SemaChecking.cpp revision 34ff062936ac78142a2c0dc8d1ae6a40df456819
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10//  This file implements extra semantic analysis beyond what is enforced
11//  by the C type system.
12//
13//===----------------------------------------------------------------------===//
14
15#include "clang/Sema/Initialization.h"
16#include "clang/Sema/Sema.h"
17#include "clang/Sema/SemaInternal.h"
18#include "clang/Sema/Initialization.h"
19#include "clang/Sema/ScopeInfo.h"
20#include "clang/Analysis/Analyses/FormatString.h"
21#include "clang/AST/ASTContext.h"
22#include "clang/AST/CharUnits.h"
23#include "clang/AST/DeclCXX.h"
24#include "clang/AST/DeclObjC.h"
25#include "clang/AST/ExprCXX.h"
26#include "clang/AST/ExprObjC.h"
27#include "clang/AST/EvaluatedExprVisitor.h"
28#include "clang/AST/DeclObjC.h"
29#include "clang/AST/StmtCXX.h"
30#include "clang/AST/StmtObjC.h"
31#include "clang/Lex/Preprocessor.h"
32#include "llvm/ADT/BitVector.h"
33#include "llvm/ADT/STLExtras.h"
34#include "llvm/Support/raw_ostream.h"
35#include "clang/Basic/TargetBuiltins.h"
36#include "clang/Basic/TargetInfo.h"
37#include "clang/Basic/ConvertUTF.h"
38#include <limits>
39using namespace clang;
40using namespace sema;
41
42SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL,
43                                                    unsigned ByteNo) const {
44  return SL->getLocationOfByte(ByteNo, PP.getSourceManager(),
45                               PP.getLangOptions(), PP.getTargetInfo());
46}
47
48/// Checks that a call expression's argument count is the desired number.
49/// This is useful when doing custom type-checking.  Returns true on error.
50static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) {
51  unsigned argCount = call->getNumArgs();
52  if (argCount == desiredArgCount) return false;
53
54  if (argCount < desiredArgCount)
55    return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args)
56        << 0 /*function call*/ << desiredArgCount << argCount
57        << call->getSourceRange();
58
59  // Highlight all the excess arguments.
60  SourceRange range(call->getArg(desiredArgCount)->getLocStart(),
61                    call->getArg(argCount - 1)->getLocEnd());
62
63  return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args)
64    << 0 /*function call*/ << desiredArgCount << argCount
65    << call->getArg(1)->getSourceRange();
66}
67
68/// CheckBuiltinAnnotationString - Checks that string argument to the builtin
69/// annotation is a non wide string literal.
70static bool CheckBuiltinAnnotationString(Sema &S, Expr *Arg) {
71  Arg = Arg->IgnoreParenCasts();
72  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
73  if (!Literal || !Literal->isAscii()) {
74    S.Diag(Arg->getLocStart(), diag::err_builtin_annotation_not_string_constant)
75      << Arg->getSourceRange();
76    return true;
77  }
78  return false;
79}
80
81ExprResult
82Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
83  ExprResult TheCallResult(Owned(TheCall));
84
85  // Find out if any arguments are required to be integer constant expressions.
86  unsigned ICEArguments = 0;
87  ASTContext::GetBuiltinTypeError Error;
88  Context.GetBuiltinType(BuiltinID, Error, &ICEArguments);
89  if (Error != ASTContext::GE_None)
90    ICEArguments = 0;  // Don't diagnose previously diagnosed errors.
91
92  // If any arguments are required to be ICE's, check and diagnose.
93  for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) {
94    // Skip arguments not required to be ICE's.
95    if ((ICEArguments & (1 << ArgNo)) == 0) continue;
96
97    llvm::APSInt Result;
98    if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
99      return true;
100    ICEArguments &= ~(1 << ArgNo);
101  }
102
103  switch (BuiltinID) {
104  case Builtin::BI__builtin___CFStringMakeConstantString:
105    assert(TheCall->getNumArgs() == 1 &&
106           "Wrong # arguments to builtin CFStringMakeConstantString");
107    if (CheckObjCString(TheCall->getArg(0)))
108      return ExprError();
109    break;
110  case Builtin::BI__builtin_stdarg_start:
111  case Builtin::BI__builtin_va_start:
112    if (SemaBuiltinVAStart(TheCall))
113      return ExprError();
114    break;
115  case Builtin::BI__builtin_isgreater:
116  case Builtin::BI__builtin_isgreaterequal:
117  case Builtin::BI__builtin_isless:
118  case Builtin::BI__builtin_islessequal:
119  case Builtin::BI__builtin_islessgreater:
120  case Builtin::BI__builtin_isunordered:
121    if (SemaBuiltinUnorderedCompare(TheCall))
122      return ExprError();
123    break;
124  case Builtin::BI__builtin_fpclassify:
125    if (SemaBuiltinFPClassification(TheCall, 6))
126      return ExprError();
127    break;
128  case Builtin::BI__builtin_isfinite:
129  case Builtin::BI__builtin_isinf:
130  case Builtin::BI__builtin_isinf_sign:
131  case Builtin::BI__builtin_isnan:
132  case Builtin::BI__builtin_isnormal:
133    if (SemaBuiltinFPClassification(TheCall, 1))
134      return ExprError();
135    break;
136  case Builtin::BI__builtin_shufflevector:
137    return SemaBuiltinShuffleVector(TheCall);
138    // TheCall will be freed by the smart pointer here, but that's fine, since
139    // SemaBuiltinShuffleVector guts it, but then doesn't release it.
140  case Builtin::BI__builtin_prefetch:
141    if (SemaBuiltinPrefetch(TheCall))
142      return ExprError();
143    break;
144  case Builtin::BI__builtin_object_size:
145    if (SemaBuiltinObjectSize(TheCall))
146      return ExprError();
147    break;
148  case Builtin::BI__builtin_longjmp:
149    if (SemaBuiltinLongjmp(TheCall))
150      return ExprError();
151    break;
152
153  case Builtin::BI__builtin_classify_type:
154    if (checkArgCount(*this, TheCall, 1)) return true;
155    TheCall->setType(Context.IntTy);
156    break;
157  case Builtin::BI__builtin_constant_p:
158    if (checkArgCount(*this, TheCall, 1)) return true;
159    TheCall->setType(Context.IntTy);
160    break;
161  case Builtin::BI__sync_fetch_and_add:
162  case Builtin::BI__sync_fetch_and_add_1:
163  case Builtin::BI__sync_fetch_and_add_2:
164  case Builtin::BI__sync_fetch_and_add_4:
165  case Builtin::BI__sync_fetch_and_add_8:
166  case Builtin::BI__sync_fetch_and_add_16:
167  case Builtin::BI__sync_fetch_and_sub:
168  case Builtin::BI__sync_fetch_and_sub_1:
169  case Builtin::BI__sync_fetch_and_sub_2:
170  case Builtin::BI__sync_fetch_and_sub_4:
171  case Builtin::BI__sync_fetch_and_sub_8:
172  case Builtin::BI__sync_fetch_and_sub_16:
173  case Builtin::BI__sync_fetch_and_or:
174  case Builtin::BI__sync_fetch_and_or_1:
175  case Builtin::BI__sync_fetch_and_or_2:
176  case Builtin::BI__sync_fetch_and_or_4:
177  case Builtin::BI__sync_fetch_and_or_8:
178  case Builtin::BI__sync_fetch_and_or_16:
179  case Builtin::BI__sync_fetch_and_and:
180  case Builtin::BI__sync_fetch_and_and_1:
181  case Builtin::BI__sync_fetch_and_and_2:
182  case Builtin::BI__sync_fetch_and_and_4:
183  case Builtin::BI__sync_fetch_and_and_8:
184  case Builtin::BI__sync_fetch_and_and_16:
185  case Builtin::BI__sync_fetch_and_xor:
186  case Builtin::BI__sync_fetch_and_xor_1:
187  case Builtin::BI__sync_fetch_and_xor_2:
188  case Builtin::BI__sync_fetch_and_xor_4:
189  case Builtin::BI__sync_fetch_and_xor_8:
190  case Builtin::BI__sync_fetch_and_xor_16:
191  case Builtin::BI__sync_add_and_fetch:
192  case Builtin::BI__sync_add_and_fetch_1:
193  case Builtin::BI__sync_add_and_fetch_2:
194  case Builtin::BI__sync_add_and_fetch_4:
195  case Builtin::BI__sync_add_and_fetch_8:
196  case Builtin::BI__sync_add_and_fetch_16:
197  case Builtin::BI__sync_sub_and_fetch:
198  case Builtin::BI__sync_sub_and_fetch_1:
199  case Builtin::BI__sync_sub_and_fetch_2:
200  case Builtin::BI__sync_sub_and_fetch_4:
201  case Builtin::BI__sync_sub_and_fetch_8:
202  case Builtin::BI__sync_sub_and_fetch_16:
203  case Builtin::BI__sync_and_and_fetch:
204  case Builtin::BI__sync_and_and_fetch_1:
205  case Builtin::BI__sync_and_and_fetch_2:
206  case Builtin::BI__sync_and_and_fetch_4:
207  case Builtin::BI__sync_and_and_fetch_8:
208  case Builtin::BI__sync_and_and_fetch_16:
209  case Builtin::BI__sync_or_and_fetch:
210  case Builtin::BI__sync_or_and_fetch_1:
211  case Builtin::BI__sync_or_and_fetch_2:
212  case Builtin::BI__sync_or_and_fetch_4:
213  case Builtin::BI__sync_or_and_fetch_8:
214  case Builtin::BI__sync_or_and_fetch_16:
215  case Builtin::BI__sync_xor_and_fetch:
216  case Builtin::BI__sync_xor_and_fetch_1:
217  case Builtin::BI__sync_xor_and_fetch_2:
218  case Builtin::BI__sync_xor_and_fetch_4:
219  case Builtin::BI__sync_xor_and_fetch_8:
220  case Builtin::BI__sync_xor_and_fetch_16:
221  case Builtin::BI__sync_val_compare_and_swap:
222  case Builtin::BI__sync_val_compare_and_swap_1:
223  case Builtin::BI__sync_val_compare_and_swap_2:
224  case Builtin::BI__sync_val_compare_and_swap_4:
225  case Builtin::BI__sync_val_compare_and_swap_8:
226  case Builtin::BI__sync_val_compare_and_swap_16:
227  case Builtin::BI__sync_bool_compare_and_swap:
228  case Builtin::BI__sync_bool_compare_and_swap_1:
229  case Builtin::BI__sync_bool_compare_and_swap_2:
230  case Builtin::BI__sync_bool_compare_and_swap_4:
231  case Builtin::BI__sync_bool_compare_and_swap_8:
232  case Builtin::BI__sync_bool_compare_and_swap_16:
233  case Builtin::BI__sync_lock_test_and_set:
234  case Builtin::BI__sync_lock_test_and_set_1:
235  case Builtin::BI__sync_lock_test_and_set_2:
236  case Builtin::BI__sync_lock_test_and_set_4:
237  case Builtin::BI__sync_lock_test_and_set_8:
238  case Builtin::BI__sync_lock_test_and_set_16:
239  case Builtin::BI__sync_lock_release:
240  case Builtin::BI__sync_lock_release_1:
241  case Builtin::BI__sync_lock_release_2:
242  case Builtin::BI__sync_lock_release_4:
243  case Builtin::BI__sync_lock_release_8:
244  case Builtin::BI__sync_lock_release_16:
245  case Builtin::BI__sync_swap:
246  case Builtin::BI__sync_swap_1:
247  case Builtin::BI__sync_swap_2:
248  case Builtin::BI__sync_swap_4:
249  case Builtin::BI__sync_swap_8:
250  case Builtin::BI__sync_swap_16:
251    return SemaBuiltinAtomicOverloaded(move(TheCallResult));
252  case Builtin::BI__atomic_load:
253    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Load);
254  case Builtin::BI__atomic_store:
255    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Store);
256  case Builtin::BI__atomic_init:
257    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Init);
258  case Builtin::BI__atomic_exchange:
259    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xchg);
260  case Builtin::BI__atomic_compare_exchange_strong:
261    return SemaAtomicOpsOverloaded(move(TheCallResult),
262                                   AtomicExpr::CmpXchgStrong);
263  case Builtin::BI__atomic_compare_exchange_weak:
264    return SemaAtomicOpsOverloaded(move(TheCallResult),
265                                   AtomicExpr::CmpXchgWeak);
266  case Builtin::BI__atomic_fetch_add:
267    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Add);
268  case Builtin::BI__atomic_fetch_sub:
269    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Sub);
270  case Builtin::BI__atomic_fetch_and:
271    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::And);
272  case Builtin::BI__atomic_fetch_or:
273    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Or);
274  case Builtin::BI__atomic_fetch_xor:
275    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xor);
276  case Builtin::BI__builtin_annotation:
277    if (CheckBuiltinAnnotationString(*this, TheCall->getArg(1)))
278      return ExprError();
279    break;
280  }
281
282  // Since the target specific builtins for each arch overlap, only check those
283  // of the arch we are compiling for.
284  if (BuiltinID >= Builtin::FirstTSBuiltin) {
285    switch (Context.getTargetInfo().getTriple().getArch()) {
286      case llvm::Triple::arm:
287      case llvm::Triple::thumb:
288        if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall))
289          return ExprError();
290        break;
291      default:
292        break;
293    }
294  }
295
296  return move(TheCallResult);
297}
298
299// Get the valid immediate range for the specified NEON type code.
300static unsigned RFT(unsigned t, bool shift = false) {
301  NeonTypeFlags Type(t);
302  int IsQuad = Type.isQuad();
303  switch (Type.getEltType()) {
304  case NeonTypeFlags::Int8:
305  case NeonTypeFlags::Poly8:
306    return shift ? 7 : (8 << IsQuad) - 1;
307  case NeonTypeFlags::Int16:
308  case NeonTypeFlags::Poly16:
309    return shift ? 15 : (4 << IsQuad) - 1;
310  case NeonTypeFlags::Int32:
311    return shift ? 31 : (2 << IsQuad) - 1;
312  case NeonTypeFlags::Int64:
313    return shift ? 63 : (1 << IsQuad) - 1;
314  case NeonTypeFlags::Float16:
315    assert(!shift && "cannot shift float types!");
316    return (4 << IsQuad) - 1;
317  case NeonTypeFlags::Float32:
318    assert(!shift && "cannot shift float types!");
319    return (2 << IsQuad) - 1;
320  }
321  llvm_unreachable("Invalid NeonTypeFlag!");
322}
323
324/// getNeonEltType - Return the QualType corresponding to the elements of
325/// the vector type specified by the NeonTypeFlags.  This is used to check
326/// the pointer arguments for Neon load/store intrinsics.
327static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) {
328  switch (Flags.getEltType()) {
329  case NeonTypeFlags::Int8:
330    return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy;
331  case NeonTypeFlags::Int16:
332    return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy;
333  case NeonTypeFlags::Int32:
334    return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy;
335  case NeonTypeFlags::Int64:
336    return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy;
337  case NeonTypeFlags::Poly8:
338    return Context.SignedCharTy;
339  case NeonTypeFlags::Poly16:
340    return Context.ShortTy;
341  case NeonTypeFlags::Float16:
342    return Context.UnsignedShortTy;
343  case NeonTypeFlags::Float32:
344    return Context.FloatTy;
345  }
346  llvm_unreachable("Invalid NeonTypeFlag!");
347}
348
349bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
350  llvm::APSInt Result;
351
352  unsigned mask = 0;
353  unsigned TV = 0;
354  int PtrArgNum = -1;
355  bool HasConstPtr = false;
356  switch (BuiltinID) {
357#define GET_NEON_OVERLOAD_CHECK
358#include "clang/Basic/arm_neon.inc"
359#undef GET_NEON_OVERLOAD_CHECK
360  }
361
362  // For NEON intrinsics which are overloaded on vector element type, validate
363  // the immediate which specifies which variant to emit.
364  unsigned ImmArg = TheCall->getNumArgs()-1;
365  if (mask) {
366    if (SemaBuiltinConstantArg(TheCall, ImmArg, Result))
367      return true;
368
369    TV = Result.getLimitedValue(64);
370    if ((TV > 63) || (mask & (1 << TV)) == 0)
371      return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code)
372        << TheCall->getArg(ImmArg)->getSourceRange();
373  }
374
375  if (PtrArgNum >= 0) {
376    // Check that pointer arguments have the specified type.
377    Expr *Arg = TheCall->getArg(PtrArgNum);
378    if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg))
379      Arg = ICE->getSubExpr();
380    ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg);
381    QualType RHSTy = RHS.get()->getType();
382    QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context);
383    if (HasConstPtr)
384      EltTy = EltTy.withConst();
385    QualType LHSTy = Context.getPointerType(EltTy);
386    AssignConvertType ConvTy;
387    ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS);
388    if (RHS.isInvalid())
389      return true;
390    if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy,
391                                 RHS.get(), AA_Assigning))
392      return true;
393  }
394
395  // For NEON intrinsics which take an immediate value as part of the
396  // instruction, range check them here.
397  unsigned i = 0, l = 0, u = 0;
398  switch (BuiltinID) {
399  default: return false;
400  case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break;
401  case ARM::BI__builtin_arm_usat: i = 1; u = 31; break;
402  case ARM::BI__builtin_arm_vcvtr_f:
403  case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break;
404#define GET_NEON_IMMEDIATE_CHECK
405#include "clang/Basic/arm_neon.inc"
406#undef GET_NEON_IMMEDIATE_CHECK
407  };
408
409  // Check that the immediate argument is actually a constant.
410  if (SemaBuiltinConstantArg(TheCall, i, Result))
411    return true;
412
413  // Range check against the upper/lower values for this isntruction.
414  unsigned Val = Result.getZExtValue();
415  if (Val < l || Val > (u + l))
416    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
417      << l << u+l << TheCall->getArg(i)->getSourceRange();
418
419  // FIXME: VFP Intrinsics should error if VFP not present.
420  return false;
421}
422
423/// CheckFunctionCall - Check a direct function call for various correctness
424/// and safety properties not strictly enforced by the C type system.
425bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) {
426  // Get the IdentifierInfo* for the called function.
427  IdentifierInfo *FnInfo = FDecl->getIdentifier();
428
429  // None of the checks below are needed for functions that don't have
430  // simple names (e.g., C++ conversion functions).
431  if (!FnInfo)
432    return false;
433
434  // FIXME: This mechanism should be abstracted to be less fragile and
435  // more efficient. For example, just map function ids to custom
436  // handlers.
437
438  // Printf and scanf checking.
439  for (specific_attr_iterator<FormatAttr>
440         i = FDecl->specific_attr_begin<FormatAttr>(),
441         e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) {
442    CheckFormatArguments(*i, TheCall);
443  }
444
445  for (specific_attr_iterator<NonNullAttr>
446         i = FDecl->specific_attr_begin<NonNullAttr>(),
447         e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) {
448    CheckNonNullArguments(*i, TheCall->getArgs(),
449                          TheCall->getCallee()->getLocStart());
450  }
451
452  unsigned CMId = FDecl->getMemoryFunctionKind();
453  if (CMId == 0)
454    return false;
455
456  // Handle memory setting and copying functions.
457  if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat)
458    CheckStrlcpycatArguments(TheCall, FnInfo);
459  else if (CMId == Builtin::BIstrncat)
460    CheckStrncatArguments(TheCall, FnInfo);
461  else
462    CheckMemaccessArguments(TheCall, CMId, FnInfo);
463
464  return false;
465}
466
467bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac,
468                               Expr **Args, unsigned NumArgs) {
469  for (specific_attr_iterator<FormatAttr>
470       i = Method->specific_attr_begin<FormatAttr>(),
471       e = Method->specific_attr_end<FormatAttr>(); i != e ; ++i) {
472
473    CheckFormatArguments(*i, Args, NumArgs, false, lbrac,
474                         Method->getSourceRange());
475  }
476
477  // diagnose nonnull arguments.
478  for (specific_attr_iterator<NonNullAttr>
479       i = Method->specific_attr_begin<NonNullAttr>(),
480       e = Method->specific_attr_end<NonNullAttr>(); i != e; ++i) {
481    CheckNonNullArguments(*i, Args, lbrac);
482  }
483
484  return false;
485}
486
487bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
488  const VarDecl *V = dyn_cast<VarDecl>(NDecl);
489  if (!V)
490    return false;
491
492  QualType Ty = V->getType();
493  if (!Ty->isBlockPointerType())
494    return false;
495
496  // format string checking.
497  for (specific_attr_iterator<FormatAttr>
498       i = NDecl->specific_attr_begin<FormatAttr>(),
499       e = NDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) {
500    CheckFormatArguments(*i, TheCall);
501  }
502
503  return false;
504}
505
506ExprResult
507Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, AtomicExpr::AtomicOp Op) {
508  CallExpr *TheCall = cast<CallExpr>(TheCallResult.get());
509  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
510
511  // All these operations take one of the following four forms:
512  // T   __atomic_load(_Atomic(T)*, int)                              (loads)
513  // T*  __atomic_add(_Atomic(T*)*, ptrdiff_t, int)         (pointer add/sub)
514  // int __atomic_compare_exchange_strong(_Atomic(T)*, T*, T, int, int)
515  //                                                                (cmpxchg)
516  // T   __atomic_exchange(_Atomic(T)*, T, int)             (everything else)
517  // where T is an appropriate type, and the int paremeterss are for orderings.
518  unsigned NumVals = 1;
519  unsigned NumOrders = 1;
520  if (Op == AtomicExpr::Load) {
521    NumVals = 0;
522  } else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) {
523    NumVals = 2;
524    NumOrders = 2;
525  }
526  if (Op == AtomicExpr::Init)
527    NumOrders = 0;
528
529  if (TheCall->getNumArgs() < NumVals+NumOrders+1) {
530    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
531      << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs()
532      << TheCall->getCallee()->getSourceRange();
533    return ExprError();
534  } else if (TheCall->getNumArgs() > NumVals+NumOrders+1) {
535    Diag(TheCall->getArg(NumVals+NumOrders+1)->getLocStart(),
536         diag::err_typecheck_call_too_many_args)
537      << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs()
538      << TheCall->getCallee()->getSourceRange();
539    return ExprError();
540  }
541
542  // Inspect the first argument of the atomic operation.  This should always be
543  // a pointer to an _Atomic type.
544  Expr *Ptr = TheCall->getArg(0);
545  Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get();
546  const PointerType *pointerType = Ptr->getType()->getAs<PointerType>();
547  if (!pointerType) {
548    Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic)
549      << Ptr->getType() << Ptr->getSourceRange();
550    return ExprError();
551  }
552
553  QualType AtomTy = pointerType->getPointeeType();
554  if (!AtomTy->isAtomicType()) {
555    Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic)
556      << Ptr->getType() << Ptr->getSourceRange();
557    return ExprError();
558  }
559  QualType ValType = AtomTy->getAs<AtomicType>()->getValueType();
560
561  if ((Op == AtomicExpr::Add || Op == AtomicExpr::Sub) &&
562      !ValType->isIntegerType() && !ValType->isPointerType()) {
563    Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr)
564      << Ptr->getType() << Ptr->getSourceRange();
565    return ExprError();
566  }
567
568  if (!ValType->isIntegerType() &&
569      (Op == AtomicExpr::And || Op == AtomicExpr::Or || Op == AtomicExpr::Xor)){
570    Diag(DRE->getLocStart(), diag::err_atomic_op_logical_needs_atomic_int)
571      << Ptr->getType() << Ptr->getSourceRange();
572    return ExprError();
573  }
574
575  switch (ValType.getObjCLifetime()) {
576  case Qualifiers::OCL_None:
577  case Qualifiers::OCL_ExplicitNone:
578    // okay
579    break;
580
581  case Qualifiers::OCL_Weak:
582  case Qualifiers::OCL_Strong:
583  case Qualifiers::OCL_Autoreleasing:
584    Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
585      << ValType << Ptr->getSourceRange();
586    return ExprError();
587  }
588
589  QualType ResultType = ValType;
590  if (Op == AtomicExpr::Store || Op == AtomicExpr::Init)
591    ResultType = Context.VoidTy;
592  else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong)
593    ResultType = Context.BoolTy;
594
595  // The first argument --- the pointer --- has a fixed type; we
596  // deduce the types of the rest of the arguments accordingly.  Walk
597  // the remaining arguments, converting them to the deduced value type.
598  for (unsigned i = 1; i != NumVals+NumOrders+1; ++i) {
599    ExprResult Arg = TheCall->getArg(i);
600    QualType Ty;
601    if (i < NumVals+1) {
602      // The second argument to a cmpxchg is a pointer to the data which will
603      // be exchanged. The second argument to a pointer add/subtract is the
604      // amount to add/subtract, which must be a ptrdiff_t.  The third
605      // argument to a cmpxchg and the second argument in all other cases
606      // is the type of the value.
607      if (i == 1 && (Op == AtomicExpr::CmpXchgWeak ||
608                     Op == AtomicExpr::CmpXchgStrong))
609         Ty = Context.getPointerType(ValType.getUnqualifiedType());
610      else if (!ValType->isIntegerType() &&
611               (Op == AtomicExpr::Add || Op == AtomicExpr::Sub))
612        Ty = Context.getPointerDiffType();
613      else
614        Ty = ValType;
615    } else {
616      // The order(s) are always converted to int.
617      Ty = Context.IntTy;
618    }
619    InitializedEntity Entity =
620        InitializedEntity::InitializeParameter(Context, Ty, false);
621    Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
622    if (Arg.isInvalid())
623      return true;
624    TheCall->setArg(i, Arg.get());
625  }
626
627  SmallVector<Expr*, 5> SubExprs;
628  SubExprs.push_back(Ptr);
629  if (Op == AtomicExpr::Load) {
630    SubExprs.push_back(TheCall->getArg(1)); // Order
631  } else if (Op == AtomicExpr::Init) {
632    SubExprs.push_back(TheCall->getArg(1)); // Val1
633  } else if (Op != AtomicExpr::CmpXchgWeak && Op != AtomicExpr::CmpXchgStrong) {
634    SubExprs.push_back(TheCall->getArg(2)); // Order
635    SubExprs.push_back(TheCall->getArg(1)); // Val1
636  } else {
637    SubExprs.push_back(TheCall->getArg(3)); // Order
638    SubExprs.push_back(TheCall->getArg(1)); // Val1
639    SubExprs.push_back(TheCall->getArg(2)); // Val2
640    SubExprs.push_back(TheCall->getArg(4)); // OrderFail
641  }
642
643  return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(),
644                                        SubExprs.data(), SubExprs.size(),
645                                        ResultType, Op,
646                                        TheCall->getRParenLoc()));
647}
648
649
650/// checkBuiltinArgument - Given a call to a builtin function, perform
651/// normal type-checking on the given argument, updating the call in
652/// place.  This is useful when a builtin function requires custom
653/// type-checking for some of its arguments but not necessarily all of
654/// them.
655///
656/// Returns true on error.
657static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) {
658  FunctionDecl *Fn = E->getDirectCallee();
659  assert(Fn && "builtin call without direct callee!");
660
661  ParmVarDecl *Param = Fn->getParamDecl(ArgIndex);
662  InitializedEntity Entity =
663    InitializedEntity::InitializeParameter(S.Context, Param);
664
665  ExprResult Arg = E->getArg(0);
666  Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg);
667  if (Arg.isInvalid())
668    return true;
669
670  E->setArg(ArgIndex, Arg.take());
671  return false;
672}
673
674/// SemaBuiltinAtomicOverloaded - We have a call to a function like
675/// __sync_fetch_and_add, which is an overloaded function based on the pointer
676/// type of its first argument.  The main ActOnCallExpr routines have already
677/// promoted the types of arguments because all of these calls are prototyped as
678/// void(...).
679///
680/// This function goes through and does final semantic checking for these
681/// builtins,
682ExprResult
683Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
684  CallExpr *TheCall = (CallExpr *)TheCallResult.get();
685  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
686  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
687
688  // Ensure that we have at least one argument to do type inference from.
689  if (TheCall->getNumArgs() < 1) {
690    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
691      << 0 << 1 << TheCall->getNumArgs()
692      << TheCall->getCallee()->getSourceRange();
693    return ExprError();
694  }
695
696  // Inspect the first argument of the atomic builtin.  This should always be
697  // a pointer type, whose element is an integral scalar or pointer type.
698  // Because it is a pointer type, we don't have to worry about any implicit
699  // casts here.
700  // FIXME: We don't allow floating point scalars as input.
701  Expr *FirstArg = TheCall->getArg(0);
702  ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg);
703  if (FirstArgResult.isInvalid())
704    return ExprError();
705  FirstArg = FirstArgResult.take();
706  TheCall->setArg(0, FirstArg);
707
708  const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>();
709  if (!pointerType) {
710    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
711      << FirstArg->getType() << FirstArg->getSourceRange();
712    return ExprError();
713  }
714
715  QualType ValType = pointerType->getPointeeType();
716  if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
717      !ValType->isBlockPointerType()) {
718    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
719      << FirstArg->getType() << FirstArg->getSourceRange();
720    return ExprError();
721  }
722
723  switch (ValType.getObjCLifetime()) {
724  case Qualifiers::OCL_None:
725  case Qualifiers::OCL_ExplicitNone:
726    // okay
727    break;
728
729  case Qualifiers::OCL_Weak:
730  case Qualifiers::OCL_Strong:
731  case Qualifiers::OCL_Autoreleasing:
732    Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
733      << ValType << FirstArg->getSourceRange();
734    return ExprError();
735  }
736
737  // Strip any qualifiers off ValType.
738  ValType = ValType.getUnqualifiedType();
739
740  // The majority of builtins return a value, but a few have special return
741  // types, so allow them to override appropriately below.
742  QualType ResultType = ValType;
743
744  // We need to figure out which concrete builtin this maps onto.  For example,
745  // __sync_fetch_and_add with a 2 byte object turns into
746  // __sync_fetch_and_add_2.
747#define BUILTIN_ROW(x) \
748  { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
749    Builtin::BI##x##_8, Builtin::BI##x##_16 }
750
751  static const unsigned BuiltinIndices[][5] = {
752    BUILTIN_ROW(__sync_fetch_and_add),
753    BUILTIN_ROW(__sync_fetch_and_sub),
754    BUILTIN_ROW(__sync_fetch_and_or),
755    BUILTIN_ROW(__sync_fetch_and_and),
756    BUILTIN_ROW(__sync_fetch_and_xor),
757
758    BUILTIN_ROW(__sync_add_and_fetch),
759    BUILTIN_ROW(__sync_sub_and_fetch),
760    BUILTIN_ROW(__sync_and_and_fetch),
761    BUILTIN_ROW(__sync_or_and_fetch),
762    BUILTIN_ROW(__sync_xor_and_fetch),
763
764    BUILTIN_ROW(__sync_val_compare_and_swap),
765    BUILTIN_ROW(__sync_bool_compare_and_swap),
766    BUILTIN_ROW(__sync_lock_test_and_set),
767    BUILTIN_ROW(__sync_lock_release),
768    BUILTIN_ROW(__sync_swap)
769  };
770#undef BUILTIN_ROW
771
772  // Determine the index of the size.
773  unsigned SizeIndex;
774  switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
775  case 1: SizeIndex = 0; break;
776  case 2: SizeIndex = 1; break;
777  case 4: SizeIndex = 2; break;
778  case 8: SizeIndex = 3; break;
779  case 16: SizeIndex = 4; break;
780  default:
781    Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
782      << FirstArg->getType() << FirstArg->getSourceRange();
783    return ExprError();
784  }
785
786  // Each of these builtins has one pointer argument, followed by some number of
787  // values (0, 1 or 2) followed by a potentially empty varags list of stuff
788  // that we ignore.  Find out which row of BuiltinIndices to read from as well
789  // as the number of fixed args.
790  unsigned BuiltinID = FDecl->getBuiltinID();
791  unsigned BuiltinIndex, NumFixed = 1;
792  switch (BuiltinID) {
793  default: llvm_unreachable("Unknown overloaded atomic builtin!");
794  case Builtin::BI__sync_fetch_and_add:
795  case Builtin::BI__sync_fetch_and_add_1:
796  case Builtin::BI__sync_fetch_and_add_2:
797  case Builtin::BI__sync_fetch_and_add_4:
798  case Builtin::BI__sync_fetch_and_add_8:
799  case Builtin::BI__sync_fetch_and_add_16:
800    BuiltinIndex = 0;
801    break;
802
803  case Builtin::BI__sync_fetch_and_sub:
804  case Builtin::BI__sync_fetch_and_sub_1:
805  case Builtin::BI__sync_fetch_and_sub_2:
806  case Builtin::BI__sync_fetch_and_sub_4:
807  case Builtin::BI__sync_fetch_and_sub_8:
808  case Builtin::BI__sync_fetch_and_sub_16:
809    BuiltinIndex = 1;
810    break;
811
812  case Builtin::BI__sync_fetch_and_or:
813  case Builtin::BI__sync_fetch_and_or_1:
814  case Builtin::BI__sync_fetch_and_or_2:
815  case Builtin::BI__sync_fetch_and_or_4:
816  case Builtin::BI__sync_fetch_and_or_8:
817  case Builtin::BI__sync_fetch_and_or_16:
818    BuiltinIndex = 2;
819    break;
820
821  case Builtin::BI__sync_fetch_and_and:
822  case Builtin::BI__sync_fetch_and_and_1:
823  case Builtin::BI__sync_fetch_and_and_2:
824  case Builtin::BI__sync_fetch_and_and_4:
825  case Builtin::BI__sync_fetch_and_and_8:
826  case Builtin::BI__sync_fetch_and_and_16:
827    BuiltinIndex = 3;
828    break;
829
830  case Builtin::BI__sync_fetch_and_xor:
831  case Builtin::BI__sync_fetch_and_xor_1:
832  case Builtin::BI__sync_fetch_and_xor_2:
833  case Builtin::BI__sync_fetch_and_xor_4:
834  case Builtin::BI__sync_fetch_and_xor_8:
835  case Builtin::BI__sync_fetch_and_xor_16:
836    BuiltinIndex = 4;
837    break;
838
839  case Builtin::BI__sync_add_and_fetch:
840  case Builtin::BI__sync_add_and_fetch_1:
841  case Builtin::BI__sync_add_and_fetch_2:
842  case Builtin::BI__sync_add_and_fetch_4:
843  case Builtin::BI__sync_add_and_fetch_8:
844  case Builtin::BI__sync_add_and_fetch_16:
845    BuiltinIndex = 5;
846    break;
847
848  case Builtin::BI__sync_sub_and_fetch:
849  case Builtin::BI__sync_sub_and_fetch_1:
850  case Builtin::BI__sync_sub_and_fetch_2:
851  case Builtin::BI__sync_sub_and_fetch_4:
852  case Builtin::BI__sync_sub_and_fetch_8:
853  case Builtin::BI__sync_sub_and_fetch_16:
854    BuiltinIndex = 6;
855    break;
856
857  case Builtin::BI__sync_and_and_fetch:
858  case Builtin::BI__sync_and_and_fetch_1:
859  case Builtin::BI__sync_and_and_fetch_2:
860  case Builtin::BI__sync_and_and_fetch_4:
861  case Builtin::BI__sync_and_and_fetch_8:
862  case Builtin::BI__sync_and_and_fetch_16:
863    BuiltinIndex = 7;
864    break;
865
866  case Builtin::BI__sync_or_and_fetch:
867  case Builtin::BI__sync_or_and_fetch_1:
868  case Builtin::BI__sync_or_and_fetch_2:
869  case Builtin::BI__sync_or_and_fetch_4:
870  case Builtin::BI__sync_or_and_fetch_8:
871  case Builtin::BI__sync_or_and_fetch_16:
872    BuiltinIndex = 8;
873    break;
874
875  case Builtin::BI__sync_xor_and_fetch:
876  case Builtin::BI__sync_xor_and_fetch_1:
877  case Builtin::BI__sync_xor_and_fetch_2:
878  case Builtin::BI__sync_xor_and_fetch_4:
879  case Builtin::BI__sync_xor_and_fetch_8:
880  case Builtin::BI__sync_xor_and_fetch_16:
881    BuiltinIndex = 9;
882    break;
883
884  case Builtin::BI__sync_val_compare_and_swap:
885  case Builtin::BI__sync_val_compare_and_swap_1:
886  case Builtin::BI__sync_val_compare_and_swap_2:
887  case Builtin::BI__sync_val_compare_and_swap_4:
888  case Builtin::BI__sync_val_compare_and_swap_8:
889  case Builtin::BI__sync_val_compare_and_swap_16:
890    BuiltinIndex = 10;
891    NumFixed = 2;
892    break;
893
894  case Builtin::BI__sync_bool_compare_and_swap:
895  case Builtin::BI__sync_bool_compare_and_swap_1:
896  case Builtin::BI__sync_bool_compare_and_swap_2:
897  case Builtin::BI__sync_bool_compare_and_swap_4:
898  case Builtin::BI__sync_bool_compare_and_swap_8:
899  case Builtin::BI__sync_bool_compare_and_swap_16:
900    BuiltinIndex = 11;
901    NumFixed = 2;
902    ResultType = Context.BoolTy;
903    break;
904
905  case Builtin::BI__sync_lock_test_and_set:
906  case Builtin::BI__sync_lock_test_and_set_1:
907  case Builtin::BI__sync_lock_test_and_set_2:
908  case Builtin::BI__sync_lock_test_and_set_4:
909  case Builtin::BI__sync_lock_test_and_set_8:
910  case Builtin::BI__sync_lock_test_and_set_16:
911    BuiltinIndex = 12;
912    break;
913
914  case Builtin::BI__sync_lock_release:
915  case Builtin::BI__sync_lock_release_1:
916  case Builtin::BI__sync_lock_release_2:
917  case Builtin::BI__sync_lock_release_4:
918  case Builtin::BI__sync_lock_release_8:
919  case Builtin::BI__sync_lock_release_16:
920    BuiltinIndex = 13;
921    NumFixed = 0;
922    ResultType = Context.VoidTy;
923    break;
924
925  case Builtin::BI__sync_swap:
926  case Builtin::BI__sync_swap_1:
927  case Builtin::BI__sync_swap_2:
928  case Builtin::BI__sync_swap_4:
929  case Builtin::BI__sync_swap_8:
930  case Builtin::BI__sync_swap_16:
931    BuiltinIndex = 14;
932    break;
933  }
934
935  // Now that we know how many fixed arguments we expect, first check that we
936  // have at least that many.
937  if (TheCall->getNumArgs() < 1+NumFixed) {
938    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
939      << 0 << 1+NumFixed << TheCall->getNumArgs()
940      << TheCall->getCallee()->getSourceRange();
941    return ExprError();
942  }
943
944  // Get the decl for the concrete builtin from this, we can tell what the
945  // concrete integer type we should convert to is.
946  unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
947  const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
948  IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
949  FunctionDecl *NewBuiltinDecl =
950    cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
951                                           TUScope, false, DRE->getLocStart()));
952
953  // The first argument --- the pointer --- has a fixed type; we
954  // deduce the types of the rest of the arguments accordingly.  Walk
955  // the remaining arguments, converting them to the deduced value type.
956  for (unsigned i = 0; i != NumFixed; ++i) {
957    ExprResult Arg = TheCall->getArg(i+1);
958
959    // GCC does an implicit conversion to the pointer or integer ValType.  This
960    // can fail in some cases (1i -> int**), check for this error case now.
961    // Initialize the argument.
962    InitializedEntity Entity = InitializedEntity::InitializeParameter(Context,
963                                                   ValType, /*consume*/ false);
964    Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
965    if (Arg.isInvalid())
966      return ExprError();
967
968    // Okay, we have something that *can* be converted to the right type.  Check
969    // to see if there is a potentially weird extension going on here.  This can
970    // happen when you do an atomic operation on something like an char* and
971    // pass in 42.  The 42 gets converted to char.  This is even more strange
972    // for things like 45.123 -> char, etc.
973    // FIXME: Do this check.
974    TheCall->setArg(i+1, Arg.take());
975  }
976
977  ASTContext& Context = this->getASTContext();
978
979  // Create a new DeclRefExpr to refer to the new decl.
980  DeclRefExpr* NewDRE = DeclRefExpr::Create(
981      Context,
982      DRE->getQualifierLoc(),
983      SourceLocation(),
984      NewBuiltinDecl,
985      DRE->getLocation(),
986      NewBuiltinDecl->getType(),
987      DRE->getValueKind());
988
989  // Set the callee in the CallExpr.
990  // FIXME: This leaks the original parens and implicit casts.
991  ExprResult PromotedCall = UsualUnaryConversions(NewDRE);
992  if (PromotedCall.isInvalid())
993    return ExprError();
994  TheCall->setCallee(PromotedCall.take());
995
996  // Change the result type of the call to match the original value type. This
997  // is arbitrary, but the codegen for these builtins ins design to handle it
998  // gracefully.
999  TheCall->setType(ResultType);
1000
1001  return move(TheCallResult);
1002}
1003
1004/// CheckObjCString - Checks that the argument to the builtin
1005/// CFString constructor is correct
1006/// Note: It might also make sense to do the UTF-16 conversion here (would
1007/// simplify the backend).
1008bool Sema::CheckObjCString(Expr *Arg) {
1009  Arg = Arg->IgnoreParenCasts();
1010  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
1011
1012  if (!Literal || !Literal->isAscii()) {
1013    Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
1014      << Arg->getSourceRange();
1015    return true;
1016  }
1017
1018  if (Literal->containsNonAsciiOrNull()) {
1019    StringRef String = Literal->getString();
1020    unsigned NumBytes = String.size();
1021    SmallVector<UTF16, 128> ToBuf(NumBytes);
1022    const UTF8 *FromPtr = (UTF8 *)String.data();
1023    UTF16 *ToPtr = &ToBuf[0];
1024
1025    ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
1026                                                 &ToPtr, ToPtr + NumBytes,
1027                                                 strictConversion);
1028    // Check for conversion failure.
1029    if (Result != conversionOK)
1030      Diag(Arg->getLocStart(),
1031           diag::warn_cfstring_truncated) << Arg->getSourceRange();
1032  }
1033  return false;
1034}
1035
1036/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
1037/// Emit an error and return true on failure, return false on success.
1038bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
1039  Expr *Fn = TheCall->getCallee();
1040  if (TheCall->getNumArgs() > 2) {
1041    Diag(TheCall->getArg(2)->getLocStart(),
1042         diag::err_typecheck_call_too_many_args)
1043      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1044      << Fn->getSourceRange()
1045      << SourceRange(TheCall->getArg(2)->getLocStart(),
1046                     (*(TheCall->arg_end()-1))->getLocEnd());
1047    return true;
1048  }
1049
1050  if (TheCall->getNumArgs() < 2) {
1051    return Diag(TheCall->getLocEnd(),
1052      diag::err_typecheck_call_too_few_args_at_least)
1053      << 0 /*function call*/ << 2 << TheCall->getNumArgs();
1054  }
1055
1056  // Type-check the first argument normally.
1057  if (checkBuiltinArgument(*this, TheCall, 0))
1058    return true;
1059
1060  // Determine whether the current function is variadic or not.
1061  BlockScopeInfo *CurBlock = getCurBlock();
1062  bool isVariadic;
1063  if (CurBlock)
1064    isVariadic = CurBlock->TheDecl->isVariadic();
1065  else if (FunctionDecl *FD = getCurFunctionDecl())
1066    isVariadic = FD->isVariadic();
1067  else
1068    isVariadic = getCurMethodDecl()->isVariadic();
1069
1070  if (!isVariadic) {
1071    Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
1072    return true;
1073  }
1074
1075  // Verify that the second argument to the builtin is the last argument of the
1076  // current function or method.
1077  bool SecondArgIsLastNamedArgument = false;
1078  const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
1079
1080  if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
1081    if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
1082      // FIXME: This isn't correct for methods (results in bogus warning).
1083      // Get the last formal in the current function.
1084      const ParmVarDecl *LastArg;
1085      if (CurBlock)
1086        LastArg = *(CurBlock->TheDecl->param_end()-1);
1087      else if (FunctionDecl *FD = getCurFunctionDecl())
1088        LastArg = *(FD->param_end()-1);
1089      else
1090        LastArg = *(getCurMethodDecl()->param_end()-1);
1091      SecondArgIsLastNamedArgument = PV == LastArg;
1092    }
1093  }
1094
1095  if (!SecondArgIsLastNamedArgument)
1096    Diag(TheCall->getArg(1)->getLocStart(),
1097         diag::warn_second_parameter_of_va_start_not_last_named_argument);
1098  return false;
1099}
1100
1101/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
1102/// friends.  This is declared to take (...), so we have to check everything.
1103bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
1104  if (TheCall->getNumArgs() < 2)
1105    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1106      << 0 << 2 << TheCall->getNumArgs()/*function call*/;
1107  if (TheCall->getNumArgs() > 2)
1108    return Diag(TheCall->getArg(2)->getLocStart(),
1109                diag::err_typecheck_call_too_many_args)
1110      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1111      << SourceRange(TheCall->getArg(2)->getLocStart(),
1112                     (*(TheCall->arg_end()-1))->getLocEnd());
1113
1114  ExprResult OrigArg0 = TheCall->getArg(0);
1115  ExprResult OrigArg1 = TheCall->getArg(1);
1116
1117  // Do standard promotions between the two arguments, returning their common
1118  // type.
1119  QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
1120  if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
1121    return true;
1122
1123  // Make sure any conversions are pushed back into the call; this is
1124  // type safe since unordered compare builtins are declared as "_Bool
1125  // foo(...)".
1126  TheCall->setArg(0, OrigArg0.get());
1127  TheCall->setArg(1, OrigArg1.get());
1128
1129  if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
1130    return false;
1131
1132  // If the common type isn't a real floating type, then the arguments were
1133  // invalid for this operation.
1134  if (!Res->isRealFloatingType())
1135    return Diag(OrigArg0.get()->getLocStart(),
1136                diag::err_typecheck_call_invalid_ordered_compare)
1137      << OrigArg0.get()->getType() << OrigArg1.get()->getType()
1138      << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
1139
1140  return false;
1141}
1142
1143/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
1144/// __builtin_isnan and friends.  This is declared to take (...), so we have
1145/// to check everything. We expect the last argument to be a floating point
1146/// value.
1147bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
1148  if (TheCall->getNumArgs() < NumArgs)
1149    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1150      << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
1151  if (TheCall->getNumArgs() > NumArgs)
1152    return Diag(TheCall->getArg(NumArgs)->getLocStart(),
1153                diag::err_typecheck_call_too_many_args)
1154      << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
1155      << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
1156                     (*(TheCall->arg_end()-1))->getLocEnd());
1157
1158  Expr *OrigArg = TheCall->getArg(NumArgs-1);
1159
1160  if (OrigArg->isTypeDependent())
1161    return false;
1162
1163  // This operation requires a non-_Complex floating-point number.
1164  if (!OrigArg->getType()->isRealFloatingType())
1165    return Diag(OrigArg->getLocStart(),
1166                diag::err_typecheck_call_invalid_unary_fp)
1167      << OrigArg->getType() << OrigArg->getSourceRange();
1168
1169  // If this is an implicit conversion from float -> double, remove it.
1170  if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
1171    Expr *CastArg = Cast->getSubExpr();
1172    if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
1173      assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
1174             "promotion from float to double is the only expected cast here");
1175      Cast->setSubExpr(0);
1176      TheCall->setArg(NumArgs-1, CastArg);
1177      OrigArg = CastArg;
1178    }
1179  }
1180
1181  return false;
1182}
1183
1184/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
1185// This is declared to take (...), so we have to check everything.
1186ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
1187  if (TheCall->getNumArgs() < 2)
1188    return ExprError(Diag(TheCall->getLocEnd(),
1189                          diag::err_typecheck_call_too_few_args_at_least)
1190      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1191      << TheCall->getSourceRange());
1192
1193  // Determine which of the following types of shufflevector we're checking:
1194  // 1) unary, vector mask: (lhs, mask)
1195  // 2) binary, vector mask: (lhs, rhs, mask)
1196  // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
1197  QualType resType = TheCall->getArg(0)->getType();
1198  unsigned numElements = 0;
1199
1200  if (!TheCall->getArg(0)->isTypeDependent() &&
1201      !TheCall->getArg(1)->isTypeDependent()) {
1202    QualType LHSType = TheCall->getArg(0)->getType();
1203    QualType RHSType = TheCall->getArg(1)->getType();
1204
1205    if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
1206      Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
1207        << SourceRange(TheCall->getArg(0)->getLocStart(),
1208                       TheCall->getArg(1)->getLocEnd());
1209      return ExprError();
1210    }
1211
1212    numElements = LHSType->getAs<VectorType>()->getNumElements();
1213    unsigned numResElements = TheCall->getNumArgs() - 2;
1214
1215    // Check to see if we have a call with 2 vector arguments, the unary shuffle
1216    // with mask.  If so, verify that RHS is an integer vector type with the
1217    // same number of elts as lhs.
1218    if (TheCall->getNumArgs() == 2) {
1219      if (!RHSType->hasIntegerRepresentation() ||
1220          RHSType->getAs<VectorType>()->getNumElements() != numElements)
1221        Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1222          << SourceRange(TheCall->getArg(1)->getLocStart(),
1223                         TheCall->getArg(1)->getLocEnd());
1224      numResElements = numElements;
1225    }
1226    else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
1227      Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1228        << SourceRange(TheCall->getArg(0)->getLocStart(),
1229                       TheCall->getArg(1)->getLocEnd());
1230      return ExprError();
1231    } else if (numElements != numResElements) {
1232      QualType eltType = LHSType->getAs<VectorType>()->getElementType();
1233      resType = Context.getVectorType(eltType, numResElements,
1234                                      VectorType::GenericVector);
1235    }
1236  }
1237
1238  for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
1239    if (TheCall->getArg(i)->isTypeDependent() ||
1240        TheCall->getArg(i)->isValueDependent())
1241      continue;
1242
1243    llvm::APSInt Result(32);
1244    if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
1245      return ExprError(Diag(TheCall->getLocStart(),
1246                  diag::err_shufflevector_nonconstant_argument)
1247                << TheCall->getArg(i)->getSourceRange());
1248
1249    if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
1250      return ExprError(Diag(TheCall->getLocStart(),
1251                  diag::err_shufflevector_argument_too_large)
1252               << TheCall->getArg(i)->getSourceRange());
1253  }
1254
1255  SmallVector<Expr*, 32> exprs;
1256
1257  for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
1258    exprs.push_back(TheCall->getArg(i));
1259    TheCall->setArg(i, 0);
1260  }
1261
1262  return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
1263                                            exprs.size(), resType,
1264                                            TheCall->getCallee()->getLocStart(),
1265                                            TheCall->getRParenLoc()));
1266}
1267
1268/// SemaBuiltinPrefetch - Handle __builtin_prefetch.
1269// This is declared to take (const void*, ...) and can take two
1270// optional constant int args.
1271bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
1272  unsigned NumArgs = TheCall->getNumArgs();
1273
1274  if (NumArgs > 3)
1275    return Diag(TheCall->getLocEnd(),
1276             diag::err_typecheck_call_too_many_args_at_most)
1277             << 0 /*function call*/ << 3 << NumArgs
1278             << TheCall->getSourceRange();
1279
1280  // Argument 0 is checked for us and the remaining arguments must be
1281  // constant integers.
1282  for (unsigned i = 1; i != NumArgs; ++i) {
1283    Expr *Arg = TheCall->getArg(i);
1284
1285    llvm::APSInt Result;
1286    if (SemaBuiltinConstantArg(TheCall, i, Result))
1287      return true;
1288
1289    // FIXME: gcc issues a warning and rewrites these to 0. These
1290    // seems especially odd for the third argument since the default
1291    // is 3.
1292    if (i == 1) {
1293      if (Result.getLimitedValue() > 1)
1294        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1295             << "0" << "1" << Arg->getSourceRange();
1296    } else {
1297      if (Result.getLimitedValue() > 3)
1298        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1299            << "0" << "3" << Arg->getSourceRange();
1300    }
1301  }
1302
1303  return false;
1304}
1305
1306/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
1307/// TheCall is a constant expression.
1308bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
1309                                  llvm::APSInt &Result) {
1310  Expr *Arg = TheCall->getArg(ArgNum);
1311  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
1312  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
1313
1314  if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
1315
1316  if (!Arg->isIntegerConstantExpr(Result, Context))
1317    return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
1318                << FDecl->getDeclName() <<  Arg->getSourceRange();
1319
1320  return false;
1321}
1322
1323/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
1324/// int type). This simply type checks that type is one of the defined
1325/// constants (0-3).
1326// For compatibility check 0-3, llvm only handles 0 and 2.
1327bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
1328  llvm::APSInt Result;
1329
1330  // Check constant-ness first.
1331  if (SemaBuiltinConstantArg(TheCall, 1, Result))
1332    return true;
1333
1334  Expr *Arg = TheCall->getArg(1);
1335  if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
1336    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1337             << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1338  }
1339
1340  return false;
1341}
1342
1343/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
1344/// This checks that val is a constant 1.
1345bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
1346  Expr *Arg = TheCall->getArg(1);
1347  llvm::APSInt Result;
1348
1349  // TODO: This is less than ideal. Overload this to take a value.
1350  if (SemaBuiltinConstantArg(TheCall, 1, Result))
1351    return true;
1352
1353  if (Result != 1)
1354    return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
1355             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1356
1357  return false;
1358}
1359
1360// Handle i > 1 ? "x" : "y", recursively.
1361bool Sema::SemaCheckStringLiteral(const Expr *E, Expr **Args,
1362                                  unsigned NumArgs, bool HasVAListArg,
1363                                  unsigned format_idx, unsigned firstDataArg,
1364                                  FormatStringType Type, bool inFunctionCall) {
1365 tryAgain:
1366  if (E->isTypeDependent() || E->isValueDependent())
1367    return false;
1368
1369  E = E->IgnoreParenCasts();
1370
1371  switch (E->getStmtClass()) {
1372  case Stmt::BinaryConditionalOperatorClass:
1373  case Stmt::ConditionalOperatorClass: {
1374    const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E);
1375    return SemaCheckStringLiteral(C->getTrueExpr(), Args, NumArgs, HasVAListArg,
1376                                  format_idx, firstDataArg, Type,
1377                                  inFunctionCall)
1378       && SemaCheckStringLiteral(C->getFalseExpr(), Args, NumArgs, HasVAListArg,
1379                                 format_idx, firstDataArg, Type,
1380                                 inFunctionCall);
1381  }
1382
1383  case Stmt::IntegerLiteralClass:
1384    // Technically -Wformat-nonliteral does not warn about this case.
1385    // The behavior of printf and friends in this case is implementation
1386    // dependent.  Ideally if the format string cannot be null then
1387    // it should have a 'nonnull' attribute in the function prototype.
1388    return true;
1389
1390  case Stmt::ImplicitCastExprClass: {
1391    E = cast<ImplicitCastExpr>(E)->getSubExpr();
1392    goto tryAgain;
1393  }
1394
1395  case Stmt::OpaqueValueExprClass:
1396    if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
1397      E = src;
1398      goto tryAgain;
1399    }
1400    return false;
1401
1402  case Stmt::PredefinedExprClass:
1403    // While __func__, etc., are technically not string literals, they
1404    // cannot contain format specifiers and thus are not a security
1405    // liability.
1406    return true;
1407
1408  case Stmt::DeclRefExprClass: {
1409    const DeclRefExpr *DR = cast<DeclRefExpr>(E);
1410
1411    // As an exception, do not flag errors for variables binding to
1412    // const string literals.
1413    if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
1414      bool isConstant = false;
1415      QualType T = DR->getType();
1416
1417      if (const ArrayType *AT = Context.getAsArrayType(T)) {
1418        isConstant = AT->getElementType().isConstant(Context);
1419      } else if (const PointerType *PT = T->getAs<PointerType>()) {
1420        isConstant = T.isConstant(Context) &&
1421                     PT->getPointeeType().isConstant(Context);
1422      } else if (T->isObjCObjectPointerType()) {
1423        // In ObjC, there is usually no "const ObjectPointer" type,
1424        // so don't check if the pointee type is constant.
1425        isConstant = T.isConstant(Context);
1426      }
1427
1428      if (isConstant) {
1429        if (const Expr *Init = VD->getAnyInitializer())
1430          return SemaCheckStringLiteral(Init, Args, NumArgs,
1431                                        HasVAListArg, format_idx, firstDataArg,
1432                                        Type, /*inFunctionCall*/false);
1433      }
1434
1435      // For vprintf* functions (i.e., HasVAListArg==true), we add a
1436      // special check to see if the format string is a function parameter
1437      // of the function calling the printf function.  If the function
1438      // has an attribute indicating it is a printf-like function, then we
1439      // should suppress warnings concerning non-literals being used in a call
1440      // to a vprintf function.  For example:
1441      //
1442      // void
1443      // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
1444      //      va_list ap;
1445      //      va_start(ap, fmt);
1446      //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
1447      //      ...
1448      //
1449      //
1450      //  FIXME: We don't have full attribute support yet, so just check to see
1451      //    if the argument is a DeclRefExpr that references a parameter.  We'll
1452      //    add proper support for checking the attribute later.
1453      if (HasVAListArg)
1454        if (isa<ParmVarDecl>(VD))
1455          return true;
1456    }
1457
1458    return false;
1459  }
1460
1461  case Stmt::CallExprClass: {
1462    const CallExpr *CE = cast<CallExpr>(E);
1463    if (const ImplicitCastExpr *ICE
1464          = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
1465      if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
1466        if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
1467          if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
1468            unsigned ArgIndex = FA->getFormatIdx();
1469            const Expr *Arg = CE->getArg(ArgIndex - 1);
1470
1471            return SemaCheckStringLiteral(Arg, Args, NumArgs, HasVAListArg,
1472                                          format_idx, firstDataArg, Type,
1473                                          inFunctionCall);
1474          }
1475        }
1476      }
1477    }
1478
1479    return false;
1480  }
1481  case Stmt::ObjCStringLiteralClass:
1482  case Stmt::StringLiteralClass: {
1483    const StringLiteral *StrE = NULL;
1484
1485    if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1486      StrE = ObjCFExpr->getString();
1487    else
1488      StrE = cast<StringLiteral>(E);
1489
1490    if (StrE) {
1491      CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx,
1492                        firstDataArg, Type, inFunctionCall);
1493      return true;
1494    }
1495
1496    return false;
1497  }
1498
1499  default:
1500    return false;
1501  }
1502}
1503
1504void
1505Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1506                            const Expr * const *ExprArgs,
1507                            SourceLocation CallSiteLoc) {
1508  for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1509                                  e = NonNull->args_end();
1510       i != e; ++i) {
1511    const Expr *ArgExpr = ExprArgs[*i];
1512    if (ArgExpr->isNullPointerConstant(Context,
1513                                       Expr::NPC_ValueDependentIsNotNull))
1514      Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1515  }
1516}
1517
1518Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) {
1519  return llvm::StringSwitch<FormatStringType>(Format->getType())
1520  .Case("scanf", FST_Scanf)
1521  .Cases("printf", "printf0", FST_Printf)
1522  .Cases("NSString", "CFString", FST_NSString)
1523  .Case("strftime", FST_Strftime)
1524  .Case("strfmon", FST_Strfmon)
1525  .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf)
1526  .Default(FST_Unknown);
1527}
1528
1529/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar
1530/// functions) for correct use of format strings.
1531void Sema::CheckFormatArguments(const FormatAttr *Format, CallExpr *TheCall) {
1532  bool IsCXXMember = false;
1533  // The way the format attribute works in GCC, the implicit this argument
1534  // of member functions is counted. However, it doesn't appear in our own
1535  // lists, so decrement format_idx in that case.
1536  if (isa<CXXMemberCallExpr>(TheCall)) {
1537    const CXXMethodDecl *method_decl =
1538    dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl());
1539    IsCXXMember = method_decl && method_decl->isInstance();
1540  }
1541  CheckFormatArguments(Format, TheCall->getArgs(), TheCall->getNumArgs(),
1542                       IsCXXMember, TheCall->getRParenLoc(),
1543                       TheCall->getCallee()->getSourceRange());
1544}
1545
1546void Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args,
1547                                unsigned NumArgs, bool IsCXXMember,
1548                                SourceLocation Loc, SourceRange Range) {
1549  bool HasVAListArg = Format->getFirstArg() == 0;
1550  unsigned format_idx = Format->getFormatIdx() - 1;
1551  unsigned firstDataArg = HasVAListArg ? 0 : Format->getFirstArg() - 1;
1552  if (IsCXXMember) {
1553    if (format_idx == 0)
1554      return;
1555    --format_idx;
1556    if(firstDataArg != 0)
1557      --firstDataArg;
1558  }
1559  CheckFormatArguments(Args, NumArgs, HasVAListArg, format_idx,
1560                       firstDataArg, GetFormatStringType(Format), Loc, Range);
1561}
1562
1563void Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs,
1564                                bool HasVAListArg, unsigned format_idx,
1565                                unsigned firstDataArg, FormatStringType Type,
1566                                SourceLocation Loc, SourceRange Range) {
1567  // CHECK: printf/scanf-like function is called with no format string.
1568  if (format_idx >= NumArgs) {
1569    Diag(Loc, diag::warn_missing_format_string) << Range;
1570    return;
1571  }
1572
1573  const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts();
1574
1575  // CHECK: format string is not a string literal.
1576  //
1577  // Dynamically generated format strings are difficult to
1578  // automatically vet at compile time.  Requiring that format strings
1579  // are string literals: (1) permits the checking of format strings by
1580  // the compiler and thereby (2) can practically remove the source of
1581  // many format string exploits.
1582
1583  // Format string can be either ObjC string (e.g. @"%d") or
1584  // C string (e.g. "%d")
1585  // ObjC string uses the same format specifiers as C string, so we can use
1586  // the same format string checking logic for both ObjC and C strings.
1587  if (SemaCheckStringLiteral(OrigFormatExpr, Args, NumArgs, HasVAListArg,
1588                             format_idx, firstDataArg, Type))
1589    return;  // Literal format string found, check done!
1590
1591  // Do not emit diag when the string param is a macro expansion and the
1592  // format is either NSString or CFString. This is a hack to prevent
1593  // diag when using the NSLocalizedString and CFCopyLocalizedString macros
1594  // which are usually used in place of NS and CF string literals.
1595  if (Type == FST_NSString && Args[format_idx]->getLocStart().isMacroID())
1596    return;
1597
1598  // If there are no arguments specified, warn with -Wformat-security, otherwise
1599  // warn only with -Wformat-nonliteral.
1600  if (NumArgs == format_idx+1)
1601    Diag(Args[format_idx]->getLocStart(),
1602         diag::warn_format_nonliteral_noargs)
1603      << OrigFormatExpr->getSourceRange();
1604  else
1605    Diag(Args[format_idx]->getLocStart(),
1606         diag::warn_format_nonliteral)
1607           << OrigFormatExpr->getSourceRange();
1608}
1609
1610namespace {
1611class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1612protected:
1613  Sema &S;
1614  const StringLiteral *FExpr;
1615  const Expr *OrigFormatExpr;
1616  const unsigned FirstDataArg;
1617  const unsigned NumDataArgs;
1618  const bool IsObjCLiteral;
1619  const char *Beg; // Start of format string.
1620  const bool HasVAListArg;
1621  const Expr * const *Args;
1622  const unsigned NumArgs;
1623  unsigned FormatIdx;
1624  llvm::BitVector CoveredArgs;
1625  bool usesPositionalArgs;
1626  bool atFirstArg;
1627  bool inFunctionCall;
1628public:
1629  CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1630                     const Expr *origFormatExpr, unsigned firstDataArg,
1631                     unsigned numDataArgs, bool isObjCLiteral,
1632                     const char *beg, bool hasVAListArg,
1633                     Expr **args, unsigned numArgs,
1634                     unsigned formatIdx, bool inFunctionCall)
1635    : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1636      FirstDataArg(firstDataArg),
1637      NumDataArgs(numDataArgs),
1638      IsObjCLiteral(isObjCLiteral), Beg(beg),
1639      HasVAListArg(hasVAListArg),
1640      Args(args), NumArgs(numArgs), FormatIdx(formatIdx),
1641      usesPositionalArgs(false), atFirstArg(true),
1642      inFunctionCall(inFunctionCall) {
1643        CoveredArgs.resize(numDataArgs);
1644        CoveredArgs.reset();
1645      }
1646
1647  void DoneProcessing();
1648
1649  void HandleIncompleteSpecifier(const char *startSpecifier,
1650                                 unsigned specifierLen);
1651
1652  virtual void HandleInvalidPosition(const char *startSpecifier,
1653                                     unsigned specifierLen,
1654                                     analyze_format_string::PositionContext p);
1655
1656  virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
1657
1658  void HandleNullChar(const char *nullCharacter);
1659
1660  template <typename Range>
1661  static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall,
1662                                   const Expr *ArgumentExpr,
1663                                   PartialDiagnostic PDiag,
1664                                   SourceLocation StringLoc,
1665                                   bool IsStringLocation, Range StringRange,
1666                                   FixItHint Fixit = FixItHint());
1667
1668protected:
1669  bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
1670                                        const char *startSpec,
1671                                        unsigned specifierLen,
1672                                        const char *csStart, unsigned csLen);
1673
1674  void HandlePositionalNonpositionalArgs(SourceLocation Loc,
1675                                         const char *startSpec,
1676                                         unsigned specifierLen);
1677
1678  SourceRange getFormatStringRange();
1679  CharSourceRange getSpecifierRange(const char *startSpecifier,
1680                                    unsigned specifierLen);
1681  SourceLocation getLocationOfByte(const char *x);
1682
1683  const Expr *getDataArg(unsigned i) const;
1684
1685  bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
1686                    const analyze_format_string::ConversionSpecifier &CS,
1687                    const char *startSpecifier, unsigned specifierLen,
1688                    unsigned argIndex);
1689
1690  template <typename Range>
1691  void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc,
1692                            bool IsStringLocation, Range StringRange,
1693                            FixItHint Fixit = FixItHint());
1694
1695  void CheckPositionalAndNonpositionalArgs(
1696      const analyze_format_string::FormatSpecifier *FS);
1697};
1698}
1699
1700SourceRange CheckFormatHandler::getFormatStringRange() {
1701  return OrigFormatExpr->getSourceRange();
1702}
1703
1704CharSourceRange CheckFormatHandler::
1705getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
1706  SourceLocation Start = getLocationOfByte(startSpecifier);
1707  SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
1708
1709  // Advance the end SourceLocation by one due to half-open ranges.
1710  End = End.getLocWithOffset(1);
1711
1712  return CharSourceRange::getCharRange(Start, End);
1713}
1714
1715SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
1716  return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
1717}
1718
1719void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
1720                                                   unsigned specifierLen){
1721  EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier),
1722                       getLocationOfByte(startSpecifier),
1723                       /*IsStringLocation*/true,
1724                       getSpecifierRange(startSpecifier, specifierLen));
1725}
1726
1727void
1728CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
1729                                     analyze_format_string::PositionContext p) {
1730  EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier)
1731                         << (unsigned) p,
1732                       getLocationOfByte(startPos), /*IsStringLocation*/true,
1733                       getSpecifierRange(startPos, posLen));
1734}
1735
1736void CheckFormatHandler::HandleZeroPosition(const char *startPos,
1737                                            unsigned posLen) {
1738  EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier),
1739                               getLocationOfByte(startPos),
1740                               /*IsStringLocation*/true,
1741                               getSpecifierRange(startPos, posLen));
1742}
1743
1744void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
1745  if (!IsObjCLiteral) {
1746    // The presence of a null character is likely an error.
1747    EmitFormatDiagnostic(
1748      S.PDiag(diag::warn_printf_format_string_contains_null_char),
1749      getLocationOfByte(nullCharacter), /*IsStringLocation*/true,
1750      getFormatStringRange());
1751  }
1752}
1753
1754const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
1755  return Args[FirstDataArg + i];
1756}
1757
1758void CheckFormatHandler::DoneProcessing() {
1759    // Does the number of data arguments exceed the number of
1760    // format conversions in the format string?
1761  if (!HasVAListArg) {
1762      // Find any arguments that weren't covered.
1763    CoveredArgs.flip();
1764    signed notCoveredArg = CoveredArgs.find_first();
1765    if (notCoveredArg >= 0) {
1766      assert((unsigned)notCoveredArg < NumDataArgs);
1767      EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used),
1768                           getDataArg((unsigned) notCoveredArg)->getLocStart(),
1769                           /*IsStringLocation*/false, getFormatStringRange());
1770    }
1771  }
1772}
1773
1774bool
1775CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
1776                                                     SourceLocation Loc,
1777                                                     const char *startSpec,
1778                                                     unsigned specifierLen,
1779                                                     const char *csStart,
1780                                                     unsigned csLen) {
1781
1782  bool keepGoing = true;
1783  if (argIndex < NumDataArgs) {
1784    // Consider the argument coverered, even though the specifier doesn't
1785    // make sense.
1786    CoveredArgs.set(argIndex);
1787  }
1788  else {
1789    // If argIndex exceeds the number of data arguments we
1790    // don't issue a warning because that is just a cascade of warnings (and
1791    // they may have intended '%%' anyway). We don't want to continue processing
1792    // the format string after this point, however, as we will like just get
1793    // gibberish when trying to match arguments.
1794    keepGoing = false;
1795  }
1796
1797  EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion)
1798                         << StringRef(csStart, csLen),
1799                       Loc, /*IsStringLocation*/true,
1800                       getSpecifierRange(startSpec, specifierLen));
1801
1802  return keepGoing;
1803}
1804
1805void
1806CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc,
1807                                                      const char *startSpec,
1808                                                      unsigned specifierLen) {
1809  EmitFormatDiagnostic(
1810    S.PDiag(diag::warn_format_mix_positional_nonpositional_args),
1811    Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen));
1812}
1813
1814bool
1815CheckFormatHandler::CheckNumArgs(
1816  const analyze_format_string::FormatSpecifier &FS,
1817  const analyze_format_string::ConversionSpecifier &CS,
1818  const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
1819
1820  if (argIndex >= NumDataArgs) {
1821    PartialDiagnostic PDiag = FS.usesPositionalArg()
1822      ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args)
1823           << (argIndex+1) << NumDataArgs)
1824      : S.PDiag(diag::warn_printf_insufficient_data_args);
1825    EmitFormatDiagnostic(
1826      PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true,
1827      getSpecifierRange(startSpecifier, specifierLen));
1828    return false;
1829  }
1830  return true;
1831}
1832
1833template<typename Range>
1834void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag,
1835                                              SourceLocation Loc,
1836                                              bool IsStringLocation,
1837                                              Range StringRange,
1838                                              FixItHint FixIt) {
1839  EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag,
1840                       Loc, IsStringLocation, StringRange, FixIt);
1841}
1842
1843/// \brief If the format string is not within the funcion call, emit a note
1844/// so that the function call and string are in diagnostic messages.
1845///
1846/// \param inFunctionCall if true, the format string is within the function
1847/// call and only one diagnostic message will be produced.  Otherwise, an
1848/// extra note will be emitted pointing to location of the format string.
1849///
1850/// \param ArgumentExpr the expression that is passed as the format string
1851/// argument in the function call.  Used for getting locations when two
1852/// diagnostics are emitted.
1853///
1854/// \param PDiag the callee should already have provided any strings for the
1855/// diagnostic message.  This function only adds locations and fixits
1856/// to diagnostics.
1857///
1858/// \param Loc primary location for diagnostic.  If two diagnostics are
1859/// required, one will be at Loc and a new SourceLocation will be created for
1860/// the other one.
1861///
1862/// \param IsStringLocation if true, Loc points to the format string should be
1863/// used for the note.  Otherwise, Loc points to the argument list and will
1864/// be used with PDiag.
1865///
1866/// \param StringRange some or all of the string to highlight.  This is
1867/// templated so it can accept either a CharSourceRange or a SourceRange.
1868///
1869/// \param Fixit optional fix it hint for the format string.
1870template<typename Range>
1871void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall,
1872                                              const Expr *ArgumentExpr,
1873                                              PartialDiagnostic PDiag,
1874                                              SourceLocation Loc,
1875                                              bool IsStringLocation,
1876                                              Range StringRange,
1877                                              FixItHint FixIt) {
1878  if (InFunctionCall)
1879    S.Diag(Loc, PDiag) << StringRange << FixIt;
1880  else {
1881    S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag)
1882      << ArgumentExpr->getSourceRange();
1883    S.Diag(IsStringLocation ? Loc : StringRange.getBegin(),
1884           diag::note_format_string_defined)
1885      << StringRange << FixIt;
1886  }
1887}
1888
1889//===--- CHECK: Printf format string checking ------------------------------===//
1890
1891namespace {
1892class CheckPrintfHandler : public CheckFormatHandler {
1893public:
1894  CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
1895                     const Expr *origFormatExpr, unsigned firstDataArg,
1896                     unsigned numDataArgs, bool isObjCLiteral,
1897                     const char *beg, bool hasVAListArg,
1898                     Expr **Args, unsigned NumArgs,
1899                     unsigned formatIdx, bool inFunctionCall)
1900  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1901                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1902                       Args, NumArgs, formatIdx, inFunctionCall) {}
1903
1904
1905  bool HandleInvalidPrintfConversionSpecifier(
1906                                      const analyze_printf::PrintfSpecifier &FS,
1907                                      const char *startSpecifier,
1908                                      unsigned specifierLen);
1909
1910  bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
1911                             const char *startSpecifier,
1912                             unsigned specifierLen);
1913
1914  bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
1915                    const char *startSpecifier, unsigned specifierLen);
1916  void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
1917                           const analyze_printf::OptionalAmount &Amt,
1918                           unsigned type,
1919                           const char *startSpecifier, unsigned specifierLen);
1920  void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1921                  const analyze_printf::OptionalFlag &flag,
1922                  const char *startSpecifier, unsigned specifierLen);
1923  void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
1924                         const analyze_printf::OptionalFlag &ignoredFlag,
1925                         const analyze_printf::OptionalFlag &flag,
1926                         const char *startSpecifier, unsigned specifierLen);
1927};
1928}
1929
1930bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
1931                                      const analyze_printf::PrintfSpecifier &FS,
1932                                      const char *startSpecifier,
1933                                      unsigned specifierLen) {
1934  const analyze_printf::PrintfConversionSpecifier &CS =
1935    FS.getConversionSpecifier();
1936
1937  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1938                                          getLocationOfByte(CS.getStart()),
1939                                          startSpecifier, specifierLen,
1940                                          CS.getStart(), CS.getLength());
1941}
1942
1943bool CheckPrintfHandler::HandleAmount(
1944                               const analyze_format_string::OptionalAmount &Amt,
1945                               unsigned k, const char *startSpecifier,
1946                               unsigned specifierLen) {
1947
1948  if (Amt.hasDataArgument()) {
1949    if (!HasVAListArg) {
1950      unsigned argIndex = Amt.getArgIndex();
1951      if (argIndex >= NumDataArgs) {
1952        EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg)
1953                               << k,
1954                             getLocationOfByte(Amt.getStart()),
1955                             /*IsStringLocation*/true,
1956                             getSpecifierRange(startSpecifier, specifierLen));
1957        // Don't do any more checking.  We will just emit
1958        // spurious errors.
1959        return false;
1960      }
1961
1962      // Type check the data argument.  It should be an 'int'.
1963      // Although not in conformance with C99, we also allow the argument to be
1964      // an 'unsigned int' as that is a reasonably safe case.  GCC also
1965      // doesn't emit a warning for that case.
1966      CoveredArgs.set(argIndex);
1967      const Expr *Arg = getDataArg(argIndex);
1968      QualType T = Arg->getType();
1969
1970      const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context);
1971      assert(ATR.isValid());
1972
1973      if (!ATR.matchesType(S.Context, T)) {
1974        EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type)
1975                               << k << ATR.getRepresentativeTypeName(S.Context)
1976                               << T << Arg->getSourceRange(),
1977                             getLocationOfByte(Amt.getStart()),
1978                             /*IsStringLocation*/true,
1979                             getSpecifierRange(startSpecifier, specifierLen));
1980        // Don't do any more checking.  We will just emit
1981        // spurious errors.
1982        return false;
1983      }
1984    }
1985  }
1986  return true;
1987}
1988
1989void CheckPrintfHandler::HandleInvalidAmount(
1990                                      const analyze_printf::PrintfSpecifier &FS,
1991                                      const analyze_printf::OptionalAmount &Amt,
1992                                      unsigned type,
1993                                      const char *startSpecifier,
1994                                      unsigned specifierLen) {
1995  const analyze_printf::PrintfConversionSpecifier &CS =
1996    FS.getConversionSpecifier();
1997
1998  FixItHint fixit =
1999    Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant
2000      ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
2001                                 Amt.getConstantLength()))
2002      : FixItHint();
2003
2004  EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount)
2005                         << type << CS.toString(),
2006                       getLocationOfByte(Amt.getStart()),
2007                       /*IsStringLocation*/true,
2008                       getSpecifierRange(startSpecifier, specifierLen),
2009                       fixit);
2010}
2011
2012void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
2013                                    const analyze_printf::OptionalFlag &flag,
2014                                    const char *startSpecifier,
2015                                    unsigned specifierLen) {
2016  // Warn about pointless flag with a fixit removal.
2017  const analyze_printf::PrintfConversionSpecifier &CS =
2018    FS.getConversionSpecifier();
2019  EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag)
2020                         << flag.toString() << CS.toString(),
2021                       getLocationOfByte(flag.getPosition()),
2022                       /*IsStringLocation*/true,
2023                       getSpecifierRange(startSpecifier, specifierLen),
2024                       FixItHint::CreateRemoval(
2025                         getSpecifierRange(flag.getPosition(), 1)));
2026}
2027
2028void CheckPrintfHandler::HandleIgnoredFlag(
2029                                const analyze_printf::PrintfSpecifier &FS,
2030                                const analyze_printf::OptionalFlag &ignoredFlag,
2031                                const analyze_printf::OptionalFlag &flag,
2032                                const char *startSpecifier,
2033                                unsigned specifierLen) {
2034  // Warn about ignored flag with a fixit removal.
2035  EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag)
2036                         << ignoredFlag.toString() << flag.toString(),
2037                       getLocationOfByte(ignoredFlag.getPosition()),
2038                       /*IsStringLocation*/true,
2039                       getSpecifierRange(startSpecifier, specifierLen),
2040                       FixItHint::CreateRemoval(
2041                         getSpecifierRange(ignoredFlag.getPosition(), 1)));
2042}
2043
2044bool
2045CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
2046                                            &FS,
2047                                          const char *startSpecifier,
2048                                          unsigned specifierLen) {
2049
2050  using namespace analyze_format_string;
2051  using namespace analyze_printf;
2052  const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
2053
2054  if (FS.consumesDataArgument()) {
2055    if (atFirstArg) {
2056        atFirstArg = false;
2057        usesPositionalArgs = FS.usesPositionalArg();
2058    }
2059    else if (usesPositionalArgs != FS.usesPositionalArg()) {
2060      HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()),
2061                                        startSpecifier, specifierLen);
2062      return false;
2063    }
2064  }
2065
2066  // First check if the field width, precision, and conversion specifier
2067  // have matching data arguments.
2068  if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
2069                    startSpecifier, specifierLen)) {
2070    return false;
2071  }
2072
2073  if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
2074                    startSpecifier, specifierLen)) {
2075    return false;
2076  }
2077
2078  if (!CS.consumesDataArgument()) {
2079    // FIXME: Technically specifying a precision or field width here
2080    // makes no sense.  Worth issuing a warning at some point.
2081    return true;
2082  }
2083
2084  // Consume the argument.
2085  unsigned argIndex = FS.getArgIndex();
2086  if (argIndex < NumDataArgs) {
2087    // The check to see if the argIndex is valid will come later.
2088    // We set the bit here because we may exit early from this
2089    // function if we encounter some other error.
2090    CoveredArgs.set(argIndex);
2091  }
2092
2093  // Check for using an Objective-C specific conversion specifier
2094  // in a non-ObjC literal.
2095  if (!IsObjCLiteral && CS.isObjCArg()) {
2096    return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
2097                                                  specifierLen);
2098  }
2099
2100  // Check for invalid use of field width
2101  if (!FS.hasValidFieldWidth()) {
2102    HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
2103        startSpecifier, specifierLen);
2104  }
2105
2106  // Check for invalid use of precision
2107  if (!FS.hasValidPrecision()) {
2108    HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
2109        startSpecifier, specifierLen);
2110  }
2111
2112  // Check each flag does not conflict with any other component.
2113  if (!FS.hasValidThousandsGroupingPrefix())
2114    HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
2115  if (!FS.hasValidLeadingZeros())
2116    HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
2117  if (!FS.hasValidPlusPrefix())
2118    HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
2119  if (!FS.hasValidSpacePrefix())
2120    HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
2121  if (!FS.hasValidAlternativeForm())
2122    HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
2123  if (!FS.hasValidLeftJustified())
2124    HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
2125
2126  // Check that flags are not ignored by another flag
2127  if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
2128    HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
2129        startSpecifier, specifierLen);
2130  if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
2131    HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
2132            startSpecifier, specifierLen);
2133
2134  // Check the length modifier is valid with the given conversion specifier.
2135  const LengthModifier &LM = FS.getLengthModifier();
2136  if (!FS.hasValidLengthModifier())
2137    EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length)
2138                           << LM.toString() << CS.toString(),
2139                         getLocationOfByte(LM.getStart()),
2140                         /*IsStringLocation*/true,
2141                         getSpecifierRange(startSpecifier, specifierLen),
2142                         FixItHint::CreateRemoval(
2143                           getSpecifierRange(LM.getStart(),
2144                                             LM.getLength())));
2145
2146  // Are we using '%n'?
2147  if (CS.getKind() == ConversionSpecifier::nArg) {
2148    // Issue a warning about this being a possible security issue.
2149    EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back),
2150                         getLocationOfByte(CS.getStart()),
2151                         /*IsStringLocation*/true,
2152                         getSpecifierRange(startSpecifier, specifierLen));
2153    // Continue checking the other format specifiers.
2154    return true;
2155  }
2156
2157  // The remaining checks depend on the data arguments.
2158  if (HasVAListArg)
2159    return true;
2160
2161  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
2162    return false;
2163
2164  // Now type check the data expression that matches the
2165  // format specifier.
2166  const Expr *Ex = getDataArg(argIndex);
2167  const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context,
2168                                                           IsObjCLiteral);
2169  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
2170    // Check if we didn't match because of an implicit cast from a 'char'
2171    // or 'short' to an 'int'.  This is done because printf is a varargs
2172    // function.
2173    if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex))
2174      if (ICE->getType() == S.Context.IntTy) {
2175        // All further checking is done on the subexpression.
2176        Ex = ICE->getSubExpr();
2177        if (ATR.matchesType(S.Context, Ex->getType()))
2178          return true;
2179      }
2180
2181    // We may be able to offer a FixItHint if it is a supported type.
2182    PrintfSpecifier fixedFS = FS;
2183    bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions());
2184
2185    if (success) {
2186      // Get the fix string from the fixed format specifier
2187      llvm::SmallString<128> buf;
2188      llvm::raw_svector_ostream os(buf);
2189      fixedFS.toString(os);
2190
2191      EmitFormatDiagnostic(
2192        S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2193          << ATR.getRepresentativeTypeName(S.Context) << Ex->getType()
2194          << Ex->getSourceRange(),
2195        getLocationOfByte(CS.getStart()),
2196        /*IsStringLocation*/true,
2197        getSpecifierRange(startSpecifier, specifierLen),
2198        FixItHint::CreateReplacement(
2199          getSpecifierRange(startSpecifier, specifierLen),
2200          os.str()));
2201    }
2202    else {
2203      EmitFormatDiagnostic(
2204        S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2205          << ATR.getRepresentativeTypeName(S.Context) << Ex->getType()
2206          << getSpecifierRange(startSpecifier, specifierLen)
2207          << Ex->getSourceRange(),
2208        getLocationOfByte(CS.getStart()),
2209        true,
2210        getSpecifierRange(startSpecifier, specifierLen));
2211    }
2212  }
2213
2214  return true;
2215}
2216
2217//===--- CHECK: Scanf format string checking ------------------------------===//
2218
2219namespace {
2220class CheckScanfHandler : public CheckFormatHandler {
2221public:
2222  CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
2223                    const Expr *origFormatExpr, unsigned firstDataArg,
2224                    unsigned numDataArgs, bool isObjCLiteral,
2225                    const char *beg, bool hasVAListArg,
2226                    Expr **Args, unsigned NumArgs,
2227                    unsigned formatIdx, bool inFunctionCall)
2228  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
2229                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
2230                       Args, NumArgs, formatIdx, inFunctionCall) {}
2231
2232  bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
2233                            const char *startSpecifier,
2234                            unsigned specifierLen);
2235
2236  bool HandleInvalidScanfConversionSpecifier(
2237          const analyze_scanf::ScanfSpecifier &FS,
2238          const char *startSpecifier,
2239          unsigned specifierLen);
2240
2241  void HandleIncompleteScanList(const char *start, const char *end);
2242};
2243}
2244
2245void CheckScanfHandler::HandleIncompleteScanList(const char *start,
2246                                                 const char *end) {
2247  EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete),
2248                       getLocationOfByte(end), /*IsStringLocation*/true,
2249                       getSpecifierRange(start, end - start));
2250}
2251
2252bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
2253                                        const analyze_scanf::ScanfSpecifier &FS,
2254                                        const char *startSpecifier,
2255                                        unsigned specifierLen) {
2256
2257  const analyze_scanf::ScanfConversionSpecifier &CS =
2258    FS.getConversionSpecifier();
2259
2260  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
2261                                          getLocationOfByte(CS.getStart()),
2262                                          startSpecifier, specifierLen,
2263                                          CS.getStart(), CS.getLength());
2264}
2265
2266bool CheckScanfHandler::HandleScanfSpecifier(
2267                                       const analyze_scanf::ScanfSpecifier &FS,
2268                                       const char *startSpecifier,
2269                                       unsigned specifierLen) {
2270
2271  using namespace analyze_scanf;
2272  using namespace analyze_format_string;
2273
2274  const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
2275
2276  // Handle case where '%' and '*' don't consume an argument.  These shouldn't
2277  // be used to decide if we are using positional arguments consistently.
2278  if (FS.consumesDataArgument()) {
2279    if (atFirstArg) {
2280      atFirstArg = false;
2281      usesPositionalArgs = FS.usesPositionalArg();
2282    }
2283    else if (usesPositionalArgs != FS.usesPositionalArg()) {
2284      HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()),
2285                                        startSpecifier, specifierLen);
2286      return false;
2287    }
2288  }
2289
2290  // Check if the field with is non-zero.
2291  const OptionalAmount &Amt = FS.getFieldWidth();
2292  if (Amt.getHowSpecified() == OptionalAmount::Constant) {
2293    if (Amt.getConstantAmount() == 0) {
2294      const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
2295                                                   Amt.getConstantLength());
2296      EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width),
2297                           getLocationOfByte(Amt.getStart()),
2298                           /*IsStringLocation*/true, R,
2299                           FixItHint::CreateRemoval(R));
2300    }
2301  }
2302
2303  if (!FS.consumesDataArgument()) {
2304    // FIXME: Technically specifying a precision or field width here
2305    // makes no sense.  Worth issuing a warning at some point.
2306    return true;
2307  }
2308
2309  // Consume the argument.
2310  unsigned argIndex = FS.getArgIndex();
2311  if (argIndex < NumDataArgs) {
2312      // The check to see if the argIndex is valid will come later.
2313      // We set the bit here because we may exit early from this
2314      // function if we encounter some other error.
2315    CoveredArgs.set(argIndex);
2316  }
2317
2318  // Check the length modifier is valid with the given conversion specifier.
2319  const LengthModifier &LM = FS.getLengthModifier();
2320  if (!FS.hasValidLengthModifier()) {
2321    const CharSourceRange &R = getSpecifierRange(LM.getStart(), LM.getLength());
2322    EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length)
2323                         << LM.toString() << CS.toString()
2324                         << getSpecifierRange(startSpecifier, specifierLen),
2325                         getLocationOfByte(LM.getStart()),
2326                         /*IsStringLocation*/true, R,
2327                         FixItHint::CreateRemoval(R));
2328  }
2329
2330  // The remaining checks depend on the data arguments.
2331  if (HasVAListArg)
2332    return true;
2333
2334  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
2335    return false;
2336
2337  // Check that the argument type matches the format specifier.
2338  const Expr *Ex = getDataArg(argIndex);
2339  const analyze_scanf::ScanfArgTypeResult &ATR = FS.getArgType(S.Context);
2340  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
2341    ScanfSpecifier fixedFS = FS;
2342    bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions());
2343
2344    if (success) {
2345      // Get the fix string from the fixed format specifier.
2346      llvm::SmallString<128> buf;
2347      llvm::raw_svector_ostream os(buf);
2348      fixedFS.toString(os);
2349
2350      EmitFormatDiagnostic(
2351        S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2352          << ATR.getRepresentativeTypeName(S.Context) << Ex->getType()
2353          << Ex->getSourceRange(),
2354        getLocationOfByte(CS.getStart()),
2355        /*IsStringLocation*/true,
2356        getSpecifierRange(startSpecifier, specifierLen),
2357        FixItHint::CreateReplacement(
2358          getSpecifierRange(startSpecifier, specifierLen),
2359          os.str()));
2360    } else {
2361      EmitFormatDiagnostic(
2362        S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2363          << ATR.getRepresentativeTypeName(S.Context) << Ex->getType()
2364          << Ex->getSourceRange(),
2365        getLocationOfByte(CS.getStart()),
2366        /*IsStringLocation*/true,
2367        getSpecifierRange(startSpecifier, specifierLen));
2368    }
2369  }
2370
2371  return true;
2372}
2373
2374void Sema::CheckFormatString(const StringLiteral *FExpr,
2375                             const Expr *OrigFormatExpr,
2376                             Expr **Args, unsigned NumArgs,
2377                             bool HasVAListArg, unsigned format_idx,
2378                             unsigned firstDataArg, FormatStringType Type,
2379                             bool inFunctionCall) {
2380
2381  // CHECK: is the format string a wide literal?
2382  if (!FExpr->isAscii()) {
2383    CheckFormatHandler::EmitFormatDiagnostic(
2384      *this, inFunctionCall, Args[format_idx],
2385      PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(),
2386      /*IsStringLocation*/true, OrigFormatExpr->getSourceRange());
2387    return;
2388  }
2389
2390  // Str - The format string.  NOTE: this is NOT null-terminated!
2391  StringRef StrRef = FExpr->getString();
2392  const char *Str = StrRef.data();
2393  unsigned StrLen = StrRef.size();
2394  const unsigned numDataArgs = NumArgs - firstDataArg;
2395
2396  // CHECK: empty format string?
2397  if (StrLen == 0 && numDataArgs > 0) {
2398    CheckFormatHandler::EmitFormatDiagnostic(
2399      *this, inFunctionCall, Args[format_idx],
2400      PDiag(diag::warn_empty_format_string), FExpr->getLocStart(),
2401      /*IsStringLocation*/true, OrigFormatExpr->getSourceRange());
2402    return;
2403  }
2404
2405  if (Type == FST_Printf || Type == FST_NSString) {
2406    CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
2407                         numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr),
2408                         Str, HasVAListArg, Args, NumArgs, format_idx,
2409                         inFunctionCall);
2410
2411    if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen,
2412                                                  getLangOptions()))
2413      H.DoneProcessing();
2414  } else if (Type == FST_Scanf) {
2415    CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
2416                        numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr),
2417                        Str, HasVAListArg, Args, NumArgs, format_idx,
2418                        inFunctionCall);
2419
2420    if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen,
2421                                                 getLangOptions()))
2422      H.DoneProcessing();
2423  } // TODO: handle other formats
2424}
2425
2426//===--- CHECK: Standard memory functions ---------------------------------===//
2427
2428/// \brief Determine whether the given type is a dynamic class type (e.g.,
2429/// whether it has a vtable).
2430static bool isDynamicClassType(QualType T) {
2431  if (CXXRecordDecl *Record = T->getAsCXXRecordDecl())
2432    if (CXXRecordDecl *Definition = Record->getDefinition())
2433      if (Definition->isDynamicClass())
2434        return true;
2435
2436  return false;
2437}
2438
2439/// \brief If E is a sizeof expression, returns its argument expression,
2440/// otherwise returns NULL.
2441static const Expr *getSizeOfExprArg(const Expr* E) {
2442  if (const UnaryExprOrTypeTraitExpr *SizeOf =
2443      dyn_cast<UnaryExprOrTypeTraitExpr>(E))
2444    if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType())
2445      return SizeOf->getArgumentExpr()->IgnoreParenImpCasts();
2446
2447  return 0;
2448}
2449
2450/// \brief If E is a sizeof expression, returns its argument type.
2451static QualType getSizeOfArgType(const Expr* E) {
2452  if (const UnaryExprOrTypeTraitExpr *SizeOf =
2453      dyn_cast<UnaryExprOrTypeTraitExpr>(E))
2454    if (SizeOf->getKind() == clang::UETT_SizeOf)
2455      return SizeOf->getTypeOfArgument();
2456
2457  return QualType();
2458}
2459
2460/// \brief Check for dangerous or invalid arguments to memset().
2461///
2462/// This issues warnings on known problematic, dangerous or unspecified
2463/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp'
2464/// function calls.
2465///
2466/// \param Call The call expression to diagnose.
2467void Sema::CheckMemaccessArguments(const CallExpr *Call,
2468                                   unsigned BId,
2469                                   IdentifierInfo *FnName) {
2470  assert(BId != 0);
2471
2472  // It is possible to have a non-standard definition of memset.  Validate
2473  // we have enough arguments, and if not, abort further checking.
2474  unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3);
2475  if (Call->getNumArgs() < ExpectedNumArgs)
2476    return;
2477
2478  unsigned LastArg = (BId == Builtin::BImemset ||
2479                      BId == Builtin::BIstrndup ? 1 : 2);
2480  unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2);
2481  const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts();
2482
2483  // We have special checking when the length is a sizeof expression.
2484  QualType SizeOfArgTy = getSizeOfArgType(LenExpr);
2485  const Expr *SizeOfArg = getSizeOfExprArg(LenExpr);
2486  llvm::FoldingSetNodeID SizeOfArgID;
2487
2488  for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) {
2489    const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts();
2490    SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange();
2491
2492    QualType DestTy = Dest->getType();
2493    if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
2494      QualType PointeeTy = DestPtrTy->getPointeeType();
2495
2496      // Never warn about void type pointers. This can be used to suppress
2497      // false positives.
2498      if (PointeeTy->isVoidType())
2499        continue;
2500
2501      // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by
2502      // actually comparing the expressions for equality. Because computing the
2503      // expression IDs can be expensive, we only do this if the diagnostic is
2504      // enabled.
2505      if (SizeOfArg &&
2506          Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess,
2507                                   SizeOfArg->getExprLoc())) {
2508        // We only compute IDs for expressions if the warning is enabled, and
2509        // cache the sizeof arg's ID.
2510        if (SizeOfArgID == llvm::FoldingSetNodeID())
2511          SizeOfArg->Profile(SizeOfArgID, Context, true);
2512        llvm::FoldingSetNodeID DestID;
2513        Dest->Profile(DestID, Context, true);
2514        if (DestID == SizeOfArgID) {
2515          // TODO: For strncpy() and friends, this could suggest sizeof(dst)
2516          //       over sizeof(src) as well.
2517          unsigned ActionIdx = 0; // Default is to suggest dereferencing.
2518          if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest))
2519            if (UnaryOp->getOpcode() == UO_AddrOf)
2520              ActionIdx = 1; // If its an address-of operator, just remove it.
2521          if (Context.getTypeSize(PointeeTy) == Context.getCharWidth())
2522            ActionIdx = 2; // If the pointee's size is sizeof(char),
2523                           // suggest an explicit length.
2524          unsigned DestSrcSelect =
2525            (BId == Builtin::BIstrndup ? 1 : ArgIdx);
2526          DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest,
2527                              PDiag(diag::warn_sizeof_pointer_expr_memaccess)
2528                                << FnName << DestSrcSelect << ActionIdx
2529                                << Dest->getSourceRange()
2530                                << SizeOfArg->getSourceRange());
2531          break;
2532        }
2533      }
2534
2535      // Also check for cases where the sizeof argument is the exact same
2536      // type as the memory argument, and where it points to a user-defined
2537      // record type.
2538      if (SizeOfArgTy != QualType()) {
2539        if (PointeeTy->isRecordType() &&
2540            Context.typesAreCompatible(SizeOfArgTy, DestTy)) {
2541          DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest,
2542                              PDiag(diag::warn_sizeof_pointer_type_memaccess)
2543                                << FnName << SizeOfArgTy << ArgIdx
2544                                << PointeeTy << Dest->getSourceRange()
2545                                << LenExpr->getSourceRange());
2546          break;
2547        }
2548      }
2549
2550      // Always complain about dynamic classes.
2551      if (isDynamicClassType(PointeeTy)) {
2552
2553        unsigned OperationType = 0;
2554        // "overwritten" if we're warning about the destination for any call
2555        // but memcmp; otherwise a verb appropriate to the call.
2556        if (ArgIdx != 0 || BId == Builtin::BImemcmp) {
2557          if (BId == Builtin::BImemcpy)
2558            OperationType = 1;
2559          else if(BId == Builtin::BImemmove)
2560            OperationType = 2;
2561          else if (BId == Builtin::BImemcmp)
2562            OperationType = 3;
2563        }
2564
2565        DiagRuntimeBehavior(
2566          Dest->getExprLoc(), Dest,
2567          PDiag(diag::warn_dyn_class_memaccess)
2568            << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx)
2569            << FnName << PointeeTy
2570            << OperationType
2571            << Call->getCallee()->getSourceRange());
2572      } else if (PointeeTy.hasNonTrivialObjCLifetime() &&
2573               BId != Builtin::BImemset)
2574        DiagRuntimeBehavior(
2575          Dest->getExprLoc(), Dest,
2576          PDiag(diag::warn_arc_object_memaccess)
2577            << ArgIdx << FnName << PointeeTy
2578            << Call->getCallee()->getSourceRange());
2579      else
2580        continue;
2581
2582      DiagRuntimeBehavior(
2583        Dest->getExprLoc(), Dest,
2584        PDiag(diag::note_bad_memaccess_silence)
2585          << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
2586      break;
2587    }
2588  }
2589}
2590
2591// A little helper routine: ignore addition and subtraction of integer literals.
2592// This intentionally does not ignore all integer constant expressions because
2593// we don't want to remove sizeof().
2594static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) {
2595  Ex = Ex->IgnoreParenCasts();
2596
2597  for (;;) {
2598    const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex);
2599    if (!BO || !BO->isAdditiveOp())
2600      break;
2601
2602    const Expr *RHS = BO->getRHS()->IgnoreParenCasts();
2603    const Expr *LHS = BO->getLHS()->IgnoreParenCasts();
2604
2605    if (isa<IntegerLiteral>(RHS))
2606      Ex = LHS;
2607    else if (isa<IntegerLiteral>(LHS))
2608      Ex = RHS;
2609    else
2610      break;
2611  }
2612
2613  return Ex;
2614}
2615
2616// Warn if the user has made the 'size' argument to strlcpy or strlcat
2617// be the size of the source, instead of the destination.
2618void Sema::CheckStrlcpycatArguments(const CallExpr *Call,
2619                                    IdentifierInfo *FnName) {
2620
2621  // Don't crash if the user has the wrong number of arguments
2622  if (Call->getNumArgs() != 3)
2623    return;
2624
2625  const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context);
2626  const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context);
2627  const Expr *CompareWithSrc = NULL;
2628
2629  // Look for 'strlcpy(dst, x, sizeof(x))'
2630  if (const Expr *Ex = getSizeOfExprArg(SizeArg))
2631    CompareWithSrc = Ex;
2632  else {
2633    // Look for 'strlcpy(dst, x, strlen(x))'
2634    if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) {
2635      if (SizeCall->isBuiltinCall() == Builtin::BIstrlen
2636          && SizeCall->getNumArgs() == 1)
2637        CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context);
2638    }
2639  }
2640
2641  if (!CompareWithSrc)
2642    return;
2643
2644  // Determine if the argument to sizeof/strlen is equal to the source
2645  // argument.  In principle there's all kinds of things you could do
2646  // here, for instance creating an == expression and evaluating it with
2647  // EvaluateAsBooleanCondition, but this uses a more direct technique:
2648  const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg);
2649  if (!SrcArgDRE)
2650    return;
2651
2652  const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc);
2653  if (!CompareWithSrcDRE ||
2654      SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl())
2655    return;
2656
2657  const Expr *OriginalSizeArg = Call->getArg(2);
2658  Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size)
2659    << OriginalSizeArg->getSourceRange() << FnName;
2660
2661  // Output a FIXIT hint if the destination is an array (rather than a
2662  // pointer to an array).  This could be enhanced to handle some
2663  // pointers if we know the actual size, like if DstArg is 'array+2'
2664  // we could say 'sizeof(array)-2'.
2665  const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts();
2666  QualType DstArgTy = DstArg->getType();
2667
2668  // Only handle constant-sized or VLAs, but not flexible members.
2669  if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) {
2670    // Only issue the FIXIT for arrays of size > 1.
2671    if (CAT->getSize().getSExtValue() <= 1)
2672      return;
2673  } else if (!DstArgTy->isVariableArrayType()) {
2674    return;
2675  }
2676
2677  llvm::SmallString<128> sizeString;
2678  llvm::raw_svector_ostream OS(sizeString);
2679  OS << "sizeof(";
2680  DstArg->printPretty(OS, Context, 0, getPrintingPolicy());
2681  OS << ")";
2682
2683  Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size)
2684    << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(),
2685                                    OS.str());
2686}
2687
2688/// Check if two expressions refer to the same declaration.
2689static bool referToTheSameDecl(const Expr *E1, const Expr *E2) {
2690  if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1))
2691    if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2))
2692      return D1->getDecl() == D2->getDecl();
2693  return false;
2694}
2695
2696static const Expr *getStrlenExprArg(const Expr *E) {
2697  if (const CallExpr *CE = dyn_cast<CallExpr>(E)) {
2698    const FunctionDecl *FD = CE->getDirectCallee();
2699    if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen)
2700      return 0;
2701    return CE->getArg(0)->IgnoreParenCasts();
2702  }
2703  return 0;
2704}
2705
2706// Warn on anti-patterns as the 'size' argument to strncat.
2707// The correct size argument should look like following:
2708//   strncat(dst, src, sizeof(dst) - strlen(dest) - 1);
2709void Sema::CheckStrncatArguments(const CallExpr *CE,
2710                                 IdentifierInfo *FnName) {
2711  // Don't crash if the user has the wrong number of arguments.
2712  if (CE->getNumArgs() < 3)
2713    return;
2714  const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts();
2715  const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts();
2716  const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts();
2717
2718  // Identify common expressions, which are wrongly used as the size argument
2719  // to strncat and may lead to buffer overflows.
2720  unsigned PatternType = 0;
2721  if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) {
2722    // - sizeof(dst)
2723    if (referToTheSameDecl(SizeOfArg, DstArg))
2724      PatternType = 1;
2725    // - sizeof(src)
2726    else if (referToTheSameDecl(SizeOfArg, SrcArg))
2727      PatternType = 2;
2728  } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) {
2729    if (BE->getOpcode() == BO_Sub) {
2730      const Expr *L = BE->getLHS()->IgnoreParenCasts();
2731      const Expr *R = BE->getRHS()->IgnoreParenCasts();
2732      // - sizeof(dst) - strlen(dst)
2733      if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) &&
2734          referToTheSameDecl(DstArg, getStrlenExprArg(R)))
2735        PatternType = 1;
2736      // - sizeof(src) - (anything)
2737      else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L)))
2738        PatternType = 2;
2739    }
2740  }
2741
2742  if (PatternType == 0)
2743    return;
2744
2745  if (PatternType == 1)
2746    Diag(DstArg->getLocStart(), diag::warn_strncat_large_size)
2747      << LenArg->getSourceRange();
2748  else
2749    Diag(DstArg->getLocStart(), diag::warn_strncat_src_size)
2750      << LenArg->getSourceRange();
2751
2752  // Output a FIXIT hint if the destination is an array (rather than a
2753  // pointer to an array).  This could be enhanced to handle some
2754  // pointers if we know the actual size, like if DstArg is 'array+2'
2755  // we could say 'sizeof(array)-2'.
2756  QualType DstArgTy = DstArg->getType();
2757
2758  // Only handle constant-sized or VLAs, but not flexible members.
2759  if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) {
2760    // Only issue the FIXIT for arrays of size > 1.
2761    if (CAT->getSize().getSExtValue() <= 1)
2762      return;
2763  } else if (!DstArgTy->isVariableArrayType()) {
2764    return;
2765  }
2766
2767  llvm::SmallString<128> sizeString;
2768  llvm::raw_svector_ostream OS(sizeString);
2769  OS << "sizeof(";
2770  DstArg->printPretty(OS, Context, 0, getPrintingPolicy());
2771  OS << ") - ";
2772  OS << "strlen(";
2773  DstArg->printPretty(OS, Context, 0, getPrintingPolicy());
2774  OS << ") - 1";
2775
2776  Diag(LenArg->getLocStart(), diag::note_strncat_wrong_size)
2777    << FixItHint::CreateReplacement(LenArg->getSourceRange(),
2778                                    OS.str());
2779}
2780
2781//===--- CHECK: Return Address of Stack Variable --------------------------===//
2782
2783static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars);
2784static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars);
2785
2786/// CheckReturnStackAddr - Check if a return statement returns the address
2787///   of a stack variable.
2788void
2789Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
2790                           SourceLocation ReturnLoc) {
2791
2792  Expr *stackE = 0;
2793  SmallVector<DeclRefExpr *, 8> refVars;
2794
2795  // Perform checking for returned stack addresses, local blocks,
2796  // label addresses or references to temporaries.
2797  if (lhsType->isPointerType() ||
2798      (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) {
2799    stackE = EvalAddr(RetValExp, refVars);
2800  } else if (lhsType->isReferenceType()) {
2801    stackE = EvalVal(RetValExp, refVars);
2802  }
2803
2804  if (stackE == 0)
2805    return; // Nothing suspicious was found.
2806
2807  SourceLocation diagLoc;
2808  SourceRange diagRange;
2809  if (refVars.empty()) {
2810    diagLoc = stackE->getLocStart();
2811    diagRange = stackE->getSourceRange();
2812  } else {
2813    // We followed through a reference variable. 'stackE' contains the
2814    // problematic expression but we will warn at the return statement pointing
2815    // at the reference variable. We will later display the "trail" of
2816    // reference variables using notes.
2817    diagLoc = refVars[0]->getLocStart();
2818    diagRange = refVars[0]->getSourceRange();
2819  }
2820
2821  if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
2822    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
2823                                             : diag::warn_ret_stack_addr)
2824     << DR->getDecl()->getDeclName() << diagRange;
2825  } else if (isa<BlockExpr>(stackE)) { // local block.
2826    Diag(diagLoc, diag::err_ret_local_block) << diagRange;
2827  } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
2828    Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
2829  } else { // local temporary.
2830    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
2831                                             : diag::warn_ret_local_temp_addr)
2832     << diagRange;
2833  }
2834
2835  // Display the "trail" of reference variables that we followed until we
2836  // found the problematic expression using notes.
2837  for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
2838    VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
2839    // If this var binds to another reference var, show the range of the next
2840    // var, otherwise the var binds to the problematic expression, in which case
2841    // show the range of the expression.
2842    SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
2843                                  : stackE->getSourceRange();
2844    Diag(VD->getLocation(), diag::note_ref_var_local_bind)
2845      << VD->getDeclName() << range;
2846  }
2847}
2848
2849/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
2850///  check if the expression in a return statement evaluates to an address
2851///  to a location on the stack, a local block, an address of a label, or a
2852///  reference to local temporary. The recursion is used to traverse the
2853///  AST of the return expression, with recursion backtracking when we
2854///  encounter a subexpression that (1) clearly does not lead to one of the
2855///  above problematic expressions (2) is something we cannot determine leads to
2856///  a problematic expression based on such local checking.
2857///
2858///  Both EvalAddr and EvalVal follow through reference variables to evaluate
2859///  the expression that they point to. Such variables are added to the
2860///  'refVars' vector so that we know what the reference variable "trail" was.
2861///
2862///  EvalAddr processes expressions that are pointers that are used as
2863///  references (and not L-values).  EvalVal handles all other values.
2864///  At the base case of the recursion is a check for the above problematic
2865///  expressions.
2866///
2867///  This implementation handles:
2868///
2869///   * pointer-to-pointer casts
2870///   * implicit conversions from array references to pointers
2871///   * taking the address of fields
2872///   * arbitrary interplay between "&" and "*" operators
2873///   * pointer arithmetic from an address of a stack variable
2874///   * taking the address of an array element where the array is on the stack
2875static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) {
2876  if (E->isTypeDependent())
2877      return NULL;
2878
2879  // We should only be called for evaluating pointer expressions.
2880  assert((E->getType()->isAnyPointerType() ||
2881          E->getType()->isBlockPointerType() ||
2882          E->getType()->isObjCQualifiedIdType()) &&
2883         "EvalAddr only works on pointers");
2884
2885  E = E->IgnoreParens();
2886
2887  // Our "symbolic interpreter" is just a dispatch off the currently
2888  // viewed AST node.  We then recursively traverse the AST by calling
2889  // EvalAddr and EvalVal appropriately.
2890  switch (E->getStmtClass()) {
2891  case Stmt::DeclRefExprClass: {
2892    DeclRefExpr *DR = cast<DeclRefExpr>(E);
2893
2894    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2895      // If this is a reference variable, follow through to the expression that
2896      // it points to.
2897      if (V->hasLocalStorage() &&
2898          V->getType()->isReferenceType() && V->hasInit()) {
2899        // Add the reference variable to the "trail".
2900        refVars.push_back(DR);
2901        return EvalAddr(V->getInit(), refVars);
2902      }
2903
2904    return NULL;
2905  }
2906
2907  case Stmt::UnaryOperatorClass: {
2908    // The only unary operator that make sense to handle here
2909    // is AddrOf.  All others don't make sense as pointers.
2910    UnaryOperator *U = cast<UnaryOperator>(E);
2911
2912    if (U->getOpcode() == UO_AddrOf)
2913      return EvalVal(U->getSubExpr(), refVars);
2914    else
2915      return NULL;
2916  }
2917
2918  case Stmt::BinaryOperatorClass: {
2919    // Handle pointer arithmetic.  All other binary operators are not valid
2920    // in this context.
2921    BinaryOperator *B = cast<BinaryOperator>(E);
2922    BinaryOperatorKind op = B->getOpcode();
2923
2924    if (op != BO_Add && op != BO_Sub)
2925      return NULL;
2926
2927    Expr *Base = B->getLHS();
2928
2929    // Determine which argument is the real pointer base.  It could be
2930    // the RHS argument instead of the LHS.
2931    if (!Base->getType()->isPointerType()) Base = B->getRHS();
2932
2933    assert (Base->getType()->isPointerType());
2934    return EvalAddr(Base, refVars);
2935  }
2936
2937  // For conditional operators we need to see if either the LHS or RHS are
2938  // valid DeclRefExpr*s.  If one of them is valid, we return it.
2939  case Stmt::ConditionalOperatorClass: {
2940    ConditionalOperator *C = cast<ConditionalOperator>(E);
2941
2942    // Handle the GNU extension for missing LHS.
2943    if (Expr *lhsExpr = C->getLHS()) {
2944    // In C++, we can have a throw-expression, which has 'void' type.
2945      if (!lhsExpr->getType()->isVoidType())
2946        if (Expr* LHS = EvalAddr(lhsExpr, refVars))
2947          return LHS;
2948    }
2949
2950    // In C++, we can have a throw-expression, which has 'void' type.
2951    if (C->getRHS()->getType()->isVoidType())
2952      return NULL;
2953
2954    return EvalAddr(C->getRHS(), refVars);
2955  }
2956
2957  case Stmt::BlockExprClass:
2958    if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
2959      return E; // local block.
2960    return NULL;
2961
2962  case Stmt::AddrLabelExprClass:
2963    return E; // address of label.
2964
2965  case Stmt::ExprWithCleanupsClass:
2966    return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars);
2967
2968  // For casts, we need to handle conversions from arrays to
2969  // pointer values, and pointer-to-pointer conversions.
2970  case Stmt::ImplicitCastExprClass:
2971  case Stmt::CStyleCastExprClass:
2972  case Stmt::CXXFunctionalCastExprClass:
2973  case Stmt::ObjCBridgedCastExprClass: {
2974    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
2975    QualType T = SubExpr->getType();
2976
2977    if (SubExpr->getType()->isPointerType() ||
2978        SubExpr->getType()->isBlockPointerType() ||
2979        SubExpr->getType()->isObjCQualifiedIdType())
2980      return EvalAddr(SubExpr, refVars);
2981    else if (T->isArrayType())
2982      return EvalVal(SubExpr, refVars);
2983    else
2984      return 0;
2985  }
2986
2987  // C++ casts.  For dynamic casts, static casts, and const casts, we
2988  // are always converting from a pointer-to-pointer, so we just blow
2989  // through the cast.  In the case the dynamic cast doesn't fail (and
2990  // return NULL), we take the conservative route and report cases
2991  // where we return the address of a stack variable.  For Reinterpre
2992  // FIXME: The comment about is wrong; we're not always converting
2993  // from pointer to pointer. I'm guessing that this code should also
2994  // handle references to objects.
2995  case Stmt::CXXStaticCastExprClass:
2996  case Stmt::CXXDynamicCastExprClass:
2997  case Stmt::CXXConstCastExprClass:
2998  case Stmt::CXXReinterpretCastExprClass: {
2999      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
3000      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
3001        return EvalAddr(S, refVars);
3002      else
3003        return NULL;
3004  }
3005
3006  case Stmt::MaterializeTemporaryExprClass:
3007    if (Expr *Result = EvalAddr(
3008                         cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
3009                                refVars))
3010      return Result;
3011
3012    return E;
3013
3014  // Everything else: we simply don't reason about them.
3015  default:
3016    return NULL;
3017  }
3018}
3019
3020
3021///  EvalVal - This function is complements EvalAddr in the mutual recursion.
3022///   See the comments for EvalAddr for more details.
3023static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) {
3024do {
3025  // We should only be called for evaluating non-pointer expressions, or
3026  // expressions with a pointer type that are not used as references but instead
3027  // are l-values (e.g., DeclRefExpr with a pointer type).
3028
3029  // Our "symbolic interpreter" is just a dispatch off the currently
3030  // viewed AST node.  We then recursively traverse the AST by calling
3031  // EvalAddr and EvalVal appropriately.
3032
3033  E = E->IgnoreParens();
3034  switch (E->getStmtClass()) {
3035  case Stmt::ImplicitCastExprClass: {
3036    ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
3037    if (IE->getValueKind() == VK_LValue) {
3038      E = IE->getSubExpr();
3039      continue;
3040    }
3041    return NULL;
3042  }
3043
3044  case Stmt::ExprWithCleanupsClass:
3045    return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars);
3046
3047  case Stmt::DeclRefExprClass: {
3048    // When we hit a DeclRefExpr we are looking at code that refers to a
3049    // variable's name. If it's not a reference variable we check if it has
3050    // local storage within the function, and if so, return the expression.
3051    DeclRefExpr *DR = cast<DeclRefExpr>(E);
3052
3053    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
3054      if (V->hasLocalStorage()) {
3055        if (!V->getType()->isReferenceType())
3056          return DR;
3057
3058        // Reference variable, follow through to the expression that
3059        // it points to.
3060        if (V->hasInit()) {
3061          // Add the reference variable to the "trail".
3062          refVars.push_back(DR);
3063          return EvalVal(V->getInit(), refVars);
3064        }
3065      }
3066
3067    return NULL;
3068  }
3069
3070  case Stmt::UnaryOperatorClass: {
3071    // The only unary operator that make sense to handle here
3072    // is Deref.  All others don't resolve to a "name."  This includes
3073    // handling all sorts of rvalues passed to a unary operator.
3074    UnaryOperator *U = cast<UnaryOperator>(E);
3075
3076    if (U->getOpcode() == UO_Deref)
3077      return EvalAddr(U->getSubExpr(), refVars);
3078
3079    return NULL;
3080  }
3081
3082  case Stmt::ArraySubscriptExprClass: {
3083    // Array subscripts are potential references to data on the stack.  We
3084    // retrieve the DeclRefExpr* for the array variable if it indeed
3085    // has local storage.
3086    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars);
3087  }
3088
3089  case Stmt::ConditionalOperatorClass: {
3090    // For conditional operators we need to see if either the LHS or RHS are
3091    // non-NULL Expr's.  If one is non-NULL, we return it.
3092    ConditionalOperator *C = cast<ConditionalOperator>(E);
3093
3094    // Handle the GNU extension for missing LHS.
3095    if (Expr *lhsExpr = C->getLHS())
3096      if (Expr *LHS = EvalVal(lhsExpr, refVars))
3097        return LHS;
3098
3099    return EvalVal(C->getRHS(), refVars);
3100  }
3101
3102  // Accesses to members are potential references to data on the stack.
3103  case Stmt::MemberExprClass: {
3104    MemberExpr *M = cast<MemberExpr>(E);
3105
3106    // Check for indirect access.  We only want direct field accesses.
3107    if (M->isArrow())
3108      return NULL;
3109
3110    // Check whether the member type is itself a reference, in which case
3111    // we're not going to refer to the member, but to what the member refers to.
3112    if (M->getMemberDecl()->getType()->isReferenceType())
3113      return NULL;
3114
3115    return EvalVal(M->getBase(), refVars);
3116  }
3117
3118  case Stmt::MaterializeTemporaryExprClass:
3119    if (Expr *Result = EvalVal(
3120                          cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
3121                               refVars))
3122      return Result;
3123
3124    return E;
3125
3126  default:
3127    // Check that we don't return or take the address of a reference to a
3128    // temporary. This is only useful in C++.
3129    if (!E->isTypeDependent() && E->isRValue())
3130      return E;
3131
3132    // Everything else: we simply don't reason about them.
3133    return NULL;
3134  }
3135} while (true);
3136}
3137
3138//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
3139
3140/// Check for comparisons of floating point operands using != and ==.
3141/// Issue a warning if these are no self-comparisons, as they are not likely
3142/// to do what the programmer intended.
3143void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) {
3144  bool EmitWarning = true;
3145
3146  Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts();
3147  Expr* RightExprSansParen = RHS->IgnoreParenImpCasts();
3148
3149  // Special case: check for x == x (which is OK).
3150  // Do not emit warnings for such cases.
3151  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
3152    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
3153      if (DRL->getDecl() == DRR->getDecl())
3154        EmitWarning = false;
3155
3156
3157  // Special case: check for comparisons against literals that can be exactly
3158  //  represented by APFloat.  In such cases, do not emit a warning.  This
3159  //  is a heuristic: often comparison against such literals are used to
3160  //  detect if a value in a variable has not changed.  This clearly can
3161  //  lead to false negatives.
3162  if (EmitWarning) {
3163    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
3164      if (FLL->isExact())
3165        EmitWarning = false;
3166    } else
3167      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
3168        if (FLR->isExact())
3169          EmitWarning = false;
3170    }
3171  }
3172
3173  // Check for comparisons with builtin types.
3174  if (EmitWarning)
3175    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
3176      if (CL->isBuiltinCall())
3177        EmitWarning = false;
3178
3179  if (EmitWarning)
3180    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
3181      if (CR->isBuiltinCall())
3182        EmitWarning = false;
3183
3184  // Emit the diagnostic.
3185  if (EmitWarning)
3186    Diag(Loc, diag::warn_floatingpoint_eq)
3187      << LHS->getSourceRange() << RHS->getSourceRange();
3188}
3189
3190//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
3191//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
3192
3193namespace {
3194
3195/// Structure recording the 'active' range of an integer-valued
3196/// expression.
3197struct IntRange {
3198  /// The number of bits active in the int.
3199  unsigned Width;
3200
3201  /// True if the int is known not to have negative values.
3202  bool NonNegative;
3203
3204  IntRange(unsigned Width, bool NonNegative)
3205    : Width(Width), NonNegative(NonNegative)
3206  {}
3207
3208  /// Returns the range of the bool type.
3209  static IntRange forBoolType() {
3210    return IntRange(1, true);
3211  }
3212
3213  /// Returns the range of an opaque value of the given integral type.
3214  static IntRange forValueOfType(ASTContext &C, QualType T) {
3215    return forValueOfCanonicalType(C,
3216                          T->getCanonicalTypeInternal().getTypePtr());
3217  }
3218
3219  /// Returns the range of an opaque value of a canonical integral type.
3220  static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
3221    assert(T->isCanonicalUnqualified());
3222
3223    if (const VectorType *VT = dyn_cast<VectorType>(T))
3224      T = VT->getElementType().getTypePtr();
3225    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
3226      T = CT->getElementType().getTypePtr();
3227
3228    // For enum types, use the known bit width of the enumerators.
3229    if (const EnumType *ET = dyn_cast<EnumType>(T)) {
3230      EnumDecl *Enum = ET->getDecl();
3231      if (!Enum->isCompleteDefinition())
3232        return IntRange(C.getIntWidth(QualType(T, 0)), false);
3233
3234      unsigned NumPositive = Enum->getNumPositiveBits();
3235      unsigned NumNegative = Enum->getNumNegativeBits();
3236
3237      return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
3238    }
3239
3240    const BuiltinType *BT = cast<BuiltinType>(T);
3241    assert(BT->isInteger());
3242
3243    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
3244  }
3245
3246  /// Returns the "target" range of a canonical integral type, i.e.
3247  /// the range of values expressible in the type.
3248  ///
3249  /// This matches forValueOfCanonicalType except that enums have the
3250  /// full range of their type, not the range of their enumerators.
3251  static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
3252    assert(T->isCanonicalUnqualified());
3253
3254    if (const VectorType *VT = dyn_cast<VectorType>(T))
3255      T = VT->getElementType().getTypePtr();
3256    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
3257      T = CT->getElementType().getTypePtr();
3258    if (const EnumType *ET = dyn_cast<EnumType>(T))
3259      T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr();
3260
3261    const BuiltinType *BT = cast<BuiltinType>(T);
3262    assert(BT->isInteger());
3263
3264    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
3265  }
3266
3267  /// Returns the supremum of two ranges: i.e. their conservative merge.
3268  static IntRange join(IntRange L, IntRange R) {
3269    return IntRange(std::max(L.Width, R.Width),
3270                    L.NonNegative && R.NonNegative);
3271  }
3272
3273  /// Returns the infinum of two ranges: i.e. their aggressive merge.
3274  static IntRange meet(IntRange L, IntRange R) {
3275    return IntRange(std::min(L.Width, R.Width),
3276                    L.NonNegative || R.NonNegative);
3277  }
3278};
3279
3280static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value,
3281                              unsigned MaxWidth) {
3282  if (value.isSigned() && value.isNegative())
3283    return IntRange(value.getMinSignedBits(), false);
3284
3285  if (value.getBitWidth() > MaxWidth)
3286    value = value.trunc(MaxWidth);
3287
3288  // isNonNegative() just checks the sign bit without considering
3289  // signedness.
3290  return IntRange(value.getActiveBits(), true);
3291}
3292
3293static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
3294                              unsigned MaxWidth) {
3295  if (result.isInt())
3296    return GetValueRange(C, result.getInt(), MaxWidth);
3297
3298  if (result.isVector()) {
3299    IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
3300    for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
3301      IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
3302      R = IntRange::join(R, El);
3303    }
3304    return R;
3305  }
3306
3307  if (result.isComplexInt()) {
3308    IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
3309    IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
3310    return IntRange::join(R, I);
3311  }
3312
3313  // This can happen with lossless casts to intptr_t of "based" lvalues.
3314  // Assume it might use arbitrary bits.
3315  // FIXME: The only reason we need to pass the type in here is to get
3316  // the sign right on this one case.  It would be nice if APValue
3317  // preserved this.
3318  assert(result.isLValue() || result.isAddrLabelDiff());
3319  return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType());
3320}
3321
3322/// Pseudo-evaluate the given integer expression, estimating the
3323/// range of values it might take.
3324///
3325/// \param MaxWidth - the width to which the value will be truncated
3326static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
3327  E = E->IgnoreParens();
3328
3329  // Try a full evaluation first.
3330  Expr::EvalResult result;
3331  if (E->EvaluateAsRValue(result, C))
3332    return GetValueRange(C, result.Val, E->getType(), MaxWidth);
3333
3334  // I think we only want to look through implicit casts here; if the
3335  // user has an explicit widening cast, we should treat the value as
3336  // being of the new, wider type.
3337  if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
3338    if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue)
3339      return GetExprRange(C, CE->getSubExpr(), MaxWidth);
3340
3341    IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
3342
3343    bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
3344
3345    // Assume that non-integer casts can span the full range of the type.
3346    if (!isIntegerCast)
3347      return OutputTypeRange;
3348
3349    IntRange SubRange
3350      = GetExprRange(C, CE->getSubExpr(),
3351                     std::min(MaxWidth, OutputTypeRange.Width));
3352
3353    // Bail out if the subexpr's range is as wide as the cast type.
3354    if (SubRange.Width >= OutputTypeRange.Width)
3355      return OutputTypeRange;
3356
3357    // Otherwise, we take the smaller width, and we're non-negative if
3358    // either the output type or the subexpr is.
3359    return IntRange(SubRange.Width,
3360                    SubRange.NonNegative || OutputTypeRange.NonNegative);
3361  }
3362
3363  if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
3364    // If we can fold the condition, just take that operand.
3365    bool CondResult;
3366    if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
3367      return GetExprRange(C, CondResult ? CO->getTrueExpr()
3368                                        : CO->getFalseExpr(),
3369                          MaxWidth);
3370
3371    // Otherwise, conservatively merge.
3372    IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
3373    IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
3374    return IntRange::join(L, R);
3375  }
3376
3377  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
3378    switch (BO->getOpcode()) {
3379
3380    // Boolean-valued operations are single-bit and positive.
3381    case BO_LAnd:
3382    case BO_LOr:
3383    case BO_LT:
3384    case BO_GT:
3385    case BO_LE:
3386    case BO_GE:
3387    case BO_EQ:
3388    case BO_NE:
3389      return IntRange::forBoolType();
3390
3391    // The type of the assignments is the type of the LHS, so the RHS
3392    // is not necessarily the same type.
3393    case BO_MulAssign:
3394    case BO_DivAssign:
3395    case BO_RemAssign:
3396    case BO_AddAssign:
3397    case BO_SubAssign:
3398    case BO_XorAssign:
3399    case BO_OrAssign:
3400      // TODO: bitfields?
3401      return IntRange::forValueOfType(C, E->getType());
3402
3403    // Simple assignments just pass through the RHS, which will have
3404    // been coerced to the LHS type.
3405    case BO_Assign:
3406      // TODO: bitfields?
3407      return GetExprRange(C, BO->getRHS(), MaxWidth);
3408
3409    // Operations with opaque sources are black-listed.
3410    case BO_PtrMemD:
3411    case BO_PtrMemI:
3412      return IntRange::forValueOfType(C, E->getType());
3413
3414    // Bitwise-and uses the *infinum* of the two source ranges.
3415    case BO_And:
3416    case BO_AndAssign:
3417      return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
3418                            GetExprRange(C, BO->getRHS(), MaxWidth));
3419
3420    // Left shift gets black-listed based on a judgement call.
3421    case BO_Shl:
3422      // ...except that we want to treat '1 << (blah)' as logically
3423      // positive.  It's an important idiom.
3424      if (IntegerLiteral *I
3425            = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
3426        if (I->getValue() == 1) {
3427          IntRange R = IntRange::forValueOfType(C, E->getType());
3428          return IntRange(R.Width, /*NonNegative*/ true);
3429        }
3430      }
3431      // fallthrough
3432
3433    case BO_ShlAssign:
3434      return IntRange::forValueOfType(C, E->getType());
3435
3436    // Right shift by a constant can narrow its left argument.
3437    case BO_Shr:
3438    case BO_ShrAssign: {
3439      IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
3440
3441      // If the shift amount is a positive constant, drop the width by
3442      // that much.
3443      llvm::APSInt shift;
3444      if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
3445          shift.isNonNegative()) {
3446        unsigned zext = shift.getZExtValue();
3447        if (zext >= L.Width)
3448          L.Width = (L.NonNegative ? 0 : 1);
3449        else
3450          L.Width -= zext;
3451      }
3452
3453      return L;
3454    }
3455
3456    // Comma acts as its right operand.
3457    case BO_Comma:
3458      return GetExprRange(C, BO->getRHS(), MaxWidth);
3459
3460    // Black-list pointer subtractions.
3461    case BO_Sub:
3462      if (BO->getLHS()->getType()->isPointerType())
3463        return IntRange::forValueOfType(C, E->getType());
3464      break;
3465
3466    // The width of a division result is mostly determined by the size
3467    // of the LHS.
3468    case BO_Div: {
3469      // Don't 'pre-truncate' the operands.
3470      unsigned opWidth = C.getIntWidth(E->getType());
3471      IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
3472
3473      // If the divisor is constant, use that.
3474      llvm::APSInt divisor;
3475      if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) {
3476        unsigned log2 = divisor.logBase2(); // floor(log_2(divisor))
3477        if (log2 >= L.Width)
3478          L.Width = (L.NonNegative ? 0 : 1);
3479        else
3480          L.Width = std::min(L.Width - log2, MaxWidth);
3481        return L;
3482      }
3483
3484      // Otherwise, just use the LHS's width.
3485      IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
3486      return IntRange(L.Width, L.NonNegative && R.NonNegative);
3487    }
3488
3489    // The result of a remainder can't be larger than the result of
3490    // either side.
3491    case BO_Rem: {
3492      // Don't 'pre-truncate' the operands.
3493      unsigned opWidth = C.getIntWidth(E->getType());
3494      IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
3495      IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
3496
3497      IntRange meet = IntRange::meet(L, R);
3498      meet.Width = std::min(meet.Width, MaxWidth);
3499      return meet;
3500    }
3501
3502    // The default behavior is okay for these.
3503    case BO_Mul:
3504    case BO_Add:
3505    case BO_Xor:
3506    case BO_Or:
3507      break;
3508    }
3509
3510    // The default case is to treat the operation as if it were closed
3511    // on the narrowest type that encompasses both operands.
3512    IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
3513    IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
3514    return IntRange::join(L, R);
3515  }
3516
3517  if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
3518    switch (UO->getOpcode()) {
3519    // Boolean-valued operations are white-listed.
3520    case UO_LNot:
3521      return IntRange::forBoolType();
3522
3523    // Operations with opaque sources are black-listed.
3524    case UO_Deref:
3525    case UO_AddrOf: // should be impossible
3526      return IntRange::forValueOfType(C, E->getType());
3527
3528    default:
3529      return GetExprRange(C, UO->getSubExpr(), MaxWidth);
3530    }
3531  }
3532
3533  if (dyn_cast<OffsetOfExpr>(E)) {
3534    IntRange::forValueOfType(C, E->getType());
3535  }
3536
3537  if (FieldDecl *BitField = E->getBitField())
3538    return IntRange(BitField->getBitWidthValue(C),
3539                    BitField->getType()->isUnsignedIntegerOrEnumerationType());
3540
3541  return IntRange::forValueOfType(C, E->getType());
3542}
3543
3544static IntRange GetExprRange(ASTContext &C, Expr *E) {
3545  return GetExprRange(C, E, C.getIntWidth(E->getType()));
3546}
3547
3548/// Checks whether the given value, which currently has the given
3549/// source semantics, has the same value when coerced through the
3550/// target semantics.
3551static bool IsSameFloatAfterCast(const llvm::APFloat &value,
3552                                 const llvm::fltSemantics &Src,
3553                                 const llvm::fltSemantics &Tgt) {
3554  llvm::APFloat truncated = value;
3555
3556  bool ignored;
3557  truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
3558  truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
3559
3560  return truncated.bitwiseIsEqual(value);
3561}
3562
3563/// Checks whether the given value, which currently has the given
3564/// source semantics, has the same value when coerced through the
3565/// target semantics.
3566///
3567/// The value might be a vector of floats (or a complex number).
3568static bool IsSameFloatAfterCast(const APValue &value,
3569                                 const llvm::fltSemantics &Src,
3570                                 const llvm::fltSemantics &Tgt) {
3571  if (value.isFloat())
3572    return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
3573
3574  if (value.isVector()) {
3575    for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
3576      if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
3577        return false;
3578    return true;
3579  }
3580
3581  assert(value.isComplexFloat());
3582  return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
3583          IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
3584}
3585
3586static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
3587
3588static bool IsZero(Sema &S, Expr *E) {
3589  // Suppress cases where we are comparing against an enum constant.
3590  if (const DeclRefExpr *DR =
3591      dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
3592    if (isa<EnumConstantDecl>(DR->getDecl()))
3593      return false;
3594
3595  // Suppress cases where the '0' value is expanded from a macro.
3596  if (E->getLocStart().isMacroID())
3597    return false;
3598
3599  llvm::APSInt Value;
3600  return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
3601}
3602
3603static bool HasEnumType(Expr *E) {
3604  // Strip off implicit integral promotions.
3605  while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
3606    if (ICE->getCastKind() != CK_IntegralCast &&
3607        ICE->getCastKind() != CK_NoOp)
3608      break;
3609    E = ICE->getSubExpr();
3610  }
3611
3612  return E->getType()->isEnumeralType();
3613}
3614
3615static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
3616  BinaryOperatorKind op = E->getOpcode();
3617  if (E->isValueDependent())
3618    return;
3619
3620  if (op == BO_LT && IsZero(S, E->getRHS())) {
3621    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
3622      << "< 0" << "false" << HasEnumType(E->getLHS())
3623      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3624  } else if (op == BO_GE && IsZero(S, E->getRHS())) {
3625    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
3626      << ">= 0" << "true" << HasEnumType(E->getLHS())
3627      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3628  } else if (op == BO_GT && IsZero(S, E->getLHS())) {
3629    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
3630      << "0 >" << "false" << HasEnumType(E->getRHS())
3631      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3632  } else if (op == BO_LE && IsZero(S, E->getLHS())) {
3633    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
3634      << "0 <=" << "true" << HasEnumType(E->getRHS())
3635      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3636  }
3637}
3638
3639/// Analyze the operands of the given comparison.  Implements the
3640/// fallback case from AnalyzeComparison.
3641static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
3642  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
3643  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
3644}
3645
3646/// \brief Implements -Wsign-compare.
3647///
3648/// \param E the binary operator to check for warnings
3649static void AnalyzeComparison(Sema &S, BinaryOperator *E) {
3650  // The type the comparison is being performed in.
3651  QualType T = E->getLHS()->getType();
3652  assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
3653         && "comparison with mismatched types");
3654
3655  // We don't do anything special if this isn't an unsigned integral
3656  // comparison:  we're only interested in integral comparisons, and
3657  // signed comparisons only happen in cases we don't care to warn about.
3658  //
3659  // We also don't care about value-dependent expressions or expressions
3660  // whose result is a constant.
3661  if (!T->hasUnsignedIntegerRepresentation()
3662      || E->isValueDependent() || E->isIntegerConstantExpr(S.Context))
3663    return AnalyzeImpConvsInComparison(S, E);
3664
3665  Expr *LHS = E->getLHS()->IgnoreParenImpCasts();
3666  Expr *RHS = E->getRHS()->IgnoreParenImpCasts();
3667
3668  // Check to see if one of the (unmodified) operands is of different
3669  // signedness.
3670  Expr *signedOperand, *unsignedOperand;
3671  if (LHS->getType()->hasSignedIntegerRepresentation()) {
3672    assert(!RHS->getType()->hasSignedIntegerRepresentation() &&
3673           "unsigned comparison between two signed integer expressions?");
3674    signedOperand = LHS;
3675    unsignedOperand = RHS;
3676  } else if (RHS->getType()->hasSignedIntegerRepresentation()) {
3677    signedOperand = RHS;
3678    unsignedOperand = LHS;
3679  } else {
3680    CheckTrivialUnsignedComparison(S, E);
3681    return AnalyzeImpConvsInComparison(S, E);
3682  }
3683
3684  // Otherwise, calculate the effective range of the signed operand.
3685  IntRange signedRange = GetExprRange(S.Context, signedOperand);
3686
3687  // Go ahead and analyze implicit conversions in the operands.  Note
3688  // that we skip the implicit conversions on both sides.
3689  AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc());
3690  AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc());
3691
3692  // If the signed range is non-negative, -Wsign-compare won't fire,
3693  // but we should still check for comparisons which are always true
3694  // or false.
3695  if (signedRange.NonNegative)
3696    return CheckTrivialUnsignedComparison(S, E);
3697
3698  // For (in)equality comparisons, if the unsigned operand is a
3699  // constant which cannot collide with a overflowed signed operand,
3700  // then reinterpreting the signed operand as unsigned will not
3701  // change the result of the comparison.
3702  if (E->isEqualityOp()) {
3703    unsigned comparisonWidth = S.Context.getIntWidth(T);
3704    IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
3705
3706    // We should never be unable to prove that the unsigned operand is
3707    // non-negative.
3708    assert(unsignedRange.NonNegative && "unsigned range includes negative?");
3709
3710    if (unsignedRange.Width < comparisonWidth)
3711      return;
3712  }
3713
3714  S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison)
3715    << LHS->getType() << RHS->getType()
3716    << LHS->getSourceRange() << RHS->getSourceRange();
3717}
3718
3719/// Analyzes an attempt to assign the given value to a bitfield.
3720///
3721/// Returns true if there was something fishy about the attempt.
3722static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
3723                                      SourceLocation InitLoc) {
3724  assert(Bitfield->isBitField());
3725  if (Bitfield->isInvalidDecl())
3726    return false;
3727
3728  // White-list bool bitfields.
3729  if (Bitfield->getType()->isBooleanType())
3730    return false;
3731
3732  // Ignore value- or type-dependent expressions.
3733  if (Bitfield->getBitWidth()->isValueDependent() ||
3734      Bitfield->getBitWidth()->isTypeDependent() ||
3735      Init->isValueDependent() ||
3736      Init->isTypeDependent())
3737    return false;
3738
3739  Expr *OriginalInit = Init->IgnoreParenImpCasts();
3740
3741  llvm::APSInt Value;
3742  if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects))
3743    return false;
3744
3745  unsigned OriginalWidth = Value.getBitWidth();
3746  unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context);
3747
3748  if (OriginalWidth <= FieldWidth)
3749    return false;
3750
3751  // Compute the value which the bitfield will contain.
3752  llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
3753  TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType());
3754
3755  // Check whether the stored value is equal to the original value.
3756  TruncatedValue = TruncatedValue.extend(OriginalWidth);
3757  if (Value == TruncatedValue)
3758    return false;
3759
3760  // Special-case bitfields of width 1: booleans are naturally 0/1, and
3761  // therefore don't strictly fit into a signed bitfield of width 1.
3762  if (FieldWidth == 1 && Value == 1)
3763    return false;
3764
3765  std::string PrettyValue = Value.toString(10);
3766  std::string PrettyTrunc = TruncatedValue.toString(10);
3767
3768  S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
3769    << PrettyValue << PrettyTrunc << OriginalInit->getType()
3770    << Init->getSourceRange();
3771
3772  return true;
3773}
3774
3775/// Analyze the given simple or compound assignment for warning-worthy
3776/// operations.
3777static void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
3778  // Just recurse on the LHS.
3779  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
3780
3781  // We want to recurse on the RHS as normal unless we're assigning to
3782  // a bitfield.
3783  if (FieldDecl *Bitfield = E->getLHS()->getBitField()) {
3784    if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
3785                                  E->getOperatorLoc())) {
3786      // Recurse, ignoring any implicit conversions on the RHS.
3787      return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
3788                                        E->getOperatorLoc());
3789    }
3790  }
3791
3792  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
3793}
3794
3795/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
3796static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
3797                            SourceLocation CContext, unsigned diag,
3798                            bool pruneControlFlow = false) {
3799  if (pruneControlFlow) {
3800    S.DiagRuntimeBehavior(E->getExprLoc(), E,
3801                          S.PDiag(diag)
3802                            << SourceType << T << E->getSourceRange()
3803                            << SourceRange(CContext));
3804    return;
3805  }
3806  S.Diag(E->getExprLoc(), diag)
3807    << SourceType << T << E->getSourceRange() << SourceRange(CContext);
3808}
3809
3810/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
3811static void DiagnoseImpCast(Sema &S, Expr *E, QualType T,
3812                            SourceLocation CContext, unsigned diag,
3813                            bool pruneControlFlow = false) {
3814  DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow);
3815}
3816
3817/// Diagnose an implicit cast from a literal expression. Does not warn when the
3818/// cast wouldn't lose information.
3819void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
3820                                    SourceLocation CContext) {
3821  // Try to convert the literal exactly to an integer. If we can, don't warn.
3822  bool isExact = false;
3823  const llvm::APFloat &Value = FL->getValue();
3824  llvm::APSInt IntegerValue(S.Context.getIntWidth(T),
3825                            T->hasUnsignedIntegerRepresentation());
3826  if (Value.convertToInteger(IntegerValue,
3827                             llvm::APFloat::rmTowardZero, &isExact)
3828      == llvm::APFloat::opOK && isExact)
3829    return;
3830
3831  S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
3832    << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext);
3833}
3834
3835std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
3836  if (!Range.Width) return "0";
3837
3838  llvm::APSInt ValueInRange = Value;
3839  ValueInRange.setIsSigned(!Range.NonNegative);
3840  ValueInRange = ValueInRange.trunc(Range.Width);
3841  return ValueInRange.toString(10);
3842}
3843
3844void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
3845                             SourceLocation CC, bool *ICContext = 0) {
3846  if (E->isTypeDependent() || E->isValueDependent()) return;
3847
3848  const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
3849  const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
3850  if (Source == Target) return;
3851  if (Target->isDependentType()) return;
3852
3853  // If the conversion context location is invalid don't complain. We also
3854  // don't want to emit a warning if the issue occurs from the expansion of
3855  // a system macro. The problem is that 'getSpellingLoc()' is slow, so we
3856  // delay this check as long as possible. Once we detect we are in that
3857  // scenario, we just return.
3858  if (CC.isInvalid())
3859    return;
3860
3861  // Diagnose implicit casts to bool.
3862  if (Target->isSpecificBuiltinType(BuiltinType::Bool)) {
3863    if (isa<StringLiteral>(E))
3864      // Warn on string literal to bool.  Checks for string literals in logical
3865      // expressions, for instances, assert(0 && "error here"), is prevented
3866      // by a check in AnalyzeImplicitConversions().
3867      return DiagnoseImpCast(S, E, T, CC,
3868                             diag::warn_impcast_string_literal_to_bool);
3869    if (Source->isFunctionType()) {
3870      // Warn on function to bool. Checks free functions and static member
3871      // functions. Weakly imported functions are excluded from the check,
3872      // since it's common to test their value to check whether the linker
3873      // found a definition for them.
3874      ValueDecl *D = 0;
3875      if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) {
3876        D = R->getDecl();
3877      } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) {
3878        D = M->getMemberDecl();
3879      }
3880
3881      if (D && !D->isWeak()) {
3882        if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) {
3883          S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool)
3884            << F << E->getSourceRange() << SourceRange(CC);
3885          S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence)
3886            << FixItHint::CreateInsertion(E->getExprLoc(), "&");
3887          QualType ReturnType;
3888          UnresolvedSet<4> NonTemplateOverloads;
3889          S.isExprCallable(*E, ReturnType, NonTemplateOverloads);
3890          if (!ReturnType.isNull()
3891              && ReturnType->isSpecificBuiltinType(BuiltinType::Bool))
3892            S.Diag(E->getExprLoc(), diag::note_function_to_bool_call)
3893              << FixItHint::CreateInsertion(
3894                 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()");
3895          return;
3896        }
3897      }
3898    }
3899    return; // Other casts to bool are not checked.
3900  }
3901
3902  // Strip vector types.
3903  if (isa<VectorType>(Source)) {
3904    if (!isa<VectorType>(Target)) {
3905      if (S.SourceMgr.isInSystemMacro(CC))
3906        return;
3907      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
3908    }
3909
3910    // If the vector cast is cast between two vectors of the same size, it is
3911    // a bitcast, not a conversion.
3912    if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target))
3913      return;
3914
3915    Source = cast<VectorType>(Source)->getElementType().getTypePtr();
3916    Target = cast<VectorType>(Target)->getElementType().getTypePtr();
3917  }
3918
3919  // Strip complex types.
3920  if (isa<ComplexType>(Source)) {
3921    if (!isa<ComplexType>(Target)) {
3922      if (S.SourceMgr.isInSystemMacro(CC))
3923        return;
3924
3925      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
3926    }
3927
3928    Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
3929    Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
3930  }
3931
3932  const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
3933  const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
3934
3935  // If the source is floating point...
3936  if (SourceBT && SourceBT->isFloatingPoint()) {
3937    // ...and the target is floating point...
3938    if (TargetBT && TargetBT->isFloatingPoint()) {
3939      // ...then warn if we're dropping FP rank.
3940
3941      // Builtin FP kinds are ordered by increasing FP rank.
3942      if (SourceBT->getKind() > TargetBT->getKind()) {
3943        // Don't warn about float constants that are precisely
3944        // representable in the target type.
3945        Expr::EvalResult result;
3946        if (E->EvaluateAsRValue(result, S.Context)) {
3947          // Value might be a float, a float vector, or a float complex.
3948          if (IsSameFloatAfterCast(result.Val,
3949                   S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
3950                   S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
3951            return;
3952        }
3953
3954        if (S.SourceMgr.isInSystemMacro(CC))
3955          return;
3956
3957        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
3958      }
3959      return;
3960    }
3961
3962    // If the target is integral, always warn.
3963    if ((TargetBT && TargetBT->isInteger())) {
3964      if (S.SourceMgr.isInSystemMacro(CC))
3965        return;
3966
3967      Expr *InnerE = E->IgnoreParenImpCasts();
3968      // We also want to warn on, e.g., "int i = -1.234"
3969      if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE))
3970        if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus)
3971          InnerE = UOp->getSubExpr()->IgnoreParenImpCasts();
3972
3973      if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
3974        DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
3975      } else {
3976        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
3977      }
3978    }
3979
3980    return;
3981  }
3982
3983  if (!Source->isIntegerType() || !Target->isIntegerType())
3984    return;
3985
3986  if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)
3987           == Expr::NPCK_GNUNull) && Target->isIntegerType()) {
3988    S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer)
3989        << E->getSourceRange() << clang::SourceRange(CC);
3990    return;
3991  }
3992
3993  IntRange SourceRange = GetExprRange(S.Context, E);
3994  IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
3995
3996  if (SourceRange.Width > TargetRange.Width) {
3997    // If the source is a constant, use a default-on diagnostic.
3998    // TODO: this should happen for bitfield stores, too.
3999    llvm::APSInt Value(32);
4000    if (E->isIntegerConstantExpr(Value, S.Context)) {
4001      if (S.SourceMgr.isInSystemMacro(CC))
4002        return;
4003
4004      std::string PrettySourceValue = Value.toString(10);
4005      std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
4006
4007      S.DiagRuntimeBehavior(E->getExprLoc(), E,
4008        S.PDiag(diag::warn_impcast_integer_precision_constant)
4009            << PrettySourceValue << PrettyTargetValue
4010            << E->getType() << T << E->getSourceRange()
4011            << clang::SourceRange(CC));
4012      return;
4013    }
4014
4015    // People want to build with -Wshorten-64-to-32 and not -Wconversion.
4016    if (S.SourceMgr.isInSystemMacro(CC))
4017      return;
4018
4019    if (SourceRange.Width == 64 && TargetRange.Width == 32)
4020      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32,
4021                             /* pruneControlFlow */ true);
4022    return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
4023  }
4024
4025  if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
4026      (!TargetRange.NonNegative && SourceRange.NonNegative &&
4027       SourceRange.Width == TargetRange.Width)) {
4028
4029    if (S.SourceMgr.isInSystemMacro(CC))
4030      return;
4031
4032    unsigned DiagID = diag::warn_impcast_integer_sign;
4033
4034    // Traditionally, gcc has warned about this under -Wsign-compare.
4035    // We also want to warn about it in -Wconversion.
4036    // So if -Wconversion is off, use a completely identical diagnostic
4037    // in the sign-compare group.
4038    // The conditional-checking code will
4039    if (ICContext) {
4040      DiagID = diag::warn_impcast_integer_sign_conditional;
4041      *ICContext = true;
4042    }
4043
4044    return DiagnoseImpCast(S, E, T, CC, DiagID);
4045  }
4046
4047  // Diagnose conversions between different enumeration types.
4048  // In C, we pretend that the type of an EnumConstantDecl is its enumeration
4049  // type, to give us better diagnostics.
4050  QualType SourceType = E->getType();
4051  if (!S.getLangOptions().CPlusPlus) {
4052    if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
4053      if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
4054        EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
4055        SourceType = S.Context.getTypeDeclType(Enum);
4056        Source = S.Context.getCanonicalType(SourceType).getTypePtr();
4057      }
4058  }
4059
4060  if (const EnumType *SourceEnum = Source->getAs<EnumType>())
4061    if (const EnumType *TargetEnum = Target->getAs<EnumType>())
4062      if ((SourceEnum->getDecl()->getIdentifier() ||
4063           SourceEnum->getDecl()->getTypedefNameForAnonDecl()) &&
4064          (TargetEnum->getDecl()->getIdentifier() ||
4065           TargetEnum->getDecl()->getTypedefNameForAnonDecl()) &&
4066          SourceEnum != TargetEnum) {
4067        if (S.SourceMgr.isInSystemMacro(CC))
4068          return;
4069
4070        return DiagnoseImpCast(S, E, SourceType, T, CC,
4071                               diag::warn_impcast_different_enum_types);
4072      }
4073
4074  return;
4075}
4076
4077void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T);
4078
4079void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
4080                             SourceLocation CC, bool &ICContext) {
4081  E = E->IgnoreParenImpCasts();
4082
4083  if (isa<ConditionalOperator>(E))
4084    return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T);
4085
4086  AnalyzeImplicitConversions(S, E, CC);
4087  if (E->getType() != T)
4088    return CheckImplicitConversion(S, E, T, CC, &ICContext);
4089  return;
4090}
4091
4092void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) {
4093  SourceLocation CC = E->getQuestionLoc();
4094
4095  AnalyzeImplicitConversions(S, E->getCond(), CC);
4096
4097  bool Suspicious = false;
4098  CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
4099  CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
4100
4101  // If -Wconversion would have warned about either of the candidates
4102  // for a signedness conversion to the context type...
4103  if (!Suspicious) return;
4104
4105  // ...but it's currently ignored...
4106  if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
4107                                 CC))
4108    return;
4109
4110  // ...then check whether it would have warned about either of the
4111  // candidates for a signedness conversion to the condition type.
4112  if (E->getType() == T) return;
4113
4114  Suspicious = false;
4115  CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
4116                          E->getType(), CC, &Suspicious);
4117  if (!Suspicious)
4118    CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
4119                            E->getType(), CC, &Suspicious);
4120}
4121
4122/// AnalyzeImplicitConversions - Find and report any interesting
4123/// implicit conversions in the given expression.  There are a couple
4124/// of competing diagnostics here, -Wconversion and -Wsign-compare.
4125void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
4126  QualType T = OrigE->getType();
4127  Expr *E = OrigE->IgnoreParenImpCasts();
4128
4129  if (E->isTypeDependent() || E->isValueDependent())
4130    return;
4131
4132  // For conditional operators, we analyze the arguments as if they
4133  // were being fed directly into the output.
4134  if (isa<ConditionalOperator>(E)) {
4135    ConditionalOperator *CO = cast<ConditionalOperator>(E);
4136    CheckConditionalOperator(S, CO, T);
4137    return;
4138  }
4139
4140  // Go ahead and check any implicit conversions we might have skipped.
4141  // The non-canonical typecheck is just an optimization;
4142  // CheckImplicitConversion will filter out dead implicit conversions.
4143  if (E->getType() != T)
4144    CheckImplicitConversion(S, E, T, CC);
4145
4146  // Now continue drilling into this expression.
4147
4148  // Skip past explicit casts.
4149  if (isa<ExplicitCastExpr>(E)) {
4150    E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
4151    return AnalyzeImplicitConversions(S, E, CC);
4152  }
4153
4154  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
4155    // Do a somewhat different check with comparison operators.
4156    if (BO->isComparisonOp())
4157      return AnalyzeComparison(S, BO);
4158
4159    // And with simple assignments.
4160    if (BO->getOpcode() == BO_Assign)
4161      return AnalyzeAssignment(S, BO);
4162  }
4163
4164  // These break the otherwise-useful invariant below.  Fortunately,
4165  // we don't really need to recurse into them, because any internal
4166  // expressions should have been analyzed already when they were
4167  // built into statements.
4168  if (isa<StmtExpr>(E)) return;
4169
4170  // Don't descend into unevaluated contexts.
4171  if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
4172
4173  // Now just recurse over the expression's children.
4174  CC = E->getExprLoc();
4175  BinaryOperator *BO = dyn_cast<BinaryOperator>(E);
4176  bool IsLogicalOperator = BO && BO->isLogicalOp();
4177  for (Stmt::child_range I = E->children(); I; ++I) {
4178    Expr *ChildExpr = cast<Expr>(*I);
4179    if (IsLogicalOperator &&
4180        isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts()))
4181      // Ignore checking string literals that are in logical operators.
4182      continue;
4183    AnalyzeImplicitConversions(S, ChildExpr, CC);
4184  }
4185}
4186
4187} // end anonymous namespace
4188
4189/// Diagnoses "dangerous" implicit conversions within the given
4190/// expression (which is a full expression).  Implements -Wconversion
4191/// and -Wsign-compare.
4192///
4193/// \param CC the "context" location of the implicit conversion, i.e.
4194///   the most location of the syntactic entity requiring the implicit
4195///   conversion
4196void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
4197  // Don't diagnose in unevaluated contexts.
4198  if (ExprEvalContexts.back().Context == Sema::Unevaluated)
4199    return;
4200
4201  // Don't diagnose for value- or type-dependent expressions.
4202  if (E->isTypeDependent() || E->isValueDependent())
4203    return;
4204
4205  // Check for array bounds violations in cases where the check isn't triggered
4206  // elsewhere for other Expr types (like BinaryOperators), e.g. when an
4207  // ArraySubscriptExpr is on the RHS of a variable initialization.
4208  CheckArrayAccess(E);
4209
4210  // This is not the right CC for (e.g.) a variable initialization.
4211  AnalyzeImplicitConversions(*this, E, CC);
4212}
4213
4214void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
4215                                       FieldDecl *BitField,
4216                                       Expr *Init) {
4217  (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
4218}
4219
4220/// CheckParmsForFunctionDef - Check that the parameters of the given
4221/// function are appropriate for the definition of a function. This
4222/// takes care of any checks that cannot be performed on the
4223/// declaration itself, e.g., that the types of each of the function
4224/// parameters are complete.
4225bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
4226                                    bool CheckParameterNames) {
4227  bool HasInvalidParm = false;
4228  for (; P != PEnd; ++P) {
4229    ParmVarDecl *Param = *P;
4230
4231    // C99 6.7.5.3p4: the parameters in a parameter type list in a
4232    // function declarator that is part of a function definition of
4233    // that function shall not have incomplete type.
4234    //
4235    // This is also C++ [dcl.fct]p6.
4236    if (!Param->isInvalidDecl() &&
4237        RequireCompleteType(Param->getLocation(), Param->getType(),
4238                               diag::err_typecheck_decl_incomplete_type)) {
4239      Param->setInvalidDecl();
4240      HasInvalidParm = true;
4241    }
4242
4243    // C99 6.9.1p5: If the declarator includes a parameter type list, the
4244    // declaration of each parameter shall include an identifier.
4245    if (CheckParameterNames &&
4246        Param->getIdentifier() == 0 &&
4247        !Param->isImplicit() &&
4248        !getLangOptions().CPlusPlus)
4249      Diag(Param->getLocation(), diag::err_parameter_name_omitted);
4250
4251    // C99 6.7.5.3p12:
4252    //   If the function declarator is not part of a definition of that
4253    //   function, parameters may have incomplete type and may use the [*]
4254    //   notation in their sequences of declarator specifiers to specify
4255    //   variable length array types.
4256    QualType PType = Param->getOriginalType();
4257    if (const ArrayType *AT = Context.getAsArrayType(PType)) {
4258      if (AT->getSizeModifier() == ArrayType::Star) {
4259        // FIXME: This diagnosic should point the the '[*]' if source-location
4260        // information is added for it.
4261        Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
4262      }
4263    }
4264  }
4265
4266  return HasInvalidParm;
4267}
4268
4269/// CheckCastAlign - Implements -Wcast-align, which warns when a
4270/// pointer cast increases the alignment requirements.
4271void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
4272  // This is actually a lot of work to potentially be doing on every
4273  // cast; don't do it if we're ignoring -Wcast_align (as is the default).
4274  if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
4275                                          TRange.getBegin())
4276        == DiagnosticsEngine::Ignored)
4277    return;
4278
4279  // Ignore dependent types.
4280  if (T->isDependentType() || Op->getType()->isDependentType())
4281    return;
4282
4283  // Require that the destination be a pointer type.
4284  const PointerType *DestPtr = T->getAs<PointerType>();
4285  if (!DestPtr) return;
4286
4287  // If the destination has alignment 1, we're done.
4288  QualType DestPointee = DestPtr->getPointeeType();
4289  if (DestPointee->isIncompleteType()) return;
4290  CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
4291  if (DestAlign.isOne()) return;
4292
4293  // Require that the source be a pointer type.
4294  const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
4295  if (!SrcPtr) return;
4296  QualType SrcPointee = SrcPtr->getPointeeType();
4297
4298  // Whitelist casts from cv void*.  We already implicitly
4299  // whitelisted casts to cv void*, since they have alignment 1.
4300  // Also whitelist casts involving incomplete types, which implicitly
4301  // includes 'void'.
4302  if (SrcPointee->isIncompleteType()) return;
4303
4304  CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
4305  if (SrcAlign >= DestAlign) return;
4306
4307  Diag(TRange.getBegin(), diag::warn_cast_align)
4308    << Op->getType() << T
4309    << static_cast<unsigned>(SrcAlign.getQuantity())
4310    << static_cast<unsigned>(DestAlign.getQuantity())
4311    << TRange << Op->getSourceRange();
4312}
4313
4314static const Type* getElementType(const Expr *BaseExpr) {
4315  const Type* EltType = BaseExpr->getType().getTypePtr();
4316  if (EltType->isAnyPointerType())
4317    return EltType->getPointeeType().getTypePtr();
4318  else if (EltType->isArrayType())
4319    return EltType->getBaseElementTypeUnsafe();
4320  return EltType;
4321}
4322
4323/// \brief Check whether this array fits the idiom of a size-one tail padded
4324/// array member of a struct.
4325///
4326/// We avoid emitting out-of-bounds access warnings for such arrays as they are
4327/// commonly used to emulate flexible arrays in C89 code.
4328static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size,
4329                                    const NamedDecl *ND) {
4330  if (Size != 1 || !ND) return false;
4331
4332  const FieldDecl *FD = dyn_cast<FieldDecl>(ND);
4333  if (!FD) return false;
4334
4335  // Don't consider sizes resulting from macro expansions or template argument
4336  // substitution to form C89 tail-padded arrays.
4337  ConstantArrayTypeLoc TL =
4338    cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc());
4339  const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr());
4340  if (!SizeExpr || SizeExpr->getExprLoc().isMacroID())
4341    return false;
4342
4343  const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext());
4344  if (!RD) return false;
4345  if (RD->isUnion()) return false;
4346  if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) {
4347    if (!CRD->isStandardLayout()) return false;
4348  }
4349
4350  // See if this is the last field decl in the record.
4351  const Decl *D = FD;
4352  while ((D = D->getNextDeclInContext()))
4353    if (isa<FieldDecl>(D))
4354      return false;
4355  return true;
4356}
4357
4358void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr,
4359                            const ArraySubscriptExpr *ASE,
4360                            bool AllowOnePastEnd, bool IndexNegated) {
4361  IndexExpr = IndexExpr->IgnoreParenCasts();
4362  if (IndexExpr->isValueDependent())
4363    return;
4364
4365  const Type *EffectiveType = getElementType(BaseExpr);
4366  BaseExpr = BaseExpr->IgnoreParenCasts();
4367  const ConstantArrayType *ArrayTy =
4368    Context.getAsConstantArrayType(BaseExpr->getType());
4369  if (!ArrayTy)
4370    return;
4371
4372  llvm::APSInt index;
4373  if (!IndexExpr->EvaluateAsInt(index, Context))
4374    return;
4375  if (IndexNegated)
4376    index = -index;
4377
4378  const NamedDecl *ND = NULL;
4379  if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
4380    ND = dyn_cast<NamedDecl>(DRE->getDecl());
4381  if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
4382    ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
4383
4384  if (index.isUnsigned() || !index.isNegative()) {
4385    llvm::APInt size = ArrayTy->getSize();
4386    if (!size.isStrictlyPositive())
4387      return;
4388
4389    const Type* BaseType = getElementType(BaseExpr);
4390    if (BaseType != EffectiveType) {
4391      // Make sure we're comparing apples to apples when comparing index to size
4392      uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType);
4393      uint64_t array_typesize = Context.getTypeSize(BaseType);
4394      // Handle ptrarith_typesize being zero, such as when casting to void*
4395      if (!ptrarith_typesize) ptrarith_typesize = 1;
4396      if (ptrarith_typesize != array_typesize) {
4397        // There's a cast to a different size type involved
4398        uint64_t ratio = array_typesize / ptrarith_typesize;
4399        // TODO: Be smarter about handling cases where array_typesize is not a
4400        // multiple of ptrarith_typesize
4401        if (ptrarith_typesize * ratio == array_typesize)
4402          size *= llvm::APInt(size.getBitWidth(), ratio);
4403      }
4404    }
4405
4406    if (size.getBitWidth() > index.getBitWidth())
4407      index = index.sext(size.getBitWidth());
4408    else if (size.getBitWidth() < index.getBitWidth())
4409      size = size.sext(index.getBitWidth());
4410
4411    // For array subscripting the index must be less than size, but for pointer
4412    // arithmetic also allow the index (offset) to be equal to size since
4413    // computing the next address after the end of the array is legal and
4414    // commonly done e.g. in C++ iterators and range-based for loops.
4415    if (AllowOnePastEnd ? index.sle(size) : index.slt(size))
4416      return;
4417
4418    // Also don't warn for arrays of size 1 which are members of some
4419    // structure. These are often used to approximate flexible arrays in C89
4420    // code.
4421    if (IsTailPaddedMemberArray(*this, size, ND))
4422      return;
4423
4424    // Suppress the warning if the subscript expression (as identified by the
4425    // ']' location) and the index expression are both from macro expansions
4426    // within a system header.
4427    if (ASE) {
4428      SourceLocation RBracketLoc = SourceMgr.getSpellingLoc(
4429          ASE->getRBracketLoc());
4430      if (SourceMgr.isInSystemHeader(RBracketLoc)) {
4431        SourceLocation IndexLoc = SourceMgr.getSpellingLoc(
4432            IndexExpr->getLocStart());
4433        if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc))
4434          return;
4435      }
4436    }
4437
4438    unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds;
4439    if (ASE)
4440      DiagID = diag::warn_array_index_exceeds_bounds;
4441
4442    DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
4443                        PDiag(DiagID) << index.toString(10, true)
4444                          << size.toString(10, true)
4445                          << (unsigned)size.getLimitedValue(~0U)
4446                          << IndexExpr->getSourceRange());
4447  } else {
4448    unsigned DiagID = diag::warn_array_index_precedes_bounds;
4449    if (!ASE) {
4450      DiagID = diag::warn_ptr_arith_precedes_bounds;
4451      if (index.isNegative()) index = -index;
4452    }
4453
4454    DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
4455                        PDiag(DiagID) << index.toString(10, true)
4456                          << IndexExpr->getSourceRange());
4457  }
4458
4459  if (!ND) {
4460    // Try harder to find a NamedDecl to point at in the note.
4461    while (const ArraySubscriptExpr *ASE =
4462           dyn_cast<ArraySubscriptExpr>(BaseExpr))
4463      BaseExpr = ASE->getBase()->IgnoreParenCasts();
4464    if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
4465      ND = dyn_cast<NamedDecl>(DRE->getDecl());
4466    if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
4467      ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
4468  }
4469
4470  if (ND)
4471    DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
4472                        PDiag(diag::note_array_index_out_of_bounds)
4473                          << ND->getDeclName());
4474}
4475
4476void Sema::CheckArrayAccess(const Expr *expr) {
4477  int AllowOnePastEnd = 0;
4478  while (expr) {
4479    expr = expr->IgnoreParenImpCasts();
4480    switch (expr->getStmtClass()) {
4481      case Stmt::ArraySubscriptExprClass: {
4482        const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr);
4483        CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE,
4484                         AllowOnePastEnd > 0);
4485        return;
4486      }
4487      case Stmt::UnaryOperatorClass: {
4488        // Only unwrap the * and & unary operators
4489        const UnaryOperator *UO = cast<UnaryOperator>(expr);
4490        expr = UO->getSubExpr();
4491        switch (UO->getOpcode()) {
4492          case UO_AddrOf:
4493            AllowOnePastEnd++;
4494            break;
4495          case UO_Deref:
4496            AllowOnePastEnd--;
4497            break;
4498          default:
4499            return;
4500        }
4501        break;
4502      }
4503      case Stmt::ConditionalOperatorClass: {
4504        const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
4505        if (const Expr *lhs = cond->getLHS())
4506          CheckArrayAccess(lhs);
4507        if (const Expr *rhs = cond->getRHS())
4508          CheckArrayAccess(rhs);
4509        return;
4510      }
4511      default:
4512        return;
4513    }
4514  }
4515}
4516
4517//===--- CHECK: Objective-C retain cycles ----------------------------------//
4518
4519namespace {
4520  struct RetainCycleOwner {
4521    RetainCycleOwner() : Variable(0), Indirect(false) {}
4522    VarDecl *Variable;
4523    SourceRange Range;
4524    SourceLocation Loc;
4525    bool Indirect;
4526
4527    void setLocsFrom(Expr *e) {
4528      Loc = e->getExprLoc();
4529      Range = e->getSourceRange();
4530    }
4531  };
4532}
4533
4534/// Consider whether capturing the given variable can possibly lead to
4535/// a retain cycle.
4536static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) {
4537  // In ARC, it's captured strongly iff the variable has __strong
4538  // lifetime.  In MRR, it's captured strongly if the variable is
4539  // __block and has an appropriate type.
4540  if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
4541    return false;
4542
4543  owner.Variable = var;
4544  owner.setLocsFrom(ref);
4545  return true;
4546}
4547
4548static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) {
4549  while (true) {
4550    e = e->IgnoreParens();
4551    if (CastExpr *cast = dyn_cast<CastExpr>(e)) {
4552      switch (cast->getCastKind()) {
4553      case CK_BitCast:
4554      case CK_LValueBitCast:
4555      case CK_LValueToRValue:
4556      case CK_ARCReclaimReturnedObject:
4557        e = cast->getSubExpr();
4558        continue;
4559
4560      default:
4561        return false;
4562      }
4563    }
4564
4565    if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) {
4566      ObjCIvarDecl *ivar = ref->getDecl();
4567      if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
4568        return false;
4569
4570      // Try to find a retain cycle in the base.
4571      if (!findRetainCycleOwner(S, ref->getBase(), owner))
4572        return false;
4573
4574      if (ref->isFreeIvar()) owner.setLocsFrom(ref);
4575      owner.Indirect = true;
4576      return true;
4577    }
4578
4579    if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) {
4580      VarDecl *var = dyn_cast<VarDecl>(ref->getDecl());
4581      if (!var) return false;
4582      return considerVariable(var, ref, owner);
4583    }
4584
4585    if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) {
4586      owner.Variable = ref->getDecl();
4587      owner.setLocsFrom(ref);
4588      return true;
4589    }
4590
4591    if (MemberExpr *member = dyn_cast<MemberExpr>(e)) {
4592      if (member->isArrow()) return false;
4593
4594      // Don't count this as an indirect ownership.
4595      e = member->getBase();
4596      continue;
4597    }
4598
4599    if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) {
4600      // Only pay attention to pseudo-objects on property references.
4601      ObjCPropertyRefExpr *pre
4602        = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm()
4603                                              ->IgnoreParens());
4604      if (!pre) return false;
4605      if (pre->isImplicitProperty()) return false;
4606      ObjCPropertyDecl *property = pre->getExplicitProperty();
4607      if (!property->isRetaining() &&
4608          !(property->getPropertyIvarDecl() &&
4609            property->getPropertyIvarDecl()->getType()
4610              .getObjCLifetime() == Qualifiers::OCL_Strong))
4611          return false;
4612
4613      owner.Indirect = true;
4614      if (pre->isSuperReceiver()) {
4615        owner.Variable = S.getCurMethodDecl()->getSelfDecl();
4616        if (!owner.Variable)
4617          return false;
4618        owner.Loc = pre->getLocation();
4619        owner.Range = pre->getSourceRange();
4620        return true;
4621      }
4622      e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase())
4623                              ->getSourceExpr());
4624      continue;
4625    }
4626
4627    // Array ivars?
4628
4629    return false;
4630  }
4631}
4632
4633namespace {
4634  struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> {
4635    FindCaptureVisitor(ASTContext &Context, VarDecl *variable)
4636      : EvaluatedExprVisitor<FindCaptureVisitor>(Context),
4637        Variable(variable), Capturer(0) {}
4638
4639    VarDecl *Variable;
4640    Expr *Capturer;
4641
4642    void VisitDeclRefExpr(DeclRefExpr *ref) {
4643      if (ref->getDecl() == Variable && !Capturer)
4644        Capturer = ref;
4645    }
4646
4647    void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) {
4648      if (ref->getDecl() == Variable && !Capturer)
4649        Capturer = ref;
4650    }
4651
4652    void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) {
4653      if (Capturer) return;
4654      Visit(ref->getBase());
4655      if (Capturer && ref->isFreeIvar())
4656        Capturer = ref;
4657    }
4658
4659    void VisitBlockExpr(BlockExpr *block) {
4660      // Look inside nested blocks
4661      if (block->getBlockDecl()->capturesVariable(Variable))
4662        Visit(block->getBlockDecl()->getBody());
4663    }
4664  };
4665}
4666
4667/// Check whether the given argument is a block which captures a
4668/// variable.
4669static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) {
4670  assert(owner.Variable && owner.Loc.isValid());
4671
4672  e = e->IgnoreParenCasts();
4673  BlockExpr *block = dyn_cast<BlockExpr>(e);
4674  if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable))
4675    return 0;
4676
4677  FindCaptureVisitor visitor(S.Context, owner.Variable);
4678  visitor.Visit(block->getBlockDecl()->getBody());
4679  return visitor.Capturer;
4680}
4681
4682static void diagnoseRetainCycle(Sema &S, Expr *capturer,
4683                                RetainCycleOwner &owner) {
4684  assert(capturer);
4685  assert(owner.Variable && owner.Loc.isValid());
4686
4687  S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle)
4688    << owner.Variable << capturer->getSourceRange();
4689  S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner)
4690    << owner.Indirect << owner.Range;
4691}
4692
4693/// Check for a keyword selector that starts with the word 'add' or
4694/// 'set'.
4695static bool isSetterLikeSelector(Selector sel) {
4696  if (sel.isUnarySelector()) return false;
4697
4698  StringRef str = sel.getNameForSlot(0);
4699  while (!str.empty() && str.front() == '_') str = str.substr(1);
4700  if (str.startswith("set"))
4701    str = str.substr(3);
4702  else if (str.startswith("add")) {
4703    // Specially whitelist 'addOperationWithBlock:'.
4704    if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock"))
4705      return false;
4706    str = str.substr(3);
4707  }
4708  else
4709    return false;
4710
4711  if (str.empty()) return true;
4712  return !islower(str.front());
4713}
4714
4715/// Check a message send to see if it's likely to cause a retain cycle.
4716void Sema::checkRetainCycles(ObjCMessageExpr *msg) {
4717  // Only check instance methods whose selector looks like a setter.
4718  if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector()))
4719    return;
4720
4721  // Try to find a variable that the receiver is strongly owned by.
4722  RetainCycleOwner owner;
4723  if (msg->getReceiverKind() == ObjCMessageExpr::Instance) {
4724    if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner))
4725      return;
4726  } else {
4727    assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance);
4728    owner.Variable = getCurMethodDecl()->getSelfDecl();
4729    owner.Loc = msg->getSuperLoc();
4730    owner.Range = msg->getSuperLoc();
4731  }
4732
4733  // Check whether the receiver is captured by any of the arguments.
4734  for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i)
4735    if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner))
4736      return diagnoseRetainCycle(*this, capturer, owner);
4737}
4738
4739/// Check a property assign to see if it's likely to cause a retain cycle.
4740void Sema::checkRetainCycles(Expr *receiver, Expr *argument) {
4741  RetainCycleOwner owner;
4742  if (!findRetainCycleOwner(*this, receiver, owner))
4743    return;
4744
4745  if (Expr *capturer = findCapturingExpr(*this, argument, owner))
4746    diagnoseRetainCycle(*this, capturer, owner);
4747}
4748
4749bool Sema::checkUnsafeAssigns(SourceLocation Loc,
4750                              QualType LHS, Expr *RHS) {
4751  Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime();
4752  if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone)
4753    return false;
4754  // strip off any implicit cast added to get to the one arc-specific
4755  while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
4756    if (cast->getCastKind() == CK_ARCConsumeObject) {
4757      Diag(Loc, diag::warn_arc_retained_assign)
4758        << (LT == Qualifiers::OCL_ExplicitNone)
4759        << RHS->getSourceRange();
4760      return true;
4761    }
4762    RHS = cast->getSubExpr();
4763  }
4764  return false;
4765}
4766
4767void Sema::checkUnsafeExprAssigns(SourceLocation Loc,
4768                              Expr *LHS, Expr *RHS) {
4769  QualType LHSType;
4770  // PropertyRef on LHS type need be directly obtained from
4771  // its declaration as it has a PsuedoType.
4772  ObjCPropertyRefExpr *PRE
4773    = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens());
4774  if (PRE && !PRE->isImplicitProperty()) {
4775    const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
4776    if (PD)
4777      LHSType = PD->getType();
4778  }
4779
4780  if (LHSType.isNull())
4781    LHSType = LHS->getType();
4782  if (checkUnsafeAssigns(Loc, LHSType, RHS))
4783    return;
4784  Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime();
4785  // FIXME. Check for other life times.
4786  if (LT != Qualifiers::OCL_None)
4787    return;
4788
4789  if (PRE) {
4790    if (PRE->isImplicitProperty())
4791      return;
4792    const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
4793    if (!PD)
4794      return;
4795
4796    unsigned Attributes = PD->getPropertyAttributes();
4797    if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) {
4798      // when 'assign' attribute was not explicitly specified
4799      // by user, ignore it and rely on property type itself
4800      // for lifetime info.
4801      unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten();
4802      if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) &&
4803          LHSType->isObjCRetainableType())
4804        return;
4805
4806      while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
4807        if (cast->getCastKind() == CK_ARCConsumeObject) {
4808          Diag(Loc, diag::warn_arc_retained_property_assign)
4809          << RHS->getSourceRange();
4810          return;
4811        }
4812        RHS = cast->getSubExpr();
4813      }
4814    }
4815  }
4816}
4817