Owner);
6240 }
6241 
6242 static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc,
6243                                      Expr *RHS, bool isProperty) {
6244   // Check if RHS is an Objective-C object literal, which also can get
6245   // immediately zapped in a weak reference.  Note that we explicitly
6246   // allow ObjCStringLiterals, since those are designed to never really die.
6247   RHS = RHS->IgnoreParenImpCasts();
6248 
6249   // This enum needs to match with the 'select' in
6250   // warn_objc_arc_literal_assign (off-by-1).
6251   Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS);
6252   if (Kind == Sema::LK_String || Kind == Sema::LK_None)
6253     return false;
6254 
6255   S.Diag(Loc, diag::warn_arc_literal_assign)
6256     << (unsigned) Kind
6257     << (isProperty ? 0 : 1)
6258     << RHS->getSourceRange();
6259 
6260   return true;
6261 }
6262 
6263 static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc,
6264                                     Qualifiers::ObjCLifetime LT,
6265                                     Expr *RHS, bool isProperty) {
6266   // Strip off any implicit cast added to get to the one ARC-specific.
6267   while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
6268     if (cast->getCastKind() == CK_ARCConsumeObject) {
6269       S.Diag(Loc, diag::warn_arc_retained_assign)
6270         << (LT == Qualifiers::OCL_ExplicitNone)
6271         << (isProperty ? 0 : 1)
6272         << RHS->getSourceRange();
6273       return true;
6274     }
6275     RHS = cast->getSubExpr();
6276   }
6277 
6278   if (LT == Qualifiers::OCL_Weak &&
6279       checkUnsafeAssignLiteral(S, Loc, RHS, isProperty))
6280     return true;
6281 
6282   return false;
6283 }
6284 
6285 bool Sema::checkUnsafeAssigns(SourceLocation Loc,
6286                               QualType LHS, Expr *RHS) {
6287   Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime();
6288 
6289   if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone)
6290     return false;
6291 
6292   if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false))
6293     return true;
6294 
6295   return false;
6296 }
6297 
6298 void Sema::checkUnsafeExprAssigns(SourceLocation Loc,
6299                               Expr *LHS, Expr *RHS) {
6300   QualType LHSType;
6301   // PropertyRef on LHS type need be directly obtained from
6302   // its declaration as it has a PsuedoType.
6303   ObjCPropertyRefExpr *PRE
6304     = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens());
6305   if (PRE && !PRE->isImplicitProperty()) {
6306     const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
6307     if (PD)
6308       LHSType = PD->getType();
6309   }
6310 
6311   if (LHSType.isNull())
6312     LHSType = LHS->getType();
6313 
6314   Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime();
6315 
6316   if (LT == Qualifiers::OCL_Weak) {
6317     DiagnosticsEngine::Level Level =
6318       Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc);
6319     if (Level != DiagnosticsEngine::Ignored)
6320       getCurFunction()->markSafeWeakUse(LHS);
6321   }
6322 
6323   if (checkUnsafeAssigns(Loc, LHSType, RHS))
6324     return;
6325 
6326   // FIXME. Check for other life times.
6327   if (LT != Qualifiers::OCL_None)
6328     return;
6329 
6330   if (PRE) {
6331     if (PRE->isImplicitProperty())
6332       return;
6333     const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
6334     if (!PD)
6335       return;
6336 
6337     unsigned Attributes = PD->getPropertyAttributes();
6338     if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) {
6339       // when 'assign' attribute was not explicitly specified
6340       // by user, ignore it and rely on property type itself
6341       // for lifetime info.
6342       unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten();
6343       if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) &&
6344           LHSType->isObjCRetainableType())
6345         return;
6346 
6347       while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
6348         if (cast->getCastKind() == CK_ARCConsumeObject) {
6349           Diag(Loc, diag::warn_arc_retained_property_assign)
6350           << RHS->getSourceRange();
6351           return;
6352         }
6353         RHS = cast->getSubExpr();
6354       }
6355     }
6356     else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) {
6357       if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true))
6358         return;
6359     }
6360   }
6361 }
6362 
6363 //===--- CHECK: Empty statement body (-Wempty-body) ---------------------===//
6364 
6365 namespace {
6366 bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr,
6367                                  SourceLocation StmtLoc,
6368                                  const NullStmt *Body) {
6369   // Do not warn if the body is a macro that expands to nothing, e.g:
6370   //
6371   // #define CALL(x)
6372   // if (condition)
6373   //   CALL(0);
6374   //
6375   if (Body->hasLeadingEmptyMacro())
6376     return false;
6377 
6378   // Get line numbers of statement and body.
6379   bool StmtLineInvalid;
6380   unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc,
6381                                                       &StmtLineInvalid);
6382   if (StmtLineInvalid)
6383     return false;
6384 
6385   bool BodyLineInvalid;
6386   unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(),
6387                                                       &BodyLineInvalid);
6388   if (BodyLineInvalid)
6389     return false;
6390 
6391   // Warn if null statement and body are on the same line.
6392   if (StmtLine != BodyLine)
6393     return false;
6394 
6395   return true;
6396 }
6397 } // Unnamed namespace
6398 
6399 void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc,
6400                                  const Stmt *Body,
6401                                  unsigned DiagID) {
6402   // Since this is a syntactic check, don't emit diagnostic for template
6403   // instantiations, this just adds noise.
6404   if (CurrentInstantiationScope)
6405     return;
6406 
6407   // The body should be a null statement.
6408   const NullStmt *NBody = dyn_cast<NullStmt>(Body);
6409   if (!NBody)
6410     return;
6411 
6412   // Do the usual checks.
6413   if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody))
6414     return;
6415 
6416   Diag(NBody->getSemiLoc(), DiagID);
6417   Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line);
6418 }
6419 
6420 void Sema::DiagnoseEmptyLoopBody(const Stmt *S,
6421                                  const Stmt *PossibleBody) {
6422   assert(!CurrentInstantiationScope); // Ensured by caller
6423 
6424   SourceLocation StmtLoc;
6425   const Stmt *Body;
6426   unsigned DiagID;
6427   if (const ForStmt *FS = dyn_cast<ForStmt>(S)) {
6428     StmtLoc = FS->getRParenLoc();
6429     Body = FS->getBody();
6430     DiagID = diag::warn_empty_for_body;
6431   } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) {
6432     StmtLoc = WS->getCond()->getSourceRange().getEnd();
6433     Body = WS->getBody();
6434     DiagID = diag::warn_empty_while_body;
6435   } else
6436     return; // Neither `for' nor `while'.
6437 
6438   // The body should be a null statement.
6439   const NullStmt *NBody = dyn_cast<NullStmt>(Body);
6440   if (!NBody)
6441     return;
6442 
6443   // Skip expensive checks if diagnostic is disabled.
6444   if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) ==
6445           DiagnosticsEngine::Ignored)
6446     return;
6447 
6448   // Do the usual checks.
6449   if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody))
6450     return;
6451 
6452   // `for(...);' and `while(...);' are popular idioms, so in order to keep
6453   // noise level low, emit diagnostics only if for/while is followed by a
6454   // CompoundStmt, e.g.:
6455   //    for (int i = 0; i < n; i++);
6456   //    {
6457   //      a(i);
6458   //    }
6459   // or if for/while is followed by a statement with more indentation
6460   // than for/while itself:
6461   //    for (int i = 0; i < n; i++);
6462   //      a(i);
6463   bool ProbableTypo = isa<CompoundStmt>(PossibleBody);
6464   if (!ProbableTypo) {
6465     bool BodyColInvalid;
6466     unsigned BodyCol = SourceMgr.getPresumedColumnNumber(
6467                              PossibleBody->getLocStart(),
6468                              &BodyColInvalid);
6469     if (BodyColInvalid)
6470       return;
6471 
6472     bool StmtColInvalid;
6473     unsigned StmtCol = SourceMgr.getPresumedColumnNumber(
6474                              S->getLocStart(),
6475                              &StmtColInvalid);
6476     if (StmtColInvalid)
6477       return;
6478 
6479     if (BodyCol > StmtCol)
6480       ProbableTypo = true;
6481   }
6482 
6483   if (ProbableTypo) {
6484     Diag(NBody->getSemiLoc(), DiagID);
6485     Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line);
6486   }
6487 }
6488 
6489 //===--- Layout compatibility ----------------------------------------------//
6490 
6491 namespace {
6492 
6493 bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2);
6494 
6495 /// \brief Check if two enumeration types are layout-compatible.
6496 bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) {
6497   // C++11 [dcl.enum] p8:
6498   // Two enumeration types are layout-compatible if they have the same
6499   // underlying type.
6500   return ED1->isComplete() && ED2->isComplete() &&
6501          C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType());
6502 }
6503 
6504 /// \brief Check if two fields are layout-compatible.
6505 bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) {
6506   if (!isLayoutCompatible(C, Field1->getType(), Field2->getType()))
6507     return false;
6508 
6509   if (Field1->isBitField() != Field2->isBitField())
6510     return false;
6511 
6512   if (Field1->isBitField()) {
6513     // Make sure that the bit-fields are the same length.
6514     unsigned Bits1 = Field1->getBitWidthValue(C);
6515     unsigned Bits2 = Field2->getBitWidthValue(C);
6516 
6517     if (Bits1 != Bits2)
6518       return false;
6519   }
6520 
6521   return true;
6522 }
6523 
6524 /// \brief Check if two standard-layout structs are layout-compatible.
6525 /// (C++11 [class.mem] p17)
6526 bool isLayoutCompatibleStruct(ASTContext &C,
6527                               RecordDecl *RD1,
6528                               RecordDecl *RD2) {
6529   // If both records are C++ classes, check that base classes match.
6530   if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) {
6531     // If one of records is a CXXRecordDecl we are in C++ mode,
6532     // thus the other one is a CXXRecordDecl, too.
6533     const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2);
6534     // Check number of base classes.
6535     if (D1CXX->getNumBases() != D2CXX->getNumBases())
6536       return false;
6537 
6538     // Check the base classes.
6539     for (CXXRecordDecl::base_class_const_iterator
6540                Base1 = D1CXX->bases_begin(),
6541            BaseEnd1 = D1CXX->bases_end(),
6542               Base2 = D2CXX->bases_begin();
6543          Base1 != BaseEnd1;
6544          ++Base1, ++Base2) {
6545       if (!isLayoutCompatible(C, Base1->getType(), Base2->getType()))
6546         return false;
6547     }
6548   } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) {
6549     // If only RD2 is a C++ class, it should have zero base classes.
6550     if (D2CXX->getNumBases() > 0)
6551       return false;
6552   }
6553 
6554   // Check the fields.
6555   RecordDecl::field_iterator Field2 = RD2->field_begin(),
6556                              Field2End = RD2->field_end(),
6557                              Field1 = RD1->field_begin(),
6558                              Field1End = RD1->field_end();
6559   for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) {
6560     if (!isLayoutCompatible(C, *Field1, *Field2))
6561       return false;
6562   }
6563   if (Field1 != Field1End || Field2 != Field2End)
6564     return false;
6565 
6566   return true;
6567 }
6568 
6569 /// \brief Check if two standard-layout unions are layout-compatible.
6570 /// (C++11 [class.mem] p18)
6571 bool isLayoutCompatibleUnion(ASTContext &C,
6572                              RecordDecl *RD1,
6573                              RecordDecl *RD2) {
6574   llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields;
6575   for (RecordDecl::field_iterator Field2 = RD2->field_begin(),
6576                                   Field2End = RD2->field_end();
6577        Field2 != Field2End; ++Field2) {
6578     UnmatchedFields.insert(*Field2);
6579   }
6580 
6581   for (RecordDecl::field_iterator Field1 = RD1->field_begin(),
6582                                   Field1End = RD1->field_end();
6583        Field1 != Field1End; ++Field1) {
6584     llvm::SmallPtrSet<FieldDecl *, 8>::iterator
6585         I = UnmatchedFields.begin(),
6586         E = UnmatchedFields.end();
6587 
6588     for ( ; I != E; ++I) {
6589       if (isLayoutCompatible(C, *Field1, *I)) {
6590         bool Result = UnmatchedFields.erase(*I);
6591         (void) Result;
6592         assert(Result);
6593         break;
6594       }
6595     }
6596     if (I == E)
6597       return false;
6598   }
6599 
6600   return UnmatchedFields.empty();
6601 }
6602 
6603 bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) {
6604   if (RD1->isUnion() != RD2->isUnion())
6605     return false;
6606 
6607   if (RD1->isUnion())
6608     return isLayoutCompatibleUnion(C, RD1, RD2);
6609   else
6610     return isLayoutCompatibleStruct(C, RD1, RD2);
6611 }
6612 
6613 /// \brief Check if two types are layout-compatible in C++11 sense.
6614 bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) {
6615   if (T1.isNull() || T2.isNull())
6616     return false;
6617 
6618   // C++11 [basic.types] p11:
6619   // If two types T1 and T2 are the same type, then T1 and T2 are
6620   // layout-compatible types.
6621   if (C.hasSameType(T1, T2))
6622     return true;
6623 
6624   T1 = T1.getCanonicalType().getUnqualifiedType();
6625   T2 = T2.getCanonicalType().getUnqualifiedType();
6626 
6627   const Type::TypeClass TC1 = T1->getTypeClass();
6628   const Type::TypeClass TC2 = T2->getTypeClass();
6629 
6630   if (TC1 != TC2)
6631     return false;
6632 
6633   if (TC1 == Type::Enum) {
6634     return isLayoutCompatible(C,
6635                               cast<EnumType>(T1)->getDecl(),
6636                               cast<EnumType>(T2)->getDecl());
6637   } else if (TC1 == Type::Record) {
6638     if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType())
6639       return false;
6640 
6641     return isLayoutCompatible(C,
6642                               cast<RecordType>(T1)->getDecl(),
6643                               cast<RecordType>(T2)->getDecl());
6644   }
6645 
6646   return false;
6647 }
6648 }
6649 
6650 //===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----//
6651 
6652 namespace {
6653 /// \brief Given a type tag expression find the type tag itself.
6654 ///
6655 /// \param TypeExpr Type tag expression, as it appears in user's code.
6656 ///
6657 /// \param VD Declaration of an identifier that appears in a type tag.
6658 ///
6659 /// \param MagicValue Type tag magic value.
6660 bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx,
6661                      const ValueDecl **VD, uint64_t *MagicValue) {
6662   while(true) {
6663     if (!TypeExpr)
6664       return false;
6665 
6666     TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts();
6667 
6668     switch (TypeExpr->getStmtClass()) {
6669     case Stmt::UnaryOperatorClass: {
6670       const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr);
6671       if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) {
6672         TypeExpr = UO->getSubExpr();
6673         continue;
6674       }
6675       return false;
6676     }
6677 
6678     case Stmt::DeclRefExprClass: {
6679       const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr);
6680       *VD = DRE->getDecl();
6681       return true;
6682     }
6683 
6684     case Stmt::IntegerLiteralClass: {
6685       const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr);
6686       llvm::APInt MagicValueAPInt = IL->getValue();
6687       if (MagicValueAPInt.getActiveBits() <= 64) {
6688         *MagicValue = MagicValueAPInt.getZExtValue();
6689         return true;
6690       } else
6691         return false;
6692     }
6693 
6694     case Stmt::BinaryConditionalOperatorClass:
6695     case Stmt::ConditionalOperatorClass: {
6696       const AbstractConditionalOperator *ACO =
6697           cast<AbstractConditionalOperator>(TypeExpr);
6698       bool Result;
6699       if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) {
6700         if (Result)
6701           TypeExpr = ACO->getTrueExpr();
6702         else
6703           TypeExpr = ACO->getFalseExpr();
6704         continue;
6705       }
6706       return false;
6707     }
6708 
6709     case Stmt::BinaryOperatorClass: {
6710       const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr);
6711       if (BO->getOpcode() == BO_Comma) {
6712         TypeExpr = BO->getRHS();
6713         continue;
6714       }
6715       return false;
6716     }
6717 
6718     default:
6719       return false;
6720     }
6721   }
6722 }
6723 
6724 /// \brief Retrieve the C type corresponding to type tag TypeExpr.
6725 ///
6726 /// \param TypeExpr Expression that specifies a type tag.
6727 ///
6728 /// \param MagicValues Registered magic values.
6729 ///
6730 /// \param FoundWrongKind Set to true if a type tag was found, but of a wrong
6731 ///        kind.
6732 ///
6733 /// \param TypeInfo Information about the corresponding C type.
6734 ///
6735 /// \returns true if the corresponding C type was found.
6736 bool GetMatchingCType(
6737         const IdentifierInfo *ArgumentKind,
6738         const Expr *TypeExpr, const ASTContext &Ctx,
6739         const llvm::DenseMap<Sema::TypeTagMagicValue,
6740                              Sema::TypeTagData> *MagicValues,
6741         bool &FoundWrongKind,
6742         Sema::TypeTagData &TypeInfo) {
6743   FoundWrongKind = false;
6744 
6745   // Variable declaration that has type_tag_for_datatype attribute.
6746   const ValueDecl *VD = NULL;
6747 
6748   uint64_t MagicValue;
6749 
6750   if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue))
6751     return false;
6752 
6753   if (VD) {
6754     for (specific_attr_iterator<TypeTagForDatatypeAttr>
6755              I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(),
6756              E = VD->specific_attr_end<TypeTagForDatatypeAttr>();
6757          I != E; ++I) {
6758       if (I->getArgumentKind() != ArgumentKind) {
6759         FoundWrongKind = true;
6760         return false;
6761       }
6762       TypeInfo.Type = I->getMatchingCType();
6763       TypeInfo.LayoutCompatible = I->getLayoutCompatible();
6764       TypeInfo.MustBeNull = I->getMustBeNull();
6765       return true;
6766     }
6767     return false;
6768   }
6769 
6770   if (!MagicValues)
6771     return false;
6772 
6773   llvm::DenseMap<Sema::TypeTagMagicValue,
6774                  Sema::TypeTagData>::const_iterator I =
6775       MagicValues->find(std::make_pair(ArgumentKind, MagicValue));
6776   if (I == MagicValues->end())
6777     return false;
6778 
6779   TypeInfo = I->second;
6780   return true;
6781 }
6782 } // unnamed namespace
6783 
6784 void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind,
6785                                       uint64_t MagicValue, QualType Type,
6786                                       bool LayoutCompatible,
6787                                       bool MustBeNull) {
6788   if (!TypeTagForDatatypeMagicValues)
6789     TypeTagForDatatypeMagicValues.reset(
6790         new llvm::DenseMap<TypeTagMagicValue, TypeTagData>);
6791 
6792   TypeTagMagicValue Magic(ArgumentKind, MagicValue);
6793   (*TypeTagForDatatypeMagicValues)[Magic] =
6794       TypeTagData(Type, LayoutCompatible, MustBeNull);
6795 }
6796 
6797 namespace {
6798 bool IsSameCharType(QualType T1, QualType T2) {
6799   const BuiltinType *BT1 = T1->getAs<BuiltinType>();
6800   if (!BT1)
6801     return false;
6802 
6803   const BuiltinType *BT2 = T2->getAs<BuiltinType>();
6804   if (!BT2)
6805     return false;
6806 
6807   BuiltinType::Kind T1Kind = BT1->getKind();
6808   BuiltinType::Kind T2Kind = BT2->getKind();
6809 
6810   return (T1Kind == BuiltinType::SChar  && T2Kind == BuiltinType::Char_S) ||
6811          (T1Kind == BuiltinType::UChar  && T2Kind == BuiltinType::Char_U) ||
6812          (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) ||
6813          (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar);
6814 }
6815 } // unnamed namespace
6816 
6817 void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr,
6818                                     const Expr * const *ExprArgs) {
6819   const IdentifierInfo *ArgumentKind = Attr->getArgumentKind();
6820   bool IsPointerAttr = Attr->getIsPointer();
6821 
6822   const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()];
6823   bool FoundWrongKind;
6824   TypeTagData TypeInfo;
6825   if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context,
6826                         TypeTagForDatatypeMagicValues.get(),
6827                         FoundWrongKind, TypeInfo)) {
6828     if (FoundWrongKind)
6829       Diag(TypeTagExpr->getExprLoc(),
6830            diag::warn_type_tag_for_datatype_wrong_kind)
6831         << TypeTagExpr->getSourceRange();
6832     return;
6833   }
6834 
6835   const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()];
6836   if (IsPointerAttr) {
6837     // Skip implicit cast of pointer to `void *' (as a function argument).
6838     if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr))
6839       if (ICE->getType()->isVoidPointerType() &&
6840           ICE->getCastKind() == CK_BitCast)
6841         ArgumentExpr = ICE->getSubExpr();
6842   }
6843   QualType ArgumentType = ArgumentExpr->getType();
6844 
6845   // Passing a `void*' pointer shouldn't trigger a warning.
6846   if (IsPointerAttr && ArgumentType->isVoidPointerType())
6847     return;
6848 
6849   if (TypeInfo.MustBeNull) {
6850     // Type tag with matching void type requires a null pointer.
6851     if (!ArgumentExpr->isNullPointerConstant(Context,
6852                                              Expr::NPC_ValueDependentIsNotNull)) {
6853       Diag(ArgumentExpr->getExprLoc(),
6854            diag::warn_type_safety_null_pointer_required)
6855           << ArgumentKind->getName()
6856           << ArgumentExpr->getSourceRange()
6857           << TypeTagExpr->getSourceRange();
6858     }
6859     return;
6860   }
6861 
6862   QualType RequiredType = TypeInfo.Type;
6863   if (IsPointerAttr)
6864     RequiredType = Context.getPointerType(RequiredType);
6865 
6866   bool mismatch = false;
6867   if (!TypeInfo.LayoutCompatible) {
6868     mismatch = !Context.hasSameType(ArgumentType, RequiredType);
6869 
6870     // C++11 [basic.fundamental] p1:
6871     // Plain char, signed char, and unsigned char are three distinct types.
6872     //
6873     // But we treat plain `char' as equivalent to `signed char' or `unsigned
6874     // char' depending on the current char signedness mode.
6875     if (mismatch)
6876       if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(),
6877                                            RequiredType->getPointeeType())) ||
6878           (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType)))
6879         mismatch = false;
6880   } else
6881     if (IsPointerAttr)
6882       mismatch = !isLayoutCompatible(Context,
6883                                      ArgumentType->getPointeeType(),
6884                                      RequiredType->getPointeeType());
6885     else
6886       mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType);
6887 
6888   if (mismatch)
6889     Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch)
6890         << ArgumentType << ArgumentKind->getName()
6891         << TypeInfo.LayoutCompatible << RequiredType
6892         << ArgumentExpr->getSourceRange()
6893         << TypeTagExpr->getSourceRange();
6894 }
6895