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::CheckPointerCall(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() && !Ty->isFunctionPointerType())
608     return false;
609 
610   VariadicCallType CallType;
611   if (!Proto || !Proto->isVariadic()) {
612     CallType = VariadicDoesNotApply;
613   } else if (Ty->isBlockPointerType()) {
614     CallType = VariadicBlock;
615   } else { // Ty->isFunctionPointerType()
616     CallType = VariadicFunction;
617   }
618   unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0;
619 
620   checkCall(NDecl,
621             llvm::makeArrayRef<const Expr *>(TheCall->getArgs(),
622                                              TheCall->getNumArgs()),
623             NumProtoArgs, /*IsMemberFunction=*/false,
624             TheCall->getRParenLoc(),
625             TheCall->getCallee()->getSourceRange(), CallType);
626 
627   return false;
628 }
629 
630 ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult,
631                                          AtomicExpr::AtomicOp Op) {
632   CallExpr *TheCall = cast<CallExpr>(TheCallResult.get());
633   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
634 
635   // All these operations take one of the following forms:
636   enum {
637     // C    __c11_atomic_init(A *, C)
638     Init,
639     // C    __c11_atomic_load(A *, int)
640     Load,
641     // void __atomic_load(A *, CP, int)
642     Copy,
643     // C    __c11_atomic_add(A *, M, int)
644     Arithmetic,
645     // C    __atomic_exchange_n(A *, CP, int)
646     Xchg,
647     // void __atomic_exchange(A *, C *, CP, int)
648     GNUXchg,
649     // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int)
650     C11CmpXchg,
651     // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int)
652     GNUCmpXchg
653   } Form = Init;
654   const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 };
655   const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 };
656   // where:
657   //   C is an appropriate type,
658   //   A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins,
659   //   CP is C for __c11 builtins and GNU _n builtins and is C * otherwise,
660   //   M is C if C is an integer, and ptrdiff_t if C is a pointer, and
661   //   the int parameters are for orderings.
662 
663   assert(AtomicExpr::AO__c11_atomic_init == 0 &&
664          AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load
665          && "need to update code for modified C11 atomics");
666   bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init &&
667                Op <= AtomicExpr::AO__c11_atomic_fetch_xor;
668   bool IsN = Op == AtomicExpr::AO__atomic_load_n ||
669              Op == AtomicExpr::AO__atomic_store_n ||
670              Op == AtomicExpr::AO__atomic_exchange_n ||
671              Op == AtomicExpr::AO__atomic_compare_exchange_n;
672   bool IsAddSub = false;
673 
674   switch (Op) {
675   case AtomicExpr::AO__c11_atomic_init:
676     Form = Init;
677     break;
678 
679   case AtomicExpr::AO__c11_atomic_load:
680   case AtomicExpr::AO__atomic_load_n:
681     Form = Load;
682     break;
683 
684   case AtomicExpr::AO__c11_atomic_store:
685   case AtomicExpr::AO__atomic_load:
686   case AtomicExpr::AO__atomic_store:
687   case AtomicExpr::AO__atomic_store_n:
688     Form = Copy;
689     break;
690 
691   case AtomicExpr::AO__c11_atomic_fetch_add:
692   case AtomicExpr::AO__c11_atomic_fetch_sub:
693   case AtomicExpr::AO__atomic_fetch_add:
694   case AtomicExpr::AO__atomic_fetch_sub:
695   case AtomicExpr::AO__atomic_add_fetch:
696   case AtomicExpr::AO__atomic_sub_fetch:
697     IsAddSub = true;
698     // Fall through.
699   case AtomicExpr::AO__c11_atomic_fetch_and:
700   case AtomicExpr::AO__c11_atomic_fetch_or:
701   case AtomicExpr::AO__c11_atomic_fetch_xor:
702   case AtomicExpr::AO__atomic_fetch_and:
703   case AtomicExpr::AO__atomic_fetch_or:
704   case AtomicExpr::AO__atomic_fetch_xor:
705   case AtomicExpr::AO__atomic_fetch_nand:
706   case AtomicExpr::AO__atomic_and_fetch:
707   case AtomicExpr::AO__atomic_or_fetch:
708   case AtomicExpr::AO__atomic_xor_fetch:
709   case AtomicExpr::AO__atomic_nand_fetch:
710     Form = Arithmetic;
711     break;
712 
713   case AtomicExpr::AO__c11_atomic_exchange:
714   case AtomicExpr::AO__atomic_exchange_n:
715     Form = Xchg;
716     break;
717 
718   case AtomicExpr::AO__atomic_exchange:
719     Form = GNUXchg;
720     break;
721 
722   case AtomicExpr::AO__c11_atomic_compare_exchange_strong:
723   case AtomicExpr::AO__c11_atomic_compare_exchange_weak:
724     Form = C11CmpXchg;
725     break;
726 
727   case AtomicExpr::AO__atomic_compare_exchange:
728   case AtomicExpr::AO__atomic_compare_exchange_n:
729     Form = GNUCmpXchg;
730     break;
731   }
732 
733   // Check we have the right number of arguments.
734   if (TheCall->getNumArgs() < NumArgs[Form]) {
735     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
736       << 0 << NumArgs[Form] << TheCall->getNumArgs()
737       << TheCall->getCallee()->getSourceRange();
738     return ExprError();
739   } else if (TheCall->getNumArgs() > NumArgs[Form]) {
740     Diag(TheCall->getArg(NumArgs[Form])->getLocStart(),
741          diag::err_typecheck_call_too_many_args)
742       << 0 << NumArgs[Form] << TheCall->getNumArgs()
743       << TheCall->getCallee()->getSourceRange();
744     return ExprError();
745   }
746 
747   // Inspect the first argument of the atomic operation.
748   Expr *Ptr = TheCall->getArg(0);
749   Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get();
750   const PointerType *pointerType = Ptr->getType()->getAs<PointerType>();
751   if (!pointerType) {
752     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
753       << Ptr->getType() << Ptr->getSourceRange();
754     return ExprError();
755   }
756 
757   // For a __c11 builtin, this should be a pointer to an _Atomic type.
758   QualType AtomTy = pointerType->getPointeeType(); // 'A'
759   QualType ValType = AtomTy; // 'C'
760   if (IsC11) {
761     if (!AtomTy->isAtomicType()) {
762       Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic)
763         << Ptr->getType() << Ptr->getSourceRange();
764       return ExprError();
765     }
766     if (AtomTy.isConstQualified()) {
767       Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic)
768         << Ptr->getType() << Ptr->getSourceRange();
769       return ExprError();
770     }
771     ValType = AtomTy->getAs<AtomicType>()->getValueType();
772   }
773 
774   // For an arithmetic operation, the implied arithmetic must be well-formed.
775   if (Form == Arithmetic) {
776     // gcc does not enforce these rules for GNU atomics, but we do so for sanity.
777     if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) {
778       Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr)
779         << IsC11 << Ptr->getType() << Ptr->getSourceRange();
780       return ExprError();
781     }
782     if (!IsAddSub && !ValType->isIntegerType()) {
783       Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int)
784         << IsC11 << Ptr->getType() << Ptr->getSourceRange();
785       return ExprError();
786     }
787   } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) {
788     // For __atomic_*_n operations, the value type must be a scalar integral or
789     // pointer type which is 1, 2, 4, 8 or 16 bytes in length.
790     Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr)
791       << IsC11 << Ptr->getType() << Ptr->getSourceRange();
792     return ExprError();
793   }
794 
795   if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) {
796     // For GNU atomics, require a trivially-copyable type. This is not part of
797     // the GNU atomics specification, but we enforce it for sanity.
798     Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy)
799       << Ptr->getType() << Ptr->getSourceRange();
800     return ExprError();
801   }
802 
803   // FIXME: For any builtin other than a load, the ValType must not be
804   // const-qualified.
805 
806   switch (ValType.getObjCLifetime()) {
807   case Qualifiers::OCL_None:
808   case Qualifiers::OCL_ExplicitNone:
809     // okay
810     break;
811 
812   case Qualifiers::OCL_Weak:
813   case Qualifiers::OCL_Strong:
814   case Qualifiers::OCL_Autoreleasing:
815     // FIXME: Can this happen? By this point, ValType should be known
816     // to be trivially copyable.
817     Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
818       << ValType << Ptr->getSourceRange();
819     return ExprError();
820   }
821 
822   QualType ResultType = ValType;
823   if (Form == Copy || Form == GNUXchg || Form == Init)
824     ResultType = Context.VoidTy;
825   else if (Form == C11CmpXchg || Form == GNUCmpXchg)
826     ResultType = Context.BoolTy;
827 
828   // The type of a parameter passed 'by value'. In the GNU atomics, such
829   // arguments are actually passed as pointers.
830   QualType ByValType = ValType; // 'CP'
831   if (!IsC11 && !IsN)
832     ByValType = Ptr->getType();
833 
834   // The first argument --- the pointer --- has a fixed type; we
835   // deduce the types of the rest of the arguments accordingly.  Walk
836   // the remaining arguments, converting them to the deduced value type.
837   for (unsigned i = 1; i != NumArgs[Form]; ++i) {
838     QualType Ty;
839     if (i < NumVals[Form] + 1) {
840       switch (i) {
841       case 1:
842         // The second argument is the non-atomic operand. For arithmetic, this
843         // is always passed by value, and for a compare_exchange it is always
844         // passed by address. For the rest, GNU uses by-address and C11 uses
845         // by-value.
846         assert(Form != Load);
847         if (Form == Init || (Form == Arithmetic && ValType->isIntegerType()))
848           Ty = ValType;
849         else if (Form == Copy || Form == Xchg)
850           Ty = ByValType;
851         else if (Form == Arithmetic)
852           Ty = Context.getPointerDiffType();
853         else
854           Ty = Context.getPointerType(ValType.getUnqualifiedType());
855         break;
856       case 2:
857         // The third argument to compare_exchange / GNU exchange is a
858         // (pointer to a) desired value.
859         Ty = ByValType;
860         break;
861       case 3:
862         // The fourth argument to GNU compare_exchange is a 'weak' flag.
863         Ty = Context.BoolTy;
864         break;
865       }
866     } else {
867       // The order(s) are always converted to int.
868       Ty = Context.IntTy;
869     }
870 
871     InitializedEntity Entity =
872         InitializedEntity::InitializeParameter(Context, Ty, false);
873     ExprResult Arg = TheCall->getArg(i);
874     Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
875     if (Arg.isInvalid())
876       return true;
877     TheCall->setArg(i, Arg.get());
878   }
879 
880   // Permute the arguments into a 'consistent' order.
881   SmallVector<Expr*, 5> SubExprs;
882   SubExprs.push_back(Ptr);
883   switch (Form) {
884   case Init:
885     // Note, AtomicExpr::getVal1() has a special case for this atomic.
886     SubExprs.push_back(TheCall->getArg(1)); // Val1
887     break;
888   case Load:
889     SubExprs.push_back(TheCall->getArg(1)); // Order
890     break;
891   case Copy:
892   case Arithmetic:
893   case Xchg:
894     SubExprs.push_back(TheCall->getArg(2)); // Order
895     SubExprs.push_back(TheCall->getArg(1)); // Val1
896     break;
897   case GNUXchg:
898     // Note, AtomicExpr::getVal2() has a special case for this atomic.
899     SubExprs.push_back(TheCall->getArg(3)); // Order
900     SubExprs.push_back(TheCall->getArg(1)); // Val1
901     SubExprs.push_back(TheCall->getArg(2)); // Val2
902     break;
903   case C11CmpXchg:
904     SubExprs.push_back(TheCall->getArg(3)); // Order
905     SubExprs.push_back(TheCall->getArg(1)); // Val1
906     SubExprs.push_back(TheCall->getArg(4)); // OrderFail
907     SubExprs.push_back(TheCall->getArg(2)); // Val2
908     break;
909   case GNUCmpXchg:
910     SubExprs.push_back(TheCall->getArg(4)); // Order
911     SubExprs.push_back(TheCall->getArg(1)); // Val1
912     SubExprs.push_back(TheCall->getArg(5)); // OrderFail
913     SubExprs.push_back(TheCall->getArg(2)); // Val2
914     SubExprs.push_back(TheCall->getArg(3)); // Weak
915     break;
916   }
917 
918   AtomicExpr *AE = new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(),
919                                             SubExprs, ResultType, Op,
920                                             TheCall->getRParenLoc());
921 
922   if ((Op == AtomicExpr::AO__c11_atomic_load ||
923        (Op == AtomicExpr::AO__c11_atomic_store)) &&
924       Context.AtomicUsesUnsupportedLibcall(AE))
925     Diag(AE->getLocStart(), diag::err_atomic_load_store_uses_lib) <<
926     ((Op == AtomicExpr::AO__c11_atomic_load) ? 0 : 1);
927 
928   return Owned(AE);
929 }
930 
931 
932 /// checkBuiltinArgument - Given a call to a builtin function, perform
933 /// normal type-checking on the given argument, updating the call in
934 /// place.  This is useful when a builtin function requires custom
935 /// type-checking for some of its arguments but not necessarily all of
936 /// them.
937 ///
938 /// Returns true on error.
939 static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) {
940   FunctionDecl *Fn = E->getDirectCallee();
941   assert(Fn && "builtin call without direct callee!");
942 
943   ParmVarDecl *Param = Fn->getParamDecl(ArgIndex);
944   InitializedEntity Entity =
945     InitializedEntity::InitializeParameter(S.Context, Param);
946 
947   ExprResult Arg = E->getArg(0);
948   Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg);
949   if (Arg.isInvalid())
950     return true;
951 
952   E->setArg(ArgIndex, Arg.take());
953   return false;
954 }
955 
956 /// SemaBuiltinAtomicOverloaded - We have a call to a function like
957 /// __sync_fetch_and_add, which is an overloaded function based on the pointer
958 /// type of its first argument.  The main ActOnCallExpr routines have already
959 /// promoted the types of arguments because all of these calls are prototyped as
960 /// void(...).
961 ///
962 /// This function goes through and does final semantic checking for these
963 /// builtins,
964 ExprResult
965 Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
966   CallExpr *TheCall = (CallExpr *)TheCallResult.get();
967   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
968   FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
969 
970   // Ensure that we have at least one argument to do type inference from.
971   if (TheCall->getNumArgs() < 1) {
972     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
973       << 0 << 1 << TheCall->getNumArgs()
974       << TheCall->getCallee()->getSourceRange();
975     return ExprError();
976   }
977 
978   // Inspect the first argument of the atomic builtin.  This should always be
979   // a pointer type, whose element is an integral scalar or pointer type.
980   // Because it is a pointer type, we don't have to worry about any implicit
981   // casts here.
982   // FIXME: We don't allow floating point scalars as input.
983   Expr *FirstArg = TheCall->getArg(0);
984   ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg);
985   if (FirstArgResult.isInvalid())
986     return ExprError();
987   FirstArg = FirstArgResult.take();
988   TheCall->setArg(0, FirstArg);
989 
990   const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>();
991   if (!pointerType) {
992     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
993       << FirstArg->getType() << FirstArg->getSourceRange();
994     return ExprError();
995   }
996 
997   QualType ValType = pointerType->getPointeeType();
998   if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
999       !ValType->isBlockPointerType()) {
1000     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
1001       << FirstArg->getType() << FirstArg->getSourceRange();
1002     return ExprError();
1003   }
1004 
1005   switch (ValType.getObjCLifetime()) {
1006   case Qualifiers::OCL_None:
1007   case Qualifiers::OCL_ExplicitNone:
1008     // okay
1009     break;
1010 
1011   case Qualifiers::OCL_Weak:
1012   case Qualifiers::OCL_Strong:
1013   case Qualifiers::OCL_Autoreleasing:
1014     Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
1015       << ValType << FirstArg->getSourceRange();
1016     return ExprError();
1017   }
1018 
1019   // Strip any qualifiers off ValType.
1020   ValType = ValType.getUnqualifiedType();
1021 
1022   // The majority of builtins return a value, but a few have special return
1023   // types, so allow them to override appropriately below.
1024   QualType ResultType = ValType;
1025 
1026   // We need to figure out which concrete builtin this maps onto.  For example,
1027   // __sync_fetch_and_add with a 2 byte object turns into
1028   // __sync_fetch_and_add_2.
1029 #define BUILTIN_ROW(x) \
1030   { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
1031     Builtin::BI##x##_8, Builtin::BI##x##_16 }
1032 
1033   static const unsigned BuiltinIndices[][5] = {
1034     BUILTIN_ROW(__sync_fetch_and_add),
1035     BUILTIN_ROW(__sync_fetch_and_sub),
1036     BUILTIN_ROW(__sync_fetch_and_or),
1037     BUILTIN_ROW(__sync_fetch_and_and),
1038     BUILTIN_ROW(__sync_fetch_and_xor),
1039 
1040     BUILTIN_ROW(__sync_add_and_fetch),
1041     BUILTIN_ROW(__sync_sub_and_fetch),
1042     BUILTIN_ROW(__sync_and_and_fetch),
1043     BUILTIN_ROW(__sync_or_and_fetch),
1044     BUILTIN_ROW(__sync_xor_and_fetch),
1045 
1046     BUILTIN_ROW(__sync_val_compare_and_swap),
1047     BUILTIN_ROW(__sync_bool_compare_and_swap),
1048     BUILTIN_ROW(__sync_lock_test_and_set),
1049     BUILTIN_ROW(__sync_lock_release),
1050     BUILTIN_ROW(__sync_swap)
1051   };
1052 #undef BUILTIN_ROW
1053 
1054   // Determine the index of the size.
1055   unsigned SizeIndex;
1056   switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
1057   case 1: SizeIndex = 0; break;
1058   case 2: SizeIndex = 1; break;
1059   case 4: SizeIndex = 2; break;
1060   case 8: SizeIndex = 3; break;
1061   case 16: SizeIndex = 4; break;
1062   default:
1063     Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
1064       << FirstArg->getType() << FirstArg->getSourceRange();
1065     return ExprError();
1066   }
1067 
1068   // Each of these builtins has one pointer argument, followed by some number of
1069   // values (0, 1 or 2) followed by a potentially empty varags list of stuff
1070   // that we ignore.  Find out which row of BuiltinIndices to read from as well
1071   // as the number of fixed args.
1072   unsigned BuiltinID = FDecl->getBuiltinID();
1073   unsigned BuiltinIndex, NumFixed = 1;
1074   switch (BuiltinID) {
1075   default: llvm_unreachable("Unknown overloaded atomic builtin!");
1076   case Builtin::BI__sync_fetch_and_add:
1077   case Builtin::BI__sync_fetch_and_add_1:
1078   case Builtin::BI__sync_fetch_and_add_2:
1079   case Builtin::BI__sync_fetch_and_add_4:
1080   case Builtin::BI__sync_fetch_and_add_8:
1081   case Builtin::BI__sync_fetch_and_add_16:
1082     BuiltinIndex = 0;
1083     break;
1084 
1085   case Builtin::BI__sync_fetch_and_sub:
1086   case Builtin::BI__sync_fetch_and_sub_1:
1087   case Builtin::BI__sync_fetch_and_sub_2:
1088   case Builtin::BI__sync_fetch_and_sub_4:
1089   case Builtin::BI__sync_fetch_and_sub_8:
1090   case Builtin::BI__sync_fetch_and_sub_16:
1091     BuiltinIndex = 1;
1092     break;
1093 
1094   case Builtin::BI__sync_fetch_and_or:
1095   case Builtin::BI__sync_fetch_and_or_1:
1096   case Builtin::BI__sync_fetch_and_or_2:
1097   case Builtin::BI__sync_fetch_and_or_4:
1098   case Builtin::BI__sync_fetch_and_or_8:
1099   case Builtin::BI__sync_fetch_and_or_16:
1100     BuiltinIndex = 2;
1101     break;
1102 
1103   case Builtin::BI__sync_fetch_and_and:
1104   case Builtin::BI__sync_fetch_and_and_1:
1105   case Builtin::BI__sync_fetch_and_and_2:
1106   case Builtin::BI__sync_fetch_and_and_4:
1107   case Builtin::BI__sync_fetch_and_and_8:
1108   case Builtin::BI__sync_fetch_and_and_16:
1109     BuiltinIndex = 3;
1110     break;
1111 
1112   case Builtin::BI__sync_fetch_and_xor:
1113   case Builtin::BI__sync_fetch_and_xor_1:
1114   case Builtin::BI__sync_fetch_and_xor_2:
1115   case Builtin::BI__sync_fetch_and_xor_4:
1116   case Builtin::BI__sync_fetch_and_xor_8:
1117   case Builtin::BI__sync_fetch_and_xor_16:
1118     BuiltinIndex = 4;
1119     break;
1120 
1121   case Builtin::BI__sync_add_and_fetch:
1122   case Builtin::BI__sync_add_and_fetch_1:
1123   case Builtin::BI__sync_add_and_fetch_2:
1124   case Builtin::BI__sync_add_and_fetch_4:
1125   case Builtin::BI__sync_add_and_fetch_8:
1126   case Builtin::BI__sync_add_and_fetch_16:
1127     BuiltinIndex = 5;
1128     break;
1129 
1130   case Builtin::BI__sync_sub_and_fetch:
1131   case Builtin::BI__sync_sub_and_fetch_1:
1132   case Builtin::BI__sync_sub_and_fetch_2:
1133   case Builtin::BI__sync_sub_and_fetch_4:
1134   case Builtin::BI__sync_sub_and_fetch_8:
1135   case Builtin::BI__sync_sub_and_fetch_16:
1136     BuiltinIndex = 6;
1137     break;
1138 
1139   case Builtin::BI__sync_and_and_fetch:
1140   case Builtin::BI__sync_and_and_fetch_1:
1141   case Builtin::BI__sync_and_and_fetch_2:
1142   case Builtin::BI__sync_and_and_fetch_4:
1143   case Builtin::BI__sync_and_and_fetch_8:
1144   case Builtin::BI__sync_and_and_fetch_16:
1145     BuiltinIndex = 7;
1146     break;
1147 
1148   case Builtin::BI__sync_or_and_fetch:
1149   case Builtin::BI__sync_or_and_fetch_1:
1150   case Builtin::BI__sync_or_and_fetch_2:
1151   case Builtin::BI__sync_or_and_fetch_4:
1152   case Builtin::BI__sync_or_and_fetch_8:
1153   case Builtin::BI__sync_or_and_fetch_16:
1154     BuiltinIndex = 8;
1155     break;
1156 
1157   case Builtin::BI__sync_xor_and_fetch:
1158   case Builtin::BI__sync_xor_and_fetch_1:
1159   case Builtin::BI__sync_xor_and_fetch_2:
1160   case Builtin::BI__sync_xor_and_fetch_4:
1161   case Builtin::BI__sync_xor_and_fetch_8:
1162   case Builtin::BI__sync_xor_and_fetch_16:
1163     BuiltinIndex = 9;
1164     break;
1165 
1166   case Builtin::BI__sync_val_compare_and_swap:
1167   case Builtin::BI__sync_val_compare_and_swap_1:
1168   case Builtin::BI__sync_val_compare_and_swap_2:
1169   case Builtin::BI__sync_val_compare_and_swap_4:
1170   case Builtin::BI__sync_val_compare_and_swap_8:
1171   case Builtin::BI__sync_val_compare_and_swap_16:
1172     BuiltinIndex = 10;
1173     NumFixed = 2;
1174     break;
1175 
1176   case Builtin::BI__sync_bool_compare_and_swap:
1177   case Builtin::BI__sync_bool_compare_and_swap_1:
1178   case Builtin::BI__sync_bool_compare_and_swap_2:
1179   case Builtin::BI__sync_bool_compare_and_swap_4:
1180   case Builtin::BI__sync_bool_compare_and_swap_8:
1181   case Builtin::BI__sync_bool_compare_and_swap_16:
1182     BuiltinIndex = 11;
1183     NumFixed = 2;
1184     ResultType = Context.BoolTy;
1185     break;
1186 
1187   case Builtin::BI__sync_lock_test_and_set:
1188   case Builtin::BI__sync_lock_test_and_set_1:
1189   case Builtin::BI__sync_lock_test_and_set_2:
1190   case Builtin::BI__sync_lock_test_and_set_4:
1191   case Builtin::BI__sync_lock_test_and_set_8:
1192   case Builtin::BI__sync_lock_test_and_set_16:
1193     BuiltinIndex = 12;
1194     break;
1195 
1196   case Builtin::BI__sync_lock_release:
1197   case Builtin::BI__sync_lock_release_1:
1198   case Builtin::BI__sync_lock_release_2:
1199   case Builtin::BI__sync_lock_release_4:
1200   case Builtin::BI__sync_lock_release_8:
1201   case Builtin::BI__sync_lock_release_16:
1202     BuiltinIndex = 13;
1203     NumFixed = 0;
1204     ResultType = Context.VoidTy;
1205     break;
1206 
1207   case Builtin::BI__sync_swap:
1208   case Builtin::BI__sync_swap_1:
1209   case Builtin::BI__sync_swap_2:
1210   case Builtin::BI__sync_swap_4:
1211   case Builtin::BI__sync_swap_8:
1212   case Builtin::BI__sync_swap_16:
1213     BuiltinIndex = 14;
1214     break;
1215   }
1216 
1217   // Now that we know how many fixed arguments we expect, first check that we
1218   // have at least that many.
1219   if (TheCall->getNumArgs() < 1+NumFixed) {
1220     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
1221       << 0 << 1+NumFixed << TheCall->getNumArgs()
1222       << TheCall->getCallee()->getSourceRange();
1223     return ExprError();
1224   }
1225 
1226   // Get the decl for the concrete builtin from this, we can tell what the
1227   // concrete integer type we should convert to is.
1228   unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
1229   const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
1230   FunctionDecl *NewBuiltinDecl;
1231   if (NewBuiltinID == BuiltinID)
1232     NewBuiltinDecl = FDecl;
1233   else {
1234     // Perform builtin lookup to avoid redeclaring it.
1235     DeclarationName DN(&Context.Idents.get(NewBuiltinName));
1236     LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName);
1237     LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true);
1238     assert(Res.getFoundDecl());
1239     NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl());
1240     if (NewBuiltinDecl == 0)
1241       return ExprError();
1242   }
1243 
1244   // The first argument --- the pointer --- has a fixed type; we
1245   // deduce the types of the rest of the arguments accordingly.  Walk
1246   // the remaining arguments, converting them to the deduced value type.
1247   for (unsigned i = 0; i != NumFixed; ++i) {
1248     ExprResult Arg = TheCall->getArg(i+1);
1249 
1250     // GCC does an implicit conversion to the pointer or integer ValType.  This
1251     // can fail in some cases (1i -> int**), check for this error case now.
1252     // Initialize the argument.
1253     InitializedEntity Entity = InitializedEntity::InitializeParameter(Context,
1254                                                    ValType, /*consume*/ false);
1255     Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
1256     if (Arg.isInvalid())
1257       return ExprError();
1258 
1259     // Okay, we have something that *can* be converted to the right type.  Check
1260     // to see if there is a potentially weird extension going on here.  This can
1261     // happen when you do an atomic operation on something like an char* and
1262     // pass in 42.  The 42 gets converted to char.  This is even more strange
1263     // for things like 45.123 -> char, etc.
1264     // FIXME: Do this check.
1265     TheCall->setArg(i+1, Arg.take());
1266   }
1267 
1268   ASTContext& Context = this->getASTContext();
1269 
1270   // Create a new DeclRefExpr to refer to the new decl.
1271   DeclRefExpr* NewDRE = DeclRefExpr::Create(
1272       Context,
1273       DRE->getQualifierLoc(),
1274       SourceLocation(),
1275       NewBuiltinDecl,
1276       /*enclosing*/ false,
1277       DRE->getLocation(),
1278       Context.BuiltinFnTy,
1279       DRE->getValueKind());
1280 
1281   // Set the callee in the CallExpr.
1282   // FIXME: This loses syntactic information.
1283   QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType());
1284   ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy,
1285                                               CK_BuiltinFnToFnPtr);
1286   TheCall->setCallee(PromotedCall.take());
1287 
1288   // Change the result type of the call to match the original value type. This
1289   // is arbitrary, but the codegen for these builtins ins design to handle it
1290   // gracefully.
1291   TheCall->setType(ResultType);
1292 
1293   return TheCallResult;
1294 }
1295 
1296 /// CheckObjCString - Checks that the argument to the builtin
1297 /// CFString constructor is correct
1298 /// Note: It might also make sense to do the UTF-16 conversion here (would
1299 /// simplify the backend).
1300 bool Sema::CheckObjCString(Expr *Arg) {
1301   Arg = Arg->IgnoreParenCasts();
1302   StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
1303 
1304   if (!Literal || !Literal->isAscii()) {
1305     Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
1306       << Arg->getSourceRange();
1307     return true;
1308   }
1309 
1310   if (Literal->containsNonAsciiOrNull()) {
1311     StringRef String = Literal->getString();
1312     unsigned NumBytes = String.size();
1313     SmallVector<UTF16, 128> ToBuf(NumBytes);
1314     const UTF8 *FromPtr = (const UTF8 *)String.data();
1315     UTF16 *ToPtr = &ToBuf[0];
1316 
1317     ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
1318                                                  &ToPtr, ToPtr + NumBytes,
1319                                                  strictConversion);
1320     // Check for conversion failure.
1321     if (Result != conversionOK)
1322       Diag(Arg->getLocStart(),
1323            diag::warn_cfstring_truncated) << Arg->getSourceRange();
1324   }
1325   return false;
1326 }
1327 
1328 /// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
1329 /// Emit an error and return true on failure, return false on success.
1330 bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
1331   Expr *Fn = TheCall->getCallee();
1332   if (TheCall->getNumArgs() > 2) {
1333     Diag(TheCall->getArg(2)->getLocStart(),
1334          diag::err_typecheck_call_too_many_args)
1335       << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1336       << Fn->getSourceRange()
1337       << SourceRange(TheCall->getArg(2)->getLocStart(),
1338                      (*(TheCall->arg_end()-1))->getLocEnd());
1339     return true;
1340   }
1341 
1342   if (TheCall->getNumArgs() < 2) {
1343     return Diag(TheCall->getLocEnd(),
1344       diag::err_typecheck_call_too_few_args_at_least)
1345       << 0 /*function call*/ << 2 << TheCall->getNumArgs();
1346   }
1347 
1348   // Type-check the first argument normally.
1349   if (checkBuiltinArgument(*this, TheCall, 0))
1350     return true;
1351 
1352   // Determine whether the current function is variadic or not.
1353   BlockScopeInfo *CurBlock = getCurBlock();
1354   bool isVariadic;
1355   if (CurBlock)
1356     isVariadic = CurBlock->TheDecl->isVariadic();
1357   else if (FunctionDecl *FD = getCurFunctionDecl())
1358     isVariadic = FD->isVariadic();
1359   else
1360     isVariadic = getCurMethodDecl()->isVariadic();
1361 
1362   if (!isVariadic) {
1363     Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
1364     return true;
1365   }
1366 
1367   // Verify that the second argument to the builtin is the last argument of the
1368   // current function or method.
1369   bool SecondArgIsLastNamedArgument = false;
1370   const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
1371 
1372   // These are valid if SecondArgIsLastNamedArgument is false after the next
1373   // block.
1374   QualType Type;
1375   SourceLocation ParamLoc;
1376 
1377   if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
1378     if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
1379       // FIXME: This isn't correct for methods (results in bogus warning).
1380       // Get the last formal in the current function.
1381       const ParmVarDecl *LastArg;
1382       if (CurBlock)
1383         LastArg = *(CurBlock->TheDecl->param_end()-1);
1384       else if (FunctionDecl *FD = getCurFunctionDecl())
1385         LastArg = *(FD->param_end()-1);
1386       else
1387         LastArg = *(getCurMethodDecl()->param_end()-1);
1388       SecondArgIsLastNamedArgument = PV == LastArg;
1389 
1390       Type = PV->getType();
1391       ParamLoc = PV->getLocation();
1392     }
1393   }
1394 
1395   if (!SecondArgIsLastNamedArgument)
1396     Diag(TheCall->getArg(1)->getLocStart(),
1397          diag::warn_second_parameter_of_va_start_not_last_named_argument);
1398   else if (Type->isReferenceType()) {
1399     Diag(Arg->getLocStart(),
1400          diag::warn_va_start_of_reference_type_is_undefined);
1401     Diag(ParamLoc, diag::note_parameter_type) << Type;
1402   }
1403 
1404   return false;
1405 }
1406 
1407 /// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
1408 /// friends.  This is declared to take (...), so we have to check everything.
1409 bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
1410   if (TheCall->getNumArgs() < 2)
1411     return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1412       << 0 << 2 << TheCall->getNumArgs()/*function call*/;
1413   if (TheCall->getNumArgs() > 2)
1414     return Diag(TheCall->getArg(2)->getLocStart(),
1415                 diag::err_typecheck_call_too_many_args)
1416       << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1417       << SourceRange(TheCall->getArg(2)->getLocStart(),
1418                      (*(TheCall->arg_end()-1))->getLocEnd());
1419 
1420   ExprResult OrigArg0 = TheCall->getArg(0);
1421   ExprResult OrigArg1 = TheCall->getArg(1);
1422 
1423   // Do standard promotions between the two arguments, returning their common
1424   // type.
1425   QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
1426   if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
1427     return true;
1428 
1429   // Make sure any conversions are pushed back into the call; this is
1430   // type safe since unordered compare builtins are declared as "_Bool
1431   // foo(...)".
1432   TheCall->setArg(0, OrigArg0.get());
1433   TheCall->setArg(1, OrigArg1.get());
1434 
1435   if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
1436     return false;
1437 
1438   // If the common type isn't a real floating type, then the arguments were
1439   // invalid for this operation.
1440   if (Res.isNull() || !Res->isRealFloatingType())
1441     return Diag(OrigArg0.get()->getLocStart(),
1442                 diag::err_typecheck_call_invalid_ordered_compare)
1443       << OrigArg0.get()->getType() << OrigArg1.get()->getType()
1444       << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
1445 
1446   return false;
1447 }
1448 
1449 /// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
1450 /// __builtin_isnan and friends.  This is declared to take (...), so we have
1451 /// to check everything. We expect the last argument to be a floating point
1452 /// value.
1453 bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
1454   if (TheCall->getNumArgs() < NumArgs)
1455     return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1456       << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
1457   if (TheCall->getNumArgs() > NumArgs)
1458     return Diag(TheCall->getArg(NumArgs)->getLocStart(),
1459                 diag::err_typecheck_call_too_many_args)
1460       << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
1461       << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
1462                      (*(TheCall->arg_end()-1))->getLocEnd());
1463 
1464   Expr *OrigArg = TheCall->getArg(NumArgs-1);
1465 
1466   if (OrigArg->isTypeDependent())
1467     return false;
1468 
1469   // This operation requires a non-_Complex floating-point number.
1470   if (!OrigArg->getType()->isRealFloatingType())
1471     return Diag(OrigArg->getLocStart(),
1472                 diag::err_typecheck_call_invalid_unary_fp)
1473       << OrigArg->getType() << OrigArg->getSourceRange();
1474 
1475   // If this is an implicit conversion from float -> double, remove it.
1476   if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
1477     Expr *CastArg = Cast->getSubExpr();
1478     if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
1479       assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
1480              "promotion from float to double is the only expected cast here");
1481       Cast->setSubExpr(0);
1482       TheCall->setArg(NumArgs-1, CastArg);
1483     }
1484   }
1485 
1486   return false;
1487 }
1488 
1489 /// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
1490 // This is declared to take (...), so we have to check everything.
1491 ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
1492   if (TheCall->getNumArgs() < 2)
1493     return ExprError(Diag(TheCall->getLocEnd(),
1494                           diag::err_typecheck_call_too_few_args_at_least)
1495       << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1496       << TheCall->getSourceRange());
1497 
1498   // Determine which of the following types of shufflevector we're checking:
1499   // 1) unary, vector mask: (lhs, mask)
1500   // 2) binary, vector mask: (lhs, rhs, mask)
1501   // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
1502   QualType resType = TheCall->getArg(0)->getType();
1503   unsigned numElements = 0;
1504 
1505   if (!TheCall->getArg(0)->isTypeDependent() &&
1506       !TheCall->getArg(1)->isTypeDependent()) {
1507     QualType LHSType = TheCall->getArg(0)->getType();
1508     QualType RHSType = TheCall->getArg(1)->getType();
1509 
1510     if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
1511       Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
1512         << SourceRange(TheCall->getArg(0)->getLocStart(),
1513                        TheCall->getArg(1)->getLocEnd());
1514       return ExprError();
1515     }
1516 
1517     numElements = LHSType->getAs<VectorType>()->getNumElements();
1518     unsigned numResElements = TheCall->getNumArgs() - 2;
1519 
1520     // Check to see if we have a call with 2 vector arguments, the unary shuffle
1521     // with mask.  If so, verify that RHS is an integer vector type with the
1522     // same number of elts as lhs.
1523     if (TheCall->getNumArgs() == 2) {
1524       if (!RHSType->hasIntegerRepresentation() ||
1525           RHSType->getAs<VectorType>()->getNumElements() != numElements)
1526         Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1527           << SourceRange(TheCall->getArg(1)->getLocStart(),
1528                          TheCall->getArg(1)->getLocEnd());
1529       numResElements = numElements;
1530     }
1531     else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
1532       Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1533         << SourceRange(TheCall->getArg(0)->getLocStart(),
1534                        TheCall->getArg(1)->getLocEnd());
1535       return ExprError();
1536     } else if (numElements != numResElements) {
1537       QualType eltType = LHSType->getAs<VectorType>()->getElementType();
1538       resType = Context.getVectorType(eltType, numResElements,
1539                                       VectorType::GenericVector);
1540     }
1541   }
1542 
1543   for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
1544     if (TheCall->getArg(i)->isTypeDependent() ||
1545         TheCall->getArg(i)->isValueDependent())
1546       continue;
1547 
1548     llvm::APSInt Result(32);
1549     if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
1550       return ExprError(Diag(TheCall->getLocStart(),
1551                   diag::err_shufflevector_nonconstant_argument)
1552                 << TheCall->getArg(i)->getSourceRange());
1553 
1554     if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
1555       return ExprError(Diag(TheCall->getLocStart(),
1556                   diag::err_shufflevector_argument_too_large)
1557                << TheCall->getArg(i)->getSourceRange());
1558   }
1559 
1560   SmallVector<Expr*, 32> exprs;
1561 
1562   for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
1563     exprs.push_back(TheCall->getArg(i));
1564     TheCall->setArg(i, 0);
1565   }
1566 
1567   return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType,
1568                                             TheCall->getCallee()->getLocStart(),
1569                                             TheCall->getRParenLoc()));
1570 }
1571 
1572 /// SemaBuiltinPrefetch - Handle __builtin_prefetch.
1573 // This is declared to take (const void*, ...) and can take two
1574 // optional constant int args.
1575 bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
1576   unsigned NumArgs = TheCall->getNumArgs();
1577 
1578   if (NumArgs > 3)
1579     return Diag(TheCall->getLocEnd(),
1580              diag::err_typecheck_call_too_many_args_at_most)
1581              << 0 /*function call*/ << 3 << NumArgs
1582              << TheCall->getSourceRange();
1583 
1584   // Argument 0 is checked for us and the remaining arguments must be
1585   // constant integers.
1586   for (unsigned i = 1; i != NumArgs; ++i) {
1587     Expr *Arg = TheCall->getArg(i);
1588 
1589     // We can't check the value of a dependent argument.
1590     if (Arg->isTypeDependent() || Arg->isValueDependent())
1591       continue;
1592 
1593     llvm::APSInt Result;
1594     if (SemaBuiltinConstantArg(TheCall, i, Result))
1595       return true;
1596 
1597     // FIXME: gcc issues a warning and rewrites these to 0. These
1598     // seems especially odd for the third argument since the default
1599     // is 3.
1600     if (i == 1) {
1601       if (Result.getLimitedValue() > 1)
1602         return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1603              << "0" << "1" << Arg->getSourceRange();
1604     } else {
1605       if (Result.getLimitedValue() > 3)
1606         return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1607             << "0" << "3" << Arg->getSourceRange();
1608     }
1609   }
1610 
1611   return false;
1612 }
1613 
1614 /// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
1615 /// TheCall is a constant expression.
1616 bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
1617                                   llvm::APSInt &Result) {
1618   Expr *Arg = TheCall->getArg(ArgNum);
1619   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
1620   FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
1621 
1622   if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
1623 
1624   if (!Arg->isIntegerConstantExpr(Result, Context))
1625     return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
1626                 << FDecl->getDeclName() <<  Arg->getSourceRange();
1627 
1628   return false;
1629 }
1630 
1631 /// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
1632 /// int type). This simply type checks that type is one of the defined
1633 /// constants (0-3).
1634 // For compatibility check 0-3, llvm only handles 0 and 2.
1635 bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
1636   llvm::APSInt Result;
1637 
1638   // We can't check the value of a dependent argument.
1639   if (TheCall->getArg(1)->isTypeDependent() ||
1640       TheCall->getArg(1)->isValueDependent())
1641     return false;
1642 
1643   // Check constant-ness first.
1644   if (SemaBuiltinConstantArg(TheCall, 1, Result))
1645     return true;
1646 
1647   Expr *Arg = TheCall->getArg(1);
1648   if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
1649     return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1650              << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1651   }
1652 
1653   return false;
1654 }
1655 
1656 /// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
1657 /// This checks that val is a constant 1.
1658 bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
1659   Expr *Arg = TheCall->getArg(1);
1660   llvm::APSInt Result;
1661 
1662   // TODO: This is less than ideal. Overload this to take a value.
1663   if (SemaBuiltinConstantArg(TheCall, 1, Result))
1664     return true;
1665 
1666   if (Result != 1)
1667     return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
1668              << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1669 
1670   return false;
1671 }
1672 
1673 // Determine if an expression is a string literal or constant string.
1674 // If this function returns false on the arguments to a function expecting a
1675 // format string, we will usually need to emit a warning.
1676 // True string literals are then checked by CheckFormatString.
1677 Sema::StringLiteralCheckType
1678 Sema::checkFormatStringExpr(const Expr *E, ArrayRef<const Expr *> Args,
1679                             bool HasVAListArg,
1680                             unsigned format_idx, unsigned firstDataArg,
1681                             FormatStringType Type, VariadicCallType CallType,
1682                             bool inFunctionCall) {
1683  tryAgain:
1684   if (E->isTypeDependent() || E->isValueDependent())
1685     return SLCT_NotALiteral;
1686 
1687   E = E->IgnoreParenCasts();
1688 
1689   if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull))
1690     // Technically -Wformat-nonliteral does not warn about this case.
1691     // The behavior of printf and friends in this case is implementation
1692     // dependent.  Ideally if the format string cannot be null then
1693     // it should have a 'nonnull' attribute in the function prototype.
1694     return SLCT_CheckedLiteral;
1695 
1696   switch (E->getStmtClass()) {
1697   case Stmt::BinaryConditionalOperatorClass:
1698   case Stmt::ConditionalOperatorClass: {
1699     // The expression is a literal if both sub-expressions were, and it was
1700     // completely checked only if both sub-expressions were checked.
1701     const AbstractConditionalOperator *C =
1702         cast<AbstractConditionalOperator>(E);
1703     StringLiteralCheckType Left =
1704         checkFormatStringExpr(C->getTrueExpr(), Args,
1705                               HasVAListArg, format_idx, firstDataArg,
1706                               Type, CallType, inFunctionCall);
1707     if (Left == SLCT_NotALiteral)
1708       return SLCT_NotALiteral;
1709     StringLiteralCheckType Right =
1710         checkFormatStringExpr(C->getFalseExpr(), Args,
1711                               HasVAListArg, format_idx, firstDataArg,
1712                               Type, CallType, inFunctionCall);
1713     return Left < Right ? Left : Right;
1714   }
1715 
1716   case Stmt::ImplicitCastExprClass: {
1717     E = cast<ImplicitCastExpr>(E)->getSubExpr();
1718     goto tryAgain;
1719   }
1720 
1721   case Stmt::OpaqueValueExprClass:
1722     if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
1723       E = src;
1724       goto tryAgain;
1725     }
1726     return SLCT_NotALiteral;
1727 
1728   case Stmt::PredefinedExprClass:
1729     // While __func__, etc., are technically not string literals, they
1730     // cannot contain format specifiers and thus are not a security
1731     // liability.
1732     return SLCT_UncheckedLiteral;
1733 
1734   case Stmt::DeclRefExprClass: {
1735     const DeclRefExpr *DR = cast<DeclRefExpr>(E);
1736 
1737     // As an exception, do not flag errors for variables binding to
1738     // const string literals.
1739     if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
1740       bool isConstant = false;
1741       QualType T = DR->getType();
1742 
1743       if (const ArrayType *AT = Context.getAsArrayType(T)) {
1744         isConstant = AT->getElementType().isConstant(Context);
1745       } else if (const PointerType *PT = T->getAs<PointerType>()) {
1746         isConstant = T.isConstant(Context) &&
1747                      PT->getPointeeType().isConstant(Context);
1748       } else if (T->isObjCObjectPointerType()) {
1749         // In ObjC, there is usually no "const ObjectPointer" type,
1750         // so don't check if the pointee type is constant.
1751         isConstant = T.isConstant(Context);
1752       }
1753 
1754       if (isConstant) {
1755         if (const Expr *Init = VD->getAnyInitializer()) {
1756           // Look through initializers like const char c[] = { "foo" }
1757           if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) {
1758             if (InitList->isStringLiteralInit())
1759               Init = InitList->getInit(0)->IgnoreParenImpCasts();
1760           }
1761           return checkFormatStringExpr(Init, Args,
1762                                        HasVAListArg, format_idx,
1763                                        firstDataArg, Type, CallType,
1764                                        /*inFunctionCall*/false);
1765         }
1766       }
1767 
1768       // For vprintf* functions (i.e., HasVAListArg==true), we add a
1769       // special check to see if the format string is a function parameter
1770       // of the function calling the printf function.  If the function
1771       // has an attribute indicating it is a printf-like function, then we
1772       // should suppress warnings concerning non-literals being used in a call
1773       // to a vprintf function.  For example:
1774       //
1775       // void
1776       // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
1777       //      va_list ap;
1778       //      va_start(ap, fmt);
1779       //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
1780       //      ...
1781       //
1782       if (HasVAListArg) {
1783         if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) {
1784           if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) {
1785             int PVIndex = PV->getFunctionScopeIndex() + 1;
1786             for (specific_attr_iterator<FormatAttr>
1787                  i = ND->specific_attr_begin<FormatAttr>(),
1788                  e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) {
1789               FormatAttr *PVFormat = *i;
1790               // adjust for implicit parameter
1791               if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND))
1792                 if (MD->isInstance())
1793                   ++PVIndex;
1794               // We also check if the formats are compatible.
1795               // We can't pass a 'scanf' string to a 'printf' function.
1796               if (PVIndex == PVFormat->getFormatIdx() &&
1797                   Type == GetFormatStringType(PVFormat))
1798                 return SLCT_UncheckedLiteral;
1799             }
1800           }
1801         }
1802       }
1803     }
1804 
1805     return SLCT_NotALiteral;
1806   }
1807 
1808   case Stmt::CallExprClass:
1809   case Stmt::CXXMemberCallExprClass: {
1810     const CallExpr *CE = cast<CallExpr>(E);
1811     if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) {
1812       if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) {
1813         unsigned ArgIndex = FA->getFormatIdx();
1814         if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND))
1815           if (MD->isInstance())
1816             --ArgIndex;
1817         const Expr *Arg = CE->getArg(ArgIndex - 1);
1818 
1819         return checkFormatStringExpr(Arg, Args,
1820                                      HasVAListArg, format_idx, firstDataArg,
1821                                      Type, CallType, inFunctionCall);
1822       } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) {
1823         unsigned BuiltinID = FD->getBuiltinID();
1824         if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString ||
1825             BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) {
1826           const Expr *Arg = CE->getArg(0);
1827           return checkFormatStringExpr(Arg, Args,
1828                                        HasVAListArg, format_idx,
1829                                        firstDataArg, Type, CallType,
1830                                        inFunctionCall);
1831         }
1832       }
1833     }
1834 
1835     return SLCT_NotALiteral;
1836   }
1837   case Stmt::ObjCStringLiteralClass:
1838   case Stmt::StringLiteralClass: {
1839     const StringLiteral *StrE = NULL;
1840 
1841     if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1842       StrE = ObjCFExpr->getString();
1843     else
1844       StrE = cast<StringLiteral>(E);
1845 
1846     if (StrE) {
1847       CheckFormatString(StrE, E, Args, HasVAListArg, format_idx,
1848                         firstDataArg, Type, inFunctionCall, CallType);
1849       return SLCT_CheckedLiteral;
1850     }
1851 
1852     return SLCT_NotALiteral;
1853   }
1854 
1855   default:
1856     return SLCT_NotALiteral;
1857   }
1858 }
1859 
1860 void
1861 Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1862                             const Expr * const *ExprArgs,
1863                             SourceLocation CallSiteLoc) {
1864   for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1865                                   e = NonNull->args_end();
1866        i != e; ++i) {
1867     const Expr *ArgExpr = ExprArgs[*i];
1868 
1869     // As a special case, transparent unions initialized with zero are
1870     // considered null for the purposes of the nonnull attribute.
1871     if (const RecordType *UT = ArgExpr->getType()->getAsUnionType()) {
1872       if (UT->getDecl()->hasAttr<TransparentUnionAttr>())
1873         if (const CompoundLiteralExpr *CLE =
1874             dyn_cast<CompoundLiteralExpr>(ArgExpr))
1875           if (const InitListExpr *ILE =
1876               dyn_cast<InitListExpr>(CLE->getInitializer()))
1877             ArgExpr = ILE->getInit(0);
1878     }
1879 
1880     bool Result;
1881     if (ArgExpr->EvaluateAsBooleanCondition(Result, Context) && !Result)
1882       Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1883   }
1884 }
1885 
1886 Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) {
1887   return llvm::StringSwitch<FormatStringType>(Format->getType())
1888   .Case("scanf", FST_Scanf)
1889   .Cases("printf", "printf0", FST_Printf)
1890   .Cases("NSString", "CFString", FST_NSString)
1891   .Case("strftime", FST_Strftime)
1892   .Case("strfmon", FST_Strfmon)
1893   .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf)
1894   .Default(FST_Unknown);
1895 }
1896 
1897 /// CheckFormatArguments - Check calls to printf and scanf (and similar
1898 /// functions) for correct use of format strings.
1899 /// Returns true if a format string has been fully checked.
1900 bool Sema::CheckFormatArguments(const FormatAttr *Format,
1901                                 ArrayRef<const Expr *> Args,
1902                                 bool IsCXXMember,
1903                                 VariadicCallType CallType,
1904                                 SourceLocation Loc, SourceRange Range) {
1905   FormatStringInfo FSI;
1906   if (getFormatStringInfo(Format, IsCXXMember, &FSI))
1907     return CheckFormatArguments(Args, FSI.HasVAListArg, FSI.FormatIdx,
1908                                 FSI.FirstDataArg, GetFormatStringType(Format),
1909                                 CallType, Loc, Range);
1910   return false;
1911 }
1912 
1913 bool Sema::CheckFormatArguments(ArrayRef<const Expr *> Args,
1914                                 bool HasVAListArg, unsigned format_idx,
1915                                 unsigned firstDataArg, FormatStringType Type,
1916                                 VariadicCallType CallType,
1917                                 SourceLocation Loc, SourceRange Range) {
1918   // CHECK: printf/scanf-like function is called with no format string.
1919   if (format_idx >= Args.size()) {
1920     Diag(Loc, diag::warn_missing_format_string) << Range;
1921     return false;
1922   }
1923 
1924   const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts();
1925 
1926   // CHECK: format string is not a string literal.
1927   //
1928   // Dynamically generated format strings are difficult to
1929   // automatically vet at compile time.  Requiring that format strings
1930   // are string literals: (1) permits the checking of format strings by
1931   // the compiler and thereby (2) can practically remove the source of
1932   // many format string exploits.
1933 
1934   // Format string can be either ObjC string (e.g. @"%d") or
1935   // C string (e.g. "%d")
1936   // ObjC string uses the same format specifiers as C string, so we can use
1937   // the same format string checking logic for both ObjC and C strings.
1938   StringLiteralCheckType CT =
1939       checkFormatStringExpr(OrigFormatExpr, Args, HasVAListArg,
1940                             format_idx, firstDataArg, Type, CallType);
1941   if (CT != SLCT_NotALiteral)
1942     // Literal format string found, check done!
1943     return CT == SLCT_CheckedLiteral;
1944 
1945   // Strftime is particular as it always uses a single 'time' argument,
1946   // so it is safe to pass a non-literal string.
1947   if (Type == FST_Strftime)
1948     return false;
1949 
1950   // Do not emit diag when the string param is a macro expansion and the
1951   // format is either NSString or CFString. This is a hack to prevent
1952   // diag when using the NSLocalizedString and CFCopyLocalizedString macros
1953   // which are usually used in place of NS and CF string literals.
1954   if (Type == FST_NSString &&
1955       SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart()))
1956     return false;
1957 
1958   // If there are no arguments specified, warn with -Wformat-security, otherwise
1959   // warn only with -Wformat-nonliteral.
1960   if (Args.size() == firstDataArg)
1961     Diag(Args[format_idx]->getLocStart(),
1962          diag::warn_format_nonliteral_noargs)
1963       << OrigFormatExpr->getSourceRange();
1964   else
1965     Diag(Args[format_idx]->getLocStart(),
1966          diag::warn_format_nonliteral)
1967            << OrigFormatExpr->getSourceRange();
1968   return false;
1969 }
1970 
1971 namespace {
1972 class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1973 protected:
1974   Sema &S;
1975   const StringLiteral *FExpr;
1976   const Expr *OrigFormatExpr;
1977   const unsigned FirstDataArg;
1978   const unsigned NumDataArgs;
1979   const char *Beg; // Start of format string.
1980   const bool HasVAListArg;
1981   ArrayRef<const Expr *> Args;
1982   unsigned FormatIdx;
1983   llvm::BitVector CoveredArgs;
1984   bool usesPositionalArgs;
1985   bool atFirstArg;
1986   bool inFunctionCall;
1987   Sema::VariadicCallType CallType;
1988 public:
1989   CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1990                      const Expr *origFormatExpr, unsigned firstDataArg,
1991                      unsigned numDataArgs, const char *beg, bool hasVAListArg,
1992                      ArrayRef<const Expr *> Args,
1993                      unsigned formatIdx, bool inFunctionCall,
1994                      Sema::VariadicCallType callType)
1995     : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1996       FirstDataArg(firstDataArg), NumDataArgs(numDataArgs),
1997       Beg(beg), HasVAListArg(hasVAListArg),
1998       Args(Args), FormatIdx(formatIdx),
1999       usesPositionalArgs(false), atFirstArg(true),
2000       inFunctionCall(inFunctionCall), CallType(callType) {
2001         CoveredArgs.resize(numDataArgs);
2002         CoveredArgs.reset();
2003       }
2004 
2005   void DoneProcessing();
2006 
2007   void HandleIncompleteSpecifier(const char *startSpecifier,
2008                                  unsigned specifierLen);
2009 
2010   void HandleInvalidLengthModifier(
2011       const analyze_format_string::FormatSpecifier &FS,
2012       const analyze_format_string::ConversionSpecifier &CS,
2013       const char *startSpecifier, unsigned specifierLen, unsigned DiagID);
2014 
2015   void HandleNonStandardLengthModifier(
2016       const analyze_format_string::FormatSpecifier &FS,
2017       const char *startSpecifier, unsigned specifierLen);
2018 
2019   void HandleNonStandardConversionSpecifier(
2020       const analyze_format_string::ConversionSpecifier &CS,
2021       const char *startSpecifier, unsigned specifierLen);
2022 
2023   virtual void HandlePosition(const char *startPos, unsigned posLen);
2024 
2025   virtual void HandleInvalidPosition(const char *startSpecifier,
2026                                      unsigned specifierLen,
2027                                      analyze_format_string::PositionContext p);
2028 
2029   virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
2030 
2031   void HandleNullChar(const char *nullCharacter);
2032 
2033   template <typename Range>
2034   static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall,
2035                                    const Expr *ArgumentExpr,
2036                                    PartialDiagnostic PDiag,
2037                                    SourceLocation StringLoc,
2038                                    bool IsStringLocation, Range StringRange,
2039                                    ArrayRef<FixItHint> Fixit = None);
2040 
2041 protected:
2042   bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
2043                                         const char *startSpec,
2044                                         unsigned specifierLen,
2045                                         const char *csStart, unsigned csLen);
2046 
2047   void HandlePositionalNonpositionalArgs(SourceLocation Loc,
2048                                          const char *startSpec,
2049                                          unsigned specifierLen);
2050 
2051   SourceRange getFormatStringRange();
2052   CharSourceRange getSpecifierRange(const char *startSpecifier,
2053                                     unsigned specifierLen);
2054   SourceLocation getLocationOfByte(const char *x);
2055 
2056   const Expr *getDataArg(unsigned i) const;
2057 
2058   bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
2059                     const analyze_format_string::ConversionSpecifier &CS,
2060                     const char *startSpecifier, unsigned specifierLen,
2061                     unsigned argIndex);
2062 
2063   template <typename Range>
2064   void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc,
2065                             bool IsStringLocation, Range StringRange,
2066                             ArrayRef<FixItHint> Fixit = None);
2067 
2068   void CheckPositionalAndNonpositionalArgs(
2069       const analyze_format_string::FormatSpecifier *FS);
2070 };
2071 }
2072 
2073 SourceRange CheckFormatHandler::getFormatStringRange() {
2074   return OrigFormatExpr->getSourceRange();
2075 }
2076 
2077 CharSourceRange CheckFormatHandler::
2078 getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
2079   SourceLocation Start = getLocationOfByte(startSpecifier);
2080   SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
2081 
2082   // Advance the end SourceLocation by one due to half-open ranges.
2083   End = End.getLocWithOffset(1);
2084 
2085   return CharSourceRange::getCharRange(Start, End);
2086 }
2087 
2088 SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
2089   return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
2090 }
2091 
2092 void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
2093                                                    unsigned specifierLen){
2094   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier),
2095                        getLocationOfByte(startSpecifier),
2096                        /*IsStringLocation*/true,
2097                        getSpecifierRange(startSpecifier, specifierLen));
2098 }
2099 
2100 void CheckFormatHandler::HandleInvalidLengthModifier(
2101     const analyze_format_string::FormatSpecifier &FS,
2102     const analyze_format_string::ConversionSpecifier &CS,
2103     const char *startSpecifier, unsigned specifierLen, unsigned DiagID) {
2104   using namespace analyze_format_string;
2105 
2106   const LengthModifier &LM = FS.getLengthModifier();
2107   CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength());
2108 
2109   // See if we know how to fix this length modifier.
2110   Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier();
2111   if (FixedLM) {
2112     EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(),
2113                          getLocationOfByte(LM.getStart()),
2114                          /*IsStringLocation*/true,
2115                          getSpecifierRange(startSpecifier, specifierLen));
2116 
2117     S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier)
2118       << FixedLM->toString()
2119       << FixItHint::CreateReplacement(LMRange, FixedLM->toString());
2120 
2121   } else {
2122     FixItHint Hint;
2123     if (DiagID == diag::warn_format_nonsensical_length)
2124       Hint = FixItHint::CreateRemoval(LMRange);
2125 
2126     EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(),
2127                          getLocationOfByte(LM.getStart()),
2128                          /*IsStringLocation*/true,
2129                          getSpecifierRange(startSpecifier, specifierLen),
2130                          Hint);
2131   }
2132 }
2133 
2134 void CheckFormatHandler::HandleNonStandardLengthModifier(
2135     const analyze_format_string::FormatSpecifier &FS,
2136     const char *startSpecifier, unsigned specifierLen) {
2137   using namespace analyze_format_string;
2138 
2139   const LengthModifier &LM = FS.getLengthModifier();
2140   CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength());
2141 
2142   // See if we know how to fix this length modifier.
2143   Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier();
2144   if (FixedLM) {
2145     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2146                            << LM.toString() << 0,
2147                          getLocationOfByte(LM.getStart()),
2148                          /*IsStringLocation*/true,
2149                          getSpecifierRange(startSpecifier, specifierLen));
2150 
2151     S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier)
2152       << FixedLM->toString()
2153       << FixItHint::CreateReplacement(LMRange, FixedLM->toString());
2154 
2155   } else {
2156     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2157                            << LM.toString() << 0,
2158                          getLocationOfByte(LM.getStart()),
2159                          /*IsStringLocation*/true,
2160                          getSpecifierRange(startSpecifier, specifierLen));
2161   }
2162 }
2163 
2164 void CheckFormatHandler::HandleNonStandardConversionSpecifier(
2165     const analyze_format_string::ConversionSpecifier &CS,
2166     const char *startSpecifier, unsigned specifierLen) {
2167   using namespace analyze_format_string;
2168 
2169   // See if we know how to fix this conversion specifier.
2170   Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier();
2171   if (FixedCS) {
2172     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2173                           << CS.toString() << /*conversion specifier*/1,
2174                          getLocationOfByte(CS.getStart()),
2175                          /*IsStringLocation*/true,
2176                          getSpecifierRange(startSpecifier, specifierLen));
2177 
2178     CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength());
2179     S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier)
2180       << FixedCS->toString()
2181       << FixItHint::CreateReplacement(CSRange, FixedCS->toString());
2182   } else {
2183     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2184                           << CS.toString() << /*conversion specifier*/1,
2185                          getLocationOfByte(CS.getStart()),
2186                          /*IsStringLocation*/true,
2187                          getSpecifierRange(startSpecifier, specifierLen));
2188   }
2189 }
2190 
2191 void CheckFormatHandler::HandlePosition(const char *startPos,
2192                                         unsigned posLen) {
2193   EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg),
2194                                getLocationOfByte(startPos),
2195                                /*IsStringLocation*/true,
2196                                getSpecifierRange(startPos, posLen));
2197 }
2198 
2199 void
2200 CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
2201                                      analyze_format_string::PositionContext p) {
2202   EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier)
2203                          << (unsigned) p,
2204                        getLocationOfByte(startPos), /*IsStringLocation*/true,
2205                        getSpecifierRange(startPos, posLen));
2206 }
2207 
2208 void CheckFormatHandler::HandleZeroPosition(const char *startPos,
2209                                             unsigned posLen) {
2210   EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier),
2211                                getLocationOfByte(startPos),
2212                                /*IsStringLocation*/true,
2213                                getSpecifierRange(startPos, posLen));
2214 }
2215 
2216 void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
2217   if (!isa<ObjCStringLiteral>(OrigFormatExpr)) {
2218     // The presence of a null character is likely an error.
2219     EmitFormatDiagnostic(
2220       S.PDiag(diag::warn_printf_format_string_contains_null_char),
2221       getLocationOfByte(nullCharacter), /*IsStringLocation*/true,
2222       getFormatStringRange());
2223   }
2224 }
2225 
2226 // Note that this may return NULL if there was an error parsing or building
2227 // one of the argument expressions.
2228 const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
2229   return Args[FirstDataArg + i];
2230 }
2231 
2232 void CheckFormatHandler::DoneProcessing() {
2233     // Does the number of data arguments exceed the number of
2234     // format conversions in the format string?
2235   if (!HasVAListArg) {
2236       // Find any arguments that weren't covered.
2237     CoveredArgs.flip();
2238     signed notCoveredArg = CoveredArgs.find_first();
2239     if (notCoveredArg >= 0) {
2240       assert((unsigned)notCoveredArg < NumDataArgs);
2241       if (const Expr *E = getDataArg((unsigned) notCoveredArg)) {
2242         SourceLocation Loc = E->getLocStart();
2243         if (!S.getSourceManager().isInSystemMacro(Loc)) {
2244           EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used),
2245                                Loc, /*IsStringLocation*/false,
2246                                getFormatStringRange());
2247         }
2248       }
2249     }
2250   }
2251 }
2252 
2253 bool
2254 CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
2255                                                      SourceLocation Loc,
2256                                                      const char *startSpec,
2257                                                      unsigned specifierLen,
2258                                                      const char *csStart,
2259                                                      unsigned csLen) {
2260 
2261   bool keepGoing = true;
2262   if (argIndex < NumDataArgs) {
2263     // Consider the argument coverered, even though the specifier doesn't
2264     // make sense.
2265     CoveredArgs.set(argIndex);
2266   }
2267   else {
2268     // If argIndex exceeds the number of data arguments we
2269     // don't issue a warning because that is just a cascade of warnings (and
2270     // they may have intended '%%' anyway). We don't want to continue processing
2271     // the format string after this point, however, as we will like just get
2272     // gibberish when trying to match arguments.
2273     keepGoing = false;
2274   }
2275 
2276   EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion)
2277                          << StringRef(csStart, csLen),
2278                        Loc, /*IsStringLocation*/true,
2279                        getSpecifierRange(startSpec, specifierLen));
2280 
2281   return keepGoing;
2282 }
2283 
2284 void
2285 CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc,
2286                                                       const char *startSpec,
2287                                                       unsigned specifierLen) {
2288   EmitFormatDiagnostic(
2289     S.PDiag(diag::warn_format_mix_positional_nonpositional_args),
2290     Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen));
2291 }
2292 
2293 bool
2294 CheckFormatHandler::CheckNumArgs(
2295   const analyze_format_string::FormatSpecifier &FS,
2296   const analyze_format_string::ConversionSpecifier &CS,
2297   const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
2298 
2299   if (argIndex >= NumDataArgs) {
2300     PartialDiagnostic PDiag = FS.usesPositionalArg()
2301       ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args)
2302            << (argIndex+1) << NumDataArgs)
2303       : S.PDiag(diag::warn_printf_insufficient_data_args);
2304     EmitFormatDiagnostic(
2305       PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true,
2306       getSpecifierRange(startSpecifier, specifierLen));
2307     return false;
2308   }
2309   return true;
2310 }
2311 
2312 template<typename Range>
2313 void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag,
2314                                               SourceLocation Loc,
2315                                               bool IsStringLocation,
2316                                               Range StringRange,
2317                                               ArrayRef<FixItHint> FixIt) {
2318   EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag,
2319                        Loc, IsStringLocation, StringRange, FixIt);
2320 }
2321 
2322 /// \brief If the format string is not within the funcion call, emit a note
2323 /// so that the function call and string are in diagnostic messages.
2324 ///
2325 /// \param InFunctionCall if true, the format string is within the function
2326 /// call and only one diagnostic message will be produced.  Otherwise, an
2327 /// extra note will be emitted pointing to location of the format string.
2328 ///
2329 /// \param ArgumentExpr the expression that is passed as the format string
2330 /// argument in the function call.  Used for getting locations when two
2331 /// diagnostics are emitted.
2332 ///
2333 /// \param PDiag the callee should already have provided any strings for the
2334 /// diagnostic message.  This function only adds locations and fixits
2335 /// to diagnostics.
2336 ///
2337 /// \param Loc primary location for diagnostic.  If two diagnostics are
2338 /// required, one will be at Loc and a new SourceLocation will be created for
2339 /// the other one.
2340 ///
2341 /// \param IsStringLocation if true, Loc points to the format string should be
2342 /// used for the note.  Otherwise, Loc points to the argument list and will
2343 /// be used with PDiag.
2344 ///
2345 /// \param StringRange some or all of the string to highlight.  This is
2346 /// templated so it can accept either a CharSourceRange or a SourceRange.
2347 ///
2348 /// \param FixIt optional fix it hint for the format string.
2349 template<typename Range>
2350 void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall,
2351                                               const Expr *ArgumentExpr,
2352                                               PartialDiagnostic PDiag,
2353                                               SourceLocation Loc,
2354                                               bool IsStringLocation,
2355                                               Range StringRange,
2356                                               ArrayRef<FixItHint> FixIt) {
2357   if (InFunctionCall) {
2358     const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag);
2359     D << StringRange;
2360     for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end();
2361          I != E; ++I) {
2362       D << *I;
2363     }
2364   } else {
2365     S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag)
2366       << ArgumentExpr->getSourceRange();
2367 
2368     const Sema::SemaDiagnosticBuilder &Note =
2369       S.Diag(IsStringLocation ? Loc : StringRange.getBegin(),
2370              diag::note_format_string_defined);
2371 
2372     Note << StringRange;
2373     for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end();
2374          I != E; ++I) {
2375       Note << *I;
2376     }
2377   }
2378 }
2379 
2380 //===--- CHECK: Printf format string checking ------------------------------===//
2381 
2382 namespace {
2383 class CheckPrintfHandler : public CheckFormatHandler {
2384   bool ObjCContext;
2385 public:
2386   CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
2387                      const Expr *origFormatExpr, unsigned firstDataArg,
2388                      unsigned numDataArgs, bool isObjC,
2389                      const char *beg, bool hasVAListArg,
2390                      ArrayRef<const Expr *> Args,
2391                      unsigned formatIdx, bool inFunctionCall,
2392                      Sema::VariadicCallType CallType)
2393   : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
2394                        numDataArgs, beg, hasVAListArg, Args,
2395                        formatIdx, inFunctionCall, CallType), ObjCContext(isObjC)
2396   {}
2397 
2398 
2399   bool HandleInvalidPrintfConversionSpecifier(
2400                                       const analyze_printf::PrintfSpecifier &FS,
2401                                       const char *startSpecifier,
2402                                       unsigned specifierLen);
2403 
2404   bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
2405                              const char *startSpecifier,
2406                              unsigned specifierLen);
2407   bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS,
2408                        const char *StartSpecifier,
2409                        unsigned SpecifierLen,
2410                        const Expr *E);
2411 
2412   bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
2413                     const char *startSpecifier, unsigned specifierLen);
2414   void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
2415                            const analyze_printf::OptionalAmount &Amt,
2416                            unsigned type,
2417                            const char *startSpecifier, unsigned specifierLen);
2418   void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
2419                   const analyze_printf::OptionalFlag &flag,
2420                   const char *startSpecifier, unsigned specifierLen);
2421   void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
2422                          const analyze_printf::OptionalFlag &ignoredFlag,
2423                          const analyze_printf::OptionalFlag &flag,
2424                          const char *startSpecifier, unsigned specifierLen);
2425   bool checkForCStrMembers(const analyze_printf::ArgType &AT,
2426                            const Expr *E, const CharSourceRange &CSR);
2427 
2428 };
2429 }
2430 
2431 bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
2432                                       const analyze_printf::PrintfSpecifier &FS,
2433                                       const char *startSpecifier,
2434                                       unsigned specifierLen) {
2435   const analyze_printf::PrintfConversionSpecifier &CS =
2436     FS.getConversionSpecifier();
2437 
2438   return HandleInvalidConversionSpecifier(FS.getArgIndex(),
2439                                           getLocationOfByte(CS.getStart()),
2440                                           startSpecifier, specifierLen,
2441                                           CS.getStart(), CS.getLength());
2442 }
2443 
2444 bool CheckPrintfHandler::HandleAmount(
2445                                const analyze_format_string::OptionalAmount &Amt,
2446                                unsigned k, const char *startSpecifier,
2447                                unsigned specifierLen) {
2448 
2449   if (Amt.hasDataArgument()) {
2450     if (!HasVAListArg) {
2451       unsigned argIndex = Amt.getArgIndex();
2452       if (argIndex >= NumDataArgs) {
2453         EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg)
2454                                << k,
2455                              getLocationOfByte(Amt.getStart()),
2456                              /*IsStringLocation*/true,
2457                              getSpecifierRange(startSpecifier, specifierLen));
2458         // Don't do any more checking.  We will just emit
2459         // spurious errors.
2460         return false;
2461       }
2462 
2463       // Type check the data argument.  It should be an 'int'.
2464       // Although not in conformance with C99, we also allow the argument to be
2465       // an 'unsigned int' as that is a reasonably safe case.  GCC also
2466       // doesn't emit a warning for that case.
2467       CoveredArgs.set(argIndex);
2468       const Expr *Arg = getDataArg(argIndex);
2469       if (!Arg)
2470         return false;
2471 
2472       QualType T = Arg->getType();
2473 
2474       const analyze_printf::ArgType &AT = Amt.getArgType(S.Context);
2475       assert(AT.isValid());
2476 
2477       if (!AT.matchesType(S.Context, T)) {
2478         EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type)
2479                                << k << AT.getRepresentativeTypeName(S.Context)
2480                                << T << Arg->getSourceRange(),
2481                              getLocationOfByte(Amt.getStart()),
2482                              /*IsStringLocation*/true,
2483                              getSpecifierRange(startSpecifier, specifierLen));
2484         // Don't do any more checking.  We will just emit
2485         // spurious errors.
2486         return false;
2487       }
2488     }
2489   }
2490   return true;
2491 }
2492 
2493 void CheckPrintfHandler::HandleInvalidAmount(
2494                                       const analyze_printf::PrintfSpecifier &FS,
2495                                       const analyze_printf::OptionalAmount &Amt,
2496                                       unsigned type,
2497                                       const char *startSpecifier,
2498                                       unsigned specifierLen) {
2499   const analyze_printf::PrintfConversionSpecifier &CS =
2500     FS.getConversionSpecifier();
2501 
2502   FixItHint fixit =
2503     Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant
2504       ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
2505                                  Amt.getConstantLength()))
2506       : FixItHint();
2507 
2508   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount)
2509                          << type << CS.toString(),
2510                        getLocationOfByte(Amt.getStart()),
2511                        /*IsStringLocation*/true,
2512                        getSpecifierRange(startSpecifier, specifierLen),
2513                        fixit);
2514 }
2515 
2516 void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
2517                                     const analyze_printf::OptionalFlag &flag,
2518                                     const char *startSpecifier,
2519                                     unsigned specifierLen) {
2520   // Warn about pointless flag with a fixit removal.
2521   const analyze_printf::PrintfConversionSpecifier &CS =
2522     FS.getConversionSpecifier();
2523   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag)
2524                          << flag.toString() << CS.toString(),
2525                        getLocationOfByte(flag.getPosition()),
2526                        /*IsStringLocation*/true,
2527                        getSpecifierRange(startSpecifier, specifierLen),
2528                        FixItHint::CreateRemoval(
2529                          getSpecifierRange(flag.getPosition(), 1)));
2530 }
2531 
2532 void CheckPrintfHandler::HandleIgnoredFlag(
2533                                 const analyze_printf::PrintfSpecifier &FS,
2534                                 const analyze_printf::OptionalFlag &ignoredFlag,
2535                                 const analyze_printf::OptionalFlag &flag,
2536                                 const char *startSpecifier,
2537                                 unsigned specifierLen) {
2538   // Warn about ignored flag with a fixit removal.
2539   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag)
2540                          << ignoredFlag.toString() << flag.toString(),
2541                        getLocationOfByte(ignoredFlag.getPosition()),
2542                        /*IsStringLocation*/true,
2543                        getSpecifierRange(startSpecifier, specifierLen),
2544                        FixItHint::CreateRemoval(
2545                          getSpecifierRange(ignoredFlag.getPosition(), 1)));
2546 }
2547 
2548 // Determines if the specified is a C++ class or struct containing
2549 // a member with the specified name and kind (e.g. a CXXMethodDecl named
2550 // "c_str()").
2551 template<typename MemberKind>
2552 static llvm::SmallPtrSet<MemberKind*, 1>
2553 CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) {
2554   const RecordType *RT = Ty->getAs<RecordType>();
2555   llvm::SmallPtrSet<MemberKind*, 1> Results;
2556 
2557   if (!RT)
2558     return Results;
2559   const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl());
2560   if (!RD)
2561     return Results;
2562 
2563   LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(),
2564                  Sema::LookupMemberName);
2565 
2566   // We just need to include all members of the right kind turned up by the
2567   // filter, at this point.
2568   if (S.LookupQualifiedName(R, RT->getDecl()))
2569     for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) {
2570       NamedDecl *decl = (*I)->getUnderlyingDecl();
2571       if (MemberKind *FK = dyn_cast<MemberKind>(decl))
2572         Results.insert(FK);
2573     }
2574   return Results;
2575 }
2576 
2577 // Check if a (w)string was passed when a (w)char* was needed, and offer a
2578 // better diagnostic if so. AT is assumed to be valid.
2579 // Returns true when a c_str() conversion method is found.
2580 bool CheckPrintfHandler::checkForCStrMembers(
2581     const analyze_printf::ArgType &AT, const Expr *E,
2582     const CharSourceRange &CSR) {
2583   typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet;
2584 
2585   MethodSet Results =
2586       CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType());
2587 
2588   for (MethodSet::iterator MI = Results.begin(), ME = Results.end();
2589        MI != ME; ++MI) {
2590     const CXXMethodDecl *Method = *MI;
2591     if (Method->getNumParams() == 0 &&
2592           AT.matchesType(S.Context, Method->getResultType())) {
2593       // FIXME: Suggest parens if the expression needs them.
2594       SourceLocation EndLoc =
2595           S.getPreprocessor().getLocForEndOfToken(E->getLocEnd());
2596       S.Diag(E->getLocStart(), diag::note_printf_c_str)
2597           << "c_str()"
2598           << FixItHint::CreateInsertion(EndLoc, ".c_str()");
2599       return true;
2600     }
2601   }
2602 
2603   return false;
2604 }
2605 
2606 bool
2607 CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
2608                                             &FS,
2609                                           const char *startSpecifier,
2610                                           unsigned specifierLen) {
2611 
2612   using namespace analyze_format_string;
2613   using namespace analyze_printf;
2614   const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
2615 
2616   if (FS.consumesDataArgument()) {
2617     if (atFirstArg) {
2618         atFirstArg = false;
2619         usesPositionalArgs = FS.usesPositionalArg();
2620     }
2621     else if (usesPositionalArgs != FS.usesPositionalArg()) {
2622       HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()),
2623                                         startSpecifier, specifierLen);
2624       return false;
2625     }
2626   }
2627 
2628   // First check if the field width, precision, and conversion specifier
2629   // have matching data arguments.
2630   if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
2631                     startSpecifier, specifierLen)) {
2632     return false;
2633   }
2634 
2635   if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
2636                     startSpecifier, specifierLen)) {
2637     return false;
2638   }
2639 
2640   if (!CS.consumesDataArgument()) {
2641     // FIXME: Technically specifying a precision or field width here
2642     // makes no sense.  Worth issuing a warning at some point.
2643     return true;
2644   }
2645 
2646   // Consume the argument.
2647   unsigned argIndex = FS.getArgIndex();
2648   if (argIndex < NumDataArgs) {
2649     // The check to see if the argIndex is valid will come later.
2650     // We set the bit here because we may exit early from this
2651     // function if we encounter some other error.
2652     CoveredArgs.set(argIndex);
2653   }
2654 
2655   // Check for using an Objective-C specific conversion specifier
2656   // in a non-ObjC literal.
2657   if (!ObjCContext && CS.isObjCArg()) {
2658     return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
2659                                                   specifierLen);
2660   }
2661 
2662   // Check for invalid use of field width
2663   if (!FS.hasValidFieldWidth()) {
2664     HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
2665         startSpecifier, specifierLen);
2666   }
2667 
2668   // Check for invalid use of precision
2669   if (!FS.hasValidPrecision()) {
2670     HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
2671         startSpecifier, specifierLen);
2672   }
2673 
2674   // Check each flag does not conflict with any other component.
2675   if (!FS.hasValidThousandsGroupingPrefix())
2676     HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
2677   if (!FS.hasValidLeadingZeros())
2678     HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
2679   if (!FS.hasValidPlusPrefix())
2680     HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
2681   if (!FS.hasValidSpacePrefix())
2682     HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
2683   if (!FS.hasValidAlternativeForm())
2684     HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
2685   if (!FS.hasValidLeftJustified())
2686     HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
2687 
2688   // Check that flags are not ignored by another flag
2689   if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
2690     HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
2691         startSpecifier, specifierLen);
2692   if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
2693     HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
2694             startSpecifier, specifierLen);
2695 
2696   // Check the length modifier is valid with the given conversion specifier.
2697   if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo()))
2698     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
2699                                 diag::warn_format_nonsensical_length);
2700   else if (!FS.hasStandardLengthModifier())
2701     HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen);
2702   else if (!FS.hasStandardLengthConversionCombination())
2703     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
2704                                 diag::warn_format_non_standard_conversion_spec);
2705 
2706   if (!FS.hasStandardConversionSpecifier(S.getLangOpts()))
2707     HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen);
2708 
2709   // The remaining checks depend on the data arguments.
2710   if (HasVAListArg)
2711     return true;
2712 
2713   if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
2714     return false;
2715 
2716   const Expr *Arg = getDataArg(argIndex);
2717   if (!Arg)
2718     return true;
2719 
2720   return checkFormatExpr(FS, startSpecifier, specifierLen, Arg);
2721 }
2722 
2723 static bool requiresParensToAddCast(const Expr *E) {
2724   // FIXME: We should have a general way to reason about operator
2725   // precedence and whether parens are actually needed here.
2726   // Take care of a few common cases where they aren't.
2727   const Expr *Inside = E->IgnoreImpCasts();
2728   if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside))
2729     Inside = POE->getSyntacticForm()->IgnoreImpCasts();
2730 
2731   switch (Inside->getStmtClass()) {
2732   case Stmt::ArraySubscriptExprClass:
2733   case Stmt::CallExprClass:
2734   case Stmt::CharacterLiteralClass:
2735   case Stmt::CXXBoolLiteralExprClass:
2736   case Stmt::DeclRefExprClass:
2737   case Stmt::FloatingLiteralClass:
2738   case Stmt::IntegerLiteralClass:
2739   case Stmt::MemberExprClass:
2740   case Stmt::ObjCArrayLiteralClass:
2741   case Stmt::ObjCBoolLiteralExprClass:
2742   case Stmt::ObjCBoxedExprClass:
2743   case Stmt::ObjCDictionaryLiteralClass:
2744   case Stmt::ObjCEncodeExprClass:
2745   case Stmt::ObjCIvarRefExprClass:
2746   case Stmt::ObjCMessageExprClass:
2747   case Stmt::ObjCPropertyRefExprClass:
2748   case Stmt::ObjCStringLiteralClass:
2749   case Stmt::ObjCSubscriptRefExprClass:
2750   case Stmt::ParenExprClass:
2751   case Stmt::StringLiteralClass:
2752   case Stmt::UnaryOperatorClass:
2753     return false;
2754   default:
2755     return true;
2756   }
2757 }
2758 
2759 bool
2760 CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS,
2761                                     const char *StartSpecifier,
2762                                     unsigned SpecifierLen,
2763                                     const Expr *E) {
2764   using namespace analyze_format_string;
2765   using namespace analyze_printf;
2766   // Now type check the data expression that matches the
2767   // format specifier.
2768   const analyze_printf::ArgType &AT = FS.getArgType(S.Context,
2769                                                     ObjCContext);
2770   if (!AT.isValid())
2771     return true;
2772 
2773   QualType ExprTy = E->getType();
2774   while (const TypeOfExprType *TET = dyn_cast<TypeOfExprType>(ExprTy)) {
2775     ExprTy = TET->getUnderlyingExpr()->getType();
2776   }
2777 
2778   if (AT.matchesType(S.Context, ExprTy))
2779     return true;
2780 
2781   // Look through argument promotions for our error message's reported type.
2782   // This includes the integral and floating promotions, but excludes array
2783   // and function pointer decay; seeing that an argument intended to be a
2784   // string has type 'char [6]' is probably more confusing than 'char *'.
2785   if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
2786     if (ICE->getCastKind() == CK_IntegralCast ||
2787         ICE->getCastKind() == CK_FloatingCast) {
2788       E = ICE->getSubExpr();
2789       ExprTy = E->getType();
2790 
2791       // Check if we didn't match because of an implicit cast from a 'char'
2792       // or 'short' to an 'int'.  This is done because printf is a varargs
2793       // function.
2794       if (ICE->getType() == S.Context.IntTy ||
2795           ICE->getType() == S.Context.UnsignedIntTy) {
2796         // All further checking is done on the subexpression.
2797         if (AT.matchesType(S.Context, ExprTy))
2798           return true;
2799       }
2800     }
2801   } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) {
2802     // Special case for 'a', which has type 'int' in C.
2803     // Note, however, that we do /not/ want to treat multibyte constants like
2804     // 'MooV' as characters! This form is deprecated but still exists.
2805     if (ExprTy == S.Context.IntTy)
2806       if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue()))
2807         ExprTy = S.Context.CharTy;
2808   }
2809 
2810   // %C in an Objective-C context prints a unichar, not a wchar_t.
2811   // If the argument is an integer of some kind, believe the %C and suggest
2812   // a cast instead of changing the conversion specifier.
2813   QualType IntendedTy = ExprTy;
2814   if (ObjCContext &&
2815       FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) {
2816     if (ExprTy->isIntegralOrUnscopedEnumerationType() &&
2817         !ExprTy->isCharType()) {
2818       // 'unichar' is defined as a typedef of unsigned short, but we should
2819       // prefer using the typedef if it is visible.
2820       IntendedTy = S.Context.UnsignedShortTy;
2821 
2822       LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(),
2823                           Sema::LookupOrdinaryName);
2824       if (S.LookupName(Result, S.getCurScope())) {
2825         NamedDecl *ND = Result.getFoundDecl();
2826         if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND))
2827           if (TD->getUnderlyingType() == IntendedTy)
2828             IntendedTy = S.Context.getTypedefType(TD);
2829       }
2830     }
2831   }
2832 
2833   // Special-case some of Darwin's platform-independence types by suggesting
2834   // casts to primitive types that are known to be large enough.
2835   bool ShouldNotPrintDirectly = false;
2836   if (S.Context.getTargetInfo().getTriple().isOSDarwin()) {
2837     // Use a 'while' to peel off layers of typedefs.
2838     QualType TyTy = IntendedTy;
2839     while (const TypedefType *UserTy = TyTy->getAs<TypedefType>()) {
2840       StringRef Name = UserTy->getDecl()->getName();
2841       QualType CastTy = llvm::StringSwitch<QualType>(Name)
2842         .Case("NSInteger", S.Context.LongTy)
2843         .Case("NSUInteger", S.Context.UnsignedLongTy)
2844         .Case("SInt32", S.Context.IntTy)
2845         .Case("UInt32", S.Context.UnsignedIntTy)
2846         .Default(QualType());
2847 
2848       if (!CastTy.isNull()) {
2849         ShouldNotPrintDirectly = true;
2850         IntendedTy = CastTy;
2851         break;
2852       }
2853       TyTy = UserTy->desugar();
2854     }
2855   }
2856 
2857   // We may be able to offer a FixItHint if it is a supported type.
2858   PrintfSpecifier fixedFS = FS;
2859   bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(),
2860                                  S.Context, ObjCContext);
2861 
2862   if (success) {
2863     // Get the fix string from the fixed format specifier
2864     SmallString<16> buf;
2865     llvm::raw_svector_ostream os(buf);
2866     fixedFS.toString(os);
2867 
2868     CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen);
2869 
2870     if (IntendedTy == ExprTy) {
2871       // In this case, the specifier is wrong and should be changed to match
2872       // the argument.
2873       EmitFormatDiagnostic(
2874         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2875           << AT.getRepresentativeTypeName(S.Context) << IntendedTy
2876           << E->getSourceRange(),
2877         E->getLocStart(),
2878         /*IsStringLocation*/false,
2879         SpecRange,
2880         FixItHint::CreateReplacement(SpecRange, os.str()));
2881 
2882     } else {
2883       // The canonical type for formatting this value is different from the
2884       // actual type of the expression. (This occurs, for example, with Darwin's
2885       // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but
2886       // should be printed as 'long' for 64-bit compatibility.)
2887       // Rather than emitting a normal format/argument mismatch, we want to
2888       // add a cast to the recommended type (and correct the format string
2889       // if necessary).
2890       SmallString<16> CastBuf;
2891       llvm::raw_svector_ostream CastFix(CastBuf);
2892       CastFix << "(";
2893       IntendedTy.print(CastFix, S.Context.getPrintingPolicy());
2894       CastFix << ")";
2895 
2896       SmallVector<FixItHint,4> Hints;
2897       if (!AT.matchesType(S.Context, IntendedTy))
2898         Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str()));
2899 
2900       if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) {
2901         // If there's already a cast present, just replace it.
2902         SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc());
2903         Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str()));
2904 
2905       } else if (!requiresParensToAddCast(E)) {
2906         // If the expression has high enough precedence,
2907         // just write the C-style cast.
2908         Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(),
2909                                                    CastFix.str()));
2910       } else {
2911         // Otherwise, add parens around the expression as well as the cast.
2912         CastFix << "(";
2913         Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(),
2914                                                    CastFix.str()));
2915 
2916         SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd());
2917         Hints.push_back(FixItHint::CreateInsertion(After, ")"));
2918       }
2919 
2920       if (ShouldNotPrintDirectly) {
2921         // The expression has a type that should not be printed directly.
2922         // We extract the name from the typedef because we don't want to show
2923         // the underlying type in the diagnostic.
2924         StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName();
2925 
2926         EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast)
2927                                << Name << IntendedTy
2928                                << E->getSourceRange(),
2929                              E->getLocStart(), /*IsStringLocation=*/false,
2930                              SpecRange, Hints);
2931       } else {
2932         // In this case, the expression could be printed using a different
2933         // specifier, but we've decided that the specifier is probably correct
2934         // and we should cast instead. Just use the normal warning message.
2935         EmitFormatDiagnostic(
2936           S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2937             << AT.getRepresentativeTypeName(S.Context) << ExprTy
2938             << E->getSourceRange(),
2939           E->getLocStart(), /*IsStringLocation*/false,
2940           SpecRange, Hints);
2941       }
2942     }
2943   } else {
2944     const CharSourceRange &CSR = getSpecifierRange(StartSpecifier,
2945                                                    SpecifierLen);
2946     // Since the warning for passing non-POD types to variadic functions
2947     // was deferred until now, we emit a warning for non-POD
2948     // arguments here.
2949     if (S.isValidVarArgType(ExprTy) == Sema::VAK_Invalid) {
2950       unsigned DiagKind;
2951       if (ExprTy->isObjCObjectType())
2952         DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format;
2953       else
2954         DiagKind = diag::warn_non_pod_vararg_with_format_string;
2955 
2956       EmitFormatDiagnostic(
2957         S.PDiag(DiagKind)
2958           << S.getLangOpts().CPlusPlus11
2959           << ExprTy
2960           << CallType
2961           << AT.getRepresentativeTypeName(S.Context)
2962           << CSR
2963           << E->getSourceRange(),
2964         E->getLocStart(), /*IsStringLocation*/false, CSR);
2965 
2966       checkForCStrMembers(AT, E, CSR);
2967     } else
2968       EmitFormatDiagnostic(
2969         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2970           << AT.getRepresentativeTypeName(S.Context) << ExprTy
2971           << CSR
2972           << E->getSourceRange(),
2973         E->getLocStart(), /*IsStringLocation*/false, CSR);
2974   }
2975 
2976   return true;
2977 }
2978 
2979 //===--- CHECK: Scanf format string checking ------------------------------===//
2980 
2981 namespace {
2982 class CheckScanfHandler : public CheckFormatHandler {
2983 public:
2984   CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
2985                     const Expr *origFormatExpr, unsigned firstDataArg,
2986                     unsigned numDataArgs, const char *beg, bool hasVAListArg,
2987                     ArrayRef<const Expr *> Args,
2988                     unsigned formatIdx, bool inFunctionCall,
2989                     Sema::VariadicCallType CallType)
2990   : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
2991                        numDataArgs, beg, hasVAListArg,
2992                        Args, formatIdx, inFunctionCall, CallType)
2993   {}
2994 
2995   bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
2996                             const char *startSpecifier,
2997                             unsigned specifierLen);
2998 
2999   bool HandleInvalidScanfConversionSpecifier(
3000           const analyze_scanf::ScanfSpecifier &FS,
3001           const char *startSpecifier,
3002           unsigned specifierLen);
3003 
3004   void HandleIncompleteScanList(const char *start, const char *end);
3005 };
3006 }
3007 
3008 void CheckScanfHandler::HandleIncompleteScanList(const char *start,
3009                                                  const char *end) {
3010   EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete),
3011                        getLocationOfByte(end), /*IsStringLocation*/true,
3012                        getSpecifierRange(start, end - start));
3013 }
3014 
3015 bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
3016                                         const analyze_scanf::ScanfSpecifier &FS,
3017                                         const char *startSpecifier,
3018                                         unsigned specifierLen) {
3019 
3020   const analyze_scanf::ScanfConversionSpecifier &CS =
3021     FS.getConversionSpecifier();
3022 
3023   return HandleInvalidConversionSpecifier(FS.getArgIndex(),
3024                                           getLocationOfByte(CS.getStart()),
3025                                           startSpecifier, specifierLen,
3026                                           CS.getStart(), CS.getLength());
3027 }
3028 
3029 bool CheckScanfHandler::HandleScanfSpecifier(
3030                                        const analyze_scanf::ScanfSpecifier &FS,
3031                                        const char *startSpecifier,
3032                                        unsigned specifierLen) {
3033 
3034   using namespace analyze_scanf;
3035   using namespace analyze_format_string;
3036 
3037   const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
3038 
3039   // Handle case where '%' and '*' don't consume an argument.  These shouldn't
3040   // be used to decide if we are using positional arguments consistently.
3041   if (FS.consumesDataArgument()) {
3042     if (atFirstArg) {
3043       atFirstArg = false;
3044       usesPositionalArgs = FS.usesPositionalArg();
3045     }
3046     else if (usesPositionalArgs != FS.usesPositionalArg()) {
3047       HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()),
3048                                         startSpecifier, specifierLen);
3049       return false;
3050     }
3051   }
3052 
3053   // Check if the field with is non-zero.
3054   const OptionalAmount &Amt = FS.getFieldWidth();
3055   if (Amt.getHowSpecified() == OptionalAmount::Constant) {
3056     if (Amt.getConstantAmount() == 0) {
3057       const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
3058                                                    Amt.getConstantLength());
3059       EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width),
3060                            getLocationOfByte(Amt.getStart()),
3061                            /*IsStringLocation*/true, R,
3062                            FixItHint::CreateRemoval(R));
3063     }
3064   }
3065 
3066   if (!FS.consumesDataArgument()) {
3067     // FIXME: Technically specifying a precision or field width here
3068     // makes no sense.  Worth issuing a warning at some point.
3069     return true;
3070   }
3071 
3072   // Consume the argument.
3073   unsigned argIndex = FS.getArgIndex();
3074   if (argIndex < NumDataArgs) {
3075       // The check to see if the argIndex is valid will come later.
3076       // We set the bit here because we may exit early from this
3077       // function if we encounter some other error.
3078     CoveredArgs.set(argIndex);
3079   }
3080 
3081   // Check the length modifier is valid with the given conversion specifier.
3082   if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo()))
3083     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
3084                                 diag::warn_format_nonsensical_length);
3085   else if (!FS.hasStandardLengthModifier())
3086     HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen);
3087   else if (!FS.hasStandardLengthConversionCombination())
3088     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
3089                                 diag::warn_format_non_standard_conversion_spec);
3090 
3091   if (!FS.hasStandardConversionSpecifier(S.getLangOpts()))
3092     HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen);
3093 
3094   // The remaining checks depend on the data arguments.
3095   if (HasVAListArg)
3096     return true;
3097 
3098   if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
3099     return false;
3100 
3101   // Check that the argument type matches the format specifier.
3102   const Expr *Ex = getDataArg(argIndex);
3103   if (!Ex)
3104     return true;
3105 
3106   const analyze_format_string::ArgType &AT = FS.getArgType(S.Context);
3107   if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) {
3108     ScanfSpecifier fixedFS = FS;
3109     bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(),
3110                                    S.Context);
3111 
3112     if (success) {
3113       // Get the fix string from the fixed format specifier.
3114       SmallString<128> buf;
3115       llvm::raw_svector_ostream os(buf);
3116       fixedFS.toString(os);
3117 
3118       EmitFormatDiagnostic(
3119         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
3120           << AT.getRepresentativeTypeName(S.Context) << Ex->getType()
3121           << Ex->getSourceRange(),
3122         Ex->getLocStart(),
3123         /*IsStringLocation*/false,
3124         getSpecifierRange(startSpecifier, specifierLen),
3125         FixItHint::CreateReplacement(
3126           getSpecifierRange(startSpecifier, specifierLen),
3127           os.str()));
3128     } else {
3129       EmitFormatDiagnostic(
3130         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
3131           << AT.getRepresentativeTypeName(S.Context) << Ex->getType()
3132           << Ex->getSourceRange(),
3133         Ex->getLocStart(),
3134         /*IsStringLocation*/false,
3135         getSpecifierRange(startSpecifier, specifierLen));
3136     }
3137   }
3138 
3139   return true;
3140 }
3141 
3142 void Sema::CheckFormatString(const StringLiteral *FExpr,
3143                              const Expr *OrigFormatExpr,
3144                              ArrayRef<const Expr *> Args,
3145                              bool HasVAListArg, unsigned format_idx,
3146                              unsigned firstDataArg, FormatStringType Type,
3147                              bool inFunctionCall, VariadicCallType CallType) {
3148 
3149   // CHECK: is the format string a wide literal?
3150   if (!FExpr->isAscii() && !FExpr->isUTF8()) {
3151     CheckFormatHandler::EmitFormatDiagnostic(
3152       *this, inFunctionCall, Args[format_idx],
3153       PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(),
3154       /*IsStringLocation*/true, OrigFormatExpr->getSourceRange());
3155     return;
3156   }
3157 
3158   // Str - The format string.  NOTE: this is NOT null-terminated!
3159   StringRef StrRef = FExpr->getString();
3160   const char *Str = StrRef.data();
3161   unsigned StrLen = StrRef.size();
3162   const unsigned numDataArgs = Args.size() - firstDataArg;
3163 
3164   // CHECK: empty format string?
3165   if (StrLen == 0 && numDataArgs > 0) {
3166     CheckFormatHandler::EmitFormatDiagnostic(
3167       *this, inFunctionCall, Args[format_idx],
3168       PDiag(diag::warn_empty_format_string), FExpr->getLocStart(),
3169       /*IsStringLocation*/true, OrigFormatExpr->getSourceRange());
3170     return;
3171   }
3172 
3173   if (Type == FST_Printf || Type == FST_NSString) {
3174     CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
3175                          numDataArgs, (Type == FST_NSString),
3176                          Str, HasVAListArg, Args, format_idx,
3177                          inFunctionCall, CallType);
3178 
3179     if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen,
3180                                                   getLangOpts(),
3181                                                   Context.getTargetInfo()))
3182       H.DoneProcessing();
3183   } else if (Type == FST_Scanf) {
3184     CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs,
3185                         Str, HasVAListArg, Args, format_idx,
3186                         inFunctionCall, CallType);
3187 
3188     if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen,
3189                                                  getLangOpts(),
3190                                                  Context.getTargetInfo()))
3191       H.DoneProcessing();
3192   } // TODO: handle other formats
3193 }
3194 
3195 //===--- CHECK: Standard memory functions ---------------------------------===//
3196 
3197 /// \brief Determine whether the given type is a dynamic class type (e.g.,
3198 /// whether it has a vtable).
3199 static bool isDynamicClassType(QualType T) {
3200   if (CXXRecordDecl *Record = T->getAsCXXRecordDecl())
3201     if (CXXRecordDecl *Definition = Record->getDefinition())
3202       if (Definition->isDynamicClass())
3203         return true;
3204 
3205   return false;
3206 }
3207 
3208 /// \brief If E is a sizeof expression, returns its argument expression,
3209 /// otherwise returns NULL.
3210 static const Expr *getSizeOfExprArg(const Expr* E) {
3211   if (const UnaryExprOrTypeTraitExpr *SizeOf =
3212       dyn_cast<UnaryExprOrTypeTraitExpr>(E))
3213     if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType())
3214       return SizeOf->getArgumentExpr()->IgnoreParenImpCasts();
3215 
3216   return 0;
3217 }
3218 
3219 /// \brief If E is a sizeof expression, returns its argument type.
3220 static QualType getSizeOfArgType(const Expr* E) {
3221   if (const UnaryExprOrTypeTraitExpr *SizeOf =
3222       dyn_cast<UnaryExprOrTypeTraitExpr>(E))
3223     if (SizeOf->getKind() == clang::UETT_SizeOf)
3224       return SizeOf->getTypeOfArgument();
3225 
3226   return QualType();
3227 }
3228 
3229 /// \brief Check for dangerous or invalid arguments to memset().
3230 ///
3231 /// This issues warnings on known problematic, dangerous or unspecified
3232 /// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp'
3233 /// function calls.
3234 ///
3235 /// \param Call The call expression to diagnose.
3236 void Sema::CheckMemaccessArguments(const CallExpr *Call,
3237                                    unsigned BId,
3238                                    IdentifierInfo *FnName) {
3239   assert(BId != 0);
3240 
3241   // It is possible to have a non-standard definition of memset.  Validate
3242   // we have enough arguments, and if not, abort further checking.
3243   unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3);
3244   if (Call->getNumArgs() < ExpectedNumArgs)
3245     return;
3246 
3247   unsigned LastArg = (BId == Builtin::BImemset ||
3248                       BId == Builtin::BIstrndup ? 1 : 2);
3249   unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2);
3250   const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts();
3251 
3252   // We have special checking when the length is a sizeof expression.
3253   QualType SizeOfArgTy = getSizeOfArgType(LenExpr);
3254   const Expr *SizeOfArg = getSizeOfExprArg(LenExpr);
3255   llvm::FoldingSetNodeID SizeOfArgID;
3256 
3257   for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) {
3258     const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts();
3259     SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange();
3260 
3261     QualType DestTy = Dest->getType();
3262     if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
3263       QualType PointeeTy = DestPtrTy->getPointeeType();
3264 
3265       // Never warn about void type pointers. This can be used to suppress
3266       // false positives.
3267       if (PointeeTy->isVoidType())
3268         continue;
3269 
3270       // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by
3271       // actually comparing the expressions for equality. Because computing the
3272       // expression IDs can be expensive, we only do this if the diagnostic is
3273       // enabled.
3274       if (SizeOfArg &&
3275           Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess,
3276                                    SizeOfArg->getExprLoc())) {
3277         // We only compute IDs for expressions if the warning is enabled, and
3278         // cache the sizeof arg's ID.
3279         if (SizeOfArgID == llvm::FoldingSetNodeID())
3280           SizeOfArg->Profile(SizeOfArgID, Context, true);
3281         llvm::FoldingSetNodeID DestID;
3282         Dest->Profile(DestID, Context, true);
3283         if (DestID == SizeOfArgID) {
3284           // TODO: For strncpy() and friends, this could suggest sizeof(dst)
3285           //       over sizeof(src) as well.
3286           unsigned ActionIdx = 0; // Default is to suggest dereferencing.
3287           StringRef ReadableName = FnName->getName();
3288 
3289           if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest))
3290             if (UnaryOp->getOpcode() == UO_AddrOf)
3291               ActionIdx = 1; // If its an address-of operator, just remove it.
3292           if (!PointeeTy->isIncompleteType() &&
3293               (Context.getTypeSize(PointeeTy) == Context.getCharWidth()))
3294             ActionIdx = 2; // If the pointee's size is sizeof(char),
3295                            // suggest an explicit length.
3296 
3297           // If the function is defined as a builtin macro, do not show macro
3298           // expansion.
3299           SourceLocation SL = SizeOfArg->getExprLoc();
3300           SourceRange DSR = Dest->getSourceRange();
3301           SourceRange SSR = SizeOfArg->getSourceRange();
3302           SourceManager &SM  = PP.getSourceManager();
3303 
3304           if (SM.isMacroArgExpansion(SL)) {
3305             ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts);
3306             SL = SM.getSpellingLoc(SL);
3307             DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()),
3308                              SM.getSpellingLoc(DSR.getEnd()));
3309             SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()),
3310                              SM.getSpellingLoc(SSR.getEnd()));
3311           }
3312 
3313           DiagRuntimeBehavior(SL, SizeOfArg,
3314                               PDiag(diag::warn_sizeof_pointer_expr_memaccess)
3315                                 << ReadableName
3316                                 << PointeeTy
3317                                 << DestTy
3318                                 << DSR
3319                                 << SSR);
3320           DiagRuntimeBehavior(SL, SizeOfArg,
3321                          PDiag(diag::warn_sizeof_pointer_expr_memaccess_note)
3322                                 << ActionIdx
3323                                 << SSR);
3324 
3325           break;
3326         }
3327       }
3328 
3329       // Also check for cases where the sizeof argument is the exact same
3330       // type as the memory argument, and where it points to a user-defined
3331       // record type.
3332       if (SizeOfArgTy != QualType()) {
3333         if (PointeeTy->isRecordType() &&
3334             Context.typesAreCompatible(SizeOfArgTy, DestTy)) {
3335           DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest,
3336                               PDiag(diag::warn_sizeof_pointer_type_memaccess)
3337                                 << FnName << SizeOfArgTy << ArgIdx
3338                                 << PointeeTy << Dest->getSourceRange()
3339                                 << LenExpr->getSourceRange());
3340           break;
3341         }
3342       }
3343 
3344       // Always complain about dynamic classes.
3345       if (isDynamicClassType(PointeeTy)) {
3346 
3347         unsigned OperationType = 0;
3348         // "overwritten" if we're warning about the destination for any call
3349         // but memcmp; otherwise a verb appropriate to the call.
3350         if (ArgIdx != 0 || BId == Builtin::BImemcmp) {
3351           if (BId == Builtin::BImemcpy)
3352             OperationType = 1;
3353           else if(BId == Builtin::BImemmove)
3354             OperationType = 2;
3355           else if (BId == Builtin::BImemcmp)
3356             OperationType = 3;
3357         }
3358 
3359         DiagRuntimeBehavior(
3360           Dest->getExprLoc(), Dest,
3361           PDiag(diag::warn_dyn_class_memaccess)
3362             << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx)
3363             << FnName << PointeeTy
3364             << OperationType
3365             << Call->getCallee()->getSourceRange());
3366       } else if (PointeeTy.hasNonTrivialObjCLifetime() &&
3367                BId != Builtin::BImemset)
3368         DiagRuntimeBehavior(
3369           Dest->getExprLoc(), Dest,
3370           PDiag(diag::warn_arc_object_memaccess)
3371             << ArgIdx << FnName << PointeeTy
3372             << Call->getCallee()->getSourceRange());
3373       else
3374         continue;
3375 
3376       DiagRuntimeBehavior(
3377         Dest->getExprLoc(), Dest,
3378         PDiag(diag::note_bad_memaccess_silence)
3379           << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
3380       break;
3381     }
3382   }
3383 }
3384 
3385 // A little helper routine: ignore addition and subtraction of integer literals.
3386 // This intentionally does not ignore all integer constant expressions because
3387 // we don't want to remove sizeof().
3388 static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) {
3389   Ex = Ex->IgnoreParenCasts();
3390 
3391   for (;;) {
3392     const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex);
3393     if (!BO || !BO->isAdditiveOp())
3394       break;
3395 
3396     const Expr *RHS = BO->getRHS()->IgnoreParenCasts();
3397     const Expr *LHS = BO->getLHS()->IgnoreParenCasts();
3398 
3399     if (isa<IntegerLiteral>(RHS))
3400       Ex = LHS;
3401     else if (isa<IntegerLiteral>(LHS))
3402       Ex = RHS;
3403     else
3404       break;
3405   }
3406 
3407   return Ex;
3408 }
3409 
3410 static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty,
3411                                                       ASTContext &Context) {
3412   // Only handle constant-sized or VLAs, but not flexible members.
3413   if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) {
3414     // Only issue the FIXIT for arrays of size > 1.
3415     if (CAT->getSize().getSExtValue() <= 1)
3416       return false;
3417   } else if (!Ty->isVariableArrayType()) {
3418     return false;
3419   }
3420   return true;
3421 }
3422 
3423 // Warn if the user has made the 'size' argument to strlcpy or strlcat
3424 // be the size of the source, instead of the destination.
3425 void Sema::CheckStrlcpycatArguments(const CallExpr *Call,
3426                                     IdentifierInfo *FnName) {
3427 
3428   // Don't crash if the user has the wrong number of arguments
3429   if (Call->getNumArgs() != 3)
3430     return;
3431 
3432   const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context);
3433   const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context);
3434   const Expr *CompareWithSrc = NULL;
3435 
3436   // Look for 'strlcpy(dst, x, sizeof(x))'
3437   if (const Expr *Ex = getSizeOfExprArg(SizeArg))
3438     CompareWithSrc = Ex;
3439   else {
3440     // Look for 'strlcpy(dst, x, strlen(x))'
3441     if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) {
3442       if (SizeCall->isBuiltinCall() == Builtin::BIstrlen
3443           && SizeCall->getNumArgs() == 1)
3444         CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context);
3445     }
3446   }
3447 
3448   if (!CompareWithSrc)
3449     return;
3450 
3451   // Determine if the argument to sizeof/strlen is equal to the source
3452   // argument.  In principle there's all kinds of things you could do
3453   // here, for instance creating an == expression and evaluating it with
3454   // EvaluateAsBooleanCondition, but this uses a more direct technique:
3455   const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg);
3456   if (!SrcArgDRE)
3457     return;
3458 
3459   const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc);
3460   if (!CompareWithSrcDRE ||
3461       SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl())
3462     return;
3463 
3464   const Expr *OriginalSizeArg = Call->getArg(2);
3465   Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size)
3466     << OriginalSizeArg->getSourceRange() << FnName;
3467 
3468   // Output a FIXIT hint if the destination is an array (rather than a
3469   // pointer to an array).  This could be enhanced to handle some
3470   // pointers if we know the actual size, like if DstArg is 'array+2'
3471   // we could say 'sizeof(array)-2'.
3472   const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts();
3473   if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context))
3474     return;
3475 
3476   SmallString<128> sizeString;
3477   llvm::raw_svector_ostream OS(sizeString);
3478   OS << "sizeof(";
3479   DstArg->printPretty(OS, 0, getPrintingPolicy());
3480   OS << ")";
3481 
3482   Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size)
3483     << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(),
3484                                     OS.str());
3485 }
3486 
3487 /// Check if two expressions refer to the same declaration.
3488 static bool referToTheSameDecl(const Expr *E1, const Expr *E2) {
3489   if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1))
3490     if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2))
3491       return D1->getDecl() == D2->getDecl();
3492   return false;
3493 }
3494 
3495 static const Expr *getStrlenExprArg(const Expr *E) {
3496   if (const CallExpr *CE = dyn_cast<CallExpr>(E)) {
3497     const FunctionDecl *FD = CE->getDirectCallee();
3498     if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen)
3499       return 0;
3500     return CE->getArg(0)->IgnoreParenCasts();
3501   }
3502   return 0;
3503 }
3504 
3505 // Warn on anti-patterns as the 'size' argument to strncat.
3506 // The correct size argument should look like following:
3507 //   strncat(dst, src, sizeof(dst) - strlen(dest) - 1);
3508 void Sema::CheckStrncatArguments(const CallExpr *CE,
3509                                  IdentifierInfo *FnName) {
3510   // Don't crash if the user has the wrong number of arguments.
3511   if (CE->getNumArgs() < 3)
3512     return;
3513   const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts();
3514   const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts();
3515   const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts();
3516 
3517   // Identify common expressions, which are wrongly used as the size argument
3518   // to strncat and may lead to buffer overflows.
3519   unsigned PatternType = 0;
3520   if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) {
3521     // - sizeof(dst)
3522     if (referToTheSameDecl(SizeOfArg, DstArg))
3523       PatternType = 1;
3524     // - sizeof(src)
3525     else if (referToTheSameDecl(SizeOfArg, SrcArg))
3526       PatternType = 2;
3527   } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) {
3528     if (BE->getOpcode() == BO_Sub) {
3529       const Expr *L = BE->getLHS()->IgnoreParenCasts();
3530       const Expr *R = BE->getRHS()->IgnoreParenCasts();
3531       // - sizeof(dst) - strlen(dst)
3532       if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) &&
3533           referToTheSameDecl(DstArg, getStrlenExprArg(R)))
3534         PatternType = 1;
3535       // - sizeof(src) - (anything)
3536       else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L)))
3537         PatternType = 2;
3538     }
3539   }
3540 
3541   if (PatternType == 0)
3542     return;
3543 
3544   // Generate the diagnostic.
3545   SourceLocation SL = LenArg->getLocStart();
3546   SourceRange SR = LenArg->getSourceRange();
3547   SourceManager &SM  = PP.getSourceManager();
3548 
3549   // If the function is defined as a builtin macro, do not show macro expansion.
3550   if (SM.isMacroArgExpansion(SL)) {
3551     SL = SM.getSpellingLoc(SL);
3552     SR = SourceRange(SM.getSpellingLoc(SR.getBegin()),
3553                      SM.getSpellingLoc(SR.getEnd()));
3554   }
3555 
3556   // Check if the destination is an array (rather than a pointer to an array).
3557   QualType DstTy = DstArg->getType();
3558   bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy,
3559                                                                     Context);
3560   if (!isKnownSizeArray) {
3561     if (PatternType == 1)
3562       Diag(SL, diag::warn_strncat_wrong_size) << SR;
3563     else
3564       Diag(SL, diag::warn_strncat_src_size) << SR;
3565     return;
3566   }
3567 
3568   if (PatternType == 1)
3569     Diag(SL, diag::warn_strncat_large_size) << SR;
3570   else
3571     Diag(SL, diag::warn_strncat_src_size) << SR;
3572 
3573   SmallString<128> sizeString;
3574   llvm::raw_svector_ostream OS(sizeString);
3575   OS << "sizeof(";
3576   DstArg->printPretty(OS, 0, getPrintingPolicy());
3577   OS << ") - ";
3578   OS << "strlen(";
3579   DstArg->printPretty(OS, 0, getPrintingPolicy());
3580   OS << ") - 1";
3581 
3582   Diag(SL, diag::note_strncat_wrong_size)
3583     << FixItHint::CreateReplacement(SR, OS.str());
3584 }
3585 
3586 //===--- CHECK: Return Address of Stack Variable --------------------------===//
3587 
3588 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
3589                      Decl *ParentDecl);
3590 static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars,
3591                       Decl *ParentDecl);
3592 
3593 /// CheckReturnStackAddr - Check if a return statement returns the address
3594 ///   of a stack variable.
3595 void
3596 Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
3597                            SourceLocation ReturnLoc) {
3598 
3599   Expr *stackE = 0;
3600   SmallVector<DeclRefExpr *, 8> refVars;
3601 
3602   // Perform checking for returned stack addresses, local blocks,
3603   // label addresses or references to temporaries.
3604   if (lhsType->isPointerType() ||
3605       (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) {
3606     stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0);
3607   } else if (lhsType->isReferenceType()) {
3608     stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0);
3609   }
3610 
3611   if (stackE == 0)
3612     return; // Nothing suspicious was found.
3613 
3614   SourceLocation diagLoc;
3615   SourceRange diagRange;
3616   if (refVars.empty()) {
3617     diagLoc = stackE->getLocStart();
3618     diagRange = stackE->getSourceRange();
3619   } else {
3620     // We followed through a reference variable. 'stackE' contains the
3621     // problematic expression but we will warn at the return statement pointing
3622     // at the reference variable. We will later display the "trail" of
3623     // reference variables using notes.
3624     diagLoc = refVars[0]->getLocStart();
3625     diagRange = refVars[0]->getSourceRange();
3626   }
3627 
3628   if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
3629     Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
3630                                              : diag::warn_ret_stack_addr)
3631      << DR->getDecl()->getDeclName() << diagRange;
3632   } else if (isa<BlockExpr>(stackE)) { // local block.
3633     Diag(diagLoc, diag::err_ret_local_block) << diagRange;
3634   } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
3635     Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
3636   } else { // local temporary.
3637     Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
3638                                              : diag::warn_ret_local_temp_addr)
3639      << diagRange;
3640   }
3641 
3642   // Display the "trail" of reference variables that we followed until we
3643   // found the problematic expression using notes.
3644   for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
3645     VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
3646     // If this var binds to another reference var, show the range of the next
3647     // var, otherwise the var binds to the problematic expression, in which case
3648     // show the range of the expression.
3649     SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
3650                                   : stackE->getSourceRange();
3651     Diag(VD->getLocation(), diag::note_ref_var_local_bind)
3652       << VD->getDeclName() << range;
3653   }
3654 }
3655 
3656 /// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
3657 ///  check if the expression in a return statement evaluates to an address
3658 ///  to a location on the stack, a local block, an address of a label, or a
3659 ///  reference to local temporary. The recursion is used to traverse the
3660 ///  AST of the return expression, with recursion backtracking when we
3661 ///  encounter a subexpression that (1) clearly does not lead to one of the
3662 ///  above problematic expressions (2) is something we cannot determine leads to
3663 ///  a problematic expression based on such local checking.
3664 ///
3665 ///  Both EvalAddr and EvalVal follow through reference variables to evaluate
3666 ///  the expression that they point to. Such variables are added to the
3667 ///  'refVars' vector so that we know what the reference variable "trail" was.
3668 ///
3669 ///  EvalAddr processes expressions that are pointers that are used as
3670 ///  references (and not L-values).  EvalVal handles all other values.
3671 ///  At the base case of the recursion is a check for the above problematic
3672 ///  expressions.
3673 ///
3674 ///  This implementation handles:
3675 ///
3676 ///   * pointer-to-pointer casts
3677 ///   * implicit conversions from array references to pointers
3678 ///   * taking the address of fields
3679 ///   * arbitrary interplay between "&" and "*" operators
3680 ///   * pointer arithmetic from an address of a stack variable
3681 ///   * taking the address of an array element where the array is on the stack
3682 static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
3683                       Decl *ParentDecl) {
3684   if (E->isTypeDependent())
3685       return NULL;
3686 
3687   // We should only be called for evaluating pointer expressions.
3688   assert((E->getType()->isAnyPointerType() ||
3689           E->getType()->isBlockPointerType() ||
3690           E->getType()->isObjCQualifiedIdType()) &&
3691          "EvalAddr only works on pointers");
3692 
3693   E = E->IgnoreParens();
3694 
3695   // Our "symbolic interpreter" is just a dispatch off the currently
3696   // viewed AST node.  We then recursively traverse the AST by calling
3697   // EvalAddr and EvalVal appropriately.
3698   switch (E->getStmtClass()) {
3699   case Stmt::DeclRefExprClass: {
3700     DeclRefExpr *DR = cast<DeclRefExpr>(E);
3701 
3702     if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
3703       // If this is a reference variable, follow through to the expression that
3704       // it points to.
3705       if (V->hasLocalStorage() &&
3706           V->getType()->isReferenceType() && V->hasInit()) {
3707         // Add the reference variable to the "trail".
3708         refVars.push_back(DR);
3709         return EvalAddr(V->getInit(), refVars, ParentDecl);
3710       }
3711 
3712     return NULL;
3713   }
3714 
3715   case Stmt::UnaryOperatorClass: {
3716     // The only unary operator that make sense to handle here
3717     // is AddrOf.  All others don't make sense as pointers.
3718     UnaryOperator *U = cast<UnaryOperator>(E);
3719 
3720     if (U->getOpcode() == UO_AddrOf)
3721       return EvalVal(U->getSubExpr(), refVars, ParentDecl);
3722     else
3723       return NULL;
3724   }
3725 
3726   case Stmt::BinaryOperatorClass: {
3727     // Handle pointer arithmetic.  All other binary operators are not valid
3728     // in this context.
3729     BinaryOperator *B = cast<BinaryOperator>(E);
3730     BinaryOperatorKind op = B->getOpcode();
3731 
3732     if (op != BO_Add && op != BO_Sub)
3733       return NULL;
3734 
3735     Expr *Base = B->getLHS();
3736 
3737     // Determine which argument is the real pointer base.  It could be
3738     // the RHS argument instead of the LHS.
3739     if (!Base->getType()->isPointerType()) Base = B->getRHS();
3740 
3741     assert (Base->getType()->isPointerType());
3742     return EvalAddr(Base, refVars, ParentDecl);
3743   }
3744 
3745   // For conditional operators we need to see if either the LHS or RHS are
3746   // valid DeclRefExpr*s.  If one of them is valid, we return it.
3747   case Stmt::ConditionalOperatorClass: {
3748     ConditionalOperator *C = cast<ConditionalOperator>(E);
3749 
3750     // Handle the GNU extension for missing LHS.
3751     if (Expr *lhsExpr = C->getLHS()) {
3752     // In C++, we can have a throw-expression, which has 'void' type.
3753       if (!lhsExpr->getType()->isVoidType())
3754         if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl))
3755           return LHS;
3756     }
3757 
3758     // In C++, we can have a throw-expression, which has 'void' type.
3759     if (C->getRHS()->getType()->isVoidType())
3760       return NULL;
3761 
3762     return EvalAddr(C->getRHS(), refVars, ParentDecl);
3763   }
3764 
3765   case Stmt::BlockExprClass:
3766     if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
3767       return E; // local block.
3768     return NULL;
3769 
3770   case Stmt::AddrLabelExprClass:
3771     return E; // address of label.
3772 
3773   case Stmt::ExprWithCleanupsClass:
3774     return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,
3775                     ParentDecl);
3776 
3777   // For casts, we need to handle conversions from arrays to
3778   // pointer values, and pointer-to-pointer conversions.
3779   case Stmt::ImplicitCastExprClass:
3780   case Stmt::CStyleCastExprClass:
3781   case Stmt::CXXFunctionalCastExprClass:
3782   case Stmt::ObjCBridgedCastExprClass:
3783   case Stmt::CXXStaticCastExprClass:
3784   case Stmt::CXXDynamicCastExprClass:
3785   case Stmt::CXXConstCastExprClass:
3786   case Stmt::CXXReinterpretCastExprClass: {
3787     Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
3788     switch (cast<CastExpr>(E)->getCastKind()) {
3789     case CK_BitCast:
3790     case CK_LValueToRValue:
3791     case CK_NoOp:
3792     case CK_BaseToDerived:
3793     case CK_DerivedToBase:
3794     case CK_UncheckedDerivedToBase:
3795     case CK_Dynamic:
3796     case CK_CPointerToObjCPointerCast:
3797     case CK_BlockPointerToObjCPointerCast:
3798     case CK_AnyPointerToBlockPointerCast:
3799       return EvalAddr(SubExpr, refVars, ParentDecl);
3800 
3801     case CK_ArrayToPointerDecay:
3802       return EvalVal(SubExpr, refVars, ParentDecl);
3803 
3804     default:
3805       return 0;
3806     }
3807   }
3808 
3809   case Stmt::MaterializeTemporaryExprClass:
3810     if (Expr *Result = EvalAddr(
3811                          cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
3812                                 refVars, ParentDecl))
3813       return Result;
3814 
3815     return E;
3816 
3817   // Everything else: we simply don't reason about them.
3818   default:
3819     return NULL;
3820   }
3821 }
3822 
3823 
3824 ///  EvalVal - This function is complements EvalAddr in the mutual recursion.
3825 ///   See the comments for EvalAddr for more details.
3826 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
3827                      Decl *ParentDecl) {
3828 do {
3829   // We should only be called for evaluating non-pointer expressions, or
3830   // expressions with a pointer type that are not used as references but instead
3831   // are l-values (e.g., DeclRefExpr with a pointer type).
3832 
3833   // Our "symbolic interpreter" is just a dispatch off the currently
3834   // viewed AST node.  We then recursively traverse the AST by calling
3835   // EvalAddr and EvalVal appropriately.
3836 
3837   E = E->IgnoreParens();
3838   switch (E->getStmtClass()) {
3839   case Stmt::ImplicitCastExprClass: {
3840     ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
3841     if (IE->getValueKind() == VK_LValue) {
3842       E = IE->getSubExpr();
3843       continue;
3844     }
3845     return NULL;
3846   }
3847 
3848   case Stmt::ExprWithCleanupsClass:
3849     return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl);
3850 
3851   case Stmt::DeclRefExprClass: {
3852     // When we hit a DeclRefExpr we are looking at code that refers to a
3853     // variable's name. If it's not a reference variable we check if it has
3854     // local storage within the function, and if so, return the expression.
3855     DeclRefExpr *DR = cast<DeclRefExpr>(E);
3856 
3857     if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) {
3858       // Check if it refers to itself, e.g. "int& i = i;".
3859       if (V == ParentDecl)
3860         return DR;
3861 
3862       if (V->hasLocalStorage()) {
3863         if (!V->getType()->isReferenceType())
3864           return DR;
3865 
3866         // Reference variable, follow through to the expression that
3867         // it points to.
3868         if (V->hasInit()) {
3869           // Add the reference variable to the "trail".
3870           refVars.push_back(DR);
3871           return EvalVal(V->getInit(), refVars, V);
3872         }
3873       }
3874     }
3875 
3876     return NULL;
3877   }
3878 
3879   case Stmt::UnaryOperatorClass: {
3880     // The only unary operator that make sense to handle here
3881     // is Deref.  All others don't resolve to a "name."  This includes
3882     // handling all sorts of rvalues passed to a unary operator.
3883     UnaryOperator *U = cast<UnaryOperator>(E);
3884 
3885     if (U->getOpcode() == UO_Deref)
3886       return EvalAddr(U->getSubExpr(), refVars, ParentDecl);
3887 
3888     return NULL;
3889   }
3890 
3891   case Stmt::ArraySubscriptExprClass: {
3892     // Array subscripts are potential references to data on the stack.  We
3893     // retrieve the DeclRefExpr* for the array variable if it indeed
3894     // has local storage.
3895     return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl);
3896   }
3897 
3898   case Stmt::ConditionalOperatorClass: {
3899     // For conditional operators we need to see if either the LHS or RHS are
3900     // non-NULL Expr's.  If one is non-NULL, we return it.
3901     ConditionalOperator *C = cast<ConditionalOperator>(E);
3902 
3903     // Handle the GNU extension for missing LHS.
3904     if (Expr *lhsExpr = C->getLHS())
3905       if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl))
3906         return LHS;
3907 
3908     return EvalVal(C->getRHS(), refVars, ParentDecl);
3909   }
3910 
3911   // Accesses to members are potential references to data on the stack.
3912   case Stmt::MemberExprClass: {
3913     MemberExpr *M = cast<MemberExpr>(E);
3914 
3915     // Check for indirect access.  We only want direct field accesses.
3916     if (M->isArrow())
3917       return NULL;
3918 
3919     // Check whether the member type is itself a reference, in which case
3920     // we're not going to refer to the member, but to what the member refers to.
3921     if (M->getMemberDecl()->getType()->isReferenceType())
3922       return NULL;
3923 
3924     return EvalVal(M->getBase(), refVars, ParentDecl);
3925   }
3926 
3927   case Stmt::MaterializeTemporaryExprClass:
3928     if (Expr *Result = EvalVal(
3929                           cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
3930                                refVars, ParentDecl))
3931       return Result;
3932 
3933     return E;
3934 
3935   default:
3936     // Check that we don't return or take the address of a reference to a
3937     // temporary. This is only useful in C++.
3938     if (!E->isTypeDependent() && E->isRValue())
3939       return E;
3940 
3941     // Everything else: we simply don't reason about them.
3942     return NULL;
3943   }
3944 } while (true);
3945 }
3946 
3947 //===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
3948 
3949 /// Check for comparisons of floating point operands using != and ==.
3950 /// Issue a warning if these are no self-comparisons, as they are not likely
3951 /// to do what the programmer intended.
3952 void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) {
3953   Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts();
3954   Expr* RightExprSansParen = RHS->IgnoreParenImpCasts();
3955 
3956   // Special case: check for x == x (which is OK).
3957   // Do not emit warnings for such cases.
3958   if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
3959     if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
3960       if (DRL->getDecl() == DRR->getDecl())
3961         return;
3962 
3963 
3964   // Special case: check for comparisons against literals that can be exactly
3965   //  represented by APFloat.  In such cases, do not emit a warning.  This
3966   //  is a heuristic: often comparison against such literals are used to
3967   //  detect if a value in a variable has not changed.  This clearly can
3968   //  lead to false negatives.
3969   if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
3970     if (FLL->isExact())
3971       return;
3972   } else
3973     if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen))
3974       if (FLR->isExact())
3975         return;
3976 
3977   // Check for comparisons with builtin types.
3978   if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
3979     if (CL->isBuiltinCall())
3980       return;
3981 
3982   if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
3983     if (CR->isBuiltinCall())
3984       return;
3985 
3986   // Emit the diagnostic.
3987   Diag(Loc, diag::warn_floatingpoint_eq)
3988     << LHS->getSourceRange() << RHS->getSourceRange();
3989 }
3990 
3991 //===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
3992 //===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
3993 
3994 namespace {
3995 
3996 /// Structure recording the 'active' range of an integer-valued
3997 /// expression.
3998 struct IntRange {
3999   /// The number of bits active in the int.
4000   unsigned Width;
4001 
4002   /// True if the int is known not to have negative values.
4003   bool NonNegative;
4004 
4005   IntRange(unsigned Width, bool NonNegative)
4006     : Width(Width), NonNegative(NonNegative)
4007   {}
4008 
4009   /// Returns the range of the bool type.
4010   static IntRange forBoolType() {
4011     return IntRange(1, true);
4012   }
4013 
4014   /// Returns the range of an opaque value of the given integral type.
4015   static IntRange forValueOfType(ASTContext &C, QualType T) {
4016     return forValueOfCanonicalType(C,
4017                           T->getCanonicalTypeInternal().getTypePtr());
4018   }
4019 
4020   /// Returns the range of an opaque value of a canonical integral type.
4021   static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
4022     assert(T->isCanonicalUnqualified());
4023 
4024     if (const VectorType *VT = dyn_cast<VectorType>(T))
4025       T = VT->getElementType().getTypePtr();
4026     if (const ComplexType *CT = dyn_cast<ComplexType>(T))
4027       T = CT->getElementType().getTypePtr();
4028 
4029     // For enum types, use the known bit width of the enumerators.
4030     if (const EnumType *ET = dyn_cast<EnumType>(T)) {
4031       EnumDecl *Enum = ET->getDecl();
4032       if (!Enum->isCompleteDefinition())
4033         return IntRange(C.getIntWidth(QualType(T, 0)), false);
4034 
4035       unsigned NumPositive = Enum->getNumPositiveBits();
4036       unsigned NumNegative = Enum->getNumNegativeBits();
4037 
4038       if (NumNegative == 0)
4039         return IntRange(NumPositive, true/*NonNegative*/);
4040       else
4041         return IntRange(std::max(NumPositive + 1, NumNegative),
4042                         false/*NonNegative*/);
4043     }
4044 
4045     const BuiltinType *BT = cast<BuiltinType>(T);
4046     assert(BT->isInteger());
4047 
4048     return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
4049   }
4050 
4051   /// Returns the "target" range of a canonical integral type, i.e.
4052   /// the range of values expressible in the type.
4053   ///
4054   /// This matches forValueOfCanonicalType except that enums have the
4055   /// full range of their type, not the range of their enumerators.
4056   static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
4057     assert(T->isCanonicalUnqualified());
4058 
4059     if (const VectorType *VT = dyn_cast<VectorType>(T))
4060       T = VT->getElementType().getTypePtr();
4061     if (const ComplexType *CT = dyn_cast<ComplexType>(T))
4062       T = CT->getElementType().getTypePtr();
4063     if (const EnumType *ET = dyn_cast<EnumType>(T))
4064       T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr();
4065 
4066     const BuiltinType *BT = cast<BuiltinType>(T);
4067     assert(BT->isInteger());
4068 
4069     return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
4070   }
4071 
4072   /// Returns the supremum of two ranges: i.e. their conservative merge.
4073   static IntRange join(IntRange L, IntRange R) {
4074     return IntRange(std::max(L.Width, R.Width),
4075                     L.NonNegative && R.NonNegative);
4076   }
4077 
4078   /// Returns the infinum of two ranges: i.e. their aggressive merge.
4079   static IntRange meet(IntRange L, IntRange R) {
4080     return IntRange(std::min(L.Width, R.Width),
4081                     L.NonNegative || R.NonNegative);
4082   }
4083 };
4084 
4085 static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value,
4086                               unsigned MaxWidth) {
4087   if (value.isSigned() && value.isNegative())
4088     return IntRange(value.getMinSignedBits(), false);
4089 
4090   if (value.getBitWidth() > MaxWidth)
4091     value = value.trunc(MaxWidth);
4092 
4093   // isNonNegative() just checks the sign bit without considering
4094   // signedness.
4095   return IntRange(value.getActiveBits(), true);
4096 }
4097 
4098 static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
4099                               unsigned MaxWidth) {
4100   if (result.isInt())
4101     return GetValueRange(C, result.getInt(), MaxWidth);
4102 
4103   if (result.isVector()) {
4104     IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
4105     for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
4106       IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
4107       R = IntRange::join(R, El);
4108     }
4109     return R;
4110   }
4111 
4112   if (result.isComplexInt()) {
4113     IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
4114     IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
4115     return IntRange::join(R, I);
4116   }
4117 
4118   // This can happen with lossless casts to intptr_t of "based" lvalues.
4119   // Assume it might use arbitrary bits.
4120   // FIXME: The only reason we need to pass the type in here is to get
4121   // the sign right on this one case.  It would be nice if APValue
4122   // preserved this.
4123   assert(result.isLValue() || result.isAddrLabelDiff());
4124   return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType());
4125 }
4126 
4127 /// Pseudo-evaluate the given integer expression, estimating the
4128 /// range of values it might take.
4129 ///
4130 /// \param MaxWidth - the width to which the value will be truncated
4131 static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
4132   E = E->IgnoreParens();
4133 
4134   // Try a full evaluation first.
4135   Expr::EvalResult result;
4136   if (E->EvaluateAsRValue(result, C))
4137     return GetValueRange(C, result.Val, E->getType(), MaxWidth);
4138 
4139   // I think we only want to look through implicit casts here; if the
4140   // user has an explicit widening cast, we should treat the value as
4141   // being of the new, wider type.
4142   if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
4143     if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue)
4144       return GetExprRange(C, CE->getSubExpr(), MaxWidth);
4145 
4146     IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
4147 
4148     bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
4149 
4150     // Assume that non-integer casts can span the full range of the type.
4151     if (!isIntegerCast)
4152       return OutputTypeRange;
4153 
4154     IntRange SubRange
4155       = GetExprRange(C, CE->getSubExpr(),
4156                      std::min(MaxWidth, OutputTypeRange.Width));
4157 
4158     // Bail out if the subexpr's range is as wide as the cast type.
4159     if (SubRange.Width >= OutputTypeRange.Width)
4160       return OutputTypeRange;
4161 
4162     // Otherwise, we take the smaller width, and we're non-negative if
4163     // either the output type or the subexpr is.
4164     return IntRange(SubRange.Width,
4165                     SubRange.NonNegative || OutputTypeRange.NonNegative);
4166   }
4167 
4168   if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
4169     // If we can fold the condition, just take that operand.
4170     bool CondResult;
4171     if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
4172       return GetExprRange(C, CondResult ? CO->getTrueExpr()
4173                                         : CO->getFalseExpr(),
4174                           MaxWidth);
4175 
4176     // Otherwise, conservatively merge.
4177     IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
4178     IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
4179     return IntRange::join(L, R);
4180   }
4181 
4182   if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
4183     switch (BO->getOpcode()) {
4184 
4185     // Boolean-valued operations are single-bit and positive.
4186     case BO_LAnd:
4187     case BO_LOr:
4188     case BO_LT:
4189     case BO_GT:
4190     case BO_LE:
4191     case BO_GE:
4192     case BO_EQ:
4193     case BO_NE:
4194       return IntRange::forBoolType();
4195 
4196     // The type of the assignments is the type of the LHS, so the RHS
4197     // is not necessarily the same type.
4198     case BO_MulAssign:
4199     case BO_DivAssign:
4200     case BO_RemAssign:
4201     case BO_AddAssign:
4202     case BO_SubAssign:
4203     case BO_XorAssign:
4204     case BO_OrAssign:
4205       // TODO: bitfields?
4206       return IntRange::forValueOfType(C, E->getType());
4207 
4208     // Simple assignments just pass through the RHS, which will have
4209     // been coerced to the LHS type.
4210     case BO_Assign:
4211       // TODO: bitfields?
4212       return GetExprRange(C, BO->getRHS(), MaxWidth);
4213 
4214     // Operations with opaque sources are black-listed.
4215     case BO_PtrMemD:
4216     case BO_PtrMemI:
4217       return IntRange::forValueOfType(C, E->getType());
4218 
4219     // Bitwise-and uses the *infinum* of the two source ranges.
4220     case BO_And:
4221     case BO_AndAssign:
4222       return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
4223                             GetExprRange(C, BO->getRHS(), MaxWidth));
4224 
4225     // Left shift gets black-listed based on a judgement call.
4226     case BO_Shl:
4227       // ...except that we want to treat '1 << (blah)' as logically
4228       // positive.  It's an important idiom.
4229       if (IntegerLiteral *I
4230             = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
4231         if (I->getValue() == 1) {
4232           IntRange R = IntRange::forValueOfType(C, E->getType());
4233           return IntRange(R.Width, /*NonNegative*/ true);
4234         }
4235       }
4236       // fallthrough
4237 
4238     case BO_ShlAssign:
4239       return IntRange::forValueOfType(C, E->getType());
4240 
4241     // Right shift by a constant can narrow its left argument.
4242     case BO_Shr:
4243     case BO_ShrAssign: {
4244       IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
4245 
4246       // If the shift amount is a positive constant, drop the width by
4247       // that much.
4248       llvm::APSInt shift;
4249       if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
4250           shift.isNonNegative()) {
4251         unsigned zext = shift.getZExtValue();
4252         if (zext >= L.Width)
4253           L.Width = (L.NonNegative ? 0 : 1);
4254         else
4255           L.Width -= zext;
4256       }
4257 
4258       return L;
4259     }
4260 
4261     // Comma acts as its right operand.
4262     case BO_Comma:
4263       return GetExprRange(C, BO->getRHS(), MaxWidth);
4264 
4265     // Black-list pointer subtractions.
4266     case BO_Sub:
4267       if (BO->getLHS()->getType()->isPointerType())
4268         return IntRange::forValueOfType(C, E->getType());
4269       break;
4270 
4271     // The width of a division result is mostly determined by the size
4272     // of the LHS.
4273     case BO_Div: {
4274       // Don't 'pre-truncate' the operands.
4275       unsigned opWidth = C.getIntWidth(E->getType());
4276       IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
4277 
4278       // If the divisor is constant, use that.
4279       llvm::APSInt divisor;
4280       if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) {
4281         unsigned log2 = divisor.logBase2(); // floor(log_2(divisor))
4282         if (log2 >= L.Width)
4283           L.Width = (L.NonNegative ? 0 : 1);
4284         else
4285           L.Width = std::min(L.Width - log2, MaxWidth);
4286         return L;
4287       }
4288 
4289       // Otherwise, just use the LHS's width.
4290       IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
4291       return IntRange(L.Width, L.NonNegative && R.NonNegative);
4292     }
4293 
4294     // The result of a remainder can't be larger than the result of
4295     // either side.
4296     case BO_Rem: {
4297       // Don't 'pre-truncate' the operands.
4298       unsigned opWidth = C.getIntWidth(E->getType());
4299       IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
4300       IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
4301 
4302       IntRange meet = IntRange::meet(L, R);
4303       meet.Width = std::min(meet.Width, MaxWidth);
4304       return meet;
4305     }
4306 
4307     // The default behavior is okay for these.
4308     case BO_Mul:
4309     case BO_Add:
4310     case BO_Xor:
4311     case BO_Or:
4312       break;
4313     }
4314 
4315     // The default case is to treat the operation as if it were closed
4316     // on the narrowest type that encompasses both operands.
4317     IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
4318     IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
4319     return IntRange::join(L, R);
4320   }
4321 
4322   if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
4323     switch (UO->getOpcode()) {
4324     // Boolean-valued operations are white-listed.
4325     case UO_LNot:
4326       return IntRange::forBoolType();
4327 
4328     // Operations with opaque sources are black-listed.
4329     case UO_Deref:
4330     case UO_AddrOf: // should be impossible
4331       return IntRange::forValueOfType(C, E->getType());
4332 
4333     default:
4334       return GetExprRange(C, UO->getSubExpr(), MaxWidth);
4335     }
4336   }
4337 
4338   if (dyn_cast<OffsetOfExpr>(E)) {
4339     IntRange::forValueOfType(C, E->getType());
4340   }
4341 
4342   if (FieldDecl *BitField = E->getSourceBitField())
4343     return IntRange(BitField->getBitWidthValue(C),
4344                     BitField->getType()->isUnsignedIntegerOrEnumerationType());
4345 
4346   return IntRange::forValueOfType(C, E->getType());
4347 }
4348 
4349 static IntRange GetExprRange(ASTContext &C, Expr *E) {
4350   return GetExprRange(C, E, C.getIntWidth(E->getType()));
4351 }
4352 
4353 /// Checks whether the given value, which currently has the given
4354 /// source semantics, has the same value when coerced through the
4355 /// target semantics.
4356 static bool IsSameFloatAfterCast(const llvm::APFloat &value,
4357                                  const llvm::fltSemantics &Src,
4358                                  const llvm::fltSemantics &Tgt) {
4359   llvm::APFloat truncated = value;
4360 
4361   bool ignored;
4362   truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
4363   truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
4364 
4365   return truncated.bitwiseIsEqual(value);
4366 }
4367 
4368 /// Checks whether the given value, which currently has the given
4369 /// source semantics, has the same value when coerced through the
4370 /// target semantics.
4371 ///
4372 /// The value might be a vector of floats (or a complex number).
4373 static bool IsSameFloatAfterCast(const APValue &value,
4374                                  const llvm::fltSemantics &Src,
4375                                  const llvm::fltSemantics &Tgt) {
4376   if (value.isFloat())
4377     return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
4378 
4379   if (value.isVector()) {
4380     for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
4381       if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
4382         return false;
4383     return true;
4384   }
4385 
4386   assert(value.isComplexFloat());
4387   return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
4388           IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
4389 }
4390 
4391 static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
4392 
4393 static bool IsZero(Sema &S, Expr *E) {
4394   // Suppress cases where we are comparing against an enum constant.
4395   if (const DeclRefExpr *DR =
4396       dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
4397     if (isa<EnumConstantDecl>(DR->getDecl()))
4398       return false;
4399 
4400   // Suppress cases where the '0' value is expanded from a macro.
4401   if (E->getLocStart().isMacroID())
4402     return false;
4403 
4404   llvm::APSInt Value;
4405   return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
4406 }
4407 
4408 static bool HasEnumType(Expr *E) {
4409   // Strip off implicit integral promotions.
4410   while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
4411     if (ICE->getCastKind() != CK_IntegralCast &&
4412         ICE->getCastKind() != CK_NoOp)
4413       break;
4414     E = ICE->getSubExpr();
4415   }
4416 
4417   return E->getType()->isEnumeralType();
4418 }
4419 
4420 static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
4421   BinaryOperatorKind op = E->getOpcode();
4422   if (E->isValueDependent())
4423     return;
4424 
4425   if (op == BO_LT && IsZero(S, E->getRHS())) {
4426     S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
4427       << "< 0" << "false" << HasEnumType(E->getLHS())
4428       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4429   } else if (op == BO_GE && IsZero(S, E->getRHS())) {
4430     S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
4431       << ">= 0" << "true" << HasEnumType(E->getLHS())
4432       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4433   } else if (op == BO_GT && IsZero(S, E->getLHS())) {
4434     S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
4435       << "0 >" << "false" << HasEnumType(E->getRHS())
4436       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4437   } else if (op == BO_LE && IsZero(S, E->getLHS())) {
4438     S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
4439       << "0 <=" << "true" << HasEnumType(E->getRHS())
4440       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4441   }
4442 }
4443 
4444 static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E,
4445                                          Expr *Constant, Expr *Other,
4446                                          llvm::APSInt Value,
4447                                          bool RhsConstant) {
4448   // 0 values are handled later by CheckTrivialUnsignedComparison().
4449   if (Value == 0)
4450     return;
4451 
4452   BinaryOperatorKind op = E->getOpcode();
4453   QualType OtherT = Other->getType();
4454   QualType ConstantT = Constant->getType();
4455   QualType CommonT = E->getLHS()->getType();
4456   if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT))
4457     return;
4458   assert((OtherT->isIntegerType() && ConstantT->isIntegerType())
4459          && "comparison with non-integer type");
4460 
4461   bool ConstantSigned = ConstantT->isSignedIntegerType();
4462   bool CommonSigned = CommonT->isSignedIntegerType();
4463 
4464   bool EqualityOnly = false;
4465 
4466   // TODO: Investigate using GetExprRange() to get tighter bounds on
4467   // on the bit ranges.
4468   IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT);
4469   unsigned OtherWidth = OtherRange.Width;
4470 
4471   if (CommonSigned) {
4472     // The common type is signed, therefore no signed to unsigned conversion.
4473     if (!OtherRange.NonNegative) {
4474       // Check that the constant is representable in type OtherT.
4475       if (ConstantSigned) {
4476         if (OtherWidth >= Value.getMinSignedBits())
4477           return;
4478       } else { // !ConstantSigned
4479         if (OtherWidth >= Value.getActiveBits() + 1)
4480           return;
4481       }
4482     } else { // !OtherSigned
4483       // Check that the constant is representable in type OtherT.
4484       // Negative values are out of range.
4485       if (ConstantSigned) {
4486         if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits())
4487           return;
4488       } else { // !ConstantSigned
4489         if (OtherWidth >= Value.getActiveBits())
4490           return;
4491       }
4492     }
4493   } else {  // !CommonSigned
4494     if (OtherRange.NonNegative) {
4495       if (OtherWidth >= Value.getActiveBits())
4496         return;
4497     } else if (!OtherRange.NonNegative && !ConstantSigned) {
4498       // Check to see if the constant is representable in OtherT.
4499       if (OtherWidth > Value.getActiveBits())
4500         return;
4501       // Check to see if the constant is equivalent to a negative value
4502       // cast to CommonT.
4503       if (S.Context.getIntWidth(ConstantT) == S.Context.getIntWidth(CommonT) &&
4504           Value.isNegative() && Value.getMinSignedBits() <= OtherWidth)
4505         return;
4506       // The constant value rests between values that OtherT can represent after
4507       // conversion.  Relational comparison still works, but equality
4508       // comparisons will be tautological.
4509       EqualityOnly = true;
4510     } else { // OtherSigned && ConstantSigned
4511       assert(0 && "Two signed types converted to unsigned types.");
4512     }
4513   }
4514 
4515   bool PositiveConstant = !ConstantSigned || Value.isNonNegative();
4516 
4517   bool IsTrue = true;
4518   if (op == BO_EQ || op == BO_NE) {
4519     IsTrue = op == BO_NE;
4520   } else if (EqualityOnly) {
4521     return;
4522   } else if (RhsConstant) {
4523     if (op == BO_GT || op == BO_GE)
4524       IsTrue = !PositiveConstant;
4525     else // op == BO_LT || op == BO_LE
4526       IsTrue = PositiveConstant;
4527   } else {
4528     if (op == BO_LT || op == BO_LE)
4529       IsTrue = !PositiveConstant;
4530     else // op == BO_GT || op == BO_GE
4531       IsTrue = PositiveConstant;
4532   }
4533 
4534   // If this is a comparison to an enum constant, include that
4535   // constant in the diagnostic.
4536   const EnumConstantDecl *ED = 0;
4537   if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Constant))
4538     ED = dyn_cast<EnumConstantDecl>(DR->getDecl());
4539 
4540   SmallString<64> PrettySourceValue;
4541   llvm::raw_svector_ostream OS(PrettySourceValue);
4542   if (ED)
4543     OS << '\'' << *ED << "' (" << Value << ")";
4544   else
4545     OS << Value;
4546 
4547   S.Diag(E->getOperatorLoc(), diag::warn_out_of_range_compare)
4548       << OS.str() << OtherT << IsTrue
4549       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4550 }
4551 
4552 /// Analyze the operands of the given comparison.  Implements the
4553 /// fallback case from AnalyzeComparison.
4554 static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
4555   AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
4556   AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
4557 }
4558 
4559 /// \brief Implements -Wsign-compare.
4560 ///
4561 /// \param E the binary operator to check for warnings
4562 static void AnalyzeComparison(Sema &S, BinaryOperator *E) {
4563   // The type the comparison is being performed in.
4564   QualType T = E->getLHS()->getType();
4565   assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
4566          && "comparison with mismatched types");
4567   if (E->isValueDependent())
4568     return AnalyzeImpConvsInComparison(S, E);
4569 
4570   Expr *LHS = E->getLHS()->IgnoreParenImpCasts();
4571   Expr *RHS = E->getRHS()->IgnoreParenImpCasts();
4572 
4573   bool IsComparisonConstant = false;
4574 
4575   // Check whether an integer constant comparison results in a value
4576   // of 'true' or 'false'.
4577   if (T->isIntegralType(S.Context)) {
4578     llvm::APSInt RHSValue;
4579     bool IsRHSIntegralLiteral =
4580       RHS->isIntegerConstantExpr(RHSValue, S.Context);
4581     llvm::APSInt LHSValue;
4582     bool IsLHSIntegralLiteral =
4583       LHS->isIntegerConstantExpr(LHSValue, S.Context);
4584     if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral)
4585         DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true);
4586     else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral)
4587       DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false);
4588     else
4589       IsComparisonConstant =
4590         (IsRHSIntegralLiteral && IsLHSIntegralLiteral);
4591   } else if (!T->hasUnsignedIntegerRepresentation())
4592       IsComparisonConstant = E->isIntegerConstantExpr(S.Context);
4593 
4594   // We don't do anything special if this isn't an unsigned integral
4595   // comparison:  we're only interested in integral comparisons, and
4596   // signed comparisons only happen in cases we don't care to warn about.
4597   //
4598   // We also don't care about value-dependent expressions or expressions
4599   // whose result is a constant.
4600   if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant)
4601     return AnalyzeImpConvsInComparison(S, E);
4602 
4603   // Check to see if one of the (unmodified) operands is of different
4604   // signedness.
4605   Expr *signedOperand, *unsignedOperand;
4606   if (LHS->getType()->hasSignedIntegerRepresentation()) {
4607     assert(!RHS->getType()->hasSignedIntegerRepresentation() &&
4608            "unsigned comparison between two signed integer expressions?");
4609     signedOperand = LHS;
4610     unsignedOperand = RHS;
4611   } else if (RHS->getType()->hasSignedIntegerRepresentation()) {
4612     signedOperand = RHS;
4613     unsignedOperand = LHS;
4614   } else {
4615     CheckTrivialUnsignedComparison(S, E);
4616     return AnalyzeImpConvsInComparison(S, E);
4617   }
4618 
4619   // Otherwise, calculate the effective range of the signed operand.
4620   IntRange signedRange = GetExprRange(S.Context, signedOperand);
4621 
4622   // Go ahead and analyze implicit conversions in the operands.  Note
4623   // that we skip the implicit conversions on both sides.
4624   AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc());
4625   AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc());
4626 
4627   // If the signed range is non-negative, -Wsign-compare won't fire,
4628   // but we should still check for comparisons which are always true
4629   // or false.
4630   if (signedRange.NonNegative)
4631     return CheckTrivialUnsignedComparison(S, E);
4632 
4633   // For (in)equality comparisons, if the unsigned operand is a
4634   // constant which cannot collide with a overflowed signed operand,
4635   // then reinterpreting the signed operand as unsigned will not
4636   // change the result of the comparison.
4637   if (E->isEqualityOp()) {
4638     unsigned comparisonWidth = S.Context.getIntWidth(T);
4639     IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
4640 
4641     // We should never be unable to prove that the unsigned operand is
4642     // non-negative.
4643     assert(unsignedRange.NonNegative && "unsigned range includes negative?");
4644 
4645     if (unsignedRange.Width < comparisonWidth)
4646       return;
4647   }
4648 
4649   S.DiagRuntimeBehavior(E->getOperatorLoc(), E,
4650     S.PDiag(diag::warn_mixed_sign_comparison)
4651       << LHS->getType() << RHS->getType()
4652       << LHS->getSourceRange() << RHS->getSourceRange());
4653 }
4654 
4655 /// Analyzes an attempt to assign the given value to a bitfield.
4656 ///
4657 /// Returns true if there was something fishy about the attempt.
4658 static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
4659                                       SourceLocation InitLoc) {
4660   assert(Bitfield->isBitField());
4661   if (Bitfield->isInvalidDecl())
4662     return false;
4663 
4664   // White-list bool bitfields.
4665   if (Bitfield->getType()->isBooleanType())
4666     return false;
4667 
4668   // Ignore value- or type-dependent expressions.
4669   if (Bitfield->getBitWidth()->isValueDependent() ||
4670       Bitfield->getBitWidth()->isTypeDependent() ||
4671       Init->isValueDependent() ||
4672       Init->isTypeDependent())
4673     return false;
4674 
4675   Expr *OriginalInit = Init->IgnoreParenImpCasts();
4676 
4677   llvm::APSInt Value;
4678   if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects))
4679     return false;
4680 
4681   unsigned OriginalWidth = Value.getBitWidth();
4682   unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context);
4683 
4684   if (OriginalWidth <= FieldWidth)
4685     return false;
4686 
4687   // Compute the value which the bitfield will contain.
4688   llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
4689   TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType());
4690 
4691   // Check whether the stored value is equal to the original value.
4692   TruncatedValue = TruncatedValue.extend(OriginalWidth);
4693   if (llvm::APSInt::isSameValue(Value, TruncatedValue))
4694     return false;
4695 
4696   // Special-case bitfields of width 1: booleans are naturally 0/1, and
4697   // therefore don't strictly fit into a signed bitfield of width 1.
4698   if (FieldWidth == 1 && Value == 1)
4699     return false;
4700 
4701   std::string PrettyValue = Value.toString(10);
4702   std::string PrettyTrunc = TruncatedValue.toString(10);
4703 
4704   S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
4705     << PrettyValue << PrettyTrunc << OriginalInit->getType()
4706     << Init->getSourceRange();
4707 
4708   return true;
4709 }
4710 
4711 /// Analyze the given simple or compound assignment for warning-worthy
4712 /// operations.
4713 static void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
4714   // Just recurse on the LHS.
4715   AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
4716 
4717   // We want to recurse on the RHS as normal unless we're assigning to
4718   // a bitfield.
4719   if (FieldDecl *Bitfield = E->getLHS()->getSourceBitField()) {
4720     if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
4721                                   E->getOperatorLoc())) {
4722       // Recurse, ignoring any implicit conversions on the RHS.
4723       return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
4724                                         E->getOperatorLoc());
4725     }
4726   }
4727 
4728   AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
4729 }
4730 
4731 /// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
4732 static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
4733                             SourceLocation CContext, unsigned diag,
4734                             bool pruneControlFlow = false) {
4735   if (pruneControlFlow) {
4736     S.DiagRuntimeBehavior(E->getExprLoc(), E,
4737                           S.PDiag(diag)
4738                             << SourceType << T << E->getSourceRange()
4739                             << SourceRange(CContext));
4740     return;
4741   }
4742   S.Diag(E->getExprLoc(), diag)
4743     << SourceType << T << E->getSourceRange() << SourceRange(CContext);
4744 }
4745 
4746 /// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
4747 static void DiagnoseImpCast(Sema &S, Expr *E, QualType T,
4748                             SourceLocation CContext, unsigned diag,
4749                             bool pruneControlFlow = false) {
4750   DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow);
4751 }
4752 
4753 /// Diagnose an implicit cast from a literal expression. Does not warn when the
4754 /// cast wouldn't lose information.
4755 void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
4756                                     SourceLocation CContext) {
4757   // Try to convert the literal exactly to an integer. If we can, don't warn.
4758   bool isExact = false;
4759   const llvm::APFloat &Value = FL->getValue();
4760   llvm::APSInt IntegerValue(S.Context.getIntWidth(T),
4761                             T->hasUnsignedIntegerRepresentation());
4762   if (Value.convertToInteger(IntegerValue,
4763                              llvm::APFloat::rmTowardZero, &isExact)
4764       == llvm::APFloat::opOK && isExact)
4765     return;
4766 
4767   SmallString<16> PrettySourceValue;
4768   Value.toString(PrettySourceValue);
4769   SmallString<16> PrettyTargetValue;
4770   if (T->isSpecificBuiltinType(BuiltinType::Bool))
4771     PrettyTargetValue = IntegerValue == 0 ? "false" : "true";
4772   else
4773     IntegerValue.toString(PrettyTargetValue);
4774 
4775   S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
4776     << FL->getType() << T.getUnqualifiedType() << PrettySourceValue
4777     << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext);
4778 }
4779 
4780 std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
4781   if (!Range.Width) return "0";
4782 
4783   llvm::APSInt ValueInRange = Value;
4784   ValueInRange.setIsSigned(!Range.NonNegative);
4785   ValueInRange = ValueInRange.trunc(Range.Width);
4786   return ValueInRange.toString(10);
4787 }
4788 
4789 static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) {
4790   if (!isa<ImplicitCastExpr>(Ex))
4791     return false;
4792 
4793   Expr *InnerE = Ex->IgnoreParenImpCasts();
4794   const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr();
4795   const Type *Source =
4796     S.Context.getCanonicalType(InnerE->getType()).getTypePtr();
4797   if (Target->isDependentType())
4798     return false;
4799 
4800   const BuiltinType *FloatCandidateBT =
4801     dyn_cast<BuiltinType>(ToBool ? Source : Target);
4802   const Type *BoolCandidateType = ToBool ? Target : Source;
4803 
4804   return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) &&
4805           FloatCandidateBT && (FloatCandidateBT->isFloatingPoint()));
4806 }
4807 
4808 void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall,
4809                                       SourceLocation CC) {
4810   unsigned NumArgs = TheCall->getNumArgs();
4811   for (unsigned i = 0; i < NumArgs; ++i) {
4812     Expr *CurrA = TheCall->getArg(i);
4813     if (!IsImplicitBoolFloatConversion(S, CurrA, true))
4814       continue;
4815 
4816     bool IsSwapped = ((i > 0) &&
4817         IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false));
4818     IsSwapped |= ((i < (NumArgs - 1)) &&
4819         IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false));
4820     if (IsSwapped) {
4821       // Warn on this floating-point to bool conversion.
4822       DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(),
4823                       CurrA->getType(), CC,
4824                       diag::warn_impcast_floating_point_to_bool);
4825     }
4826   }
4827 }
4828 
4829 void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
4830                              SourceLocation CC, bool *ICContext = 0) {
4831   if (E->isTypeDependent() || E->isValueDependent()) return;
4832 
4833   const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
4834   const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
4835   if (Source == Target) return;
4836   if (Target->isDependentType()) return;
4837 
4838   // If the conversion context location is invalid don't complain. We also
4839   // don't want to emit a warning if the issue occurs from the expansion of
4840   // a system macro. The problem is that 'getSpellingLoc()' is slow, so we
4841   // delay this check as long as possible. Once we detect we are in that
4842   // scenario, we just return.
4843   if (CC.isInvalid())
4844     return;
4845 
4846   // Diagnose implicit casts to bool.
4847   if (Target->isSpecificBuiltinType(BuiltinType::Bool)) {
4848     if (isa<StringLiteral>(E))
4849       // Warn on string literal to bool.  Checks for string literals in logical
4850       // expressions, for instances, assert(0 && "error here"), is prevented
4851       // by a check in AnalyzeImplicitConversions().
4852       return DiagnoseImpCast(S, E, T, CC,
4853                              diag::warn_impcast_string_literal_to_bool);
4854     if (Source->isFunctionType()) {
4855       // Warn on function to bool. Checks free functions and static member
4856       // functions. Weakly imported functions are excluded from the check,
4857       // since it's common to test their value to check whether the linker
4858       // found a definition for them.
4859       ValueDecl *D = 0;
4860       if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) {
4861         D = R->getDecl();
4862       } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) {
4863         D = M->getMemberDecl();
4864       }
4865 
4866       if (D && !D->isWeak()) {
4867         if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) {
4868           S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool)
4869             << F << E->getSourceRange() << SourceRange(CC);
4870           S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence)
4871             << FixItHint::CreateInsertion(E->getExprLoc(), "&");
4872           QualType ReturnType;
4873           UnresolvedSet<4> NonTemplateOverloads;
4874           S.isExprCallable(*E, ReturnType, NonTemplateOverloads);
4875           if (!ReturnType.isNull()
4876               && ReturnType->isSpecificBuiltinType(BuiltinType::Bool))
4877             S.Diag(E->getExprLoc(), diag::note_function_to_bool_call)
4878               << FixItHint::CreateInsertion(
4879                  S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()");
4880           return;
4881         }
4882       }
4883     }
4884   }
4885 
4886   // Strip vector types.
4887   if (isa<VectorType>(Source)) {
4888     if (!isa<VectorType>(Target)) {
4889       if (S.SourceMgr.isInSystemMacro(CC))
4890         return;
4891       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
4892     }
4893 
4894     // If the vector cast is cast between two vectors of the same size, it is
4895     // a bitcast, not a conversion.
4896     if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target))
4897       return;
4898 
4899     Source = cast<VectorType>(Source)->getElementType().getTypePtr();
4900     Target = cast<VectorType>(Target)->getElementType().getTypePtr();
4901   }
4902 
4903   // Strip complex types.
4904   if (isa<ComplexType>(Source)) {
4905     if (!isa<ComplexType>(Target)) {
4906       if (S.SourceMgr.isInSystemMacro(CC))
4907         return;
4908 
4909       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
4910     }
4911 
4912     Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
4913     Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
4914   }
4915 
4916   const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
4917   const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
4918 
4919   // If the source is floating point...
4920   if (SourceBT && SourceBT->isFloatingPoint()) {
4921     // ...and the target is floating point...
4922     if (TargetBT && TargetBT->isFloatingPoint()) {
4923       // ...then warn if we're dropping FP rank.
4924 
4925       // Builtin FP kinds are ordered by increasing FP rank.
4926       if (SourceBT->getKind() > TargetBT->getKind()) {
4927         // Don't warn about float constants that are precisely
4928         // representable in the target type.
4929         Expr::EvalResult result;
4930         if (E->EvaluateAsRValue(result, S.Context)) {
4931           // Value might be a float, a float vector, or a float complex.
4932           if (IsSameFloatAfterCast(result.Val,
4933                    S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
4934                    S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
4935             return;
4936         }
4937 
4938         if (S.SourceMgr.isInSystemMacro(CC))
4939           return;
4940 
4941         DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
4942       }
4943       return;
4944     }
4945 
4946     // If the target is integral, always warn.
4947     if (TargetBT && TargetBT->isInteger()) {
4948       if (S.SourceMgr.isInSystemMacro(CC))
4949         return;
4950 
4951       Expr *InnerE = E->IgnoreParenImpCasts();
4952       // We also want to warn on, e.g., "int i = -1.234"
4953       if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE))
4954         if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus)
4955           InnerE = UOp->getSubExpr()->IgnoreParenImpCasts();
4956 
4957       if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
4958         DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
4959       } else {
4960         DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
4961       }
4962     }
4963 
4964     // If the target is bool, warn if expr is a function or method call.
4965     if (Target->isSpecificBuiltinType(BuiltinType::Bool) &&
4966         isa<CallExpr>(E)) {
4967       // Check last argument of function call to see if it is an
4968       // implicit cast from a type matching the type the result
4969       // is being cast to.
4970       CallExpr *CEx = cast<CallExpr>(E);
4971       unsigned NumArgs = CEx->getNumArgs();
4972       if (NumArgs > 0) {
4973         Expr *LastA = CEx->getArg(NumArgs - 1);
4974         Expr *InnerE = LastA->IgnoreParenImpCasts();
4975         const Type *InnerType =
4976           S.Context.getCanonicalType(InnerE->getType()).getTypePtr();
4977         if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) {
4978           // Warn on this floating-point to bool conversion
4979           DiagnoseImpCast(S, E, T, CC,
4980                           diag::warn_impcast_floating_point_to_bool);
4981         }
4982       }
4983     }
4984     return;
4985   }
4986 
4987   if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)
4988            == Expr::NPCK_GNUNull) && !Target->isAnyPointerType()
4989       && !Target->isBlockPointerType() && !Target->isMemberPointerType()
4990       && Target->isScalarType() && !Target->isNullPtrType()) {
4991     SourceLocation Loc = E->getSourceRange().getBegin();
4992     if (Loc.isMacroID())
4993       Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first;
4994     if (!Loc.isMacroID() || CC.isMacroID())
4995       S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer)
4996           << T << clang::SourceRange(CC)
4997           << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T));
4998   }
4999 
5000   if (!Source->isIntegerType() || !Target->isIntegerType())
5001     return;
5002 
5003   // TODO: remove this early return once the false positives for constant->bool
5004   // in templates, macros, etc, are reduced or removed.
5005   if (Target->isSpecificBuiltinType(BuiltinType::Bool))
5006     return;
5007 
5008   IntRange SourceRange = GetExprRange(S.Context, E);
5009   IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
5010 
5011   if (SourceRange.Width > TargetRange.Width) {
5012     // If the source is a constant, use a default-on diagnostic.
5013     // TODO: this should happen for bitfield stores, too.
5014     llvm::APSInt Value(32);
5015     if (E->isIntegerConstantExpr(Value, S.Context)) {
5016       if (S.SourceMgr.isInSystemMacro(CC))
5017         return;
5018 
5019       std::string PrettySourceValue = Value.toString(10);
5020       std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
5021 
5022       S.DiagRuntimeBehavior(E->getExprLoc(), E,
5023         S.PDiag(diag::warn_impcast_integer_precision_constant)
5024             << PrettySourceValue << PrettyTargetValue
5025             << E->getType() << T << E->getSourceRange()
5026             << clang::SourceRange(CC));
5027       return;
5028     }
5029 
5030     // People want to build with -Wshorten-64-to-32 and not -Wconversion.
5031     if (S.SourceMgr.isInSystemMacro(CC))
5032       return;
5033 
5034     if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64)
5035       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32,
5036                              /* pruneControlFlow */ true);
5037     return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
5038   }
5039 
5040   if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
5041       (!TargetRange.NonNegative && SourceRange.NonNegative &&
5042        SourceRange.Width == TargetRange.Width)) {
5043 
5044     if (S.SourceMgr.isInSystemMacro(CC))
5045       return;
5046 
5047     unsigned DiagID = diag::warn_impcast_integer_sign;
5048 
5049     // Traditionally, gcc has warned about this under -Wsign-compare.
5050     // We also want to warn about it in -Wconversion.
5051     // So if -Wconversion is off, use a completely identical diagnostic
5052     // in the sign-compare group.
5053     // The conditional-checking code will
5054     if (ICContext) {
5055       DiagID = diag::warn_impcast_integer_sign_conditional;
5056       *ICContext = true;
5057     }
5058 
5059     return DiagnoseImpCast(S, E, T, CC, DiagID);
5060   }
5061 
5062   // Diagnose conversions between different enumeration types.
5063   // In C, we pretend that the type of an EnumConstantDecl is its enumeration
5064   // type, to give us better diagnostics.
5065   QualType SourceType = E->getType();
5066   if (!S.getLangOpts().CPlusPlus) {
5067     if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
5068       if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
5069         EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
5070         SourceType = S.Context.getTypeDeclType(Enum);
5071         Source = S.Context.getCanonicalType(SourceType).getTypePtr();
5072       }
5073   }
5074 
5075   if (const EnumType *SourceEnum = Source->getAs<EnumType>())
5076     if (const EnumType *TargetEnum = Target->getAs<EnumType>())
5077       if (SourceEnum->getDecl()->hasNameForLinkage() &&
5078           TargetEnum->getDecl()->hasNameForLinkage() &&
5079           SourceEnum != TargetEnum) {
5080         if (S.SourceMgr.isInSystemMacro(CC))
5081           return;
5082 
5083         return DiagnoseImpCast(S, E, SourceType, T, CC,
5084                                diag::warn_impcast_different_enum_types);
5085       }
5086 
5087   return;
5088 }
5089 
5090 void CheckConditionalOperator(Sema &S, ConditionalOperator *E,
5091                               SourceLocation CC, QualType T);
5092 
5093 void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
5094                              SourceLocation CC, bool &ICContext) {
5095   E = E->IgnoreParenImpCasts();
5096 
5097   if (isa<ConditionalOperator>(E))
5098     return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T);
5099 
5100   AnalyzeImplicitConversions(S, E, CC);
5101   if (E->getType() != T)
5102     return CheckImplicitConversion(S, E, T, CC, &ICContext);
5103   return;
5104 }
5105 
5106 void CheckConditionalOperator(Sema &S, ConditionalOperator *E,
5107                               SourceLocation CC, QualType T) {
5108   AnalyzeImplicitConversions(S, E->getCond(), CC);
5109 
5110   bool Suspicious = false;
5111   CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
5112   CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
5113 
5114   // If -Wconversion would have warned about either of the candidates
5115   // for a signedness conversion to the context type...
5116   if (!Suspicious) return;
5117 
5118   // ...but it's currently ignored...
5119   if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
5120                                  CC))
5121     return;
5122 
5123   // ...then check whether it would have warned about either of the
5124   // candidates for a signedness conversion to the condition type.
5125   if (E->getType() == T) return;
5126 
5127   Suspicious = false;
5128   CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
5129                           E->getType(), CC, &Suspicious);
5130   if (!Suspicious)
5131     CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
5132                             E->getType(), CC, &Suspicious);
5133 }
5134 
5135 /// AnalyzeImplicitConversions - Find and report any interesting
5136 /// implicit conversions in the given expression.  There are a couple
5137 /// of competing diagnostics here, -Wconversion and -Wsign-compare.
5138 void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
5139   QualType T = OrigE->getType();
5140   Expr *E = OrigE->IgnoreParenImpCasts();
5141 
5142   if (E->isTypeDependent() || E->isValueDependent())
5143     return;
5144 
5145   // For conditional operators, we analyze the arguments as if they
5146   // were being fed directly into the output.
5147   if (isa<ConditionalOperator>(E)) {
5148     ConditionalOperator *CO = cast<ConditionalOperator>(E);
5149     CheckConditionalOperator(S, CO, CC, T);
5150     return;
5151   }
5152 
5153   // Check implicit argument conversions for function calls.
5154   if (CallExpr *Call = dyn_cast<CallExpr>(E))
5155     CheckImplicitArgumentConversions(S, Call, CC);
5156 
5157   // Go ahead and check any implicit conversions we might have skipped.
5158   // The non-canonical typecheck is just an optimization;
5159   // CheckImplicitConversion will filter out dead implicit conversions.
5160   if (E->getType() != T)
5161     CheckImplicitConversion(S, E, T, CC);
5162 
5163   // Now continue drilling into this expression.
5164 
5165   if (PseudoObjectExpr * POE = dyn_cast<PseudoObjectExpr>(E)) {
5166     if (POE->getResultExpr())
5167       E = POE->getResultExpr();
5168   }
5169 
5170   if (const OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E))
5171     return AnalyzeImplicitConversions(S, OVE->getSourceExpr(), CC);
5172 
5173   // Skip past explicit casts.
5174   if (isa<ExplicitCastExpr>(E)) {
5175     E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
5176     return AnalyzeImplicitConversions(S, E, CC);
5177   }
5178 
5179   if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
5180     // Do a somewhat different check with comparison operators.
5181     if (BO->isComparisonOp())
5182       return AnalyzeComparison(S, BO);
5183 
5184     // And with simple assignments.
5185     if (BO->getOpcode() == BO_Assign)
5186       return AnalyzeAssignment(S, BO);
5187   }
5188 
5189   // These break the otherwise-useful invariant below.  Fortunately,
5190   // we don't really need to recurse into them, because any internal
5191   // expressions should have been analyzed already when they were
5192   // built into statements.
5193   if (isa<StmtExpr>(E)) return;
5194 
5195   // Don't descend into unevaluated contexts.
5196   if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
5197 
5198   // Now just recurse over the expression's children.
5199   CC = E->getExprLoc();
5200   BinaryOperator *BO = dyn_cast<BinaryOperator>(E);
5201   bool IsLogicalOperator = BO && BO->isLogicalOp();
5202   for (Stmt::child_range I = E->children(); I; ++I) {
5203     Expr *ChildExpr = dyn_cast_or_null<Expr>(*I);
5204     if (!ChildExpr)
5205       continue;
5206 
5207     if (IsLogicalOperator &&
5208         isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts()))
5209       // Ignore checking string literals that are in logical operators.
5210       continue;
5211     AnalyzeImplicitConversions(S, ChildExpr, CC);
5212   }
5213 }
5214 
5215 } // end anonymous namespace
5216 
5217 /// Diagnoses "dangerous" implicit conversions within the given
5218 /// expression (which is a full expression).  Implements -Wconversion
5219 /// and -Wsign-compare.
5220 ///
5221 /// \param CC the "context" location of the implicit conversion, i.e.
5222 ///   the most location of the syntactic entity requiring the implicit
5223 ///   conversion
5224 void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
5225   // Don't diagnose in unevaluated contexts.
5226   if (isUnevaluatedContext())
5227     return;
5228 
5229   // Don't diagnose for value- or type-dependent expressions.
5230   if (E->isTypeDependent() || E->isValueDependent())
5231     return;
5232 
5233   // Check for array bounds violations in cases where the check isn't triggered
5234   // elsewhere for other Expr types (like BinaryOperators), e.g. when an
5235   // ArraySubscriptExpr is on the RHS of a variable initialization.
5236   CheckArrayAccess(E);
5237 
5238   // This is not the right CC for (e.g.) a variable initialization.
5239   AnalyzeImplicitConversions(*this, E, CC);
5240 }
5241 
5242 /// Diagnose when expression is an integer constant expression and its evaluation
5243 /// results in integer overflow
5244 void Sema::CheckForIntOverflow (Expr *E) {
5245   if (isa<BinaryOperator>(E->IgnoreParens())) {
5246     llvm::SmallVector<PartialDiagnosticAt, 4> Diags;
5247     E->EvaluateForOverflow(Context, &Diags);
5248   }
5249 }
5250 
5251 namespace {
5252 /// \brief Visitor for expressions which looks for unsequenced operations on the
5253 /// same object.
5254 class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> {
5255   /// \brief A tree of sequenced regions within an expression. Two regions are
5256   /// unsequenced if one is an ancestor or a descendent of the other. When we
5257   /// finish processing an expression with sequencing, such as a comma
5258   /// expression, we fold its tree nodes into its parent, since they are
5259   /// unsequenced with respect to nodes we will visit later.
5260   class SequenceTree {
5261     struct Value {
5262       explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {}
5263       unsigned Parent : 31;
5264       bool Merged : 1;
5265     };
5266     llvm::SmallVector<Value, 8> Values;
5267 
5268   public:
5269     /// \brief A region within an expression which may be sequenced with respect
5270     /// to some other region.
5271     class Seq {
5272       explicit Seq(unsigned N) : Index(N) {}
5273       unsigned Index;
5274       friend class SequenceTree;
5275     public:
5276       Seq() : Index(0) {}
5277     };
5278 
5279     SequenceTree() { Values.push_back(Value(0)); }
5280     Seq root() const { return Seq(0); }
5281 
5282     /// \brief Create a new sequence of operations, which is an unsequenced
5283     /// subset of \p Parent. This sequence of operations is sequenced with
5284     /// respect to other children of \p Parent.
5285     Seq allocate(Seq Parent) {
5286       Values.push_back(Value(Parent.Index));
5287       return Seq(Values.size() - 1);
5288     }
5289 
5290     /// \brief Merge a sequence of operations into its parent.
5291     void merge(Seq S) {
5292       Values[S.Index].Merged = true;
5293     }
5294 
5295     /// \brief Determine whether two operations are unsequenced. This operation
5296     /// is asymmetric: \p Cur should be the more recent sequence, and \p Old
5297     /// should have been merged into its parent as appropriate.
5298     bool isUnsequenced(Seq Cur, Seq Old) {
5299       unsigned C = representative(Cur.Index);
5300       unsigned Target = representative(Old.Index);
5301       while (C >= Target) {
5302         if (C == Target)
5303           return true;
5304         C = Values[C].Parent;
5305       }
5306       return false;
5307     }
5308 
5309   private:
5310     /// \brief Pick a representative for a sequence.
5311     unsigned representative(unsigned K) {
5312       if (Values[K].Merged)
5313         // Perform path compression as we go.
5314         return Values[K].Parent = representative(Values[K].Parent);
5315       return K;
5316     }
5317   };
5318 
5319   /// An object for which we can track unsequenced uses.
5320   typedef NamedDecl *Object;
5321 
5322   /// Different flavors of object usage which we track. We only track the
5323   /// least-sequenced usage of each kind.
5324   enum UsageKind {
5325     /// A read of an object. Multiple unsequenced reads are OK.
5326     UK_Use,
5327     /// A modification of an object which is sequenced before the value
5328     /// computation of the expression, such as ++n.
5329     UK_ModAsValue,
5330     /// A modification of an object which is not sequenced before the value
5331     /// computation of the expression, such as n++.
5332     UK_ModAsSideEffect,
5333 
5334     UK_Count = UK_ModAsSideEffect + 1
5335   };
5336 
5337   struct Usage {
5338     Usage() : Use(0), Seq() {}
5339     Expr *Use;
5340     SequenceTree::Seq Seq;
5341   };
5342 
5343   struct UsageInfo {
5344     UsageInfo() : Diagnosed(false) {}
5345     Usage Uses[UK_Count];
5346     /// Have we issued a diagnostic for this variable already?
5347     bool Diagnosed;
5348   };
5349   typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap;
5350 
5351   Sema &SemaRef;
5352   /// Sequenced regions within the expression.
5353   SequenceTree Tree;
5354   /// Declaration modifications and references which we have seen.
5355   UsageInfoMap UsageMap;
5356   /// The region we are currently within.
5357   SequenceTree::Seq Region;
5358   /// Filled in with declarations which were modified as a side-effect
5359   /// (that is, post-increment operations).
5360   llvm::SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect;
5361   /// Expressions to check later. We defer checking these to reduce
5362   /// stack usage.
5363   llvm::SmallVectorImpl<Expr*> &WorkList;
5364 
5365   /// RAII object wrapping the visitation of a sequenced subexpression of an
5366   /// expression. At the end of this process, the side-effects of the evaluation
5367   /// become sequenced with respect to the value computation of the result, so
5368   /// we downgrade any UK_ModAsSideEffect within the evaluation to
5369   /// UK_ModAsValue.
5370   struct SequencedSubexpression {
5371     SequencedSubexpression(SequenceChecker &Self)
5372       : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) {
5373       Self.ModAsSideEffect = &ModAsSideEffect;
5374     }
5375     ~SequencedSubexpression() {
5376       for (unsigned I = 0, E = ModAsSideEffect.size(); I != E; ++I) {
5377         UsageInfo &U = Self.UsageMap[ModAsSideEffect[I].first];
5378         U.Uses[UK_ModAsSideEffect] = ModAsSideEffect[I].second;
5379         Self.addUsage(U, ModAsSideEffect[I].first,
5380                       ModAsSideEffect[I].second.Use, UK_ModAsValue);
5381       }
5382       Self.ModAsSideEffect = OldModAsSideEffect;
5383     }
5384 
5385     SequenceChecker &Self;
5386     llvm::SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect;
5387     llvm::SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect;
5388   };
5389 
5390   /// RAII object wrapping the visitation of a subexpression which we might
5391   /// choose to evaluate as a constant. If any subexpression is evaluated and
5392   /// found to be non-constant, this allows us to suppress the evaluation of
5393   /// the outer expression.
5394   class EvaluationTracker {
5395   public:
5396     EvaluationTracker(SequenceChecker &Self)
5397         : Self(Self), Prev(Self.EvalTracker), EvalOK(true) {
5398       Self.EvalTracker = this;
5399     }
5400     ~EvaluationTracker() {
5401       Self.EvalTracker = Prev;
5402       if (Prev)
5403         Prev->EvalOK &= EvalOK;
5404     }
5405 
5406     bool evaluate(const Expr *E, bool &Result) {
5407       if (!EvalOK || E->isValueDependent())
5408         return false;
5409       EvalOK = E->EvaluateAsBooleanCondition(Result, Self.SemaRef.Context);
5410       return EvalOK;
5411     }
5412 
5413   private:
5414     SequenceChecker &Self;
5415     EvaluationTracker *Prev;
5416     bool EvalOK;
5417   } *EvalTracker;
5418 
5419   /// \brief Find the object which is produced by the specified expression,
5420   /// if any.
5421   Object getObject(Expr *E, bool Mod) const {
5422     E = E->IgnoreParenCasts();
5423     if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
5424       if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec))
5425         return getObject(UO->getSubExpr(), Mod);
5426     } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
5427       if (BO->getOpcode() == BO_Comma)
5428         return getObject(BO->getRHS(), Mod);
5429       if (Mod && BO->isAssignmentOp())
5430         return getObject(BO->getLHS(), Mod);
5431     } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) {
5432       // FIXME: Check for more interesting cases, like "x.n = ++x.n".
5433       if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts()))
5434         return ME->getMemberDecl();
5435     } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
5436       // FIXME: If this is a reference, map through to its value.
5437       return DRE->getDecl();
5438     return 0;
5439   }
5440 
5441   /// \brief Note that an object was modified or used by an expression.
5442   void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) {
5443     Usage &U = UI.Uses[UK];
5444     if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) {
5445       if (UK == UK_ModAsSideEffect && ModAsSideEffect)
5446         ModAsSideEffect->push_back(std::make_pair(O, U));
5447       U.Use = Ref;
5448       U.Seq = Region;
5449     }
5450   }
5451   /// \brief Check whether a modification or use conflicts with a prior usage.
5452   void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind,
5453                   bool IsModMod) {
5454     if (UI.Diagnosed)
5455       return;
5456 
5457     const Usage &U = UI.Uses[OtherKind];
5458     if (!U.Use || !Tree.isUnsequenced(Region, U.Seq))
5459       return;
5460 
5461     Expr *Mod = U.Use;
5462     Expr *ModOrUse = Ref;
5463     if (OtherKind == UK_Use)
5464       std::swap(Mod, ModOrUse);
5465 
5466     SemaRef.Diag(Mod->getExprLoc(),
5467                  IsModMod ? diag::warn_unsequenced_mod_mod
5468                           : diag::warn_unsequenced_mod_use)
5469       << O << SourceRange(ModOrUse->getExprLoc());
5470     UI.Diagnosed = true;
5471   }
5472 
5473   void notePreUse(Object O, Expr *Use) {
5474     UsageInfo &U = UsageMap[O];
5475     // Uses conflict with other modifications.
5476     checkUsage(O, U, Use, UK_ModAsValue, false);
5477   }
5478   void notePostUse(Object O, Expr *Use) {
5479     UsageInfo &U = UsageMap[O];
5480     checkUsage(O, U, Use, UK_ModAsSideEffect, false);
5481     addUsage(U, O, Use, UK_Use);
5482   }
5483 
5484   void notePreMod(Object O, Expr *Mod) {
5485     UsageInfo &U = UsageMap[O];
5486     // Modifications conflict with other modifications and with uses.
5487     checkUsage(O, U, Mod, UK_ModAsValue, true);
5488     checkUsage(O, U, Mod, UK_Use, false);
5489   }
5490   void notePostMod(Object O, Expr *Use, UsageKind UK) {
5491     UsageInfo &U = UsageMap[O];
5492     checkUsage(O, U, Use, UK_ModAsSideEffect, true);
5493     addUsage(U, O, Use, UK);
5494   }
5495 
5496 public:
5497   SequenceChecker(Sema &S, Expr *E,
5498                   llvm::SmallVectorImpl<Expr*> &WorkList)
5499     : EvaluatedExprVisitor<SequenceChecker>(S.Context), SemaRef(S),
5500       Region(Tree.root()), ModAsSideEffect(0), WorkList(WorkList),
5501       EvalTracker(0) {
5502     Visit(E);
5503   }
5504 
5505   void VisitStmt(Stmt *S) {
5506     // Skip all statements which aren't expressions for now.
5507   }
5508 
5509   void VisitExpr(Expr *E) {
5510     // By default, just recurse to evaluated subexpressions.
5511     EvaluatedExprVisitor<SequenceChecker>::VisitStmt(E);
5512   }
5513 
5514   void VisitCastExpr(CastExpr *E) {
5515     Object O = Object();
5516     if (E->getCastKind() == CK_LValueToRValue)
5517       O = getObject(E->getSubExpr(), false);
5518 
5519     if (O)
5520       notePreUse(O, E);
5521     VisitExpr(E);
5522     if (O)
5523       notePostUse(O, E);
5524   }
5525 
5526   void VisitBinComma(BinaryOperator *BO) {
5527     // C++11 [expr.comma]p1:
5528     //   Every value computation and side effect associated with the left
5529     //   expression is sequenced before every value computation and side
5530     //   effect associated with the right expression.
5531     SequenceTree::Seq LHS = Tree.allocate(Region);
5532     SequenceTree::Seq RHS = Tree.allocate(Region);
5533     SequenceTree::Seq OldRegion = Region;
5534 
5535     {
5536       SequencedSubexpression SeqLHS(*this);
5537       Region = LHS;
5538       Visit(BO->getLHS());
5539     }
5540 
5541     Region = RHS;
5542     Visit(BO->getRHS());
5543 
5544     Region = OldRegion;
5545 
5546     // Forget that LHS and RHS are sequenced. They are both unsequenced
5547     // with respect to other stuff.
5548     Tree.merge(LHS);
5549     Tree.merge(RHS);
5550   }
5551 
5552   void VisitBinAssign(BinaryOperator *BO) {
5553     // The modification is sequenced after the value computation of the LHS
5554     // and RHS, so check it before inspecting the operands and update the
5555     // map afterwards.
5556     Object O = getObject(BO->getLHS(), true);
5557     if (!O)
5558       return VisitExpr(BO);
5559 
5560     notePreMod(O, BO);
5561 
5562     // C++11 [expr.ass]p7:
5563     //   E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated
5564     //   only once.
5565     //
5566     // Therefore, for a compound assignment operator, O is considered used
5567     // everywhere except within the evaluation of E1 itself.
5568     if (isa<CompoundAssignOperator>(BO))
5569       notePreUse(O, BO);
5570 
5571     Visit(BO->getLHS());
5572 
5573     if (isa<CompoundAssignOperator>(BO))
5574       notePostUse(O, BO);
5575 
5576     Visit(BO->getRHS());
5577 
5578     notePostMod(O, BO, UK_ModAsValue);
5579   }
5580   void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) {
5581     VisitBinAssign(CAO);
5582   }
5583 
5584   void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); }
5585   void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); }
5586   void VisitUnaryPreIncDec(UnaryOperator *UO) {
5587     Object O = getObject(UO->getSubExpr(), true);
5588     if (!O)
5589       return VisitExpr(UO);
5590 
5591     notePreMod(O, UO);
5592     Visit(UO->getSubExpr());
5593     notePostMod(O, UO, UK_ModAsValue);
5594   }
5595 
5596   void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); }
5597   void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); }
5598   void VisitUnaryPostIncDec(UnaryOperator *UO) {
5599     Object O = getObject(UO->getSubExpr(), true);
5600     if (!O)
5601       return VisitExpr(UO);
5602 
5603     notePreMod(O, UO);
5604     Visit(UO->getSubExpr());
5605     notePostMod(O, UO, UK_ModAsSideEffect);
5606   }
5607 
5608   /// Don't visit the RHS of '&&' or '||' if it might not be evaluated.
5609   void VisitBinLOr(BinaryOperator *BO) {
5610     // The side-effects of the LHS of an '&&' are sequenced before the
5611     // value computation of the RHS, and hence before the value computation
5612     // of the '&&' itself, unless the LHS evaluates to zero. We treat them
5613     // as if they were unconditionally sequenced.
5614     EvaluationTracker Eval(*this);
5615     {
5616       SequencedSubexpression Sequenced(*this);
5617       Visit(BO->getLHS());
5618     }
5619 
5620     bool Result;
5621     if (Eval.evaluate(BO->getLHS(), Result)) {
5622       if (!Result)
5623         Visit(BO->getRHS());
5624     } else {
5625       // Check for unsequenced operations in the RHS, treating it as an
5626       // entirely separate evaluation.
5627       //
5628       // FIXME: If there are operations in the RHS which are unsequenced
5629       // with respect to operations outside the RHS, and those operations
5630       // are unconditionally evaluated, diagnose them.
5631       WorkList.push_back(BO->getRHS());
5632     }
5633   }
5634   void VisitBinLAnd(BinaryOperator *BO) {
5635     EvaluationTracker Eval(*this);
5636     {
5637       SequencedSubexpression Sequenced(*this);
5638       Visit(BO->getLHS());
5639     }
5640 
5641     bool Result;
5642     if (Eval.evaluate(BO->getLHS(), Result)) {
5643       if (Result)
5644         Visit(BO->getRHS());
5645     } else {
5646       WorkList.push_back(BO->getRHS());
5647     }
5648   }
5649 
5650   // Only visit the condition, unless we can be sure which subexpression will
5651   // be chosen.
5652   void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) {
5653     EvaluationTracker Eval(*this);
5654     SequencedSubexpression Sequenced(*this);
5655     Visit(CO->getCond());
5656 
5657     bool Result;
5658     if (Eval.evaluate(CO->getCond(), Result))
5659       Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr());
5660     else {
5661       WorkList.push_back(CO->getTrueExpr());
5662       WorkList.push_back(CO->getFalseExpr());
5663     }
5664   }
5665 
5666   void VisitCXXConstructExpr(CXXConstructExpr *CCE) {
5667     if (!CCE->isListInitialization())
5668       return VisitExpr(CCE);
5669 
5670     // In C++11, list initializations are sequenced.
5671     llvm::SmallVector<SequenceTree::Seq, 32> Elts;
5672     SequenceTree::Seq Parent = Region;
5673     for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(),
5674                                         E = CCE->arg_end();
5675          I != E; ++I) {
5676       Region = Tree.allocate(Parent);
5677       Elts.push_back(Region);
5678       Visit(*I);
5679     }
5680 
5681     // Forget that the initializers are sequenced.
5682     Region = Parent;
5683     for (unsigned I = 0; I < Elts.size(); ++I)
5684       Tree.merge(Elts[I]);
5685   }
5686 
5687   void VisitInitListExpr(InitListExpr *ILE) {
5688     if (!SemaRef.getLangOpts().CPlusPlus11)
5689       return VisitExpr(ILE);
5690 
5691     // In C++11, list initializations are sequenced.
5692     llvm::SmallVector<SequenceTree::Seq, 32> Elts;
5693     SequenceTree::Seq Parent = Region;
5694     for (unsigned I = 0; I < ILE->getNumInits(); ++I) {
5695       Expr *E = ILE->getInit(I);
5696       if (!E) continue;
5697       Region = Tree.allocate(Parent);
5698       Elts.push_back(Region);
5699       Visit(E);
5700     }
5701 
5702     // Forget that the initializers are sequenced.
5703     Region = Parent;
5704     for (unsigned I = 0; I < Elts.size(); ++I)
5705       Tree.merge(Elts[I]);
5706   }
5707 };
5708 }
5709 
5710 void Sema::CheckUnsequencedOperations(Expr *E) {
5711   llvm::SmallVector<Expr*, 8> WorkList;
5712   WorkList.push_back(E);
5713   while (!WorkList.empty()) {
5714     Expr *Item = WorkList.back();
5715     WorkList.pop_back();
5716     SequenceChecker(*this, Item, WorkList);
5717   }
5718 }
5719 
5720 void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc,
5721                               bool IsConstexpr) {
5722   CheckImplicitConversions(E, CheckLoc);
5723   CheckUnsequencedOperations(E);
5724   if (!IsConstexpr && !E->isValueDependent())
5725     CheckForIntOverflow(E);
5726 }
5727 
5728 void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
5729                                        FieldDecl *BitField,
5730                                        Expr *Init) {
5731   (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
5732 }
5733 
5734 /// CheckParmsForFunctionDef - Check that the parameters of the given
5735 /// function are appropriate for the definition of a function. This
5736 /// takes care of any checks that cannot be performed on the
5737 /// declaration itself, e.g., that the types of each of the function
5738 /// parameters are complete.
5739 bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
5740                                     bool CheckParameterNames) {
5741   bool HasInvalidParm = false;
5742   for (; P != PEnd; ++P) {
5743     ParmVarDecl *Param = *P;
5744 
5745     // C99 6.7.5.3p4: the parameters in a parameter type list in a
5746     // function declarator that is part of a function definition of
5747     // that function shall not have incomplete type.
5748     //
5749     // This is also C++ [dcl.fct]p6.
5750     if (!Param->isInvalidDecl() &&
5751         RequireCompleteType(Param->getLocation(), Param->getType(),
5752                             diag::err_typecheck_decl_incomplete_type)) {
5753       Param->setInvalidDecl();
5754       HasInvalidParm = true;
5755     }
5756 
5757     // C99 6.9.1p5: If the declarator includes a parameter type list, the
5758     // declaration of each parameter shall include an identifier.
5759     if (CheckParameterNames &&
5760         Param->getIdentifier() == 0 &&
5761         !Param->isImplicit() &&
5762         !getLangOpts().CPlusPlus)
5763       Diag(Param->getLocation(), diag::err_parameter_name_omitted);
5764 
5765     // C99 6.7.5.3p12:
5766     //   If the function declarator is not part of a definition of that
5767     //   function, parameters may have incomplete type and may use the [*]
5768     //   notation in their sequences of declarator specifiers to specify
5769     //   variable length array types.
5770     QualType PType = Param->getOriginalType();
5771     while (const ArrayType *AT = Context.getAsArrayType(PType)) {
5772       if (AT->getSizeModifier() == ArrayType::Star) {
5773         // FIXME: This diagnostic should point the '[*]' if source-location
5774         // information is added for it.
5775         Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
5776         break;
5777       }
5778       PType= AT->getElementType();
5779     }
5780   }
5781 
5782   return HasInvalidParm;
5783 }
5784 
5785 /// CheckCastAlign - Implements -Wcast-align, which warns when a
5786 /// pointer cast increases the alignment requirements.
5787 void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
5788   // This is actually a lot of work to potentially be doing on every
5789   // cast; don't do it if we're ignoring -Wcast_align (as is the default).
5790   if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
5791                                           TRange.getBegin())
5792         == DiagnosticsEngine::Ignored)
5793     return;
5794 
5795   // Ignore dependent types.
5796   if (T->isDependentType() || Op->getType()->isDependentType())
5797     return;
5798 
5799   // Require that the destination be a pointer type.
5800   const PointerType *DestPtr = T->getAs<PointerType>();
5801   if (!DestPtr) return;
5802 
5803   // If the destination has alignment 1, we're done.
5804   QualType DestPointee = DestPtr->getPointeeType();
5805   if (DestPointee->isIncompleteType()) return;
5806   CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
5807   if (DestAlign.isOne()) return;
5808 
5809   // Require that the source be a pointer type.
5810   const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
5811   if (!SrcPtr) return;
5812   QualType SrcPointee = SrcPtr->getPointeeType();
5813 
5814   // Whitelist casts from cv void*.  We already implicitly
5815   // whitelisted casts to cv void*, since they have alignment 1.
5816   // Also whitelist casts involving incomplete types, which implicitly
5817   // includes 'void'.
5818   if (SrcPointee->isIncompleteType()) return;
5819 
5820   CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
5821   if (SrcAlign >= DestAlign) return;
5822 
5823   Diag(TRange.getBegin(), diag::warn_cast_align)
5824     << Op->getType() << T
5825     << static_cast<unsigned>(SrcAlign.getQuantity())
5826     << static_cast<unsigned>(DestAlign.getQuantity())
5827     << TRange << Op->getSourceRange();
5828 }
5829 
5830 static const Type* getElementType(const Expr *BaseExpr) {
5831   const Type* EltType = BaseExpr->getType().getTypePtr();
5832   if (EltType->isAnyPointerType())
5833     return EltType->getPointeeType().getTypePtr();
5834   else if (EltType->isArrayType())
5835     return EltType->getBaseElementTypeUnsafe();
5836   return EltType;
5837 }
5838 
5839 /// \brief Check whether this array fits the idiom of a size-one tail padded
5840 /// array member of a struct.
5841 ///
5842 /// We avoid emitting out-of-bounds access warnings for such arrays as they are
5843 /// commonly used to emulate flexible arrays in C89 code.
5844 static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size,
5845                                     const NamedDecl *ND) {
5846   if (Size != 1 || !ND) return false;
5847 
5848   const FieldDecl *FD = dyn_cast<FieldDecl>(ND);
5849   if (!FD) return false;
5850 
5851   // Don't consider sizes resulting from macro expansions or template argument
5852   // substitution to form C89 tail-padded arrays.
5853 
5854   TypeSourceInfo *TInfo = FD->getTypeSourceInfo();
5855   while (TInfo) {
5856     TypeLoc TL = TInfo->getTypeLoc();
5857     // Look through typedefs.
5858     if (TypedefTypeLoc TTL = TL.getAs<TypedefTypeLoc>()) {
5859       const TypedefNameDecl *TDL = TTL.getTypedefNameDecl();
5860       TInfo = TDL->getTypeSourceInfo();
5861       continue;
5862     }
5863     if (ConstantArrayTypeLoc CTL = TL.getAs<ConstantArrayTypeLoc>()) {
5864       const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr());
5865       if (!SizeExpr || SizeExpr->getExprLoc().isMacroID())
5866         return false;
5867     }
5868     break;
5869   }
5870 
5871   const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext());
5872   if (!RD) return false;
5873   if (RD->isUnion()) return false;
5874   if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) {
5875     if (!CRD->isStandardLayout()) return false;
5876   }
5877 
5878   // See if this is the last field decl in the record.
5879   const Decl *D = FD;
5880   while ((D = D->getNextDeclInContext()))
5881     if (isa<FieldDecl>(D))
5882       return false;
5883   return true;
5884 }
5885 
5886 void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr,
5887                             const ArraySubscriptExpr *ASE,
5888                             bool AllowOnePastEnd, bool IndexNegated) {
5889   IndexExpr = IndexExpr->IgnoreParenImpCasts();
5890   if (IndexExpr->isValueDependent())
5891     return;
5892 
5893   const Type *EffectiveType = getElementType(BaseExpr);
5894   BaseExpr = BaseExpr->IgnoreParenCasts();
5895   const ConstantArrayType *ArrayTy =
5896     Context.getAsConstantArrayType(BaseExpr->getType());
5897   if (!ArrayTy)
5898     return;
5899 
5900   llvm::APSInt index;
5901   if (!IndexExpr->EvaluateAsInt(index, Context))
5902     return;
5903   if (IndexNegated)
5904     index = -index;
5905 
5906   const NamedDecl *ND = NULL;
5907   if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
5908     ND = dyn_cast<NamedDecl>(DRE->getDecl());
5909   if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
5910     ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
5911 
5912   if (index.isUnsigned() || !index.isNegative()) {
5913     llvm::APInt size = ArrayTy->getSize();
5914     if (!size.isStrictlyPositive())
5915       return;
5916 
5917     const Type* BaseType = getElementType(BaseExpr);
5918     if (BaseType != EffectiveType) {
5919       // Make sure we're comparing apples to apples when comparing index to size
5920       uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType);
5921       uint64_t array_typesize = Context.getTypeSize(BaseType);
5922       // Handle ptrarith_typesize being zero, such as when casting to void*
5923       if (!ptrarith_typesize) ptrarith_typesize = 1;
5924       if (ptrarith_typesize != array_typesize) {
5925         // There's a cast to a different size type involved
5926         uint64_t ratio = array_typesize / ptrarith_typesize;
5927         // TODO: Be smarter about handling cases where array_typesize is not a
5928         // multiple of ptrarith_typesize
5929         if (ptrarith_typesize * ratio == array_typesize)
5930           size *= llvm::APInt(size.getBitWidth(), ratio);
5931       }
5932     }
5933 
5934     if (size.getBitWidth() > index.getBitWidth())
5935       index = index.zext(size.getBitWidth());
5936     else if (size.getBitWidth() < index.getBitWidth())
5937       size = size.zext(index.getBitWidth());
5938 
5939     // For array subscripting the index must be less than size, but for pointer
5940     // arithmetic also allow the index (offset) to be equal to size since
5941     // computing the next address after the end of the array is legal and
5942     // commonly done e.g. in C++ iterators and range-based for loops.
5943     if (AllowOnePastEnd ? index.ule(size) : index.ult(size))
5944       return;
5945 
5946     // Also don't warn for arrays of size 1 which are members of some
5947     // structure. These are often used to approximate flexible arrays in C89
5948     // code.
5949     if (IsTailPaddedMemberArray(*this, size, ND))
5950       return;
5951 
5952     // Suppress the warning if the subscript expression (as identified by the
5953     // ']' location) and the index expression are both from macro expansions
5954     // within a system header.
5955     if (ASE) {
5956       SourceLocation RBracketLoc = SourceMgr.getSpellingLoc(
5957           ASE->getRBracketLoc());
5958       if (SourceMgr.isInSystemHeader(RBracketLoc)) {
5959         SourceLocation IndexLoc = SourceMgr.getSpellingLoc(
5960             IndexExpr->getLocStart());
5961         if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc))
5962           return;
5963       }
5964     }
5965 
5966     unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds;
5967     if (ASE)
5968       DiagID = diag::warn_array_index_exceeds_bounds;
5969 
5970     DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
5971                         PDiag(DiagID) << index.toString(10, true)
5972                           << size.toString(10, true)
5973                           << (unsigned)size.getLimitedValue(~0U)
5974                           << IndexExpr->getSourceRange());
5975   } else {
5976     unsigned DiagID = diag::warn_array_index_precedes_bounds;
5977     if (!ASE) {
5978       DiagID = diag::warn_ptr_arith_precedes_bounds;
5979       if (index.isNegative()) index = -index;
5980     }
5981 
5982     DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
5983                         PDiag(DiagID) << index.toString(10, true)
5984                           << IndexExpr->getSourceRange());
5985   }
5986 
5987   if (!ND) {
5988     // Try harder to find a NamedDecl to point at in the note.
5989     while (const ArraySubscriptExpr *ASE =
5990            dyn_cast<ArraySubscriptExpr>(BaseExpr))
5991       BaseExpr = ASE->getBase()->IgnoreParenCasts();
5992     if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
5993       ND = dyn_cast<NamedDecl>(DRE->getDecl());
5994     if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
5995       ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
5996   }
5997 
5998   if (ND)
5999     DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
6000                         PDiag(diag::note_array_index_out_of_bounds)
6001                           << ND->getDeclName());
6002 }
6003 
6004 void Sema::CheckArrayAccess(const Expr *expr) {
6005   int AllowOnePastEnd = 0;
6006   while (expr) {
6007     expr = expr->IgnoreParenImpCasts();
6008     switch (expr->getStmtClass()) {
6009       case Stmt::ArraySubscriptExprClass: {
6010         const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr);
6011         CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE,
6012                          AllowOnePastEnd > 0);
6013         return;
6014       }
6015       case Stmt::UnaryOperatorClass: {
6016         // Only unwrap the * and & unary operators
6017         const UnaryOperator *UO = cast<UnaryOperator>(expr);
6018         expr = UO->getSubExpr();
6019         switch (UO->getOpcode()) {
6020           case UO_AddrOf:
6021             AllowOnePastEnd++;
6022             break;
6023           case UO_Deref:
6024             AllowOnePastEnd--;
6025             break;
6026           default:
6027             return;
6028         }
6029         break;
6030       }
6031       case Stmt::ConditionalOperatorClass: {
6032         const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
6033         if (const Expr *lhs = cond->getLHS())
6034           CheckArrayAccess(lhs);
6035         if (const Expr *rhs = cond->getRHS())
6036           CheckArrayAccess(rhs);
6037         return;
6038       }
6039       default:
6040         return;
6041     }
6042   }
6043 }
6044 
6045 //===--- CHECK: Objective-C retain cycles ----------------------------------//
6046 
6047 namespace {
6048   struct RetainCycleOwner {
6049     RetainCycleOwner() : Variable(0), Indirect(false) {}
6050     VarDecl *Variable;
6051     SourceRange Range;
6052     SourceLocation Loc;
6053     bool Indirect;
6054 
6055     void setLocsFrom(Expr *e) {
6056       Loc = e->getExprLoc();
6057       Range = e->getSourceRange();
6058     }
6059   };
6060 }
6061 
6062 /// Consider whether capturing the given variable can possibly lead to
6063 /// a retain cycle.
6064 static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) {
6065   // In ARC, it's captured strongly iff the variable has __strong
6066   // lifetime.  In MRR, it's captured strongly if the variable is
6067   // __block and has an appropriate type.
6068   if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
6069     return false;
6070 
6071   owner.Variable = var;
6072   if (ref)
6073     owner.setLocsFrom(ref);
6074   return true;
6075 }
6076 
6077 static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) {
6078   while (true) {
6079     e = e->IgnoreParens();
6080     if (CastExpr *cast = dyn_cast<CastExpr>(e)) {
6081       switch (cast->getCastKind()) {
6082       case CK_BitCast:
6083       case CK_LValueBitCast:
6084       case CK_LValueToRValue:
6085       case CK_ARCReclaimReturnedObject:
6086         e = cast->getSubExpr();
6087         continue;
6088 
6089       default:
6090         return false;
6091       }
6092     }
6093 
6094     if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) {
6095       ObjCIvarDecl *ivar = ref->getDecl();
6096       if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
6097         return false;
6098 
6099       // Try to find a retain cycle in the base.
6100       if (!findRetainCycleOwner(S, ref->getBase(), owner))
6101         return false;
6102 
6103       if (ref->isFreeIvar()) owner.setLocsFrom(ref);
6104       owner.Indirect = true;
6105       return true;
6106     }
6107 
6108     if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) {
6109       VarDecl *var = dyn_cast<VarDecl>(ref->getDecl());
6110       if (!var) return false;
6111       return considerVariable(var, ref, owner);
6112     }
6113 
6114     if (MemberExpr *member = dyn_cast<MemberExpr>(e)) {
6115       if (member->isArrow()) return false;
6116 
6117       // Don't count this as an indirect ownership.
6118       e = member->getBase();
6119       continue;
6120     }
6121 
6122     if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) {
6123       // Only pay attention to pseudo-objects on property references.
6124       ObjCPropertyRefExpr *pre
6125         = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm()
6126                                               ->IgnoreParens());
6127       if (!pre) return false;
6128       if (pre->isImplicitProperty()) return false;
6129       ObjCPropertyDecl *property = pre->getExplicitProperty();
6130       if (!property->isRetaining() &&
6131           !(property->getPropertyIvarDecl() &&
6132             property->getPropertyIvarDecl()->getType()
6133               .getObjCLifetime() == Qualifiers::OCL_Strong))
6134           return false;
6135 
6136       owner.Indirect = true;
6137       if (pre->isSuperReceiver()) {
6138         owner.Variable = S.getCurMethodDecl()->getSelfDecl();
6139         if (!owner.Variable)
6140           return false;
6141         owner.Loc = pre->getLocation();
6142         owner.Range = pre->getSourceRange();
6143         return true;
6144       }
6145       e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase())
6146                               ->getSourceExpr());
6147       continue;
6148     }
6149 
6150     // Array ivars?
6151 
6152     return false;
6153   }
6154 }
6155 
6156 namespace {
6157   struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> {
6158     FindCaptureVisitor(ASTContext &Context, VarDecl *variable)
6159       : EvaluatedExprVisitor<FindCaptureVisitor>(Context),
6160         Variable(variable), Capturer(0) {}
6161 
6162     VarDecl *Variable;
6163     Expr *Capturer;
6164 
6165     void VisitDeclRefExpr(DeclRefExpr *ref) {
6166       if (ref->getDecl() == Variable && !Capturer)
6167         Capturer = ref;
6168     }
6169 
6170     void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) {
6171       if (Capturer) return;
6172       Visit(ref->getBase());
6173       if (Capturer && ref->isFreeIvar())
6174         Capturer = ref;
6175     }
6176 
6177     void VisitBlockExpr(BlockExpr *block) {
6178       // Look inside nested blocks
6179       if (block->getBlockDecl()->capturesVariable(Variable))
6180         Visit(block->getBlockDecl()->getBody());
6181     }
6182 
6183     void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) {
6184       if (Capturer) return;
6185       if (OVE->getSourceExpr())
6186         Visit(OVE->getSourceExpr());
6187     }
6188   };
6189 }
6190 
6191 /// Check whether the given argument is a block which captures a
6192 /// variable.
6193 static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) {
6194   assert(owner.Variable && owner.Loc.isValid());
6195 
6196   e = e->IgnoreParenCasts();
6197 
6198   // Look through [^{...} copy] and Block_copy(^{...}).
6199   if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) {
6200     Selector Cmd = ME->getSelector();
6201     if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") {
6202       e = ME->getInstanceReceiver();
6203       if (!e)
6204         return 0;
6205       e = e->IgnoreParenCasts();
6206     }
6207   } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) {
6208     if (CE->getNumArgs() == 1) {
6209       FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl());
6210       if (Fn) {
6211         const IdentifierInfo *FnI = Fn->getIdentifier();
6212         if (FnI && FnI->isStr("_Block_copy")) {
6213           e = CE->getArg(0)->IgnoreParenCasts();
6214         }
6215       }
6216     }
6217   }
6218 
6219   BlockExpr *block = dyn_cast<BlockExpr>(e);
6220   if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable))
6221     return 0;
6222 
6223   FindCaptureVisitor visitor(S.Context, owner.Variable);
6224   visitor.Visit(block->getBlockDecl()->getBody());
6225   return visitor.Capturer;
6226 }
6227 
6228 static void diagnoseRetainCycle(Sema &S, Expr *capturer,
6229                                 RetainCycleOwner &owner) {
6230   assert(capturer);
6231   assert(owner.Variable && owner.Loc.isValid());
6232 
6233   S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle)
6234     << owner.Variable << capturer->getSourceRange();
6235   S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner)
6236     << owner.Indirect << owner.Range;
6237 }
6238 
6239 /// Check for a keyword selector that starts with the word 'add' or
6240 /// 'set'.
6241 static bool isSetterLikeSelector(Selector sel) {
6242   if (sel.isUnarySelector()) return false;
6243 
6244   StringRef str = sel.getNameForSlot(0);
6245   while (!str.empty() && str.front() == '_') str = str.substr(1);
6246   if (str.startswith("set"))
6247     str = str.substr(3);
6248   else if (str.startswith("add")) {
6249     // Specially whitelist 'addOperationWithBlock:'.
6250     if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock"))
6251       return false;
6252     str = str.substr(3);
6253   }
6254   else
6255     return false;
6256 
6257   if (str.empty()) return true;
6258   return !isLowercase(str.front());
6259 }
6260 
6261 /// Check a message send to see if it's likely to cause a retain cycle.
6262 void Sema::checkRetainCycles(ObjCMessageExpr *msg) {
6263   // Only check instance methods whose selector looks like a setter.
6264   if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector()))
6265     return;
6266 
6267   // Try to find a variable that the receiver is strongly owned by.
6268   RetainCycleOwner owner;
6269   if (msg->getReceiverKind() == ObjCMessageExpr::Instance) {
6270     if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner))
6271       return;
6272   } else {
6273     assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance);
6274     owner.Variable = getCurMethodDecl()->getSelfDecl();
6275     owner.Loc = msg->getSuperLoc();
6276     owner.Range = msg->getSuperLoc();
6277   }
6278 
6279   // Check whether the receiver is captured by any of the arguments.
6280   for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i)
6281     if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner))
6282       return diagnoseRetainCycle(*this, capturer, owner);
6283 }
6284 
6285 /// Check a property assign to see if it's likely to cause a retain cycle.
6286 void Sema::checkRetainCycles(Expr *receiver, Expr *argument) {
6287   RetainCycleOwner owner;
6288   if (!findRetainCycleOwner(*this, receiver, owner))
6289     return;
6290 
6291   if (Expr *capturer = findCapturingExpr(*this, argument, owner))
6292     diagnoseRetainCycle(*this, capturer, owner);
6293 }
6294 
6295 void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) {
6296   RetainCycleOwner Owner;
6297   if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner))
6298     return;
6299 
6300   // Because we don't have an expression for the variable, we have to set the
6301   // location explicitly here.
6302   Owner.Loc = Var->getLocation();
6303   Owner.Range = Var->getSourceRange();
6304 
6305   if (Expr *Capturer = findCapturingExpr(*this, Init, Owner))
6306     diagnoseRetainCycle(*this, Capturer, Owner);
6307 }
6308 
6309 static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc,
6310                                      Expr *RHS, bool isProperty) {
6311   // Check if RHS is an Objective-C object literal, which also can get
6312   // immediately zapped in a weak reference.  Note that we explicitly
6313   // allow ObjCStringLiterals, since those are designed to never really die.
6314   RHS = RHS->IgnoreParenImpCasts();
6315 
6316   // This enum needs to match with the 'select' in
6317   // warn_objc_arc_literal_assign (off-by-1).
6318   Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS);
6319   if (Kind == Sema::LK_String || Kind == Sema::LK_None)
6320     return false;
6321 
6322   S.Diag(Loc, diag::warn_arc_literal_assign)
6323     << (unsigned) Kind
6324     << (isProperty ? 0 : 1)
6325     << RHS->getSourceRange();
6326 
6327   return true;
6328 }
6329 
6330 static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc,
6331                                     Qualifiers::ObjCLifetime LT,
6332                                     Expr *RHS, bool isProperty) {
6333   // Strip off any implicit cast added to get to the one ARC-specific.
6334   while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
6335     if (cast->getCastKind() == CK_ARCConsumeObject) {
6336       S.Diag(Loc, diag::warn_arc_retained_assign)
6337         << (LT == Qualifiers::OCL_ExplicitNone)
6338         << (isProperty ? 0 : 1)
6339         << RHS->getSourceRange();
6340       return true;
6341     }
6342     RHS = cast->getSubExpr();
6343   }
6344 
6345   if (LT == Qualifiers::OCL_Weak &&
6346       checkUnsafeAssignLiteral(S, Loc, RHS, isProperty))
6347     return true;
6348 
6349   return false;
6350 }
6351 
6352 bool Sema::checkUnsafeAssigns(SourceLocation Loc,
6353                               QualType LHS, Expr *RHS) {
6354   Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime();
6355 
6356   if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone)
6357     return false;
6358 
6359   if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false))
6360     return true;
6361 
6362   return false;
6363 }
6364 
6365 void Sema::checkUnsafeExprAssigns(SourceLocation Loc,
6366                               Expr *LHS, Expr *RHS) {
6367   QualType LHSType;
6368   // PropertyRef on LHS type need be directly obtained from
6369   // its declaration as it has a PsuedoType.
6370   ObjCPropertyRefExpr *PRE
6371     = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens());
6372   if (PRE && !PRE->isImplicitProperty()) {
6373     const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
6374     if (PD)
6375       LHSType = PD->getType();
6376   }
6377 
6378   if (LHSType.isNull())
6379     LHSType = LHS->getType();
6380 
6381   Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime();
6382 
6383   if (LT == Qualifiers::OCL_Weak) {
6384     DiagnosticsEngine::Level Level =
6385       Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc);
6386     if (Level != DiagnosticsEngine::Ignored)
6387       getCurFunction()->markSafeWeakUse(LHS);
6388   }
6389 
6390   if (checkUnsafeAssigns(Loc, LHSType, RHS))
6391     return;
6392 
6393   // FIXME. Check for other life times.
6394   if (LT != Qualifiers::OCL_None)
6395     return;
6396 
6397   if (PRE) {
6398     if (PRE->isImplicitProperty())
6399       return;
6400     const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
6401     if (!PD)
6402       return;
6403 
6404     unsigned Attributes = PD->getPropertyAttributes();
6405     if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) {
6406       // when 'assign' attribute was not explicitly specified
6407       // by user, ignore it and rely on property type itself
6408       // for lifetime info.
6409       unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten();
6410       if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) &&
6411           LHSType->isObjCRetainableType())
6412         return;
6413 
6414       while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
6415         if (cast->getCastKind() == CK_ARCConsumeObject) {
6416           Diag(Loc, diag::warn_arc_retained_property_assign)
6417           << RHS->getSourceRange();
6418           return;
6419         }
6420         RHS = cast->getSubExpr();
6421       }
6422     }
6423     else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) {
6424       if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true))
6425         return;
6426     }
6427   }
6428 }
6429 
6430 //===--- CHECK: Empty statement body (-Wempty-body) ---------------------===//
6431 
6432 namespace {
6433 bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr,
6434                                  SourceLocation StmtLoc,
6435                                  const NullStmt *Body) {
6436   // Do not warn if the body is a macro that expands to nothing, e.g:
6437   //
6438   // #define CALL(x)
6439   // if (condition)
6440   //   CALL(0);
6441   //
6442   if (Body->hasLeadingEmptyMacro())
6443     return false;
6444 
6445   // Get line numbers of statement and body.
6446   bool StmtLineInvalid;
6447   unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc,
6448                                                       &StmtLineInvalid);
6449   if (StmtLineInvalid)
6450     return false;
6451 
6452   bool BodyLineInvalid;
6453   unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(),
6454                                                       &BodyLineInvalid);
6455   if (BodyLineInvalid)
6456     return false;
6457 
6458   // Warn if null statement and body are on the same line.
6459   if (StmtLine != BodyLine)
6460     return false;
6461 
6462   return true;
6463 }
6464 } // Unnamed namespace
6465 
6466 void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc,
6467                                  const Stmt *Body,
6468                                  unsigned DiagID) {
6469   // Since this is a syntactic check, don't emit diagnostic for template
6470   // instantiations, this just adds noise.
6471   if (CurrentInstantiationScope)
6472     return;
6473 
6474   // The body should be a null statement.
6475   const NullStmt *NBody = dyn_cast<NullStmt>(Body);
6476   if (!NBody)
6477     return;
6478 
6479   // Do the usual checks.
6480   if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody))
6481     return;
6482 
6483   Diag(NBody->getSemiLoc(), DiagID);
6484   Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line);
6485 }
6486 
6487 void Sema::DiagnoseEmptyLoopBody(const Stmt *S,
6488                                  const Stmt *PossibleBody) {
6489   assert(!CurrentInstantiationScope); // Ensured by caller
6490 
6491   SourceLocation StmtLoc;
6492   const Stmt *Body;
6493   unsigned DiagID;
6494   if (const ForStmt *FS = dyn_cast<ForStmt>(S)) {
6495     StmtLoc = FS->getRParenLoc();
6496     Body = FS->getBody();
6497     DiagID = diag::warn_empty_for_body;
6498   } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) {
6499     StmtLoc = WS->getCond()->getSourceRange().getEnd();
6500     Body = WS->getBody();
6501     DiagID = diag::warn_empty_while_body;
6502   } else
6503     return; // Neither `for' nor `while'.
6504 
6505   // The body should be a null statement.
6506   const NullStmt *NBody = dyn_cast<NullStmt>(Body);
6507   if (!NBody)
6508     return;
6509 
6510   // Skip expensive checks if diagnostic is disabled.
6511   if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) ==
6512           DiagnosticsEngine::Ignored)
6513     return;
6514 
6515   // Do the usual checks.
6516   if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody))
6517     return;
6518 
6519   // `for(...);' and `while(...);' are popular idioms, so in order to keep
6520   // noise level low, emit diagnostics only if for/while is followed by a
6521   // CompoundStmt, e.g.:
6522   //    for (int i = 0; i < n; i++);
6523   //    {
6524   //      a(i);
6525   //    }
6526   // or if for/while is followed by a statement with more indentation
6527   // than for/while itself:
6528   //    for (int i = 0; i < n; i++);
6529   //      a(i);
6530   bool ProbableTypo = isa<CompoundStmt>(PossibleBody);
6531   if (!ProbableTypo) {
6532     bool BodyColInvalid;
6533     unsigned BodyCol = SourceMgr.getPresumedColumnNumber(
6534                              PossibleBody->getLocStart(),
6535                              &BodyColInvalid);
6536     if (BodyColInvalid)
6537       return;
6538 
6539     bool StmtColInvalid;
6540     unsigned StmtCol = SourceMgr.getPresumedColumnNumber(
6541                              S->getLocStart(),
6542                              &StmtColInvalid);
6543     if (StmtColInvalid)
6544       return;
6545 
6546     if (BodyCol > StmtCol)
6547       ProbableTypo = true;
6548   }
6549 
6550   if (ProbableTypo) {
6551     Diag(NBody->getSemiLoc(), DiagID);
6552     Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line);
6553   }
6554 }
6555 
6556 //===--- Layout compatibility ----------------------------------------------//
6557 
6558 namespace {
6559 
6560 bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2);
6561 
6562 /// \brief Check if two enumeration types are layout-compatible.
6563 bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) {
6564   // C++11 [dcl.enum] p8:
6565   // Two enumeration types are layout-compatible if they have the same
6566   // underlying type.
6567   return ED1->isComplete() && ED2->isComplete() &&
6568          C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType());
6569 }
6570 
6571 /// \brief Check if two fields are layout-compatible.
6572 bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) {
6573   if (!isLayoutCompatible(C, Field1->getType(), Field2->getType()))
6574     return false;
6575 
6576   if (Field1->isBitField() != Field2->isBitField())
6577     return false;
6578 
6579   if (Field1->isBitField()) {
6580     // Make sure that the bit-fields are the same length.
6581     unsigned Bits1 = Field1->getBitWidthValue(C);
6582     unsigned Bits2 = Field2->getBitWidthValue(C);
6583 
6584     if (Bits1 != Bits2)
6585       return false;
6586   }
6587 
6588   return true;
6589 }
6590 
6591 /// \brief Check if two standard-layout structs are layout-compatible.
6592 /// (C++11 [class.mem] p17)
6593 bool isLayoutCompatibleStruct(ASTContext &C,
6594                               RecordDecl *RD1,
6595                               RecordDecl *RD2) {
6596   // If both records are C++ classes, check that base classes match.
6597   if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) {
6598     // If one of records is a CXXRecordDecl we are in C++ mode,
6599     // thus the other one is a CXXRecordDecl, too.
6600     const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2);
6601     // Check number of base classes.
6602     if (D1CXX->getNumBases() != D2CXX->getNumBases())
6603       return false;
6604 
6605     // Check the base classes.
6606     for (CXXRecordDecl::base_class_const_iterator
6607                Base1 = D1CXX->bases_begin(),
6608            BaseEnd1 = D1CXX->bases_end(),
6609               Base2 = D2CXX->bases_begin();
6610          Base1 != BaseEnd1;
6611          ++Base1, ++Base2) {
6612       if (!isLayoutCompatible(C, Base1->getType(), Base2->getType()))
6613         return false;
6614     }
6615   } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) {
6616     // If only RD2 is a C++ class, it should have zero base classes.
6617     if (D2CXX->getNumBases() > 0)
6618       return false;
6619   }
6620 
6621   // Check the fields.
6622   RecordDecl::field_iterator Field2 = RD2->field_begin(),
6623                              Field2End = RD2->field_end(),
6624                              Field1 = RD1->field_begin(),
6625                              Field1End = RD1->field_end();
6626   for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) {
6627     if (!isLayoutCompatible(C, *Field1, *Field2))
6628       return false;
6629   }
6630   if (Field1 != Field1End || Field2 != Field2End)
6631     return false;
6632 
6633   return true;
6634 }
6635 
6636 /// \brief Check if two standard-layout unions are layout-compatible.
6637 /// (C++11 [class.mem] p18)
6638 bool isLayoutCompatibleUnion(ASTContext &C,
6639                              RecordDecl *RD1,
6640                              RecordDecl *RD2) {
6641   llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields;
6642   for (RecordDecl::field_iterator Field2 = RD2->field_begin(),
6643                                   Field2End = RD2->field_end();
6644        Field2 != Field2End; ++Field2) {
6645     UnmatchedFields.insert(*Field2);
6646   }
6647 
6648   for (RecordDecl::field_iterator Field1 = RD1->field_begin(),
6649                                   Field1End = RD1->field_end();
6650        Field1 != Field1End; ++Field1) {
6651     llvm::SmallPtrSet<FieldDecl *, 8>::iterator
6652         I = UnmatchedFields.begin(),
6653         E = UnmatchedFields.end();
6654 
6655     for ( ; I != E; ++I) {
6656       if (isLayoutCompatible(C, *Field1, *I)) {
6657         bool Result = UnmatchedFields.erase(*I);
6658         (void) Result;
6659         assert(Result);
6660         break;
6661       }
6662     }
6663     if (I == E)
6664       return false;
6665   }
6666 
6667   return UnmatchedFields.empty();
6668 }
6669 
6670 bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) {
6671   if (RD1->isUnion() != RD2->isUnion())
6672     return false;
6673 
6674   if (RD1->isUnion())
6675     return isLayoutCompatibleUnion(C, RD1, RD2);
6676   else
6677     return isLayoutCompatibleStruct(C, RD1, RD2);
6678 }
6679 
6680 /// \brief Check if two types are layout-compatible in C++11 sense.
6681 bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) {
6682   if (T1.isNull() || T2.isNull())
6683     return false;
6684 
6685   // C++11 [basic.types] p11:
6686   // If two types T1 and T2 are the same type, then T1 and T2 are
6687   // layout-compatible types.
6688   if (C.hasSameType(T1, T2))
6689     return true;
6690 
6691   T1 = T1.getCanonicalType().getUnqualifiedType();
6692   T2 = T2.getCanonicalType().getUnqualifiedType();
6693 
6694   const Type::TypeClass TC1 = T1->getTypeClass();
6695   const Type::TypeClass TC2 = T2->getTypeClass();
6696 
6697   if (TC1 != TC2)
6698     return false;
6699 
6700   if (TC1 == Type::Enum) {
6701     return isLayoutCompatible(C,
6702                               cast<EnumType>(T1)->getDecl(),
6703                               cast<EnumType>(T2)->getDecl());
6704   } else if (TC1 == Type::Record) {
6705     if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType())
6706       return false;
6707 
6708     return isLayoutCompatible(C,
6709                               cast<RecordType>(T1)->getDecl(),
6710                               cast<RecordType>(T2)->getDecl());
6711   }
6712 
6713   return false;
6714 }
6715 }
6716 
6717 //===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----//
6718 
6719 namespace {
6720 /// \brief Given a type tag expression find the type tag itself.
6721 ///
6722 /// \param TypeExpr Type tag expression, as it appears in user's code.
6723 ///
6724 /// \param VD Declaration of an identifier that appears in a type tag.
6725 ///
6726 /// \param MagicValue Type tag magic value.
6727 bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx,
6728                      const ValueDecl **VD, uint64_t *MagicValue) {
6729   while(true) {
6730     if (!TypeExpr)
6731       return false;
6732 
6733     TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts();
6734 
6735     switch (TypeExpr->getStmtClass()) {
6736     case Stmt::UnaryOperatorClass: {
6737       const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr);
6738       if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) {
6739         TypeExpr = UO->getSubExpr();
6740         continue;
6741       }
6742       return false;
6743     }
6744 
6745     case Stmt::DeclRefExprClass: {
6746       const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr);
6747       *VD = DRE->getDecl();
6748       return true;
6749     }
6750 
6751     case Stmt::IntegerLiteralClass: {
6752       const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr);
6753       llvm::APInt MagicValueAPInt = IL->getValue();
6754       if (MagicValueAPInt.getActiveBits() <= 64) {
6755         *MagicValue = MagicValueAPInt.getZExtValue();
6756         return true;
6757       } else
6758         return false;
6759     }
6760 
6761     case Stmt::BinaryConditionalOperatorClass:
6762     case Stmt::ConditionalOperatorClass: {
6763       const AbstractConditionalOperator *ACO =
6764           cast<AbstractConditionalOperator>(TypeExpr);
6765       bool Result;
6766       if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) {
6767         if (Result)
6768           TypeExpr = ACO->getTrueExpr();
6769         else
6770           TypeExpr = ACO->getFalseExpr();
6771         continue;
6772       }
6773       return false;
6774     }
6775 
6776     case Stmt::BinaryOperatorClass: {
6777       const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr);
6778       if (BO->getOpcode() == BO_Comma) {
6779         TypeExpr = BO->getRHS();
6780         continue;
6781       }
6782       return false;
6783     }
6784 
6785     default:
6786       return false;
6787     }
6788   }
6789 }
6790 
6791 /// \brief Retrieve the C type corresponding to type tag TypeExpr.
6792 ///
6793 /// \param TypeExpr Expression that specifies a type tag.
6794 ///
6795 /// \param MagicValues Registered magic values.
6796 ///
6797 /// \param FoundWrongKind Set to true if a type tag was found, but of a wrong
6798 ///        kind.
6799 ///
6800 /// \param TypeInfo Information about the corresponding C type.
6801 ///
6802 /// \returns true if the corresponding C type was found.
6803 bool GetMatchingCType(
6804         const IdentifierInfo *ArgumentKind,
6805         const Expr *TypeExpr, const ASTContext &Ctx,
6806         const llvm::DenseMap<Sema::TypeTagMagicValue,
6807                              Sema::TypeTagData> *MagicValues,
6808         bool &FoundWrongKind,
6809         Sema::TypeTagData &TypeInfo) {
6810   FoundWrongKind = false;
6811 
6812   // Variable declaration that has type_tag_for_datatype attribute.
6813   const ValueDecl *VD = NULL;
6814 
6815   uint64_t MagicValue;
6816 
6817   if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue))
6818     return false;
6819 
6820   if (VD) {
6821     for (specific_attr_iterator<TypeTagForDatatypeAttr>
6822              I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(),
6823              E = VD->specific_attr_end<TypeTagForDatatypeAttr>();
6824          I != E; ++I) {
6825       if (I->getArgumentKind() != ArgumentKind) {
6826         FoundWrongKind = true;
6827         return false;
6828       }
6829       TypeInfo.Type = I->getMatchingCType();
6830       TypeInfo.LayoutCompatible = I->getLayoutCompatible();
6831       TypeInfo.MustBeNull = I->getMustBeNull();
6832       return true;
6833     }
6834     return false;
6835   }
6836 
6837   if (!MagicValues)
6838     return false;
6839 
6840   llvm::DenseMap<Sema::TypeTagMagicValue,
6841                  Sema::TypeTagData>::const_iterator I =
6842       MagicValues->find(std::make_pair(ArgumentKind, MagicValue));
6843   if (I == MagicValues->end())
6844     return false;
6845 
6846   TypeInfo = I->second;
6847   return true;
6848 }
6849 } // unnamed namespace
6850 
6851 void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind,
6852                                       uint64_t MagicValue, QualType Type,
6853                                       bool LayoutCompatible,
6854                                       bool MustBeNull) {
6855   if (!TypeTagForDatatypeMagicValues)
6856     TypeTagForDatatypeMagicValues.reset(
6857         new llvm::DenseMap<TypeTagMagicValue, TypeTagData>);
6858 
6859   TypeTagMagicValue Magic(ArgumentKind, MagicValue);
6860   (*TypeTagForDatatypeMagicValues)[Magic] =
6861       TypeTagData(Type, LayoutCompatible, MustBeNull);
6862 }
6863 
6864 namespace {
6865 bool IsSameCharType(QualType T1, QualType T2) {
6866   const BuiltinType *BT1 = T1->getAs<BuiltinType>();
6867   if (!BT1)
6868     return false;
6869 
6870   const BuiltinType *BT2 = T2->getAs<BuiltinType>();
6871   if (!BT2)
6872     return false;
6873 
6874   BuiltinType::Kind T1Kind = BT1->getKind();
6875   BuiltinType::Kind T2Kind = BT2->getKind();
6876 
6877   return (T1Kind == BuiltinType::SChar  && T2Kind == BuiltinType::Char_S) ||
6878          (T1Kind == BuiltinType::UChar  && T2Kind == BuiltinType::Char_U) ||
6879          (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) ||
6880          (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar);
6881 }
6882 } // unnamed namespace
6883 
6884 void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr,
6885                                     const Expr * const *ExprArgs) {
6886   const IdentifierInfo *ArgumentKind = Attr->getArgumentKind();
6887   bool IsPointerAttr = Attr->getIsPointer();
6888 
6889   const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()];
6890   bool FoundWrongKind;
6891   TypeTagData TypeInfo;
6892   if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context,
6893                         TypeTagForDatatypeMagicValues.get(),
6894                         FoundWrongKind, TypeInfo)) {
6895     if (FoundWrongKind)
6896       Diag(TypeTagExpr->getExprLoc(),
6897            diag::warn_type_tag_for_datatype_wrong_kind)
6898         << TypeTagExpr->getSourceRange();
6899     return;
6900   }
6901 
6902   const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()];
6903   if (IsPointerAttr) {
6904     // Skip implicit cast of pointer to `void *' (as a function argument).
6905     if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr))
6906       if (ICE->getType()->isVoidPointerType() &&
6907           ICE->getCastKind() == CK_BitCast)
6908         ArgumentExpr = ICE->getSubExpr();
6909   }
6910   QualType ArgumentType = ArgumentExpr->getType();
6911 
6912   // Passing a `void*' pointer shouldn't trigger a warning.
6913   if (IsPointerAttr && ArgumentType->isVoidPointerType())
6914     return;
6915 
6916   if (TypeInfo.MustBeNull) {
6917     // Type tag with matching void type requires a null pointer.
6918     if (!ArgumentExpr->isNullPointerConstant(Context,
6919                                              Expr::NPC_ValueDependentIsNotNull)) {
6920       Diag(ArgumentExpr->getExprLoc(),
6921            diag::warn_type_safety_null_pointer_required)
6922           << ArgumentKind->getName()
6923           << ArgumentExpr->getSourceRange()
6924           << TypeTagExpr->getSourceRange();
6925     }
6926     return;
6927   }
6928 
6929   QualType RequiredType = TypeInfo.Type;
6930   if (IsPointerAttr)
6931     RequiredType = Context.getPointerType(RequiredType);
6932 
6933   bool mismatch = false;
6934   if (!TypeInfo.LayoutCompatible) {
6935     mismatch = !Context.hasSameType(ArgumentType, RequiredType);
6936 
6937     // C++11 [basic.fundamental] p1:
6938     // Plain char, signed char, and unsigned char are three distinct types.
6939     //
6940     // But we treat plain `char' as equivalent to `signed char' or `unsigned
6941     // char' depending on the current char signedness mode.
6942     if (mismatch)
6943       if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(),
6944                                            RequiredType->getPointeeType())) ||
6945           (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType)))
6946         mismatch = false;
6947   } else
6948     if (IsPointerAttr)
6949       mismatch = !isLayoutCompatible(Context,
6950                                      ArgumentType->getPointeeType(),
6951                                      RequiredType->getPointeeType());
6952     else
6953       mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType);
6954 
6955   if (mismatch)
6956     Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch)
6957         << ArgumentType << ArgumentKind->getName()
6958         << TypeInfo.LayoutCompatible << RequiredType
6959         << ArgumentExpr->getSourceRange()
6960         << TypeTagExpr->getSourceRange();
6961 }
6962