SemaChecking.cpp revision 33e56f3273457bfa22c7c50bc46cf5a18216863d
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/ScopeInfo.h"
19#include "clang/Analysis/Analyses/FormatString.h"
20#include "clang/AST/ASTContext.h"
21#include "clang/AST/CharUnits.h"
22#include "clang/AST/DeclCXX.h"
23#include "clang/AST/DeclObjC.h"
24#include "clang/AST/ExprCXX.h"
25#include "clang/AST/ExprObjC.h"
26#include "clang/AST/EvaluatedExprVisitor.h"
27#include "clang/AST/DeclObjC.h"
28#include "clang/AST/StmtCXX.h"
29#include "clang/AST/StmtObjC.h"
30#include "clang/Lex/Preprocessor.h"
31#include "llvm/ADT/BitVector.h"
32#include "llvm/ADT/STLExtras.h"
33#include "llvm/Support/raw_ostream.h"
34#include "clang/Basic/TargetBuiltins.h"
35#include "clang/Basic/TargetInfo.h"
36#include "clang/Basic/ConvertUTF.h"
37#include <limits>
38using namespace clang;
39using namespace sema;
40
41SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL,
42                                                    unsigned ByteNo) const {
43  return SL->getLocationOfByte(ByteNo, PP.getSourceManager(),
44                               PP.getLangOptions(), PP.getTargetInfo());
45}
46
47
48/// CheckablePrintfAttr - does a function call have a "printf" attribute
49/// and arguments that merit checking?
50bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) {
51  if (Format->getType() == "printf") return true;
52  if (Format->getType() == "printf0") {
53    // printf0 allows null "format" string; if so don't check format/args
54    unsigned format_idx = Format->getFormatIdx() - 1;
55    // Does the index refer to the implicit object argument?
56    if (isa<CXXMemberCallExpr>(TheCall)) {
57      if (format_idx == 0)
58        return false;
59      --format_idx;
60    }
61    if (format_idx < TheCall->getNumArgs()) {
62      Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts();
63      if (!Format->isNullPointerConstant(Context,
64                                         Expr::NPC_ValueDependentIsNull))
65        return true;
66    }
67  }
68  return false;
69}
70
71/// Checks that a call expression's argument count is the desired number.
72/// This is useful when doing custom type-checking.  Returns true on error.
73static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) {
74  unsigned argCount = call->getNumArgs();
75  if (argCount == desiredArgCount) return false;
76
77  if (argCount < desiredArgCount)
78    return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args)
79        << 0 /*function call*/ << desiredArgCount << argCount
80        << call->getSourceRange();
81
82  // Highlight all the excess arguments.
83  SourceRange range(call->getArg(desiredArgCount)->getLocStart(),
84                    call->getArg(argCount - 1)->getLocEnd());
85
86  return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args)
87    << 0 /*function call*/ << desiredArgCount << argCount
88    << call->getArg(1)->getSourceRange();
89}
90
91/// CheckBuiltinAnnotationString - Checks that string argument to the builtin
92/// annotation is a non wide string literal.
93static bool CheckBuiltinAnnotationString(Sema &S, Expr *Arg) {
94  Arg = Arg->IgnoreParenCasts();
95  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
96  if (!Literal || !Literal->isAscii()) {
97    S.Diag(Arg->getLocStart(), diag::err_builtin_annotation_not_string_constant)
98      << Arg->getSourceRange();
99    return true;
100  }
101  return false;
102}
103
104ExprResult
105Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
106  ExprResult TheCallResult(Owned(TheCall));
107
108  // Find out if any arguments are required to be integer constant expressions.
109  unsigned ICEArguments = 0;
110  ASTContext::GetBuiltinTypeError Error;
111  Context.GetBuiltinType(BuiltinID, Error, &ICEArguments);
112  if (Error != ASTContext::GE_None)
113    ICEArguments = 0;  // Don't diagnose previously diagnosed errors.
114
115  // If any arguments are required to be ICE's, check and diagnose.
116  for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) {
117    // Skip arguments not required to be ICE's.
118    if ((ICEArguments & (1 << ArgNo)) == 0) continue;
119
120    llvm::APSInt Result;
121    if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
122      return true;
123    ICEArguments &= ~(1 << ArgNo);
124  }
125
126  switch (BuiltinID) {
127  case Builtin::BI__builtin___CFStringMakeConstantString:
128    assert(TheCall->getNumArgs() == 1 &&
129           "Wrong # arguments to builtin CFStringMakeConstantString");
130    if (CheckObjCString(TheCall->getArg(0)))
131      return ExprError();
132    break;
133  case Builtin::BI__builtin_stdarg_start:
134  case Builtin::BI__builtin_va_start:
135    if (SemaBuiltinVAStart(TheCall))
136      return ExprError();
137    break;
138  case Builtin::BI__builtin_isgreater:
139  case Builtin::BI__builtin_isgreaterequal:
140  case Builtin::BI__builtin_isless:
141  case Builtin::BI__builtin_islessequal:
142  case Builtin::BI__builtin_islessgreater:
143  case Builtin::BI__builtin_isunordered:
144    if (SemaBuiltinUnorderedCompare(TheCall))
145      return ExprError();
146    break;
147  case Builtin::BI__builtin_fpclassify:
148    if (SemaBuiltinFPClassification(TheCall, 6))
149      return ExprError();
150    break;
151  case Builtin::BI__builtin_isfinite:
152  case Builtin::BI__builtin_isinf:
153  case Builtin::BI__builtin_isinf_sign:
154  case Builtin::BI__builtin_isnan:
155  case Builtin::BI__builtin_isnormal:
156    if (SemaBuiltinFPClassification(TheCall, 1))
157      return ExprError();
158    break;
159  case Builtin::BI__builtin_shufflevector:
160    return SemaBuiltinShuffleVector(TheCall);
161    // TheCall will be freed by the smart pointer here, but that's fine, since
162    // SemaBuiltinShuffleVector guts it, but then doesn't release it.
163  case Builtin::BI__builtin_prefetch:
164    if (SemaBuiltinPrefetch(TheCall))
165      return ExprError();
166    break;
167  case Builtin::BI__builtin_object_size:
168    if (SemaBuiltinObjectSize(TheCall))
169      return ExprError();
170    break;
171  case Builtin::BI__builtin_longjmp:
172    if (SemaBuiltinLongjmp(TheCall))
173      return ExprError();
174    break;
175
176  case Builtin::BI__builtin_classify_type:
177    if (checkArgCount(*this, TheCall, 1)) return true;
178    TheCall->setType(Context.IntTy);
179    break;
180  case Builtin::BI__builtin_constant_p:
181    if (checkArgCount(*this, TheCall, 1)) return true;
182    TheCall->setType(Context.IntTy);
183    break;
184  case Builtin::BI__sync_fetch_and_add:
185  case Builtin::BI__sync_fetch_and_sub:
186  case Builtin::BI__sync_fetch_and_or:
187  case Builtin::BI__sync_fetch_and_and:
188  case Builtin::BI__sync_fetch_and_xor:
189  case Builtin::BI__sync_add_and_fetch:
190  case Builtin::BI__sync_sub_and_fetch:
191  case Builtin::BI__sync_and_and_fetch:
192  case Builtin::BI__sync_or_and_fetch:
193  case Builtin::BI__sync_xor_and_fetch:
194  case Builtin::BI__sync_val_compare_and_swap:
195  case Builtin::BI__sync_bool_compare_and_swap:
196  case Builtin::BI__sync_lock_test_and_set:
197  case Builtin::BI__sync_lock_release:
198  case Builtin::BI__sync_swap:
199    return SemaBuiltinAtomicOverloaded(move(TheCallResult));
200  case Builtin::BI__builtin_annotation:
201    if (CheckBuiltinAnnotationString(*this, TheCall->getArg(1)))
202      return ExprError();
203    break;
204  }
205
206  // Since the target specific builtins for each arch overlap, only check those
207  // of the arch we are compiling for.
208  if (BuiltinID >= Builtin::FirstTSBuiltin) {
209    switch (Context.getTargetInfo().getTriple().getArch()) {
210      case llvm::Triple::arm:
211      case llvm::Triple::thumb:
212        if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall))
213          return ExprError();
214        break;
215      default:
216        break;
217    }
218  }
219
220  return move(TheCallResult);
221}
222
223// Get the valid immediate range for the specified NEON type code.
224static unsigned RFT(unsigned t, bool shift = false) {
225  bool quad = t & 0x10;
226
227  switch (t & 0x7) {
228    case 0: // i8
229      return shift ? 7 : (8 << (int)quad) - 1;
230    case 1: // i16
231      return shift ? 15 : (4 << (int)quad) - 1;
232    case 2: // i32
233      return shift ? 31 : (2 << (int)quad) - 1;
234    case 3: // i64
235      return shift ? 63 : (1 << (int)quad) - 1;
236    case 4: // f32
237      assert(!shift && "cannot shift float types!");
238      return (2 << (int)quad) - 1;
239    case 5: // poly8
240      return shift ? 7 : (8 << (int)quad) - 1;
241    case 6: // poly16
242      return shift ? 15 : (4 << (int)quad) - 1;
243    case 7: // float16
244      assert(!shift && "cannot shift float types!");
245      return (4 << (int)quad) - 1;
246  }
247  return 0;
248}
249
250bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
251  llvm::APSInt Result;
252
253  unsigned mask = 0;
254  unsigned TV = 0;
255  switch (BuiltinID) {
256#define GET_NEON_OVERLOAD_CHECK
257#include "clang/Basic/arm_neon.inc"
258#undef GET_NEON_OVERLOAD_CHECK
259  }
260
261  // For NEON intrinsics which are overloaded on vector element type, validate
262  // the immediate which specifies which variant to emit.
263  if (mask) {
264    unsigned ArgNo = TheCall->getNumArgs()-1;
265    if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
266      return true;
267
268    TV = Result.getLimitedValue(32);
269    if ((TV > 31) || (mask & (1 << TV)) == 0)
270      return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code)
271        << TheCall->getArg(ArgNo)->getSourceRange();
272  }
273
274  // For NEON intrinsics which take an immediate value as part of the
275  // instruction, range check them here.
276  unsigned i = 0, l = 0, u = 0;
277  switch (BuiltinID) {
278  default: return false;
279  case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break;
280  case ARM::BI__builtin_arm_usat: i = 1; u = 31; break;
281  case ARM::BI__builtin_arm_vcvtr_f:
282  case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break;
283#define GET_NEON_IMMEDIATE_CHECK
284#include "clang/Basic/arm_neon.inc"
285#undef GET_NEON_IMMEDIATE_CHECK
286  };
287
288  // Check that the immediate argument is actually a constant.
289  if (SemaBuiltinConstantArg(TheCall, i, Result))
290    return true;
291
292  // Range check against the upper/lower values for this isntruction.
293  unsigned Val = Result.getZExtValue();
294  if (Val < l || Val > (u + l))
295    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
296      << l << u+l << TheCall->getArg(i)->getSourceRange();
297
298  // FIXME: VFP Intrinsics should error if VFP not present.
299  return false;
300}
301
302/// CheckFunctionCall - Check a direct function call for various correctness
303/// and safety properties not strictly enforced by the C type system.
304bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) {
305  // Get the IdentifierInfo* for the called function.
306  IdentifierInfo *FnInfo = FDecl->getIdentifier();
307
308  // None of the checks below are needed for functions that don't have
309  // simple names (e.g., C++ conversion functions).
310  if (!FnInfo)
311    return false;
312
313  // FIXME: This mechanism should be abstracted to be less fragile and
314  // more efficient. For example, just map function ids to custom
315  // handlers.
316
317  // Printf and scanf checking.
318  for (specific_attr_iterator<FormatAttr>
319         i = FDecl->specific_attr_begin<FormatAttr>(),
320         e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) {
321
322    const FormatAttr *Format = *i;
323    const bool b = Format->getType() == "scanf";
324    if (b || CheckablePrintfAttr(Format, TheCall)) {
325      bool HasVAListArg = Format->getFirstArg() == 0;
326      CheckPrintfScanfArguments(TheCall, HasVAListArg,
327                                Format->getFormatIdx() - 1,
328                                HasVAListArg ? 0 : Format->getFirstArg() - 1,
329                                !b);
330    }
331  }
332
333  for (specific_attr_iterator<NonNullAttr>
334         i = FDecl->specific_attr_begin<NonNullAttr>(),
335         e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) {
336    CheckNonNullArguments(*i, TheCall->getArgs(),
337                          TheCall->getCallee()->getLocStart());
338  }
339
340  // Builtin handling
341  int CMF = -1;
342  switch (FDecl->getBuiltinID()) {
343  case Builtin::BI__builtin_memset:
344  case Builtin::BI__builtin___memset_chk:
345  case Builtin::BImemset:
346    CMF = CMF_Memset;
347    break;
348
349  case Builtin::BI__builtin_memcpy:
350  case Builtin::BI__builtin___memcpy_chk:
351  case Builtin::BImemcpy:
352    CMF = CMF_Memcpy;
353    break;
354
355  case Builtin::BI__builtin_memmove:
356  case Builtin::BI__builtin___memmove_chk:
357  case Builtin::BImemmove:
358    CMF = CMF_Memmove;
359    break;
360
361  case Builtin::BIstrlcpy:
362  case Builtin::BIstrlcat:
363    CheckStrlcpycatArguments(TheCall, FnInfo);
364    break;
365
366  case Builtin::BI__builtin_memcmp:
367    CMF = CMF_Memcmp;
368    break;
369
370  default:
371    if (FDecl->getLinkage() == ExternalLinkage &&
372        (!getLangOptions().CPlusPlus || FDecl->isExternC())) {
373      if (FnInfo->isStr("memset"))
374        CMF = CMF_Memset;
375      else if (FnInfo->isStr("memcpy"))
376        CMF = CMF_Memcpy;
377      else if (FnInfo->isStr("memmove"))
378        CMF = CMF_Memmove;
379      else if (FnInfo->isStr("memcmp"))
380        CMF = CMF_Memcmp;
381    }
382    break;
383  }
384
385  // Memset/memcpy/memmove handling
386  if (CMF != -1)
387    CheckMemaccessArguments(TheCall, CheckedMemoryFunction(CMF), FnInfo);
388
389  return false;
390}
391
392bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
393  // Printf checking.
394  const FormatAttr *Format = NDecl->getAttr<FormatAttr>();
395  if (!Format)
396    return false;
397
398  const VarDecl *V = dyn_cast<VarDecl>(NDecl);
399  if (!V)
400    return false;
401
402  QualType Ty = V->getType();
403  if (!Ty->isBlockPointerType())
404    return false;
405
406  const bool b = Format->getType() == "scanf";
407  if (!b && !CheckablePrintfAttr(Format, TheCall))
408    return false;
409
410  bool HasVAListArg = Format->getFirstArg() == 0;
411  CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
412                            HasVAListArg ? 0 : Format->getFirstArg() - 1, !b);
413
414  return false;
415}
416
417/// checkBuiltinArgument - Given a call to a builtin function, perform
418/// normal type-checking on the given argument, updating the call in
419/// place.  This is useful when a builtin function requires custom
420/// type-checking for some of its arguments but not necessarily all of
421/// them.
422///
423/// Returns true on error.
424static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) {
425  FunctionDecl *Fn = E->getDirectCallee();
426  assert(Fn && "builtin call without direct callee!");
427
428  ParmVarDecl *Param = Fn->getParamDecl(ArgIndex);
429  InitializedEntity Entity =
430    InitializedEntity::InitializeParameter(S.Context, Param);
431
432  ExprResult Arg = E->getArg(0);
433  Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg);
434  if (Arg.isInvalid())
435    return true;
436
437  E->setArg(ArgIndex, Arg.take());
438  return false;
439}
440
441/// SemaBuiltinAtomicOverloaded - We have a call to a function like
442/// __sync_fetch_and_add, which is an overloaded function based on the pointer
443/// type of its first argument.  The main ActOnCallExpr routines have already
444/// promoted the types of arguments because all of these calls are prototyped as
445/// void(...).
446///
447/// This function goes through and does final semantic checking for these
448/// builtins,
449ExprResult
450Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
451  CallExpr *TheCall = (CallExpr *)TheCallResult.get();
452  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
453  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
454
455  // Ensure that we have at least one argument to do type inference from.
456  if (TheCall->getNumArgs() < 1) {
457    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
458      << 0 << 1 << TheCall->getNumArgs()
459      << TheCall->getCallee()->getSourceRange();
460    return ExprError();
461  }
462
463  // Inspect the first argument of the atomic builtin.  This should always be
464  // a pointer type, whose element is an integral scalar or pointer type.
465  // Because it is a pointer type, we don't have to worry about any implicit
466  // casts here.
467  // FIXME: We don't allow floating point scalars as input.
468  Expr *FirstArg = TheCall->getArg(0);
469  const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>();
470  if (!pointerType) {
471    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
472      << FirstArg->getType() << FirstArg->getSourceRange();
473    return ExprError();
474  }
475
476  QualType ValType = pointerType->getPointeeType();
477  if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
478      !ValType->isBlockPointerType()) {
479    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
480      << FirstArg->getType() << FirstArg->getSourceRange();
481    return ExprError();
482  }
483
484  switch (ValType.getObjCLifetime()) {
485  case Qualifiers::OCL_None:
486  case Qualifiers::OCL_ExplicitNone:
487    // okay
488    break;
489
490  case Qualifiers::OCL_Weak:
491  case Qualifiers::OCL_Strong:
492  case Qualifiers::OCL_Autoreleasing:
493    Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
494      << ValType << FirstArg->getSourceRange();
495    return ExprError();
496  }
497
498  // The majority of builtins return a value, but a few have special return
499  // types, so allow them to override appropriately below.
500  QualType ResultType = ValType;
501
502  // We need to figure out which concrete builtin this maps onto.  For example,
503  // __sync_fetch_and_add with a 2 byte object turns into
504  // __sync_fetch_and_add_2.
505#define BUILTIN_ROW(x) \
506  { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
507    Builtin::BI##x##_8, Builtin::BI##x##_16 }
508
509  static const unsigned BuiltinIndices[][5] = {
510    BUILTIN_ROW(__sync_fetch_and_add),
511    BUILTIN_ROW(__sync_fetch_and_sub),
512    BUILTIN_ROW(__sync_fetch_and_or),
513    BUILTIN_ROW(__sync_fetch_and_and),
514    BUILTIN_ROW(__sync_fetch_and_xor),
515
516    BUILTIN_ROW(__sync_add_and_fetch),
517    BUILTIN_ROW(__sync_sub_and_fetch),
518    BUILTIN_ROW(__sync_and_and_fetch),
519    BUILTIN_ROW(__sync_or_and_fetch),
520    BUILTIN_ROW(__sync_xor_and_fetch),
521
522    BUILTIN_ROW(__sync_val_compare_and_swap),
523    BUILTIN_ROW(__sync_bool_compare_and_swap),
524    BUILTIN_ROW(__sync_lock_test_and_set),
525    BUILTIN_ROW(__sync_lock_release),
526    BUILTIN_ROW(__sync_swap)
527  };
528#undef BUILTIN_ROW
529
530  // Determine the index of the size.
531  unsigned SizeIndex;
532  switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
533  case 1: SizeIndex = 0; break;
534  case 2: SizeIndex = 1; break;
535  case 4: SizeIndex = 2; break;
536  case 8: SizeIndex = 3; break;
537  case 16: SizeIndex = 4; break;
538  default:
539    Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
540      << FirstArg->getType() << FirstArg->getSourceRange();
541    return ExprError();
542  }
543
544  // Each of these builtins has one pointer argument, followed by some number of
545  // values (0, 1 or 2) followed by a potentially empty varags list of stuff
546  // that we ignore.  Find out which row of BuiltinIndices to read from as well
547  // as the number of fixed args.
548  unsigned BuiltinID = FDecl->getBuiltinID();
549  unsigned BuiltinIndex, NumFixed = 1;
550  switch (BuiltinID) {
551  default: assert(0 && "Unknown overloaded atomic builtin!");
552  case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break;
553  case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break;
554  case Builtin::BI__sync_fetch_and_or:  BuiltinIndex = 2; break;
555  case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break;
556  case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break;
557
558  case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break;
559  case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break;
560  case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break;
561  case Builtin::BI__sync_or_and_fetch:  BuiltinIndex = 8; break;
562  case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break;
563
564  case Builtin::BI__sync_val_compare_and_swap:
565    BuiltinIndex = 10;
566    NumFixed = 2;
567    break;
568  case Builtin::BI__sync_bool_compare_and_swap:
569    BuiltinIndex = 11;
570    NumFixed = 2;
571    ResultType = Context.BoolTy;
572    break;
573  case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break;
574  case Builtin::BI__sync_lock_release:
575    BuiltinIndex = 13;
576    NumFixed = 0;
577    ResultType = Context.VoidTy;
578    break;
579  case Builtin::BI__sync_swap: BuiltinIndex = 14; break;
580  }
581
582  // Now that we know how many fixed arguments we expect, first check that we
583  // have at least that many.
584  if (TheCall->getNumArgs() < 1+NumFixed) {
585    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
586      << 0 << 1+NumFixed << TheCall->getNumArgs()
587      << TheCall->getCallee()->getSourceRange();
588    return ExprError();
589  }
590
591  // Get the decl for the concrete builtin from this, we can tell what the
592  // concrete integer type we should convert to is.
593  unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
594  const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
595  IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
596  FunctionDecl *NewBuiltinDecl =
597    cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
598                                           TUScope, false, DRE->getLocStart()));
599
600  // The first argument --- the pointer --- has a fixed type; we
601  // deduce the types of the rest of the arguments accordingly.  Walk
602  // the remaining arguments, converting them to the deduced value type.
603  for (unsigned i = 0; i != NumFixed; ++i) {
604    ExprResult Arg = TheCall->getArg(i+1);
605
606    // If the argument is an implicit cast, then there was a promotion due to
607    // "...", just remove it now.
608    if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) {
609      Arg = ICE->getSubExpr();
610      ICE->setSubExpr(0);
611      TheCall->setArg(i+1, Arg.get());
612    }
613
614    // GCC does an implicit conversion to the pointer or integer ValType.  This
615    // can fail in some cases (1i -> int**), check for this error case now.
616    CastKind Kind = CK_Invalid;
617    ExprValueKind VK = VK_RValue;
618    CXXCastPath BasePath;
619    Arg = CheckCastTypes(Arg.get()->getLocStart(), Arg.get()->getSourceRange(),
620                         ValType, Arg.take(), Kind, VK, BasePath);
621    if (Arg.isInvalid())
622      return ExprError();
623
624    // Okay, we have something that *can* be converted to the right type.  Check
625    // to see if there is a potentially weird extension going on here.  This can
626    // happen when you do an atomic operation on something like an char* and
627    // pass in 42.  The 42 gets converted to char.  This is even more strange
628    // for things like 45.123 -> char, etc.
629    // FIXME: Do this check.
630    Arg = ImpCastExprToType(Arg.take(), ValType, Kind, VK, &BasePath);
631    TheCall->setArg(i+1, Arg.get());
632  }
633
634  ASTContext& Context = this->getASTContext();
635
636  // Create a new DeclRefExpr to refer to the new decl.
637  DeclRefExpr* NewDRE = DeclRefExpr::Create(
638      Context,
639      DRE->getQualifierLoc(),
640      NewBuiltinDecl,
641      DRE->getLocation(),
642      NewBuiltinDecl->getType(),
643      DRE->getValueKind());
644
645  // Set the callee in the CallExpr.
646  // FIXME: This leaks the original parens and implicit casts.
647  ExprResult PromotedCall = UsualUnaryConversions(NewDRE);
648  if (PromotedCall.isInvalid())
649    return ExprError();
650  TheCall->setCallee(PromotedCall.take());
651
652  // Change the result type of the call to match the original value type. This
653  // is arbitrary, but the codegen for these builtins ins design to handle it
654  // gracefully.
655  TheCall->setType(ResultType);
656
657  return move(TheCallResult);
658}
659
660/// CheckObjCString - Checks that the argument to the builtin
661/// CFString constructor is correct
662/// Note: It might also make sense to do the UTF-16 conversion here (would
663/// simplify the backend).
664bool Sema::CheckObjCString(Expr *Arg) {
665  Arg = Arg->IgnoreParenCasts();
666  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
667
668  if (!Literal || !Literal->isAscii()) {
669    Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
670      << Arg->getSourceRange();
671    return true;
672  }
673
674  if (Literal->containsNonAsciiOrNull()) {
675    StringRef String = Literal->getString();
676    unsigned NumBytes = String.size();
677    SmallVector<UTF16, 128> ToBuf(NumBytes);
678    const UTF8 *FromPtr = (UTF8 *)String.data();
679    UTF16 *ToPtr = &ToBuf[0];
680
681    ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
682                                                 &ToPtr, ToPtr + NumBytes,
683                                                 strictConversion);
684    // Check for conversion failure.
685    if (Result != conversionOK)
686      Diag(Arg->getLocStart(),
687           diag::warn_cfstring_truncated) << Arg->getSourceRange();
688  }
689  return false;
690}
691
692/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
693/// Emit an error and return true on failure, return false on success.
694bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
695  Expr *Fn = TheCall->getCallee();
696  if (TheCall->getNumArgs() > 2) {
697    Diag(TheCall->getArg(2)->getLocStart(),
698         diag::err_typecheck_call_too_many_args)
699      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
700      << Fn->getSourceRange()
701      << SourceRange(TheCall->getArg(2)->getLocStart(),
702                     (*(TheCall->arg_end()-1))->getLocEnd());
703    return true;
704  }
705
706  if (TheCall->getNumArgs() < 2) {
707    return Diag(TheCall->getLocEnd(),
708      diag::err_typecheck_call_too_few_args_at_least)
709      << 0 /*function call*/ << 2 << TheCall->getNumArgs();
710  }
711
712  // Type-check the first argument normally.
713  if (checkBuiltinArgument(*this, TheCall, 0))
714    return true;
715
716  // Determine whether the current function is variadic or not.
717  BlockScopeInfo *CurBlock = getCurBlock();
718  bool isVariadic;
719  if (CurBlock)
720    isVariadic = CurBlock->TheDecl->isVariadic();
721  else if (FunctionDecl *FD = getCurFunctionDecl())
722    isVariadic = FD->isVariadic();
723  else
724    isVariadic = getCurMethodDecl()->isVariadic();
725
726  if (!isVariadic) {
727    Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
728    return true;
729  }
730
731  // Verify that the second argument to the builtin is the last argument of the
732  // current function or method.
733  bool SecondArgIsLastNamedArgument = false;
734  const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
735
736  if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
737    if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
738      // FIXME: This isn't correct for methods (results in bogus warning).
739      // Get the last formal in the current function.
740      const ParmVarDecl *LastArg;
741      if (CurBlock)
742        LastArg = *(CurBlock->TheDecl->param_end()-1);
743      else if (FunctionDecl *FD = getCurFunctionDecl())
744        LastArg = *(FD->param_end()-1);
745      else
746        LastArg = *(getCurMethodDecl()->param_end()-1);
747      SecondArgIsLastNamedArgument = PV == LastArg;
748    }
749  }
750
751  if (!SecondArgIsLastNamedArgument)
752    Diag(TheCall->getArg(1)->getLocStart(),
753         diag::warn_second_parameter_of_va_start_not_last_named_argument);
754  return false;
755}
756
757/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
758/// friends.  This is declared to take (...), so we have to check everything.
759bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
760  if (TheCall->getNumArgs() < 2)
761    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
762      << 0 << 2 << TheCall->getNumArgs()/*function call*/;
763  if (TheCall->getNumArgs() > 2)
764    return Diag(TheCall->getArg(2)->getLocStart(),
765                diag::err_typecheck_call_too_many_args)
766      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
767      << SourceRange(TheCall->getArg(2)->getLocStart(),
768                     (*(TheCall->arg_end()-1))->getLocEnd());
769
770  ExprResult OrigArg0 = TheCall->getArg(0);
771  ExprResult OrigArg1 = TheCall->getArg(1);
772
773  // Do standard promotions between the two arguments, returning their common
774  // type.
775  QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
776  if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
777    return true;
778
779  // Make sure any conversions are pushed back into the call; this is
780  // type safe since unordered compare builtins are declared as "_Bool
781  // foo(...)".
782  TheCall->setArg(0, OrigArg0.get());
783  TheCall->setArg(1, OrigArg1.get());
784
785  if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
786    return false;
787
788  // If the common type isn't a real floating type, then the arguments were
789  // invalid for this operation.
790  if (!Res->isRealFloatingType())
791    return Diag(OrigArg0.get()->getLocStart(),
792                diag::err_typecheck_call_invalid_ordered_compare)
793      << OrigArg0.get()->getType() << OrigArg1.get()->getType()
794      << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
795
796  return false;
797}
798
799/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
800/// __builtin_isnan and friends.  This is declared to take (...), so we have
801/// to check everything. We expect the last argument to be a floating point
802/// value.
803bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
804  if (TheCall->getNumArgs() < NumArgs)
805    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
806      << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
807  if (TheCall->getNumArgs() > NumArgs)
808    return Diag(TheCall->getArg(NumArgs)->getLocStart(),
809                diag::err_typecheck_call_too_many_args)
810      << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
811      << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
812                     (*(TheCall->arg_end()-1))->getLocEnd());
813
814  Expr *OrigArg = TheCall->getArg(NumArgs-1);
815
816  if (OrigArg->isTypeDependent())
817    return false;
818
819  // This operation requires a non-_Complex floating-point number.
820  if (!OrigArg->getType()->isRealFloatingType())
821    return Diag(OrigArg->getLocStart(),
822                diag::err_typecheck_call_invalid_unary_fp)
823      << OrigArg->getType() << OrigArg->getSourceRange();
824
825  // If this is an implicit conversion from float -> double, remove it.
826  if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
827    Expr *CastArg = Cast->getSubExpr();
828    if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
829      assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
830             "promotion from float to double is the only expected cast here");
831      Cast->setSubExpr(0);
832      TheCall->setArg(NumArgs-1, CastArg);
833      OrigArg = CastArg;
834    }
835  }
836
837  return false;
838}
839
840/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
841// This is declared to take (...), so we have to check everything.
842ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
843  if (TheCall->getNumArgs() < 2)
844    return ExprError(Diag(TheCall->getLocEnd(),
845                          diag::err_typecheck_call_too_few_args_at_least)
846      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
847      << TheCall->getSourceRange());
848
849  // Determine which of the following types of shufflevector we're checking:
850  // 1) unary, vector mask: (lhs, mask)
851  // 2) binary, vector mask: (lhs, rhs, mask)
852  // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
853  QualType resType = TheCall->getArg(0)->getType();
854  unsigned numElements = 0;
855
856  if (!TheCall->getArg(0)->isTypeDependent() &&
857      !TheCall->getArg(1)->isTypeDependent()) {
858    QualType LHSType = TheCall->getArg(0)->getType();
859    QualType RHSType = TheCall->getArg(1)->getType();
860
861    if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
862      Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
863        << SourceRange(TheCall->getArg(0)->getLocStart(),
864                       TheCall->getArg(1)->getLocEnd());
865      return ExprError();
866    }
867
868    numElements = LHSType->getAs<VectorType>()->getNumElements();
869    unsigned numResElements = TheCall->getNumArgs() - 2;
870
871    // Check to see if we have a call with 2 vector arguments, the unary shuffle
872    // with mask.  If so, verify that RHS is an integer vector type with the
873    // same number of elts as lhs.
874    if (TheCall->getNumArgs() == 2) {
875      if (!RHSType->hasIntegerRepresentation() ||
876          RHSType->getAs<VectorType>()->getNumElements() != numElements)
877        Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
878          << SourceRange(TheCall->getArg(1)->getLocStart(),
879                         TheCall->getArg(1)->getLocEnd());
880      numResElements = numElements;
881    }
882    else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
883      Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
884        << SourceRange(TheCall->getArg(0)->getLocStart(),
885                       TheCall->getArg(1)->getLocEnd());
886      return ExprError();
887    } else if (numElements != numResElements) {
888      QualType eltType = LHSType->getAs<VectorType>()->getElementType();
889      resType = Context.getVectorType(eltType, numResElements,
890                                      VectorType::GenericVector);
891    }
892  }
893
894  for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
895    if (TheCall->getArg(i)->isTypeDependent() ||
896        TheCall->getArg(i)->isValueDependent())
897      continue;
898
899    llvm::APSInt Result(32);
900    if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
901      return ExprError(Diag(TheCall->getLocStart(),
902                  diag::err_shufflevector_nonconstant_argument)
903                << TheCall->getArg(i)->getSourceRange());
904
905    if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
906      return ExprError(Diag(TheCall->getLocStart(),
907                  diag::err_shufflevector_argument_too_large)
908               << TheCall->getArg(i)->getSourceRange());
909  }
910
911  SmallVector<Expr*, 32> exprs;
912
913  for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
914    exprs.push_back(TheCall->getArg(i));
915    TheCall->setArg(i, 0);
916  }
917
918  return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
919                                            exprs.size(), resType,
920                                            TheCall->getCallee()->getLocStart(),
921                                            TheCall->getRParenLoc()));
922}
923
924/// SemaBuiltinPrefetch - Handle __builtin_prefetch.
925// This is declared to take (const void*, ...) and can take two
926// optional constant int args.
927bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
928  unsigned NumArgs = TheCall->getNumArgs();
929
930  if (NumArgs > 3)
931    return Diag(TheCall->getLocEnd(),
932             diag::err_typecheck_call_too_many_args_at_most)
933             << 0 /*function call*/ << 3 << NumArgs
934             << TheCall->getSourceRange();
935
936  // Argument 0 is checked for us and the remaining arguments must be
937  // constant integers.
938  for (unsigned i = 1; i != NumArgs; ++i) {
939    Expr *Arg = TheCall->getArg(i);
940
941    llvm::APSInt Result;
942    if (SemaBuiltinConstantArg(TheCall, i, Result))
943      return true;
944
945    // FIXME: gcc issues a warning and rewrites these to 0. These
946    // seems especially odd for the third argument since the default
947    // is 3.
948    if (i == 1) {
949      if (Result.getLimitedValue() > 1)
950        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
951             << "0" << "1" << Arg->getSourceRange();
952    } else {
953      if (Result.getLimitedValue() > 3)
954        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
955            << "0" << "3" << Arg->getSourceRange();
956    }
957  }
958
959  return false;
960}
961
962/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
963/// TheCall is a constant expression.
964bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
965                                  llvm::APSInt &Result) {
966  Expr *Arg = TheCall->getArg(ArgNum);
967  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
968  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
969
970  if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
971
972  if (!Arg->isIntegerConstantExpr(Result, Context))
973    return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
974                << FDecl->getDeclName() <<  Arg->getSourceRange();
975
976  return false;
977}
978
979/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
980/// int type). This simply type checks that type is one of the defined
981/// constants (0-3).
982// For compatibility check 0-3, llvm only handles 0 and 2.
983bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
984  llvm::APSInt Result;
985
986  // Check constant-ness first.
987  if (SemaBuiltinConstantArg(TheCall, 1, Result))
988    return true;
989
990  Expr *Arg = TheCall->getArg(1);
991  if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
992    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
993             << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
994  }
995
996  return false;
997}
998
999/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
1000/// This checks that val is a constant 1.
1001bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
1002  Expr *Arg = TheCall->getArg(1);
1003  llvm::APSInt Result;
1004
1005  // TODO: This is less than ideal. Overload this to take a value.
1006  if (SemaBuiltinConstantArg(TheCall, 1, Result))
1007    return true;
1008
1009  if (Result != 1)
1010    return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
1011             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1012
1013  return false;
1014}
1015
1016// Handle i > 1 ? "x" : "y", recursively.
1017bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall,
1018                                  bool HasVAListArg,
1019                                  unsigned format_idx, unsigned firstDataArg,
1020                                  bool isPrintf) {
1021 tryAgain:
1022  if (E->isTypeDependent() || E->isValueDependent())
1023    return false;
1024
1025  E = E->IgnoreParens();
1026
1027  switch (E->getStmtClass()) {
1028  case Stmt::BinaryConditionalOperatorClass:
1029  case Stmt::ConditionalOperatorClass: {
1030    const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E);
1031    return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg,
1032                                  format_idx, firstDataArg, isPrintf)
1033        && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg,
1034                                  format_idx, firstDataArg, isPrintf);
1035  }
1036
1037  case Stmt::IntegerLiteralClass:
1038    // Technically -Wformat-nonliteral does not warn about this case.
1039    // The behavior of printf and friends in this case is implementation
1040    // dependent.  Ideally if the format string cannot be null then
1041    // it should have a 'nonnull' attribute in the function prototype.
1042    return true;
1043
1044  case Stmt::ImplicitCastExprClass: {
1045    E = cast<ImplicitCastExpr>(E)->getSubExpr();
1046    goto tryAgain;
1047  }
1048
1049  case Stmt::OpaqueValueExprClass:
1050    if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
1051      E = src;
1052      goto tryAgain;
1053    }
1054    return false;
1055
1056  case Stmt::PredefinedExprClass:
1057    // While __func__, etc., are technically not string literals, they
1058    // cannot contain format specifiers and thus are not a security
1059    // liability.
1060    return true;
1061
1062  case Stmt::DeclRefExprClass: {
1063    const DeclRefExpr *DR = cast<DeclRefExpr>(E);
1064
1065    // As an exception, do not flag errors for variables binding to
1066    // const string literals.
1067    if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
1068      bool isConstant = false;
1069      QualType T = DR->getType();
1070
1071      if (const ArrayType *AT = Context.getAsArrayType(T)) {
1072        isConstant = AT->getElementType().isConstant(Context);
1073      } else if (const PointerType *PT = T->getAs<PointerType>()) {
1074        isConstant = T.isConstant(Context) &&
1075                     PT->getPointeeType().isConstant(Context);
1076      }
1077
1078      if (isConstant) {
1079        if (const Expr *Init = VD->getAnyInitializer())
1080          return SemaCheckStringLiteral(Init, TheCall,
1081                                        HasVAListArg, format_idx, firstDataArg,
1082                                        isPrintf);
1083      }
1084
1085      // For vprintf* functions (i.e., HasVAListArg==true), we add a
1086      // special check to see if the format string is a function parameter
1087      // of the function calling the printf function.  If the function
1088      // has an attribute indicating it is a printf-like function, then we
1089      // should suppress warnings concerning non-literals being used in a call
1090      // to a vprintf function.  For example:
1091      //
1092      // void
1093      // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
1094      //      va_list ap;
1095      //      va_start(ap, fmt);
1096      //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
1097      //      ...
1098      //
1099      //
1100      //  FIXME: We don't have full attribute support yet, so just check to see
1101      //    if the argument is a DeclRefExpr that references a parameter.  We'll
1102      //    add proper support for checking the attribute later.
1103      if (HasVAListArg)
1104        if (isa<ParmVarDecl>(VD))
1105          return true;
1106    }
1107
1108    return false;
1109  }
1110
1111  case Stmt::CallExprClass: {
1112    const CallExpr *CE = cast<CallExpr>(E);
1113    if (const ImplicitCastExpr *ICE
1114          = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
1115      if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
1116        if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
1117          if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
1118            unsigned ArgIndex = FA->getFormatIdx();
1119            const Expr *Arg = CE->getArg(ArgIndex - 1);
1120
1121            return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg,
1122                                          format_idx, firstDataArg, isPrintf);
1123          }
1124        }
1125      }
1126    }
1127
1128    return false;
1129  }
1130  case Stmt::ObjCStringLiteralClass:
1131  case Stmt::StringLiteralClass: {
1132    const StringLiteral *StrE = NULL;
1133
1134    if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1135      StrE = ObjCFExpr->getString();
1136    else
1137      StrE = cast<StringLiteral>(E);
1138
1139    if (StrE) {
1140      CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx,
1141                        firstDataArg, isPrintf);
1142      return true;
1143    }
1144
1145    return false;
1146  }
1147
1148  default:
1149    return false;
1150  }
1151}
1152
1153void
1154Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1155                            const Expr * const *ExprArgs,
1156                            SourceLocation CallSiteLoc) {
1157  for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1158                                  e = NonNull->args_end();
1159       i != e; ++i) {
1160    const Expr *ArgExpr = ExprArgs[*i];
1161    if (ArgExpr->isNullPointerConstant(Context,
1162                                       Expr::NPC_ValueDependentIsNotNull))
1163      Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1164  }
1165}
1166
1167/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar
1168/// functions) for correct use of format strings.
1169void
1170Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg,
1171                                unsigned format_idx, unsigned firstDataArg,
1172                                bool isPrintf) {
1173
1174  const Expr *Fn = TheCall->getCallee();
1175
1176  // The way the format attribute works in GCC, the implicit this argument
1177  // of member functions is counted. However, it doesn't appear in our own
1178  // lists, so decrement format_idx in that case.
1179  if (isa<CXXMemberCallExpr>(TheCall)) {
1180    const CXXMethodDecl *method_decl =
1181      dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl());
1182    if (method_decl && method_decl->isInstance()) {
1183      // Catch a format attribute mistakenly referring to the object argument.
1184      if (format_idx == 0)
1185        return;
1186      --format_idx;
1187      if(firstDataArg != 0)
1188        --firstDataArg;
1189    }
1190  }
1191
1192  // CHECK: printf/scanf-like function is called with no format string.
1193  if (format_idx >= TheCall->getNumArgs()) {
1194    Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string)
1195      << Fn->getSourceRange();
1196    return;
1197  }
1198
1199  const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts();
1200
1201  // CHECK: format string is not a string literal.
1202  //
1203  // Dynamically generated format strings are difficult to
1204  // automatically vet at compile time.  Requiring that format strings
1205  // are string literals: (1) permits the checking of format strings by
1206  // the compiler and thereby (2) can practically remove the source of
1207  // many format string exploits.
1208
1209  // Format string can be either ObjC string (e.g. @"%d") or
1210  // C string (e.g. "%d")
1211  // ObjC string uses the same format specifiers as C string, so we can use
1212  // the same format string checking logic for both ObjC and C strings.
1213  if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx,
1214                             firstDataArg, isPrintf))
1215    return;  // Literal format string found, check done!
1216
1217  // If there are no arguments specified, warn with -Wformat-security, otherwise
1218  // warn only with -Wformat-nonliteral.
1219  if (TheCall->getNumArgs() == format_idx+1)
1220    Diag(TheCall->getArg(format_idx)->getLocStart(),
1221         diag::warn_format_nonliteral_noargs)
1222      << OrigFormatExpr->getSourceRange();
1223  else
1224    Diag(TheCall->getArg(format_idx)->getLocStart(),
1225         diag::warn_format_nonliteral)
1226           << OrigFormatExpr->getSourceRange();
1227}
1228
1229namespace {
1230class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1231protected:
1232  Sema &S;
1233  const StringLiteral *FExpr;
1234  const Expr *OrigFormatExpr;
1235  const unsigned FirstDataArg;
1236  const unsigned NumDataArgs;
1237  const bool IsObjCLiteral;
1238  const char *Beg; // Start of format string.
1239  const bool HasVAListArg;
1240  const CallExpr *TheCall;
1241  unsigned FormatIdx;
1242  llvm::BitVector CoveredArgs;
1243  bool usesPositionalArgs;
1244  bool atFirstArg;
1245public:
1246  CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1247                     const Expr *origFormatExpr, unsigned firstDataArg,
1248                     unsigned numDataArgs, bool isObjCLiteral,
1249                     const char *beg, bool hasVAListArg,
1250                     const CallExpr *theCall, unsigned formatIdx)
1251    : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1252      FirstDataArg(firstDataArg),
1253      NumDataArgs(numDataArgs),
1254      IsObjCLiteral(isObjCLiteral), Beg(beg),
1255      HasVAListArg(hasVAListArg),
1256      TheCall(theCall), FormatIdx(formatIdx),
1257      usesPositionalArgs(false), atFirstArg(true) {
1258        CoveredArgs.resize(numDataArgs);
1259        CoveredArgs.reset();
1260      }
1261
1262  void DoneProcessing();
1263
1264  void HandleIncompleteSpecifier(const char *startSpecifier,
1265                                 unsigned specifierLen);
1266
1267  virtual void HandleInvalidPosition(const char *startSpecifier,
1268                                     unsigned specifierLen,
1269                                     analyze_format_string::PositionContext p);
1270
1271  virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
1272
1273  void HandleNullChar(const char *nullCharacter);
1274
1275protected:
1276  bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
1277                                        const char *startSpec,
1278                                        unsigned specifierLen,
1279                                        const char *csStart, unsigned csLen);
1280
1281  SourceRange getFormatStringRange();
1282  CharSourceRange getSpecifierRange(const char *startSpecifier,
1283                                    unsigned specifierLen);
1284  SourceLocation getLocationOfByte(const char *x);
1285
1286  const Expr *getDataArg(unsigned i) const;
1287
1288  bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
1289                    const analyze_format_string::ConversionSpecifier &CS,
1290                    const char *startSpecifier, unsigned specifierLen,
1291                    unsigned argIndex);
1292};
1293}
1294
1295SourceRange CheckFormatHandler::getFormatStringRange() {
1296  return OrigFormatExpr->getSourceRange();
1297}
1298
1299CharSourceRange CheckFormatHandler::
1300getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
1301  SourceLocation Start = getLocationOfByte(startSpecifier);
1302  SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
1303
1304  // Advance the end SourceLocation by one due to half-open ranges.
1305  End = End.getFileLocWithOffset(1);
1306
1307  return CharSourceRange::getCharRange(Start, End);
1308}
1309
1310SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
1311  return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
1312}
1313
1314void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
1315                                                   unsigned specifierLen){
1316  SourceLocation Loc = getLocationOfByte(startSpecifier);
1317  S.Diag(Loc, diag::warn_printf_incomplete_specifier)
1318    << getSpecifierRange(startSpecifier, specifierLen);
1319}
1320
1321void
1322CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
1323                                     analyze_format_string::PositionContext p) {
1324  SourceLocation Loc = getLocationOfByte(startPos);
1325  S.Diag(Loc, diag::warn_format_invalid_positional_specifier)
1326    << (unsigned) p << getSpecifierRange(startPos, posLen);
1327}
1328
1329void CheckFormatHandler::HandleZeroPosition(const char *startPos,
1330                                            unsigned posLen) {
1331  SourceLocation Loc = getLocationOfByte(startPos);
1332  S.Diag(Loc, diag::warn_format_zero_positional_specifier)
1333    << getSpecifierRange(startPos, posLen);
1334}
1335
1336void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
1337  if (!IsObjCLiteral) {
1338    // The presence of a null character is likely an error.
1339    S.Diag(getLocationOfByte(nullCharacter),
1340           diag::warn_printf_format_string_contains_null_char)
1341      << getFormatStringRange();
1342  }
1343}
1344
1345const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
1346  return TheCall->getArg(FirstDataArg + i);
1347}
1348
1349void CheckFormatHandler::DoneProcessing() {
1350    // Does the number of data arguments exceed the number of
1351    // format conversions in the format string?
1352  if (!HasVAListArg) {
1353      // Find any arguments that weren't covered.
1354    CoveredArgs.flip();
1355    signed notCoveredArg = CoveredArgs.find_first();
1356    if (notCoveredArg >= 0) {
1357      assert((unsigned)notCoveredArg < NumDataArgs);
1358      S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(),
1359             diag::warn_printf_data_arg_not_used)
1360      << getFormatStringRange();
1361    }
1362  }
1363}
1364
1365bool
1366CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
1367                                                     SourceLocation Loc,
1368                                                     const char *startSpec,
1369                                                     unsigned specifierLen,
1370                                                     const char *csStart,
1371                                                     unsigned csLen) {
1372
1373  bool keepGoing = true;
1374  if (argIndex < NumDataArgs) {
1375    // Consider the argument coverered, even though the specifier doesn't
1376    // make sense.
1377    CoveredArgs.set(argIndex);
1378  }
1379  else {
1380    // If argIndex exceeds the number of data arguments we
1381    // don't issue a warning because that is just a cascade of warnings (and
1382    // they may have intended '%%' anyway). We don't want to continue processing
1383    // the format string after this point, however, as we will like just get
1384    // gibberish when trying to match arguments.
1385    keepGoing = false;
1386  }
1387
1388  S.Diag(Loc, diag::warn_format_invalid_conversion)
1389    << StringRef(csStart, csLen)
1390    << getSpecifierRange(startSpec, specifierLen);
1391
1392  return keepGoing;
1393}
1394
1395bool
1396CheckFormatHandler::CheckNumArgs(
1397  const analyze_format_string::FormatSpecifier &FS,
1398  const analyze_format_string::ConversionSpecifier &CS,
1399  const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
1400
1401  if (argIndex >= NumDataArgs) {
1402    if (FS.usesPositionalArg())  {
1403      S.Diag(getLocationOfByte(CS.getStart()),
1404             diag::warn_printf_positional_arg_exceeds_data_args)
1405      << (argIndex+1) << NumDataArgs
1406      << getSpecifierRange(startSpecifier, specifierLen);
1407    }
1408    else {
1409      S.Diag(getLocationOfByte(CS.getStart()),
1410             diag::warn_printf_insufficient_data_args)
1411      << getSpecifierRange(startSpecifier, specifierLen);
1412    }
1413
1414    return false;
1415  }
1416  return true;
1417}
1418
1419//===--- CHECK: Printf format string checking ------------------------------===//
1420
1421namespace {
1422class CheckPrintfHandler : public CheckFormatHandler {
1423public:
1424  CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
1425                     const Expr *origFormatExpr, unsigned firstDataArg,
1426                     unsigned numDataArgs, bool isObjCLiteral,
1427                     const char *beg, bool hasVAListArg,
1428                     const CallExpr *theCall, unsigned formatIdx)
1429  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1430                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1431                       theCall, formatIdx) {}
1432
1433
1434  bool HandleInvalidPrintfConversionSpecifier(
1435                                      const analyze_printf::PrintfSpecifier &FS,
1436                                      const char *startSpecifier,
1437                                      unsigned specifierLen);
1438
1439  bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
1440                             const char *startSpecifier,
1441                             unsigned specifierLen);
1442
1443  bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
1444                    const char *startSpecifier, unsigned specifierLen);
1445  void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
1446                           const analyze_printf::OptionalAmount &Amt,
1447                           unsigned type,
1448                           const char *startSpecifier, unsigned specifierLen);
1449  void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1450                  const analyze_printf::OptionalFlag &flag,
1451                  const char *startSpecifier, unsigned specifierLen);
1452  void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
1453                         const analyze_printf::OptionalFlag &ignoredFlag,
1454                         const analyze_printf::OptionalFlag &flag,
1455                         const char *startSpecifier, unsigned specifierLen);
1456};
1457}
1458
1459bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
1460                                      const analyze_printf::PrintfSpecifier &FS,
1461                                      const char *startSpecifier,
1462                                      unsigned specifierLen) {
1463  const analyze_printf::PrintfConversionSpecifier &CS =
1464    FS.getConversionSpecifier();
1465
1466  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1467                                          getLocationOfByte(CS.getStart()),
1468                                          startSpecifier, specifierLen,
1469                                          CS.getStart(), CS.getLength());
1470}
1471
1472bool CheckPrintfHandler::HandleAmount(
1473                               const analyze_format_string::OptionalAmount &Amt,
1474                               unsigned k, const char *startSpecifier,
1475                               unsigned specifierLen) {
1476
1477  if (Amt.hasDataArgument()) {
1478    if (!HasVAListArg) {
1479      unsigned argIndex = Amt.getArgIndex();
1480      if (argIndex >= NumDataArgs) {
1481        S.Diag(getLocationOfByte(Amt.getStart()),
1482               diag::warn_printf_asterisk_missing_arg)
1483          << k << getSpecifierRange(startSpecifier, specifierLen);
1484        // Don't do any more checking.  We will just emit
1485        // spurious errors.
1486        return false;
1487      }
1488
1489      // Type check the data argument.  It should be an 'int'.
1490      // Although not in conformance with C99, we also allow the argument to be
1491      // an 'unsigned int' as that is a reasonably safe case.  GCC also
1492      // doesn't emit a warning for that case.
1493      CoveredArgs.set(argIndex);
1494      const Expr *Arg = getDataArg(argIndex);
1495      QualType T = Arg->getType();
1496
1497      const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context);
1498      assert(ATR.isValid());
1499
1500      if (!ATR.matchesType(S.Context, T)) {
1501        S.Diag(getLocationOfByte(Amt.getStart()),
1502               diag::warn_printf_asterisk_wrong_type)
1503          << k
1504          << ATR.getRepresentativeType(S.Context) << T
1505          << getSpecifierRange(startSpecifier, specifierLen)
1506          << Arg->getSourceRange();
1507        // Don't do any more checking.  We will just emit
1508        // spurious errors.
1509        return false;
1510      }
1511    }
1512  }
1513  return true;
1514}
1515
1516void CheckPrintfHandler::HandleInvalidAmount(
1517                                      const analyze_printf::PrintfSpecifier &FS,
1518                                      const analyze_printf::OptionalAmount &Amt,
1519                                      unsigned type,
1520                                      const char *startSpecifier,
1521                                      unsigned specifierLen) {
1522  const analyze_printf::PrintfConversionSpecifier &CS =
1523    FS.getConversionSpecifier();
1524  switch (Amt.getHowSpecified()) {
1525  case analyze_printf::OptionalAmount::Constant:
1526    S.Diag(getLocationOfByte(Amt.getStart()),
1527        diag::warn_printf_nonsensical_optional_amount)
1528      << type
1529      << CS.toString()
1530      << getSpecifierRange(startSpecifier, specifierLen)
1531      << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
1532          Amt.getConstantLength()));
1533    break;
1534
1535  default:
1536    S.Diag(getLocationOfByte(Amt.getStart()),
1537        diag::warn_printf_nonsensical_optional_amount)
1538      << type
1539      << CS.toString()
1540      << getSpecifierRange(startSpecifier, specifierLen);
1541    break;
1542  }
1543}
1544
1545void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1546                                    const analyze_printf::OptionalFlag &flag,
1547                                    const char *startSpecifier,
1548                                    unsigned specifierLen) {
1549  // Warn about pointless flag with a fixit removal.
1550  const analyze_printf::PrintfConversionSpecifier &CS =
1551    FS.getConversionSpecifier();
1552  S.Diag(getLocationOfByte(flag.getPosition()),
1553      diag::warn_printf_nonsensical_flag)
1554    << flag.toString() << CS.toString()
1555    << getSpecifierRange(startSpecifier, specifierLen)
1556    << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1));
1557}
1558
1559void CheckPrintfHandler::HandleIgnoredFlag(
1560                                const analyze_printf::PrintfSpecifier &FS,
1561                                const analyze_printf::OptionalFlag &ignoredFlag,
1562                                const analyze_printf::OptionalFlag &flag,
1563                                const char *startSpecifier,
1564                                unsigned specifierLen) {
1565  // Warn about ignored flag with a fixit removal.
1566  S.Diag(getLocationOfByte(ignoredFlag.getPosition()),
1567      diag::warn_printf_ignored_flag)
1568    << ignoredFlag.toString() << flag.toString()
1569    << getSpecifierRange(startSpecifier, specifierLen)
1570    << FixItHint::CreateRemoval(getSpecifierRange(
1571        ignoredFlag.getPosition(), 1));
1572}
1573
1574bool
1575CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
1576                                            &FS,
1577                                          const char *startSpecifier,
1578                                          unsigned specifierLen) {
1579
1580  using namespace analyze_format_string;
1581  using namespace analyze_printf;
1582  const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
1583
1584  if (FS.consumesDataArgument()) {
1585    if (atFirstArg) {
1586        atFirstArg = false;
1587        usesPositionalArgs = FS.usesPositionalArg();
1588    }
1589    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1590      // Cannot mix-and-match positional and non-positional arguments.
1591      S.Diag(getLocationOfByte(CS.getStart()),
1592             diag::warn_format_mix_positional_nonpositional_args)
1593        << getSpecifierRange(startSpecifier, specifierLen);
1594      return false;
1595    }
1596  }
1597
1598  // First check if the field width, precision, and conversion specifier
1599  // have matching data arguments.
1600  if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
1601                    startSpecifier, specifierLen)) {
1602    return false;
1603  }
1604
1605  if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
1606                    startSpecifier, specifierLen)) {
1607    return false;
1608  }
1609
1610  if (!CS.consumesDataArgument()) {
1611    // FIXME: Technically specifying a precision or field width here
1612    // makes no sense.  Worth issuing a warning at some point.
1613    return true;
1614  }
1615
1616  // Consume the argument.
1617  unsigned argIndex = FS.getArgIndex();
1618  if (argIndex < NumDataArgs) {
1619    // The check to see if the argIndex is valid will come later.
1620    // We set the bit here because we may exit early from this
1621    // function if we encounter some other error.
1622    CoveredArgs.set(argIndex);
1623  }
1624
1625  // Check for using an Objective-C specific conversion specifier
1626  // in a non-ObjC literal.
1627  if (!IsObjCLiteral && CS.isObjCArg()) {
1628    return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
1629                                                  specifierLen);
1630  }
1631
1632  // Check for invalid use of field width
1633  if (!FS.hasValidFieldWidth()) {
1634    HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
1635        startSpecifier, specifierLen);
1636  }
1637
1638  // Check for invalid use of precision
1639  if (!FS.hasValidPrecision()) {
1640    HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
1641        startSpecifier, specifierLen);
1642  }
1643
1644  // Check each flag does not conflict with any other component.
1645  if (!FS.hasValidThousandsGroupingPrefix())
1646    HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
1647  if (!FS.hasValidLeadingZeros())
1648    HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
1649  if (!FS.hasValidPlusPrefix())
1650    HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
1651  if (!FS.hasValidSpacePrefix())
1652    HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
1653  if (!FS.hasValidAlternativeForm())
1654    HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
1655  if (!FS.hasValidLeftJustified())
1656    HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
1657
1658  // Check that flags are not ignored by another flag
1659  if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
1660    HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
1661        startSpecifier, specifierLen);
1662  if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
1663    HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
1664            startSpecifier, specifierLen);
1665
1666  // Check the length modifier is valid with the given conversion specifier.
1667  const LengthModifier &LM = FS.getLengthModifier();
1668  if (!FS.hasValidLengthModifier())
1669    S.Diag(getLocationOfByte(LM.getStart()),
1670        diag::warn_format_nonsensical_length)
1671      << LM.toString() << CS.toString()
1672      << getSpecifierRange(startSpecifier, specifierLen)
1673      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1674          LM.getLength()));
1675
1676  // Are we using '%n'?
1677  if (CS.getKind() == ConversionSpecifier::nArg) {
1678    // Issue a warning about this being a possible security issue.
1679    S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back)
1680      << getSpecifierRange(startSpecifier, specifierLen);
1681    // Continue checking the other format specifiers.
1682    return true;
1683  }
1684
1685  // The remaining checks depend on the data arguments.
1686  if (HasVAListArg)
1687    return true;
1688
1689  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1690    return false;
1691
1692  // Now type check the data expression that matches the
1693  // format specifier.
1694  const Expr *Ex = getDataArg(argIndex);
1695  const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context);
1696  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
1697    // Check if we didn't match because of an implicit cast from a 'char'
1698    // or 'short' to an 'int'.  This is done because printf is a varargs
1699    // function.
1700    if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex))
1701      if (ICE->getType() == S.Context.IntTy) {
1702        // All further checking is done on the subexpression.
1703        Ex = ICE->getSubExpr();
1704        if (ATR.matchesType(S.Context, Ex->getType()))
1705          return true;
1706      }
1707
1708    // We may be able to offer a FixItHint if it is a supported type.
1709    PrintfSpecifier fixedFS = FS;
1710    bool success = fixedFS.fixType(Ex->getType());
1711
1712    if (success) {
1713      // Get the fix string from the fixed format specifier
1714      llvm::SmallString<128> buf;
1715      llvm::raw_svector_ostream os(buf);
1716      fixedFS.toString(os);
1717
1718      // FIXME: getRepresentativeType() perhaps should return a string
1719      // instead of a QualType to better handle when the representative
1720      // type is 'wint_t' (which is defined in the system headers).
1721      S.Diag(getLocationOfByte(CS.getStart()),
1722          diag::warn_printf_conversion_argument_type_mismatch)
1723        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1724        << getSpecifierRange(startSpecifier, specifierLen)
1725        << Ex->getSourceRange()
1726        << FixItHint::CreateReplacement(
1727            getSpecifierRange(startSpecifier, specifierLen),
1728            os.str());
1729    }
1730    else {
1731      S.Diag(getLocationOfByte(CS.getStart()),
1732             diag::warn_printf_conversion_argument_type_mismatch)
1733        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1734        << getSpecifierRange(startSpecifier, specifierLen)
1735        << Ex->getSourceRange();
1736    }
1737  }
1738
1739  return true;
1740}
1741
1742//===--- CHECK: Scanf format string checking ------------------------------===//
1743
1744namespace {
1745class CheckScanfHandler : public CheckFormatHandler {
1746public:
1747  CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
1748                    const Expr *origFormatExpr, unsigned firstDataArg,
1749                    unsigned numDataArgs, bool isObjCLiteral,
1750                    const char *beg, bool hasVAListArg,
1751                    const CallExpr *theCall, unsigned formatIdx)
1752  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1753                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1754                       theCall, formatIdx) {}
1755
1756  bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
1757                            const char *startSpecifier,
1758                            unsigned specifierLen);
1759
1760  bool HandleInvalidScanfConversionSpecifier(
1761          const analyze_scanf::ScanfSpecifier &FS,
1762          const char *startSpecifier,
1763          unsigned specifierLen);
1764
1765  void HandleIncompleteScanList(const char *start, const char *end);
1766};
1767}
1768
1769void CheckScanfHandler::HandleIncompleteScanList(const char *start,
1770                                                 const char *end) {
1771  S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete)
1772    << getSpecifierRange(start, end - start);
1773}
1774
1775bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
1776                                        const analyze_scanf::ScanfSpecifier &FS,
1777                                        const char *startSpecifier,
1778                                        unsigned specifierLen) {
1779
1780  const analyze_scanf::ScanfConversionSpecifier &CS =
1781    FS.getConversionSpecifier();
1782
1783  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1784                                          getLocationOfByte(CS.getStart()),
1785                                          startSpecifier, specifierLen,
1786                                          CS.getStart(), CS.getLength());
1787}
1788
1789bool CheckScanfHandler::HandleScanfSpecifier(
1790                                       const analyze_scanf::ScanfSpecifier &FS,
1791                                       const char *startSpecifier,
1792                                       unsigned specifierLen) {
1793
1794  using namespace analyze_scanf;
1795  using namespace analyze_format_string;
1796
1797  const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
1798
1799  // Handle case where '%' and '*' don't consume an argument.  These shouldn't
1800  // be used to decide if we are using positional arguments consistently.
1801  if (FS.consumesDataArgument()) {
1802    if (atFirstArg) {
1803      atFirstArg = false;
1804      usesPositionalArgs = FS.usesPositionalArg();
1805    }
1806    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1807      // Cannot mix-and-match positional and non-positional arguments.
1808      S.Diag(getLocationOfByte(CS.getStart()),
1809             diag::warn_format_mix_positional_nonpositional_args)
1810        << getSpecifierRange(startSpecifier, specifierLen);
1811      return false;
1812    }
1813  }
1814
1815  // Check if the field with is non-zero.
1816  const OptionalAmount &Amt = FS.getFieldWidth();
1817  if (Amt.getHowSpecified() == OptionalAmount::Constant) {
1818    if (Amt.getConstantAmount() == 0) {
1819      const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
1820                                                   Amt.getConstantLength());
1821      S.Diag(getLocationOfByte(Amt.getStart()),
1822             diag::warn_scanf_nonzero_width)
1823        << R << FixItHint::CreateRemoval(R);
1824    }
1825  }
1826
1827  if (!FS.consumesDataArgument()) {
1828    // FIXME: Technically specifying a precision or field width here
1829    // makes no sense.  Worth issuing a warning at some point.
1830    return true;
1831  }
1832
1833  // Consume the argument.
1834  unsigned argIndex = FS.getArgIndex();
1835  if (argIndex < NumDataArgs) {
1836      // The check to see if the argIndex is valid will come later.
1837      // We set the bit here because we may exit early from this
1838      // function if we encounter some other error.
1839    CoveredArgs.set(argIndex);
1840  }
1841
1842  // Check the length modifier is valid with the given conversion specifier.
1843  const LengthModifier &LM = FS.getLengthModifier();
1844  if (!FS.hasValidLengthModifier()) {
1845    S.Diag(getLocationOfByte(LM.getStart()),
1846           diag::warn_format_nonsensical_length)
1847      << LM.toString() << CS.toString()
1848      << getSpecifierRange(startSpecifier, specifierLen)
1849      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1850                                                    LM.getLength()));
1851  }
1852
1853  // The remaining checks depend on the data arguments.
1854  if (HasVAListArg)
1855    return true;
1856
1857  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1858    return false;
1859
1860  // FIXME: Check that the argument type matches the format specifier.
1861
1862  return true;
1863}
1864
1865void Sema::CheckFormatString(const StringLiteral *FExpr,
1866                             const Expr *OrigFormatExpr,
1867                             const CallExpr *TheCall, bool HasVAListArg,
1868                             unsigned format_idx, unsigned firstDataArg,
1869                             bool isPrintf) {
1870
1871  // CHECK: is the format string a wide literal?
1872  if (!FExpr->isAscii()) {
1873    Diag(FExpr->getLocStart(),
1874         diag::warn_format_string_is_wide_literal)
1875    << OrigFormatExpr->getSourceRange();
1876    return;
1877  }
1878
1879  // Str - The format string.  NOTE: this is NOT null-terminated!
1880  StringRef StrRef = FExpr->getString();
1881  const char *Str = StrRef.data();
1882  unsigned StrLen = StrRef.size();
1883
1884  // CHECK: empty format string?
1885  if (StrLen == 0) {
1886    Diag(FExpr->getLocStart(), diag::warn_empty_format_string)
1887    << OrigFormatExpr->getSourceRange();
1888    return;
1889  }
1890
1891  if (isPrintf) {
1892    CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1893                         TheCall->getNumArgs() - firstDataArg,
1894                         isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1895                         HasVAListArg, TheCall, format_idx);
1896
1897    if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen))
1898      H.DoneProcessing();
1899  }
1900  else {
1901    CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1902                        TheCall->getNumArgs() - firstDataArg,
1903                        isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1904                        HasVAListArg, TheCall, format_idx);
1905
1906    if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen))
1907      H.DoneProcessing();
1908  }
1909}
1910
1911//===--- CHECK: Standard memory functions ---------------------------------===//
1912
1913/// \brief Determine whether the given type is a dynamic class type (e.g.,
1914/// whether it has a vtable).
1915static bool isDynamicClassType(QualType T) {
1916  if (CXXRecordDecl *Record = T->getAsCXXRecordDecl())
1917    if (CXXRecordDecl *Definition = Record->getDefinition())
1918      if (Definition->isDynamicClass())
1919        return true;
1920
1921  return false;
1922}
1923
1924/// \brief If E is a sizeof expression, returns its argument expression,
1925/// otherwise returns NULL.
1926static const Expr *getSizeOfExprArg(const Expr* E) {
1927  if (const UnaryExprOrTypeTraitExpr *SizeOf =
1928      dyn_cast<UnaryExprOrTypeTraitExpr>(E))
1929    if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType())
1930      return SizeOf->getArgumentExpr()->IgnoreParenImpCasts();
1931
1932  return 0;
1933}
1934
1935/// \brief If E is a sizeof expression, returns its argument type.
1936static QualType getSizeOfArgType(const Expr* E) {
1937  if (const UnaryExprOrTypeTraitExpr *SizeOf =
1938      dyn_cast<UnaryExprOrTypeTraitExpr>(E))
1939    if (SizeOf->getKind() == clang::UETT_SizeOf)
1940      return SizeOf->getTypeOfArgument();
1941
1942  return QualType();
1943}
1944
1945/// \brief Check for dangerous or invalid arguments to memset().
1946///
1947/// This issues warnings on known problematic, dangerous or unspecified
1948/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp'
1949/// function calls.
1950///
1951/// \param Call The call expression to diagnose.
1952void Sema::CheckMemaccessArguments(const CallExpr *Call,
1953                                   CheckedMemoryFunction CMF,
1954                                   IdentifierInfo *FnName) {
1955  // It is possible to have a non-standard definition of memset.  Validate
1956  // we have enough arguments, and if not, abort further checking.
1957  if (Call->getNumArgs() < 3)
1958    return;
1959
1960  unsigned LastArg = (CMF == CMF_Memset? 1 : 2);
1961  const Expr *LenExpr = Call->getArg(2)->IgnoreParenImpCasts();
1962
1963  // We have special checking when the length is a sizeof expression.
1964  QualType SizeOfArgTy = getSizeOfArgType(LenExpr);
1965  const Expr *SizeOfArg = getSizeOfExprArg(LenExpr);
1966  llvm::FoldingSetNodeID SizeOfArgID;
1967
1968  for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) {
1969    const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts();
1970    SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange();
1971
1972    QualType DestTy = Dest->getType();
1973    if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
1974      QualType PointeeTy = DestPtrTy->getPointeeType();
1975
1976      // Never warn about void type pointers. This can be used to suppress
1977      // false positives.
1978      if (PointeeTy->isVoidType())
1979        continue;
1980
1981      // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by
1982      // actually comparing the expressions for equality. Because computing the
1983      // expression IDs can be expensive, we only do this if the diagnostic is
1984      // enabled.
1985      if (SizeOfArg &&
1986          Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess,
1987                                   SizeOfArg->getExprLoc())) {
1988        // We only compute IDs for expressions if the warning is enabled, and
1989        // cache the sizeof arg's ID.
1990        if (SizeOfArgID == llvm::FoldingSetNodeID())
1991          SizeOfArg->Profile(SizeOfArgID, Context, true);
1992        llvm::FoldingSetNodeID DestID;
1993        Dest->Profile(DestID, Context, true);
1994        if (DestID == SizeOfArgID) {
1995          unsigned ActionIdx = 0; // Default is to suggest dereferencing.
1996          if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest))
1997            if (UnaryOp->getOpcode() == UO_AddrOf)
1998              ActionIdx = 1; // If its an address-of operator, just remove it.
1999          if (Context.getTypeSize(PointeeTy) == Context.getCharWidth())
2000            ActionIdx = 2; // If the pointee's size is sizeof(char),
2001                           // suggest an explicit length.
2002          DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest,
2003                              PDiag(diag::warn_sizeof_pointer_expr_memaccess)
2004                                << FnName << ArgIdx << ActionIdx
2005                                << Dest->getSourceRange()
2006                                << SizeOfArg->getSourceRange());
2007          break;
2008        }
2009      }
2010
2011      // Also check for cases where the sizeof argument is the exact same
2012      // type as the memory argument, and where it points to a user-defined
2013      // record type.
2014      if (SizeOfArgTy != QualType()) {
2015        if (PointeeTy->isRecordType() &&
2016            Context.typesAreCompatible(SizeOfArgTy, DestTy)) {
2017          DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest,
2018                              PDiag(diag::warn_sizeof_pointer_type_memaccess)
2019                                << FnName << SizeOfArgTy << ArgIdx
2020                                << PointeeTy << Dest->getSourceRange()
2021                                << LenExpr->getSourceRange());
2022          break;
2023        }
2024      }
2025
2026      // Always complain about dynamic classes.
2027      if (isDynamicClassType(PointeeTy))
2028        DiagRuntimeBehavior(
2029          Dest->getExprLoc(), Dest,
2030          PDiag(diag::warn_dyn_class_memaccess)
2031            << (CMF == CMF_Memcmp ? ArgIdx + 2 : ArgIdx) << FnName << PointeeTy
2032            // "overwritten" if we're warning about the destination for any call
2033            // but memcmp; otherwise a verb appropriate to the call.
2034            << (ArgIdx == 0 && CMF != CMF_Memcmp ? 0 : (unsigned)CMF)
2035            << Call->getCallee()->getSourceRange());
2036      else if (PointeeTy.hasNonTrivialObjCLifetime() && CMF != CMF_Memset)
2037        DiagRuntimeBehavior(
2038          Dest->getExprLoc(), Dest,
2039          PDiag(diag::warn_arc_object_memaccess)
2040            << ArgIdx << FnName << PointeeTy
2041            << Call->getCallee()->getSourceRange());
2042      else
2043        continue;
2044
2045      DiagRuntimeBehavior(
2046        Dest->getExprLoc(), Dest,
2047        PDiag(diag::note_bad_memaccess_silence)
2048          << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
2049      break;
2050    }
2051  }
2052}
2053
2054// A little helper routine: ignore addition and subtraction of integer literals.
2055// This intentionally does not ignore all integer constant expressions because
2056// we don't want to remove sizeof().
2057static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) {
2058  Ex = Ex->IgnoreParenCasts();
2059
2060  for (;;) {
2061    const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex);
2062    if (!BO || !BO->isAdditiveOp())
2063      break;
2064
2065    const Expr *RHS = BO->getRHS()->IgnoreParenCasts();
2066    const Expr *LHS = BO->getLHS()->IgnoreParenCasts();
2067
2068    if (isa<IntegerLiteral>(RHS))
2069      Ex = LHS;
2070    else if (isa<IntegerLiteral>(LHS))
2071      Ex = RHS;
2072    else
2073      break;
2074  }
2075
2076  return Ex;
2077}
2078
2079// Warn if the user has made the 'size' argument to strlcpy or strlcat
2080// be the size of the source, instead of the destination.
2081void Sema::CheckStrlcpycatArguments(const CallExpr *Call,
2082                                    IdentifierInfo *FnName) {
2083
2084  // Don't crash if the user has the wrong number of arguments
2085  if (Call->getNumArgs() != 3)
2086    return;
2087
2088  const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context);
2089  const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context);
2090  const Expr *CompareWithSrc = NULL;
2091
2092  // Look for 'strlcpy(dst, x, sizeof(x))'
2093  if (const Expr *Ex = getSizeOfExprArg(SizeArg))
2094    CompareWithSrc = Ex;
2095  else {
2096    // Look for 'strlcpy(dst, x, strlen(x))'
2097    if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) {
2098      if (SizeCall->isBuiltinCall(Context) == Builtin::BIstrlen
2099          && SizeCall->getNumArgs() == 1)
2100        CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context);
2101    }
2102  }
2103
2104  if (!CompareWithSrc)
2105    return;
2106
2107  // Determine if the argument to sizeof/strlen is equal to the source
2108  // argument.  In principle there's all kinds of things you could do
2109  // here, for instance creating an == expression and evaluating it with
2110  // EvaluateAsBooleanCondition, but this uses a more direct technique:
2111  const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg);
2112  if (!SrcArgDRE)
2113    return;
2114
2115  const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc);
2116  if (!CompareWithSrcDRE ||
2117      SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl())
2118    return;
2119
2120  const Expr *OriginalSizeArg = Call->getArg(2);
2121  Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size)
2122    << OriginalSizeArg->getSourceRange() << FnName;
2123
2124  // Output a FIXIT hint if the destination is an array (rather than a
2125  // pointer to an array).  This could be enhanced to handle some
2126  // pointers if we know the actual size, like if DstArg is 'array+2'
2127  // we could say 'sizeof(array)-2'.
2128  const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts();
2129  QualType DstArgTy = DstArg->getType();
2130
2131  // Only handle constant-sized or VLAs, but not flexible members.
2132  if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) {
2133    // Only issue the FIXIT for arrays of size > 1.
2134    if (CAT->getSize().getSExtValue() <= 1)
2135      return;
2136  } else if (!DstArgTy->isVariableArrayType()) {
2137    return;
2138  }
2139
2140  llvm::SmallString<128> sizeString;
2141  llvm::raw_svector_ostream OS(sizeString);
2142  OS << "sizeof(";
2143  DstArg->printPretty(OS, Context, 0, Context.PrintingPolicy);
2144  OS << ")";
2145
2146  Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size)
2147    << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(),
2148                                    OS.str());
2149}
2150
2151//===--- CHECK: Return Address of Stack Variable --------------------------===//
2152
2153static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars);
2154static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars);
2155
2156/// CheckReturnStackAddr - Check if a return statement returns the address
2157///   of a stack variable.
2158void
2159Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
2160                           SourceLocation ReturnLoc) {
2161
2162  Expr *stackE = 0;
2163  SmallVector<DeclRefExpr *, 8> refVars;
2164
2165  // Perform checking for returned stack addresses, local blocks,
2166  // label addresses or references to temporaries.
2167  if (lhsType->isPointerType() ||
2168      (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) {
2169    stackE = EvalAddr(RetValExp, refVars);
2170  } else if (lhsType->isReferenceType()) {
2171    stackE = EvalVal(RetValExp, refVars);
2172  }
2173
2174  if (stackE == 0)
2175    return; // Nothing suspicious was found.
2176
2177  SourceLocation diagLoc;
2178  SourceRange diagRange;
2179  if (refVars.empty()) {
2180    diagLoc = stackE->getLocStart();
2181    diagRange = stackE->getSourceRange();
2182  } else {
2183    // We followed through a reference variable. 'stackE' contains the
2184    // problematic expression but we will warn at the return statement pointing
2185    // at the reference variable. We will later display the "trail" of
2186    // reference variables using notes.
2187    diagLoc = refVars[0]->getLocStart();
2188    diagRange = refVars[0]->getSourceRange();
2189  }
2190
2191  if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
2192    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
2193                                             : diag::warn_ret_stack_addr)
2194     << DR->getDecl()->getDeclName() << diagRange;
2195  } else if (isa<BlockExpr>(stackE)) { // local block.
2196    Diag(diagLoc, diag::err_ret_local_block) << diagRange;
2197  } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
2198    Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
2199  } else { // local temporary.
2200    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
2201                                             : diag::warn_ret_local_temp_addr)
2202     << diagRange;
2203  }
2204
2205  // Display the "trail" of reference variables that we followed until we
2206  // found the problematic expression using notes.
2207  for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
2208    VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
2209    // If this var binds to another reference var, show the range of the next
2210    // var, otherwise the var binds to the problematic expression, in which case
2211    // show the range of the expression.
2212    SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
2213                                  : stackE->getSourceRange();
2214    Diag(VD->getLocation(), diag::note_ref_var_local_bind)
2215      << VD->getDeclName() << range;
2216  }
2217}
2218
2219/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
2220///  check if the expression in a return statement evaluates to an address
2221///  to a location on the stack, a local block, an address of a label, or a
2222///  reference to local temporary. The recursion is used to traverse the
2223///  AST of the return expression, with recursion backtracking when we
2224///  encounter a subexpression that (1) clearly does not lead to one of the
2225///  above problematic expressions (2) is something we cannot determine leads to
2226///  a problematic expression based on such local checking.
2227///
2228///  Both EvalAddr and EvalVal follow through reference variables to evaluate
2229///  the expression that they point to. Such variables are added to the
2230///  'refVars' vector so that we know what the reference variable "trail" was.
2231///
2232///  EvalAddr processes expressions that are pointers that are used as
2233///  references (and not L-values).  EvalVal handles all other values.
2234///  At the base case of the recursion is a check for the above problematic
2235///  expressions.
2236///
2237///  This implementation handles:
2238///
2239///   * pointer-to-pointer casts
2240///   * implicit conversions from array references to pointers
2241///   * taking the address of fields
2242///   * arbitrary interplay between "&" and "*" operators
2243///   * pointer arithmetic from an address of a stack variable
2244///   * taking the address of an array element where the array is on the stack
2245static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) {
2246  if (E->isTypeDependent())
2247      return NULL;
2248
2249  // We should only be called for evaluating pointer expressions.
2250  assert((E->getType()->isAnyPointerType() ||
2251          E->getType()->isBlockPointerType() ||
2252          E->getType()->isObjCQualifiedIdType()) &&
2253         "EvalAddr only works on pointers");
2254
2255  E = E->IgnoreParens();
2256
2257  // Our "symbolic interpreter" is just a dispatch off the currently
2258  // viewed AST node.  We then recursively traverse the AST by calling
2259  // EvalAddr and EvalVal appropriately.
2260  switch (E->getStmtClass()) {
2261  case Stmt::DeclRefExprClass: {
2262    DeclRefExpr *DR = cast<DeclRefExpr>(E);
2263
2264    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2265      // If this is a reference variable, follow through to the expression that
2266      // it points to.
2267      if (V->hasLocalStorage() &&
2268          V->getType()->isReferenceType() && V->hasInit()) {
2269        // Add the reference variable to the "trail".
2270        refVars.push_back(DR);
2271        return EvalAddr(V->getInit(), refVars);
2272      }
2273
2274    return NULL;
2275  }
2276
2277  case Stmt::UnaryOperatorClass: {
2278    // The only unary operator that make sense to handle here
2279    // is AddrOf.  All others don't make sense as pointers.
2280    UnaryOperator *U = cast<UnaryOperator>(E);
2281
2282    if (U->getOpcode() == UO_AddrOf)
2283      return EvalVal(U->getSubExpr(), refVars);
2284    else
2285      return NULL;
2286  }
2287
2288  case Stmt::BinaryOperatorClass: {
2289    // Handle pointer arithmetic.  All other binary operators are not valid
2290    // in this context.
2291    BinaryOperator *B = cast<BinaryOperator>(E);
2292    BinaryOperatorKind op = B->getOpcode();
2293
2294    if (op != BO_Add && op != BO_Sub)
2295      return NULL;
2296
2297    Expr *Base = B->getLHS();
2298
2299    // Determine which argument is the real pointer base.  It could be
2300    // the RHS argument instead of the LHS.
2301    if (!Base->getType()->isPointerType()) Base = B->getRHS();
2302
2303    assert (Base->getType()->isPointerType());
2304    return EvalAddr(Base, refVars);
2305  }
2306
2307  // For conditional operators we need to see if either the LHS or RHS are
2308  // valid DeclRefExpr*s.  If one of them is valid, we return it.
2309  case Stmt::ConditionalOperatorClass: {
2310    ConditionalOperator *C = cast<ConditionalOperator>(E);
2311
2312    // Handle the GNU extension for missing LHS.
2313    if (Expr *lhsExpr = C->getLHS()) {
2314    // In C++, we can have a throw-expression, which has 'void' type.
2315      if (!lhsExpr->getType()->isVoidType())
2316        if (Expr* LHS = EvalAddr(lhsExpr, refVars))
2317          return LHS;
2318    }
2319
2320    // In C++, we can have a throw-expression, which has 'void' type.
2321    if (C->getRHS()->getType()->isVoidType())
2322      return NULL;
2323
2324    return EvalAddr(C->getRHS(), refVars);
2325  }
2326
2327  case Stmt::BlockExprClass:
2328    if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
2329      return E; // local block.
2330    return NULL;
2331
2332  case Stmt::AddrLabelExprClass:
2333    return E; // address of label.
2334
2335  // For casts, we need to handle conversions from arrays to
2336  // pointer values, and pointer-to-pointer conversions.
2337  case Stmt::ImplicitCastExprClass:
2338  case Stmt::CStyleCastExprClass:
2339  case Stmt::CXXFunctionalCastExprClass:
2340  case Stmt::ObjCBridgedCastExprClass: {
2341    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
2342    QualType T = SubExpr->getType();
2343
2344    if (SubExpr->getType()->isPointerType() ||
2345        SubExpr->getType()->isBlockPointerType() ||
2346        SubExpr->getType()->isObjCQualifiedIdType())
2347      return EvalAddr(SubExpr, refVars);
2348    else if (T->isArrayType())
2349      return EvalVal(SubExpr, refVars);
2350    else
2351      return 0;
2352  }
2353
2354  // C++ casts.  For dynamic casts, static casts, and const casts, we
2355  // are always converting from a pointer-to-pointer, so we just blow
2356  // through the cast.  In the case the dynamic cast doesn't fail (and
2357  // return NULL), we take the conservative route and report cases
2358  // where we return the address of a stack variable.  For Reinterpre
2359  // FIXME: The comment about is wrong; we're not always converting
2360  // from pointer to pointer. I'm guessing that this code should also
2361  // handle references to objects.
2362  case Stmt::CXXStaticCastExprClass:
2363  case Stmt::CXXDynamicCastExprClass:
2364  case Stmt::CXXConstCastExprClass:
2365  case Stmt::CXXReinterpretCastExprClass: {
2366      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
2367      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
2368        return EvalAddr(S, refVars);
2369      else
2370        return NULL;
2371  }
2372
2373  case Stmt::MaterializeTemporaryExprClass:
2374    if (Expr *Result = EvalAddr(
2375                         cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
2376                                refVars))
2377      return Result;
2378
2379    return E;
2380
2381  // Everything else: we simply don't reason about them.
2382  default:
2383    return NULL;
2384  }
2385}
2386
2387
2388///  EvalVal - This function is complements EvalAddr in the mutual recursion.
2389///   See the comments for EvalAddr for more details.
2390static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) {
2391do {
2392  // We should only be called for evaluating non-pointer expressions, or
2393  // expressions with a pointer type that are not used as references but instead
2394  // are l-values (e.g., DeclRefExpr with a pointer type).
2395
2396  // Our "symbolic interpreter" is just a dispatch off the currently
2397  // viewed AST node.  We then recursively traverse the AST by calling
2398  // EvalAddr and EvalVal appropriately.
2399
2400  E = E->IgnoreParens();
2401  switch (E->getStmtClass()) {
2402  case Stmt::ImplicitCastExprClass: {
2403    ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
2404    if (IE->getValueKind() == VK_LValue) {
2405      E = IE->getSubExpr();
2406      continue;
2407    }
2408    return NULL;
2409  }
2410
2411  case Stmt::DeclRefExprClass: {
2412    // When we hit a DeclRefExpr we are looking at code that refers to a
2413    // variable's name. If it's not a reference variable we check if it has
2414    // local storage within the function, and if so, return the expression.
2415    DeclRefExpr *DR = cast<DeclRefExpr>(E);
2416
2417    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2418      if (V->hasLocalStorage()) {
2419        if (!V->getType()->isReferenceType())
2420          return DR;
2421
2422        // Reference variable, follow through to the expression that
2423        // it points to.
2424        if (V->hasInit()) {
2425          // Add the reference variable to the "trail".
2426          refVars.push_back(DR);
2427          return EvalVal(V->getInit(), refVars);
2428        }
2429      }
2430
2431    return NULL;
2432  }
2433
2434  case Stmt::UnaryOperatorClass: {
2435    // The only unary operator that make sense to handle here
2436    // is Deref.  All others don't resolve to a "name."  This includes
2437    // handling all sorts of rvalues passed to a unary operator.
2438    UnaryOperator *U = cast<UnaryOperator>(E);
2439
2440    if (U->getOpcode() == UO_Deref)
2441      return EvalAddr(U->getSubExpr(), refVars);
2442
2443    return NULL;
2444  }
2445
2446  case Stmt::ArraySubscriptExprClass: {
2447    // Array subscripts are potential references to data on the stack.  We
2448    // retrieve the DeclRefExpr* for the array variable if it indeed
2449    // has local storage.
2450    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars);
2451  }
2452
2453  case Stmt::ConditionalOperatorClass: {
2454    // For conditional operators we need to see if either the LHS or RHS are
2455    // non-NULL Expr's.  If one is non-NULL, we return it.
2456    ConditionalOperator *C = cast<ConditionalOperator>(E);
2457
2458    // Handle the GNU extension for missing LHS.
2459    if (Expr *lhsExpr = C->getLHS())
2460      if (Expr *LHS = EvalVal(lhsExpr, refVars))
2461        return LHS;
2462
2463    return EvalVal(C->getRHS(), refVars);
2464  }
2465
2466  // Accesses to members are potential references to data on the stack.
2467  case Stmt::MemberExprClass: {
2468    MemberExpr *M = cast<MemberExpr>(E);
2469
2470    // Check for indirect access.  We only want direct field accesses.
2471    if (M->isArrow())
2472      return NULL;
2473
2474    // Check whether the member type is itself a reference, in which case
2475    // we're not going to refer to the member, but to what the member refers to.
2476    if (M->getMemberDecl()->getType()->isReferenceType())
2477      return NULL;
2478
2479    return EvalVal(M->getBase(), refVars);
2480  }
2481
2482  case Stmt::MaterializeTemporaryExprClass:
2483    if (Expr *Result = EvalVal(
2484                          cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
2485                               refVars))
2486      return Result;
2487
2488    return E;
2489
2490  default:
2491    // Check that we don't return or take the address of a reference to a
2492    // temporary. This is only useful in C++.
2493    if (!E->isTypeDependent() && E->isRValue())
2494      return E;
2495
2496    // Everything else: we simply don't reason about them.
2497    return NULL;
2498  }
2499} while (true);
2500}
2501
2502//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
2503
2504/// Check for comparisons of floating point operands using != and ==.
2505/// Issue a warning if these are no self-comparisons, as they are not likely
2506/// to do what the programmer intended.
2507void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) {
2508  bool EmitWarning = true;
2509
2510  Expr* LeftExprSansParen = lex->IgnoreParenImpCasts();
2511  Expr* RightExprSansParen = rex->IgnoreParenImpCasts();
2512
2513  // Special case: check for x == x (which is OK).
2514  // Do not emit warnings for such cases.
2515  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
2516    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
2517      if (DRL->getDecl() == DRR->getDecl())
2518        EmitWarning = false;
2519
2520
2521  // Special case: check for comparisons against literals that can be exactly
2522  //  represented by APFloat.  In such cases, do not emit a warning.  This
2523  //  is a heuristic: often comparison against such literals are used to
2524  //  detect if a value in a variable has not changed.  This clearly can
2525  //  lead to false negatives.
2526  if (EmitWarning) {
2527    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
2528      if (FLL->isExact())
2529        EmitWarning = false;
2530    } else
2531      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
2532        if (FLR->isExact())
2533          EmitWarning = false;
2534    }
2535  }
2536
2537  // Check for comparisons with builtin types.
2538  if (EmitWarning)
2539    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
2540      if (CL->isBuiltinCall(Context))
2541        EmitWarning = false;
2542
2543  if (EmitWarning)
2544    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
2545      if (CR->isBuiltinCall(Context))
2546        EmitWarning = false;
2547
2548  // Emit the diagnostic.
2549  if (EmitWarning)
2550    Diag(loc, diag::warn_floatingpoint_eq)
2551      << lex->getSourceRange() << rex->getSourceRange();
2552}
2553
2554//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
2555//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
2556
2557namespace {
2558
2559/// Structure recording the 'active' range of an integer-valued
2560/// expression.
2561struct IntRange {
2562  /// The number of bits active in the int.
2563  unsigned Width;
2564
2565  /// True if the int is known not to have negative values.
2566  bool NonNegative;
2567
2568  IntRange(unsigned Width, bool NonNegative)
2569    : Width(Width), NonNegative(NonNegative)
2570  {}
2571
2572  /// Returns the range of the bool type.
2573  static IntRange forBoolType() {
2574    return IntRange(1, true);
2575  }
2576
2577  /// Returns the range of an opaque value of the given integral type.
2578  static IntRange forValueOfType(ASTContext &C, QualType T) {
2579    return forValueOfCanonicalType(C,
2580                          T->getCanonicalTypeInternal().getTypePtr());
2581  }
2582
2583  /// Returns the range of an opaque value of a canonical integral type.
2584  static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
2585    assert(T->isCanonicalUnqualified());
2586
2587    if (const VectorType *VT = dyn_cast<VectorType>(T))
2588      T = VT->getElementType().getTypePtr();
2589    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2590      T = CT->getElementType().getTypePtr();
2591
2592    // For enum types, use the known bit width of the enumerators.
2593    if (const EnumType *ET = dyn_cast<EnumType>(T)) {
2594      EnumDecl *Enum = ET->getDecl();
2595      if (!Enum->isDefinition())
2596        return IntRange(C.getIntWidth(QualType(T, 0)), false);
2597
2598      unsigned NumPositive = Enum->getNumPositiveBits();
2599      unsigned NumNegative = Enum->getNumNegativeBits();
2600
2601      return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
2602    }
2603
2604    const BuiltinType *BT = cast<BuiltinType>(T);
2605    assert(BT->isInteger());
2606
2607    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2608  }
2609
2610  /// Returns the "target" range of a canonical integral type, i.e.
2611  /// the range of values expressible in the type.
2612  ///
2613  /// This matches forValueOfCanonicalType except that enums have the
2614  /// full range of their type, not the range of their enumerators.
2615  static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
2616    assert(T->isCanonicalUnqualified());
2617
2618    if (const VectorType *VT = dyn_cast<VectorType>(T))
2619      T = VT->getElementType().getTypePtr();
2620    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2621      T = CT->getElementType().getTypePtr();
2622    if (const EnumType *ET = dyn_cast<EnumType>(T))
2623      T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr();
2624
2625    const BuiltinType *BT = cast<BuiltinType>(T);
2626    assert(BT->isInteger());
2627
2628    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2629  }
2630
2631  /// Returns the supremum of two ranges: i.e. their conservative merge.
2632  static IntRange join(IntRange L, IntRange R) {
2633    return IntRange(std::max(L.Width, R.Width),
2634                    L.NonNegative && R.NonNegative);
2635  }
2636
2637  /// Returns the infinum of two ranges: i.e. their aggressive merge.
2638  static IntRange meet(IntRange L, IntRange R) {
2639    return IntRange(std::min(L.Width, R.Width),
2640                    L.NonNegative || R.NonNegative);
2641  }
2642};
2643
2644IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) {
2645  if (value.isSigned() && value.isNegative())
2646    return IntRange(value.getMinSignedBits(), false);
2647
2648  if (value.getBitWidth() > MaxWidth)
2649    value = value.trunc(MaxWidth);
2650
2651  // isNonNegative() just checks the sign bit without considering
2652  // signedness.
2653  return IntRange(value.getActiveBits(), true);
2654}
2655
2656IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
2657                       unsigned MaxWidth) {
2658  if (result.isInt())
2659    return GetValueRange(C, result.getInt(), MaxWidth);
2660
2661  if (result.isVector()) {
2662    IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
2663    for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
2664      IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
2665      R = IntRange::join(R, El);
2666    }
2667    return R;
2668  }
2669
2670  if (result.isComplexInt()) {
2671    IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
2672    IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
2673    return IntRange::join(R, I);
2674  }
2675
2676  // This can happen with lossless casts to intptr_t of "based" lvalues.
2677  // Assume it might use arbitrary bits.
2678  // FIXME: The only reason we need to pass the type in here is to get
2679  // the sign right on this one case.  It would be nice if APValue
2680  // preserved this.
2681  assert(result.isLValue());
2682  return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType());
2683}
2684
2685/// Pseudo-evaluate the given integer expression, estimating the
2686/// range of values it might take.
2687///
2688/// \param MaxWidth - the width to which the value will be truncated
2689IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
2690  E = E->IgnoreParens();
2691
2692  // Try a full evaluation first.
2693  Expr::EvalResult result;
2694  if (E->Evaluate(result, C))
2695    return GetValueRange(C, result.Val, E->getType(), MaxWidth);
2696
2697  // I think we only want to look through implicit casts here; if the
2698  // user has an explicit widening cast, we should treat the value as
2699  // being of the new, wider type.
2700  if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
2701    if (CE->getCastKind() == CK_NoOp)
2702      return GetExprRange(C, CE->getSubExpr(), MaxWidth);
2703
2704    IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
2705
2706    bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
2707
2708    // Assume that non-integer casts can span the full range of the type.
2709    if (!isIntegerCast)
2710      return OutputTypeRange;
2711
2712    IntRange SubRange
2713      = GetExprRange(C, CE->getSubExpr(),
2714                     std::min(MaxWidth, OutputTypeRange.Width));
2715
2716    // Bail out if the subexpr's range is as wide as the cast type.
2717    if (SubRange.Width >= OutputTypeRange.Width)
2718      return OutputTypeRange;
2719
2720    // Otherwise, we take the smaller width, and we're non-negative if
2721    // either the output type or the subexpr is.
2722    return IntRange(SubRange.Width,
2723                    SubRange.NonNegative || OutputTypeRange.NonNegative);
2724  }
2725
2726  if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
2727    // If we can fold the condition, just take that operand.
2728    bool CondResult;
2729    if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
2730      return GetExprRange(C, CondResult ? CO->getTrueExpr()
2731                                        : CO->getFalseExpr(),
2732                          MaxWidth);
2733
2734    // Otherwise, conservatively merge.
2735    IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
2736    IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
2737    return IntRange::join(L, R);
2738  }
2739
2740  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
2741    switch (BO->getOpcode()) {
2742
2743    // Boolean-valued operations are single-bit and positive.
2744    case BO_LAnd:
2745    case BO_LOr:
2746    case BO_LT:
2747    case BO_GT:
2748    case BO_LE:
2749    case BO_GE:
2750    case BO_EQ:
2751    case BO_NE:
2752      return IntRange::forBoolType();
2753
2754    // The type of the assignments is the type of the LHS, so the RHS
2755    // is not necessarily the same type.
2756    case BO_MulAssign:
2757    case BO_DivAssign:
2758    case BO_RemAssign:
2759    case BO_AddAssign:
2760    case BO_SubAssign:
2761    case BO_XorAssign:
2762    case BO_OrAssign:
2763      // TODO: bitfields?
2764      return IntRange::forValueOfType(C, E->getType());
2765
2766    // Simple assignments just pass through the RHS, which will have
2767    // been coerced to the LHS type.
2768    case BO_Assign:
2769      // TODO: bitfields?
2770      return GetExprRange(C, BO->getRHS(), MaxWidth);
2771
2772    // Operations with opaque sources are black-listed.
2773    case BO_PtrMemD:
2774    case BO_PtrMemI:
2775      return IntRange::forValueOfType(C, E->getType());
2776
2777    // Bitwise-and uses the *infinum* of the two source ranges.
2778    case BO_And:
2779    case BO_AndAssign:
2780      return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
2781                            GetExprRange(C, BO->getRHS(), MaxWidth));
2782
2783    // Left shift gets black-listed based on a judgement call.
2784    case BO_Shl:
2785      // ...except that we want to treat '1 << (blah)' as logically
2786      // positive.  It's an important idiom.
2787      if (IntegerLiteral *I
2788            = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
2789        if (I->getValue() == 1) {
2790          IntRange R = IntRange::forValueOfType(C, E->getType());
2791          return IntRange(R.Width, /*NonNegative*/ true);
2792        }
2793      }
2794      // fallthrough
2795
2796    case BO_ShlAssign:
2797      return IntRange::forValueOfType(C, E->getType());
2798
2799    // Right shift by a constant can narrow its left argument.
2800    case BO_Shr:
2801    case BO_ShrAssign: {
2802      IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2803
2804      // If the shift amount is a positive constant, drop the width by
2805      // that much.
2806      llvm::APSInt shift;
2807      if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
2808          shift.isNonNegative()) {
2809        unsigned zext = shift.getZExtValue();
2810        if (zext >= L.Width)
2811          L.Width = (L.NonNegative ? 0 : 1);
2812        else
2813          L.Width -= zext;
2814      }
2815
2816      return L;
2817    }
2818
2819    // Comma acts as its right operand.
2820    case BO_Comma:
2821      return GetExprRange(C, BO->getRHS(), MaxWidth);
2822
2823    // Black-list pointer subtractions.
2824    case BO_Sub:
2825      if (BO->getLHS()->getType()->isPointerType())
2826        return IntRange::forValueOfType(C, E->getType());
2827      break;
2828
2829    // The width of a division result is mostly determined by the size
2830    // of the LHS.
2831    case BO_Div: {
2832      // Don't 'pre-truncate' the operands.
2833      unsigned opWidth = C.getIntWidth(E->getType());
2834      IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
2835
2836      // If the divisor is constant, use that.
2837      llvm::APSInt divisor;
2838      if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) {
2839        unsigned log2 = divisor.logBase2(); // floor(log_2(divisor))
2840        if (log2 >= L.Width)
2841          L.Width = (L.NonNegative ? 0 : 1);
2842        else
2843          L.Width = std::min(L.Width - log2, MaxWidth);
2844        return L;
2845      }
2846
2847      // Otherwise, just use the LHS's width.
2848      IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
2849      return IntRange(L.Width, L.NonNegative && R.NonNegative);
2850    }
2851
2852    // The result of a remainder can't be larger than the result of
2853    // either side.
2854    case BO_Rem: {
2855      // Don't 'pre-truncate' the operands.
2856      unsigned opWidth = C.getIntWidth(E->getType());
2857      IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
2858      IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
2859
2860      IntRange meet = IntRange::meet(L, R);
2861      meet.Width = std::min(meet.Width, MaxWidth);
2862      return meet;
2863    }
2864
2865    // The default behavior is okay for these.
2866    case BO_Mul:
2867    case BO_Add:
2868    case BO_Xor:
2869    case BO_Or:
2870      break;
2871    }
2872
2873    // The default case is to treat the operation as if it were closed
2874    // on the narrowest type that encompasses both operands.
2875    IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2876    IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
2877    return IntRange::join(L, R);
2878  }
2879
2880  if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
2881    switch (UO->getOpcode()) {
2882    // Boolean-valued operations are white-listed.
2883    case UO_LNot:
2884      return IntRange::forBoolType();
2885
2886    // Operations with opaque sources are black-listed.
2887    case UO_Deref:
2888    case UO_AddrOf: // should be impossible
2889      return IntRange::forValueOfType(C, E->getType());
2890
2891    default:
2892      return GetExprRange(C, UO->getSubExpr(), MaxWidth);
2893    }
2894  }
2895
2896  if (dyn_cast<OffsetOfExpr>(E)) {
2897    IntRange::forValueOfType(C, E->getType());
2898  }
2899
2900  FieldDecl *BitField = E->getBitField();
2901  if (BitField) {
2902    llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C);
2903    unsigned BitWidth = BitWidthAP.getZExtValue();
2904
2905    return IntRange(BitWidth,
2906                    BitField->getType()->isUnsignedIntegerOrEnumerationType());
2907  }
2908
2909  return IntRange::forValueOfType(C, E->getType());
2910}
2911
2912IntRange GetExprRange(ASTContext &C, Expr *E) {
2913  return GetExprRange(C, E, C.getIntWidth(E->getType()));
2914}
2915
2916/// Checks whether the given value, which currently has the given
2917/// source semantics, has the same value when coerced through the
2918/// target semantics.
2919bool IsSameFloatAfterCast(const llvm::APFloat &value,
2920                          const llvm::fltSemantics &Src,
2921                          const llvm::fltSemantics &Tgt) {
2922  llvm::APFloat truncated = value;
2923
2924  bool ignored;
2925  truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
2926  truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
2927
2928  return truncated.bitwiseIsEqual(value);
2929}
2930
2931/// Checks whether the given value, which currently has the given
2932/// source semantics, has the same value when coerced through the
2933/// target semantics.
2934///
2935/// The value might be a vector of floats (or a complex number).
2936bool IsSameFloatAfterCast(const APValue &value,
2937                          const llvm::fltSemantics &Src,
2938                          const llvm::fltSemantics &Tgt) {
2939  if (value.isFloat())
2940    return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
2941
2942  if (value.isVector()) {
2943    for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
2944      if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
2945        return false;
2946    return true;
2947  }
2948
2949  assert(value.isComplexFloat());
2950  return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
2951          IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
2952}
2953
2954void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
2955
2956static bool IsZero(Sema &S, Expr *E) {
2957  // Suppress cases where we are comparing against an enum constant.
2958  if (const DeclRefExpr *DR =
2959      dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
2960    if (isa<EnumConstantDecl>(DR->getDecl()))
2961      return false;
2962
2963  // Suppress cases where the '0' value is expanded from a macro.
2964  if (E->getLocStart().isMacroID())
2965    return false;
2966
2967  llvm::APSInt Value;
2968  return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
2969}
2970
2971static bool HasEnumType(Expr *E) {
2972  // Strip off implicit integral promotions.
2973  while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
2974    if (ICE->getCastKind() != CK_IntegralCast &&
2975        ICE->getCastKind() != CK_NoOp)
2976      break;
2977    E = ICE->getSubExpr();
2978  }
2979
2980  return E->getType()->isEnumeralType();
2981}
2982
2983void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
2984  BinaryOperatorKind op = E->getOpcode();
2985  if (E->isValueDependent())
2986    return;
2987
2988  if (op == BO_LT && IsZero(S, E->getRHS())) {
2989    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2990      << "< 0" << "false" << HasEnumType(E->getLHS())
2991      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2992  } else if (op == BO_GE && IsZero(S, E->getRHS())) {
2993    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2994      << ">= 0" << "true" << HasEnumType(E->getLHS())
2995      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2996  } else if (op == BO_GT && IsZero(S, E->getLHS())) {
2997    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2998      << "0 >" << "false" << HasEnumType(E->getRHS())
2999      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3000  } else if (op == BO_LE && IsZero(S, E->getLHS())) {
3001    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
3002      << "0 <=" << "true" << HasEnumType(E->getRHS())
3003      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3004  }
3005}
3006
3007/// Analyze the operands of the given comparison.  Implements the
3008/// fallback case from AnalyzeComparison.
3009void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
3010  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
3011  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
3012}
3013
3014/// \brief Implements -Wsign-compare.
3015///
3016/// \param lex the left-hand expression
3017/// \param rex the right-hand expression
3018/// \param OpLoc the location of the joining operator
3019/// \param BinOpc binary opcode or 0
3020void AnalyzeComparison(Sema &S, BinaryOperator *E) {
3021  // The type the comparison is being performed in.
3022  QualType T = E->getLHS()->getType();
3023  assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
3024         && "comparison with mismatched types");
3025
3026  // We don't do anything special if this isn't an unsigned integral
3027  // comparison:  we're only interested in integral comparisons, and
3028  // signed comparisons only happen in cases we don't care to warn about.
3029  //
3030  // We also don't care about value-dependent expressions or expressions
3031  // whose result is a constant.
3032  if (!T->hasUnsignedIntegerRepresentation()
3033      || E->isValueDependent() || E->isIntegerConstantExpr(S.Context))
3034    return AnalyzeImpConvsInComparison(S, E);
3035
3036  Expr *lex = E->getLHS()->IgnoreParenImpCasts();
3037  Expr *rex = E->getRHS()->IgnoreParenImpCasts();
3038
3039  // Check to see if one of the (unmodified) operands is of different
3040  // signedness.
3041  Expr *signedOperand, *unsignedOperand;
3042  if (lex->getType()->hasSignedIntegerRepresentation()) {
3043    assert(!rex->getType()->hasSignedIntegerRepresentation() &&
3044           "unsigned comparison between two signed integer expressions?");
3045    signedOperand = lex;
3046    unsignedOperand = rex;
3047  } else if (rex->getType()->hasSignedIntegerRepresentation()) {
3048    signedOperand = rex;
3049    unsignedOperand = lex;
3050  } else {
3051    CheckTrivialUnsignedComparison(S, E);
3052    return AnalyzeImpConvsInComparison(S, E);
3053  }
3054
3055  // Otherwise, calculate the effective range of the signed operand.
3056  IntRange signedRange = GetExprRange(S.Context, signedOperand);
3057
3058  // Go ahead and analyze implicit conversions in the operands.  Note
3059  // that we skip the implicit conversions on both sides.
3060  AnalyzeImplicitConversions(S, lex, E->getOperatorLoc());
3061  AnalyzeImplicitConversions(S, rex, E->getOperatorLoc());
3062
3063  // If the signed range is non-negative, -Wsign-compare won't fire,
3064  // but we should still check for comparisons which are always true
3065  // or false.
3066  if (signedRange.NonNegative)
3067    return CheckTrivialUnsignedComparison(S, E);
3068
3069  // For (in)equality comparisons, if the unsigned operand is a
3070  // constant which cannot collide with a overflowed signed operand,
3071  // then reinterpreting the signed operand as unsigned will not
3072  // change the result of the comparison.
3073  if (E->isEqualityOp()) {
3074    unsigned comparisonWidth = S.Context.getIntWidth(T);
3075    IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
3076
3077    // We should never be unable to prove that the unsigned operand is
3078    // non-negative.
3079    assert(unsignedRange.NonNegative && "unsigned range includes negative?");
3080
3081    if (unsignedRange.Width < comparisonWidth)
3082      return;
3083  }
3084
3085  S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison)
3086    << lex->getType() << rex->getType()
3087    << lex->getSourceRange() << rex->getSourceRange();
3088}
3089
3090/// Analyzes an attempt to assign the given value to a bitfield.
3091///
3092/// Returns true if there was something fishy about the attempt.
3093bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
3094                               SourceLocation InitLoc) {
3095  assert(Bitfield->isBitField());
3096  if (Bitfield->isInvalidDecl())
3097    return false;
3098
3099  // White-list bool bitfields.
3100  if (Bitfield->getType()->isBooleanType())
3101    return false;
3102
3103  // Ignore value- or type-dependent expressions.
3104  if (Bitfield->getBitWidth()->isValueDependent() ||
3105      Bitfield->getBitWidth()->isTypeDependent() ||
3106      Init->isValueDependent() ||
3107      Init->isTypeDependent())
3108    return false;
3109
3110  Expr *OriginalInit = Init->IgnoreParenImpCasts();
3111
3112  llvm::APSInt Width(32);
3113  Expr::EvalResult InitValue;
3114  if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) ||
3115      !OriginalInit->Evaluate(InitValue, S.Context) ||
3116      !InitValue.Val.isInt())
3117    return false;
3118
3119  const llvm::APSInt &Value = InitValue.Val.getInt();
3120  unsigned OriginalWidth = Value.getBitWidth();
3121  unsigned FieldWidth = Width.getZExtValue();
3122
3123  if (OriginalWidth <= FieldWidth)
3124    return false;
3125
3126  llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
3127
3128  // It's fairly common to write values into signed bitfields
3129  // that, if sign-extended, would end up becoming a different
3130  // value.  We don't want to warn about that.
3131  if (Value.isSigned() && Value.isNegative())
3132    TruncatedValue = TruncatedValue.sext(OriginalWidth);
3133  else
3134    TruncatedValue = TruncatedValue.zext(OriginalWidth);
3135
3136  if (Value == TruncatedValue)
3137    return false;
3138
3139  std::string PrettyValue = Value.toString(10);
3140  std::string PrettyTrunc = TruncatedValue.toString(10);
3141
3142  S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
3143    << PrettyValue << PrettyTrunc << OriginalInit->getType()
3144    << Init->getSourceRange();
3145
3146  return true;
3147}
3148
3149/// Analyze the given simple or compound assignment for warning-worthy
3150/// operations.
3151void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
3152  // Just recurse on the LHS.
3153  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
3154
3155  // We want to recurse on the RHS as normal unless we're assigning to
3156  // a bitfield.
3157  if (FieldDecl *Bitfield = E->getLHS()->getBitField()) {
3158    if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
3159                                  E->getOperatorLoc())) {
3160      // Recurse, ignoring any implicit conversions on the RHS.
3161      return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
3162                                        E->getOperatorLoc());
3163    }
3164  }
3165
3166  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
3167}
3168
3169/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
3170void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
3171                     SourceLocation CContext, unsigned diag) {
3172  S.Diag(E->getExprLoc(), diag)
3173    << SourceType << T << E->getSourceRange() << SourceRange(CContext);
3174}
3175
3176/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
3177void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext,
3178                     unsigned diag) {
3179  DiagnoseImpCast(S, E, E->getType(), T, CContext, diag);
3180}
3181
3182/// Diagnose an implicit cast from a literal expression. Also attemps to supply
3183/// fixit hints when the cast wouldn't lose information to simply write the
3184/// expression with the expected type.
3185void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
3186                                    SourceLocation CContext) {
3187  // Emit the primary warning first, then try to emit a fixit hint note if
3188  // reasonable.
3189  S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
3190    << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext);
3191
3192  const llvm::APFloat &Value = FL->getValue();
3193
3194  // Don't attempt to fix PPC double double literals.
3195  if (&Value.getSemantics() == &llvm::APFloat::PPCDoubleDouble)
3196    return;
3197
3198  // Try to convert this exactly to an integer.
3199  bool isExact = false;
3200  llvm::APSInt IntegerValue(S.Context.getIntWidth(T),
3201                            T->hasUnsignedIntegerRepresentation());
3202  if (Value.convertToInteger(IntegerValue,
3203                             llvm::APFloat::rmTowardZero, &isExact)
3204      != llvm::APFloat::opOK || !isExact)
3205    return;
3206
3207  std::string LiteralValue = IntegerValue.toString(10);
3208  S.Diag(FL->getExprLoc(), diag::note_fix_integral_float_as_integer)
3209    << FixItHint::CreateReplacement(FL->getSourceRange(), LiteralValue);
3210}
3211
3212std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
3213  if (!Range.Width) return "0";
3214
3215  llvm::APSInt ValueInRange = Value;
3216  ValueInRange.setIsSigned(!Range.NonNegative);
3217  ValueInRange = ValueInRange.trunc(Range.Width);
3218  return ValueInRange.toString(10);
3219}
3220
3221static bool isFromSystemMacro(Sema &S, SourceLocation loc) {
3222  SourceManager &smgr = S.Context.getSourceManager();
3223  return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc));
3224}
3225
3226void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
3227                             SourceLocation CC, bool *ICContext = 0) {
3228  if (E->isTypeDependent() || E->isValueDependent()) return;
3229
3230  const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
3231  const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
3232  if (Source == Target) return;
3233  if (Target->isDependentType()) return;
3234
3235  // If the conversion context location is invalid don't complain. We also
3236  // don't want to emit a warning if the issue occurs from the expansion of
3237  // a system macro. The problem is that 'getSpellingLoc()' is slow, so we
3238  // delay this check as long as possible. Once we detect we are in that
3239  // scenario, we just return.
3240  if (CC.isInvalid())
3241    return;
3242
3243  // Never diagnose implicit casts to bool.
3244  if (Target->isSpecificBuiltinType(BuiltinType::Bool))
3245    return;
3246
3247  // Strip vector types.
3248  if (isa<VectorType>(Source)) {
3249    if (!isa<VectorType>(Target)) {
3250      if (isFromSystemMacro(S, CC))
3251        return;
3252      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
3253    }
3254
3255    // If the vector cast is cast between two vectors of the same size, it is
3256    // a bitcast, not a conversion.
3257    if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target))
3258      return;
3259
3260    Source = cast<VectorType>(Source)->getElementType().getTypePtr();
3261    Target = cast<VectorType>(Target)->getElementType().getTypePtr();
3262  }
3263
3264  // Strip complex types.
3265  if (isa<ComplexType>(Source)) {
3266    if (!isa<ComplexType>(Target)) {
3267      if (isFromSystemMacro(S, CC))
3268        return;
3269
3270      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
3271    }
3272
3273    Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
3274    Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
3275  }
3276
3277  const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
3278  const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
3279
3280  // If the source is floating point...
3281  if (SourceBT && SourceBT->isFloatingPoint()) {
3282    // ...and the target is floating point...
3283    if (TargetBT && TargetBT->isFloatingPoint()) {
3284      // ...then warn if we're dropping FP rank.
3285
3286      // Builtin FP kinds are ordered by increasing FP rank.
3287      if (SourceBT->getKind() > TargetBT->getKind()) {
3288        // Don't warn about float constants that are precisely
3289        // representable in the target type.
3290        Expr::EvalResult result;
3291        if (E->Evaluate(result, S.Context)) {
3292          // Value might be a float, a float vector, or a float complex.
3293          if (IsSameFloatAfterCast(result.Val,
3294                   S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
3295                   S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
3296            return;
3297        }
3298
3299        if (isFromSystemMacro(S, CC))
3300          return;
3301
3302        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
3303      }
3304      return;
3305    }
3306
3307    // If the target is integral, always warn.
3308    if ((TargetBT && TargetBT->isInteger())) {
3309      if (isFromSystemMacro(S, CC))
3310        return;
3311
3312      Expr *InnerE = E->IgnoreParenImpCasts();
3313      // We also want to warn on, e.g., "int i = -1.234"
3314      if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE))
3315        if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus)
3316          InnerE = UOp->getSubExpr()->IgnoreParenImpCasts();
3317
3318      if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
3319        DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
3320      } else {
3321        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
3322      }
3323    }
3324
3325    return;
3326  }
3327
3328  if (!Source->isIntegerType() || !Target->isIntegerType())
3329    return;
3330
3331  if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)
3332           == Expr::NPCK_GNUNull) && Target->isIntegerType()) {
3333    S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer)
3334        << E->getSourceRange() << clang::SourceRange(CC);
3335    return;
3336  }
3337
3338  IntRange SourceRange = GetExprRange(S.Context, E);
3339  IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
3340
3341  if (SourceRange.Width > TargetRange.Width) {
3342    // If the source is a constant, use a default-on diagnostic.
3343    // TODO: this should happen for bitfield stores, too.
3344    llvm::APSInt Value(32);
3345    if (E->isIntegerConstantExpr(Value, S.Context)) {
3346      if (isFromSystemMacro(S, CC))
3347        return;
3348
3349      std::string PrettySourceValue = Value.toString(10);
3350      std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
3351
3352      S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant)
3353        << PrettySourceValue << PrettyTargetValue
3354        << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC);
3355      return;
3356    }
3357
3358    // People want to build with -Wshorten-64-to-32 and not -Wconversion.
3359    if (isFromSystemMacro(S, CC))
3360      return;
3361
3362    if (SourceRange.Width == 64 && TargetRange.Width == 32)
3363      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32);
3364    return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
3365  }
3366
3367  if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
3368      (!TargetRange.NonNegative && SourceRange.NonNegative &&
3369       SourceRange.Width == TargetRange.Width)) {
3370
3371    if (isFromSystemMacro(S, CC))
3372      return;
3373
3374    unsigned DiagID = diag::warn_impcast_integer_sign;
3375
3376    // Traditionally, gcc has warned about this under -Wsign-compare.
3377    // We also want to warn about it in -Wconversion.
3378    // So if -Wconversion is off, use a completely identical diagnostic
3379    // in the sign-compare group.
3380    // The conditional-checking code will
3381    if (ICContext) {
3382      DiagID = diag::warn_impcast_integer_sign_conditional;
3383      *ICContext = true;
3384    }
3385
3386    return DiagnoseImpCast(S, E, T, CC, DiagID);
3387  }
3388
3389  // Diagnose conversions between different enumeration types.
3390  // In C, we pretend that the type of an EnumConstantDecl is its enumeration
3391  // type, to give us better diagnostics.
3392  QualType SourceType = E->getType();
3393  if (!S.getLangOptions().CPlusPlus) {
3394    if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
3395      if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
3396        EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
3397        SourceType = S.Context.getTypeDeclType(Enum);
3398        Source = S.Context.getCanonicalType(SourceType).getTypePtr();
3399      }
3400  }
3401
3402  if (const EnumType *SourceEnum = Source->getAs<EnumType>())
3403    if (const EnumType *TargetEnum = Target->getAs<EnumType>())
3404      if ((SourceEnum->getDecl()->getIdentifier() ||
3405           SourceEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3406          (TargetEnum->getDecl()->getIdentifier() ||
3407           TargetEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3408          SourceEnum != TargetEnum) {
3409        if (isFromSystemMacro(S, CC))
3410          return;
3411
3412        return DiagnoseImpCast(S, E, SourceType, T, CC,
3413                               diag::warn_impcast_different_enum_types);
3414      }
3415
3416  return;
3417}
3418
3419void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T);
3420
3421void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
3422                             SourceLocation CC, bool &ICContext) {
3423  E = E->IgnoreParenImpCasts();
3424
3425  if (isa<ConditionalOperator>(E))
3426    return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T);
3427
3428  AnalyzeImplicitConversions(S, E, CC);
3429  if (E->getType() != T)
3430    return CheckImplicitConversion(S, E, T, CC, &ICContext);
3431  return;
3432}
3433
3434void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) {
3435  SourceLocation CC = E->getQuestionLoc();
3436
3437  AnalyzeImplicitConversions(S, E->getCond(), CC);
3438
3439  bool Suspicious = false;
3440  CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
3441  CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
3442
3443  // If -Wconversion would have warned about either of the candidates
3444  // for a signedness conversion to the context type...
3445  if (!Suspicious) return;
3446
3447  // ...but it's currently ignored...
3448  if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
3449                                 CC))
3450    return;
3451
3452  // ...then check whether it would have warned about either of the
3453  // candidates for a signedness conversion to the condition type.
3454  if (E->getType() == T) return;
3455
3456  Suspicious = false;
3457  CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
3458                          E->getType(), CC, &Suspicious);
3459  if (!Suspicious)
3460    CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
3461                            E->getType(), CC, &Suspicious);
3462}
3463
3464/// AnalyzeImplicitConversions - Find and report any interesting
3465/// implicit conversions in the given expression.  There are a couple
3466/// of competing diagnostics here, -Wconversion and -Wsign-compare.
3467void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
3468  QualType T = OrigE->getType();
3469  Expr *E = OrigE->IgnoreParenImpCasts();
3470
3471  // For conditional operators, we analyze the arguments as if they
3472  // were being fed directly into the output.
3473  if (isa<ConditionalOperator>(E)) {
3474    ConditionalOperator *CO = cast<ConditionalOperator>(E);
3475    CheckConditionalOperator(S, CO, T);
3476    return;
3477  }
3478
3479  // Go ahead and check any implicit conversions we might have skipped.
3480  // The non-canonical typecheck is just an optimization;
3481  // CheckImplicitConversion will filter out dead implicit conversions.
3482  if (E->getType() != T)
3483    CheckImplicitConversion(S, E, T, CC);
3484
3485  // Now continue drilling into this expression.
3486
3487  // Skip past explicit casts.
3488  if (isa<ExplicitCastExpr>(E)) {
3489    E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
3490    return AnalyzeImplicitConversions(S, E, CC);
3491  }
3492
3493  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
3494    // Do a somewhat different check with comparison operators.
3495    if (BO->isComparisonOp())
3496      return AnalyzeComparison(S, BO);
3497
3498    // And with assignments and compound assignments.
3499    if (BO->isAssignmentOp())
3500      return AnalyzeAssignment(S, BO);
3501  }
3502
3503  // These break the otherwise-useful invariant below.  Fortunately,
3504  // we don't really need to recurse into them, because any internal
3505  // expressions should have been analyzed already when they were
3506  // built into statements.
3507  if (isa<StmtExpr>(E)) return;
3508
3509  // Don't descend into unevaluated contexts.
3510  if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
3511
3512  // Now just recurse over the expression's children.
3513  CC = E->getExprLoc();
3514  for (Stmt::child_range I = E->children(); I; ++I)
3515    AnalyzeImplicitConversions(S, cast<Expr>(*I), CC);
3516}
3517
3518} // end anonymous namespace
3519
3520/// Diagnoses "dangerous" implicit conversions within the given
3521/// expression (which is a full expression).  Implements -Wconversion
3522/// and -Wsign-compare.
3523///
3524/// \param CC the "context" location of the implicit conversion, i.e.
3525///   the most location of the syntactic entity requiring the implicit
3526///   conversion
3527void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
3528  // Don't diagnose in unevaluated contexts.
3529  if (ExprEvalContexts.back().Context == Sema::Unevaluated)
3530    return;
3531
3532  // Don't diagnose for value- or type-dependent expressions.
3533  if (E->isTypeDependent() || E->isValueDependent())
3534    return;
3535
3536  // Check for array bounds violations in cases where the check isn't triggered
3537  // elsewhere for other Expr types (like BinaryOperators), e.g. when an
3538  // ArraySubscriptExpr is on the RHS of a variable initialization.
3539  CheckArrayAccess(E);
3540
3541  // This is not the right CC for (e.g.) a variable initialization.
3542  AnalyzeImplicitConversions(*this, E, CC);
3543}
3544
3545void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
3546                                       FieldDecl *BitField,
3547                                       Expr *Init) {
3548  (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
3549}
3550
3551/// CheckParmsForFunctionDef - Check that the parameters of the given
3552/// function are appropriate for the definition of a function. This
3553/// takes care of any checks that cannot be performed on the
3554/// declaration itself, e.g., that the types of each of the function
3555/// parameters are complete.
3556bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
3557                                    bool CheckParameterNames) {
3558  bool HasInvalidParm = false;
3559  for (; P != PEnd; ++P) {
3560    ParmVarDecl *Param = *P;
3561
3562    // C99 6.7.5.3p4: the parameters in a parameter type list in a
3563    // function declarator that is part of a function definition of
3564    // that function shall not have incomplete type.
3565    //
3566    // This is also C++ [dcl.fct]p6.
3567    if (!Param->isInvalidDecl() &&
3568        RequireCompleteType(Param->getLocation(), Param->getType(),
3569                               diag::err_typecheck_decl_incomplete_type)) {
3570      Param->setInvalidDecl();
3571      HasInvalidParm = true;
3572    }
3573
3574    // C99 6.9.1p5: If the declarator includes a parameter type list, the
3575    // declaration of each parameter shall include an identifier.
3576    if (CheckParameterNames &&
3577        Param->getIdentifier() == 0 &&
3578        !Param->isImplicit() &&
3579        !getLangOptions().CPlusPlus)
3580      Diag(Param->getLocation(), diag::err_parameter_name_omitted);
3581
3582    // C99 6.7.5.3p12:
3583    //   If the function declarator is not part of a definition of that
3584    //   function, parameters may have incomplete type and may use the [*]
3585    //   notation in their sequences of declarator specifiers to specify
3586    //   variable length array types.
3587    QualType PType = Param->getOriginalType();
3588    if (const ArrayType *AT = Context.getAsArrayType(PType)) {
3589      if (AT->getSizeModifier() == ArrayType::Star) {
3590        // FIXME: This diagnosic should point the the '[*]' if source-location
3591        // information is added for it.
3592        Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
3593      }
3594    }
3595  }
3596
3597  return HasInvalidParm;
3598}
3599
3600/// CheckCastAlign - Implements -Wcast-align, which warns when a
3601/// pointer cast increases the alignment requirements.
3602void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
3603  // This is actually a lot of work to potentially be doing on every
3604  // cast; don't do it if we're ignoring -Wcast_align (as is the default).
3605  if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
3606                                          TRange.getBegin())
3607        == Diagnostic::Ignored)
3608    return;
3609
3610  // Ignore dependent types.
3611  if (T->isDependentType() || Op->getType()->isDependentType())
3612    return;
3613
3614  // Require that the destination be a pointer type.
3615  const PointerType *DestPtr = T->getAs<PointerType>();
3616  if (!DestPtr) return;
3617
3618  // If the destination has alignment 1, we're done.
3619  QualType DestPointee = DestPtr->getPointeeType();
3620  if (DestPointee->isIncompleteType()) return;
3621  CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
3622  if (DestAlign.isOne()) return;
3623
3624  // Require that the source be a pointer type.
3625  const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
3626  if (!SrcPtr) return;
3627  QualType SrcPointee = SrcPtr->getPointeeType();
3628
3629  // Whitelist casts from cv void*.  We already implicitly
3630  // whitelisted casts to cv void*, since they have alignment 1.
3631  // Also whitelist casts involving incomplete types, which implicitly
3632  // includes 'void'.
3633  if (SrcPointee->isIncompleteType()) return;
3634
3635  CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
3636  if (SrcAlign >= DestAlign) return;
3637
3638  Diag(TRange.getBegin(), diag::warn_cast_align)
3639    << Op->getType() << T
3640    << static_cast<unsigned>(SrcAlign.getQuantity())
3641    << static_cast<unsigned>(DestAlign.getQuantity())
3642    << TRange << Op->getSourceRange();
3643}
3644
3645static const Type* getElementType(const Expr *BaseExpr) {
3646  const Type* EltType = BaseExpr->getType().getTypePtr();
3647  if (EltType->isAnyPointerType())
3648    return EltType->getPointeeType().getTypePtr();
3649  else if (EltType->isArrayType())
3650    return EltType->getBaseElementTypeUnsafe();
3651  return EltType;
3652}
3653
3654/// \brief Check whether this array fits the idiom of a size-one tail padded
3655/// array member of a struct.
3656///
3657/// We avoid emitting out-of-bounds access warnings for such arrays as they are
3658/// commonly used to emulate flexible arrays in C89 code.
3659static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size,
3660                                    const NamedDecl *ND) {
3661  if (Size != 1 || !ND) return false;
3662
3663  const FieldDecl *FD = dyn_cast<FieldDecl>(ND);
3664  if (!FD) return false;
3665
3666  // Don't consider sizes resulting from macro expansions or template argument
3667  // substitution to form C89 tail-padded arrays.
3668  ConstantArrayTypeLoc TL =
3669    cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc());
3670  const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr());
3671  if (!SizeExpr || SizeExpr->getExprLoc().isMacroID())
3672    return false;
3673
3674  const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext());
3675  if (!RD || !RD->isStruct())
3676    return false;
3677
3678  // See if this is the last field decl in the record.
3679  const Decl *D = FD;
3680  while ((D = D->getNextDeclInContext()))
3681    if (isa<FieldDecl>(D))
3682      return false;
3683  return true;
3684}
3685
3686void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr,
3687                            bool isSubscript, bool AllowOnePastEnd) {
3688  const Type* EffectiveType = getElementType(BaseExpr);
3689  BaseExpr = BaseExpr->IgnoreParenCasts();
3690  IndexExpr = IndexExpr->IgnoreParenCasts();
3691
3692  const ConstantArrayType *ArrayTy =
3693    Context.getAsConstantArrayType(BaseExpr->getType());
3694  if (!ArrayTy)
3695    return;
3696
3697  if (IndexExpr->isValueDependent())
3698    return;
3699  llvm::APSInt index;
3700  if (!IndexExpr->isIntegerConstantExpr(index, Context))
3701    return;
3702
3703  const NamedDecl *ND = NULL;
3704  if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
3705    ND = dyn_cast<NamedDecl>(DRE->getDecl());
3706  if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
3707    ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
3708
3709  if (index.isUnsigned() || !index.isNegative()) {
3710    llvm::APInt size = ArrayTy->getSize();
3711    if (!size.isStrictlyPositive())
3712      return;
3713
3714    const Type* BaseType = getElementType(BaseExpr);
3715    if (!isSubscript && BaseType != EffectiveType) {
3716      // Make sure we're comparing apples to apples when comparing index to size
3717      uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType);
3718      uint64_t array_typesize = Context.getTypeSize(BaseType);
3719      // Handle ptrarith_typesize being zero, such as when casting to void*
3720      if (!ptrarith_typesize) ptrarith_typesize = 1;
3721      if (ptrarith_typesize != array_typesize) {
3722        // There's a cast to a different size type involved
3723        uint64_t ratio = array_typesize / ptrarith_typesize;
3724        // TODO: Be smarter about handling cases where array_typesize is not a
3725        // multiple of ptrarith_typesize
3726        if (ptrarith_typesize * ratio == array_typesize)
3727          size *= llvm::APInt(size.getBitWidth(), ratio);
3728      }
3729    }
3730
3731    if (size.getBitWidth() > index.getBitWidth())
3732      index = index.sext(size.getBitWidth());
3733    else if (size.getBitWidth() < index.getBitWidth())
3734      size = size.sext(index.getBitWidth());
3735
3736    // For array subscripting the index must be less than size, but for pointer
3737    // arithmetic also allow the index (offset) to be equal to size since
3738    // computing the next address after the end of the array is legal and
3739    // commonly done e.g. in C++ iterators and range-based for loops.
3740    if (AllowOnePastEnd ? index.sle(size) : index.slt(size))
3741      return;
3742
3743    // Also don't warn for arrays of size 1 which are members of some
3744    // structure. These are often used to approximate flexible arrays in C89
3745    // code.
3746    if (IsTailPaddedMemberArray(*this, size, ND))
3747      return;
3748
3749    unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds;
3750    if (isSubscript)
3751      DiagID = diag::warn_array_index_exceeds_bounds;
3752
3753    DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
3754                        PDiag(DiagID) << index.toString(10, true)
3755                          << size.toString(10, true)
3756                          << (unsigned)size.getLimitedValue(~0U)
3757                          << IndexExpr->getSourceRange());
3758  } else {
3759    unsigned DiagID = diag::warn_array_index_precedes_bounds;
3760    if (!isSubscript) {
3761      DiagID = diag::warn_ptr_arith_precedes_bounds;
3762      if (index.isNegative()) index = -index;
3763    }
3764
3765    DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
3766                        PDiag(DiagID) << index.toString(10, true)
3767                          << IndexExpr->getSourceRange());
3768  }
3769
3770  if (ND)
3771    DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
3772                        PDiag(diag::note_array_index_out_of_bounds)
3773                          << ND->getDeclName());
3774}
3775
3776void Sema::CheckArrayAccess(const Expr *expr) {
3777  int AllowOnePastEnd = 0;
3778  while (expr) {
3779    expr = expr->IgnoreParenImpCasts();
3780    switch (expr->getStmtClass()) {
3781      case Stmt::ArraySubscriptExprClass: {
3782        const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr);
3783        CheckArrayAccess(ASE->getBase(), ASE->getIdx(), true,
3784                         AllowOnePastEnd > 0);
3785        return;
3786      }
3787      case Stmt::UnaryOperatorClass: {
3788        // Only unwrap the * and & unary operators
3789        const UnaryOperator *UO = cast<UnaryOperator>(expr);
3790        expr = UO->getSubExpr();
3791        switch (UO->getOpcode()) {
3792          case UO_AddrOf:
3793            AllowOnePastEnd++;
3794            break;
3795          case UO_Deref:
3796            AllowOnePastEnd--;
3797            break;
3798          default:
3799            return;
3800        }
3801        break;
3802      }
3803      case Stmt::ConditionalOperatorClass: {
3804        const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
3805        if (const Expr *lhs = cond->getLHS())
3806          CheckArrayAccess(lhs);
3807        if (const Expr *rhs = cond->getRHS())
3808          CheckArrayAccess(rhs);
3809        return;
3810      }
3811      default:
3812        return;
3813    }
3814  }
3815}
3816
3817//===--- CHECK: Objective-C retain cycles ----------------------------------//
3818
3819namespace {
3820  struct RetainCycleOwner {
3821    RetainCycleOwner() : Variable(0), Indirect(false) {}
3822    VarDecl *Variable;
3823    SourceRange Range;
3824    SourceLocation Loc;
3825    bool Indirect;
3826
3827    void setLocsFrom(Expr *e) {
3828      Loc = e->getExprLoc();
3829      Range = e->getSourceRange();
3830    }
3831  };
3832}
3833
3834/// Consider whether capturing the given variable can possibly lead to
3835/// a retain cycle.
3836static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) {
3837  // In ARC, it's captured strongly iff the variable has __strong
3838  // lifetime.  In MRR, it's captured strongly if the variable is
3839  // __block and has an appropriate type.
3840  if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
3841    return false;
3842
3843  owner.Variable = var;
3844  owner.setLocsFrom(ref);
3845  return true;
3846}
3847
3848static bool findRetainCycleOwner(Expr *e, RetainCycleOwner &owner) {
3849  while (true) {
3850    e = e->IgnoreParens();
3851    if (CastExpr *cast = dyn_cast<CastExpr>(e)) {
3852      switch (cast->getCastKind()) {
3853      case CK_BitCast:
3854      case CK_LValueBitCast:
3855      case CK_LValueToRValue:
3856      case CK_ARCReclaimReturnedObject:
3857        e = cast->getSubExpr();
3858        continue;
3859
3860      case CK_GetObjCProperty: {
3861        // Bail out if this isn't a strong explicit property.
3862        const ObjCPropertyRefExpr *pre = cast->getSubExpr()->getObjCProperty();
3863        if (pre->isImplicitProperty()) return false;
3864        ObjCPropertyDecl *property = pre->getExplicitProperty();
3865        if (!(property->getPropertyAttributes() &
3866              (ObjCPropertyDecl::OBJC_PR_retain |
3867               ObjCPropertyDecl::OBJC_PR_copy |
3868               ObjCPropertyDecl::OBJC_PR_strong)) &&
3869            !(property->getPropertyIvarDecl() &&
3870              property->getPropertyIvarDecl()->getType()
3871                .getObjCLifetime() == Qualifiers::OCL_Strong))
3872          return false;
3873
3874        owner.Indirect = true;
3875        e = const_cast<Expr*>(pre->getBase());
3876        continue;
3877      }
3878
3879      default:
3880        return false;
3881      }
3882    }
3883
3884    if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) {
3885      ObjCIvarDecl *ivar = ref->getDecl();
3886      if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
3887        return false;
3888
3889      // Try to find a retain cycle in the base.
3890      if (!findRetainCycleOwner(ref->getBase(), owner))
3891        return false;
3892
3893      if (ref->isFreeIvar()) owner.setLocsFrom(ref);
3894      owner.Indirect = true;
3895      return true;
3896    }
3897
3898    if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) {
3899      VarDecl *var = dyn_cast<VarDecl>(ref->getDecl());
3900      if (!var) return false;
3901      return considerVariable(var, ref, owner);
3902    }
3903
3904    if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) {
3905      owner.Variable = ref->getDecl();
3906      owner.setLocsFrom(ref);
3907      return true;
3908    }
3909
3910    if (MemberExpr *member = dyn_cast<MemberExpr>(e)) {
3911      if (member->isArrow()) return false;
3912
3913      // Don't count this as an indirect ownership.
3914      e = member->getBase();
3915      continue;
3916    }
3917
3918    // Array ivars?
3919
3920    return false;
3921  }
3922}
3923
3924namespace {
3925  struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> {
3926    FindCaptureVisitor(ASTContext &Context, VarDecl *variable)
3927      : EvaluatedExprVisitor<FindCaptureVisitor>(Context),
3928        Variable(variable), Capturer(0) {}
3929
3930    VarDecl *Variable;
3931    Expr *Capturer;
3932
3933    void VisitDeclRefExpr(DeclRefExpr *ref) {
3934      if (ref->getDecl() == Variable && !Capturer)
3935        Capturer = ref;
3936    }
3937
3938    void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) {
3939      if (ref->getDecl() == Variable && !Capturer)
3940        Capturer = ref;
3941    }
3942
3943    void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) {
3944      if (Capturer) return;
3945      Visit(ref->getBase());
3946      if (Capturer && ref->isFreeIvar())
3947        Capturer = ref;
3948    }
3949
3950    void VisitBlockExpr(BlockExpr *block) {
3951      // Look inside nested blocks
3952      if (block->getBlockDecl()->capturesVariable(Variable))
3953        Visit(block->getBlockDecl()->getBody());
3954    }
3955  };
3956}
3957
3958/// Check whether the given argument is a block which captures a
3959/// variable.
3960static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) {
3961  assert(owner.Variable && owner.Loc.isValid());
3962
3963  e = e->IgnoreParenCasts();
3964  BlockExpr *block = dyn_cast<BlockExpr>(e);
3965  if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable))
3966    return 0;
3967
3968  FindCaptureVisitor visitor(S.Context, owner.Variable);
3969  visitor.Visit(block->getBlockDecl()->getBody());
3970  return visitor.Capturer;
3971}
3972
3973static void diagnoseRetainCycle(Sema &S, Expr *capturer,
3974                                RetainCycleOwner &owner) {
3975  assert(capturer);
3976  assert(owner.Variable && owner.Loc.isValid());
3977
3978  S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle)
3979    << owner.Variable << capturer->getSourceRange();
3980  S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner)
3981    << owner.Indirect << owner.Range;
3982}
3983
3984/// Check for a keyword selector that starts with the word 'add' or
3985/// 'set'.
3986static bool isSetterLikeSelector(Selector sel) {
3987  if (sel.isUnarySelector()) return false;
3988
3989  StringRef str = sel.getNameForSlot(0);
3990  while (!str.empty() && str.front() == '_') str = str.substr(1);
3991  if (str.startswith("set") || str.startswith("add"))
3992    str = str.substr(3);
3993  else
3994    return false;
3995
3996  if (str.empty()) return true;
3997  return !islower(str.front());
3998}
3999
4000/// Check a message send to see if it's likely to cause a retain cycle.
4001void Sema::checkRetainCycles(ObjCMessageExpr *msg) {
4002  // Only check instance methods whose selector looks like a setter.
4003  if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector()))
4004    return;
4005
4006  // Try to find a variable that the receiver is strongly owned by.
4007  RetainCycleOwner owner;
4008  if (msg->getReceiverKind() == ObjCMessageExpr::Instance) {
4009    if (!findRetainCycleOwner(msg->getInstanceReceiver(), owner))
4010      return;
4011  } else {
4012    assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance);
4013    owner.Variable = getCurMethodDecl()->getSelfDecl();
4014    owner.Loc = msg->getSuperLoc();
4015    owner.Range = msg->getSuperLoc();
4016  }
4017
4018  // Check whether the receiver is captured by any of the arguments.
4019  for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i)
4020    if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner))
4021      return diagnoseRetainCycle(*this, capturer, owner);
4022}
4023
4024/// Check a property assign to see if it's likely to cause a retain cycle.
4025void Sema::checkRetainCycles(Expr *receiver, Expr *argument) {
4026  RetainCycleOwner owner;
4027  if (!findRetainCycleOwner(receiver, owner))
4028    return;
4029
4030  if (Expr *capturer = findCapturingExpr(*this, argument, owner))
4031    diagnoseRetainCycle(*this, capturer, owner);
4032}
4033
4034bool Sema::checkUnsafeAssigns(SourceLocation Loc,
4035                              QualType LHS, Expr *RHS) {
4036  Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime();
4037  if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone)
4038    return false;
4039  // strip off any implicit cast added to get to the one arc-specific
4040  while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
4041    if (cast->getCastKind() == CK_ARCConsumeObject) {
4042      Diag(Loc, diag::warn_arc_retained_assign)
4043        << (LT == Qualifiers::OCL_ExplicitNone)
4044        << RHS->getSourceRange();
4045      return true;
4046    }
4047    RHS = cast->getSubExpr();
4048  }
4049  return false;
4050}
4051
4052void Sema::checkUnsafeExprAssigns(SourceLocation Loc,
4053                              Expr *LHS, Expr *RHS) {
4054  QualType LHSType = LHS->getType();
4055  if (checkUnsafeAssigns(Loc, LHSType, RHS))
4056    return;
4057  Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime();
4058  // FIXME. Check for other life times.
4059  if (LT != Qualifiers::OCL_None)
4060    return;
4061
4062  if (ObjCPropertyRefExpr *PRE = dyn_cast<ObjCPropertyRefExpr>(LHS)) {
4063    if (PRE->isImplicitProperty())
4064      return;
4065    const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
4066    if (!PD)
4067      return;
4068
4069    unsigned Attributes = PD->getPropertyAttributes();
4070    if (Attributes & ObjCPropertyDecl::OBJC_PR_assign)
4071      while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
4072        if (cast->getCastKind() == CK_ARCConsumeObject) {
4073          Diag(Loc, diag::warn_arc_retained_property_assign)
4074          << RHS->getSourceRange();
4075          return;
4076        }
4077        RHS = cast->getSubExpr();
4078      }
4079  }
4080}
4081