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                                ArrayRef<const Expr *> Args) {
590   VariadicCallType CallType =
591       Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply;
592 
593   checkCall(Method, Args, Method->param_size(),
594             /*IsMemberFunction=*/false,
595             lbrac, Method->getSourceRange(), CallType);
596 
597   return false;
598 }
599 
600 bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall,
601                           const FunctionProtoType *Proto) {
602   const VarDecl *V = dyn_cast<VarDecl>(NDecl);
603   if (!V)
604     return false;
605 
606   QualType Ty = V->getType();
607   if (!Ty->isBlockPointerType())
608     return false;
609 
610   VariadicCallType CallType =
611       Proto && Proto->isVariadic() ? VariadicBlock : VariadicDoesNotApply ;
612   unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0;
613 
614   checkCall(NDecl,
615             llvm::makeArrayRef<const Expr *>(TheCall->getArgs(),
616                                              TheCall->getNumArgs()),
617             NumProtoArgs, /*IsMemberFunction=*/false,
618             TheCall->getRParenLoc(),
619             TheCall->getCallee()->getSourceRange(), CallType);
620 
621   return false;
622 }
623 
624 ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult,
625                                          AtomicExpr::AtomicOp Op) {
626   CallExpr *TheCall = cast<CallExpr>(TheCallResult.get());
627   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
628 
629   // All these operations take one of the following forms:
630   enum {
631     // C    __c11_atomic_init(A *, C)
632     Init,
633     // C    __c11_atomic_load(A *, int)
634     Load,
635     // void __atomic_load(A *, CP, int)
636     Copy,
637     // C    __c11_atomic_add(A *, M, int)
638     Arithmetic,
639     // C    __atomic_exchange_n(A *, CP, int)
640     Xchg,
641     // void __atomic_exchange(A *, C *, CP, int)
642     GNUXchg,
643     // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int)
644     C11CmpXchg,
645     // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int)
646     GNUCmpXchg
647   } Form = Init;
648   const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 };
649   const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 };
650   // where:
651   //   C is an appropriate type,
652   //   A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins,
653   //   CP is C for __c11 builtins and GNU _n builtins and is C * otherwise,
654   //   M is C if C is an integer, and ptrdiff_t if C is a pointer, and
655   //   the int parameters are for orderings.
656 
657   assert(AtomicExpr::AO__c11_atomic_init == 0 &&
658          AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load
659          && "need to update code for modified C11 atomics");
660   bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init &&
661                Op <= AtomicExpr::AO__c11_atomic_fetch_xor;
662   bool IsN = Op == AtomicExpr::AO__atomic_load_n ||
663              Op == AtomicExpr::AO__atomic_store_n ||
664              Op == AtomicExpr::AO__atomic_exchange_n ||
665              Op == AtomicExpr::AO__atomic_compare_exchange_n;
666   bool IsAddSub = false;
667 
668   switch (Op) {
669   case AtomicExpr::AO__c11_atomic_init:
670     Form = Init;
671     break;
672 
673   case AtomicExpr::AO__c11_atomic_load:
674   case AtomicExpr::AO__atomic_load_n:
675     Form = Load;
676     break;
677 
678   case AtomicExpr::AO__c11_atomic_store:
679   case AtomicExpr::AO__atomic_load:
680   case AtomicExpr::AO__atomic_store:
681   case AtomicExpr::AO__atomic_store_n:
682     Form = Copy;
683     break;
684 
685   case AtomicExpr::AO__c11_atomic_fetch_add:
686   case AtomicExpr::AO__c11_atomic_fetch_sub:
687   case AtomicExpr::AO__atomic_fetch_add:
688   case AtomicExpr::AO__atomic_fetch_sub:
689   case AtomicExpr::AO__atomic_add_fetch:
690   case AtomicExpr::AO__atomic_sub_fetch:
691     IsAddSub = true;
692     // Fall through.
693   case AtomicExpr::AO__c11_atomic_fetch_and:
694   case AtomicExpr::AO__c11_atomic_fetch_or:
695   case AtomicExpr::AO__c11_atomic_fetch_xor:
696   case AtomicExpr::AO__atomic_fetch_and:
697   case AtomicExpr::AO__atomic_fetch_or:
698   case AtomicExpr::AO__atomic_fetch_xor:
699   case AtomicExpr::AO__atomic_fetch_nand:
700   case AtomicExpr::AO__atomic_and_fetch:
701   case AtomicExpr::AO__atomic_or_fetch:
702   case AtomicExpr::AO__atomic_xor_fetch:
703   case AtomicExpr::AO__atomic_nand_fetch:
704     Form = Arithmetic;
705     break;
706 
707   case AtomicExpr::AO__c11_atomic_exchange:
708   case AtomicExpr::AO__atomic_exchange_n:
709     Form = Xchg;
710     break;
711 
712   case AtomicExpr::AO__atomic_exchange:
713     Form = GNUXchg;
714     break;
715 
716   case AtomicExpr::AO__c11_atomic_compare_exchange_strong:
717   case AtomicExpr::AO__c11_atomic_compare_exchange_weak:
718     Form = C11CmpXchg;
719     break;
720 
721   case AtomicExpr::AO__atomic_compare_exchange:
722   case AtomicExpr::AO__atomic_compare_exchange_n:
723     Form = GNUCmpXchg;
724     break;
725   }
726 
727   // Check we have the right number of arguments.
728   if (TheCall->getNumArgs() < NumArgs[Form]) {
729     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
730       << 0 << NumArgs[Form] << TheCall->getNumArgs()
731       << TheCall->getCallee()->getSourceRange();
732     return ExprError();
733   } else if (TheCall->getNumArgs() > NumArgs[Form]) {
734     Diag(TheCall->getArg(NumArgs[Form])->getLocStart(),
735          diag::err_typecheck_call_too_many_args)
736       << 0 << NumArgs[Form] << TheCall->getNumArgs()
737       << TheCall->getCallee()->getSourceRange();
738     return ExprError();
739   }
740 
741   // Inspect the first argument of the atomic operation.
742   Expr *Ptr = TheCall->getArg(0);
743   Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get();
744   const PointerType *pointerType = Ptr->getType()->getAs<PointerType>();
745   if (!pointerType) {
746     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
747       << Ptr->getType() << Ptr->getSourceRange();
748     return ExprError();
749   }
750 
751   // For a __c11 builtin, this should be a pointer to an _Atomic type.
752   QualType AtomTy = pointerType->getPointeeType(); // 'A'
753   QualType ValType = AtomTy; // 'C'
754   if (IsC11) {
755     if (!AtomTy->isAtomicType()) {
756       Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic)
757         << Ptr->getType() << Ptr->getSourceRange();
758       return ExprError();
759     }
760     if (AtomTy.isConstQualified()) {
761       Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic)
762         << Ptr->getType() << Ptr->getSourceRange();
763       return ExprError();
764     }
765     ValType = AtomTy->getAs<AtomicType>()->getValueType();
766   }
767 
768   // For an arithmetic operation, the implied arithmetic must be well-formed.
769   if (Form == Arithmetic) {
770     // gcc does not enforce these rules for GNU atomics, but we do so for sanity.
771     if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) {
772       Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr)
773         << IsC11 << Ptr->getType() << Ptr->getSourceRange();
774       return ExprError();
775     }
776     if (!IsAddSub && !ValType->isIntegerType()) {
777       Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int)
778         << IsC11 << Ptr->getType() << Ptr->getSourceRange();
779       return ExprError();
780     }
781   } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) {
782     // For __atomic_*_n operations, the value type must be a scalar integral or
783     // pointer type which is 1, 2, 4, 8 or 16 bytes in length.
784     Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr)
785       << IsC11 << Ptr->getType() << Ptr->getSourceRange();
786     return ExprError();
787   }
788 
789   if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) {
790     // For GNU atomics, require a trivially-copyable type. This is not part of
791     // the GNU atomics specification, but we enforce it for sanity.
792     Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy)
793       << Ptr->getType() << Ptr->getSourceRange();
794     return ExprError();
795   }
796 
797   // FIXME: For any builtin other than a load, the ValType must not be
798   // const-qualified.
799 
800   switch (ValType.getObjCLifetime()) {
801   case Qualifiers::OCL_None:
802   case Qualifiers::OCL_ExplicitNone:
803     // okay
804     break;
805 
806   case Qualifiers::OCL_Weak:
807   case Qualifiers::OCL_Strong:
808   case Qualifiers::OCL_Autoreleasing:
809     // FIXME: Can this happen? By this point, ValType should be known
810     // to be trivially copyable.
811     Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
812       << ValType << Ptr->getSourceRange();
813     return ExprError();
814   }
815 
816   QualType ResultType = ValType;
817   if (Form == Copy || Form == GNUXchg || Form == Init)
818     ResultType = Context.VoidTy;
819   else if (Form == C11CmpXchg || Form == GNUCmpXchg)
820     ResultType = Context.BoolTy;
821 
822   // The type of a parameter passed 'by value'. In the GNU atomics, such
823   // arguments are actually passed as pointers.
824   QualType ByValType = ValType; // 'CP'
825   if (!IsC11 && !IsN)
826     ByValType = Ptr->getType();
827 
828   // The first argument --- the pointer --- has a fixed type; we
829   // deduce the types of the rest of the arguments accordingly.  Walk
830   // the remaining arguments, converting them to the deduced value type.
831   for (unsigned i = 1; i != NumArgs[Form]; ++i) {
832     QualType Ty;
833     if (i < NumVals[Form] + 1) {
834       switch (i) {
835       case 1:
836         // The second argument is the non-atomic operand. For arithmetic, this
837         // is always passed by value, and for a compare_exchange it is always
838         // passed by address. For the rest, GNU uses by-address and C11 uses
839         // by-value.
840         assert(Form != Load);
841         if (Form == Init || (Form == Arithmetic && ValType->isIntegerType()))
842           Ty = ValType;
843         else if (Form == Copy || Form == Xchg)
844           Ty = ByValType;
845         else if (Form == Arithmetic)
846           Ty = Context.getPointerDiffType();
847         else
848           Ty = Context.getPointerType(ValType.getUnqualifiedType());
849         break;
850       case 2:
851         // The third argument to compare_exchange / GNU exchange is a
852         // (pointer to a) desired value.
853         Ty = ByValType;
854         break;
855       case 3:
856         // The fourth argument to GNU compare_exchange is a 'weak' flag.
857         Ty = Context.BoolTy;
858         break;
859       }
860     } else {
861       // The order(s) are always converted to int.
862       Ty = Context.IntTy;
863     }
864 
865     InitializedEntity Entity =
866         InitializedEntity::InitializeParameter(Context, Ty, false);
867     ExprResult Arg = TheCall->getArg(i);
868     Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
869     if (Arg.isInvalid())
870       return true;
871     TheCall->setArg(i, Arg.get());
872   }
873 
874   // Permute the arguments into a 'consistent' order.
875   SmallVector<Expr*, 5> SubExprs;
876   SubExprs.push_back(Ptr);
877   switch (Form) {
878   case Init:
879     // Note, AtomicExpr::getVal1() has a special case for this atomic.
880     SubExprs.push_back(TheCall->getArg(1)); // Val1
881     break;
882   case Load:
883     SubExprs.push_back(TheCall->getArg(1)); // Order
884     break;
885   case Copy:
886   case Arithmetic:
887   case Xchg:
888     SubExprs.push_back(TheCall->getArg(2)); // Order
889     SubExprs.push_back(TheCall->getArg(1)); // Val1
890     break;
891   case GNUXchg:
892     // Note, AtomicExpr::getVal2() has a special case for this atomic.
893     SubExprs.push_back(TheCall->getArg(3)); // Order
894     SubExprs.push_back(TheCall->getArg(1)); // Val1
895     SubExprs.push_back(TheCall->getArg(2)); // Val2
896     break;
897   case C11CmpXchg:
898     SubExprs.push_back(TheCall->getArg(3)); // Order
899     SubExprs.push_back(TheCall->getArg(1)); // Val1
900     SubExprs.push_back(TheCall->getArg(4)); // OrderFail
901     SubExprs.push_back(TheCall->getArg(2)); // Val2
902     break;
903   case GNUCmpXchg:
904     SubExprs.push_back(TheCall->getArg(4)); // Order
905     SubExprs.push_back(TheCall->getArg(1)); // Val1
906     SubExprs.push_back(TheCall->getArg(5)); // OrderFail
907     SubExprs.push_back(TheCall->getArg(2)); // Val2
908     SubExprs.push_back(TheCall->getArg(3)); // Weak
909     break;
910   }
911 
912   return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(),
913                                         SubExprs, ResultType, Op,
914                                         TheCall->getRParenLoc()));
915 }
916 
917 
918 /// checkBuiltinArgument - Given a call to a builtin function, perform
919 /// normal type-checking on the given argument, updating the call in
920 /// place.  This is useful when a builtin function requires custom
921 /// type-checking for some of its arguments but not necessarily all of
922 /// them.
923 ///
924 /// Returns true on error.
925 static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) {
926   FunctionDecl *Fn = E->getDirectCallee();
927   assert(Fn && "builtin call without direct callee!");
928 
929   ParmVarDecl *Param = Fn->getParamDecl(ArgIndex);
930   InitializedEntity Entity =
931     InitializedEntity::InitializeParameter(S.Context, Param);
932 
933   ExprResult Arg = E->getArg(0);
934   Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg);
935   if (Arg.isInvalid())
936     return true;
937 
938   E->setArg(ArgIndex, Arg.take());
939   return false;
940 }
941 
942 /// SemaBuiltinAtomicOverloaded - We have a call to a function like
943 /// __sync_fetch_and_add, which is an overloaded function based on the pointer
944 /// type of its first argument.  The main ActOnCallExpr routines have already
945 /// promoted the types of arguments because all of these calls are prototyped as
946 /// void(...).
947 ///
948 /// This function goes through and does final semantic checking for these
949 /// builtins,
950 ExprResult
951 Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
952   CallExpr *TheCall = (CallExpr *)TheCallResult.get();
953   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
954   FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
955 
956   // Ensure that we have at least one argument to do type inference from.
957   if (TheCall->getNumArgs() < 1) {
958     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
959       << 0 << 1 << TheCall->getNumArgs()
960       << TheCall->getCallee()->getSourceRange();
961     return ExprError();
962   }
963 
964   // Inspect the first argument of the atomic builtin.  This should always be
965   // a pointer type, whose element is an integral scalar or pointer type.
966   // Because it is a pointer type, we don't have to worry about any implicit
967   // casts here.
968   // FIXME: We don't allow floating point scalars as input.
969   Expr *FirstArg = TheCall->getArg(0);
970   ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg);
971   if (FirstArgResult.isInvalid())
972     return ExprError();
973   FirstArg = FirstArgResult.take();
974   TheCall->setArg(0, FirstArg);
975 
976   const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>();
977   if (!pointerType) {
978     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
979       << FirstArg->getType() << FirstArg->getSourceRange();
980     return ExprError();
981   }
982 
983   QualType ValType = pointerType->getPointeeType();
984   if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
985       !ValType->isBlockPointerType()) {
986     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
987       << FirstArg->getType() << FirstArg->getSourceRange();
988     return ExprError();
989   }
990 
991   switch (ValType.getObjCLifetime()) {
992   case Qualifiers::OCL_None:
993   case Qualifiers::OCL_ExplicitNone:
994     // okay
995     break;
996 
997   case Qualifiers::OCL_Weak:
998   case Qualifiers::OCL_Strong:
999   case Qualifiers::OCL_Autoreleasing:
1000     Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
1001       << ValType << FirstArg->getSourceRange();
1002     return ExprError();
1003   }
1004 
1005   // Strip any qualifiers off ValType.
1006   ValType = ValType.getUnqualifiedType();
1007 
1008   // The majority of builtins return a value, but a few have special return
1009   // types, so allow them to override appropriately below.
1010   QualType ResultType = ValType;
1011 
1012   // We need to figure out which concrete builtin this maps onto.  For example,
1013   // __sync_fetch_and_add with a 2 byte object turns into
1014   // __sync_fetch_and_add_2.
1015 #define BUILTIN_ROW(x) \
1016   { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
1017     Builtin::BI##x##_8, Builtin::BI##x##_16 }
1018 
1019   static const unsigned BuiltinIndices[][5] = {
1020     BUILTIN_ROW(__sync_fetch_and_add),
1021     BUILTIN_ROW(__sync_fetch_and_sub),
1022     BUILTIN_ROW(__sync_fetch_and_or),
1023     BUILTIN_ROW(__sync_fetch_and_and),
1024     BUILTIN_ROW(__sync_fetch_and_xor),
1025 
1026     BUILTIN_ROW(__sync_add_and_fetch),
1027     BUILTIN_ROW(__sync_sub_and_fetch),
1028     BUILTIN_ROW(__sync_and_and_fetch),
1029     BUILTIN_ROW(__sync_or_and_fetch),
1030     BUILTIN_ROW(__sync_xor_and_fetch),
1031 
1032     BUILTIN_ROW(__sync_val_compare_and_swap),
1033     BUILTIN_ROW(__sync_bool_compare_and_swap),
1034     BUILTIN_ROW(__sync_lock_test_and_set),
1035     BUILTIN_ROW(__sync_lock_release),
1036     BUILTIN_ROW(__sync_swap)
1037   };
1038 #undef BUILTIN_ROW
1039 
1040   // Determine the index of the size.
1041   unsigned SizeIndex;
1042   switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
1043   case 1: SizeIndex = 0; break;
1044   case 2: SizeIndex = 1; break;
1045   case 4: SizeIndex = 2; break;
1046   case 8: SizeIndex = 3; break;
1047   case 16: SizeIndex = 4; break;
1048   default:
1049     Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
1050       << FirstArg->getType() << FirstArg->getSourceRange();
1051     return ExprError();
1052   }
1053 
1054   // Each of these builtins has one pointer argument, followed by some number of
1055   // values (0, 1 or 2) followed by a potentially empty varags list of stuff
1056   // that we ignore.  Find out which row of BuiltinIndices to read from as well
1057   // as the number of fixed args.
1058   unsigned BuiltinID = FDecl->getBuiltinID();
1059   unsigned BuiltinIndex, NumFixed = 1;
1060   switch (BuiltinID) {
1061   default: llvm_unreachable("Unknown overloaded atomic builtin!");
1062   case Builtin::BI__sync_fetch_and_add:
1063   case Builtin::BI__sync_fetch_and_add_1:
1064   case Builtin::BI__sync_fetch_and_add_2:
1065   case Builtin::BI__sync_fetch_and_add_4:
1066   case Builtin::BI__sync_fetch_and_add_8:
1067   case Builtin::BI__sync_fetch_and_add_16:
1068     BuiltinIndex = 0;
1069     break;
1070 
1071   case Builtin::BI__sync_fetch_and_sub:
1072   case Builtin::BI__sync_fetch_and_sub_1:
1073   case Builtin::BI__sync_fetch_and_sub_2:
1074   case Builtin::BI__sync_fetch_and_sub_4:
1075   case Builtin::BI__sync_fetch_and_sub_8:
1076   case Builtin::BI__sync_fetch_and_sub_16:
1077     BuiltinIndex = 1;
1078     break;
1079 
1080   case Builtin::BI__sync_fetch_and_or:
1081   case Builtin::BI__sync_fetch_and_or_1:
1082   case Builtin::BI__sync_fetch_and_or_2:
1083   case Builtin::BI__sync_fetch_and_or_4:
1084   case Builtin::BI__sync_fetch_and_or_8:
1085   case Builtin::BI__sync_fetch_and_or_16:
1086     BuiltinIndex = 2;
1087     break;
1088 
1089   case Builtin::BI__sync_fetch_and_and:
1090   case Builtin::BI__sync_fetch_and_and_1:
1091   case Builtin::BI__sync_fetch_and_and_2:
1092   case Builtin::BI__sync_fetch_and_and_4:
1093   case Builtin::BI__sync_fetch_and_and_8:
1094   case Builtin::BI__sync_fetch_and_and_16:
1095     BuiltinIndex = 3;
1096     break;
1097 
1098   case Builtin::BI__sync_fetch_and_xor:
1099   case Builtin::BI__sync_fetch_and_xor_1:
1100   case Builtin::BI__sync_fetch_and_xor_2:
1101   case Builtin::BI__sync_fetch_and_xor_4:
1102   case Builtin::BI__sync_fetch_and_xor_8:
1103   case Builtin::BI__sync_fetch_and_xor_16:
1104     BuiltinIndex = 4;
1105     break;
1106 
1107   case Builtin::BI__sync_add_and_fetch:
1108   case Builtin::BI__sync_add_and_fetch_1:
1109   case Builtin::BI__sync_add_and_fetch_2:
1110   case Builtin::BI__sync_add_and_fetch_4:
1111   case Builtin::BI__sync_add_and_fetch_8:
1112   case Builtin::BI__sync_add_and_fetch_16:
1113     BuiltinIndex = 5;
1114     break;
1115 
1116   case Builtin::BI__sync_sub_and_fetch:
1117   case Builtin::BI__sync_sub_and_fetch_1:
1118   case Builtin::BI__sync_sub_and_fetch_2:
1119   case Builtin::BI__sync_sub_and_fetch_4:
1120   case Builtin::BI__sync_sub_and_fetch_8:
1121   case Builtin::BI__sync_sub_and_fetch_16:
1122     BuiltinIndex = 6;
1123     break;
1124 
1125   case Builtin::BI__sync_and_and_fetch:
1126   case Builtin::BI__sync_and_and_fetch_1:
1127   case Builtin::BI__sync_and_and_fetch_2:
1128   case Builtin::BI__sync_and_and_fetch_4:
1129   case Builtin::BI__sync_and_and_fetch_8:
1130   case Builtin::BI__sync_and_and_fetch_16:
1131     BuiltinIndex = 7;
1132     break;
1133 
1134   case Builtin::BI__sync_or_and_fetch:
1135   case Builtin::BI__sync_or_and_fetch_1:
1136   case Builtin::BI__sync_or_and_fetch_2:
1137   case Builtin::BI__sync_or_and_fetch_4:
1138   case Builtin::BI__sync_or_and_fetch_8:
1139   case Builtin::BI__sync_or_and_fetch_16:
1140     BuiltinIndex = 8;
1141     break;
1142 
1143   case Builtin::BI__sync_xor_and_fetch:
1144   case Builtin::BI__sync_xor_and_fetch_1:
1145   case Builtin::BI__sync_xor_and_fetch_2:
1146   case Builtin::BI__sync_xor_and_fetch_4:
1147   case Builtin::BI__sync_xor_and_fetch_8:
1148   case Builtin::BI__sync_xor_and_fetch_16:
1149     BuiltinIndex = 9;
1150     break;
1151 
1152   case Builtin::BI__sync_val_compare_and_swap:
1153   case Builtin::BI__sync_val_compare_and_swap_1:
1154   case Builtin::BI__sync_val_compare_and_swap_2:
1155   case Builtin::BI__sync_val_compare_and_swap_4:
1156   case Builtin::BI__sync_val_compare_and_swap_8:
1157   case Builtin::BI__sync_val_compare_and_swap_16:
1158     BuiltinIndex = 10;
1159     NumFixed = 2;
1160     break;
1161 
1162   case Builtin::BI__sync_bool_compare_and_swap:
1163   case Builtin::BI__sync_bool_compare_and_swap_1:
1164   case Builtin::BI__sync_bool_compare_and_swap_2:
1165   case Builtin::BI__sync_bool_compare_and_swap_4:
1166   case Builtin::BI__sync_bool_compare_and_swap_8:
1167   case Builtin::BI__sync_bool_compare_and_swap_16:
1168     BuiltinIndex = 11;
1169     NumFixed = 2;
1170     ResultType = Context.BoolTy;
1171     break;
1172 
1173   case Builtin::BI__sync_lock_test_and_set:
1174   case Builtin::BI__sync_lock_test_and_set_1:
1175   case Builtin::BI__sync_lock_test_and_set_2:
1176   case Builtin::BI__sync_lock_test_and_set_4:
1177   case Builtin::BI__sync_lock_test_and_set_8:
1178   case Builtin::BI__sync_lock_test_and_set_16:
1179     BuiltinIndex = 12;
1180     break;
1181 
1182   case Builtin::BI__sync_lock_release:
1183   case Builtin::BI__sync_lock_release_1:
1184   case Builtin::BI__sync_lock_release_2:
1185   case Builtin::BI__sync_lock_release_4:
1186   case Builtin::BI__sync_lock_release_8:
1187   case Builtin::BI__sync_lock_release_16:
1188     BuiltinIndex = 13;
1189     NumFixed = 0;
1190     ResultType = Context.VoidTy;
1191     break;
1192 
1193   case Builtin::BI__sync_swap:
1194   case Builtin::BI__sync_swap_1:
1195   case Builtin::BI__sync_swap_2:
1196   case Builtin::BI__sync_swap_4:
1197   case Builtin::BI__sync_swap_8:
1198   case Builtin::BI__sync_swap_16:
1199     BuiltinIndex = 14;
1200     break;
1201   }
1202 
1203   // Now that we know how many fixed arguments we expect, first check that we
1204   // have at least that many.
1205   if (TheCall->getNumArgs() < 1+NumFixed) {
1206     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
1207       << 0 << 1+NumFixed << TheCall->getNumArgs()
1208       << TheCall->getCallee()->getSourceRange();
1209     return ExprError();
1210   }
1211 
1212   // Get the decl for the concrete builtin from this, we can tell what the
1213   // concrete integer type we should convert to is.
1214   unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
1215   const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
1216   FunctionDecl *NewBuiltinDecl;
1217   if (NewBuiltinID == BuiltinID)
1218     NewBuiltinDecl = FDecl;
1219   else {
1220     // Perform builtin lookup to avoid redeclaring it.
1221     DeclarationName DN(&Context.Idents.get(NewBuiltinName));
1222     LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName);
1223     LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true);
1224     assert(Res.getFoundDecl());
1225     NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl());
1226     if (NewBuiltinDecl == 0)
1227       return ExprError();
1228   }
1229 
1230   // The first argument --- the pointer --- has a fixed type; we
1231   // deduce the types of the rest of the arguments accordingly.  Walk
1232   // the remaining arguments, converting them to the deduced value type.
1233   for (unsigned i = 0; i != NumFixed; ++i) {
1234     ExprResult Arg = TheCall->getArg(i+1);
1235 
1236     // GCC does an implicit conversion to the pointer or integer ValType.  This
1237     // can fail in some cases (1i -> int**), check for this error case now.
1238     // Initialize the argument.
1239     InitializedEntity Entity = InitializedEntity::InitializeParameter(Context,
1240                                                    ValType, /*consume*/ false);
1241     Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
1242     if (Arg.isInvalid())
1243       return ExprError();
1244 
1245     // Okay, we have something that *can* be converted to the right type.  Check
1246     // to see if there is a potentially weird extension going on here.  This can
1247     // happen when you do an atomic operation on something like an char* and
1248     // pass in 42.  The 42 gets converted to char.  This is even more strange
1249     // for things like 45.123 -> char, etc.
1250     // FIXME: Do this check.
1251     TheCall->setArg(i+1, Arg.take());
1252   }
1253 
1254   ASTContext& Context = this->getASTContext();
1255 
1256   // Create a new DeclRefExpr to refer to the new decl.
1257   DeclRefExpr* NewDRE = DeclRefExpr::Create(
1258       Context,
1259       DRE->getQualifierLoc(),
1260       SourceLocation(),
1261       NewBuiltinDecl,
1262       /*enclosing*/ false,
1263       DRE->getLocation(),
1264       Context.BuiltinFnTy,
1265       DRE->getValueKind());
1266 
1267   // Set the callee in the CallExpr.
1268   // FIXME: This loses syntactic information.
1269   QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType());
1270   ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy,
1271                                               CK_BuiltinFnToFnPtr);
1272   TheCall->setCallee(PromotedCall.take());
1273 
1274   // Change the result type of the call to match the original value type. This
1275   // is arbitrary, but the codegen for these builtins ins design to handle it
1276   // gracefully.
1277   TheCall->setType(ResultType);
1278 
1279   return TheCallResult;
1280 }
1281 
1282 /// CheckObjCString - Checks that the argument to the builtin
1283 /// CFString constructor is correct
1284 /// Note: It might also make sense to do the UTF-16 conversion here (would
1285 /// simplify the backend).
1286 bool Sema::CheckObjCString(Expr *Arg) {
1287   Arg = Arg->IgnoreParenCasts();
1288   StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
1289 
1290   if (!Literal || !Literal->isAscii()) {
1291     Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
1292       << Arg->getSourceRange();
1293     return true;
1294   }
1295 
1296   if (Literal->containsNonAsciiOrNull()) {
1297     StringRef String = Literal->getString();
1298     unsigned NumBytes = String.size();
1299     SmallVector<UTF16, 128> ToBuf(NumBytes);
1300     const UTF8 *FromPtr = (const UTF8 *)String.data();
1301     UTF16 *ToPtr = &ToBuf[0];
1302 
1303     ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
1304                                                  &ToPtr, ToPtr + NumBytes,
1305                                                  strictConversion);
1306     // Check for conversion failure.
1307     if (Result != conversionOK)
1308       Diag(Arg->getLocStart(),
1309            diag::warn_cfstring_truncated) << Arg->getSourceRange();
1310   }
1311   return false;
1312 }
1313 
1314 /// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
1315 /// Emit an error and return true on failure, return false on success.
1316 bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
1317   Expr *Fn = TheCall->getCallee();
1318   if (TheCall->getNumArgs() > 2) {
1319     Diag(TheCall->getArg(2)->getLocStart(),
1320          diag::err_typecheck_call_too_many_args)
1321       << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1322       << Fn->getSourceRange()
1323       << SourceRange(TheCall->getArg(2)->getLocStart(),
1324                      (*(TheCall->arg_end()-1))->getLocEnd());
1325     return true;
1326   }
1327 
1328   if (TheCall->getNumArgs() < 2) {
1329     return Diag(TheCall->getLocEnd(),
1330       diag::err_typecheck_call_too_few_args_at_least)
1331       << 0 /*function call*/ << 2 << TheCall->getNumArgs();
1332   }
1333 
1334   // Type-check the first argument normally.
1335   if (checkBuiltinArgument(*this, TheCall, 0))
1336     return true;
1337 
1338   // Determine whether the current function is variadic or not.
1339   BlockScopeInfo *CurBlock = getCurBlock();
1340   bool isVariadic;
1341   if (CurBlock)
1342     isVariadic = CurBlock->TheDecl->isVariadic();
1343   else if (FunctionDecl *FD = getCurFunctionDecl())
1344     isVariadic = FD->isVariadic();
1345   else
1346     isVariadic = getCurMethodDecl()->isVariadic();
1347 
1348   if (!isVariadic) {
1349     Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
1350     return true;
1351   }
1352 
1353   // Verify that the second argument to the builtin is the last argument of the
1354   // current function or method.
1355   bool SecondArgIsLastNamedArgument = false;
1356   const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
1357 
1358   if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
1359     if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
1360       // FIXME: This isn't correct for methods (results in bogus warning).
1361       // Get the last formal in the current function.
1362       const ParmVarDecl *LastArg;
1363       if (CurBlock)
1364         LastArg = *(CurBlock->TheDecl->param_end()-1);
1365       else if (FunctionDecl *FD = getCurFunctionDecl())
1366         LastArg = *(FD->param_end()-1);
1367       else
1368         LastArg = *(getCurMethodDecl()->param_end()-1);
1369       SecondArgIsLastNamedArgument = PV == LastArg;
1370     }
1371   }
1372 
1373   if (!SecondArgIsLastNamedArgument)
1374     Diag(TheCall->getArg(1)->getLocStart(),
1375          diag::warn_second_parameter_of_va_start_not_last_named_argument);
1376   return false;
1377 }
1378 
1379 /// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
1380 /// friends.  This is declared to take (...), so we have to check everything.
1381 bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
1382   if (TheCall->getNumArgs() < 2)
1383     return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1384       << 0 << 2 << TheCall->getNumArgs()/*function call*/;
1385   if (TheCall->getNumArgs() > 2)
1386     return Diag(TheCall->getArg(2)->getLocStart(),
1387                 diag::err_typecheck_call_too_many_args)
1388       << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1389       << SourceRange(TheCall->getArg(2)->getLocStart(),
1390                      (*(TheCall->arg_end()-1))->getLocEnd());
1391 
1392   ExprResult OrigArg0 = TheCall->getArg(0);
1393   ExprResult OrigArg1 = TheCall->getArg(1);
1394 
1395   // Do standard promotions between the two arguments, returning their common
1396   // type.
1397   QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
1398   if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
1399     return true;
1400 
1401   // Make sure any conversions are pushed back into the call; this is
1402   // type safe since unordered compare builtins are declared as "_Bool
1403   // foo(...)".
1404   TheCall->setArg(0, OrigArg0.get());
1405   TheCall->setArg(1, OrigArg1.get());
1406 
1407   if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
1408     return false;
1409 
1410   // If the common type isn't a real floating type, then the arguments were
1411   // invalid for this operation.
1412   if (Res.isNull() || !Res->isRealFloatingType())
1413     return Diag(OrigArg0.get()->getLocStart(),
1414                 diag::err_typecheck_call_invalid_ordered_compare)
1415       << OrigArg0.get()->getType() << OrigArg1.get()->getType()
1416       << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
1417 
1418   return false;
1419 }
1420 
1421 /// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
1422 /// __builtin_isnan and friends.  This is declared to take (...), so we have
1423 /// to check everything. We expect the last argument to be a floating point
1424 /// value.
1425 bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
1426   if (TheCall->getNumArgs() < NumArgs)
1427     return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1428       << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
1429   if (TheCall->getNumArgs() > NumArgs)
1430     return Diag(TheCall->getArg(NumArgs)->getLocStart(),
1431                 diag::err_typecheck_call_too_many_args)
1432       << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
1433       << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
1434                      (*(TheCall->arg_end()-1))->getLocEnd());
1435 
1436   Expr *OrigArg = TheCall->getArg(NumArgs-1);
1437 
1438   if (OrigArg->isTypeDependent())
1439     return false;
1440 
1441   // This operation requires a non-_Complex floating-point number.
1442   if (!OrigArg->getType()->isRealFloatingType())
1443     return Diag(OrigArg->getLocStart(),
1444                 diag::err_typecheck_call_invalid_unary_fp)
1445       << OrigArg->getType() << OrigArg->getSourceRange();
1446 
1447   // If this is an implicit conversion from float -> double, remove it.
1448   if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
1449     Expr *CastArg = Cast->getSubExpr();
1450     if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
1451       assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
1452              "promotion from float to double is the only expected cast here");
1453       Cast->setSubExpr(0);
1454       TheCall->setArg(NumArgs-1, CastArg);
1455     }
1456   }
1457 
1458   return false;
1459 }
1460 
1461 /// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
1462 // This is declared to take (...), so we have to check everything.
1463 ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
1464   if (TheCall->getNumArgs() < 2)
1465     return ExprError(Diag(TheCall->getLocEnd(),
1466                           diag::err_typecheck_call_too_few_args_at_least)
1467       << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1468       << TheCall->getSourceRange());
1469 
1470   // Determine which of the following types of shufflevector we're checking:
1471   // 1) unary, vector mask: (lhs, mask)
1472   // 2) binary, vector mask: (lhs, rhs, mask)
1473   // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
1474   QualType resType = TheCall->getArg(0)->getType();
1475   unsigned numElements = 0;
1476 
1477   if (!TheCall->getArg(0)->isTypeDependent() &&
1478       !TheCall->getArg(1)->isTypeDependent()) {
1479     QualType LHSType = TheCall->getArg(0)->getType();
1480     QualType RHSType = TheCall->getArg(1)->getType();
1481 
1482     if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
1483       Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
1484         << SourceRange(TheCall->getArg(0)->getLocStart(),
1485                        TheCall->getArg(1)->getLocEnd());
1486       return ExprError();
1487     }
1488 
1489     numElements = LHSType->getAs<VectorType>()->getNumElements();
1490     unsigned numResElements = TheCall->getNumArgs() - 2;
1491 
1492     // Check to see if we have a call with 2 vector arguments, the unary shuffle
1493     // with mask.  If so, verify that RHS is an integer vector type with the
1494     // same number of elts as lhs.
1495     if (TheCall->getNumArgs() == 2) {
1496       if (!RHSType->hasIntegerRepresentation() ||
1497           RHSType->getAs<VectorType>()->getNumElements() != numElements)
1498         Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1499           << SourceRange(TheCall->getArg(1)->getLocStart(),
1500                          TheCall->getArg(1)->getLocEnd());
1501       numResElements = numElements;
1502     }
1503     else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
1504       Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1505         << SourceRange(TheCall->getArg(0)->getLocStart(),
1506                        TheCall->getArg(1)->getLocEnd());
1507       return ExprError();
1508     } else if (numElements != numResElements) {
1509       QualType eltType = LHSType->getAs<VectorType>()->getElementType();
1510       resType = Context.getVectorType(eltType, numResElements,
1511                                       VectorType::GenericVector);
1512     }
1513   }
1514 
1515   for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
1516     if (TheCall->getArg(i)->isTypeDependent() ||
1517         TheCall->getArg(i)->isValueDependent())
1518       continue;
1519 
1520     llvm::APSInt Result(32);
1521     if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
1522       return ExprError(Diag(TheCall->getLocStart(),
1523                   diag::err_shufflevector_nonconstant_argument)
1524                 << TheCall->getArg(i)->getSourceRange());
1525 
1526     if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
1527       return ExprError(Diag(TheCall->getLocStart(),
1528                   diag::err_shufflevector_argument_too_large)
1529                << TheCall->getArg(i)->getSourceRange());
1530   }
1531 
1532   SmallVector<Expr*, 32> exprs;
1533 
1534   for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
1535     exprs.push_back(TheCall->getArg(i));
1536     TheCall->setArg(i, 0);
1537   }
1538 
1539   return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType,
1540                                             TheCall->getCallee()->getLocStart(),
1541                                             TheCall->getRParenLoc()));
1542 }
1543 
1544 /// SemaBuiltinPrefetch - Handle __builtin_prefetch.
1545 // This is declared to take (const void*, ...) and can take two
1546 // optional constant int args.
1547 bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
1548   unsigned NumArgs = TheCall->getNumArgs();
1549 
1550   if (NumArgs > 3)
1551     return Diag(TheCall->getLocEnd(),
1552              diag::err_typecheck_call_too_many_args_at_most)
1553              << 0 /*function call*/ << 3 << NumArgs
1554              << TheCall->getSourceRange();
1555 
1556   // Argument 0 is checked for us and the remaining arguments must be
1557   // constant integers.
1558   for (unsigned i = 1; i != NumArgs; ++i) {
1559     Expr *Arg = TheCall->getArg(i);
1560 
1561     // We can't check the value of a dependent argument.
1562     if (Arg->isTypeDependent() || Arg->isValueDependent())
1563       continue;
1564 
1565     llvm::APSInt Result;
1566     if (SemaBuiltinConstantArg(TheCall, i, Result))
1567       return true;
1568 
1569     // FIXME: gcc issues a warning and rewrites these to 0. These
1570     // seems especially odd for the third argument since the default
1571     // is 3.
1572     if (i == 1) {
1573       if (Result.getLimitedValue() > 1)
1574         return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1575              << "0" << "1" << Arg->getSourceRange();
1576     } else {
1577       if (Result.getLimitedValue() > 3)
1578         return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1579             << "0" << "3" << Arg->getSourceRange();
1580     }
1581   }
1582 
1583   return false;
1584 }
1585 
1586 /// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
1587 /// TheCall is a constant expression.
1588 bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
1589                                   llvm::APSInt &Result) {
1590   Expr *Arg = TheCall->getArg(ArgNum);
1591   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
1592   FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
1593 
1594   if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
1595 
1596   if (!Arg->isIntegerConstantExpr(Result, Context))
1597     return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
1598                 << FDecl->getDeclName() <<  Arg->getSourceRange();
1599 
1600   return false;
1601 }
1602 
1603 /// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
1604 /// int type). This simply type checks that type is one of the defined
1605 /// constants (0-3).
1606 // For compatibility check 0-3, llvm only handles 0 and 2.
1607 bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
1608   llvm::APSInt Result;
1609 
1610   // We can't check the value of a dependent argument.
1611   if (TheCall->getArg(1)->isTypeDependent() ||
1612       TheCall->getArg(1)->isValueDependent())
1613     return false;
1614 
1615   // Check constant-ness first.
1616   if (SemaBuiltinConstantArg(TheCall, 1, Result))
1617     return true;
1618 
1619   Expr *Arg = TheCall->getArg(1);
1620   if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
1621     return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1622              << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1623   }
1624 
1625   return false;
1626 }
1627 
1628 /// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
1629 /// This checks that val is a constant 1.
1630 bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
1631   Expr *Arg = TheCall->getArg(1);
1632   llvm::APSInt Result;
1633 
1634   // TODO: This is less than ideal. Overload this to take a value.
1635   if (SemaBuiltinConstantArg(TheCall, 1, Result))
1636     return true;
1637 
1638   if (Result != 1)
1639     return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
1640              << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1641 
1642   return false;
1643 }
1644 
1645 // Determine if an expression is a string literal or constant string.
1646 // If this function returns false on the arguments to a function expecting a
1647 // format string, we will usually need to emit a warning.
1648 // True string literals are then checked by CheckFormatString.
1649 Sema::StringLiteralCheckType
1650 Sema::checkFormatStringExpr(const Expr *E, ArrayRef<const Expr *> Args,
1651                             bool HasVAListArg,
1652                             unsigned format_idx, unsigned firstDataArg,
1653                             FormatStringType Type, VariadicCallType CallType,
1654                             bool inFunctionCall) {
1655  tryAgain:
1656   if (E->isTypeDependent() || E->isValueDependent())
1657     return SLCT_NotALiteral;
1658 
1659   E = E->IgnoreParenCasts();
1660 
1661   if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull))
1662     // Technically -Wformat-nonliteral does not warn about this case.
1663     // The behavior of printf and friends in this case is implementation
1664     // dependent.  Ideally if the format string cannot be null then
1665     // it should have a 'nonnull' attribute in the function prototype.
1666     return SLCT_CheckedLiteral;
1667 
1668   switch (E->getStmtClass()) {
1669   case Stmt::BinaryConditionalOperatorClass:
1670   case Stmt::ConditionalOperatorClass: {
1671     // The expression is a literal if both sub-expressions were, and it was
1672     // completely checked only if both sub-expressions were checked.
1673     const AbstractConditionalOperator *C =
1674         cast<AbstractConditionalOperator>(E);
1675     StringLiteralCheckType Left =
1676         checkFormatStringExpr(C->getTrueExpr(), Args,
1677                               HasVAListArg, format_idx, firstDataArg,
1678                               Type, CallType, inFunctionCall);
1679     if (Left == SLCT_NotALiteral)
1680       return SLCT_NotALiteral;
1681     StringLiteralCheckType Right =
1682         checkFormatStringExpr(C->getFalseExpr(), Args,
1683                               HasVAListArg, format_idx, firstDataArg,
1684                               Type, CallType, inFunctionCall);
1685     return Left < Right ? Left : Right;
1686   }
1687 
1688   case Stmt::ImplicitCastExprClass: {
1689     E = cast<ImplicitCastExpr>(E)->getSubExpr();
1690     goto tryAgain;
1691   }
1692 
1693   case Stmt::OpaqueValueExprClass:
1694     if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
1695       E = src;
1696       goto tryAgain;
1697     }
1698     return SLCT_NotALiteral;
1699 
1700   case Stmt::PredefinedExprClass:
1701     // While __func__, etc., are technically not string literals, they
1702     // cannot contain format specifiers and thus are not a security
1703     // liability.
1704     return SLCT_UncheckedLiteral;
1705 
1706   case Stmt::DeclRefExprClass: {
1707     const DeclRefExpr *DR = cast<DeclRefExpr>(E);
1708 
1709     // As an exception, do not flag errors for variables binding to
1710     // const string literals.
1711     if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
1712       bool isConstant = false;
1713       QualType T = DR->getType();
1714 
1715       if (const ArrayType *AT = Context.getAsArrayType(T)) {
1716         isConstant = AT->getElementType().isConstant(Context);
1717       } else if (const PointerType *PT = T->getAs<PointerType>()) {
1718         isConstant = T.isConstant(Context) &&
1719                      PT->getPointeeType().isConstant(Context);
1720       } else if (T->isObjCObjectPointerType()) {
1721         // In ObjC, there is usually no "const ObjectPointer" type,
1722         // so don't check if the pointee type is constant.
1723         isConstant = T.isConstant(Context);
1724       }
1725 
1726       if (isConstant) {
1727         if (const Expr *Init = VD->getAnyInitializer()) {
1728           // Look through initializers like const char c[] = { "foo" }
1729           if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) {
1730             if (InitList->isStringLiteralInit())
1731               Init = InitList->getInit(0)->IgnoreParenImpCasts();
1732           }
1733           return checkFormatStringExpr(Init, Args,
1734                                        HasVAListArg, format_idx,
1735                                        firstDataArg, Type, CallType,
1736                                        /*inFunctionCall*/false);
1737         }
1738       }
1739 
1740       // For vprintf* functions (i.e., HasVAListArg==true), we add a
1741       // special check to see if the format string is a function parameter
1742       // of the function calling the printf function.  If the function
1743       // has an attribute indicating it is a printf-like function, then we
1744       // should suppress warnings concerning non-literals being used in a call
1745       // to a vprintf function.  For example:
1746       //
1747       // void
1748       // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
1749       //      va_list ap;
1750       //      va_start(ap, fmt);
1751       //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
1752       //      ...
1753       //
1754       if (HasVAListArg) {
1755         if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) {
1756           if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) {
1757             int PVIndex = PV->getFunctionScopeIndex() + 1;
1758             for (specific_attr_iterator<FormatAttr>
1759                  i = ND->specific_attr_begin<FormatAttr>(),
1760                  e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) {
1761               FormatAttr *PVFormat = *i;
1762               // adjust for implicit parameter
1763               if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND))
1764                 if (MD->isInstance())
1765                   ++PVIndex;
1766               // We also check if the formats are compatible.
1767               // We can't pass a 'scanf' string to a 'printf' function.
1768               if (PVIndex == PVFormat->getFormatIdx() &&
1769                   Type == GetFormatStringType(PVFormat))
1770                 return SLCT_UncheckedLiteral;
1771             }
1772           }
1773         }
1774       }
1775     }
1776 
1777     return SLCT_NotALiteral;
1778   }
1779 
1780   case Stmt::CallExprClass:
1781   case Stmt::CXXMemberCallExprClass: {
1782     const CallExpr *CE = cast<CallExpr>(E);
1783     if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) {
1784       if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) {
1785         unsigned ArgIndex = FA->getFormatIdx();
1786         if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND))
1787           if (MD->isInstance())
1788             --ArgIndex;
1789         const Expr *Arg = CE->getArg(ArgIndex - 1);
1790 
1791         return checkFormatStringExpr(Arg, Args,
1792                                      HasVAListArg, format_idx, firstDataArg,
1793                                      Type, CallType, inFunctionCall);
1794       } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) {
1795         unsigned BuiltinID = FD->getBuiltinID();
1796         if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString ||
1797             BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) {
1798           const Expr *Arg = CE->getArg(0);
1799           return checkFormatStringExpr(Arg, Args,
1800                                        HasVAListArg, format_idx,
1801                                        firstDataArg, Type, CallType,
1802                                        inFunctionCall);
1803         }
1804       }
1805     }
1806 
1807     return SLCT_NotALiteral;
1808   }
1809   case Stmt::ObjCStringLiteralClass:
1810   case Stmt::StringLiteralClass: {
1811     const StringLiteral *StrE = NULL;
1812 
1813     if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1814       StrE = ObjCFExpr->getString();
1815     else
1816       StrE = cast<StringLiteral>(E);
1817 
1818     if (StrE) {
1819       CheckFormatString(StrE, E, Args, HasVAListArg, format_idx,
1820                         firstDataArg, Type, inFunctionCall, CallType);
1821       return SLCT_CheckedLiteral;
1822     }
1823 
1824     return SLCT_NotALiteral;
1825   }
1826 
1827   default:
1828     return SLCT_NotALiteral;
1829   }
1830 }
1831 
1832 void
1833 Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1834                             const Expr * const *ExprArgs,
1835                             SourceLocation CallSiteLoc) {
1836   for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1837                                   e = NonNull->args_end();
1838        i != e; ++i) {
1839     const Expr *ArgExpr = ExprArgs[*i];
1840 
1841     // As a special case, transparent unions initialized with zero are
1842     // considered null for the purposes of the nonnull attribute.
1843     if (const RecordType *UT = ArgExpr->getType()->getAsUnionType()) {
1844       if (UT->getDecl()->hasAttr<TransparentUnionAttr>())
1845         if (const CompoundLiteralExpr *CLE =
1846             dyn_cast<CompoundLiteralExpr>(ArgExpr))
1847           if (const InitListExpr *ILE =
1848               dyn_cast<InitListExpr>(CLE->getInitializer()))
1849             ArgExpr = ILE->getInit(0);
1850     }
1851 
1852     bool Result;
1853     if (ArgExpr->EvaluateAsBooleanCondition(Result, Context) && !Result)
1854       Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1855   }
1856 }
1857 
1858 Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) {
1859   return llvm::StringSwitch<FormatStringType>(Format->getType())
1860   .Case("scanf", FST_Scanf)
1861   .Cases("printf", "printf0", FST_Printf)
1862   .Cases("NSString", "CFString", FST_NSString)
1863   .Case("strftime", FST_Strftime)
1864   .Case("strfmon", FST_Strfmon)
1865   .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf)
1866   .Default(FST_Unknown);
1867 }
1868 
1869 /// CheckFormatArguments - Check calls to printf and scanf (and similar
1870 /// functions) for correct use of format strings.
1871 /// Returns true if a format string has been fully checked.
1872 bool Sema::CheckFormatArguments(const FormatAttr *Format,
1873                                 ArrayRef<const Expr *> Args,
1874                                 bool IsCXXMember,
1875                                 VariadicCallType CallType,
1876                                 SourceLocation Loc, SourceRange Range) {
1877   FormatStringInfo FSI;
1878   if (getFormatStringInfo(Format, IsCXXMember, &FSI))
1879     return CheckFormatArguments(Args, FSI.HasVAListArg, FSI.FormatIdx,
1880                                 FSI.FirstDataArg, GetFormatStringType(Format),
1881                                 CallType, Loc, Range);
1882   return false;
1883 }
1884 
1885 bool Sema::CheckFormatArguments(ArrayRef<const Expr *> Args,
1886                                 bool HasVAListArg, unsigned format_idx,
1887                                 unsigned firstDataArg, FormatStringType Type,
1888                                 VariadicCallType CallType,
1889                                 SourceLocation Loc, SourceRange Range) {
1890   // CHECK: printf/scanf-like function is called with no format string.
1891   if (format_idx >= Args.size()) {
1892     Diag(Loc, diag::warn_missing_format_string) << Range;
1893     return false;
1894   }
1895 
1896   const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts();
1897 
1898   // CHECK: format string is not a string literal.
1899   //
1900   // Dynamically generated format strings are difficult to
1901   // automatically vet at compile time.  Requiring that format strings
1902   // are string literals: (1) permits the checking of format strings by
1903   // the compiler and thereby (2) can practically remove the source of
1904   // many format string exploits.
1905 
1906   // Format string can be either ObjC string (e.g. @"%d") or
1907   // C string (e.g. "%d")
1908   // ObjC string uses the same format specifiers as C string, so we can use
1909   // the same format string checking logic for both ObjC and C strings.
1910   StringLiteralCheckType CT =
1911       checkFormatStringExpr(OrigFormatExpr, Args, HasVAListArg,
1912                             format_idx, firstDataArg, Type, CallType);
1913   if (CT != SLCT_NotALiteral)
1914     // Literal format string found, check done!
1915     return CT == SLCT_CheckedLiteral;
1916 
1917   // Strftime is particular as it always uses a single 'time' argument,
1918   // so it is safe to pass a non-literal string.
1919   if (Type == FST_Strftime)
1920     return false;
1921 
1922   // Do not emit diag when the string param is a macro expansion and the
1923   // format is either NSString or CFString. This is a hack to prevent
1924   // diag when using the NSLocalizedString and CFCopyLocalizedString macros
1925   // which are usually used in place of NS and CF string literals.
1926   if (Type == FST_NSString &&
1927       SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart()))
1928     return false;
1929 
1930   // If there are no arguments specified, warn with -Wformat-security, otherwise
1931   // warn only with -Wformat-nonliteral.
1932   if (Args.size() == format_idx+1)
1933     Diag(Args[format_idx]->getLocStart(),
1934          diag::warn_format_nonliteral_noargs)
1935       << OrigFormatExpr->getSourceRange();
1936   else
1937     Diag(Args[format_idx]->getLocStart(),
1938          diag::warn_format_nonliteral)
1939            << OrigFormatExpr->getSourceRange();
1940   return false;
1941 }
1942 
1943 namespace {
1944 class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1945 protected:
1946   Sema &S;
1947   const StringLiteral *FExpr;
1948   const Expr *OrigFormatExpr;
1949   const unsigned FirstDataArg;
1950   const unsigned NumDataArgs;
1951   const char *Beg; // Start of format string.
1952   const bool HasVAListArg;
1953   ArrayRef<const Expr *> Args;
1954   unsigned FormatIdx;
1955   llvm::BitVector CoveredArgs;
1956   bool usesPositionalArgs;
1957   bool atFirstArg;
1958   bool inFunctionCall;
1959   Sema::VariadicCallType CallType;
1960 public:
1961   CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1962                      const Expr *origFormatExpr, unsigned firstDataArg,
1963                      unsigned numDataArgs, const char *beg, bool hasVAListArg,
1964                      ArrayRef<const Expr *> Args,
1965                      unsigned formatIdx, bool inFunctionCall,
1966                      Sema::VariadicCallType callType)
1967     : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1968       FirstDataArg(firstDataArg), NumDataArgs(numDataArgs),
1969       Beg(beg), HasVAListArg(hasVAListArg),
1970       Args(Args), FormatIdx(formatIdx),
1971       usesPositionalArgs(false), atFirstArg(true),
1972       inFunctionCall(inFunctionCall), CallType(callType) {
1973         CoveredArgs.resize(numDataArgs);
1974         CoveredArgs.reset();
1975       }
1976 
1977   void DoneProcessing();
1978 
1979   void HandleIncompleteSpecifier(const char *startSpecifier,
1980                                  unsigned specifierLen);
1981 
1982   void HandleInvalidLengthModifier(
1983       const analyze_format_string::FormatSpecifier &FS,
1984       const analyze_format_string::ConversionSpecifier &CS,
1985       const char *startSpecifier, unsigned specifierLen, unsigned DiagID);
1986 
1987   void HandleNonStandardLengthModifier(
1988       const analyze_format_string::FormatSpecifier &FS,
1989       const char *startSpecifier, unsigned specifierLen);
1990 
1991   void HandleNonStandardConversionSpecifier(
1992       const analyze_format_string::ConversionSpecifier &CS,
1993       const char *startSpecifier, unsigned specifierLen);
1994 
1995   virtual void HandlePosition(const char *startPos, unsigned posLen);
1996 
1997   virtual void HandleInvalidPosition(const char *startSpecifier,
1998                                      unsigned specifierLen,
1999                                      analyze_format_string::PositionContext p);
2000 
2001   virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
2002 
2003   void HandleNullChar(const char *nullCharacter);
2004 
2005   template <typename Range>
2006   static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall,
2007                                    const Expr *ArgumentExpr,
2008                                    PartialDiagnostic PDiag,
2009                                    SourceLocation StringLoc,
2010                                    bool IsStringLocation, Range StringRange,
2011                                    ArrayRef<FixItHint> Fixit = None);
2012 
2013 protected:
2014   bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
2015                                         const char *startSpec,
2016                                         unsigned specifierLen,
2017                                         const char *csStart, unsigned csLen);
2018 
2019   void HandlePositionalNonpositionalArgs(SourceLocation Loc,
2020                                          const char *startSpec,
2021                                          unsigned specifierLen);
2022 
2023   SourceRange getFormatStringRange();
2024   CharSourceRange getSpecifierRange(const char *startSpecifier,
2025                                     unsigned specifierLen);
2026   SourceLocation getLocationOfByte(const char *x);
2027 
2028   const Expr *getDataArg(unsigned i) const;
2029 
2030   bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
2031                     const analyze_format_string::ConversionSpecifier &CS,
2032                     const char *startSpecifier, unsigned specifierLen,
2033                     unsigned argIndex);
2034 
2035   template <typename Range>
2036   void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc,
2037                             bool IsStringLocation, Range StringRange,
2038                             ArrayRef<FixItHint> Fixit = None);
2039 
2040   void CheckPositionalAndNonpositionalArgs(
2041       const analyze_format_string::FormatSpecifier *FS);
2042 };
2043 }
2044 
2045 SourceRange CheckFormatHandler::getFormatStringRange() {
2046   return OrigFormatExpr->getSourceRange();
2047 }
2048 
2049 CharSourceRange CheckFormatHandler::
2050 getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
2051   SourceLocation Start = getLocationOfByte(startSpecifier);
2052   SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
2053 
2054   // Advance the end SourceLocation by one due to half-open ranges.
2055   End = End.getLocWithOffset(1);
2056 
2057   return CharSourceRange::getCharRange(Start, End);
2058 }
2059 
2060 SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
2061   return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
2062 }
2063 
2064 void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
2065                                                    unsigned specifierLen){
2066   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier),
2067                        getLocationOfByte(startSpecifier),
2068                        /*IsStringLocation*/true,
2069                        getSpecifierRange(startSpecifier, specifierLen));
2070 }
2071 
2072 void CheckFormatHandler::HandleInvalidLengthModifier(
2073     const analyze_format_string::FormatSpecifier &FS,
2074     const analyze_format_string::ConversionSpecifier &CS,
2075     const char *startSpecifier, unsigned specifierLen, unsigned DiagID) {
2076   using namespace analyze_format_string;
2077 
2078   const LengthModifier &LM = FS.getLengthModifier();
2079   CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength());
2080 
2081   // See if we know how to fix this length modifier.
2082   Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier();
2083   if (FixedLM) {
2084     EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(),
2085                          getLocationOfByte(LM.getStart()),
2086                          /*IsStringLocation*/true,
2087                          getSpecifierRange(startSpecifier, specifierLen));
2088 
2089     S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier)
2090       << FixedLM->toString()
2091       << FixItHint::CreateReplacement(LMRange, FixedLM->toString());
2092 
2093   } else {
2094     FixItHint Hint;
2095     if (DiagID == diag::warn_format_nonsensical_length)
2096       Hint = FixItHint::CreateRemoval(LMRange);
2097 
2098     EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(),
2099                          getLocationOfByte(LM.getStart()),
2100                          /*IsStringLocation*/true,
2101                          getSpecifierRange(startSpecifier, specifierLen),
2102                          Hint);
2103   }
2104 }
2105 
2106 void CheckFormatHandler::HandleNonStandardLengthModifier(
2107     const analyze_format_string::FormatSpecifier &FS,
2108     const char *startSpecifier, unsigned specifierLen) {
2109   using namespace analyze_format_string;
2110 
2111   const LengthModifier &LM = FS.getLengthModifier();
2112   CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength());
2113 
2114   // See if we know how to fix this length modifier.
2115   Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier();
2116   if (FixedLM) {
2117     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2118                            << LM.toString() << 0,
2119                          getLocationOfByte(LM.getStart()),
2120                          /*IsStringLocation*/true,
2121                          getSpecifierRange(startSpecifier, specifierLen));
2122 
2123     S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier)
2124       << FixedLM->toString()
2125       << FixItHint::CreateReplacement(LMRange, FixedLM->toString());
2126 
2127   } else {
2128     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2129                            << LM.toString() << 0,
2130                          getLocationOfByte(LM.getStart()),
2131                          /*IsStringLocation*/true,
2132                          getSpecifierRange(startSpecifier, specifierLen));
2133   }
2134 }
2135 
2136 void CheckFormatHandler::HandleNonStandardConversionSpecifier(
2137     const analyze_format_string::ConversionSpecifier &CS,
2138     const char *startSpecifier, unsigned specifierLen) {
2139   using namespace analyze_format_string;
2140 
2141   // See if we know how to fix this conversion specifier.
2142   Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier();
2143   if (FixedCS) {
2144     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2145                           << CS.toString() << /*conversion specifier*/1,
2146                          getLocationOfByte(CS.getStart()),
2147                          /*IsStringLocation*/true,
2148                          getSpecifierRange(startSpecifier, specifierLen));
2149 
2150     CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength());
2151     S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier)
2152       << FixedCS->toString()
2153       << FixItHint::CreateReplacement(CSRange, FixedCS->toString());
2154   } else {
2155     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2156                           << CS.toString() << /*conversion specifier*/1,
2157                          getLocationOfByte(CS.getStart()),
2158                          /*IsStringLocation*/true,
2159                          getSpecifierRange(startSpecifier, specifierLen));
2160   }
2161 }
2162 
2163 void CheckFormatHandler::HandlePosition(const char *startPos,
2164                                         unsigned posLen) {
2165   EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg),
2166                                getLocationOfByte(startPos),
2167                                /*IsStringLocation*/true,
2168                                getSpecifierRange(startPos, posLen));
2169 }
2170 
2171 void
2172 CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
2173                                      analyze_format_string::PositionContext p) {
2174   EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier)
2175                          << (unsigned) p,
2176                        getLocationOfByte(startPos), /*IsStringLocation*/true,
2177                        getSpecifierRange(startPos, posLen));
2178 }
2179 
2180 void CheckFormatHandler::HandleZeroPosition(const char *startPos,
2181                                             unsigned posLen) {
2182   EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier),
2183                                getLocationOfByte(startPos),
2184                                /*IsStringLocation*/true,
2185                                getSpecifierRange(startPos, posLen));
2186 }
2187 
2188 void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
2189   if (!isa<ObjCStringLiteral>(OrigFormatExpr)) {
2190     // The presence of a null character is likely an error.
2191     EmitFormatDiagnostic(
2192       S.PDiag(diag::warn_printf_format_string_contains_null_char),
2193       getLocationOfByte(nullCharacter), /*IsStringLocation*/true,
2194       getFormatStringRange());
2195   }
2196 }
2197 
2198 // Note that this may return NULL if there was an error parsing or building
2199 // one of the argument expressions.
2200 const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
2201   return Args[FirstDataArg + i];
2202 }
2203 
2204 void CheckFormatHandler::DoneProcessing() {
2205     // Does the number of data arguments exceed the number of
2206     // format conversions in the format string?
2207   if (!HasVAListArg) {
2208       // Find any arguments that weren't covered.
2209     CoveredArgs.flip();
2210     signed notCoveredArg = CoveredArgs.find_first();
2211     if (notCoveredArg >= 0) {
2212       assert((unsigned)notCoveredArg < NumDataArgs);
2213       if (const Expr *E = getDataArg((unsigned) notCoveredArg)) {
2214         SourceLocation Loc = E->getLocStart();
2215         if (!S.getSourceManager().isInSystemMacro(Loc)) {
2216           EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used),
2217                                Loc, /*IsStringLocation*/false,
2218                                getFormatStringRange());
2219         }
2220       }
2221     }
2222   }
2223 }
2224 
2225 bool
2226 CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
2227                                                      SourceLocation Loc,
2228                                                      const char *startSpec,
2229                                                      unsigned specifierLen,
2230                                                      const char *csStart,
2231                                                      unsigned csLen) {
2232 
2233   bool keepGoing = true;
2234   if (argIndex < NumDataArgs) {
2235     // Consider the argument coverered, even though the specifier doesn't
2236     // make sense.
2237     CoveredArgs.set(argIndex);
2238   }
2239   else {
2240     // If argIndex exceeds the number of data arguments we
2241     // don't issue a warning because that is just a cascade of warnings (and
2242     // they may have intended '%%' anyway). We don't want to continue processing
2243     // the format string after this point, however, as we will like just get
2244     // gibberish when trying to match arguments.
2245     keepGoing = false;
2246   }
2247 
2248   EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion)
2249                          << StringRef(csStart, csLen),
2250                        Loc, /*IsStringLocation*/true,
2251                        getSpecifierRange(startSpec, specifierLen));
2252 
2253   return keepGoing;
2254 }
2255 
2256 void
2257 CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc,
2258                                                       const char *startSpec,
2259                                                       unsigned specifierLen) {
2260   EmitFormatDiagnostic(
2261     S.PDiag(diag::warn_format_mix_positional_nonpositional_args),
2262     Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen));
2263 }
2264 
2265 bool
2266 CheckFormatHandler::CheckNumArgs(
2267   const analyze_format_string::FormatSpecifier &FS,
2268   const analyze_format_string::ConversionSpecifier &CS,
2269   const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
2270 
2271   if (argIndex >= NumDataArgs) {
2272     PartialDiagnostic PDiag = FS.usesPositionalArg()
2273       ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args)
2274            << (argIndex+1) << NumDataArgs)
2275       : S.PDiag(diag::warn_printf_insufficient_data_args);
2276     EmitFormatDiagnostic(
2277       PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true,
2278       getSpecifierRange(startSpecifier, specifierLen));
2279     return false;
2280   }
2281   return true;
2282 }
2283 
2284 template<typename Range>
2285 void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag,
2286                                               SourceLocation Loc,
2287                                               bool IsStringLocation,
2288                                               Range StringRange,
2289                                               ArrayRef<FixItHint> FixIt) {
2290   EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag,
2291                        Loc, IsStringLocation, StringRange, FixIt);
2292 }
2293 
2294 /// \brief If the format string is not within the funcion call, emit a note
2295 /// so that the function call and string are in diagnostic messages.
2296 ///
2297 /// \param InFunctionCall if true, the format string is within the function
2298 /// call and only one diagnostic message will be produced.  Otherwise, an
2299 /// extra note will be emitted pointing to location of the format string.
2300 ///
2301 /// \param ArgumentExpr the expression that is passed as the format string
2302 /// argument in the function call.  Used for getting locations when two
2303 /// diagnostics are emitted.
2304 ///
2305 /// \param PDiag the callee should already have provided any strings for the
2306 /// diagnostic message.  This function only adds locations and fixits
2307 /// to diagnostics.
2308 ///
2309 /// \param Loc primary location for diagnostic.  If two diagnostics are
2310 /// required, one will be at Loc and a new SourceLocation will be created for
2311 /// the other one.
2312 ///
2313 /// \param IsStringLocation if true, Loc points to the format string should be
2314 /// used for the note.  Otherwise, Loc points to the argument list and will
2315 /// be used with PDiag.
2316 ///
2317 /// \param StringRange some or all of the string to highlight.  This is
2318 /// templated so it can accept either a CharSourceRange or a SourceRange.
2319 ///
2320 /// \param FixIt optional fix it hint for the format string.
2321 template<typename Range>
2322 void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall,
2323                                               const Expr *ArgumentExpr,
2324                                               PartialDiagnostic PDiag,
2325                                               SourceLocation Loc,
2326                                               bool IsStringLocation,
2327                                               Range StringRange,
2328                                               ArrayRef<FixItHint> FixIt) {
2329   if (InFunctionCall) {
2330     const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag);
2331     D << StringRange;
2332     for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end();
2333          I != E; ++I) {
2334       D << *I;
2335     }
2336   } else {
2337     S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag)
2338       << ArgumentExpr->getSourceRange();
2339 
2340     const Sema::SemaDiagnosticBuilder &Note =
2341       S.Diag(IsStringLocation ? Loc : StringRange.getBegin(),
2342              diag::note_format_string_defined);
2343 
2344     Note << StringRange;
2345     for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end();
2346          I != E; ++I) {
2347       Note << *I;
2348     }
2349   }
2350 }
2351 
2352 //===--- CHECK: Printf format string checking ------------------------------===//
2353 
2354 namespace {
2355 class CheckPrintfHandler : public CheckFormatHandler {
2356   bool ObjCContext;
2357 public:
2358   CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
2359                      const Expr *origFormatExpr, unsigned firstDataArg,
2360                      unsigned numDataArgs, bool isObjC,
2361                      const char *beg, bool hasVAListArg,
2362                      ArrayRef<const Expr *> Args,
2363                      unsigned formatIdx, bool inFunctionCall,
2364                      Sema::VariadicCallType CallType)
2365   : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
2366                        numDataArgs, beg, hasVAListArg, Args,
2367                        formatIdx, inFunctionCall, CallType), ObjCContext(isObjC)
2368   {}
2369 
2370 
2371   bool HandleInvalidPrintfConversionSpecifier(
2372                                       const analyze_printf::PrintfSpecifier &FS,
2373                                       const char *startSpecifier,
2374                                       unsigned specifierLen);
2375 
2376   bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
2377                              const char *startSpecifier,
2378                              unsigned specifierLen);
2379   bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS,
2380                        const char *StartSpecifier,
2381                        unsigned SpecifierLen,
2382                        const Expr *E);
2383 
2384   bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
2385                     const char *startSpecifier, unsigned specifierLen);
2386   void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
2387                            const analyze_printf::OptionalAmount &Amt,
2388                            unsigned type,
2389                            const char *startSpecifier, unsigned specifierLen);
2390   void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
2391                   const analyze_printf::OptionalFlag &flag,
2392                   const char *startSpecifier, unsigned specifierLen);
2393   void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
2394                          const analyze_printf::OptionalFlag &ignoredFlag,
2395                          const analyze_printf::OptionalFlag &flag,
2396                          const char *startSpecifier, unsigned specifierLen);
2397   bool checkForCStrMembers(const analyze_printf::ArgType &AT,
2398                            const Expr *E, const CharSourceRange &CSR);
2399 
2400 };
2401 }
2402 
2403 bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
2404                                       const analyze_printf::PrintfSpecifier &FS,
2405                                       const char *startSpecifier,
2406                                       unsigned specifierLen) {
2407   const analyze_printf::PrintfConversionSpecifier &CS =
2408     FS.getConversionSpecifier();
2409 
2410   return HandleInvalidConversionSpecifier(FS.getArgIndex(),
2411                                           getLocationOfByte(CS.getStart()),
2412                                           startSpecifier, specifierLen,
2413                                           CS.getStart(), CS.getLength());
2414 }
2415 
2416 bool CheckPrintfHandler::HandleAmount(
2417                                const analyze_format_string::OptionalAmount &Amt,
2418                                unsigned k, const char *startSpecifier,
2419                                unsigned specifierLen) {
2420 
2421   if (Amt.hasDataArgument()) {
2422     if (!HasVAListArg) {
2423       unsigned argIndex = Amt.getArgIndex();
2424       if (argIndex >= NumDataArgs) {
2425         EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg)
2426                                << k,
2427                              getLocationOfByte(Amt.getStart()),
2428                              /*IsStringLocation*/true,
2429                              getSpecifierRange(startSpecifier, specifierLen));
2430         // Don't do any more checking.  We will just emit
2431         // spurious errors.
2432         return false;
2433       }
2434 
2435       // Type check the data argument.  It should be an 'int'.
2436       // Although not in conformance with C99, we also allow the argument to be
2437       // an 'unsigned int' as that is a reasonably safe case.  GCC also
2438       // doesn't emit a warning for that case.
2439       CoveredArgs.set(argIndex);
2440       const Expr *Arg = getDataArg(argIndex);
2441       if (!Arg)
2442         return false;
2443 
2444       QualType T = Arg->getType();
2445 
2446       const analyze_printf::ArgType &AT = Amt.getArgType(S.Context);
2447       assert(AT.isValid());
2448 
2449       if (!AT.matchesType(S.Context, T)) {
2450         EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type)
2451                                << k << AT.getRepresentativeTypeName(S.Context)
2452                                << T << Arg->getSourceRange(),
2453                              getLocationOfByte(Amt.getStart()),
2454                              /*IsStringLocation*/true,
2455                              getSpecifierRange(startSpecifier, specifierLen));
2456         // Don't do any more checking.  We will just emit
2457         // spurious errors.
2458         return false;
2459       }
2460     }
2461   }
2462   return true;
2463 }
2464 
2465 void CheckPrintfHandler::HandleInvalidAmount(
2466                                       const analyze_printf::PrintfSpecifier &FS,
2467                                       const analyze_printf::OptionalAmount &Amt,
2468                                       unsigned type,
2469                                       const char *startSpecifier,
2470                                       unsigned specifierLen) {
2471   const analyze_printf::PrintfConversionSpecifier &CS =
2472     FS.getConversionSpecifier();
2473 
2474   FixItHint fixit =
2475     Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant
2476       ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
2477                                  Amt.getConstantLength()))
2478       : FixItHint();
2479 
2480   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount)
2481                          << type << CS.toString(),
2482                        getLocationOfByte(Amt.getStart()),
2483                        /*IsStringLocation*/true,
2484                        getSpecifierRange(startSpecifier, specifierLen),
2485                        fixit);
2486 }
2487 
2488 void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
2489                                     const analyze_printf::OptionalFlag &flag,
2490                                     const char *startSpecifier,
2491                                     unsigned specifierLen) {
2492   // Warn about pointless flag with a fixit removal.
2493   const analyze_printf::PrintfConversionSpecifier &CS =
2494     FS.getConversionSpecifier();
2495   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag)
2496                          << flag.toString() << CS.toString(),
2497                        getLocationOfByte(flag.getPosition()),
2498                        /*IsStringLocation*/true,
2499                        getSpecifierRange(startSpecifier, specifierLen),
2500                        FixItHint::CreateRemoval(
2501                          getSpecifierRange(flag.getPosition(), 1)));
2502 }
2503 
2504 void CheckPrintfHandler::HandleIgnoredFlag(
2505                                 const analyze_printf::PrintfSpecifier &FS,
2506                                 const analyze_printf::OptionalFlag &ignoredFlag,
2507                                 const analyze_printf::OptionalFlag &flag,
2508                                 const char *startSpecifier,
2509                                 unsigned specifierLen) {
2510   // Warn about ignored flag with a fixit removal.
2511   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag)
2512                          << ignoredFlag.toString() << flag.toString(),
2513                        getLocationOfByte(ignoredFlag.getPosition()),
2514                        /*IsStringLocation*/true,
2515                        getSpecifierRange(startSpecifier, specifierLen),
2516                        FixItHint::CreateRemoval(
2517                          getSpecifierRange(ignoredFlag.getPosition(), 1)));
2518 }
2519 
2520 // Determines if the specified is a C++ class or struct containing
2521 // a member with the specified name and kind (e.g. a CXXMethodDecl named
2522 // "c_str()").
2523 template<typename MemberKind>
2524 static llvm::SmallPtrSet<MemberKind*, 1>
2525 CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) {
2526   const RecordType *RT = Ty->getAs<RecordType>();
2527   llvm::SmallPtrSet<MemberKind*, 1> Results;
2528 
2529   if (!RT)
2530     return Results;
2531   const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl());
2532   if (!RD)
2533     return Results;
2534 
2535   LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(),
2536                  Sema::LookupMemberName);
2537 
2538   // We just need to include all members of the right kind turned up by the
2539   // filter, at this point.
2540   if (S.LookupQualifiedName(R, RT->getDecl()))
2541     for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) {
2542       NamedDecl *decl = (*I)->getUnderlyingDecl();
2543       if (MemberKind *FK = dyn_cast<MemberKind>(decl))
2544         Results.insert(FK);
2545     }
2546   return Results;
2547 }
2548 
2549 // Check if a (w)string was passed when a (w)char* was needed, and offer a
2550 // better diagnostic if so. AT is assumed to be valid.
2551 // Returns true when a c_str() conversion method is found.
2552 bool CheckPrintfHandler::checkForCStrMembers(
2553     const analyze_printf::ArgType &AT, const Expr *E,
2554     const CharSourceRange &CSR) {
2555   typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet;
2556 
2557   MethodSet Results =
2558       CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType());
2559 
2560   for (MethodSet::iterator MI = Results.begin(), ME = Results.end();
2561        MI != ME; ++MI) {
2562     const CXXMethodDecl *Method = *MI;
2563     if (Method->getNumParams() == 0 &&
2564           AT.matchesType(S.Context, Method->getResultType())) {
2565       // FIXME: Suggest parens if the expression needs them.
2566       SourceLocation EndLoc =
2567           S.getPreprocessor().getLocForEndOfToken(E->getLocEnd());
2568       S.Diag(E->getLocStart(), diag::note_printf_c_str)
2569           << "c_str()"
2570           << FixItHint::CreateInsertion(EndLoc, ".c_str()");
2571       return true;
2572     }
2573   }
2574 
2575   return false;
2576 }
2577 
2578 bool
2579 CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
2580                                             &FS,
2581                                           const char *startSpecifier,
2582                                           unsigned specifierLen) {
2583 
2584   using namespace analyze_format_string;
2585   using namespace analyze_printf;
2586   const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
2587 
2588   if (FS.consumesDataArgument()) {
2589     if (atFirstArg) {
2590         atFirstArg = false;
2591         usesPositionalArgs = FS.usesPositionalArg();
2592     }
2593     else if (usesPositionalArgs != FS.usesPositionalArg()) {
2594       HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()),
2595                                         startSpecifier, specifierLen);
2596       return false;
2597     }
2598   }
2599 
2600   // First check if the field width, precision, and conversion specifier
2601   // have matching data arguments.
2602   if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
2603                     startSpecifier, specifierLen)) {
2604     return false;
2605   }
2606 
2607   if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
2608                     startSpecifier, specifierLen)) {
2609     return false;
2610   }
2611 
2612   if (!CS.consumesDataArgument()) {
2613     // FIXME: Technically specifying a precision or field width here
2614     // makes no sense.  Worth issuing a warning at some point.
2615     return true;
2616   }
2617 
2618   // Consume the argument.
2619   unsigned argIndex = FS.getArgIndex();
2620   if (argIndex < NumDataArgs) {
2621     // The check to see if the argIndex is valid will come later.
2622     // We set the bit here because we may exit early from this
2623     // function if we encounter some other error.
2624     CoveredArgs.set(argIndex);
2625   }
2626 
2627   // Check for using an Objective-C specific conversion specifier
2628   // in a non-ObjC literal.
2629   if (!ObjCContext && CS.isObjCArg()) {
2630     return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
2631                                                   specifierLen);
2632   }
2633 
2634   // Check for invalid use of field width
2635   if (!FS.hasValidFieldWidth()) {
2636     HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
2637         startSpecifier, specifierLen);
2638   }
2639 
2640   // Check for invalid use of precision
2641   if (!FS.hasValidPrecision()) {
2642     HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
2643         startSpecifier, specifierLen);
2644   }
2645 
2646   // Check each flag does not conflict with any other component.
2647   if (!FS.hasValidThousandsGroupingPrefix())
2648     HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
2649   if (!FS.hasValidLeadingZeros())
2650     HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
2651   if (!FS.hasValidPlusPrefix())
2652     HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
2653   if (!FS.hasValidSpacePrefix())
2654     HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
2655   if (!FS.hasValidAlternativeForm())
2656     HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
2657   if (!FS.hasValidLeftJustified())
2658     HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
2659 
2660   // Check that flags are not ignored by another flag
2661   if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
2662     HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
2663         startSpecifier, specifierLen);
2664   if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
2665     HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
2666             startSpecifier, specifierLen);
2667 
2668   // Check the length modifier is valid with the given conversion specifier.
2669   if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo()))
2670     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
2671                                 diag::warn_format_nonsensical_length);
2672   else if (!FS.hasStandardLengthModifier())
2673     HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen);
2674   else if (!FS.hasStandardLengthConversionCombination())
2675     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
2676                                 diag::warn_format_non_standard_conversion_spec);
2677 
2678   if (!FS.hasStandardConversionSpecifier(S.getLangOpts()))
2679     HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen);
2680 
2681   // The remaining checks depend on the data arguments.
2682   if (HasVAListArg)
2683     return true;
2684 
2685   if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
2686     return false;
2687 
2688   const Expr *Arg = getDataArg(argIndex);
2689   if (!Arg)
2690     return true;
2691 
2692   return checkFormatExpr(FS, startSpecifier, specifierLen, Arg);
2693 }
2694 
2695 static bool requiresParensToAddCast(const Expr *E) {
2696   // FIXME: We should have a general way to reason about operator
2697   // precedence and whether parens are actually needed here.
2698   // Take care of a few common cases where they aren't.
2699   const Expr *Inside = E->IgnoreImpCasts();
2700   if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside))
2701     Inside = POE->getSyntacticForm()->IgnoreImpCasts();
2702 
2703   switch (Inside->getStmtClass()) {
2704   case Stmt::ArraySubscriptExprClass:
2705   case Stmt::CallExprClass:
2706   case Stmt::CharacterLiteralClass:
2707   case Stmt::CXXBoolLiteralExprClass:
2708   case Stmt::DeclRefExprClass:
2709   case Stmt::FloatingLiteralClass:
2710   case Stmt::IntegerLiteralClass:
2711   case Stmt::MemberExprClass:
2712   case Stmt::ObjCArrayLiteralClass:
2713   case Stmt::ObjCBoolLiteralExprClass:
2714   case Stmt::ObjCBoxedExprClass:
2715   case Stmt::ObjCDictionaryLiteralClass:
2716   case Stmt::ObjCEncodeExprClass:
2717   case Stmt::ObjCIvarRefExprClass:
2718   case Stmt::ObjCMessageExprClass:
2719   case Stmt::ObjCPropertyRefExprClass:
2720   case Stmt::ObjCStringLiteralClass:
2721   case Stmt::ObjCSubscriptRefExprClass:
2722   case Stmt::ParenExprClass:
2723   case Stmt::StringLiteralClass:
2724   case Stmt::UnaryOperatorClass:
2725     return false;
2726   default:
2727     return true;
2728   }
2729 }
2730 
2731 bool
2732 CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS,
2733                                     const char *StartSpecifier,
2734                                     unsigned SpecifierLen,
2735                                     const Expr *E) {
2736   using namespace analyze_format_string;
2737   using namespace analyze_printf;
2738   // Now type check the data expression that matches the
2739   // format specifier.
2740   const analyze_printf::ArgType &AT = FS.getArgType(S.Context,
2741                                                     ObjCContext);
2742   if (!AT.isValid())
2743     return true;
2744 
2745   QualType ExprTy = E->getType();
2746   while (const TypeOfExprType *TET = dyn_cast<TypeOfExprType>(ExprTy)) {
2747     ExprTy = TET->getUnderlyingExpr()->getType();
2748   }
2749 
2750   if (AT.matchesType(S.Context, ExprTy))
2751     return true;
2752 
2753   // Look through argument promotions for our error message's reported type.
2754   // This includes the integral and floating promotions, but excludes array
2755   // and function pointer decay; seeing that an argument intended to be a
2756   // string has type 'char [6]' is probably more confusing than 'char *'.
2757   if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
2758     if (ICE->getCastKind() == CK_IntegralCast ||
2759         ICE->getCastKind() == CK_FloatingCast) {
2760       E = ICE->getSubExpr();
2761       ExprTy = E->getType();
2762 
2763       // Check if we didn't match because of an implicit cast from a 'char'
2764       // or 'short' to an 'int'.  This is done because printf is a varargs
2765       // function.
2766       if (ICE->getType() == S.Context.IntTy ||
2767           ICE->getType() == S.Context.UnsignedIntTy) {
2768         // All further checking is done on the subexpression.
2769         if (AT.matchesType(S.Context, ExprTy))
2770           return true;
2771       }
2772     }
2773   } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) {
2774     // Special case for 'a', which has type 'int' in C.
2775     // Note, however, that we do /not/ want to treat multibyte constants like
2776     // 'MooV' as characters! This form is deprecated but still exists.
2777     if (ExprTy == S.Context.IntTy)
2778       if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue()))
2779         ExprTy = S.Context.CharTy;
2780   }
2781 
2782   // %C in an Objective-C context prints a unichar, not a wchar_t.
2783   // If the argument is an integer of some kind, believe the %C and suggest
2784   // a cast instead of changing the conversion specifier.
2785   QualType IntendedTy = ExprTy;
2786   if (ObjCContext &&
2787       FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) {
2788     if (ExprTy->isIntegralOrUnscopedEnumerationType() &&
2789         !ExprTy->isCharType()) {
2790       // 'unichar' is defined as a typedef of unsigned short, but we should
2791       // prefer using the typedef if it is visible.
2792       IntendedTy = S.Context.UnsignedShortTy;
2793 
2794       LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(),
2795                           Sema::LookupOrdinaryName);
2796       if (S.LookupName(Result, S.getCurScope())) {
2797         NamedDecl *ND = Result.getFoundDecl();
2798         if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND))
2799           if (TD->getUnderlyingType() == IntendedTy)
2800             IntendedTy = S.Context.getTypedefType(TD);
2801       }
2802     }
2803   }
2804 
2805   // Special-case some of Darwin's platform-independence types by suggesting
2806   // casts to primitive types that are known to be large enough.
2807   bool ShouldNotPrintDirectly = false;
2808   if (S.Context.getTargetInfo().getTriple().isOSDarwin()) {
2809     // Use a 'while' to peel off layers of typedefs.
2810     QualType TyTy = IntendedTy;
2811     while (const TypedefType *UserTy = TyTy->getAs<TypedefType>()) {
2812       StringRef Name = UserTy->getDecl()->getName();
2813       QualType CastTy = llvm::StringSwitch<QualType>(Name)
2814         .Case("NSInteger", S.Context.LongTy)
2815         .Case("NSUInteger", S.Context.UnsignedLongTy)
2816         .Case("SInt32", S.Context.IntTy)
2817         .Case("UInt32", S.Context.UnsignedIntTy)
2818         .Default(QualType());
2819 
2820       if (!CastTy.isNull()) {
2821         ShouldNotPrintDirectly = true;
2822         IntendedTy = CastTy;
2823         break;
2824       }
2825       TyTy = UserTy->desugar();
2826     }
2827   }
2828 
2829   // We may be able to offer a FixItHint if it is a supported type.
2830   PrintfSpecifier fixedFS = FS;
2831   bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(),
2832                                  S.Context, ObjCContext);
2833 
2834   if (success) {
2835     // Get the fix string from the fixed format specifier
2836     SmallString<16> buf;
2837     llvm::raw_svector_ostream os(buf);
2838     fixedFS.toString(os);
2839 
2840     CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen);
2841 
2842     if (IntendedTy == ExprTy) {
2843       // In this case, the specifier is wrong and should be changed to match
2844       // the argument.
2845       EmitFormatDiagnostic(
2846         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2847           << AT.getRepresentativeTypeName(S.Context) << IntendedTy
2848           << E->getSourceRange(),
2849         E->getLocStart(),
2850         /*IsStringLocation*/false,
2851         SpecRange,
2852         FixItHint::CreateReplacement(SpecRange, os.str()));
2853 
2854     } else {
2855       // The canonical type for formatting this value is different from the
2856       // actual type of the expression. (This occurs, for example, with Darwin's
2857       // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but
2858       // should be printed as 'long' for 64-bit compatibility.)
2859       // Rather than emitting a normal format/argument mismatch, we want to
2860       // add a cast to the recommended type (and correct the format string
2861       // if necessary).
2862       SmallString<16> CastBuf;
2863       llvm::raw_svector_ostream CastFix(CastBuf);
2864       CastFix << "(";
2865       IntendedTy.print(CastFix, S.Context.getPrintingPolicy());
2866       CastFix << ")";
2867 
2868       SmallVector<FixItHint,4> Hints;
2869       if (!AT.matchesType(S.Context, IntendedTy))
2870         Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str()));
2871 
2872       if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) {
2873         // If there's already a cast present, just replace it.
2874         SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc());
2875         Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str()));
2876 
2877       } else if (!requiresParensToAddCast(E)) {
2878         // If the expression has high enough precedence,
2879         // just write the C-style cast.
2880         Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(),
2881                                                    CastFix.str()));
2882       } else {
2883         // Otherwise, add parens around the expression as well as the cast.
2884         CastFix << "(";
2885         Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(),
2886                                                    CastFix.str()));
2887 
2888         SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd());
2889         Hints.push_back(FixItHint::CreateInsertion(After, ")"));
2890       }
2891 
2892       if (ShouldNotPrintDirectly) {
2893         // The expression has a type that should not be printed directly.
2894         // We extract the name from the typedef because we don't want to show
2895         // the underlying type in the diagnostic.
2896         StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName();
2897 
2898         EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast)
2899                                << Name << IntendedTy
2900                                << E->getSourceRange(),
2901                              E->getLocStart(), /*IsStringLocation=*/false,
2902                              SpecRange, Hints);
2903       } else {
2904         // In this case, the expression could be printed using a different
2905         // specifier, but we've decided that the specifier is probably correct
2906         // and we should cast instead. Just use the normal warning message.
2907         EmitFormatDiagnostic(
2908           S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2909             << AT.getRepresentativeTypeName(S.Context) << ExprTy
2910             << E->getSourceRange(),
2911           E->getLocStart(), /*IsStringLocation*/false,
2912           SpecRange, Hints);
2913       }
2914     }
2915   } else {
2916     const CharSourceRange &CSR = getSpecifierRange(StartSpecifier,
2917                                                    SpecifierLen);
2918     // Since the warning for passing non-POD types to variadic functions
2919     // was deferred until now, we emit a warning for non-POD
2920     // arguments here.
2921     if (S.isValidVarArgType(ExprTy) == Sema::VAK_Invalid) {
2922       unsigned DiagKind;
2923       if (ExprTy->isObjCObjectType())
2924         DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format;
2925       else
2926         DiagKind = diag::warn_non_pod_vararg_with_format_string;
2927 
2928       EmitFormatDiagnostic(
2929         S.PDiag(DiagKind)
2930           << S.getLangOpts().CPlusPlus11
2931           << ExprTy
2932           << CallType
2933           << AT.getRepresentativeTypeName(S.Context)
2934           << CSR
2935           << E->getSourceRange(),
2936         E->getLocStart(), /*IsStringLocation*/false, CSR);
2937 
2938       checkForCStrMembers(AT, E, CSR);
2939     } else
2940       EmitFormatDiagnostic(
2941         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2942           << AT.getRepresentativeTypeName(S.Context) << ExprTy
2943           << CSR
2944           << E->getSourceRange(),
2945         E->getLocStart(), /*IsStringLocation*/false, CSR);
2946   }
2947 
2948   return true;
2949 }
2950 
2951 //===--- CHECK: Scanf format string checking ------------------------------===//
2952 
2953 namespace {
2954 class CheckScanfHandler : public CheckFormatHandler {
2955 public:
2956   CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
2957                     const Expr *origFormatExpr, unsigned firstDataArg,
2958                     unsigned numDataArgs, const char *beg, bool hasVAListArg,
2959                     ArrayRef<const Expr *> Args,
2960                     unsigned formatIdx, bool inFunctionCall,
2961                     Sema::VariadicCallType CallType)
2962   : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
2963                        numDataArgs, beg, hasVAListArg,
2964                        Args, formatIdx, inFunctionCall, CallType)
2965   {}
2966 
2967   bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
2968                             const char *startSpecifier,
2969                             unsigned specifierLen);
2970 
2971   bool HandleInvalidScanfConversionSpecifier(
2972           const analyze_scanf::ScanfSpecifier &FS,
2973           const char *startSpecifier,
2974           unsigned specifierLen);
2975 
2976   void HandleIncompleteScanList(const char *start, const char *end);
2977 };
2978 }
2979 
2980 void CheckScanfHandler::HandleIncompleteScanList(const char *start,
2981                                                  const char *end) {
2982   EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete),
2983                        getLocationOfByte(end), /*IsStringLocation*/true,
2984                        getSpecifierRange(start, end - start));
2985 }
2986 
2987 bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
2988                                         const analyze_scanf::ScanfSpecifier &FS,
2989                                         const char *startSpecifier,
2990                                         unsigned specifierLen) {
2991 
2992   const analyze_scanf::ScanfConversionSpecifier &CS =
2993     FS.getConversionSpecifier();
2994 
2995   return HandleInvalidConversionSpecifier(FS.getArgIndex(),
2996                                           getLocationOfByte(CS.getStart()),
2997                                           startSpecifier, specifierLen,
2998                                           CS.getStart(), CS.getLength());
2999 }
3000 
3001 bool CheckScanfHandler::HandleScanfSpecifier(
3002                                        const analyze_scanf::ScanfSpecifier &FS,
3003                                        const char *startSpecifier,
3004                                        unsigned specifierLen) {
3005 
3006   using namespace analyze_scanf;
3007   using namespace analyze_format_string;
3008 
3009   const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
3010 
3011   // Handle case where '%' and '*' don't consume an argument.  These shouldn't
3012   // be used to decide if we are using positional arguments consistently.
3013   if (FS.consumesDataArgument()) {
3014     if (atFirstArg) {
3015       atFirstArg = false;
3016       usesPositionalArgs = FS.usesPositionalArg();
3017     }
3018     else if (usesPositionalArgs != FS.usesPositionalArg()) {
3019       HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()),
3020                                         startSpecifier, specifierLen);
3021       return false;
3022     }
3023   }
3024 
3025   // Check if the field with is non-zero.
3026   const OptionalAmount &Amt = FS.getFieldWidth();
3027   if (Amt.getHowSpecified() == OptionalAmount::Constant) {
3028     if (Amt.getConstantAmount() == 0) {
3029       const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
3030                                                    Amt.getConstantLength());
3031       EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width),
3032                            getLocationOfByte(Amt.getStart()),
3033                            /*IsStringLocation*/true, R,
3034                            FixItHint::CreateRemoval(R));
3035     }
3036   }
3037 
3038   if (!FS.consumesDataArgument()) {
3039     // FIXME: Technically specifying a precision or field width here
3040     // makes no sense.  Worth issuing a warning at some point.
3041     return true;
3042   }
3043 
3044   // Consume the argument.
3045   unsigned argIndex = FS.getArgIndex();
3046   if (argIndex < NumDataArgs) {
3047       // The check to see if the argIndex is valid will come later.
3048       // We set the bit here because we may exit early from this
3049       // function if we encounter some other error.
3050     CoveredArgs.set(argIndex);
3051   }
3052 
3053   // Check the length modifier is valid with the given conversion specifier.
3054   if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo()))
3055     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
3056                                 diag::warn_format_nonsensical_length);
3057   else if (!FS.hasStandardLengthModifier())
3058     HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen);
3059   else if (!FS.hasStandardLengthConversionCombination())
3060     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
3061                                 diag::warn_format_non_standard_conversion_spec);
3062 
3063   if (!FS.hasStandardConversionSpecifier(S.getLangOpts()))
3064     HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen);
3065 
3066   // The remaining checks depend on the data arguments.
3067   if (HasVAListArg)
3068     return true;
3069 
3070   if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
3071     return false;
3072 
3073   // Check that the argument type matches the format specifier.
3074   const Expr *Ex = getDataArg(argIndex);
3075   if (!Ex)
3076     return true;
3077 
3078   const analyze_format_string::ArgType &AT = FS.getArgType(S.Context);
3079   if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) {
3080     ScanfSpecifier fixedFS = FS;
3081     bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(),
3082                                    S.Context);
3083 
3084     if (success) {
3085       // Get the fix string from the fixed format specifier.
3086       SmallString<128> buf;
3087       llvm::raw_svector_ostream os(buf);
3088       fixedFS.toString(os);
3089 
3090       EmitFormatDiagnostic(
3091         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
3092           << AT.getRepresentativeTypeName(S.Context) << Ex->getType()
3093           << Ex->getSourceRange(),
3094         Ex->getLocStart(),
3095         /*IsStringLocation*/false,
3096         getSpecifierRange(startSpecifier, specifierLen),
3097         FixItHint::CreateReplacement(
3098           getSpecifierRange(startSpecifier, specifierLen),
3099           os.str()));
3100     } else {
3101       EmitFormatDiagnostic(
3102         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
3103           << AT.getRepresentativeTypeName(S.Context) << Ex->getType()
3104           << Ex->getSourceRange(),
3105         Ex->getLocStart(),
3106         /*IsStringLocation*/false,
3107         getSpecifierRange(startSpecifier, specifierLen));
3108     }
3109   }
3110 
3111   return true;
3112 }
3113 
3114 void Sema::CheckFormatString(const StringLiteral *FExpr,
3115                              const Expr *OrigFormatExpr,
3116                              ArrayRef<const Expr *> Args,
3117                              bool HasVAListArg, unsigned format_idx,
3118                              unsigned firstDataArg, FormatStringType Type,
3119                              bool inFunctionCall, VariadicCallType CallType) {
3120 
3121   // CHECK: is the format string a wide literal?
3122   if (!FExpr->isAscii() && !FExpr->isUTF8()) {
3123     CheckFormatHandler::EmitFormatDiagnostic(
3124       *this, inFunctionCall, Args[format_idx],
3125       PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(),
3126       /*IsStringLocation*/true, OrigFormatExpr->getSourceRange());
3127     return;
3128   }
3129 
3130   // Str - The format string.  NOTE: this is NOT null-terminated!
3131   StringRef StrRef = FExpr->getString();
3132   const char *Str = StrRef.data();
3133   unsigned StrLen = StrRef.size();
3134   const unsigned numDataArgs = Args.size() - firstDataArg;
3135 
3136   // CHECK: empty format string?
3137   if (StrLen == 0 && numDataArgs > 0) {
3138     CheckFormatHandler::EmitFormatDiagnostic(
3139       *this, inFunctionCall, Args[format_idx],
3140       PDiag(diag::warn_empty_format_string), FExpr->getLocStart(),
3141       /*IsStringLocation*/true, OrigFormatExpr->getSourceRange());
3142     return;
3143   }
3144 
3145   if (Type == FST_Printf || Type == FST_NSString) {
3146     CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
3147                          numDataArgs, (Type == FST_NSString),
3148                          Str, HasVAListArg, Args, format_idx,
3149                          inFunctionCall, CallType);
3150 
3151     if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen,
3152                                                   getLangOpts(),
3153                                                   Context.getTargetInfo()))
3154       H.DoneProcessing();
3155   } else if (Type == FST_Scanf) {
3156     CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs,
3157                         Str, HasVAListArg, Args, format_idx,
3158                         inFunctionCall, CallType);
3159 
3160     if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen,
3161                                                  getLangOpts(),
3162                                                  Context.getTargetInfo()))
3163       H.DoneProcessing();
3164   } // TODO: handle other formats
3165 }
3166 
3167 //===--- CHECK: Standard memory functions ---------------------------------===//
3168 
3169 /// \brief Determine whether the given type is a dynamic class type (e.g.,
3170 /// whether it has a vtable).
3171 static bool isDynamicClassType(QualType T) {
3172   if (CXXRecordDecl *Record = T->getAsCXXRecordDecl())
3173     if (CXXRecordDecl *Definition = Record->getDefinition())
3174       if (Definition->isDynamicClass())
3175         return true;
3176 
3177   return false;
3178 }
3179 
3180 /// \brief If E is a sizeof expression, returns its argument expression,
3181 /// otherwise returns NULL.
3182 static const Expr *getSizeOfExprArg(const Expr* E) {
3183   if (const UnaryExprOrTypeTraitExpr *SizeOf =
3184       dyn_cast<UnaryExprOrTypeTraitExpr>(E))
3185     if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType())
3186       return SizeOf->getArgumentExpr()->IgnoreParenImpCasts();
3187 
3188   return 0;
3189 }
3190 
3191 /// \brief If E is a sizeof expression, returns its argument type.
3192 static QualType getSizeOfArgType(const Expr* E) {
3193   if (const UnaryExprOrTypeTraitExpr *SizeOf =
3194       dyn_cast<UnaryExprOrTypeTraitExpr>(E))
3195     if (SizeOf->getKind() == clang::UETT_SizeOf)
3196       return SizeOf->getTypeOfArgument();
3197 
3198   return QualType();
3199 }
3200 
3201 /// \brief Check for dangerous or invalid arguments to memset().
3202 ///
3203 /// This issues warnings on known problematic, dangerous or unspecified
3204 /// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp'
3205 /// function calls.
3206 ///
3207 /// \param Call The call expression to diagnose.
3208 void Sema::CheckMemaccessArguments(const CallExpr *Call,
3209                                    unsigned BId,
3210                                    IdentifierInfo *FnName) {
3211   assert(BId != 0);
3212 
3213   // It is possible to have a non-standard definition of memset.  Validate
3214   // we have enough arguments, and if not, abort further checking.
3215   unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3);
3216   if (Call->getNumArgs() < ExpectedNumArgs)
3217     return;
3218 
3219   unsigned LastArg = (BId == Builtin::BImemset ||
3220                       BId == Builtin::BIstrndup ? 1 : 2);
3221   unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2);
3222   const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts();
3223 
3224   // We have special checking when the length is a sizeof expression.
3225   QualType SizeOfArgTy = getSizeOfArgType(LenExpr);
3226   const Expr *SizeOfArg = getSizeOfExprArg(LenExpr);
3227   llvm::FoldingSetNodeID SizeOfArgID;
3228 
3229   for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) {
3230     const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts();
3231     SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange();
3232 
3233     QualType DestTy = Dest->getType();
3234     if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
3235       QualType PointeeTy = DestPtrTy->getPointeeType();
3236 
3237       // Never warn about void type pointers. This can be used to suppress
3238       // false positives.
3239       if (PointeeTy->isVoidType())
3240         continue;
3241 
3242       // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by
3243       // actually comparing the expressions for equality. Because computing the
3244       // expression IDs can be expensive, we only do this if the diagnostic is
3245       // enabled.
3246       if (SizeOfArg &&
3247           Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess,
3248                                    SizeOfArg->getExprLoc())) {
3249         // We only compute IDs for expressions if the warning is enabled, and
3250         // cache the sizeof arg's ID.
3251         if (SizeOfArgID == llvm::FoldingSetNodeID())
3252           SizeOfArg->Profile(SizeOfArgID, Context, true);
3253         llvm::FoldingSetNodeID DestID;
3254         Dest->Profile(DestID, Context, true);
3255         if (DestID == SizeOfArgID) {
3256           // TODO: For strncpy() and friends, this could suggest sizeof(dst)
3257           //       over sizeof(src) as well.
3258           unsigned ActionIdx = 0; // Default is to suggest dereferencing.
3259           StringRef ReadableName = FnName->getName();
3260 
3261           if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest))
3262             if (UnaryOp->getOpcode() == UO_AddrOf)
3263               ActionIdx = 1; // If its an address-of operator, just remove it.
3264           if (!PointeeTy->isIncompleteType() &&
3265               (Context.getTypeSize(PointeeTy) == Context.getCharWidth()))
3266             ActionIdx = 2; // If the pointee's size is sizeof(char),
3267                            // suggest an explicit length.
3268 
3269           // If the function is defined as a builtin macro, do not show macro
3270           // expansion.
3271           SourceLocation SL = SizeOfArg->getExprLoc();
3272           SourceRange DSR = Dest->getSourceRange();
3273           SourceRange SSR = SizeOfArg->getSourceRange();
3274           SourceManager &SM  = PP.getSourceManager();
3275 
3276           if (SM.isMacroArgExpansion(SL)) {
3277             ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts);
3278             SL = SM.getSpellingLoc(SL);
3279             DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()),
3280                              SM.getSpellingLoc(DSR.getEnd()));
3281             SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()),
3282                              SM.getSpellingLoc(SSR.getEnd()));
3283           }
3284 
3285           DiagRuntimeBehavior(SL, SizeOfArg,
3286                               PDiag(diag::warn_sizeof_pointer_expr_memaccess)
3287                                 << ReadableName
3288                                 << PointeeTy
3289                                 << DestTy
3290                                 << DSR
3291                                 << SSR);
3292           DiagRuntimeBehavior(SL, SizeOfArg,
3293                          PDiag(diag::warn_sizeof_pointer_expr_memaccess_note)
3294                                 << ActionIdx
3295                                 << SSR);
3296 
3297           break;
3298         }
3299       }
3300 
3301       // Also check for cases where the sizeof argument is the exact same
3302       // type as the memory argument, and where it points to a user-defined
3303       // record type.
3304       if (SizeOfArgTy != QualType()) {
3305         if (PointeeTy->isRecordType() &&
3306             Context.typesAreCompatible(SizeOfArgTy, DestTy)) {
3307           DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest,
3308                               PDiag(diag::warn_sizeof_pointer_type_memaccess)
3309                                 << FnName << SizeOfArgTy << ArgIdx
3310                                 << PointeeTy << Dest->getSourceRange()
3311                                 << LenExpr->getSourceRange());
3312           break;
3313         }
3314       }
3315 
3316       // Always complain about dynamic classes.
3317       if (isDynamicClassType(PointeeTy)) {
3318 
3319         unsigned OperationType = 0;
3320         // "overwritten" if we're warning about the destination for any call
3321         // but memcmp; otherwise a verb appropriate to the call.
3322         if (ArgIdx != 0 || BId == Builtin::BImemcmp) {
3323           if (BId == Builtin::BImemcpy)
3324             OperationType = 1;
3325           else if(BId == Builtin::BImemmove)
3326             OperationType = 2;
3327           else if (BId == Builtin::BImemcmp)
3328             OperationType = 3;
3329         }
3330 
3331         DiagRuntimeBehavior(
3332           Dest->getExprLoc(), Dest,
3333           PDiag(diag::warn_dyn_class_memaccess)
3334             << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx)
3335             << FnName << PointeeTy
3336             << OperationType
3337             << Call->getCallee()->getSourceRange());
3338       } else if (PointeeTy.hasNonTrivialObjCLifetime() &&
3339                BId != Builtin::BImemset)
3340         DiagRuntimeBehavior(
3341           Dest->getExprLoc(), Dest,
3342           PDiag(diag::warn_arc_object_memaccess)
3343             << ArgIdx << FnName << PointeeTy
3344             << Call->getCallee()->getSourceRange());
3345       else
3346         continue;
3347 
3348       DiagRuntimeBehavior(
3349         Dest->getExprLoc(), Dest,
3350         PDiag(diag::note_bad_memaccess_silence)
3351           << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
3352       break;
3353     }
3354   }
3355 }
3356 
3357 // A little helper routine: ignore addition and subtraction of integer literals.
3358 // This intentionally does not ignore all integer constant expressions because
3359 // we don't want to remove sizeof().
3360 static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) {
3361   Ex = Ex->IgnoreParenCasts();
3362 
3363   for (;;) {
3364     const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex);
3365     if (!BO || !BO->isAdditiveOp())
3366       break;
3367 
3368     const Expr *RHS = BO->getRHS()->IgnoreParenCasts();
3369     const Expr *LHS = BO->getLHS()->IgnoreParenCasts();
3370 
3371     if (isa<IntegerLiteral>(RHS))
3372       Ex = LHS;
3373     else if (isa<IntegerLiteral>(LHS))
3374       Ex = RHS;
3375     else
3376       break;
3377   }
3378 
3379   return Ex;
3380 }
3381 
3382 static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty,
3383                                                       ASTContext &Context) {
3384   // Only handle constant-sized or VLAs, but not flexible members.
3385   if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) {
3386     // Only issue the FIXIT for arrays of size > 1.
3387     if (CAT->getSize().getSExtValue() <= 1)
3388       return false;
3389   } else if (!Ty->isVariableArrayType()) {
3390     return false;
3391   }
3392   return true;
3393 }
3394 
3395 // Warn if the user has made the 'size' argument to strlcpy or strlcat
3396 // be the size of the source, instead of the destination.
3397 void Sema::CheckStrlcpycatArguments(const CallExpr *Call,
3398                                     IdentifierInfo *FnName) {
3399 
3400   // Don't crash if the user has the wrong number of arguments
3401   if (Call->getNumArgs() != 3)
3402     return;
3403 
3404   const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context);
3405   const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context);
3406   const Expr *CompareWithSrc = NULL;
3407 
3408   // Look for 'strlcpy(dst, x, sizeof(x))'
3409   if (const Expr *Ex = getSizeOfExprArg(SizeArg))
3410     CompareWithSrc = Ex;
3411   else {
3412     // Look for 'strlcpy(dst, x, strlen(x))'
3413     if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) {
3414       if (SizeCall->isBuiltinCall() == Builtin::BIstrlen
3415           && SizeCall->getNumArgs() == 1)
3416         CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context);
3417     }
3418   }
3419 
3420   if (!CompareWithSrc)
3421     return;
3422 
3423   // Determine if the argument to sizeof/strlen is equal to the source
3424   // argument.  In principle there's all kinds of things you could do
3425   // here, for instance creating an == expression and evaluating it with
3426   // EvaluateAsBooleanCondition, but this uses a more direct technique:
3427   const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg);
3428   if (!SrcArgDRE)
3429     return;
3430 
3431   const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc);
3432   if (!CompareWithSrcDRE ||
3433       SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl())
3434     return;
3435 
3436   const Expr *OriginalSizeArg = Call->getArg(2);
3437   Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size)
3438     << OriginalSizeArg->getSourceRange() << FnName;
3439 
3440   // Output a FIXIT hint if the destination is an array (rather than a
3441   // pointer to an array).  This could be enhanced to handle some
3442   // pointers if we know the actual size, like if DstArg is 'array+2'
3443   // we could say 'sizeof(array)-2'.
3444   const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts();
3445   if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context))
3446     return;
3447 
3448   SmallString<128> sizeString;
3449   llvm::raw_svector_ostream OS(sizeString);
3450   OS << "sizeof(";
3451   DstArg->printPretty(OS, 0, getPrintingPolicy());
3452   OS << ")";
3453 
3454   Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size)
3455     << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(),
3456                                     OS.str());
3457 }
3458 
3459 /// Check if two expressions refer to the same declaration.
3460 static bool referToTheSameDecl(const Expr *E1, const Expr *E2) {
3461   if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1))
3462     if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2))
3463       return D1->getDecl() == D2->getDecl();
3464   return false;
3465 }
3466 
3467 static const Expr *getStrlenExprArg(const Expr *E) {
3468   if (const CallExpr *CE = dyn_cast<CallExpr>(E)) {
3469     const FunctionDecl *FD = CE->getDirectCallee();
3470     if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen)
3471       return 0;
3472     return CE->getArg(0)->IgnoreParenCasts();
3473   }
3474   return 0;
3475 }
3476 
3477 // Warn on anti-patterns as the 'size' argument to strncat.
3478 // The correct size argument should look like following:
3479 //   strncat(dst, src, sizeof(dst) - strlen(dest) - 1);
3480 void Sema::CheckStrncatArguments(const CallExpr *CE,
3481                                  IdentifierInfo *FnName) {
3482   // Don't crash if the user has the wrong number of arguments.
3483   if (CE->getNumArgs() < 3)
3484     return;
3485   const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts();
3486   const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts();
3487   const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts();
3488 
3489   // Identify common expressions, which are wrongly used as the size argument
3490   // to strncat and may lead to buffer overflows.
3491   unsigned PatternType = 0;
3492   if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) {
3493     // - sizeof(dst)
3494     if (referToTheSameDecl(SizeOfArg, DstArg))
3495       PatternType = 1;
3496     // - sizeof(src)
3497     else if (referToTheSameDecl(SizeOfArg, SrcArg))
3498       PatternType = 2;
3499   } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) {
3500     if (BE->getOpcode() == BO_Sub) {
3501       const Expr *L = BE->getLHS()->IgnoreParenCasts();
3502       const Expr *R = BE->getRHS()->IgnoreParenCasts();
3503       // - sizeof(dst) - strlen(dst)
3504       if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) &&
3505           referToTheSameDecl(DstArg, getStrlenExprArg(R)))
3506         PatternType = 1;
3507       // - sizeof(src) - (anything)
3508       else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L)))
3509         PatternType = 2;
3510     }
3511   }
3512 
3513   if (PatternType == 0)
3514     return;
3515 
3516   // Generate the diagnostic.
3517   SourceLocation SL = LenArg->getLocStart();
3518   SourceRange SR = LenArg->getSourceRange();
3519   SourceManager &SM  = PP.getSourceManager();
3520 
3521   // If the function is defined as a builtin macro, do not show macro expansion.
3522   if (SM.isMacroArgExpansion(SL)) {
3523     SL = SM.getSpellingLoc(SL);
3524     SR = SourceRange(SM.getSpellingLoc(SR.getBegin()),
3525                      SM.getSpellingLoc(SR.getEnd()));
3526   }
3527 
3528   // Check if the destination is an array (rather than a pointer to an array).
3529   QualType DstTy = DstArg->getType();
3530   bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy,
3531                                                                     Context);
3532   if (!isKnownSizeArray) {
3533     if (PatternType == 1)
3534       Diag(SL, diag::warn_strncat_wrong_size) << SR;
3535     else
3536       Diag(SL, diag::warn_strncat_src_size) << SR;
3537     return;
3538   }
3539 
3540   if (PatternType == 1)
3541     Diag(SL, diag::warn_strncat_large_size) << SR;
3542   else
3543     Diag(SL, diag::warn_strncat_src_size) << SR;
3544 
3545   SmallString<128> sizeString;
3546   llvm::raw_svector_ostream OS(sizeString);
3547   OS << "sizeof(";
3548   DstArg->printPretty(OS, 0, getPrintingPolicy());
3549   OS << ") - ";
3550   OS << "strlen(";
3551   DstArg->printPretty(OS, 0, getPrintingPolicy());
3552   OS << ") - 1";
3553 
3554   Diag(SL, diag::note_strncat_wrong_size)
3555     << FixItHint::CreateReplacement(SR, OS.str());
3556 }
3557 
3558 //===--- CHECK: Return Address of Stack Variable --------------------------===//
3559 
3560 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
3561                      Decl *ParentDecl);
3562 static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars,
3563                       Decl *ParentDecl);
3564 
3565 /// CheckReturnStackAddr - Check if a return statement returns the address
3566 ///   of a stack variable.
3567 void
3568 Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
3569                            SourceLocation ReturnLoc) {
3570 
3571   Expr *stackE = 0;
3572   SmallVector<DeclRefExpr *, 8> refVars;
3573 
3574   // Perform checking for returned stack addresses, local blocks,
3575   // label addresses or references to temporaries.
3576   if (lhsType->isPointerType() ||
3577       (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) {
3578     stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0);
3579   } else if (lhsType->isReferenceType()) {
3580     stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0);
3581   }
3582 
3583   if (stackE == 0)
3584     return; // Nothing suspicious was found.
3585 
3586   SourceLocation diagLoc;
3587   SourceRange diagRange;
3588   if (refVars.empty()) {
3589     diagLoc = stackE->getLocStart();
3590     diagRange = stackE->getSourceRange();
3591   } else {
3592     // We followed through a reference variable. 'stackE' contains the
3593     // problematic expression but we will warn at the return statement pointing
3594     // at the reference variable. We will later display the "trail" of
3595     // reference variables using notes.
3596     diagLoc = refVars[0]->getLocStart();
3597     diagRange = refVars[0]->getSourceRange();
3598   }
3599 
3600   if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
3601     Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
3602                                              : diag::warn_ret_stack_addr)
3603      << DR->getDecl()->getDeclName() << diagRange;
3604   } else if (isa<BlockExpr>(stackE)) { // local block.
3605     Diag(diagLoc, diag::err_ret_local_block) << diagRange;
3606   } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
3607     Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
3608   } else { // local temporary.
3609     Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
3610                                              : diag::warn_ret_local_temp_addr)
3611      << diagRange;
3612   }
3613 
3614   // Display the "trail" of reference variables that we followed until we
3615   // found the problematic expression using notes.
3616   for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
3617     VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
3618     // If this var binds to another reference var, show the range of the next
3619     // var, otherwise the var binds to the problematic expression, in which case
3620     // show the range of the expression.
3621     SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
3622                                   : stackE->getSourceRange();
3623     Diag(VD->getLocation(), diag::note_ref_var_local_bind)
3624       << VD->getDeclName() << range;
3625   }
3626 }
3627 
3628 /// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
3629 ///  check if the expression in a return statement evaluates to an address
3630 ///  to a location on the stack, a local block, an address of a label, or a
3631 ///  reference to local temporary. The recursion is used to traverse the
3632 ///  AST of the return expression, with recursion backtracking when we
3633 ///  encounter a subexpression that (1) clearly does not lead to one of the
3634 ///  above problematic expressions (2) is something we cannot determine leads to
3635 ///  a problematic expression based on such local checking.
3636 ///
3637 ///  Both EvalAddr and EvalVal follow through reference variables to evaluate
3638 ///  the expression that they point to. Such variables are added to the
3639 ///  'refVars' vector so that we know what the reference variable "trail" was.
3640 ///
3641 ///  EvalAddr processes expressions that are pointers that are used as
3642 ///  references (and not L-values).  EvalVal handles all other values.
3643 ///  At the base case of the recursion is a check for the above problematic
3644 ///  expressions.
3645 ///
3646 ///  This implementation handles:
3647 ///
3648 ///   * pointer-to-pointer casts
3649 ///   * implicit conversions from array references to pointers
3650 ///   * taking the address of fields
3651 ///   * arbitrary interplay between "&" and "*" operators
3652 ///   * pointer arithmetic from an address of a stack variable
3653 ///   * taking the address of an array element where the array is on the stack
3654 static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
3655                       Decl *ParentDecl) {
3656   if (E->isTypeDependent())
3657       return NULL;
3658 
3659   // We should only be called for evaluating pointer expressions.
3660   assert((E->getType()->isAnyPointerType() ||
3661           E->getType()->isBlockPointerType() ||
3662           E->getType()->isObjCQualifiedIdType()) &&
3663          "EvalAddr only works on pointers");
3664 
3665   E = E->IgnoreParens();
3666 
3667   // Our "symbolic interpreter" is just a dispatch off the currently
3668   // viewed AST node.  We then recursively traverse the AST by calling
3669   // EvalAddr and EvalVal appropriately.
3670   switch (E->getStmtClass()) {
3671   case Stmt::DeclRefExprClass: {
3672     DeclRefExpr *DR = cast<DeclRefExpr>(E);
3673 
3674     if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
3675       // If this is a reference variable, follow through to the expression that
3676       // it points to.
3677       if (V->hasLocalStorage() &&
3678           V->getType()->isReferenceType() && V->hasInit()) {
3679         // Add the reference variable to the "trail".
3680         refVars.push_back(DR);
3681         return EvalAddr(V->getInit(), refVars, ParentDecl);
3682       }
3683 
3684     return NULL;
3685   }
3686 
3687   case Stmt::UnaryOperatorClass: {
3688     // The only unary operator that make sense to handle here
3689     // is AddrOf.  All others don't make sense as pointers.
3690     UnaryOperator *U = cast<UnaryOperator>(E);
3691 
3692     if (U->getOpcode() == UO_AddrOf)
3693       return EvalVal(U->getSubExpr(), refVars, ParentDecl);
3694     else
3695       return NULL;
3696   }
3697 
3698   case Stmt::BinaryOperatorClass: {
3699     // Handle pointer arithmetic.  All other binary operators are not valid
3700     // in this context.
3701     BinaryOperator *B = cast<BinaryOperator>(E);
3702     BinaryOperatorKind op = B->getOpcode();
3703 
3704     if (op != BO_Add && op != BO_Sub)
3705       return NULL;
3706 
3707     Expr *Base = B->getLHS();
3708 
3709     // Determine which argument is the real pointer base.  It could be
3710     // the RHS argument instead of the LHS.
3711     if (!Base->getType()->isPointerType()) Base = B->getRHS();
3712 
3713     assert (Base->getType()->isPointerType());
3714     return EvalAddr(Base, refVars, ParentDecl);
3715   }
3716 
3717   // For conditional operators we need to see if either the LHS or RHS are
3718   // valid DeclRefExpr*s.  If one of them is valid, we return it.
3719   case Stmt::ConditionalOperatorClass: {
3720     ConditionalOperator *C = cast<ConditionalOperator>(E);
3721 
3722     // Handle the GNU extension for missing LHS.
3723     if (Expr *lhsExpr = C->getLHS()) {
3724     // In C++, we can have a throw-expression, which has 'void' type.
3725       if (!lhsExpr->getType()->isVoidType())
3726         if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl))
3727           return LHS;
3728     }
3729 
3730     // In C++, we can have a throw-expression, which has 'void' type.
3731     if (C->getRHS()->getType()->isVoidType())
3732       return NULL;
3733 
3734     return EvalAddr(C->getRHS(), refVars, ParentDecl);
3735   }
3736 
3737   case Stmt::BlockExprClass:
3738     if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
3739       return E; // local block.
3740     return NULL;
3741 
3742   case Stmt::AddrLabelExprClass:
3743     return E; // address of label.
3744 
3745   case Stmt::ExprWithCleanupsClass:
3746     return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,
3747                     ParentDecl);
3748 
3749   // For casts, we need to handle conversions from arrays to
3750   // pointer values, and pointer-to-pointer conversions.
3751   case Stmt::ImplicitCastExprClass:
3752   case Stmt::CStyleCastExprClass:
3753   case Stmt::CXXFunctionalCastExprClass:
3754   case Stmt::ObjCBridgedCastExprClass:
3755   case Stmt::CXXStaticCastExprClass:
3756   case Stmt::CXXDynamicCastExprClass:
3757   case Stmt::CXXConstCastExprClass:
3758   case Stmt::CXXReinterpretCastExprClass: {
3759     Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
3760     switch (cast<CastExpr>(E)->getCastKind()) {
3761     case CK_BitCast:
3762     case CK_LValueToRValue:
3763     case CK_NoOp:
3764     case CK_BaseToDerived:
3765     case CK_DerivedToBase:
3766     case CK_UncheckedDerivedToBase:
3767     case CK_Dynamic:
3768     case CK_CPointerToObjCPointerCast:
3769     case CK_BlockPointerToObjCPointerCast:
3770     case CK_AnyPointerToBlockPointerCast:
3771       return EvalAddr(SubExpr, refVars, ParentDecl);
3772 
3773     case CK_ArrayToPointerDecay:
3774       return EvalVal(SubExpr, refVars, ParentDecl);
3775 
3776     default:
3777       return 0;
3778     }
3779   }
3780 
3781   case Stmt::MaterializeTemporaryExprClass:
3782     if (Expr *Result = EvalAddr(
3783                          cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
3784                                 refVars, ParentDecl))
3785       return Result;
3786 
3787     return E;
3788 
3789   // Everything else: we simply don't reason about them.
3790   default:
3791     return NULL;
3792   }
3793 }
3794 
3795 
3796 ///  EvalVal - This function is complements EvalAddr in the mutual recursion.
3797 ///   See the comments for EvalAddr for more details.
3798 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
3799                      Decl *ParentDecl) {
3800 do {
3801   // We should only be called for evaluating non-pointer expressions, or
3802   // expressions with a pointer type that are not used as references but instead
3803   // are l-values (e.g., DeclRefExpr with a pointer type).
3804 
3805   // Our "symbolic interpreter" is just a dispatch off the currently
3806   // viewed AST node.  We then recursively traverse the AST by calling
3807   // EvalAddr and EvalVal appropriately.
3808 
3809   E = E->IgnoreParens();
3810   switch (E->getStmtClass()) {
3811   case Stmt::ImplicitCastExprClass: {
3812     ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
3813     if (IE->getValueKind() == VK_LValue) {
3814       E = IE->getSubExpr();
3815       continue;
3816     }
3817     return NULL;
3818   }
3819 
3820   case Stmt::ExprWithCleanupsClass:
3821     return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl);
3822 
3823   case Stmt::DeclRefExprClass: {
3824     // When we hit a DeclRefExpr we are looking at code that refers to a
3825     // variable's name. If it's not a reference variable we check if it has
3826     // local storage within the function, and if so, return the expression.
3827     DeclRefExpr *DR = cast<DeclRefExpr>(E);
3828 
3829     if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) {
3830       // Check if it refers to itself, e.g. "int& i = i;".
3831       if (V == ParentDecl)
3832         return DR;
3833 
3834       if (V->hasLocalStorage()) {
3835         if (!V->getType()->isReferenceType())
3836           return DR;
3837 
3838         // Reference variable, follow through to the expression that
3839         // it points to.
3840         if (V->hasInit()) {
3841           // Add the reference variable to the "trail".
3842           refVars.push_back(DR);
3843           return EvalVal(V->getInit(), refVars, V);
3844         }
3845       }
3846     }
3847 
3848     return NULL;
3849   }
3850 
3851   case Stmt::UnaryOperatorClass: {
3852     // The only unary operator that make sense to handle here
3853     // is Deref.  All others don't resolve to a "name."  This includes
3854     // handling all sorts of rvalues passed to a unary operator.
3855     UnaryOperator *U = cast<UnaryOperator>(E);
3856 
3857     if (U->getOpcode() == UO_Deref)
3858       return EvalAddr(U->getSubExpr(), refVars, ParentDecl);
3859 
3860     return NULL;
3861   }
3862 
3863   case Stmt::ArraySubscriptExprClass: {
3864     // Array subscripts are potential references to data on the stack.  We
3865     // retrieve the DeclRefExpr* for the array variable if it indeed
3866     // has local storage.
3867     return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl);
3868   }
3869 
3870   case Stmt::ConditionalOperatorClass: {
3871     // For conditional operators we need to see if either the LHS or RHS are
3872     // non-NULL Expr's.  If one is non-NULL, we return it.
3873     ConditionalOperator *C = cast<ConditionalOperator>(E);
3874 
3875     // Handle the GNU extension for missing LHS.
3876     if (Expr *lhsExpr = C->getLHS())
3877       if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl))
3878         return LHS;
3879 
3880     return EvalVal(C->getRHS(), refVars, ParentDecl);
3881   }
3882 
3883   // Accesses to members are potential references to data on the stack.
3884   case Stmt::MemberExprClass: {
3885     MemberExpr *M = cast<MemberExpr>(E);
3886 
3887     // Check for indirect access.  We only want direct field accesses.
3888     if (M->isArrow())
3889       return NULL;
3890 
3891     // Check whether the member type is itself a reference, in which case
3892     // we're not going to refer to the member, but to what the member refers to.
3893     if (M->getMemberDecl()->getType()->isReferenceType())
3894       return NULL;
3895 
3896     return EvalVal(M->getBase(), refVars, ParentDecl);
3897   }
3898 
3899   case Stmt::MaterializeTemporaryExprClass:
3900     if (Expr *Result = EvalVal(
3901                           cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
3902                                refVars, ParentDecl))
3903       return Result;
3904 
3905     return E;
3906 
3907   default:
3908     // Check that we don't return or take the address of a reference to a
3909     // temporary. This is only useful in C++.
3910     if (!E->isTypeDependent() && E->isRValue())
3911       return E;
3912 
3913     // Everything else: we simply don't reason about them.
3914     return NULL;
3915   }
3916 } while (true);
3917 }
3918 
3919 //===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
3920 
3921 /// Check for comparisons of floating point operands using != and ==.
3922 /// Issue a warning if these are no self-comparisons, as they are not likely
3923 /// to do what the programmer intended.
3924 void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) {
3925   Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts();
3926   Expr* RightExprSansParen = RHS->IgnoreParenImpCasts();
3927 
3928   // Special case: check for x == x (which is OK).
3929   // Do not emit warnings for such cases.
3930   if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
3931     if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
3932       if (DRL->getDecl() == DRR->getDecl())
3933         return;
3934 
3935 
3936   // Special case: check for comparisons against literals that can be exactly
3937   //  represented by APFloat.  In such cases, do not emit a warning.  This
3938   //  is a heuristic: often comparison against such literals are used to
3939   //  detect if a value in a variable has not changed.  This clearly can
3940   //  lead to false negatives.
3941   if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
3942     if (FLL->isExact())
3943       return;
3944   } else
3945     if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen))
3946       if (FLR->isExact())
3947         return;
3948 
3949   // Check for comparisons with builtin types.
3950   if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
3951     if (CL->isBuiltinCall())
3952       return;
3953 
3954   if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
3955     if (CR->isBuiltinCall())
3956       return;
3957 
3958   // Emit the diagnostic.
3959   Diag(Loc, diag::warn_floatingpoint_eq)
3960     << LHS->getSourceRange() << RHS->getSourceRange();
3961 }
3962 
3963 //===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
3964 //===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
3965 
3966 namespace {
3967 
3968 /// Structure recording the 'active' range of an integer-valued
3969 /// expression.
3970 struct IntRange {
3971   /// The number of bits active in the int.
3972   unsigned Width;
3973 
3974   /// True if the int is known not to have negative values.
3975   bool NonNegative;
3976 
3977   IntRange(unsigned Width, bool NonNegative)
3978     : Width(Width), NonNegative(NonNegative)
3979   {}
3980 
3981   /// Returns the range of the bool type.
3982   static IntRange forBoolType() {
3983     return IntRange(1, true);
3984   }
3985 
3986   /// Returns the range of an opaque value of the given integral type.
3987   static IntRange forValueOfType(ASTContext &C, QualType T) {
3988     return forValueOfCanonicalType(C,
3989                           T->getCanonicalTypeInternal().getTypePtr());
3990   }
3991 
3992   /// Returns the range of an opaque value of a canonical integral type.
3993   static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
3994     assert(T->isCanonicalUnqualified());
3995 
3996     if (const VectorType *VT = dyn_cast<VectorType>(T))
3997       T = VT->getElementType().getTypePtr();
3998     if (const ComplexType *CT = dyn_cast<ComplexType>(T))
3999       T = CT->getElementType().getTypePtr();
4000 
4001     // For enum types, use the known bit width of the enumerators.
4002     if (const EnumType *ET = dyn_cast<EnumType>(T)) {
4003       EnumDecl *Enum = ET->getDecl();
4004       if (!Enum->isCompleteDefinition())
4005         return IntRange(C.getIntWidth(QualType(T, 0)), false);
4006 
4007       unsigned NumPositive = Enum->getNumPositiveBits();
4008       unsigned NumNegative = Enum->getNumNegativeBits();
4009 
4010       if (NumNegative == 0)
4011         return IntRange(NumPositive, true/*NonNegative*/);
4012       else
4013         return IntRange(std::max(NumPositive + 1, NumNegative),
4014                         false/*NonNegative*/);
4015     }
4016 
4017     const BuiltinType *BT = cast<BuiltinType>(T);
4018     assert(BT->isInteger());
4019 
4020     return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
4021   }
4022 
4023   /// Returns the "target" range of a canonical integral type, i.e.
4024   /// the range of values expressible in the type.
4025   ///
4026   /// This matches forValueOfCanonicalType except that enums have the
4027   /// full range of their type, not the range of their enumerators.
4028   static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
4029     assert(T->isCanonicalUnqualified());
4030 
4031     if (const VectorType *VT = dyn_cast<VectorType>(T))
4032       T = VT->getElementType().getTypePtr();
4033     if (const ComplexType *CT = dyn_cast<ComplexType>(T))
4034       T = CT->getElementType().getTypePtr();
4035     if (const EnumType *ET = dyn_cast<EnumType>(T))
4036       T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr();
4037 
4038     const BuiltinType *BT = cast<BuiltinType>(T);
4039     assert(BT->isInteger());
4040 
4041     return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
4042   }
4043 
4044   /// Returns the supremum of two ranges: i.e. their conservative merge.
4045   static IntRange join(IntRange L, IntRange R) {
4046     return IntRange(std::max(L.Width, R.Width),
4047                     L.NonNegative && R.NonNegative);
4048   }
4049 
4050   /// Returns the infinum of two ranges: i.e. their aggressive merge.
4051   static IntRange meet(IntRange L, IntRange R) {
4052     return IntRange(std::min(L.Width, R.Width),
4053                     L.NonNegative || R.NonNegative);
4054   }
4055 };
4056 
4057 static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value,
4058                               unsigned MaxWidth) {
4059   if (value.isSigned() && value.isNegative())
4060     return IntRange(value.getMinSignedBits(), false);
4061 
4062   if (value.getBitWidth() > MaxWidth)
4063     value = value.trunc(MaxWidth);
4064 
4065   // isNonNegative() just checks the sign bit without considering
4066   // signedness.
4067   return IntRange(value.getActiveBits(), true);
4068 }
4069 
4070 static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
4071                               unsigned MaxWidth) {
4072   if (result.isInt())
4073     return GetValueRange(C, result.getInt(), MaxWidth);
4074 
4075   if (result.isVector()) {
4076     IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
4077     for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
4078       IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
4079       R = IntRange::join(R, El);
4080     }
4081     return R;
4082   }
4083 
4084   if (result.isComplexInt()) {
4085     IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
4086     IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
4087     return IntRange::join(R, I);
4088   }
4089 
4090   // This can happen with lossless casts to intptr_t of "based" lvalues.
4091   // Assume it might use arbitrary bits.
4092   // FIXME: The only reason we need to pass the type in here is to get
4093   // the sign right on this one case.  It would be nice if APValue
4094   // preserved this.
4095   assert(result.isLValue() || result.isAddrLabelDiff());
4096   return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType());
4097 }
4098 
4099 /// Pseudo-evaluate the given integer expression, estimating the
4100 /// range of values it might take.
4101 ///
4102 /// \param MaxWidth - the width to which the value will be truncated
4103 static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
4104   E = E->IgnoreParens();
4105 
4106   // Try a full evaluation first.
4107   Expr::EvalResult result;
4108   if (E->EvaluateAsRValue(result, C))
4109     return GetValueRange(C, result.Val, E->getType(), MaxWidth);
4110 
4111   // I think we only want to look through implicit casts here; if the
4112   // user has an explicit widening cast, we should treat the value as
4113   // being of the new, wider type.
4114   if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
4115     if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue)
4116       return GetExprRange(C, CE->getSubExpr(), MaxWidth);
4117 
4118     IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
4119 
4120     bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
4121 
4122     // Assume that non-integer casts can span the full range of the type.
4123     if (!isIntegerCast)
4124       return OutputTypeRange;
4125 
4126     IntRange SubRange
4127       = GetExprRange(C, CE->getSubExpr(),
4128                      std::min(MaxWidth, OutputTypeRange.Width));
4129 
4130     // Bail out if the subexpr's range is as wide as the cast type.
4131     if (SubRange.Width >= OutputTypeRange.Width)
4132       return OutputTypeRange;
4133 
4134     // Otherwise, we take the smaller width, and we're non-negative if
4135     // either the output type or the subexpr is.
4136     return IntRange(SubRange.Width,
4137                     SubRange.NonNegative || OutputTypeRange.NonNegative);
4138   }
4139 
4140   if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
4141     // If we can fold the condition, just take that operand.
4142     bool CondResult;
4143     if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
4144       return GetExprRange(C, CondResult ? CO->getTrueExpr()
4145                                         : CO->getFalseExpr(),
4146                           MaxWidth);
4147 
4148     // Otherwise, conservatively merge.
4149     IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
4150     IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
4151     return IntRange::join(L, R);
4152   }
4153 
4154   if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
4155     switch (BO->getOpcode()) {
4156 
4157     // Boolean-valued operations are single-bit and positive.
4158     case BO_LAnd:
4159     case BO_LOr:
4160     case BO_LT:
4161     case BO_GT:
4162     case BO_LE:
4163     case BO_GE:
4164     case BO_EQ:
4165     case BO_NE:
4166       return IntRange::forBoolType();
4167 
4168     // The type of the assignments is the type of the LHS, so the RHS
4169     // is not necessarily the same type.
4170     case BO_MulAssign:
4171     case BO_DivAssign:
4172     case BO_RemAssign:
4173     case BO_AddAssign:
4174     case BO_SubAssign:
4175     case BO_XorAssign:
4176     case BO_OrAssign:
4177       // TODO: bitfields?
4178       return IntRange::forValueOfType(C, E->getType());
4179 
4180     // Simple assignments just pass through the RHS, which will have
4181     // been coerced to the LHS type.
4182     case BO_Assign:
4183       // TODO: bitfields?
4184       return GetExprRange(C, BO->getRHS(), MaxWidth);
4185 
4186     // Operations with opaque sources are black-listed.
4187     case BO_PtrMemD:
4188     case BO_PtrMemI:
4189       return IntRange::forValueOfType(C, E->getType());
4190 
4191     // Bitwise-and uses the *infinum* of the two source ranges.
4192     case BO_And:
4193     case BO_AndAssign:
4194       return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
4195                             GetExprRange(C, BO->getRHS(), MaxWidth));
4196 
4197     // Left shift gets black-listed based on a judgement call.
4198     case BO_Shl:
4199       // ...except that we want to treat '1 << (blah)' as logically
4200       // positive.  It's an important idiom.
4201       if (IntegerLiteral *I
4202             = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
4203         if (I->getValue() == 1) {
4204           IntRange R = IntRange::forValueOfType(C, E->getType());
4205           return IntRange(R.Width, /*NonNegative*/ true);
4206         }
4207       }
4208       // fallthrough
4209 
4210     case BO_ShlAssign:
4211       return IntRange::forValueOfType(C, E->getType());
4212 
4213     // Right shift by a constant can narrow its left argument.
4214     case BO_Shr:
4215     case BO_ShrAssign: {
4216       IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
4217 
4218       // If the shift amount is a positive constant, drop the width by
4219       // that much.
4220       llvm::APSInt shift;
4221       if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
4222           shift.isNonNegative()) {
4223         unsigned zext = shift.getZExtValue();
4224         if (zext >= L.Width)
4225           L.Width = (L.NonNegative ? 0 : 1);
4226         else
4227           L.Width -= zext;
4228       }
4229 
4230       return L;
4231     }
4232 
4233     // Comma acts as its right operand.
4234     case BO_Comma:
4235       return GetExprRange(C, BO->getRHS(), MaxWidth);
4236 
4237     // Black-list pointer subtractions.
4238     case BO_Sub:
4239       if (BO->getLHS()->getType()->isPointerType())
4240         return IntRange::forValueOfType(C, E->getType());
4241       break;
4242 
4243     // The width of a division result is mostly determined by the size
4244     // of the LHS.
4245     case BO_Div: {
4246       // Don't 'pre-truncate' the operands.
4247       unsigned opWidth = C.getIntWidth(E->getType());
4248       IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
4249 
4250       // If the divisor is constant, use that.
4251       llvm::APSInt divisor;
4252       if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) {
4253         unsigned log2 = divisor.logBase2(); // floor(log_2(divisor))
4254         if (log2 >= L.Width)
4255           L.Width = (L.NonNegative ? 0 : 1);
4256         else
4257           L.Width = std::min(L.Width - log2, MaxWidth);
4258         return L;
4259       }
4260 
4261       // Otherwise, just use the LHS's width.
4262       IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
4263       return IntRange(L.Width, L.NonNegative && R.NonNegative);
4264     }
4265 
4266     // The result of a remainder can't be larger than the result of
4267     // either side.
4268     case BO_Rem: {
4269       // Don't 'pre-truncate' the operands.
4270       unsigned opWidth = C.getIntWidth(E->getType());
4271       IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
4272       IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
4273 
4274       IntRange meet = IntRange::meet(L, R);
4275       meet.Width = std::min(meet.Width, MaxWidth);
4276       return meet;
4277     }
4278 
4279     // The default behavior is okay for these.
4280     case BO_Mul:
4281     case BO_Add:
4282     case BO_Xor:
4283     case BO_Or:
4284       break;
4285     }
4286 
4287     // The default case is to treat the operation as if it were closed
4288     // on the narrowest type that encompasses both operands.
4289     IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
4290     IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
4291     return IntRange::join(L, R);
4292   }
4293 
4294   if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
4295     switch (UO->getOpcode()) {
4296     // Boolean-valued operations are white-listed.
4297     case UO_LNot:
4298       return IntRange::forBoolType();
4299 
4300     // Operations with opaque sources are black-listed.
4301     case UO_Deref:
4302     case UO_AddrOf: // should be impossible
4303       return IntRange::forValueOfType(C, E->getType());
4304 
4305     default:
4306       return GetExprRange(C, UO->getSubExpr(), MaxWidth);
4307     }
4308   }
4309 
4310   if (dyn_cast<OffsetOfExpr>(E)) {
4311     IntRange::forValueOfType(C, E->getType());
4312   }
4313 
4314   if (FieldDecl *BitField = E->getSourceBitField())
4315     return IntRange(BitField->getBitWidthValue(C),
4316                     BitField->getType()->isUnsignedIntegerOrEnumerationType());
4317 
4318   return IntRange::forValueOfType(C, E->getType());
4319 }
4320 
4321 static IntRange GetExprRange(ASTContext &C, Expr *E) {
4322   return GetExprRange(C, E, C.getIntWidth(E->getType()));
4323 }
4324 
4325 /// Checks whether the given value, which currently has the given
4326 /// source semantics, has the same value when coerced through the
4327 /// target semantics.
4328 static bool IsSameFloatAfterCast(const llvm::APFloat &value,
4329                                  const llvm::fltSemantics &Src,
4330                                  const llvm::fltSemantics &Tgt) {
4331   llvm::APFloat truncated = value;
4332 
4333   bool ignored;
4334   truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
4335   truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
4336 
4337   return truncated.bitwiseIsEqual(value);
4338 }
4339 
4340 /// Checks whether the given value, which currently has the given
4341 /// source semantics, has the same value when coerced through the
4342 /// target semantics.
4343 ///
4344 /// The value might be a vector of floats (or a complex number).
4345 static bool IsSameFloatAfterCast(const APValue &value,
4346                                  const llvm::fltSemantics &Src,
4347                                  const llvm::fltSemantics &Tgt) {
4348   if (value.isFloat())
4349     return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
4350 
4351   if (value.isVector()) {
4352     for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
4353       if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
4354         return false;
4355     return true;
4356   }
4357 
4358   assert(value.isComplexFloat());
4359   return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
4360           IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
4361 }
4362 
4363 static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
4364 
4365 static bool IsZero(Sema &S, Expr *E) {
4366   // Suppress cases where we are comparing against an enum constant.
4367   if (const DeclRefExpr *DR =
4368       dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
4369     if (isa<EnumConstantDecl>(DR->getDecl()))
4370       return false;
4371 
4372   // Suppress cases where the '0' value is expanded from a macro.
4373   if (E->getLocStart().isMacroID())
4374     return false;
4375 
4376   llvm::APSInt Value;
4377   return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
4378 }
4379 
4380 static bool HasEnumType(Expr *E) {
4381   // Strip off implicit integral promotions.
4382   while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
4383     if (ICE->getCastKind() != CK_IntegralCast &&
4384         ICE->getCastKind() != CK_NoOp)
4385       break;
4386     E = ICE->getSubExpr();
4387   }
4388 
4389   return E->getType()->isEnumeralType();
4390 }
4391 
4392 static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
4393   BinaryOperatorKind op = E->getOpcode();
4394   if (E->isValueDependent())
4395     return;
4396 
4397   if (op == BO_LT && IsZero(S, E->getRHS())) {
4398     S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
4399       << "< 0" << "false" << HasEnumType(E->getLHS())
4400       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4401   } else if (op == BO_GE && IsZero(S, E->getRHS())) {
4402     S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
4403       << ">= 0" << "true" << HasEnumType(E->getLHS())
4404       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4405   } else if (op == BO_GT && IsZero(S, E->getLHS())) {
4406     S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
4407       << "0 >" << "false" << HasEnumType(E->getRHS())
4408       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4409   } else if (op == BO_LE && IsZero(S, E->getLHS())) {
4410     S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
4411       << "0 <=" << "true" << HasEnumType(E->getRHS())
4412       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4413   }
4414 }
4415 
4416 static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E,
4417                                          Expr *Constant, Expr *Other,
4418                                          llvm::APSInt Value,
4419                                          bool RhsConstant) {
4420   // 0 values are handled later by CheckTrivialUnsignedComparison().
4421   if (Value == 0)
4422     return;
4423 
4424   BinaryOperatorKind op = E->getOpcode();
4425   QualType OtherT = Other->getType();
4426   QualType ConstantT = Constant->getType();
4427   QualType CommonT = E->getLHS()->getType();
4428   if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT))
4429     return;
4430   assert((OtherT->isIntegerType() && ConstantT->isIntegerType())
4431          && "comparison with non-integer type");
4432 
4433   bool ConstantSigned = ConstantT->isSignedIntegerType();
4434   bool CommonSigned = CommonT->isSignedIntegerType();
4435 
4436   bool EqualityOnly = false;
4437 
4438   // TODO: Investigate using GetExprRange() to get tighter bounds on
4439   // on the bit ranges.
4440   IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT);
4441   unsigned OtherWidth = OtherRange.Width;
4442 
4443   if (CommonSigned) {
4444     // The common type is signed, therefore no signed to unsigned conversion.
4445     if (!OtherRange.NonNegative) {
4446       // Check that the constant is representable in type OtherT.
4447       if (ConstantSigned) {
4448         if (OtherWidth >= Value.getMinSignedBits())
4449           return;
4450       } else { // !ConstantSigned
4451         if (OtherWidth >= Value.getActiveBits() + 1)
4452           return;
4453       }
4454     } else { // !OtherSigned
4455       // Check that the constant is representable in type OtherT.
4456       // Negative values are out of range.
4457       if (ConstantSigned) {
4458         if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits())
4459           return;
4460       } else { // !ConstantSigned
4461         if (OtherWidth >= Value.getActiveBits())
4462           return;
4463       }
4464     }
4465   } else {  // !CommonSigned
4466     if (OtherRange.NonNegative) {
4467       if (OtherWidth >= Value.getActiveBits())
4468         return;
4469     } else if (!OtherRange.NonNegative && !ConstantSigned) {
4470       // Check to see if the constant is representable in OtherT.
4471       if (OtherWidth > Value.getActiveBits())
4472         return;
4473       // Check to see if the constant is equivalent to a negative value
4474       // cast to CommonT.
4475       if (S.Context.getIntWidth(ConstantT) == S.Context.getIntWidth(CommonT) &&
4476           Value.isNegative() && Value.getMinSignedBits() <= OtherWidth)
4477         return;
4478       // The constant value rests between values that OtherT can represent after
4479       // conversion.  Relational comparison still works, but equality
4480       // comparisons will be tautological.
4481       EqualityOnly = true;
4482     } else { // OtherSigned && ConstantSigned
4483       assert(0 && "Two signed types converted to unsigned types.");
4484     }
4485   }
4486 
4487   bool PositiveConstant = !ConstantSigned || Value.isNonNegative();
4488 
4489   bool IsTrue = true;
4490   if (op == BO_EQ || op == BO_NE) {
4491     IsTrue = op == BO_NE;
4492   } else if (EqualityOnly) {
4493     return;
4494   } else if (RhsConstant) {
4495     if (op == BO_GT || op == BO_GE)
4496       IsTrue = !PositiveConstant;
4497     else // op == BO_LT || op == BO_LE
4498       IsTrue = PositiveConstant;
4499   } else {
4500     if (op == BO_LT || op == BO_LE)
4501       IsTrue = !PositiveConstant;
4502     else // op == BO_GT || op == BO_GE
4503       IsTrue = PositiveConstant;
4504   }
4505 
4506   // If this is a comparison to an enum constant, include that
4507   // constant in the diagnostic.
4508   const EnumConstantDecl *ED = 0;
4509   if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Constant))
4510     ED = dyn_cast<EnumConstantDecl>(DR->getDecl());
4511 
4512   SmallString<64> PrettySourceValue;
4513   llvm::raw_svector_ostream OS(PrettySourceValue);
4514   if (ED)
4515     OS << '\'' << *ED << "' (" << Value << ")";
4516   else
4517     OS << Value;
4518 
4519   S.Diag(E->getOperatorLoc(), diag::warn_out_of_range_compare)
4520       << OS.str() << OtherT << IsTrue
4521       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4522 }
4523 
4524 /// Analyze the operands of the given comparison.  Implements the
4525 /// fallback case from AnalyzeComparison.
4526 static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
4527   AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
4528   AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
4529 }
4530 
4531 /// \brief Implements -Wsign-compare.
4532 ///
4533 /// \param E the binary operator to check for warnings
4534 static void AnalyzeComparison(Sema &S, BinaryOperator *E) {
4535   // The type the comparison is being performed in.
4536   QualType T = E->getLHS()->getType();
4537   assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
4538          && "comparison with mismatched types");
4539   if (E->isValueDependent())
4540     return AnalyzeImpConvsInComparison(S, E);
4541 
4542   Expr *LHS = E->getLHS()->IgnoreParenImpCasts();
4543   Expr *RHS = E->getRHS()->IgnoreParenImpCasts();
4544 
4545   bool IsComparisonConstant = false;
4546 
4547   // Check whether an integer constant comparison results in a value
4548   // of 'true' or 'false'.
4549   if (T->isIntegralType(S.Context)) {
4550     llvm::APSInt RHSValue;
4551     bool IsRHSIntegralLiteral =
4552       RHS->isIntegerConstantExpr(RHSValue, S.Context);
4553     llvm::APSInt LHSValue;
4554     bool IsLHSIntegralLiteral =
4555       LHS->isIntegerConstantExpr(LHSValue, S.Context);
4556     if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral)
4557         DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true);
4558     else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral)
4559       DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false);
4560     else
4561       IsComparisonConstant =
4562         (IsRHSIntegralLiteral && IsLHSIntegralLiteral);
4563   } else if (!T->hasUnsignedIntegerRepresentation())
4564       IsComparisonConstant = E->isIntegerConstantExpr(S.Context);
4565 
4566   // We don't do anything special if this isn't an unsigned integral
4567   // comparison:  we're only interested in integral comparisons, and
4568   // signed comparisons only happen in cases we don't care to warn about.
4569   //
4570   // We also don't care about value-dependent expressions or expressions
4571   // whose result is a constant.
4572   if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant)
4573     return AnalyzeImpConvsInComparison(S, E);
4574 
4575   // Check to see if one of the (unmodified) operands is of different
4576   // signedness.
4577   Expr *signedOperand, *unsignedOperand;
4578   if (LHS->getType()->hasSignedIntegerRepresentation()) {
4579     assert(!RHS->getType()->hasSignedIntegerRepresentation() &&
4580            "unsigned comparison between two signed integer expressions?");
4581     signedOperand = LHS;
4582     unsignedOperand = RHS;
4583   } else if (RHS->getType()->hasSignedIntegerRepresentation()) {
4584     signedOperand = RHS;
4585     unsignedOperand = LHS;
4586   } else {
4587     CheckTrivialUnsignedComparison(S, E);
4588     return AnalyzeImpConvsInComparison(S, E);
4589   }
4590 
4591   // Otherwise, calculate the effective range of the signed operand.
4592   IntRange signedRange = GetExprRange(S.Context, signedOperand);
4593 
4594   // Go ahead and analyze implicit conversions in the operands.  Note
4595   // that we skip the implicit conversions on both sides.
4596   AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc());
4597   AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc());
4598 
4599   // If the signed range is non-negative, -Wsign-compare won't fire,
4600   // but we should still check for comparisons which are always true
4601   // or false.
4602   if (signedRange.NonNegative)
4603     return CheckTrivialUnsignedComparison(S, E);
4604 
4605   // For (in)equality comparisons, if the unsigned operand is a
4606   // constant which cannot collide with a overflowed signed operand,
4607   // then reinterpreting the signed operand as unsigned will not
4608   // change the result of the comparison.
4609   if (E->isEqualityOp()) {
4610     unsigned comparisonWidth = S.Context.getIntWidth(T);
4611     IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
4612 
4613     // We should never be unable to prove that the unsigned operand is
4614     // non-negative.
4615     assert(unsignedRange.NonNegative && "unsigned range includes negative?");
4616 
4617     if (unsignedRange.Width < comparisonWidth)
4618       return;
4619   }
4620 
4621   S.DiagRuntimeBehavior(E->getOperatorLoc(), E,
4622     S.PDiag(diag::warn_mixed_sign_comparison)
4623       << LHS->getType() << RHS->getType()
4624       << LHS->getSourceRange() << RHS->getSourceRange());
4625 }
4626 
4627 /// Analyzes an attempt to assign the given value to a bitfield.
4628 ///
4629 /// Returns true if there was something fishy about the attempt.
4630 static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
4631                                       SourceLocation InitLoc) {
4632   assert(Bitfield->isBitField());
4633   if (Bitfield->isInvalidDecl())
4634     return false;
4635 
4636   // White-list bool bitfields.
4637   if (Bitfield->getType()->isBooleanType())
4638     return false;
4639 
4640   // Ignore value- or type-dependent expressions.
4641   if (Bitfield->getBitWidth()->isValueDependent() ||
4642       Bitfield->getBitWidth()->isTypeDependent() ||
4643       Init->isValueDependent() ||
4644       Init->isTypeDependent())
4645     return false;
4646 
4647   Expr *OriginalInit = Init->IgnoreParenImpCasts();
4648 
4649   llvm::APSInt Value;
4650   if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects))
4651     return false;
4652 
4653   unsigned OriginalWidth = Value.getBitWidth();
4654   unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context);
4655 
4656   if (OriginalWidth <= FieldWidth)
4657     return false;
4658 
4659   // Compute the value which the bitfield will contain.
4660   llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
4661   TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType());
4662 
4663   // Check whether the stored value is equal to the original value.
4664   TruncatedValue = TruncatedValue.extend(OriginalWidth);
4665   if (llvm::APSInt::isSameValue(Value, TruncatedValue))
4666     return false;
4667 
4668   // Special-case bitfields of width 1: booleans are naturally 0/1, and
4669   // therefore don't strictly fit into a signed bitfield of width 1.
4670   if (FieldWidth == 1 && Value == 1)
4671     return false;
4672 
4673   std::string PrettyValue = Value.toString(10);
4674   std::string PrettyTrunc = TruncatedValue.toString(10);
4675 
4676   S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
4677     << PrettyValue << PrettyTrunc << OriginalInit->getType()
4678     << Init->getSourceRange();
4679 
4680   return true;
4681 }
4682 
4683 /// Analyze the given simple or compound assignment for warning-worthy
4684 /// operations.
4685 static void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
4686   // Just recurse on the LHS.
4687   AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
4688 
4689   // We want to recurse on the RHS as normal unless we're assigning to
4690   // a bitfield.
4691   if (FieldDecl *Bitfield = E->getLHS()->getSourceBitField()) {
4692     if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
4693                                   E->getOperatorLoc())) {
4694       // Recurse, ignoring any implicit conversions on the RHS.
4695       return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
4696                                         E->getOperatorLoc());
4697     }
4698   }
4699 
4700   AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
4701 }
4702 
4703 /// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
4704 static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
4705                             SourceLocation CContext, unsigned diag,
4706                             bool pruneControlFlow = false) {
4707   if (pruneControlFlow) {
4708     S.DiagRuntimeBehavior(E->getExprLoc(), E,
4709                           S.PDiag(diag)
4710                             << SourceType << T << E->getSourceRange()
4711                             << SourceRange(CContext));
4712     return;
4713   }
4714   S.Diag(E->getExprLoc(), diag)
4715     << SourceType << T << E->getSourceRange() << SourceRange(CContext);
4716 }
4717 
4718 /// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
4719 static void DiagnoseImpCast(Sema &S, Expr *E, QualType T,
4720                             SourceLocation CContext, unsigned diag,
4721                             bool pruneControlFlow = false) {
4722   DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow);
4723 }
4724 
4725 /// Diagnose an implicit cast from a literal expression. Does not warn when the
4726 /// cast wouldn't lose information.
4727 void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
4728                                     SourceLocation CContext) {
4729   // Try to convert the literal exactly to an integer. If we can, don't warn.
4730   bool isExact = false;
4731   const llvm::APFloat &Value = FL->getValue();
4732   llvm::APSInt IntegerValue(S.Context.getIntWidth(T),
4733                             T->hasUnsignedIntegerRepresentation());
4734   if (Value.convertToInteger(IntegerValue,
4735                              llvm::APFloat::rmTowardZero, &isExact)
4736       == llvm::APFloat::opOK && isExact)
4737     return;
4738 
4739   SmallString<16> PrettySourceValue;
4740   Value.toString(PrettySourceValue);
4741   SmallString<16> PrettyTargetValue;
4742   if (T->isSpecificBuiltinType(BuiltinType::Bool))
4743     PrettyTargetValue = IntegerValue == 0 ? "false" : "true";
4744   else
4745     IntegerValue.toString(PrettyTargetValue);
4746 
4747   S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
4748     << FL->getType() << T.getUnqualifiedType() << PrettySourceValue
4749     << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext);
4750 }
4751 
4752 std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
4753   if (!Range.Width) return "0";
4754 
4755   llvm::APSInt ValueInRange = Value;
4756   ValueInRange.setIsSigned(!Range.NonNegative);
4757   ValueInRange = ValueInRange.trunc(Range.Width);
4758   return ValueInRange.toString(10);
4759 }
4760 
4761 static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) {
4762   if (!isa<ImplicitCastExpr>(Ex))
4763     return false;
4764 
4765   Expr *InnerE = Ex->IgnoreParenImpCasts();
4766   const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr();
4767   const Type *Source =
4768     S.Context.getCanonicalType(InnerE->getType()).getTypePtr();
4769   if (Target->isDependentType())
4770     return false;
4771 
4772   const BuiltinType *FloatCandidateBT =
4773     dyn_cast<BuiltinType>(ToBool ? Source : Target);
4774   const Type *BoolCandidateType = ToBool ? Target : Source;
4775 
4776   return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) &&
4777           FloatCandidateBT && (FloatCandidateBT->isFloatingPoint()));
4778 }
4779 
4780 void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall,
4781                                       SourceLocation CC) {
4782   unsigned NumArgs = TheCall->getNumArgs();
4783   for (unsigned i = 0; i < NumArgs; ++i) {
4784     Expr *CurrA = TheCall->getArg(i);
4785     if (!IsImplicitBoolFloatConversion(S, CurrA, true))
4786       continue;
4787 
4788     bool IsSwapped = ((i > 0) &&
4789         IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false));
4790     IsSwapped |= ((i < (NumArgs - 1)) &&
4791         IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false));
4792     if (IsSwapped) {
4793       // Warn on this floating-point to bool conversion.
4794       DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(),
4795                       CurrA->getType(), CC,
4796                       diag::warn_impcast_floating_point_to_bool);
4797     }
4798   }
4799 }
4800 
4801 void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
4802                              SourceLocation CC, bool *ICContext = 0) {
4803   if (E->isTypeDependent() || E->isValueDependent()) return;
4804 
4805   const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
4806   const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
4807   if (Source == Target) return;
4808   if (Target->isDependentType()) return;
4809 
4810   // If the conversion context location is invalid don't complain. We also
4811   // don't want to emit a warning if the issue occurs from the expansion of
4812   // a system macro. The problem is that 'getSpellingLoc()' is slow, so we
4813   // delay this check as long as possible. Once we detect we are in that
4814   // scenario, we just return.
4815   if (CC.isInvalid())
4816     return;
4817 
4818   // Diagnose implicit casts to bool.
4819   if (Target->isSpecificBuiltinType(BuiltinType::Bool)) {
4820     if (isa<StringLiteral>(E))
4821       // Warn on string literal to bool.  Checks for string literals in logical
4822       // expressions, for instances, assert(0 && "error here"), is prevented
4823       // by a check in AnalyzeImplicitConversions().
4824       return DiagnoseImpCast(S, E, T, CC,
4825                              diag::warn_impcast_string_literal_to_bool);
4826     if (Source->isFunctionType()) {
4827       // Warn on function to bool. Checks free functions and static member
4828       // functions. Weakly imported functions are excluded from the check,
4829       // since it's common to test their value to check whether the linker
4830       // found a definition for them.
4831       ValueDecl *D = 0;
4832       if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) {
4833         D = R->getDecl();
4834       } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) {
4835         D = M->getMemberDecl();
4836       }
4837 
4838       if (D && !D->isWeak()) {
4839         if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) {
4840           S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool)
4841             << F << E->getSourceRange() << SourceRange(CC);
4842           S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence)
4843             << FixItHint::CreateInsertion(E->getExprLoc(), "&");
4844           QualType ReturnType;
4845           UnresolvedSet<4> NonTemplateOverloads;
4846           S.isExprCallable(*E, ReturnType, NonTemplateOverloads);
4847           if (!ReturnType.isNull()
4848               && ReturnType->isSpecificBuiltinType(BuiltinType::Bool))
4849             S.Diag(E->getExprLoc(), diag::note_function_to_bool_call)
4850               << FixItHint::CreateInsertion(
4851                  S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()");
4852           return;
4853         }
4854       }
4855     }
4856   }
4857 
4858   // Strip vector types.
4859   if (isa<VectorType>(Source)) {
4860     if (!isa<VectorType>(Target)) {
4861       if (S.SourceMgr.isInSystemMacro(CC))
4862         return;
4863       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
4864     }
4865 
4866     // If the vector cast is cast between two vectors of the same size, it is
4867     // a bitcast, not a conversion.
4868     if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target))
4869       return;
4870 
4871     Source = cast<VectorType>(Source)->getElementType().getTypePtr();
4872     Target = cast<VectorType>(Target)->getElementType().getTypePtr();
4873   }
4874 
4875   // Strip complex types.
4876   if (isa<ComplexType>(Source)) {
4877     if (!isa<ComplexType>(Target)) {
4878       if (S.SourceMgr.isInSystemMacro(CC))
4879         return;
4880 
4881       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
4882     }
4883 
4884     Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
4885     Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
4886   }
4887 
4888   const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
4889   const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
4890 
4891   // If the source is floating point...
4892   if (SourceBT && SourceBT->isFloatingPoint()) {
4893     // ...and the target is floating point...
4894     if (TargetBT && TargetBT->isFloatingPoint()) {
4895       // ...then warn if we're dropping FP rank.
4896 
4897       // Builtin FP kinds are ordered by increasing FP rank.
4898       if (SourceBT->getKind() > TargetBT->getKind()) {
4899         // Don't warn about float constants that are precisely
4900         // representable in the target type.
4901         Expr::EvalResult result;
4902         if (E->EvaluateAsRValue(result, S.Context)) {
4903           // Value might be a float, a float vector, or a float complex.
4904           if (IsSameFloatAfterCast(result.Val,
4905                    S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
4906                    S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
4907             return;
4908         }
4909 
4910         if (S.SourceMgr.isInSystemMacro(CC))
4911           return;
4912 
4913         DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
4914       }
4915       return;
4916     }
4917 
4918     // If the target is integral, always warn.
4919     if (TargetBT && TargetBT->isInteger()) {
4920       if (S.SourceMgr.isInSystemMacro(CC))
4921         return;
4922 
4923       Expr *InnerE = E->IgnoreParenImpCasts();
4924       // We also want to warn on, e.g., "int i = -1.234"
4925       if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE))
4926         if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus)
4927           InnerE = UOp->getSubExpr()->IgnoreParenImpCasts();
4928 
4929       if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
4930         DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
4931       } else {
4932         DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
4933       }
4934     }
4935 
4936     // If the target is bool, warn if expr is a function or method call.
4937     if (Target->isSpecificBuiltinType(BuiltinType::Bool) &&
4938         isa<CallExpr>(E)) {
4939       // Check last argument of function call to see if it is an
4940       // implicit cast from a type matching the type the result
4941       // is being cast to.
4942       CallExpr *CEx = cast<CallExpr>(E);
4943       unsigned NumArgs = CEx->getNumArgs();
4944       if (NumArgs > 0) {
4945         Expr *LastA = CEx->getArg(NumArgs - 1);
4946         Expr *InnerE = LastA->IgnoreParenImpCasts();
4947         const Type *InnerType =
4948           S.Context.getCanonicalType(InnerE->getType()).getTypePtr();
4949         if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) {
4950           // Warn on this floating-point to bool conversion
4951           DiagnoseImpCast(S, E, T, CC,
4952                           diag::warn_impcast_floating_point_to_bool);
4953         }
4954       }
4955     }
4956     return;
4957   }
4958 
4959   if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)
4960            == Expr::NPCK_GNUNull) && !Target->isAnyPointerType()
4961       && !Target->isBlockPointerType() && !Target->isMemberPointerType()
4962       && Target->isScalarType() && !Target->isNullPtrType()) {
4963     SourceLocation Loc = E->getSourceRange().getBegin();
4964     if (Loc.isMacroID())
4965       Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first;
4966     if (!Loc.isMacroID() || CC.isMacroID())
4967       S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer)
4968           << T << clang::SourceRange(CC)
4969           << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T));
4970   }
4971 
4972   if (!Source->isIntegerType() || !Target->isIntegerType())
4973     return;
4974 
4975   // TODO: remove this early return once the false positives for constant->bool
4976   // in templates, macros, etc, are reduced or removed.
4977   if (Target->isSpecificBuiltinType(BuiltinType::Bool))
4978     return;
4979 
4980   IntRange SourceRange = GetExprRange(S.Context, E);
4981   IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
4982 
4983   if (SourceRange.Width > TargetRange.Width) {
4984     // If the source is a constant, use a default-on diagnostic.
4985     // TODO: this should happen for bitfield stores, too.
4986     llvm::APSInt Value(32);
4987     if (E->isIntegerConstantExpr(Value, S.Context)) {
4988       if (S.SourceMgr.isInSystemMacro(CC))
4989         return;
4990 
4991       std::string PrettySourceValue = Value.toString(10);
4992       std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
4993 
4994       S.DiagRuntimeBehavior(E->getExprLoc(), E,
4995         S.PDiag(diag::warn_impcast_integer_precision_constant)
4996             << PrettySourceValue << PrettyTargetValue
4997             << E->getType() << T << E->getSourceRange()
4998             << clang::SourceRange(CC));
4999       return;
5000     }
5001 
5002     // People want to build with -Wshorten-64-to-32 and not -Wconversion.
5003     if (S.SourceMgr.isInSystemMacro(CC))
5004       return;
5005 
5006     if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64)
5007       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32,
5008                              /* pruneControlFlow */ true);
5009     return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
5010   }
5011 
5012   if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
5013       (!TargetRange.NonNegative && SourceRange.NonNegative &&
5014        SourceRange.Width == TargetRange.Width)) {
5015 
5016     if (S.SourceMgr.isInSystemMacro(CC))
5017       return;
5018 
5019     unsigned DiagID = diag::warn_impcast_integer_sign;
5020 
5021     // Traditionally, gcc has warned about this under -Wsign-compare.
5022     // We also want to warn about it in -Wconversion.
5023     // So if -Wconversion is off, use a completely identical diagnostic
5024     // in the sign-compare group.
5025     // The conditional-checking code will
5026     if (ICContext) {
5027       DiagID = diag::warn_impcast_integer_sign_conditional;
5028       *ICContext = true;
5029     }
5030 
5031     return DiagnoseImpCast(S, E, T, CC, DiagID);
5032   }
5033 
5034   // Diagnose conversions between different enumeration types.
5035   // In C, we pretend that the type of an EnumConstantDecl is its enumeration
5036   // type, to give us better diagnostics.
5037   QualType SourceType = E->getType();
5038   if (!S.getLangOpts().CPlusPlus) {
5039     if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
5040       if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
5041         EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
5042         SourceType = S.Context.getTypeDeclType(Enum);
5043         Source = S.Context.getCanonicalType(SourceType).getTypePtr();
5044       }
5045   }
5046 
5047   if (const EnumType *SourceEnum = Source->getAs<EnumType>())
5048     if (const EnumType *TargetEnum = Target->getAs<EnumType>())
5049       if (SourceEnum->getDecl()->hasNameForLinkage() &&
5050           TargetEnum->getDecl()->hasNameForLinkage() &&
5051           SourceEnum != TargetEnum) {
5052         if (S.SourceMgr.isInSystemMacro(CC))
5053           return;
5054 
5055         return DiagnoseImpCast(S, E, SourceType, T, CC,
5056                                diag::warn_impcast_different_enum_types);
5057       }
5058 
5059   return;
5060 }
5061 
5062 void CheckConditionalOperator(Sema &S, ConditionalOperator *E,
5063                               SourceLocation CC, QualType T);
5064 
5065 void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
5066                              SourceLocation CC, bool &ICContext) {
5067   E = E->IgnoreParenImpCasts();
5068 
5069   if (isa<ConditionalOperator>(E))
5070     return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T);
5071 
5072   AnalyzeImplicitConversions(S, E, CC);
5073   if (E->getType() != T)
5074     return CheckImplicitConversion(S, E, T, CC, &ICContext);
5075   return;
5076 }
5077 
5078 void CheckConditionalOperator(Sema &S, ConditionalOperator *E,
5079                               SourceLocation CC, QualType T) {
5080   AnalyzeImplicitConversions(S, E->getCond(), CC);
5081 
5082   bool Suspicious = false;
5083   CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
5084   CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
5085 
5086   // If -Wconversion would have warned about either of the candidates
5087   // for a signedness conversion to the context type...
5088   if (!Suspicious) return;
5089 
5090   // ...but it's currently ignored...
5091   if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
5092                                  CC))
5093     return;
5094 
5095   // ...then check whether it would have warned about either of the
5096   // candidates for a signedness conversion to the condition type.
5097   if (E->getType() == T) return;
5098 
5099   Suspicious = false;
5100   CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
5101                           E->getType(), CC, &Suspicious);
5102   if (!Suspicious)
5103     CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
5104                             E->getType(), CC, &Suspicious);
5105 }
5106 
5107 /// AnalyzeImplicitConversions - Find and report any interesting
5108 /// implicit conversions in the given expression.  There are a couple
5109 /// of competing diagnostics here, -Wconversion and -Wsign-compare.
5110 void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
5111   QualType T = OrigE->getType();
5112   Expr *E = OrigE->IgnoreParenImpCasts();
5113 
5114   if (E->isTypeDependent() || E->isValueDependent())
5115     return;
5116 
5117   // For conditional operators, we analyze the arguments as if they
5118   // were being fed directly into the output.
5119   if (isa<ConditionalOperator>(E)) {
5120     ConditionalOperator *CO = cast<ConditionalOperator>(E);
5121     CheckConditionalOperator(S, CO, CC, T);
5122     return;
5123   }
5124 
5125   // Check implicit argument conversions for function calls.
5126   if (CallExpr *Call = dyn_cast<CallExpr>(E))
5127     CheckImplicitArgumentConversions(S, Call, CC);
5128 
5129   // Go ahead and check any implicit conversions we might have skipped.
5130   // The non-canonical typecheck is just an optimization;
5131   // CheckImplicitConversion will filter out dead implicit conversions.
5132   if (E->getType() != T)
5133     CheckImplicitConversion(S, E, T, CC);
5134 
5135   // Now continue drilling into this expression.
5136 
5137   // Skip past explicit casts.
5138   if (isa<ExplicitCastExpr>(E)) {
5139     E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
5140     return AnalyzeImplicitConversions(S, E, CC);
5141   }
5142 
5143   if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
5144     // Do a somewhat different check with comparison operators.
5145     if (BO->isComparisonOp())
5146       return AnalyzeComparison(S, BO);
5147 
5148     // And with simple assignments.
5149     if (BO->getOpcode() == BO_Assign)
5150       return AnalyzeAssignment(S, BO);
5151   }
5152 
5153   // These break the otherwise-useful invariant below.  Fortunately,
5154   // we don't really need to recurse into them, because any internal
5155   // expressions should have been analyzed already when they were
5156   // built into statements.
5157   if (isa<StmtExpr>(E)) return;
5158 
5159   // Don't descend into unevaluated contexts.
5160   if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
5161 
5162   // Now just recurse over the expression's children.
5163   CC = E->getExprLoc();
5164   BinaryOperator *BO = dyn_cast<BinaryOperator>(E);
5165   bool IsLogicalOperator = BO && BO->isLogicalOp();
5166   for (Stmt::child_range I = E->children(); I; ++I) {
5167     Expr *ChildExpr = dyn_cast_or_null<Expr>(*I);
5168     if (!ChildExpr)
5169       continue;
5170 
5171     if (IsLogicalOperator &&
5172         isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts()))
5173       // Ignore checking string literals that are in logical operators.
5174       continue;
5175     AnalyzeImplicitConversions(S, ChildExpr, CC);
5176   }
5177 }
5178 
5179 } // end anonymous namespace
5180 
5181 /// Diagnoses "dangerous" implicit conversions within the given
5182 /// expression (which is a full expression).  Implements -Wconversion
5183 /// and -Wsign-compare.
5184 ///
5185 /// \param CC the "context" location of the implicit conversion, i.e.
5186 ///   the most location of the syntactic entity requiring the implicit
5187 ///   conversion
5188 void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
5189   // Don't diagnose in unevaluated contexts.
5190   if (isUnevaluatedContext())
5191     return;
5192 
5193   // Don't diagnose for value- or type-dependent expressions.
5194   if (E->isTypeDependent() || E->isValueDependent())
5195     return;
5196 
5197   // Check for array bounds violations in cases where the check isn't triggered
5198   // elsewhere for other Expr types (like BinaryOperators), e.g. when an
5199   // ArraySubscriptExpr is on the RHS of a variable initialization.
5200   CheckArrayAccess(E);
5201 
5202   // This is not the right CC for (e.g.) a variable initialization.
5203   AnalyzeImplicitConversions(*this, E, CC);
5204 }
5205 
5206 /// Diagnose when expression is an integer constant expression and its evaluation
5207 /// results in integer overflow
5208 void Sema::CheckForIntOverflow (Expr *E) {
5209   if (isa<BinaryOperator>(E->IgnoreParens())) {
5210     llvm::SmallVector<PartialDiagnosticAt, 4> Diags;
5211     E->EvaluateForOverflow(Context, &Diags);
5212   }
5213 }
5214 
5215 namespace {
5216 /// \brief Visitor for expressions which looks for unsequenced operations on the
5217 /// same object.
5218 class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> {
5219   /// \brief A tree of sequenced regions within an expression. Two regions are
5220   /// unsequenced if one is an ancestor or a descendent of the other. When we
5221   /// finish processing an expression with sequencing, such as a comma
5222   /// expression, we fold its tree nodes into its parent, since they are
5223   /// unsequenced with respect to nodes we will visit later.
5224   class SequenceTree {
5225     struct Value {
5226       explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {}
5227       unsigned Parent : 31;
5228       bool Merged : 1;
5229     };
5230     llvm::SmallVector<Value, 8> Values;
5231 
5232   public:
5233     /// \brief A region within an expression which may be sequenced with respect
5234     /// to some other region.
5235     class Seq {
5236       explicit Seq(unsigned N) : Index(N) {}
5237       unsigned Index;
5238       friend class SequenceTree;
5239     public:
5240       Seq() : Index(0) {}
5241     };
5242 
5243     SequenceTree() { Values.push_back(Value(0)); }
5244     Seq root() const { return Seq(0); }
5245 
5246     /// \brief Create a new sequence of operations, which is an unsequenced
5247     /// subset of \p Parent. This sequence of operations is sequenced with
5248     /// respect to other children of \p Parent.
5249     Seq allocate(Seq Parent) {
5250       Values.push_back(Value(Parent.Index));
5251       return Seq(Values.size() - 1);
5252     }
5253 
5254     /// \brief Merge a sequence of operations into its parent.
5255     void merge(Seq S) {
5256       Values[S.Index].Merged = true;
5257     }
5258 
5259     /// \brief Determine whether two operations are unsequenced. This operation
5260     /// is asymmetric: \p Cur should be the more recent sequence, and \p Old
5261     /// should have been merged into its parent as appropriate.
5262     bool isUnsequenced(Seq Cur, Seq Old) {
5263       unsigned C = representative(Cur.Index);
5264       unsigned Target = representative(Old.Index);
5265       while (C >= Target) {
5266         if (C == Target)
5267           return true;
5268         C = Values[C].Parent;
5269       }
5270       return false;
5271     }
5272 
5273   private:
5274     /// \brief Pick a representative for a sequence.
5275     unsigned representative(unsigned K) {
5276       if (Values[K].Merged)
5277         // Perform path compression as we go.
5278         return Values[K].Parent = representative(Values[K].Parent);
5279       return K;
5280     }
5281   };
5282 
5283   /// An object for which we can track unsequenced uses.
5284   typedef NamedDecl *Object;
5285 
5286   /// Different flavors of object usage which we track. We only track the
5287   /// least-sequenced usage of each kind.
5288   enum UsageKind {
5289     /// A read of an object. Multiple unsequenced reads are OK.
5290     UK_Use,
5291     /// A modification of an object which is sequenced before the value
5292     /// computation of the expression, such as ++n.
5293     UK_ModAsValue,
5294     /// A modification of an object which is not sequenced before the value
5295     /// computation of the expression, such as n++.
5296     UK_ModAsSideEffect,
5297 
5298     UK_Count = UK_ModAsSideEffect + 1
5299   };
5300 
5301   struct Usage {
5302     Usage() : Use(0), Seq() {}
5303     Expr *Use;
5304     SequenceTree::Seq Seq;
5305   };
5306 
5307   struct UsageInfo {
5308     UsageInfo() : Diagnosed(false) {}
5309     Usage Uses[UK_Count];
5310     /// Have we issued a diagnostic for this variable already?
5311     bool Diagnosed;
5312   };
5313   typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap;
5314 
5315   Sema &SemaRef;
5316   /// Sequenced regions within the expression.
5317   SequenceTree Tree;
5318   /// Declaration modifications and references which we have seen.
5319   UsageInfoMap UsageMap;
5320   /// The region we are currently within.
5321   SequenceTree::Seq Region;
5322   /// Filled in with declarations which were modified as a side-effect
5323   /// (that is, post-increment operations).
5324   llvm::SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect;
5325   /// Expressions to check later. We defer checking these to reduce
5326   /// stack usage.
5327   llvm::SmallVectorImpl<Expr*> &WorkList;
5328 
5329   /// RAII object wrapping the visitation of a sequenced subexpression of an
5330   /// expression. At the end of this process, the side-effects of the evaluation
5331   /// become sequenced with respect to the value computation of the result, so
5332   /// we downgrade any UK_ModAsSideEffect within the evaluation to
5333   /// UK_ModAsValue.
5334   struct SequencedSubexpression {
5335     SequencedSubexpression(SequenceChecker &Self)
5336       : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) {
5337       Self.ModAsSideEffect = &ModAsSideEffect;
5338     }
5339     ~SequencedSubexpression() {
5340       for (unsigned I = 0, E = ModAsSideEffect.size(); I != E; ++I) {
5341         UsageInfo &U = Self.UsageMap[ModAsSideEffect[I].first];
5342         U.Uses[UK_ModAsSideEffect] = ModAsSideEffect[I].second;
5343         Self.addUsage(U, ModAsSideEffect[I].first,
5344                       ModAsSideEffect[I].second.Use, UK_ModAsValue);
5345       }
5346       Self.ModAsSideEffect = OldModAsSideEffect;
5347     }
5348 
5349     SequenceChecker &Self;
5350     llvm::SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect;
5351     llvm::SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect;
5352   };
5353 
5354   /// \brief Find the object which is produced by the specified expression,
5355   /// if any.
5356   Object getObject(Expr *E, bool Mod) const {
5357     E = E->IgnoreParenCasts();
5358     if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
5359       if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec))
5360         return getObject(UO->getSubExpr(), Mod);
5361     } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
5362       if (BO->getOpcode() == BO_Comma)
5363         return getObject(BO->getRHS(), Mod);
5364       if (Mod && BO->isAssignmentOp())
5365         return getObject(BO->getLHS(), Mod);
5366     } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) {
5367       // FIXME: Check for more interesting cases, like "x.n = ++x.n".
5368       if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts()))
5369         return ME->getMemberDecl();
5370     } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
5371       // FIXME: If this is a reference, map through to its value.
5372       return DRE->getDecl();
5373     return 0;
5374   }
5375 
5376   /// \brief Note that an object was modified or used by an expression.
5377   void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) {
5378     Usage &U = UI.Uses[UK];
5379     if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) {
5380       if (UK == UK_ModAsSideEffect && ModAsSideEffect)
5381         ModAsSideEffect->push_back(std::make_pair(O, U));
5382       U.Use = Ref;
5383       U.Seq = Region;
5384     }
5385   }
5386   /// \brief Check whether a modification or use conflicts with a prior usage.
5387   void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind,
5388                   bool IsModMod) {
5389     if (UI.Diagnosed)
5390       return;
5391 
5392     const Usage &U = UI.Uses[OtherKind];
5393     if (!U.Use || !Tree.isUnsequenced(Region, U.Seq))
5394       return;
5395 
5396     Expr *Mod = U.Use;
5397     Expr *ModOrUse = Ref;
5398     if (OtherKind == UK_Use)
5399       std::swap(Mod, ModOrUse);
5400 
5401     SemaRef.Diag(Mod->getExprLoc(),
5402                  IsModMod ? diag::warn_unsequenced_mod_mod
5403                           : diag::warn_unsequenced_mod_use)
5404       << O << SourceRange(ModOrUse->getExprLoc());
5405     UI.Diagnosed = true;
5406   }
5407 
5408   void notePreUse(Object O, Expr *Use) {
5409     UsageInfo &U = UsageMap[O];
5410     // Uses conflict with other modifications.
5411     checkUsage(O, U, Use, UK_ModAsValue, false);
5412   }
5413   void notePostUse(Object O, Expr *Use) {
5414     UsageInfo &U = UsageMap[O];
5415     checkUsage(O, U, Use, UK_ModAsSideEffect, false);
5416     addUsage(U, O, Use, UK_Use);
5417   }
5418 
5419   void notePreMod(Object O, Expr *Mod) {
5420     UsageInfo &U = UsageMap[O];
5421     // Modifications conflict with other modifications and with uses.
5422     checkUsage(O, U, Mod, UK_ModAsValue, true);
5423     checkUsage(O, U, Mod, UK_Use, false);
5424   }
5425   void notePostMod(Object O, Expr *Use, UsageKind UK) {
5426     UsageInfo &U = UsageMap[O];
5427     checkUsage(O, U, Use, UK_ModAsSideEffect, true);
5428     addUsage(U, O, Use, UK);
5429   }
5430 
5431 public:
5432   SequenceChecker(Sema &S, Expr *E,
5433                   llvm::SmallVectorImpl<Expr*> &WorkList)
5434     : EvaluatedExprVisitor<SequenceChecker>(S.Context), SemaRef(S),
5435       Region(Tree.root()), ModAsSideEffect(0), WorkList(WorkList) {
5436     Visit(E);
5437   }
5438 
5439   void VisitStmt(Stmt *S) {
5440     // Skip all statements which aren't expressions for now.
5441   }
5442 
5443   void VisitExpr(Expr *E) {
5444     // By default, just recurse to evaluated subexpressions.
5445     EvaluatedExprVisitor<SequenceChecker>::VisitStmt(E);
5446   }
5447 
5448   void VisitCastExpr(CastExpr *E) {
5449     Object O = Object();
5450     if (E->getCastKind() == CK_LValueToRValue)
5451       O = getObject(E->getSubExpr(), false);
5452 
5453     if (O)
5454       notePreUse(O, E);
5455     VisitExpr(E);
5456     if (O)
5457       notePostUse(O, E);
5458   }
5459 
5460   void VisitBinComma(BinaryOperator *BO) {
5461     // C++11 [expr.comma]p1:
5462     //   Every value computation and side effect associated with the left
5463     //   expression is sequenced before every value computation and side
5464     //   effect associated with the right expression.
5465     SequenceTree::Seq LHS = Tree.allocate(Region);
5466     SequenceTree::Seq RHS = Tree.allocate(Region);
5467     SequenceTree::Seq OldRegion = Region;
5468 
5469     {
5470       SequencedSubexpression SeqLHS(*this);
5471       Region = LHS;
5472       Visit(BO->getLHS());
5473     }
5474 
5475     Region = RHS;
5476     Visit(BO->getRHS());
5477 
5478     Region = OldRegion;
5479 
5480     // Forget that LHS and RHS are sequenced. They are both unsequenced
5481     // with respect to other stuff.
5482     Tree.merge(LHS);
5483     Tree.merge(RHS);
5484   }
5485 
5486   void VisitBinAssign(BinaryOperator *BO) {
5487     // The modification is sequenced after the value computation of the LHS
5488     // and RHS, so check it before inspecting the operands and update the
5489     // map afterwards.
5490     Object O = getObject(BO->getLHS(), true);
5491     if (!O)
5492       return VisitExpr(BO);
5493 
5494     notePreMod(O, BO);
5495 
5496     // C++11 [expr.ass]p7:
5497     //   E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated
5498     //   only once.
5499     //
5500     // Therefore, for a compound assignment operator, O is considered used
5501     // everywhere except within the evaluation of E1 itself.
5502     if (isa<CompoundAssignOperator>(BO))
5503       notePreUse(O, BO);
5504 
5505     Visit(BO->getLHS());
5506 
5507     if (isa<CompoundAssignOperator>(BO))
5508       notePostUse(O, BO);
5509 
5510     Visit(BO->getRHS());
5511 
5512     notePostMod(O, BO, UK_ModAsValue);
5513   }
5514   void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) {
5515     VisitBinAssign(CAO);
5516   }
5517 
5518   void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); }
5519   void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); }
5520   void VisitUnaryPreIncDec(UnaryOperator *UO) {
5521     Object O = getObject(UO->getSubExpr(), true);
5522     if (!O)
5523       return VisitExpr(UO);
5524 
5525     notePreMod(O, UO);
5526     Visit(UO->getSubExpr());
5527     notePostMod(O, UO, UK_ModAsValue);
5528   }
5529 
5530   void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); }
5531   void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); }
5532   void VisitUnaryPostIncDec(UnaryOperator *UO) {
5533     Object O = getObject(UO->getSubExpr(), true);
5534     if (!O)
5535       return VisitExpr(UO);
5536 
5537     notePreMod(O, UO);
5538     Visit(UO->getSubExpr());
5539     notePostMod(O, UO, UK_ModAsSideEffect);
5540   }
5541 
5542   /// Don't visit the RHS of '&&' or '||' if it might not be evaluated.
5543   void VisitBinLOr(BinaryOperator *BO) {
5544     // The side-effects of the LHS of an '&&' are sequenced before the
5545     // value computation of the RHS, and hence before the value computation
5546     // of the '&&' itself, unless the LHS evaluates to zero. We treat them
5547     // as if they were unconditionally sequenced.
5548     {
5549       SequencedSubexpression Sequenced(*this);
5550       Visit(BO->getLHS());
5551     }
5552 
5553     bool Result;
5554     if (!BO->getLHS()->isValueDependent() &&
5555         BO->getLHS()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) {
5556       if (!Result)
5557         Visit(BO->getRHS());
5558     } else {
5559       // Check for unsequenced operations in the RHS, treating it as an
5560       // entirely separate evaluation.
5561       //
5562       // FIXME: If there are operations in the RHS which are unsequenced
5563       // with respect to operations outside the RHS, and those operations
5564       // are unconditionally evaluated, diagnose them.
5565       WorkList.push_back(BO->getRHS());
5566     }
5567   }
5568   void VisitBinLAnd(BinaryOperator *BO) {
5569     {
5570       SequencedSubexpression Sequenced(*this);
5571       Visit(BO->getLHS());
5572     }
5573 
5574     bool Result;
5575     if (!BO->getLHS()->isValueDependent() &&
5576         BO->getLHS()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) {
5577       if (Result)
5578         Visit(BO->getRHS());
5579     } else {
5580       WorkList.push_back(BO->getRHS());
5581     }
5582   }
5583 
5584   // Only visit the condition, unless we can be sure which subexpression will
5585   // be chosen.
5586   void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) {
5587     SequencedSubexpression Sequenced(*this);
5588     Visit(CO->getCond());
5589 
5590     bool Result;
5591     if (!CO->getCond()->isValueDependent() &&
5592         CO->getCond()->EvaluateAsBooleanCondition(Result, SemaRef.Context))
5593       Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr());
5594     else {
5595       WorkList.push_back(CO->getTrueExpr());
5596       WorkList.push_back(CO->getFalseExpr());
5597     }
5598   }
5599 
5600   void VisitCXXConstructExpr(CXXConstructExpr *CCE) {
5601     if (!CCE->isListInitialization())
5602       return VisitExpr(CCE);
5603 
5604     // In C++11, list initializations are sequenced.
5605     llvm::SmallVector<SequenceTree::Seq, 32> Elts;
5606     SequenceTree::Seq Parent = Region;
5607     for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(),
5608                                         E = CCE->arg_end();
5609          I != E; ++I) {
5610       Region = Tree.allocate(Parent);
5611       Elts.push_back(Region);
5612       Visit(*I);
5613     }
5614 
5615     // Forget that the initializers are sequenced.
5616     Region = Parent;
5617     for (unsigned I = 0; I < Elts.size(); ++I)
5618       Tree.merge(Elts[I]);
5619   }
5620 
5621   void VisitInitListExpr(InitListExpr *ILE) {
5622     if (!SemaRef.getLangOpts().CPlusPlus11)
5623       return VisitExpr(ILE);
5624 
5625     // In C++11, list initializations are sequenced.
5626     llvm::SmallVector<SequenceTree::Seq, 32> Elts;
5627     SequenceTree::Seq Parent = Region;
5628     for (unsigned I = 0; I < ILE->getNumInits(); ++I) {
5629       Expr *E = ILE->getInit(I);
5630       if (!E) continue;
5631       Region = Tree.allocate(Parent);
5632       Elts.push_back(Region);
5633       Visit(E);
5634     }
5635 
5636     // Forget that the initializers are sequenced.
5637     Region = Parent;
5638     for (unsigned I = 0; I < Elts.size(); ++I)
5639       Tree.merge(Elts[I]);
5640   }
5641 };
5642 }
5643 
5644 void Sema::CheckUnsequencedOperations(Expr *E) {
5645   llvm::SmallVector<Expr*, 8> WorkList;
5646   WorkList.push_back(E);
5647   while (!WorkList.empty()) {
5648     Expr *Item = WorkList.back();
5649     WorkList.pop_back();
5650     SequenceChecker(*this, Item, WorkList);
5651   }
5652 }
5653 
5654 void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc,
5655                               bool IsConstexpr) {
5656   CheckImplicitConversions(E, CheckLoc);
5657   CheckUnsequencedOperations(E);
5658   if (!IsConstexpr && !E->isValueDependent())
5659     CheckForIntOverflow(E);
5660 }
5661 
5662 void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
5663                                        FieldDecl *BitField,
5664                                        Expr *Init) {
5665   (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
5666 }
5667 
5668 /// CheckParmsForFunctionDef - Check that the parameters of the given
5669 /// function are appropriate for the definition of a function. This
5670 /// takes care of any checks that cannot be performed on the
5671 /// declaration itself, e.g., that the types of each of the function
5672 /// parameters are complete.
5673 bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
5674                                     bool CheckParameterNames) {
5675   bool HasInvalidParm = false;
5676   for (; P != PEnd; ++P) {
5677     ParmVarDecl *Param = *P;
5678 
5679     // C99 6.7.5.3p4: the parameters in a parameter type list in a
5680     // function declarator that is part of a function definition of
5681     // that function shall not have incomplete type.
5682     //
5683     // This is also C++ [dcl.fct]p6.
5684     if (!Param->isInvalidDecl() &&
5685         RequireCompleteType(Param->getLocation(), Param->getType(),
5686                             diag::err_typecheck_decl_incomplete_type)) {
5687       Param->setInvalidDecl();
5688       HasInvalidParm = true;
5689     }
5690 
5691     // C99 6.9.1p5: If the declarator includes a parameter type list, the
5692     // declaration of each parameter shall include an identifier.
5693     if (CheckParameterNames &&
5694         Param->getIdentifier() == 0 &&
5695         !Param->isImplicit() &&
5696         !getLangOpts().CPlusPlus)
5697       Diag(Param->getLocation(), diag::err_parameter_name_omitted);
5698 
5699     // C99 6.7.5.3p12:
5700     //   If the function declarator is not part of a definition of that
5701     //   function, parameters may have incomplete type and may use the [*]
5702     //   notation in their sequences of declarator specifiers to specify
5703     //   variable length array types.
5704     QualType PType = Param->getOriginalType();
5705     while (const ArrayType *AT = Context.getAsArrayType(PType)) {
5706       if (AT->getSizeModifier() == ArrayType::Star) {
5707         // FIXME: This diagnostic should point the '[*]' if source-location
5708         // information is added for it.
5709         Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
5710         break;
5711       }
5712       PType= AT->getElementType();
5713     }
5714   }
5715 
5716   return HasInvalidParm;
5717 }
5718 
5719 /// CheckCastAlign - Implements -Wcast-align, which warns when a
5720 /// pointer cast increases the alignment requirements.
5721 void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
5722   // This is actually a lot of work to potentially be doing on every
5723   // cast; don't do it if we're ignoring -Wcast_align (as is the default).
5724   if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
5725                                           TRange.getBegin())
5726         == DiagnosticsEngine::Ignored)
5727     return;
5728 
5729   // Ignore dependent types.
5730   if (T->isDependentType() || Op->getType()->isDependentType())
5731     return;
5732 
5733   // Require that the destination be a pointer type.
5734   const PointerType *DestPtr = T->getAs<PointerType>();
5735   if (!DestPtr) return;
5736 
5737   // If the destination has alignment 1, we're done.
5738   QualType DestPointee = DestPtr->getPointeeType();
5739   if (DestPointee->isIncompleteType()) return;
5740   CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
5741   if (DestAlign.isOne()) return;
5742 
5743   // Require that the source be a pointer type.
5744   const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
5745   if (!SrcPtr) return;
5746   QualType SrcPointee = SrcPtr->getPointeeType();
5747 
5748   // Whitelist casts from cv void*.  We already implicitly
5749   // whitelisted casts to cv void*, since they have alignment 1.
5750   // Also whitelist casts involving incomplete types, which implicitly
5751   // includes 'void'.
5752   if (SrcPointee->isIncompleteType()) return;
5753 
5754   CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
5755   if (SrcAlign >= DestAlign) return;
5756 
5757   Diag(TRange.getBegin(), diag::warn_cast_align)
5758     << Op->getType() << T
5759     << static_cast<unsigned>(SrcAlign.getQuantity())
5760     << static_cast<unsigned>(DestAlign.getQuantity())
5761     << TRange << Op->getSourceRange();
5762 }
5763 
5764 static const Type* getElementType(const Expr *BaseExpr) {
5765   const Type* EltType = BaseExpr->getType().getTypePtr();
5766   if (EltType->isAnyPointerType())
5767     return EltType->getPointeeType().getTypePtr();
5768   else if (EltType->isArrayType())
5769     return EltType->getBaseElementTypeUnsafe();
5770   return EltType;
5771 }
5772 
5773 /// \brief Check whether this array fits the idiom of a size-one tail padded
5774 /// array member of a struct.
5775 ///
5776 /// We avoid emitting out-of-bounds access warnings for such arrays as they are
5777 /// commonly used to emulate flexible arrays in C89 code.
5778 static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size,
5779                                     const NamedDecl *ND) {
5780   if (Size != 1 || !ND) return false;
5781 
5782   const FieldDecl *FD = dyn_cast<FieldDecl>(ND);
5783   if (!FD) return false;
5784 
5785   // Don't consider sizes resulting from macro expansions or template argument
5786   // substitution to form C89 tail-padded arrays.
5787 
5788   TypeSourceInfo *TInfo = FD->getTypeSourceInfo();
5789   while (TInfo) {
5790     TypeLoc TL = TInfo->getTypeLoc();
5791     // Look through typedefs.
5792     if (TypedefTypeLoc TTL = TL.getAs<TypedefTypeLoc>()) {
5793       const TypedefNameDecl *TDL = TTL.getTypedefNameDecl();
5794       TInfo = TDL->getTypeSourceInfo();
5795       continue;
5796     }
5797     if (ConstantArrayTypeLoc CTL = TL.getAs<ConstantArrayTypeLoc>()) {
5798       const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr());
5799       if (!SizeExpr || SizeExpr->getExprLoc().isMacroID())
5800         return false;
5801     }
5802     break;
5803   }
5804 
5805   const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext());
5806   if (!RD) return false;
5807   if (RD->isUnion()) return false;
5808   if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) {
5809     if (!CRD->isStandardLayout()) return false;
5810   }
5811 
5812   // See if this is the last field decl in the record.
5813   const Decl *D = FD;
5814   while ((D = D->getNextDeclInContext()))
5815     if (isa<FieldDecl>(D))
5816       return false;
5817   return true;
5818 }
5819 
5820 void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr,
5821                             const ArraySubscriptExpr *ASE,
5822                             bool AllowOnePastEnd, bool IndexNegated) {
5823   IndexExpr = IndexExpr->IgnoreParenImpCasts();
5824   if (IndexExpr->isValueDependent())
5825     return;
5826 
5827   const Type *EffectiveType = getElementType(BaseExpr);
5828   BaseExpr = BaseExpr->IgnoreParenCasts();
5829   const ConstantArrayType *ArrayTy =
5830     Context.getAsConstantArrayType(BaseExpr->getType());
5831   if (!ArrayTy)
5832     return;
5833 
5834   llvm::APSInt index;
5835   if (!IndexExpr->EvaluateAsInt(index, Context))
5836     return;
5837   if (IndexNegated)
5838     index = -index;
5839 
5840   const NamedDecl *ND = NULL;
5841   if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
5842     ND = dyn_cast<NamedDecl>(DRE->getDecl());
5843   if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
5844     ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
5845 
5846   if (index.isUnsigned() || !index.isNegative()) {
5847     llvm::APInt size = ArrayTy->getSize();
5848     if (!size.isStrictlyPositive())
5849       return;
5850 
5851     const Type* BaseType = getElementType(BaseExpr);
5852     if (BaseType != EffectiveType) {
5853       // Make sure we're comparing apples to apples when comparing index to size
5854       uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType);
5855       uint64_t array_typesize = Context.getTypeSize(BaseType);
5856       // Handle ptrarith_typesize being zero, such as when casting to void*
5857       if (!ptrarith_typesize) ptrarith_typesize = 1;
5858       if (ptrarith_typesize != array_typesize) {
5859         // There's a cast to a different size type involved
5860         uint64_t ratio = array_typesize / ptrarith_typesize;
5861         // TODO: Be smarter about handling cases where array_typesize is not a
5862         // multiple of ptrarith_typesize
5863         if (ptrarith_typesize * ratio == array_typesize)
5864           size *= llvm::APInt(size.getBitWidth(), ratio);
5865       }
5866     }
5867 
5868     if (size.getBitWidth() > index.getBitWidth())
5869       index = index.zext(size.getBitWidth());
5870     else if (size.getBitWidth() < index.getBitWidth())
5871       size = size.zext(index.getBitWidth());
5872 
5873     // For array subscripting the index must be less than size, but for pointer
5874     // arithmetic also allow the index (offset) to be equal to size since
5875     // computing the next address after the end of the array is legal and
5876     // commonly done e.g. in C++ iterators and range-based for loops.
5877     if (AllowOnePastEnd ? index.ule(size) : index.ult(size))
5878       return;
5879 
5880     // Also don't warn for arrays of size 1 which are members of some
5881     // structure. These are often used to approximate flexible arrays in C89
5882     // code.
5883     if (IsTailPaddedMemberArray(*this, size, ND))
5884       return;
5885 
5886     // Suppress the warning if the subscript expression (as identified by the
5887     // ']' location) and the index expression are both from macro expansions
5888     // within a system header.
5889     if (ASE) {
5890       SourceLocation RBracketLoc = SourceMgr.getSpellingLoc(
5891           ASE->getRBracketLoc());
5892       if (SourceMgr.isInSystemHeader(RBracketLoc)) {
5893         SourceLocation IndexLoc = SourceMgr.getSpellingLoc(
5894             IndexExpr->getLocStart());
5895         if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc))
5896           return;
5897       }
5898     }
5899 
5900     unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds;
5901     if (ASE)
5902       DiagID = diag::warn_array_index_exceeds_bounds;
5903 
5904     DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
5905                         PDiag(DiagID) << index.toString(10, true)
5906                           << size.toString(10, true)
5907                           << (unsigned)size.getLimitedValue(~0U)
5908                           << IndexExpr->getSourceRange());
5909   } else {
5910     unsigned DiagID = diag::warn_array_index_precedes_bounds;
5911     if (!ASE) {
5912       DiagID = diag::warn_ptr_arith_precedes_bounds;
5913       if (index.isNegative()) index = -index;
5914     }
5915 
5916     DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
5917                         PDiag(DiagID) << index.toString(10, true)
5918                           << IndexExpr->getSourceRange());
5919   }
5920 
5921   if (!ND) {
5922     // Try harder to find a NamedDecl to point at in the note.
5923     while (const ArraySubscriptExpr *ASE =
5924            dyn_cast<ArraySubscriptExpr>(BaseExpr))
5925       BaseExpr = ASE->getBase()->IgnoreParenCasts();
5926     if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
5927       ND = dyn_cast<NamedDecl>(DRE->getDecl());
5928     if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
5929       ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
5930   }
5931 
5932   if (ND)
5933     DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
5934                         PDiag(diag::note_array_index_out_of_bounds)
5935                           << ND->getDeclName());
5936 }
5937 
5938 void Sema::CheckArrayAccess(const Expr *expr) {
5939   int AllowOnePastEnd = 0;
5940   while (expr) {
5941     expr = expr->IgnoreParenImpCasts();
5942     switch (expr->getStmtClass()) {
5943       case Stmt::ArraySubscriptExprClass: {
5944         const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr);
5945         CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE,
5946                          AllowOnePastEnd > 0);
5947         return;
5948       }
5949       case Stmt::UnaryOperatorClass: {
5950         // Only unwrap the * and & unary operators
5951         const UnaryOperator *UO = cast<UnaryOperator>(expr);
5952         expr = UO->getSubExpr();
5953         switch (UO->getOpcode()) {
5954           case UO_AddrOf:
5955             AllowOnePastEnd++;
5956             break;
5957           case UO_Deref:
5958             AllowOnePastEnd--;
5959             break;
5960           default:
5961             return;
5962         }
5963         break;
5964       }
5965       case Stmt::ConditionalOperatorClass: {
5966         const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
5967         if (const Expr *lhs = cond->getLHS())
5968           CheckArrayAccess(lhs);
5969         if (const Expr *rhs = cond->getRHS())
5970           CheckArrayAccess(rhs);
5971         return;
5972       }
5973       default:
5974         return;
5975     }
5976   }
5977 }
5978 
5979 //===--- CHECK: Objective-C retain cycles ----------------------------------//
5980 
5981 namespace {
5982   struct RetainCycleOwner {
5983     RetainCycleOwner() : Variable(0), Indirect(false) {}
5984     VarDecl *Variable;
5985     SourceRange Range;
5986     SourceLocation Loc;
5987     bool Indirect;
5988 
5989     void setLocsFrom(Expr *e) {
5990       Loc = e->getExprLoc();
5991       Range = e->getSourceRange();
5992     }
5993   };
5994 }
5995 
5996 /// Consider whether capturing the given variable can possibly lead to
5997 /// a retain cycle.
5998 static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) {
5999   // In ARC, it's captured strongly iff the variable has __strong
6000   // lifetime.  In MRR, it's captured strongly if the variable is
6001   // __block and has an appropriate type.
6002   if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
6003     return false;
6004 
6005   owner.Variable = var;
6006   if (ref)
6007     owner.setLocsFrom(ref);
6008   return true;
6009 }
6010 
6011 static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) {
6012   while (true) {
6013     e = e->IgnoreParens();
6014     if (CastExpr *cast = dyn_cast<CastExpr>(e)) {
6015       switch (cast->getCastKind()) {
6016       case CK_BitCast:
6017       case CK_LValueBitCast:
6018       case CK_LValueToRValue:
6019       case CK_ARCReclaimReturnedObject:
6020         e = cast->getSubExpr();
6021         continue;
6022 
6023       default:
6024         return false;
6025       }
6026     }
6027 
6028     if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) {
6029       ObjCIvarDecl *ivar = ref->getDecl();
6030       if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
6031         return false;
6032 
6033       // Try to find a retain cycle in the base.
6034       if (!findRetainCycleOwner(S, ref->getBase(), owner))
6035         return false;
6036 
6037       if (ref->isFreeIvar()) owner.setLocsFrom(ref);
6038       owner.Indirect = true;
6039       return true;
6040     }
6041 
6042     if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) {
6043       VarDecl *var = dyn_cast<VarDecl>(ref->getDecl());
6044       if (!var) return false;
6045       return considerVariable(var, ref, owner);
6046     }
6047 
6048     if (MemberExpr *member = dyn_cast<MemberExpr>(e)) {
6049       if (member->isArrow()) return false;
6050 
6051       // Don't count this as an indirect ownership.
6052       e = member->getBase();
6053       continue;
6054     }
6055 
6056     if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) {
6057       // Only pay attention to pseudo-objects on property references.
6058       ObjCPropertyRefExpr *pre
6059         = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm()
6060                                               ->IgnoreParens());
6061       if (!pre) return false;
6062       if (pre->isImplicitProperty()) return false;
6063       ObjCPropertyDecl *property = pre->getExplicitProperty();
6064       if (!property->isRetaining() &&
6065           !(property->getPropertyIvarDecl() &&
6066             property->getPropertyIvarDecl()->getType()
6067               .getObjCLifetime() == Qualifiers::OCL_Strong))
6068           return false;
6069 
6070       owner.Indirect = true;
6071       if (pre->isSuperReceiver()) {
6072         owner.Variable = S.getCurMethodDecl()->getSelfDecl();
6073         if (!owner.Variable)
6074           return false;
6075         owner.Loc = pre->getLocation();
6076         owner.Range = pre->getSourceRange();
6077         return true;
6078       }
6079       e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase())
6080                               ->getSourceExpr());
6081       continue;
6082     }
6083 
6084     // Array ivars?
6085 
6086     return false;
6087   }
6088 }
6089 
6090 namespace {
6091   struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> {
6092     FindCaptureVisitor(ASTContext &Context, VarDecl *variable)
6093       : EvaluatedExprVisitor<FindCaptureVisitor>(Context),
6094         Variable(variable), Capturer(0) {}
6095 
6096     VarDecl *Variable;
6097     Expr *Capturer;
6098 
6099     void VisitDeclRefExpr(DeclRefExpr *ref) {
6100       if (ref->getDecl() == Variable && !Capturer)
6101         Capturer = ref;
6102     }
6103 
6104     void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) {
6105       if (Capturer) return;
6106       Visit(ref->getBase());
6107       if (Capturer && ref->isFreeIvar())
6108         Capturer = ref;
6109     }
6110 
6111     void VisitBlockExpr(BlockExpr *block) {
6112       // Look inside nested blocks
6113       if (block->getBlockDecl()->capturesVariable(Variable))
6114         Visit(block->getBlockDecl()->getBody());
6115     }
6116 
6117     void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) {
6118       if (Capturer) return;
6119       if (OVE->getSourceExpr())
6120         Visit(OVE->getSourceExpr());
6121     }
6122   };
6123 }
6124 
6125 /// Check whether the given argument is a block which captures a
6126 /// variable.
6127 static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) {
6128   assert(owner.Variable && owner.Loc.isValid());
6129 
6130   e = e->IgnoreParenCasts();
6131 
6132   // Look through [^{...} copy] and Block_copy(^{...}).
6133   if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) {
6134     Selector Cmd = ME->getSelector();
6135     if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") {
6136       e = ME->getInstanceReceiver();
6137       if (!e)
6138         return 0;
6139       e = e->IgnoreParenCasts();
6140     }
6141   } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) {
6142     if (CE->getNumArgs() == 1) {
6143       FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl());
6144       if (Fn) {
6145         const IdentifierInfo *FnI = Fn->getIdentifier();
6146         if (FnI && FnI->isStr("_Block_copy")) {
6147           e = CE->getArg(0)->IgnoreParenCasts();
6148         }
6149       }
6150     }
6151   }
6152 
6153   BlockExpr *block = dyn_cast<BlockExpr>(e);
6154   if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable))
6155     return 0;
6156 
6157   FindCaptureVisitor visitor(S.Context, owner.Variable);
6158   visitor.Visit(block->getBlockDecl()->getBody());
6159   return visitor.Capturer;
6160 }
6161 
6162 static void diagnoseRetainCycle(Sema &S, Expr *capturer,
6163                                 RetainCycleOwner &owner) {
6164   assert(capturer);
6165   assert(owner.Variable && owner.Loc.isValid());
6166 
6167   S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle)
6168     << owner.Variable << capturer->getSourceRange();
6169   S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner)
6170     << owner.Indirect << owner.Range;
6171 }
6172 
6173 /// Check for a keyword selector that starts with the word 'add' or
6174 /// 'set'.
6175 static bool isSetterLikeSelector(Selector sel) {
6176   if (sel.isUnarySelector()) return false;
6177 
6178   StringRef str = sel.getNameForSlot(0);
6179   while (!str.empty() && str.front() == '_') str = str.substr(1);
6180   if (str.startswith("set"))
6181     str = str.substr(3);
6182   else if (str.startswith("add")) {
6183     // Specially whitelist 'addOperationWithBlock:'.
6184     if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock"))
6185       return false;
6186     str = str.substr(3);
6187   }
6188   else
6189     return false;
6190 
6191   if (str.empty()) return true;
6192   return !isLowercase(str.front());
6193 }
6194 
6195 /// Check a message send to see if it's likely to cause a retain cycle.
6196 void Sema::checkRetainCycles(ObjCMessageExpr *msg) {
6197   // Only check instance methods whose selector looks like a setter.
6198   if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector()))
6199     return;
6200 
6201   // Try to find a variable that the receiver is strongly owned by.
6202   RetainCycleOwner owner;
6203   if (msg->getReceiverKind() == ObjCMessageExpr::Instance) {
6204     if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner))
6205       return;
6206   } else {
6207     assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance);
6208     owner.Variable = getCurMethodDecl()->getSelfDecl();
6209     owner.Loc = msg->getSuperLoc();
6210     owner.Range = msg->getSuperLoc();
6211   }
6212 
6213   // Check whether the receiver is captured by any of the arguments.
6214   for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i)
6215     if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner))
6216       return diagnoseRetainCycle(*this, capturer, owner);
6217 }
6218 
6219 /// Check a property assign to see if it's likely to cause a retain cycle.
6220 void Sema::checkRetainCycles(Expr *receiver, Expr *argument) {
6221   RetainCycleOwner owner;
6222   if (!findRetainCycleOwner(*this, receiver, owner))
6223     return;
6224 
6225   if (Expr *capturer = findCapturingExpr(*this, argument, owner))
6226     diagnoseRetainCycle(*this, capturer, owner);
6227 }
6228 
6229 void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) {
6230   RetainCycleOwner Owner;
6231   if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner))
6232     return;
6233 
6234   // Because we don't have an expression for the variable, we have to set the
6235   // location explicitly here.
6236   Owner.Loc = Var->getLocation();
6237   Owner.Range = Var->getSourceRange();
6238 
6239   if (Expr *Capturer = findCapturingExpr(*this, Init, Owner))
6240     diagnoseRetainCycle(*this, Capturer, Owner);
6241 }
6242 
6243 static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc,
6244                                      Expr *RHS, bool isProperty) {
6245   // Check if RHS is an Objective-C object literal, which also can get
6246   // immediately zapped in a weak reference.  Note that we explicitly
6247   // allow ObjCStringLiterals, since those are designed to never really die.
6248   RHS = RHS->IgnoreParenImpCasts();
6249 
6250   // This enum needs to match with the 'select' in
6251   // warn_objc_arc_literal_assign (off-by-1).
6252   Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS);
6253   if (Kind == Sema::LK_String || Kind == Sema::LK_None)
6254     return false;
6255 
6256   S.Diag(Loc, diag::warn_arc_literal_assign)
6257     << (unsigned) Kind
6258     << (isProperty ? 0 : 1)
6259     << RHS->getSourceRange();
6260 
6261   return true;
6262 }
6263 
6264 static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc,
6265                                     Qualifiers::ObjCLifetime LT,
6266                                     Expr *RHS, bool isProperty) {
6267   // Strip off any implicit cast added to get to the one ARC-specific.
6268   while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
6269     if (cast->getCastKind() == CK_ARCConsumeObject) {
6270       S.Diag(Loc, diag::warn_arc_retained_assign)
6271         << (LT == Qualifiers::OCL_ExplicitNone)
6272         << (isProperty ? 0 : 1)
6273         << RHS->getSourceRange();
6274       return true;
6275     }
6276     RHS = cast->getSubExpr();
6277   }
6278 
6279   if (LT == Qualifiers::OCL_Weak &&
6280       checkUnsafeAssignLiteral(S, Loc, RHS, isProperty))
6281     return true;
6282 
6283   return false;
6284 }
6285 
6286 bool Sema::checkUnsafeAssigns(SourceLocation Loc,
6287                               QualType LHS, Expr *RHS) {
6288   Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime();
6289 
6290   if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone)
6291     return false;
6292 
6293   if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false))
6294     return true;
6295 
6296   return false;
6297 }
6298 
6299 void Sema::checkUnsafeExprAssigns(SourceLocation Loc,
6300                               Expr *LHS, Expr *RHS) {
6301   QualType LHSType;
6302   // PropertyRef on LHS type need be directly obtained from
6303   // its declaration as it has a PsuedoType.
6304   ObjCPropertyRefExpr *PRE
6305     = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens());
6306   if (PRE && !PRE->isImplicitProperty()) {
6307     const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
6308     if (PD)
6309       LHSType = PD->getType();
6310   }
6311 
6312   if (LHSType.isNull())
6313     LHSType = LHS->getType();
6314 
6315   Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime();
6316 
6317   if (LT == Qualifiers::OCL_Weak) {
6318     DiagnosticsEngine::Level Level =
6319       Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc);
6320     if (Level != DiagnosticsEngine::Ignored)
6321       getCurFunction()->markSafeWeakUse(LHS);
6322   }
6323 
6324   if (checkUnsafeAssigns(Loc, LHSType, RHS))
6325     return;
6326 
6327   // FIXME. Check for other life times.
6328   if (LT != Qualifiers::OCL_None)
6329     return;
6330 
6331   if (PRE) {
6332     if (PRE->isImplicitProperty())
6333       return;
6334     const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
6335     if (!PD)
6336       return;
6337 
6338     unsigned Attributes = PD->getPropertyAttributes();
6339     if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) {
6340       // when 'assign' attribute was not explicitly specified
6341       // by user, ignore it and rely on property type itself
6342       // for lifetime info.
6343       unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten();
6344       if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) &&
6345           LHSType->isObjCRetainableType())
6346         return;
6347 
6348       while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
6349         if (cast->getCastKind() == CK_ARCConsumeObject) {
6350           Diag(Loc, diag::warn_arc_retained_property_assign)
6351           << RHS->getSourceRange();
6352           return;
6353         }
6354         RHS = cast->getSubExpr();
6355       }
6356     }
6357     else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) {
6358       if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true))
6359         return;
6360     }
6361   }
6362 }
6363 
6364 //===--- CHECK: Empty statement body (-Wempty-body) ---------------------===//
6365 
6366 namespace {
6367 bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr,
6368                                  SourceLocation StmtLoc,
6369                                  const NullStmt *Body) {
6370   // Do not warn if the body is a macro that expands to nothing, e.g:
6371   //
6372   // #define CALL(x)
6373   // if (condition)
6374   //   CALL(0);
6375   //
6376   if (Body->hasLeadingEmptyMacro())
6377     return false;
6378 
6379   // Get line numbers of statement and body.
6380   bool StmtLineInvalid;
6381   unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc,
6382                                                       &StmtLineInvalid);
6383   if (StmtLineInvalid)
6384     return false;
6385 
6386   bool BodyLineInvalid;
6387   unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(),
6388                                                       &BodyLineInvalid);
6389   if (BodyLineInvalid)
6390     return false;
6391 
6392   // Warn if null statement and body are on the same line.
6393   if (StmtLine != BodyLine)
6394     return false;
6395 
6396   return true;
6397 }
6398 } // Unnamed namespace
6399 
6400 void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc,
6401                                  const Stmt *Body,
6402                                  unsigned DiagID) {
6403   // Since this is a syntactic check, don't emit diagnostic for template
6404   // instantiations, this just adds noise.
6405   if (CurrentInstantiationScope)
6406     return;
6407 
6408   // The body should be a null statement.
6409   const NullStmt *NBody = dyn_cast<NullStmt>(Body);
6410   if (!NBody)
6411     return;
6412 
6413   // Do the usual checks.
6414   if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody))
6415     return;
6416 
6417   Diag(NBody->getSemiLoc(), DiagID);
6418   Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line);
6419 }
6420 
6421 void Sema::DiagnoseEmptyLoopBody(const Stmt *S,
6422                                  const Stmt *PossibleBody) {
6423   assert(!CurrentInstantiationScope); // Ensured by caller
6424 
6425   SourceLocation StmtLoc;
6426   const Stmt *Body;
6427   unsigned DiagID;
6428   if (const ForStmt *FS = dyn_cast<ForStmt>(S)) {
6429     StmtLoc = FS->getRParenLoc();
6430     Body = FS->getBody();
6431     DiagID = diag::warn_empty_for_body;
6432   } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) {
6433     StmtLoc = WS->getCond()->getSourceRange().getEnd();
6434     Body = WS->getBody();
6435     DiagID = diag::warn_empty_while_body;
6436   } else
6437     return; // Neither `for' nor `while'.
6438 
6439   // The body should be a null statement.
6440   const NullStmt *NBody = dyn_cast<NullStmt>(Body);
6441   if (!NBody)
6442     return;
6443 
6444   // Skip expensive checks if diagnostic is disabled.
6445   if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) ==
6446           DiagnosticsEngine::Ignored)
6447     return;
6448 
6449   // Do the usual checks.
6450   if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody))
6451     return;
6452 
6453   // `for(...);' and `while(...);' are popular idioms, so in order to keep
6454   // noise level low, emit diagnostics only if for/while is followed by a
6455   // CompoundStmt, e.g.:
6456   //    for (int i = 0; i < n; i++);
6457   //    {
6458   //      a(i);
6459   //    }
6460   // or if for/while is followed by a statement with more indentation
6461   // than for/while itself:
6462   //    for (int i = 0; i < n; i++);
6463   //      a(i);
6464   bool ProbableTypo = isa<CompoundStmt>(PossibleBody);
6465   if (!ProbableTypo) {
6466     bool BodyColInvalid;
6467     unsigned BodyCol = SourceMgr.getPresumedColumnNumber(
6468                              PossibleBody->getLocStart(),
6469                              &BodyColInvalid);
6470     if (BodyColInvalid)
6471       return;
6472 
6473     bool StmtColInvalid;
6474     unsigned StmtCol = SourceMgr.getPresumedColumnNumber(
6475                              S->getLocStart(),
6476                              &StmtColInvalid);
6477     if (StmtColInvalid)
6478       return;
6479 
6480     if (BodyCol > StmtCol)
6481       ProbableTypo = true;
6482   }
6483 
6484   if (ProbableTypo) {
6485     Diag(NBody->getSemiLoc(), DiagID);
6486     Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line);
6487   }
6488 }
6489 
6490 //===--- Layout compatibility ----------------------------------------------//
6491 
6492 namespace {
6493 
6494 bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2);
6495 
6496 /// \brief Check if two enumeration types are layout-compatible.
6497 bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) {
6498   // C++11 [dcl.enum] p8:
6499   // Two enumeration types are layout-compatible if they have the same
6500   // underlying type.
6501   return ED1->isComplete() && ED2->isComplete() &&
6502          C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType());
6503 }
6504 
6505 /// \brief Check if two fields are layout-compatible.
6506 bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) {
6507   if (!isLayoutCompatible(C, Field1->getType(), Field2->getType()))
6508     return false;
6509 
6510   if (Field1->isBitField() != Field2->isBitField())
6511     return false;
6512 
6513   if (Field1->isBitField()) {
6514     // Make sure that the bit-fields are the same length.
6515     unsigned Bits1 = Field1->getBitWidthValue(C);
6516     unsigned Bits2 = Field2->getBitWidthValue(C);
6517 
6518     if (Bits1 != Bits2)
6519       return false;
6520   }
6521 
6522   return true;
6523 }
6524 
6525 /// \brief Check if two standard-layout structs are layout-compatible.
6526 /// (C++11 [class.mem] p17)
6527 bool isLayoutCompatibleStruct(ASTContext &C,
6528                               RecordDecl *RD1,
6529                               RecordDecl *RD2) {
6530   // If both records are C++ classes, check that base classes match.
6531   if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) {
6532     // If one of records is a CXXRecordDecl we are in C++ mode,
6533     // thus the other one is a CXXRecordDecl, too.
6534     const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2);
6535     // Check number of base classes.
6536     if (D1CXX->getNumBases() != D2CXX->getNumBases())
6537       return false;
6538 
6539     // Check the base classes.
6540     for (CXXRecordDecl::base_class_const_iterator
6541                Base1 = D1CXX->bases_begin(),
6542            BaseEnd1 = D1CXX->bases_end(),
6543               Base2 = D2CXX->bases_begin();
6544          Base1 != BaseEnd1;
6545          ++Base1, ++Base2) {
6546       if (!isLayoutCompatible(C, Base1->getType(), Base2->getType()))
6547         return false;
6548     }
6549   } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) {
6550     // If only RD2 is a C++ class, it should have zero base classes.
6551     if (D2CXX->getNumBases() > 0)
6552       return false;
6553   }
6554 
6555   // Check the fields.
6556   RecordDecl::field_iterator Field2 = RD2->field_begin(),
6557                              Field2End = RD2->field_end(),
6558                              Field1 = RD1->field_begin(),
6559                              Field1End = RD1->field_end();
6560   for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) {
6561     if (!isLayoutCompatible(C, *Field1, *Field2))
6562       return false;
6563   }
6564   if (Field1 != Field1End || Field2 != Field2End)
6565     return false;
6566 
6567   return true;
6568 }
6569 
6570 /// \brief Check if two standard-layout unions are layout-compatible.
6571 /// (C++11 [class.mem] p18)
6572 bool isLayoutCompatibleUnion(ASTContext &C,
6573                              RecordDecl *RD1,
6574                              RecordDecl *RD2) {
6575   llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields;
6576   for (RecordDecl::field_iterator Field2 = RD2->field_begin(),
6577                                   Field2End = RD2->field_end();
6578        Field2 != Field2End; ++Field2) {
6579     UnmatchedFields.insert(*Field2);
6580   }
6581 
6582   for (RecordDecl::field_iterator Field1 = RD1->field_begin(),
6583                                   Field1End = RD1->field_end();
6584        Field1 != Field1End; ++Field1) {
6585     llvm::SmallPtrSet<FieldDecl *, 8>::iterator
6586         I = UnmatchedFields.begin(),
6587         E = UnmatchedFields.end();
6588 
6589     for ( ; I != E; ++I) {
6590       if (isLayoutCompatible(C, *Field1, *I)) {
6591         bool Result = UnmatchedFields.erase(*I);
6592         (void) Result;
6593         assert(Result);
6594         break;
6595       }
6596     }
6597     if (I == E)
6598       return false;
6599   }
6600 
6601   return UnmatchedFields.empty();
6602 }
6603 
6604 bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) {
6605   if (RD1->isUnion() != RD2->isUnion())
6606     return false;
6607 
6608   if (RD1->isUnion())
6609     return isLayoutCompatibleUnion(C, RD1, RD2);
6610   else
6611     return isLayoutCompatibleStruct(C, RD1, RD2);
6612 }
6613 
6614 /// \brief Check if two types are layout-compatible in C++11 sense.
6615 bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) {
6616   if (T1.isNull() || T2.isNull())
6617     return false;
6618 
6619   // C++11 [basic.types] p11:
6620   // If two types T1 and T2 are the same type, then T1 and T2 are
6621   // layout-compatible types.
6622   if (C.hasSameType(T1, T2))
6623     return true;
6624 
6625   T1 = T1.getCanonicalType().getUnqualifiedType();
6626   T2 = T2.getCanonicalType().getUnqualifiedType();
6627 
6628   const Type::TypeClass TC1 = T1->getTypeClass();
6629   const Type::TypeClass TC2 = T2->getTypeClass();
6630 
6631   if (TC1 != TC2)
6632     return false;
6633 
6634   if (TC1 == Type::Enum) {
6635     return isLayoutCompatible(C,
6636                               cast<EnumType>(T1)->getDecl(),
6637                               cast<EnumType>(T2)->getDecl());
6638   } else if (TC1 == Type::Record) {
6639     if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType())
6640       return false;
6641 
6642     return isLayoutCompatible(C,
6643                               cast<RecordType>(T1)->getDecl(),
6644                               cast<RecordType>(T2)->getDecl());
6645   }
6646 
6647   return false;
6648 }
6649 }
6650 
6651 //===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----//
6652 
6653 namespace {
6654 /// \brief Given a type tag expression find the type tag itself.
6655 ///
6656 /// \param TypeExpr Type tag expression, as it appears in user's code.
6657 ///
6658 /// \param VD Declaration of an identifier that appears in a type tag.
6659 ///
6660 /// \param MagicValue Type tag magic value.
6661 bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx,
6662                      const ValueDecl **VD, uint64_t *MagicValue) {
6663   while(true) {
6664     if (!TypeExpr)
6665       return false;
6666 
6667     TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts();
6668 
6669     switch (TypeExpr->getStmtClass()) {
6670     case Stmt::UnaryOperatorClass: {
6671       const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr);
6672       if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) {
6673         TypeExpr = UO->getSubExpr();
6674         continue;
6675       }
6676       return false;
6677     }
6678 
6679     case Stmt::DeclRefExprClass: {
6680       const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr);
6681       *VD = DRE->getDecl();
6682       return true;
6683     }
6684 
6685     case Stmt::IntegerLiteralClass: {
6686       const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr);
6687       llvm::APInt MagicValueAPInt = IL->getValue();
6688       if (MagicValueAPInt.getActiveBits() <= 64) {
6689         *MagicValue = MagicValueAPInt.getZExtValue();
6690         return true;
6691       } else
6692         return false;
6693     }
6694 
6695     case Stmt::BinaryConditionalOperatorClass:
6696     case Stmt::ConditionalOperatorClass: {
6697       const AbstractConditionalOperator *ACO =
6698           cast<AbstractConditionalOperator>(TypeExpr);
6699       bool Result;
6700       if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) {
6701         if (Result)
6702           TypeExpr = ACO->getTrueExpr();
6703         else
6704           TypeExpr = ACO->getFalseExpr();
6705         continue;
6706       }
6707       return false;
6708     }
6709 
6710     case Stmt::BinaryOperatorClass: {
6711       const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr);
6712       if (BO->getOpcode() == BO_Comma) {
6713         TypeExpr = BO->getRHS();
6714         continue;
6715       }
6716       return false;
6717     }
6718 
6719     default:
6720       return false;
6721     }
6722   }
6723 }
6724 
6725 /// \brief Retrieve the C type corresponding to type tag TypeExpr.
6726 ///
6727 /// \param TypeExpr Expression that specifies a type tag.
6728 ///
6729 /// \param MagicValues Registered magic values.
6730 ///
6731 /// \param FoundWrongKind Set to true if a type tag was found, but of a wrong
6732 ///        kind.
6733 ///
6734 /// \param TypeInfo Information about the corresponding C type.
6735 ///
6736 /// \returns true if the corresponding C type was found.
6737 bool GetMatchingCType(
6738         const IdentifierInfo *ArgumentKind,
6739         const Expr *TypeExpr, const ASTContext &Ctx,
6740         const llvm::DenseMap<Sema::TypeTagMagicValue,
6741                              Sema::TypeTagData> *MagicValues,
6742         bool &FoundWrongKind,
6743         Sema::TypeTagData &TypeInfo) {
6744   FoundWrongKind = false;
6745 
6746   // Variable declaration that has type_tag_for_datatype attribute.
6747   const ValueDecl *VD = NULL;
6748 
6749   uint64_t MagicValue;
6750 
6751   if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue))
6752     return false;
6753 
6754   if (VD) {
6755     for (specific_attr_iterator<TypeTagForDatatypeAttr>
6756              I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(),
6757              E = VD->specific_attr_end<TypeTagForDatatypeAttr>();
6758          I != E; ++I) {
6759       if (I->getArgumentKind() != ArgumentKind) {
6760         FoundWrongKind = true;
6761         return false;
6762       }
6763       TypeInfo.Type = I->getMatchingCType();
6764       TypeInfo.LayoutCompatible = I->getLayoutCompatible();
6765       TypeInfo.MustBeNull = I->getMustBeNull();
6766       return true;
6767     }
6768     return false;
6769   }
6770 
6771   if (!MagicValues)
6772     return false;
6773 
6774   llvm::DenseMap<Sema::TypeTagMagicValue,
6775                  Sema::TypeTagData>::const_iterator I =
6776       MagicValues->find(std::make_pair(ArgumentKind, MagicValue));
6777   if (I == MagicValues->end())
6778     return false;
6779 
6780   TypeInfo = I->second;
6781   return true;
6782 }
6783 } // unnamed namespace
6784 
6785 void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind,
6786                                       uint64_t MagicValue, QualType Type,
6787                                       bool LayoutCompatible,
6788                                       bool MustBeNull) {
6789   if (!TypeTagForDatatypeMagicValues)
6790     TypeTagForDatatypeMagicValues.reset(
6791         new llvm::DenseMap<TypeTagMagicValue, TypeTagData>);
6792 
6793   TypeTagMagicValue Magic(ArgumentKind, MagicValue);
6794   (*TypeTagForDatatypeMagicValues)[Magic] =
6795       TypeTagData(Type, LayoutCompatible, MustBeNull);
6796 }
6797 
6798 namespace {
6799 bool IsSameCharType(QualType T1, QualType T2) {
6800   const BuiltinType *BT1 = T1->getAs<BuiltinType>();
6801   if (!BT1)
6802     return false;
6803 
6804   const BuiltinType *BT2 = T2->getAs<BuiltinType>();
6805   if (!BT2)
6806     return false;
6807 
6808   BuiltinType::Kind T1Kind = BT1->getKind();
6809   BuiltinType::Kind T2Kind = BT2->getKind();
6810 
6811   return (T1Kind == BuiltinType::SChar  && T2Kind == BuiltinType::Char_S) ||
6812          (T1Kind == BuiltinType::UChar  && T2Kind == BuiltinType::Char_U) ||
6813          (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) ||
6814          (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar);
6815 }
6816 } // unnamed namespace
6817 
6818 void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr,
6819                                     const Expr * const *ExprArgs) {
6820   const IdentifierInfo *ArgumentKind = Attr->getArgumentKind();
6821   bool IsPointerAttr = Attr->getIsPointer();
6822 
6823   const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()];
6824   bool FoundWrongKind;
6825   TypeTagData TypeInfo;
6826   if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context,
6827                         TypeTagForDatatypeMagicValues.get(),
6828                         FoundWrongKind, TypeInfo)) {
6829     if (FoundWrongKind)
6830       Diag(TypeTagExpr->getExprLoc(),
6831            diag::warn_type_tag_for_datatype_wrong_kind)
6832         << TypeTagExpr->getSourceRange();
6833     return;
6834   }
6835 
6836   const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()];
6837   if (IsPointerAttr) {
6838     // Skip implicit cast of pointer to `void *' (as a function argument).
6839     if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr))
6840       if (ICE->getType()->isVoidPointerType() &&
6841           ICE->getCastKind() == CK_BitCast)
6842         ArgumentExpr = ICE->getSubExpr();
6843   }
6844   QualType ArgumentType = ArgumentExpr->getType();
6845 
6846   // Passing a `void*' pointer shouldn't trigger a warning.
6847   if (IsPointerAttr && ArgumentType->isVoidPointerType())
6848     return;
6849 
6850   if (TypeInfo.MustBeNull) {
6851     // Type tag with matching void type requires a null pointer.
6852     if (!ArgumentExpr->isNullPointerConstant(Context,
6853                                              Expr::NPC_ValueDependentIsNotNull)) {
6854       Diag(ArgumentExpr->getExprLoc(),
6855            diag::warn_type_safety_null_pointer_required)
6856           << ArgumentKind->getName()
6857           << ArgumentExpr->getSourceRange()
6858           << TypeTagExpr->getSourceRange();
6859     }
6860     return;
6861   }
6862 
6863   QualType RequiredType = TypeInfo.Type;
6864   if (IsPointerAttr)
6865     RequiredType = Context.getPointerType(RequiredType);
6866 
6867   bool mismatch = false;
6868   if (!TypeInfo.LayoutCompatible) {
6869     mismatch = !Context.hasSameType(ArgumentType, RequiredType);
6870 
6871     // C++11 [basic.fundamental] p1:
6872     // Plain char, signed char, and unsigned char are three distinct types.
6873     //
6874     // But we treat plain `char' as equivalent to `signed char' or `unsigned
6875     // char' depending on the current char signedness mode.
6876     if (mismatch)
6877       if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(),
6878                                            RequiredType->getPointeeType())) ||
6879           (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType)))
6880         mismatch = false;
6881   } else
6882     if (IsPointerAttr)
6883       mismatch = !isLayoutCompatible(Context,
6884                                      ArgumentType->getPointeeType(),
6885                                      RequiredType->getPointeeType());
6886     else
6887       mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType);
6888 
6889   if (mismatch)
6890     Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch)
6891         << ArgumentType << ArgumentKind->getName()
6892         << TypeInfo.LayoutCompatible << RequiredType
6893         << ArgumentExpr->getSourceRange()
6894         << TypeTagExpr->getSourceRange();
6895 }
6896