SemaChecking.cpp revision a7da2155167676a6a5d9fca4de947a9cab2a4908
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10//  This file implements extra semantic analysis beyond what is enforced
11//  by the C type system.
12//
13//===----------------------------------------------------------------------===//
14
15#include "clang/Sema/Initialization.h"
16#include "clang/Sema/Sema.h"
17#include "clang/Sema/SemaInternal.h"
18#include "clang/Sema/Initialization.h"
19#include "clang/Sema/ScopeInfo.h"
20#include "clang/Analysis/Analyses/FormatString.h"
21#include "clang/AST/ASTContext.h"
22#include "clang/AST/CharUnits.h"
23#include "clang/AST/DeclCXX.h"
24#include "clang/AST/DeclObjC.h"
25#include "clang/AST/ExprCXX.h"
26#include "clang/AST/ExprObjC.h"
27#include "clang/AST/EvaluatedExprVisitor.h"
28#include "clang/AST/DeclObjC.h"
29#include "clang/AST/StmtCXX.h"
30#include "clang/AST/StmtObjC.h"
31#include "clang/Lex/Preprocessor.h"
32#include "llvm/ADT/BitVector.h"
33#include "llvm/ADT/STLExtras.h"
34#include "llvm/Support/raw_ostream.h"
35#include "clang/Basic/TargetBuiltins.h"
36#include "clang/Basic/TargetInfo.h"
37#include "clang/Basic/ConvertUTF.h"
38#include <limits>
39using namespace clang;
40using namespace sema;
41
42SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL,
43                                                    unsigned ByteNo) const {
44  return SL->getLocationOfByte(ByteNo, PP.getSourceManager(),
45                               PP.getLangOptions(), PP.getTargetInfo());
46}
47
48
49/// CheckablePrintfAttr - does a function call have a "printf" attribute
50/// and arguments that merit checking?
51bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) {
52  if (Format->getType() == "printf") return true;
53  if (Format->getType() == "printf0") {
54    // printf0 allows null "format" string; if so don't check format/args
55    unsigned format_idx = Format->getFormatIdx() - 1;
56    // Does the index refer to the implicit object argument?
57    if (isa<CXXMemberCallExpr>(TheCall)) {
58      if (format_idx == 0)
59        return false;
60      --format_idx;
61    }
62    if (format_idx < TheCall->getNumArgs()) {
63      Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts();
64      if (!Format->isNullPointerConstant(Context,
65                                         Expr::NPC_ValueDependentIsNull))
66        return true;
67    }
68  }
69  return false;
70}
71
72/// Checks that a call expression's argument count is the desired number.
73/// This is useful when doing custom type-checking.  Returns true on error.
74static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) {
75  unsigned argCount = call->getNumArgs();
76  if (argCount == desiredArgCount) return false;
77
78  if (argCount < desiredArgCount)
79    return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args)
80        << 0 /*function call*/ << desiredArgCount << argCount
81        << call->getSourceRange();
82
83  // Highlight all the excess arguments.
84  SourceRange range(call->getArg(desiredArgCount)->getLocStart(),
85                    call->getArg(argCount - 1)->getLocEnd());
86
87  return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args)
88    << 0 /*function call*/ << desiredArgCount << argCount
89    << call->getArg(1)->getSourceRange();
90}
91
92/// CheckBuiltinAnnotationString - Checks that string argument to the builtin
93/// annotation is a non wide string literal.
94static bool CheckBuiltinAnnotationString(Sema &S, Expr *Arg) {
95  Arg = Arg->IgnoreParenCasts();
96  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
97  if (!Literal || !Literal->isAscii()) {
98    S.Diag(Arg->getLocStart(), diag::err_builtin_annotation_not_string_constant)
99      << Arg->getSourceRange();
100    return true;
101  }
102  return false;
103}
104
105ExprResult
106Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
107  ExprResult TheCallResult(Owned(TheCall));
108
109  // Find out if any arguments are required to be integer constant expressions.
110  unsigned ICEArguments = 0;
111  ASTContext::GetBuiltinTypeError Error;
112  Context.GetBuiltinType(BuiltinID, Error, &ICEArguments);
113  if (Error != ASTContext::GE_None)
114    ICEArguments = 0;  // Don't diagnose previously diagnosed errors.
115
116  // If any arguments are required to be ICE's, check and diagnose.
117  for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) {
118    // Skip arguments not required to be ICE's.
119    if ((ICEArguments & (1 << ArgNo)) == 0) continue;
120
121    llvm::APSInt Result;
122    if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
123      return true;
124    ICEArguments &= ~(1 << ArgNo);
125  }
126
127  switch (BuiltinID) {
128  case Builtin::BI__builtin___CFStringMakeConstantString:
129    assert(TheCall->getNumArgs() == 1 &&
130           "Wrong # arguments to builtin CFStringMakeConstantString");
131    if (CheckObjCString(TheCall->getArg(0)))
132      return ExprError();
133    break;
134  case Builtin::BI__builtin_stdarg_start:
135  case Builtin::BI__builtin_va_start:
136    if (SemaBuiltinVAStart(TheCall))
137      return ExprError();
138    break;
139  case Builtin::BI__builtin_isgreater:
140  case Builtin::BI__builtin_isgreaterequal:
141  case Builtin::BI__builtin_isless:
142  case Builtin::BI__builtin_islessequal:
143  case Builtin::BI__builtin_islessgreater:
144  case Builtin::BI__builtin_isunordered:
145    if (SemaBuiltinUnorderedCompare(TheCall))
146      return ExprError();
147    break;
148  case Builtin::BI__builtin_fpclassify:
149    if (SemaBuiltinFPClassification(TheCall, 6))
150      return ExprError();
151    break;
152  case Builtin::BI__builtin_isfinite:
153  case Builtin::BI__builtin_isinf:
154  case Builtin::BI__builtin_isinf_sign:
155  case Builtin::BI__builtin_isnan:
156  case Builtin::BI__builtin_isnormal:
157    if (SemaBuiltinFPClassification(TheCall, 1))
158      return ExprError();
159    break;
160  case Builtin::BI__builtin_shufflevector:
161    return SemaBuiltinShuffleVector(TheCall);
162    // TheCall will be freed by the smart pointer here, but that's fine, since
163    // SemaBuiltinShuffleVector guts it, but then doesn't release it.
164  case Builtin::BI__builtin_prefetch:
165    if (SemaBuiltinPrefetch(TheCall))
166      return ExprError();
167    break;
168  case Builtin::BI__builtin_object_size:
169    if (SemaBuiltinObjectSize(TheCall))
170      return ExprError();
171    break;
172  case Builtin::BI__builtin_longjmp:
173    if (SemaBuiltinLongjmp(TheCall))
174      return ExprError();
175    break;
176
177  case Builtin::BI__builtin_classify_type:
178    if (checkArgCount(*this, TheCall, 1)) return true;
179    TheCall->setType(Context.IntTy);
180    break;
181  case Builtin::BI__builtin_constant_p:
182    if (checkArgCount(*this, TheCall, 1)) return true;
183    TheCall->setType(Context.IntTy);
184    break;
185  case Builtin::BI__sync_fetch_and_add:
186  case Builtin::BI__sync_fetch_and_sub:
187  case Builtin::BI__sync_fetch_and_or:
188  case Builtin::BI__sync_fetch_and_and:
189  case Builtin::BI__sync_fetch_and_xor:
190  case Builtin::BI__sync_add_and_fetch:
191  case Builtin::BI__sync_sub_and_fetch:
192  case Builtin::BI__sync_and_and_fetch:
193  case Builtin::BI__sync_or_and_fetch:
194  case Builtin::BI__sync_xor_and_fetch:
195  case Builtin::BI__sync_val_compare_and_swap:
196  case Builtin::BI__sync_bool_compare_and_swap:
197  case Builtin::BI__sync_lock_test_and_set:
198  case Builtin::BI__sync_lock_release:
199  case Builtin::BI__sync_swap:
200    return SemaBuiltinAtomicOverloaded(move(TheCallResult));
201  case Builtin::BI__atomic_load:
202    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Load);
203  case Builtin::BI__atomic_store:
204    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Store);
205  case Builtin::BI__atomic_exchange:
206    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xchg);
207  case Builtin::BI__atomic_compare_exchange_strong:
208    return SemaAtomicOpsOverloaded(move(TheCallResult),
209                                   AtomicExpr::CmpXchgStrong);
210  case Builtin::BI__atomic_compare_exchange_weak:
211    return SemaAtomicOpsOverloaded(move(TheCallResult),
212                                   AtomicExpr::CmpXchgWeak);
213  case Builtin::BI__atomic_fetch_add:
214    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Add);
215  case Builtin::BI__atomic_fetch_sub:
216    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Sub);
217  case Builtin::BI__atomic_fetch_and:
218    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::And);
219  case Builtin::BI__atomic_fetch_or:
220    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Or);
221  case Builtin::BI__atomic_fetch_xor:
222    return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xor);
223  case Builtin::BI__builtin_annotation:
224    if (CheckBuiltinAnnotationString(*this, TheCall->getArg(1)))
225      return ExprError();
226    break;
227  }
228
229  // Since the target specific builtins for each arch overlap, only check those
230  // of the arch we are compiling for.
231  if (BuiltinID >= Builtin::FirstTSBuiltin) {
232    switch (Context.getTargetInfo().getTriple().getArch()) {
233      case llvm::Triple::arm:
234      case llvm::Triple::thumb:
235        if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall))
236          return ExprError();
237        break;
238      default:
239        break;
240    }
241  }
242
243  return move(TheCallResult);
244}
245
246// Get the valid immediate range for the specified NEON type code.
247static unsigned RFT(unsigned t, bool shift = false) {
248  bool quad = t & 0x10;
249
250  switch (t & 0x7) {
251    case 0: // i8
252      return shift ? 7 : (8 << (int)quad) - 1;
253    case 1: // i16
254      return shift ? 15 : (4 << (int)quad) - 1;
255    case 2: // i32
256      return shift ? 31 : (2 << (int)quad) - 1;
257    case 3: // i64
258      return shift ? 63 : (1 << (int)quad) - 1;
259    case 4: // f32
260      assert(!shift && "cannot shift float types!");
261      return (2 << (int)quad) - 1;
262    case 5: // poly8
263      return shift ? 7 : (8 << (int)quad) - 1;
264    case 6: // poly16
265      return shift ? 15 : (4 << (int)quad) - 1;
266    case 7: // float16
267      assert(!shift && "cannot shift float types!");
268      return (4 << (int)quad) - 1;
269  }
270  return 0;
271}
272
273bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
274  llvm::APSInt Result;
275
276  unsigned mask = 0;
277  unsigned TV = 0;
278  switch (BuiltinID) {
279#define GET_NEON_OVERLOAD_CHECK
280#include "clang/Basic/arm_neon.inc"
281#undef GET_NEON_OVERLOAD_CHECK
282  }
283
284  // For NEON intrinsics which are overloaded on vector element type, validate
285  // the immediate which specifies which variant to emit.
286  if (mask) {
287    unsigned ArgNo = TheCall->getNumArgs()-1;
288    if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
289      return true;
290
291    TV = Result.getLimitedValue(32);
292    if ((TV > 31) || (mask & (1 << TV)) == 0)
293      return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code)
294        << TheCall->getArg(ArgNo)->getSourceRange();
295  }
296
297  // For NEON intrinsics which take an immediate value as part of the
298  // instruction, range check them here.
299  unsigned i = 0, l = 0, u = 0;
300  switch (BuiltinID) {
301  default: return false;
302  case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break;
303  case ARM::BI__builtin_arm_usat: i = 1; u = 31; break;
304  case ARM::BI__builtin_arm_vcvtr_f:
305  case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break;
306#define GET_NEON_IMMEDIATE_CHECK
307#include "clang/Basic/arm_neon.inc"
308#undef GET_NEON_IMMEDIATE_CHECK
309  };
310
311  // Check that the immediate argument is actually a constant.
312  if (SemaBuiltinConstantArg(TheCall, i, Result))
313    return true;
314
315  // Range check against the upper/lower values for this isntruction.
316  unsigned Val = Result.getZExtValue();
317  if (Val < l || Val > (u + l))
318    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
319      << l << u+l << TheCall->getArg(i)->getSourceRange();
320
321  // FIXME: VFP Intrinsics should error if VFP not present.
322  return false;
323}
324
325/// CheckFunctionCall - Check a direct function call for various correctness
326/// and safety properties not strictly enforced by the C type system.
327bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) {
328  // Get the IdentifierInfo* for the called function.
329  IdentifierInfo *FnInfo = FDecl->getIdentifier();
330
331  // None of the checks below are needed for functions that don't have
332  // simple names (e.g., C++ conversion functions).
333  if (!FnInfo)
334    return false;
335
336  // FIXME: This mechanism should be abstracted to be less fragile and
337  // more efficient. For example, just map function ids to custom
338  // handlers.
339
340  // Printf and scanf checking.
341  for (specific_attr_iterator<FormatAttr>
342         i = FDecl->specific_attr_begin<FormatAttr>(),
343         e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) {
344
345    const FormatAttr *Format = *i;
346    const bool b = Format->getType() == "scanf";
347    if (b || CheckablePrintfAttr(Format, TheCall)) {
348      bool HasVAListArg = Format->getFirstArg() == 0;
349      CheckPrintfScanfArguments(TheCall, HasVAListArg,
350                                Format->getFormatIdx() - 1,
351                                HasVAListArg ? 0 : Format->getFirstArg() - 1,
352                                !b);
353    }
354  }
355
356  for (specific_attr_iterator<NonNullAttr>
357         i = FDecl->specific_attr_begin<NonNullAttr>(),
358         e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) {
359    CheckNonNullArguments(*i, TheCall->getArgs(),
360                          TheCall->getCallee()->getLocStart());
361  }
362
363  CheckStaticArrayArguments(FDecl, TheCall->getArgs(),
364                            TheCall->getCallee()->getLocStart());
365
366  // Builtin handling
367  int CMF = -1;
368  switch (FDecl->getBuiltinID()) {
369  case Builtin::BI__builtin_memset:
370  case Builtin::BI__builtin___memset_chk:
371  case Builtin::BImemset:
372    CMF = CMF_Memset;
373    break;
374
375  case Builtin::BI__builtin_memcpy:
376  case Builtin::BI__builtin___memcpy_chk:
377  case Builtin::BImemcpy:
378    CMF = CMF_Memcpy;
379    break;
380
381  case Builtin::BI__builtin_memmove:
382  case Builtin::BI__builtin___memmove_chk:
383  case Builtin::BImemmove:
384    CMF = CMF_Memmove;
385    break;
386
387  case Builtin::BIstrlcpy:
388  case Builtin::BIstrlcat:
389    CheckStrlcpycatArguments(TheCall, FnInfo);
390    break;
391
392  case Builtin::BI__builtin_memcmp:
393    CMF = CMF_Memcmp;
394    break;
395
396  case Builtin::BI__builtin_strncpy:
397  case Builtin::BI__builtin___strncpy_chk:
398  case Builtin::BIstrncpy:
399    CMF = CMF_Strncpy;
400    break;
401
402  case Builtin::BI__builtin_strncmp:
403    CMF = CMF_Strncmp;
404    break;
405
406  case Builtin::BI__builtin_strncasecmp:
407    CMF = CMF_Strncasecmp;
408    break;
409
410  case Builtin::BI__builtin_strncat:
411  case Builtin::BIstrncat:
412    CMF = CMF_Strncat;
413    break;
414
415  case Builtin::BI__builtin_strndup:
416  case Builtin::BIstrndup:
417    CMF = CMF_Strndup;
418    break;
419
420  default:
421    if (FDecl->getLinkage() == ExternalLinkage &&
422        (!getLangOptions().CPlusPlus || FDecl->isExternC())) {
423      if (FnInfo->isStr("memset"))
424        CMF = CMF_Memset;
425      else if (FnInfo->isStr("memcpy"))
426        CMF = CMF_Memcpy;
427      else if (FnInfo->isStr("memmove"))
428        CMF = CMF_Memmove;
429      else if (FnInfo->isStr("memcmp"))
430        CMF = CMF_Memcmp;
431      else if (FnInfo->isStr("strncpy"))
432        CMF = CMF_Strncpy;
433      else if (FnInfo->isStr("strncmp"))
434        CMF = CMF_Strncmp;
435      else if (FnInfo->isStr("strncasecmp"))
436        CMF = CMF_Strncasecmp;
437      else if (FnInfo->isStr("strncat"))
438        CMF = CMF_Strncat;
439      else if (FnInfo->isStr("strndup"))
440        CMF = CMF_Strndup;
441    }
442    break;
443  }
444
445  // Memset/memcpy/memmove handling
446  if (CMF != -1)
447    CheckMemaccessArguments(TheCall, CheckedMemoryFunction(CMF), FnInfo);
448
449  return false;
450}
451
452bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
453  // Printf checking.
454  const FormatAttr *Format = NDecl->getAttr<FormatAttr>();
455  if (!Format)
456    return false;
457
458  const VarDecl *V = dyn_cast<VarDecl>(NDecl);
459  if (!V)
460    return false;
461
462  QualType Ty = V->getType();
463  if (!Ty->isBlockPointerType())
464    return false;
465
466  const bool b = Format->getType() == "scanf";
467  if (!b && !CheckablePrintfAttr(Format, TheCall))
468    return false;
469
470  bool HasVAListArg = Format->getFirstArg() == 0;
471  CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
472                            HasVAListArg ? 0 : Format->getFirstArg() - 1, !b);
473
474  return false;
475}
476
477ExprResult
478Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, AtomicExpr::AtomicOp Op) {
479  CallExpr *TheCall = cast<CallExpr>(TheCallResult.get());
480  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
481
482  // All these operations take one of the following four forms:
483  // T   __atomic_load(_Atomic(T)*, int)                              (loads)
484  // T*  __atomic_add(_Atomic(T*)*, ptrdiff_t, int)         (pointer add/sub)
485  // int __atomic_compare_exchange_strong(_Atomic(T)*, T*, T, int, int)
486  //                                                                (cmpxchg)
487  // T   __atomic_exchange(_Atomic(T)*, T, int)             (everything else)
488  // where T is an appropriate type, and the int paremeterss are for orderings.
489  unsigned NumVals = 1;
490  unsigned NumOrders = 1;
491  if (Op == AtomicExpr::Load) {
492    NumVals = 0;
493  } else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) {
494    NumVals = 2;
495    NumOrders = 2;
496  }
497
498  if (TheCall->getNumArgs() < NumVals+NumOrders+1) {
499    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
500      << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs()
501      << TheCall->getCallee()->getSourceRange();
502    return ExprError();
503  } else if (TheCall->getNumArgs() > NumVals+NumOrders+1) {
504    Diag(TheCall->getArg(NumVals+NumOrders+1)->getLocStart(),
505         diag::err_typecheck_call_too_many_args)
506      << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs()
507      << TheCall->getCallee()->getSourceRange();
508    return ExprError();
509  }
510
511  // Inspect the first argument of the atomic operation.  This should always be
512  // a pointer to an _Atomic type.
513  Expr *Ptr = TheCall->getArg(0);
514  Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get();
515  const PointerType *pointerType = Ptr->getType()->getAs<PointerType>();
516  if (!pointerType) {
517    Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic)
518      << Ptr->getType() << Ptr->getSourceRange();
519    return ExprError();
520  }
521
522  QualType AtomTy = pointerType->getPointeeType();
523  if (!AtomTy->isAtomicType()) {
524    Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic)
525      << Ptr->getType() << Ptr->getSourceRange();
526    return ExprError();
527  }
528  QualType ValType = AtomTy->getAs<AtomicType>()->getValueType();
529
530  if ((Op == AtomicExpr::Add || Op == AtomicExpr::Sub) &&
531      !ValType->isIntegerType() && !ValType->isPointerType()) {
532    Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr)
533      << Ptr->getType() << Ptr->getSourceRange();
534    return ExprError();
535  }
536
537  if (!ValType->isIntegerType() &&
538      (Op == AtomicExpr::And || Op == AtomicExpr::Or || Op == AtomicExpr::Xor)){
539    Diag(DRE->getLocStart(), diag::err_atomic_op_logical_needs_atomic_int)
540      << Ptr->getType() << Ptr->getSourceRange();
541    return ExprError();
542  }
543
544  switch (ValType.getObjCLifetime()) {
545  case Qualifiers::OCL_None:
546  case Qualifiers::OCL_ExplicitNone:
547    // okay
548    break;
549
550  case Qualifiers::OCL_Weak:
551  case Qualifiers::OCL_Strong:
552  case Qualifiers::OCL_Autoreleasing:
553    Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
554      << ValType << Ptr->getSourceRange();
555    return ExprError();
556  }
557
558  QualType ResultType = ValType;
559  if (Op == AtomicExpr::Store)
560    ResultType = Context.VoidTy;
561  else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong)
562    ResultType = Context.BoolTy;
563
564  // The first argument --- the pointer --- has a fixed type; we
565  // deduce the types of the rest of the arguments accordingly.  Walk
566  // the remaining arguments, converting them to the deduced value type.
567  for (unsigned i = 1; i != NumVals+NumOrders+1; ++i) {
568    ExprResult Arg = TheCall->getArg(i);
569    QualType Ty;
570    if (i < NumVals+1) {
571      // The second argument to a cmpxchg is a pointer to the data which will
572      // be exchanged. The second argument to a pointer add/subtract is the
573      // amount to add/subtract, which must be a ptrdiff_t.  The third
574      // argument to a cmpxchg and the second argument in all other cases
575      // is the type of the value.
576      if (i == 1 && (Op == AtomicExpr::CmpXchgWeak ||
577                     Op == AtomicExpr::CmpXchgStrong))
578         Ty = Context.getPointerType(ValType.getUnqualifiedType());
579      else if (!ValType->isIntegerType() &&
580               (Op == AtomicExpr::Add || Op == AtomicExpr::Sub))
581        Ty = Context.getPointerDiffType();
582      else
583        Ty = ValType;
584    } else {
585      // The order(s) are always converted to int.
586      Ty = Context.IntTy;
587    }
588    InitializedEntity Entity =
589        InitializedEntity::InitializeParameter(Context, Ty, false);
590    Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
591    if (Arg.isInvalid())
592      return true;
593    TheCall->setArg(i, Arg.get());
594  }
595
596  SmallVector<Expr*, 5> SubExprs;
597  SubExprs.push_back(Ptr);
598  if (Op == AtomicExpr::Load) {
599    SubExprs.push_back(TheCall->getArg(1)); // Order
600  } else if (Op != AtomicExpr::CmpXchgWeak && Op != AtomicExpr::CmpXchgStrong) {
601    SubExprs.push_back(TheCall->getArg(2)); // Order
602    SubExprs.push_back(TheCall->getArg(1)); // Val1
603  } else {
604    SubExprs.push_back(TheCall->getArg(3)); // Order
605    SubExprs.push_back(TheCall->getArg(1)); // Val1
606    SubExprs.push_back(TheCall->getArg(2)); // Val2
607    SubExprs.push_back(TheCall->getArg(4)); // OrderFail
608  }
609
610  return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(),
611                                        SubExprs.data(), SubExprs.size(),
612                                        ResultType, Op,
613                                        TheCall->getRParenLoc()));
614}
615
616
617/// checkBuiltinArgument - Given a call to a builtin function, perform
618/// normal type-checking on the given argument, updating the call in
619/// place.  This is useful when a builtin function requires custom
620/// type-checking for some of its arguments but not necessarily all of
621/// them.
622///
623/// Returns true on error.
624static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) {
625  FunctionDecl *Fn = E->getDirectCallee();
626  assert(Fn && "builtin call without direct callee!");
627
628  ParmVarDecl *Param = Fn->getParamDecl(ArgIndex);
629  InitializedEntity Entity =
630    InitializedEntity::InitializeParameter(S.Context, Param);
631
632  ExprResult Arg = E->getArg(0);
633  Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg);
634  if (Arg.isInvalid())
635    return true;
636
637  E->setArg(ArgIndex, Arg.take());
638  return false;
639}
640
641/// SemaBuiltinAtomicOverloaded - We have a call to a function like
642/// __sync_fetch_and_add, which is an overloaded function based on the pointer
643/// type of its first argument.  The main ActOnCallExpr routines have already
644/// promoted the types of arguments because all of these calls are prototyped as
645/// void(...).
646///
647/// This function goes through and does final semantic checking for these
648/// builtins,
649ExprResult
650Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
651  CallExpr *TheCall = (CallExpr *)TheCallResult.get();
652  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
653  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
654
655  // Ensure that we have at least one argument to do type inference from.
656  if (TheCall->getNumArgs() < 1) {
657    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
658      << 0 << 1 << TheCall->getNumArgs()
659      << TheCall->getCallee()->getSourceRange();
660    return ExprError();
661  }
662
663  // Inspect the first argument of the atomic builtin.  This should always be
664  // a pointer type, whose element is an integral scalar or pointer type.
665  // Because it is a pointer type, we don't have to worry about any implicit
666  // casts here.
667  // FIXME: We don't allow floating point scalars as input.
668  Expr *FirstArg = TheCall->getArg(0);
669  const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>();
670  if (!pointerType) {
671    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
672      << FirstArg->getType() << FirstArg->getSourceRange();
673    return ExprError();
674  }
675
676  QualType ValType = pointerType->getPointeeType();
677  if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
678      !ValType->isBlockPointerType()) {
679    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
680      << FirstArg->getType() << FirstArg->getSourceRange();
681    return ExprError();
682  }
683
684  switch (ValType.getObjCLifetime()) {
685  case Qualifiers::OCL_None:
686  case Qualifiers::OCL_ExplicitNone:
687    // okay
688    break;
689
690  case Qualifiers::OCL_Weak:
691  case Qualifiers::OCL_Strong:
692  case Qualifiers::OCL_Autoreleasing:
693    Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
694      << ValType << FirstArg->getSourceRange();
695    return ExprError();
696  }
697
698  // Strip any qualifiers off ValType.
699  ValType = ValType.getUnqualifiedType();
700
701  // The majority of builtins return a value, but a few have special return
702  // types, so allow them to override appropriately below.
703  QualType ResultType = ValType;
704
705  // We need to figure out which concrete builtin this maps onto.  For example,
706  // __sync_fetch_and_add with a 2 byte object turns into
707  // __sync_fetch_and_add_2.
708#define BUILTIN_ROW(x) \
709  { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
710    Builtin::BI##x##_8, Builtin::BI##x##_16 }
711
712  static const unsigned BuiltinIndices[][5] = {
713    BUILTIN_ROW(__sync_fetch_and_add),
714    BUILTIN_ROW(__sync_fetch_and_sub),
715    BUILTIN_ROW(__sync_fetch_and_or),
716    BUILTIN_ROW(__sync_fetch_and_and),
717    BUILTIN_ROW(__sync_fetch_and_xor),
718
719    BUILTIN_ROW(__sync_add_and_fetch),
720    BUILTIN_ROW(__sync_sub_and_fetch),
721    BUILTIN_ROW(__sync_and_and_fetch),
722    BUILTIN_ROW(__sync_or_and_fetch),
723    BUILTIN_ROW(__sync_xor_and_fetch),
724
725    BUILTIN_ROW(__sync_val_compare_and_swap),
726    BUILTIN_ROW(__sync_bool_compare_and_swap),
727    BUILTIN_ROW(__sync_lock_test_and_set),
728    BUILTIN_ROW(__sync_lock_release),
729    BUILTIN_ROW(__sync_swap)
730  };
731#undef BUILTIN_ROW
732
733  // Determine the index of the size.
734  unsigned SizeIndex;
735  switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
736  case 1: SizeIndex = 0; break;
737  case 2: SizeIndex = 1; break;
738  case 4: SizeIndex = 2; break;
739  case 8: SizeIndex = 3; break;
740  case 16: SizeIndex = 4; break;
741  default:
742    Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
743      << FirstArg->getType() << FirstArg->getSourceRange();
744    return ExprError();
745  }
746
747  // Each of these builtins has one pointer argument, followed by some number of
748  // values (0, 1 or 2) followed by a potentially empty varags list of stuff
749  // that we ignore.  Find out which row of BuiltinIndices to read from as well
750  // as the number of fixed args.
751  unsigned BuiltinID = FDecl->getBuiltinID();
752  unsigned BuiltinIndex, NumFixed = 1;
753  switch (BuiltinID) {
754  default: llvm_unreachable("Unknown overloaded atomic builtin!");
755  case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break;
756  case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break;
757  case Builtin::BI__sync_fetch_and_or:  BuiltinIndex = 2; break;
758  case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break;
759  case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break;
760
761  case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break;
762  case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break;
763  case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break;
764  case Builtin::BI__sync_or_and_fetch:  BuiltinIndex = 8; break;
765  case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break;
766
767  case Builtin::BI__sync_val_compare_and_swap:
768    BuiltinIndex = 10;
769    NumFixed = 2;
770    break;
771  case Builtin::BI__sync_bool_compare_and_swap:
772    BuiltinIndex = 11;
773    NumFixed = 2;
774    ResultType = Context.BoolTy;
775    break;
776  case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break;
777  case Builtin::BI__sync_lock_release:
778    BuiltinIndex = 13;
779    NumFixed = 0;
780    ResultType = Context.VoidTy;
781    break;
782  case Builtin::BI__sync_swap: BuiltinIndex = 14; break;
783  }
784
785  // Now that we know how many fixed arguments we expect, first check that we
786  // have at least that many.
787  if (TheCall->getNumArgs() < 1+NumFixed) {
788    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
789      << 0 << 1+NumFixed << TheCall->getNumArgs()
790      << TheCall->getCallee()->getSourceRange();
791    return ExprError();
792  }
793
794  // Get the decl for the concrete builtin from this, we can tell what the
795  // concrete integer type we should convert to is.
796  unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
797  const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
798  IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
799  FunctionDecl *NewBuiltinDecl =
800    cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
801                                           TUScope, false, DRE->getLocStart()));
802
803  // The first argument --- the pointer --- has a fixed type; we
804  // deduce the types of the rest of the arguments accordingly.  Walk
805  // the remaining arguments, converting them to the deduced value type.
806  for (unsigned i = 0; i != NumFixed; ++i) {
807    ExprResult Arg = TheCall->getArg(i+1);
808
809    // If the argument is an implicit cast, then there was a promotion due to
810    // "...", just remove it now.
811    if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) {
812      Arg = ICE->getSubExpr();
813      ICE->setSubExpr(0);
814      TheCall->setArg(i+1, Arg.get());
815    }
816
817    // GCC does an implicit conversion to the pointer or integer ValType.  This
818    // can fail in some cases (1i -> int**), check for this error case now.
819    // Initialize the argument.
820    InitializedEntity Entity = InitializedEntity::InitializeParameter(Context,
821                                                   ValType, /*consume*/ false);
822    Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
823    if (Arg.isInvalid())
824      return ExprError();
825
826    // Okay, we have something that *can* be converted to the right type.  Check
827    // to see if there is a potentially weird extension going on here.  This can
828    // happen when you do an atomic operation on something like an char* and
829    // pass in 42.  The 42 gets converted to char.  This is even more strange
830    // for things like 45.123 -> char, etc.
831    // FIXME: Do this check.
832    TheCall->setArg(i+1, Arg.take());
833  }
834
835  ASTContext& Context = this->getASTContext();
836
837  // Create a new DeclRefExpr to refer to the new decl.
838  DeclRefExpr* NewDRE = DeclRefExpr::Create(
839      Context,
840      DRE->getQualifierLoc(),
841      NewBuiltinDecl,
842      DRE->getLocation(),
843      NewBuiltinDecl->getType(),
844      DRE->getValueKind());
845
846  // Set the callee in the CallExpr.
847  // FIXME: This leaks the original parens and implicit casts.
848  ExprResult PromotedCall = UsualUnaryConversions(NewDRE);
849  if (PromotedCall.isInvalid())
850    return ExprError();
851  TheCall->setCallee(PromotedCall.take());
852
853  // Change the result type of the call to match the original value type. This
854  // is arbitrary, but the codegen for these builtins ins design to handle it
855  // gracefully.
856  TheCall->setType(ResultType);
857
858  return move(TheCallResult);
859}
860
861/// CheckObjCString - Checks that the argument to the builtin
862/// CFString constructor is correct
863/// Note: It might also make sense to do the UTF-16 conversion here (would
864/// simplify the backend).
865bool Sema::CheckObjCString(Expr *Arg) {
866  Arg = Arg->IgnoreParenCasts();
867  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
868
869  if (!Literal || !Literal->isAscii()) {
870    Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
871      << Arg->getSourceRange();
872    return true;
873  }
874
875  if (Literal->containsNonAsciiOrNull()) {
876    StringRef String = Literal->getString();
877    unsigned NumBytes = String.size();
878    SmallVector<UTF16, 128> ToBuf(NumBytes);
879    const UTF8 *FromPtr = (UTF8 *)String.data();
880    UTF16 *ToPtr = &ToBuf[0];
881
882    ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
883                                                 &ToPtr, ToPtr + NumBytes,
884                                                 strictConversion);
885    // Check for conversion failure.
886    if (Result != conversionOK)
887      Diag(Arg->getLocStart(),
888           diag::warn_cfstring_truncated) << Arg->getSourceRange();
889  }
890  return false;
891}
892
893/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
894/// Emit an error and return true on failure, return false on success.
895bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
896  Expr *Fn = TheCall->getCallee();
897  if (TheCall->getNumArgs() > 2) {
898    Diag(TheCall->getArg(2)->getLocStart(),
899         diag::err_typecheck_call_too_many_args)
900      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
901      << Fn->getSourceRange()
902      << SourceRange(TheCall->getArg(2)->getLocStart(),
903                     (*(TheCall->arg_end()-1))->getLocEnd());
904    return true;
905  }
906
907  if (TheCall->getNumArgs() < 2) {
908    return Diag(TheCall->getLocEnd(),
909      diag::err_typecheck_call_too_few_args_at_least)
910      << 0 /*function call*/ << 2 << TheCall->getNumArgs();
911  }
912
913  // Type-check the first argument normally.
914  if (checkBuiltinArgument(*this, TheCall, 0))
915    return true;
916
917  // Determine whether the current function is variadic or not.
918  BlockScopeInfo *CurBlock = getCurBlock();
919  bool isVariadic;
920  if (CurBlock)
921    isVariadic = CurBlock->TheDecl->isVariadic();
922  else if (FunctionDecl *FD = getCurFunctionDecl())
923    isVariadic = FD->isVariadic();
924  else
925    isVariadic = getCurMethodDecl()->isVariadic();
926
927  if (!isVariadic) {
928    Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
929    return true;
930  }
931
932  // Verify that the second argument to the builtin is the last argument of the
933  // current function or method.
934  bool SecondArgIsLastNamedArgument = false;
935  const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
936
937  if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
938    if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
939      // FIXME: This isn't correct for methods (results in bogus warning).
940      // Get the last formal in the current function.
941      const ParmVarDecl *LastArg;
942      if (CurBlock)
943        LastArg = *(CurBlock->TheDecl->param_end()-1);
944      else if (FunctionDecl *FD = getCurFunctionDecl())
945        LastArg = *(FD->param_end()-1);
946      else
947        LastArg = *(getCurMethodDecl()->param_end()-1);
948      SecondArgIsLastNamedArgument = PV == LastArg;
949    }
950  }
951
952  if (!SecondArgIsLastNamedArgument)
953    Diag(TheCall->getArg(1)->getLocStart(),
954         diag::warn_second_parameter_of_va_start_not_last_named_argument);
955  return false;
956}
957
958/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
959/// friends.  This is declared to take (...), so we have to check everything.
960bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
961  if (TheCall->getNumArgs() < 2)
962    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
963      << 0 << 2 << TheCall->getNumArgs()/*function call*/;
964  if (TheCall->getNumArgs() > 2)
965    return Diag(TheCall->getArg(2)->getLocStart(),
966                diag::err_typecheck_call_too_many_args)
967      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
968      << SourceRange(TheCall->getArg(2)->getLocStart(),
969                     (*(TheCall->arg_end()-1))->getLocEnd());
970
971  ExprResult OrigArg0 = TheCall->getArg(0);
972  ExprResult OrigArg1 = TheCall->getArg(1);
973
974  // Do standard promotions between the two arguments, returning their common
975  // type.
976  QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
977  if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
978    return true;
979
980  // Make sure any conversions are pushed back into the call; this is
981  // type safe since unordered compare builtins are declared as "_Bool
982  // foo(...)".
983  TheCall->setArg(0, OrigArg0.get());
984  TheCall->setArg(1, OrigArg1.get());
985
986  if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
987    return false;
988
989  // If the common type isn't a real floating type, then the arguments were
990  // invalid for this operation.
991  if (!Res->isRealFloatingType())
992    return Diag(OrigArg0.get()->getLocStart(),
993                diag::err_typecheck_call_invalid_ordered_compare)
994      << OrigArg0.get()->getType() << OrigArg1.get()->getType()
995      << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
996
997  return false;
998}
999
1000/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
1001/// __builtin_isnan and friends.  This is declared to take (...), so we have
1002/// to check everything. We expect the last argument to be a floating point
1003/// value.
1004bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
1005  if (TheCall->getNumArgs() < NumArgs)
1006    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1007      << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
1008  if (TheCall->getNumArgs() > NumArgs)
1009    return Diag(TheCall->getArg(NumArgs)->getLocStart(),
1010                diag::err_typecheck_call_too_many_args)
1011      << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
1012      << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
1013                     (*(TheCall->arg_end()-1))->getLocEnd());
1014
1015  Expr *OrigArg = TheCall->getArg(NumArgs-1);
1016
1017  if (OrigArg->isTypeDependent())
1018    return false;
1019
1020  // This operation requires a non-_Complex floating-point number.
1021  if (!OrigArg->getType()->isRealFloatingType())
1022    return Diag(OrigArg->getLocStart(),
1023                diag::err_typecheck_call_invalid_unary_fp)
1024      << OrigArg->getType() << OrigArg->getSourceRange();
1025
1026  // If this is an implicit conversion from float -> double, remove it.
1027  if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
1028    Expr *CastArg = Cast->getSubExpr();
1029    if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
1030      assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
1031             "promotion from float to double is the only expected cast here");
1032      Cast->setSubExpr(0);
1033      TheCall->setArg(NumArgs-1, CastArg);
1034      OrigArg = CastArg;
1035    }
1036  }
1037
1038  return false;
1039}
1040
1041/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
1042// This is declared to take (...), so we have to check everything.
1043ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
1044  if (TheCall->getNumArgs() < 2)
1045    return ExprError(Diag(TheCall->getLocEnd(),
1046                          diag::err_typecheck_call_too_few_args_at_least)
1047      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1048      << TheCall->getSourceRange());
1049
1050  // Determine which of the following types of shufflevector we're checking:
1051  // 1) unary, vector mask: (lhs, mask)
1052  // 2) binary, vector mask: (lhs, rhs, mask)
1053  // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
1054  QualType resType = TheCall->getArg(0)->getType();
1055  unsigned numElements = 0;
1056
1057  if (!TheCall->getArg(0)->isTypeDependent() &&
1058      !TheCall->getArg(1)->isTypeDependent()) {
1059    QualType LHSType = TheCall->getArg(0)->getType();
1060    QualType RHSType = TheCall->getArg(1)->getType();
1061
1062    if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
1063      Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
1064        << SourceRange(TheCall->getArg(0)->getLocStart(),
1065                       TheCall->getArg(1)->getLocEnd());
1066      return ExprError();
1067    }
1068
1069    numElements = LHSType->getAs<VectorType>()->getNumElements();
1070    unsigned numResElements = TheCall->getNumArgs() - 2;
1071
1072    // Check to see if we have a call with 2 vector arguments, the unary shuffle
1073    // with mask.  If so, verify that RHS is an integer vector type with the
1074    // same number of elts as lhs.
1075    if (TheCall->getNumArgs() == 2) {
1076      if (!RHSType->hasIntegerRepresentation() ||
1077          RHSType->getAs<VectorType>()->getNumElements() != numElements)
1078        Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1079          << SourceRange(TheCall->getArg(1)->getLocStart(),
1080                         TheCall->getArg(1)->getLocEnd());
1081      numResElements = numElements;
1082    }
1083    else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
1084      Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1085        << SourceRange(TheCall->getArg(0)->getLocStart(),
1086                       TheCall->getArg(1)->getLocEnd());
1087      return ExprError();
1088    } else if (numElements != numResElements) {
1089      QualType eltType = LHSType->getAs<VectorType>()->getElementType();
1090      resType = Context.getVectorType(eltType, numResElements,
1091                                      VectorType::GenericVector);
1092    }
1093  }
1094
1095  for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
1096    if (TheCall->getArg(i)->isTypeDependent() ||
1097        TheCall->getArg(i)->isValueDependent())
1098      continue;
1099
1100    llvm::APSInt Result(32);
1101    if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
1102      return ExprError(Diag(TheCall->getLocStart(),
1103                  diag::err_shufflevector_nonconstant_argument)
1104                << TheCall->getArg(i)->getSourceRange());
1105
1106    if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
1107      return ExprError(Diag(TheCall->getLocStart(),
1108                  diag::err_shufflevector_argument_too_large)
1109               << TheCall->getArg(i)->getSourceRange());
1110  }
1111
1112  SmallVector<Expr*, 32> exprs;
1113
1114  for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
1115    exprs.push_back(TheCall->getArg(i));
1116    TheCall->setArg(i, 0);
1117  }
1118
1119  return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
1120                                            exprs.size(), resType,
1121                                            TheCall->getCallee()->getLocStart(),
1122                                            TheCall->getRParenLoc()));
1123}
1124
1125/// SemaBuiltinPrefetch - Handle __builtin_prefetch.
1126// This is declared to take (const void*, ...) and can take two
1127// optional constant int args.
1128bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
1129  unsigned NumArgs = TheCall->getNumArgs();
1130
1131  if (NumArgs > 3)
1132    return Diag(TheCall->getLocEnd(),
1133             diag::err_typecheck_call_too_many_args_at_most)
1134             << 0 /*function call*/ << 3 << NumArgs
1135             << TheCall->getSourceRange();
1136
1137  // Argument 0 is checked for us and the remaining arguments must be
1138  // constant integers.
1139  for (unsigned i = 1; i != NumArgs; ++i) {
1140    Expr *Arg = TheCall->getArg(i);
1141
1142    llvm::APSInt Result;
1143    if (SemaBuiltinConstantArg(TheCall, i, Result))
1144      return true;
1145
1146    // FIXME: gcc issues a warning and rewrites these to 0. These
1147    // seems especially odd for the third argument since the default
1148    // is 3.
1149    if (i == 1) {
1150      if (Result.getLimitedValue() > 1)
1151        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1152             << "0" << "1" << Arg->getSourceRange();
1153    } else {
1154      if (Result.getLimitedValue() > 3)
1155        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1156            << "0" << "3" << Arg->getSourceRange();
1157    }
1158  }
1159
1160  return false;
1161}
1162
1163/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
1164/// TheCall is a constant expression.
1165bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
1166                                  llvm::APSInt &Result) {
1167  Expr *Arg = TheCall->getArg(ArgNum);
1168  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
1169  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
1170
1171  if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
1172
1173  if (!Arg->isIntegerConstantExpr(Result, Context))
1174    return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
1175                << FDecl->getDeclName() <<  Arg->getSourceRange();
1176
1177  return false;
1178}
1179
1180/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
1181/// int type). This simply type checks that type is one of the defined
1182/// constants (0-3).
1183// For compatibility check 0-3, llvm only handles 0 and 2.
1184bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
1185  llvm::APSInt Result;
1186
1187  // Check constant-ness first.
1188  if (SemaBuiltinConstantArg(TheCall, 1, Result))
1189    return true;
1190
1191  Expr *Arg = TheCall->getArg(1);
1192  if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
1193    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1194             << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1195  }
1196
1197  return false;
1198}
1199
1200/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
1201/// This checks that val is a constant 1.
1202bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
1203  Expr *Arg = TheCall->getArg(1);
1204  llvm::APSInt Result;
1205
1206  // TODO: This is less than ideal. Overload this to take a value.
1207  if (SemaBuiltinConstantArg(TheCall, 1, Result))
1208    return true;
1209
1210  if (Result != 1)
1211    return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
1212             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1213
1214  return false;
1215}
1216
1217// Handle i > 1 ? "x" : "y", recursively.
1218bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall,
1219                                  bool HasVAListArg,
1220                                  unsigned format_idx, unsigned firstDataArg,
1221                                  bool isPrintf) {
1222 tryAgain:
1223  if (E->isTypeDependent() || E->isValueDependent())
1224    return false;
1225
1226  E = E->IgnoreParens();
1227
1228  switch (E->getStmtClass()) {
1229  case Stmt::BinaryConditionalOperatorClass:
1230  case Stmt::ConditionalOperatorClass: {
1231    const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E);
1232    return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg,
1233                                  format_idx, firstDataArg, isPrintf)
1234        && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg,
1235                                  format_idx, firstDataArg, isPrintf);
1236  }
1237
1238  case Stmt::IntegerLiteralClass:
1239    // Technically -Wformat-nonliteral does not warn about this case.
1240    // The behavior of printf and friends in this case is implementation
1241    // dependent.  Ideally if the format string cannot be null then
1242    // it should have a 'nonnull' attribute in the function prototype.
1243    return true;
1244
1245  case Stmt::ImplicitCastExprClass: {
1246    E = cast<ImplicitCastExpr>(E)->getSubExpr();
1247    goto tryAgain;
1248  }
1249
1250  case Stmt::OpaqueValueExprClass:
1251    if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
1252      E = src;
1253      goto tryAgain;
1254    }
1255    return false;
1256
1257  case Stmt::PredefinedExprClass:
1258    // While __func__, etc., are technically not string literals, they
1259    // cannot contain format specifiers and thus are not a security
1260    // liability.
1261    return true;
1262
1263  case Stmt::DeclRefExprClass: {
1264    const DeclRefExpr *DR = cast<DeclRefExpr>(E);
1265
1266    // As an exception, do not flag errors for variables binding to
1267    // const string literals.
1268    if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
1269      bool isConstant = false;
1270      QualType T = DR->getType();
1271
1272      if (const ArrayType *AT = Context.getAsArrayType(T)) {
1273        isConstant = AT->getElementType().isConstant(Context);
1274      } else if (const PointerType *PT = T->getAs<PointerType>()) {
1275        isConstant = T.isConstant(Context) &&
1276                     PT->getPointeeType().isConstant(Context);
1277      }
1278
1279      if (isConstant) {
1280        if (const Expr *Init = VD->getAnyInitializer())
1281          return SemaCheckStringLiteral(Init, TheCall,
1282                                        HasVAListArg, format_idx, firstDataArg,
1283                                        isPrintf);
1284      }
1285
1286      // For vprintf* functions (i.e., HasVAListArg==true), we add a
1287      // special check to see if the format string is a function parameter
1288      // of the function calling the printf function.  If the function
1289      // has an attribute indicating it is a printf-like function, then we
1290      // should suppress warnings concerning non-literals being used in a call
1291      // to a vprintf function.  For example:
1292      //
1293      // void
1294      // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
1295      //      va_list ap;
1296      //      va_start(ap, fmt);
1297      //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
1298      //      ...
1299      //
1300      //
1301      //  FIXME: We don't have full attribute support yet, so just check to see
1302      //    if the argument is a DeclRefExpr that references a parameter.  We'll
1303      //    add proper support for checking the attribute later.
1304      if (HasVAListArg)
1305        if (isa<ParmVarDecl>(VD))
1306          return true;
1307    }
1308
1309    return false;
1310  }
1311
1312  case Stmt::CallExprClass: {
1313    const CallExpr *CE = cast<CallExpr>(E);
1314    if (const ImplicitCastExpr *ICE
1315          = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
1316      if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
1317        if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
1318          if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
1319            unsigned ArgIndex = FA->getFormatIdx();
1320            const Expr *Arg = CE->getArg(ArgIndex - 1);
1321
1322            return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg,
1323                                          format_idx, firstDataArg, isPrintf);
1324          }
1325        }
1326      }
1327    }
1328
1329    return false;
1330  }
1331  case Stmt::ObjCStringLiteralClass:
1332  case Stmt::StringLiteralClass: {
1333    const StringLiteral *StrE = NULL;
1334
1335    if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1336      StrE = ObjCFExpr->getString();
1337    else
1338      StrE = cast<StringLiteral>(E);
1339
1340    if (StrE) {
1341      CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx,
1342                        firstDataArg, isPrintf);
1343      return true;
1344    }
1345
1346    return false;
1347  }
1348
1349  default:
1350    return false;
1351  }
1352}
1353
1354void
1355Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1356                            const Expr * const *ExprArgs,
1357                            SourceLocation CallSiteLoc) {
1358  for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1359                                  e = NonNull->args_end();
1360       i != e; ++i) {
1361    const Expr *ArgExpr = ExprArgs[*i];
1362    if (ArgExpr->isNullPointerConstant(Context,
1363                                       Expr::NPC_ValueDependentIsNotNull))
1364      Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1365  }
1366}
1367
1368static void DiagnoseCalleeStaticArrayParam(Sema &S, ParmVarDecl *PVD) {
1369  TypeLoc TL = PVD->getTypeSourceInfo()->getTypeLoc();
1370  if (ArrayTypeLoc *ATL = dyn_cast<ArrayTypeLoc>(&TL))
1371    S.Diag(PVD->getLocation(), diag::note_callee_static_array)
1372      << ATL->getLocalSourceRange();
1373}
1374
1375/// CheckStaticArrayArguments - Check that each argument corresponding to a
1376/// static array parameter is non-null, and that if it is formed by
1377/// array-to-pointer decay, the underlying array is sufficiently large.
1378///
1379/// C99 6.7.5.3p7: If the keyword static also appears within the [ and ] of the
1380/// array type derivation, then for each call to the function, the value of the
1381/// corresponding actual argument shall provide access to the first element of
1382/// an array with at least as many elements as specified by the size expression.
1383void
1384Sema::CheckStaticArrayArguments(const FunctionDecl *FDecl,
1385                                const Expr * const *ExprArgs,
1386                                SourceLocation CallSiteLoc) {
1387  // Static array parameters are not supported in C++.
1388  if (getLangOptions().CPlusPlus)
1389    return;
1390
1391  for (FunctionDecl::param_const_iterator i = FDecl->param_begin(),
1392       e = FDecl->param_end(); i != e; ++i, ++ExprArgs) {
1393    const Expr *ArgExpr = *ExprArgs;
1394    QualType OrigTy = (*i)->getOriginalType();
1395
1396    const ArrayType *AT = Context.getAsArrayType(OrigTy);
1397    if (!AT || AT->getSizeModifier() != ArrayType::Static)
1398      continue;
1399
1400    if (ArgExpr->isNullPointerConstant(Context,
1401                                       Expr::NPC_NeverValueDependent)) {
1402      Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1403      DiagnoseCalleeStaticArrayParam(*this, *i);
1404      continue;
1405    }
1406
1407    const ConstantArrayType *CAT = dyn_cast<ConstantArrayType>(AT);
1408    if (!CAT)
1409      continue;
1410
1411    const ConstantArrayType *ArgCAT =
1412      Context.getAsConstantArrayType(ArgExpr->IgnoreParenImpCasts()->getType());
1413    if (!ArgCAT)
1414      continue;
1415
1416    if (ArgCAT->getSize().ult(CAT->getSize())) {
1417      Diag(CallSiteLoc, diag::warn_static_array_too_small)
1418        << ArgExpr->getSourceRange()
1419        << (unsigned) ArgCAT->getSize().getZExtValue()
1420        << (unsigned) CAT->getSize().getZExtValue();
1421      DiagnoseCalleeStaticArrayParam(*this, *i);
1422    }
1423  }
1424}
1425
1426/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar
1427/// functions) for correct use of format strings.
1428void
1429Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg,
1430                                unsigned format_idx, unsigned firstDataArg,
1431                                bool isPrintf) {
1432
1433  const Expr *Fn = TheCall->getCallee();
1434
1435  // The way the format attribute works in GCC, the implicit this argument
1436  // of member functions is counted. However, it doesn't appear in our own
1437  // lists, so decrement format_idx in that case.
1438  if (isa<CXXMemberCallExpr>(TheCall)) {
1439    const CXXMethodDecl *method_decl =
1440      dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl());
1441    if (method_decl && method_decl->isInstance()) {
1442      // Catch a format attribute mistakenly referring to the object argument.
1443      if (format_idx == 0)
1444        return;
1445      --format_idx;
1446      if(firstDataArg != 0)
1447        --firstDataArg;
1448    }
1449  }
1450
1451  // CHECK: printf/scanf-like function is called with no format string.
1452  if (format_idx >= TheCall->getNumArgs()) {
1453    Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string)
1454      << Fn->getSourceRange();
1455    return;
1456  }
1457
1458  const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts();
1459
1460  // CHECK: format string is not a string literal.
1461  //
1462  // Dynamically generated format strings are difficult to
1463  // automatically vet at compile time.  Requiring that format strings
1464  // are string literals: (1) permits the checking of format strings by
1465  // the compiler and thereby (2) can practically remove the source of
1466  // many format string exploits.
1467
1468  // Format string can be either ObjC string (e.g. @"%d") or
1469  // C string (e.g. "%d")
1470  // ObjC string uses the same format specifiers as C string, so we can use
1471  // the same format string checking logic for both ObjC and C strings.
1472  if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx,
1473                             firstDataArg, isPrintf))
1474    return;  // Literal format string found, check done!
1475
1476  // If there are no arguments specified, warn with -Wformat-security, otherwise
1477  // warn only with -Wformat-nonliteral.
1478  if (TheCall->getNumArgs() == format_idx+1)
1479    Diag(TheCall->getArg(format_idx)->getLocStart(),
1480         diag::warn_format_nonliteral_noargs)
1481      << OrigFormatExpr->getSourceRange();
1482  else
1483    Diag(TheCall->getArg(format_idx)->getLocStart(),
1484         diag::warn_format_nonliteral)
1485           << OrigFormatExpr->getSourceRange();
1486}
1487
1488namespace {
1489class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1490protected:
1491  Sema &S;
1492  const StringLiteral *FExpr;
1493  const Expr *OrigFormatExpr;
1494  const unsigned FirstDataArg;
1495  const unsigned NumDataArgs;
1496  const bool IsObjCLiteral;
1497  const char *Beg; // Start of format string.
1498  const bool HasVAListArg;
1499  const CallExpr *TheCall;
1500  unsigned FormatIdx;
1501  llvm::BitVector CoveredArgs;
1502  bool usesPositionalArgs;
1503  bool atFirstArg;
1504public:
1505  CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1506                     const Expr *origFormatExpr, unsigned firstDataArg,
1507                     unsigned numDataArgs, bool isObjCLiteral,
1508                     const char *beg, bool hasVAListArg,
1509                     const CallExpr *theCall, unsigned formatIdx)
1510    : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1511      FirstDataArg(firstDataArg),
1512      NumDataArgs(numDataArgs),
1513      IsObjCLiteral(isObjCLiteral), Beg(beg),
1514      HasVAListArg(hasVAListArg),
1515      TheCall(theCall), FormatIdx(formatIdx),
1516      usesPositionalArgs(false), atFirstArg(true) {
1517        CoveredArgs.resize(numDataArgs);
1518        CoveredArgs.reset();
1519      }
1520
1521  void DoneProcessing();
1522
1523  void HandleIncompleteSpecifier(const char *startSpecifier,
1524                                 unsigned specifierLen);
1525
1526  virtual void HandleInvalidPosition(const char *startSpecifier,
1527                                     unsigned specifierLen,
1528                                     analyze_format_string::PositionContext p);
1529
1530  virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
1531
1532  void HandleNullChar(const char *nullCharacter);
1533
1534protected:
1535  bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
1536                                        const char *startSpec,
1537                                        unsigned specifierLen,
1538                                        const char *csStart, unsigned csLen);
1539
1540  SourceRange getFormatStringRange();
1541  CharSourceRange getSpecifierRange(const char *startSpecifier,
1542                                    unsigned specifierLen);
1543  SourceLocation getLocationOfByte(const char *x);
1544
1545  const Expr *getDataArg(unsigned i) const;
1546
1547  bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
1548                    const analyze_format_string::ConversionSpecifier &CS,
1549                    const char *startSpecifier, unsigned specifierLen,
1550                    unsigned argIndex);
1551};
1552}
1553
1554SourceRange CheckFormatHandler::getFormatStringRange() {
1555  return OrigFormatExpr->getSourceRange();
1556}
1557
1558CharSourceRange CheckFormatHandler::
1559getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
1560  SourceLocation Start = getLocationOfByte(startSpecifier);
1561  SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
1562
1563  // Advance the end SourceLocation by one due to half-open ranges.
1564  End = End.getLocWithOffset(1);
1565
1566  return CharSourceRange::getCharRange(Start, End);
1567}
1568
1569SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
1570  return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
1571}
1572
1573void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
1574                                                   unsigned specifierLen){
1575  SourceLocation Loc = getLocationOfByte(startSpecifier);
1576  S.Diag(Loc, diag::warn_printf_incomplete_specifier)
1577    << getSpecifierRange(startSpecifier, specifierLen);
1578}
1579
1580void
1581CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
1582                                     analyze_format_string::PositionContext p) {
1583  SourceLocation Loc = getLocationOfByte(startPos);
1584  S.Diag(Loc, diag::warn_format_invalid_positional_specifier)
1585    << (unsigned) p << getSpecifierRange(startPos, posLen);
1586}
1587
1588void CheckFormatHandler::HandleZeroPosition(const char *startPos,
1589                                            unsigned posLen) {
1590  SourceLocation Loc = getLocationOfByte(startPos);
1591  S.Diag(Loc, diag::warn_format_zero_positional_specifier)
1592    << getSpecifierRange(startPos, posLen);
1593}
1594
1595void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
1596  if (!IsObjCLiteral) {
1597    // The presence of a null character is likely an error.
1598    S.Diag(getLocationOfByte(nullCharacter),
1599           diag::warn_printf_format_string_contains_null_char)
1600      << getFormatStringRange();
1601  }
1602}
1603
1604const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
1605  return TheCall->getArg(FirstDataArg + i);
1606}
1607
1608void CheckFormatHandler::DoneProcessing() {
1609    // Does the number of data arguments exceed the number of
1610    // format conversions in the format string?
1611  if (!HasVAListArg) {
1612      // Find any arguments that weren't covered.
1613    CoveredArgs.flip();
1614    signed notCoveredArg = CoveredArgs.find_first();
1615    if (notCoveredArg >= 0) {
1616      assert((unsigned)notCoveredArg < NumDataArgs);
1617      S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(),
1618             diag::warn_printf_data_arg_not_used)
1619      << getFormatStringRange();
1620    }
1621  }
1622}
1623
1624bool
1625CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
1626                                                     SourceLocation Loc,
1627                                                     const char *startSpec,
1628                                                     unsigned specifierLen,
1629                                                     const char *csStart,
1630                                                     unsigned csLen) {
1631
1632  bool keepGoing = true;
1633  if (argIndex < NumDataArgs) {
1634    // Consider the argument coverered, even though the specifier doesn't
1635    // make sense.
1636    CoveredArgs.set(argIndex);
1637  }
1638  else {
1639    // If argIndex exceeds the number of data arguments we
1640    // don't issue a warning because that is just a cascade of warnings (and
1641    // they may have intended '%%' anyway). We don't want to continue processing
1642    // the format string after this point, however, as we will like just get
1643    // gibberish when trying to match arguments.
1644    keepGoing = false;
1645  }
1646
1647  S.Diag(Loc, diag::warn_format_invalid_conversion)
1648    << StringRef(csStart, csLen)
1649    << getSpecifierRange(startSpec, specifierLen);
1650
1651  return keepGoing;
1652}
1653
1654bool
1655CheckFormatHandler::CheckNumArgs(
1656  const analyze_format_string::FormatSpecifier &FS,
1657  const analyze_format_string::ConversionSpecifier &CS,
1658  const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
1659
1660  if (argIndex >= NumDataArgs) {
1661    if (FS.usesPositionalArg())  {
1662      S.Diag(getLocationOfByte(CS.getStart()),
1663             diag::warn_printf_positional_arg_exceeds_data_args)
1664      << (argIndex+1) << NumDataArgs
1665      << getSpecifierRange(startSpecifier, specifierLen);
1666    }
1667    else {
1668      S.Diag(getLocationOfByte(CS.getStart()),
1669             diag::warn_printf_insufficient_data_args)
1670      << getSpecifierRange(startSpecifier, specifierLen);
1671    }
1672
1673    return false;
1674  }
1675  return true;
1676}
1677
1678//===--- CHECK: Printf format string checking ------------------------------===//
1679
1680namespace {
1681class CheckPrintfHandler : public CheckFormatHandler {
1682public:
1683  CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
1684                     const Expr *origFormatExpr, unsigned firstDataArg,
1685                     unsigned numDataArgs, bool isObjCLiteral,
1686                     const char *beg, bool hasVAListArg,
1687                     const CallExpr *theCall, unsigned formatIdx)
1688  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1689                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1690                       theCall, formatIdx) {}
1691
1692
1693  bool HandleInvalidPrintfConversionSpecifier(
1694                                      const analyze_printf::PrintfSpecifier &FS,
1695                                      const char *startSpecifier,
1696                                      unsigned specifierLen);
1697
1698  bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
1699                             const char *startSpecifier,
1700                             unsigned specifierLen);
1701
1702  bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
1703                    const char *startSpecifier, unsigned specifierLen);
1704  void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
1705                           const analyze_printf::OptionalAmount &Amt,
1706                           unsigned type,
1707                           const char *startSpecifier, unsigned specifierLen);
1708  void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1709                  const analyze_printf::OptionalFlag &flag,
1710                  const char *startSpecifier, unsigned specifierLen);
1711  void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
1712                         const analyze_printf::OptionalFlag &ignoredFlag,
1713                         const analyze_printf::OptionalFlag &flag,
1714                         const char *startSpecifier, unsigned specifierLen);
1715};
1716}
1717
1718bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
1719                                      const analyze_printf::PrintfSpecifier &FS,
1720                                      const char *startSpecifier,
1721                                      unsigned specifierLen) {
1722  const analyze_printf::PrintfConversionSpecifier &CS =
1723    FS.getConversionSpecifier();
1724
1725  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1726                                          getLocationOfByte(CS.getStart()),
1727                                          startSpecifier, specifierLen,
1728                                          CS.getStart(), CS.getLength());
1729}
1730
1731bool CheckPrintfHandler::HandleAmount(
1732                               const analyze_format_string::OptionalAmount &Amt,
1733                               unsigned k, const char *startSpecifier,
1734                               unsigned specifierLen) {
1735
1736  if (Amt.hasDataArgument()) {
1737    if (!HasVAListArg) {
1738      unsigned argIndex = Amt.getArgIndex();
1739      if (argIndex >= NumDataArgs) {
1740        S.Diag(getLocationOfByte(Amt.getStart()),
1741               diag::warn_printf_asterisk_missing_arg)
1742          << k << getSpecifierRange(startSpecifier, specifierLen);
1743        // Don't do any more checking.  We will just emit
1744        // spurious errors.
1745        return false;
1746      }
1747
1748      // Type check the data argument.  It should be an 'int'.
1749      // Although not in conformance with C99, we also allow the argument to be
1750      // an 'unsigned int' as that is a reasonably safe case.  GCC also
1751      // doesn't emit a warning for that case.
1752      CoveredArgs.set(argIndex);
1753      const Expr *Arg = getDataArg(argIndex);
1754      QualType T = Arg->getType();
1755
1756      const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context);
1757      assert(ATR.isValid());
1758
1759      if (!ATR.matchesType(S.Context, T)) {
1760        S.Diag(getLocationOfByte(Amt.getStart()),
1761               diag::warn_printf_asterisk_wrong_type)
1762          << k
1763          << ATR.getRepresentativeType(S.Context) << T
1764          << getSpecifierRange(startSpecifier, specifierLen)
1765          << Arg->getSourceRange();
1766        // Don't do any more checking.  We will just emit
1767        // spurious errors.
1768        return false;
1769      }
1770    }
1771  }
1772  return true;
1773}
1774
1775void CheckPrintfHandler::HandleInvalidAmount(
1776                                      const analyze_printf::PrintfSpecifier &FS,
1777                                      const analyze_printf::OptionalAmount &Amt,
1778                                      unsigned type,
1779                                      const char *startSpecifier,
1780                                      unsigned specifierLen) {
1781  const analyze_printf::PrintfConversionSpecifier &CS =
1782    FS.getConversionSpecifier();
1783  switch (Amt.getHowSpecified()) {
1784  case analyze_printf::OptionalAmount::Constant:
1785    S.Diag(getLocationOfByte(Amt.getStart()),
1786        diag::warn_printf_nonsensical_optional_amount)
1787      << type
1788      << CS.toString()
1789      << getSpecifierRange(startSpecifier, specifierLen)
1790      << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
1791          Amt.getConstantLength()));
1792    break;
1793
1794  default:
1795    S.Diag(getLocationOfByte(Amt.getStart()),
1796        diag::warn_printf_nonsensical_optional_amount)
1797      << type
1798      << CS.toString()
1799      << getSpecifierRange(startSpecifier, specifierLen);
1800    break;
1801  }
1802}
1803
1804void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1805                                    const analyze_printf::OptionalFlag &flag,
1806                                    const char *startSpecifier,
1807                                    unsigned specifierLen) {
1808  // Warn about pointless flag with a fixit removal.
1809  const analyze_printf::PrintfConversionSpecifier &CS =
1810    FS.getConversionSpecifier();
1811  S.Diag(getLocationOfByte(flag.getPosition()),
1812      diag::warn_printf_nonsensical_flag)
1813    << flag.toString() << CS.toString()
1814    << getSpecifierRange(startSpecifier, specifierLen)
1815    << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1));
1816}
1817
1818void CheckPrintfHandler::HandleIgnoredFlag(
1819                                const analyze_printf::PrintfSpecifier &FS,
1820                                const analyze_printf::OptionalFlag &ignoredFlag,
1821                                const analyze_printf::OptionalFlag &flag,
1822                                const char *startSpecifier,
1823                                unsigned specifierLen) {
1824  // Warn about ignored flag with a fixit removal.
1825  S.Diag(getLocationOfByte(ignoredFlag.getPosition()),
1826      diag::warn_printf_ignored_flag)
1827    << ignoredFlag.toString() << flag.toString()
1828    << getSpecifierRange(startSpecifier, specifierLen)
1829    << FixItHint::CreateRemoval(getSpecifierRange(
1830        ignoredFlag.getPosition(), 1));
1831}
1832
1833bool
1834CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
1835                                            &FS,
1836                                          const char *startSpecifier,
1837                                          unsigned specifierLen) {
1838
1839  using namespace analyze_format_string;
1840  using namespace analyze_printf;
1841  const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
1842
1843  if (FS.consumesDataArgument()) {
1844    if (atFirstArg) {
1845        atFirstArg = false;
1846        usesPositionalArgs = FS.usesPositionalArg();
1847    }
1848    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1849      // Cannot mix-and-match positional and non-positional arguments.
1850      S.Diag(getLocationOfByte(CS.getStart()),
1851             diag::warn_format_mix_positional_nonpositional_args)
1852        << getSpecifierRange(startSpecifier, specifierLen);
1853      return false;
1854    }
1855  }
1856
1857  // First check if the field width, precision, and conversion specifier
1858  // have matching data arguments.
1859  if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
1860                    startSpecifier, specifierLen)) {
1861    return false;
1862  }
1863
1864  if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
1865                    startSpecifier, specifierLen)) {
1866    return false;
1867  }
1868
1869  if (!CS.consumesDataArgument()) {
1870    // FIXME: Technically specifying a precision or field width here
1871    // makes no sense.  Worth issuing a warning at some point.
1872    return true;
1873  }
1874
1875  // Consume the argument.
1876  unsigned argIndex = FS.getArgIndex();
1877  if (argIndex < NumDataArgs) {
1878    // The check to see if the argIndex is valid will come later.
1879    // We set the bit here because we may exit early from this
1880    // function if we encounter some other error.
1881    CoveredArgs.set(argIndex);
1882  }
1883
1884  // Check for using an Objective-C specific conversion specifier
1885  // in a non-ObjC literal.
1886  if (!IsObjCLiteral && CS.isObjCArg()) {
1887    return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
1888                                                  specifierLen);
1889  }
1890
1891  // Check for invalid use of field width
1892  if (!FS.hasValidFieldWidth()) {
1893    HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
1894        startSpecifier, specifierLen);
1895  }
1896
1897  // Check for invalid use of precision
1898  if (!FS.hasValidPrecision()) {
1899    HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
1900        startSpecifier, specifierLen);
1901  }
1902
1903  // Check each flag does not conflict with any other component.
1904  if (!FS.hasValidThousandsGroupingPrefix())
1905    HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
1906  if (!FS.hasValidLeadingZeros())
1907    HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
1908  if (!FS.hasValidPlusPrefix())
1909    HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
1910  if (!FS.hasValidSpacePrefix())
1911    HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
1912  if (!FS.hasValidAlternativeForm())
1913    HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
1914  if (!FS.hasValidLeftJustified())
1915    HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
1916
1917  // Check that flags are not ignored by another flag
1918  if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
1919    HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
1920        startSpecifier, specifierLen);
1921  if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
1922    HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
1923            startSpecifier, specifierLen);
1924
1925  // Check the length modifier is valid with the given conversion specifier.
1926  const LengthModifier &LM = FS.getLengthModifier();
1927  if (!FS.hasValidLengthModifier())
1928    S.Diag(getLocationOfByte(LM.getStart()),
1929        diag::warn_format_nonsensical_length)
1930      << LM.toString() << CS.toString()
1931      << getSpecifierRange(startSpecifier, specifierLen)
1932      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1933          LM.getLength()));
1934
1935  // Are we using '%n'?
1936  if (CS.getKind() == ConversionSpecifier::nArg) {
1937    // Issue a warning about this being a possible security issue.
1938    S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back)
1939      << getSpecifierRange(startSpecifier, specifierLen);
1940    // Continue checking the other format specifiers.
1941    return true;
1942  }
1943
1944  // The remaining checks depend on the data arguments.
1945  if (HasVAListArg)
1946    return true;
1947
1948  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1949    return false;
1950
1951  // Now type check the data expression that matches the
1952  // format specifier.
1953  const Expr *Ex = getDataArg(argIndex);
1954  const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context);
1955  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
1956    // Check if we didn't match because of an implicit cast from a 'char'
1957    // or 'short' to an 'int'.  This is done because printf is a varargs
1958    // function.
1959    if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex))
1960      if (ICE->getType() == S.Context.IntTy) {
1961        // All further checking is done on the subexpression.
1962        Ex = ICE->getSubExpr();
1963        if (ATR.matchesType(S.Context, Ex->getType()))
1964          return true;
1965      }
1966
1967    // We may be able to offer a FixItHint if it is a supported type.
1968    PrintfSpecifier fixedFS = FS;
1969    bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions());
1970
1971    if (success) {
1972      // Get the fix string from the fixed format specifier
1973      llvm::SmallString<128> buf;
1974      llvm::raw_svector_ostream os(buf);
1975      fixedFS.toString(os);
1976
1977      // FIXME: getRepresentativeType() perhaps should return a string
1978      // instead of a QualType to better handle when the representative
1979      // type is 'wint_t' (which is defined in the system headers).
1980      S.Diag(getLocationOfByte(CS.getStart()),
1981          diag::warn_printf_conversion_argument_type_mismatch)
1982        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1983        << getSpecifierRange(startSpecifier, specifierLen)
1984        << Ex->getSourceRange()
1985        << FixItHint::CreateReplacement(
1986            getSpecifierRange(startSpecifier, specifierLen),
1987            os.str());
1988    }
1989    else {
1990      S.Diag(getLocationOfByte(CS.getStart()),
1991             diag::warn_printf_conversion_argument_type_mismatch)
1992        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1993        << getSpecifierRange(startSpecifier, specifierLen)
1994        << Ex->getSourceRange();
1995    }
1996  }
1997
1998  return true;
1999}
2000
2001//===--- CHECK: Scanf format string checking ------------------------------===//
2002
2003namespace {
2004class CheckScanfHandler : public CheckFormatHandler {
2005public:
2006  CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
2007                    const Expr *origFormatExpr, unsigned firstDataArg,
2008                    unsigned numDataArgs, bool isObjCLiteral,
2009                    const char *beg, bool hasVAListArg,
2010                    const CallExpr *theCall, unsigned formatIdx)
2011  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
2012                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
2013                       theCall, formatIdx) {}
2014
2015  bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
2016                            const char *startSpecifier,
2017                            unsigned specifierLen);
2018
2019  bool HandleInvalidScanfConversionSpecifier(
2020          const analyze_scanf::ScanfSpecifier &FS,
2021          const char *startSpecifier,
2022          unsigned specifierLen);
2023
2024  void HandleIncompleteScanList(const char *start, const char *end);
2025};
2026}
2027
2028void CheckScanfHandler::HandleIncompleteScanList(const char *start,
2029                                                 const char *end) {
2030  S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete)
2031    << getSpecifierRange(start, end - start);
2032}
2033
2034bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
2035                                        const analyze_scanf::ScanfSpecifier &FS,
2036                                        const char *startSpecifier,
2037                                        unsigned specifierLen) {
2038
2039  const analyze_scanf::ScanfConversionSpecifier &CS =
2040    FS.getConversionSpecifier();
2041
2042  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
2043                                          getLocationOfByte(CS.getStart()),
2044                                          startSpecifier, specifierLen,
2045                                          CS.getStart(), CS.getLength());
2046}
2047
2048bool CheckScanfHandler::HandleScanfSpecifier(
2049                                       const analyze_scanf::ScanfSpecifier &FS,
2050                                       const char *startSpecifier,
2051                                       unsigned specifierLen) {
2052
2053  using namespace analyze_scanf;
2054  using namespace analyze_format_string;
2055
2056  const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
2057
2058  // Handle case where '%' and '*' don't consume an argument.  These shouldn't
2059  // be used to decide if we are using positional arguments consistently.
2060  if (FS.consumesDataArgument()) {
2061    if (atFirstArg) {
2062      atFirstArg = false;
2063      usesPositionalArgs = FS.usesPositionalArg();
2064    }
2065    else if (usesPositionalArgs != FS.usesPositionalArg()) {
2066      // Cannot mix-and-match positional and non-positional arguments.
2067      S.Diag(getLocationOfByte(CS.getStart()),
2068             diag::warn_format_mix_positional_nonpositional_args)
2069        << getSpecifierRange(startSpecifier, specifierLen);
2070      return false;
2071    }
2072  }
2073
2074  // Check if the field with is non-zero.
2075  const OptionalAmount &Amt = FS.getFieldWidth();
2076  if (Amt.getHowSpecified() == OptionalAmount::Constant) {
2077    if (Amt.getConstantAmount() == 0) {
2078      const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
2079                                                   Amt.getConstantLength());
2080      S.Diag(getLocationOfByte(Amt.getStart()),
2081             diag::warn_scanf_nonzero_width)
2082        << R << FixItHint::CreateRemoval(R);
2083    }
2084  }
2085
2086  if (!FS.consumesDataArgument()) {
2087    // FIXME: Technically specifying a precision or field width here
2088    // makes no sense.  Worth issuing a warning at some point.
2089    return true;
2090  }
2091
2092  // Consume the argument.
2093  unsigned argIndex = FS.getArgIndex();
2094  if (argIndex < NumDataArgs) {
2095      // The check to see if the argIndex is valid will come later.
2096      // We set the bit here because we may exit early from this
2097      // function if we encounter some other error.
2098    CoveredArgs.set(argIndex);
2099  }
2100
2101  // Check the length modifier is valid with the given conversion specifier.
2102  const LengthModifier &LM = FS.getLengthModifier();
2103  if (!FS.hasValidLengthModifier()) {
2104    S.Diag(getLocationOfByte(LM.getStart()),
2105           diag::warn_format_nonsensical_length)
2106      << LM.toString() << CS.toString()
2107      << getSpecifierRange(startSpecifier, specifierLen)
2108      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
2109                                                    LM.getLength()));
2110  }
2111
2112  // The remaining checks depend on the data arguments.
2113  if (HasVAListArg)
2114    return true;
2115
2116  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
2117    return false;
2118
2119  // FIXME: Check that the argument type matches the format specifier.
2120
2121  return true;
2122}
2123
2124void Sema::CheckFormatString(const StringLiteral *FExpr,
2125                             const Expr *OrigFormatExpr,
2126                             const CallExpr *TheCall, bool HasVAListArg,
2127                             unsigned format_idx, unsigned firstDataArg,
2128                             bool isPrintf) {
2129
2130  // CHECK: is the format string a wide literal?
2131  if (!FExpr->isAscii()) {
2132    Diag(FExpr->getLocStart(),
2133         diag::warn_format_string_is_wide_literal)
2134    << OrigFormatExpr->getSourceRange();
2135    return;
2136  }
2137
2138  // Str - The format string.  NOTE: this is NOT null-terminated!
2139  StringRef StrRef = FExpr->getString();
2140  const char *Str = StrRef.data();
2141  unsigned StrLen = StrRef.size();
2142  const unsigned numDataArgs = TheCall->getNumArgs() - firstDataArg;
2143
2144  // CHECK: empty format string?
2145  if (StrLen == 0 && numDataArgs > 0) {
2146    Diag(FExpr->getLocStart(), diag::warn_empty_format_string)
2147    << OrigFormatExpr->getSourceRange();
2148    return;
2149  }
2150
2151  if (isPrintf) {
2152    CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
2153                         numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr),
2154                         Str, HasVAListArg, TheCall, format_idx);
2155
2156    if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen))
2157      H.DoneProcessing();
2158  }
2159  else {
2160    CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
2161                        numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr),
2162                        Str, HasVAListArg, TheCall, format_idx);
2163
2164    if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen))
2165      H.DoneProcessing();
2166  }
2167}
2168
2169//===--- CHECK: Standard memory functions ---------------------------------===//
2170
2171/// \brief Determine whether the given type is a dynamic class type (e.g.,
2172/// whether it has a vtable).
2173static bool isDynamicClassType(QualType T) {
2174  if (CXXRecordDecl *Record = T->getAsCXXRecordDecl())
2175    if (CXXRecordDecl *Definition = Record->getDefinition())
2176      if (Definition->isDynamicClass())
2177        return true;
2178
2179  return false;
2180}
2181
2182/// \brief If E is a sizeof expression, returns its argument expression,
2183/// otherwise returns NULL.
2184static const Expr *getSizeOfExprArg(const Expr* E) {
2185  if (const UnaryExprOrTypeTraitExpr *SizeOf =
2186      dyn_cast<UnaryExprOrTypeTraitExpr>(E))
2187    if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType())
2188      return SizeOf->getArgumentExpr()->IgnoreParenImpCasts();
2189
2190  return 0;
2191}
2192
2193/// \brief If E is a sizeof expression, returns its argument type.
2194static QualType getSizeOfArgType(const Expr* E) {
2195  if (const UnaryExprOrTypeTraitExpr *SizeOf =
2196      dyn_cast<UnaryExprOrTypeTraitExpr>(E))
2197    if (SizeOf->getKind() == clang::UETT_SizeOf)
2198      return SizeOf->getTypeOfArgument();
2199
2200  return QualType();
2201}
2202
2203/// \brief Check for dangerous or invalid arguments to memset().
2204///
2205/// This issues warnings on known problematic, dangerous or unspecified
2206/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp'
2207/// function calls.
2208///
2209/// \param Call The call expression to diagnose.
2210void Sema::CheckMemaccessArguments(const CallExpr *Call,
2211                                   CheckedMemoryFunction CMF,
2212                                   IdentifierInfo *FnName) {
2213  // It is possible to have a non-standard definition of memset.  Validate
2214  // we have enough arguments, and if not, abort further checking.
2215  unsigned ExpectedNumArgs = (CMF == CMF_Strndup ? 2 : 3);
2216  if (Call->getNumArgs() < ExpectedNumArgs)
2217    return;
2218
2219  unsigned LastArg = (CMF == CMF_Memset || CMF == CMF_Strndup ? 1 : 2);
2220  unsigned LenArg = (CMF == CMF_Strndup ? 1 : 2);
2221  const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts();
2222
2223  // We have special checking when the length is a sizeof expression.
2224  QualType SizeOfArgTy = getSizeOfArgType(LenExpr);
2225  const Expr *SizeOfArg = getSizeOfExprArg(LenExpr);
2226  llvm::FoldingSetNodeID SizeOfArgID;
2227
2228  for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) {
2229    const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts();
2230    SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange();
2231
2232    QualType DestTy = Dest->getType();
2233    if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
2234      QualType PointeeTy = DestPtrTy->getPointeeType();
2235
2236      // Never warn about void type pointers. This can be used to suppress
2237      // false positives.
2238      if (PointeeTy->isVoidType())
2239        continue;
2240
2241      // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by
2242      // actually comparing the expressions for equality. Because computing the
2243      // expression IDs can be expensive, we only do this if the diagnostic is
2244      // enabled.
2245      if (SizeOfArg &&
2246          Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess,
2247                                   SizeOfArg->getExprLoc())) {
2248        // We only compute IDs for expressions if the warning is enabled, and
2249        // cache the sizeof arg's ID.
2250        if (SizeOfArgID == llvm::FoldingSetNodeID())
2251          SizeOfArg->Profile(SizeOfArgID, Context, true);
2252        llvm::FoldingSetNodeID DestID;
2253        Dest->Profile(DestID, Context, true);
2254        if (DestID == SizeOfArgID) {
2255          // TODO: For strncpy() and friends, this could suggest sizeof(dst)
2256          //       over sizeof(src) as well.
2257          unsigned ActionIdx = 0; // Default is to suggest dereferencing.
2258          if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest))
2259            if (UnaryOp->getOpcode() == UO_AddrOf)
2260              ActionIdx = 1; // If its an address-of operator, just remove it.
2261          if (Context.getTypeSize(PointeeTy) == Context.getCharWidth())
2262            ActionIdx = 2; // If the pointee's size is sizeof(char),
2263                           // suggest an explicit length.
2264          unsigned DestSrcSelect = (CMF == CMF_Strndup ? 1 : ArgIdx);
2265          DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest,
2266                              PDiag(diag::warn_sizeof_pointer_expr_memaccess)
2267                                << FnName << DestSrcSelect << ActionIdx
2268                                << Dest->getSourceRange()
2269                                << SizeOfArg->getSourceRange());
2270          break;
2271        }
2272      }
2273
2274      // Also check for cases where the sizeof argument is the exact same
2275      // type as the memory argument, and where it points to a user-defined
2276      // record type.
2277      if (SizeOfArgTy != QualType()) {
2278        if (PointeeTy->isRecordType() &&
2279            Context.typesAreCompatible(SizeOfArgTy, DestTy)) {
2280          DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest,
2281                              PDiag(diag::warn_sizeof_pointer_type_memaccess)
2282                                << FnName << SizeOfArgTy << ArgIdx
2283                                << PointeeTy << Dest->getSourceRange()
2284                                << LenExpr->getSourceRange());
2285          break;
2286        }
2287      }
2288
2289      // Always complain about dynamic classes.
2290      if (isDynamicClassType(PointeeTy))
2291        DiagRuntimeBehavior(
2292          Dest->getExprLoc(), Dest,
2293          PDiag(diag::warn_dyn_class_memaccess)
2294            << (CMF == CMF_Memcmp ? ArgIdx + 2 : ArgIdx) << FnName << PointeeTy
2295            // "overwritten" if we're warning about the destination for any call
2296            // but memcmp; otherwise a verb appropriate to the call.
2297            << (ArgIdx == 0 && CMF != CMF_Memcmp ? 0 : (unsigned)CMF)
2298            << Call->getCallee()->getSourceRange());
2299      else if (PointeeTy.hasNonTrivialObjCLifetime() && CMF != CMF_Memset)
2300        DiagRuntimeBehavior(
2301          Dest->getExprLoc(), Dest,
2302          PDiag(diag::warn_arc_object_memaccess)
2303            << ArgIdx << FnName << PointeeTy
2304            << Call->getCallee()->getSourceRange());
2305      else
2306        continue;
2307
2308      DiagRuntimeBehavior(
2309        Dest->getExprLoc(), Dest,
2310        PDiag(diag::note_bad_memaccess_silence)
2311          << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
2312      break;
2313    }
2314  }
2315}
2316
2317// A little helper routine: ignore addition and subtraction of integer literals.
2318// This intentionally does not ignore all integer constant expressions because
2319// we don't want to remove sizeof().
2320static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) {
2321  Ex = Ex->IgnoreParenCasts();
2322
2323  for (;;) {
2324    const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex);
2325    if (!BO || !BO->isAdditiveOp())
2326      break;
2327
2328    const Expr *RHS = BO->getRHS()->IgnoreParenCasts();
2329    const Expr *LHS = BO->getLHS()->IgnoreParenCasts();
2330
2331    if (isa<IntegerLiteral>(RHS))
2332      Ex = LHS;
2333    else if (isa<IntegerLiteral>(LHS))
2334      Ex = RHS;
2335    else
2336      break;
2337  }
2338
2339  return Ex;
2340}
2341
2342// Warn if the user has made the 'size' argument to strlcpy or strlcat
2343// be the size of the source, instead of the destination.
2344void Sema::CheckStrlcpycatArguments(const CallExpr *Call,
2345                                    IdentifierInfo *FnName) {
2346
2347  // Don't crash if the user has the wrong number of arguments
2348  if (Call->getNumArgs() != 3)
2349    return;
2350
2351  const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context);
2352  const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context);
2353  const Expr *CompareWithSrc = NULL;
2354
2355  // Look for 'strlcpy(dst, x, sizeof(x))'
2356  if (const Expr *Ex = getSizeOfExprArg(SizeArg))
2357    CompareWithSrc = Ex;
2358  else {
2359    // Look for 'strlcpy(dst, x, strlen(x))'
2360    if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) {
2361      if (SizeCall->isBuiltinCall(Context) == Builtin::BIstrlen
2362          && SizeCall->getNumArgs() == 1)
2363        CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context);
2364    }
2365  }
2366
2367  if (!CompareWithSrc)
2368    return;
2369
2370  // Determine if the argument to sizeof/strlen is equal to the source
2371  // argument.  In principle there's all kinds of things you could do
2372  // here, for instance creating an == expression and evaluating it with
2373  // EvaluateAsBooleanCondition, but this uses a more direct technique:
2374  const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg);
2375  if (!SrcArgDRE)
2376    return;
2377
2378  const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc);
2379  if (!CompareWithSrcDRE ||
2380      SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl())
2381    return;
2382
2383  const Expr *OriginalSizeArg = Call->getArg(2);
2384  Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size)
2385    << OriginalSizeArg->getSourceRange() << FnName;
2386
2387  // Output a FIXIT hint if the destination is an array (rather than a
2388  // pointer to an array).  This could be enhanced to handle some
2389  // pointers if we know the actual size, like if DstArg is 'array+2'
2390  // we could say 'sizeof(array)-2'.
2391  const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts();
2392  QualType DstArgTy = DstArg->getType();
2393
2394  // Only handle constant-sized or VLAs, but not flexible members.
2395  if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) {
2396    // Only issue the FIXIT for arrays of size > 1.
2397    if (CAT->getSize().getSExtValue() <= 1)
2398      return;
2399  } else if (!DstArgTy->isVariableArrayType()) {
2400    return;
2401  }
2402
2403  llvm::SmallString<128> sizeString;
2404  llvm::raw_svector_ostream OS(sizeString);
2405  OS << "sizeof(";
2406  DstArg->printPretty(OS, Context, 0, getPrintingPolicy());
2407  OS << ")";
2408
2409  Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size)
2410    << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(),
2411                                    OS.str());
2412}
2413
2414//===--- CHECK: Return Address of Stack Variable --------------------------===//
2415
2416static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars);
2417static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars);
2418
2419/// CheckReturnStackAddr - Check if a return statement returns the address
2420///   of a stack variable.
2421void
2422Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
2423                           SourceLocation ReturnLoc) {
2424
2425  Expr *stackE = 0;
2426  SmallVector<DeclRefExpr *, 8> refVars;
2427
2428  // Perform checking for returned stack addresses, local blocks,
2429  // label addresses or references to temporaries.
2430  if (lhsType->isPointerType() ||
2431      (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) {
2432    stackE = EvalAddr(RetValExp, refVars);
2433  } else if (lhsType->isReferenceType()) {
2434    stackE = EvalVal(RetValExp, refVars);
2435  }
2436
2437  if (stackE == 0)
2438    return; // Nothing suspicious was found.
2439
2440  SourceLocation diagLoc;
2441  SourceRange diagRange;
2442  if (refVars.empty()) {
2443    diagLoc = stackE->getLocStart();
2444    diagRange = stackE->getSourceRange();
2445  } else {
2446    // We followed through a reference variable. 'stackE' contains the
2447    // problematic expression but we will warn at the return statement pointing
2448    // at the reference variable. We will later display the "trail" of
2449    // reference variables using notes.
2450    diagLoc = refVars[0]->getLocStart();
2451    diagRange = refVars[0]->getSourceRange();
2452  }
2453
2454  if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
2455    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
2456                                             : diag::warn_ret_stack_addr)
2457     << DR->getDecl()->getDeclName() << diagRange;
2458  } else if (isa<BlockExpr>(stackE)) { // local block.
2459    Diag(diagLoc, diag::err_ret_local_block) << diagRange;
2460  } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
2461    Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
2462  } else { // local temporary.
2463    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
2464                                             : diag::warn_ret_local_temp_addr)
2465     << diagRange;
2466  }
2467
2468  // Display the "trail" of reference variables that we followed until we
2469  // found the problematic expression using notes.
2470  for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
2471    VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
2472    // If this var binds to another reference var, show the range of the next
2473    // var, otherwise the var binds to the problematic expression, in which case
2474    // show the range of the expression.
2475    SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
2476                                  : stackE->getSourceRange();
2477    Diag(VD->getLocation(), diag::note_ref_var_local_bind)
2478      << VD->getDeclName() << range;
2479  }
2480}
2481
2482/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
2483///  check if the expression in a return statement evaluates to an address
2484///  to a location on the stack, a local block, an address of a label, or a
2485///  reference to local temporary. The recursion is used to traverse the
2486///  AST of the return expression, with recursion backtracking when we
2487///  encounter a subexpression that (1) clearly does not lead to one of the
2488///  above problematic expressions (2) is something we cannot determine leads to
2489///  a problematic expression based on such local checking.
2490///
2491///  Both EvalAddr and EvalVal follow through reference variables to evaluate
2492///  the expression that they point to. Such variables are added to the
2493///  'refVars' vector so that we know what the reference variable "trail" was.
2494///
2495///  EvalAddr processes expressions that are pointers that are used as
2496///  references (and not L-values).  EvalVal handles all other values.
2497///  At the base case of the recursion is a check for the above problematic
2498///  expressions.
2499///
2500///  This implementation handles:
2501///
2502///   * pointer-to-pointer casts
2503///   * implicit conversions from array references to pointers
2504///   * taking the address of fields
2505///   * arbitrary interplay between "&" and "*" operators
2506///   * pointer arithmetic from an address of a stack variable
2507///   * taking the address of an array element where the array is on the stack
2508static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) {
2509  if (E->isTypeDependent())
2510      return NULL;
2511
2512  // We should only be called for evaluating pointer expressions.
2513  assert((E->getType()->isAnyPointerType() ||
2514          E->getType()->isBlockPointerType() ||
2515          E->getType()->isObjCQualifiedIdType()) &&
2516         "EvalAddr only works on pointers");
2517
2518  E = E->IgnoreParens();
2519
2520  // Our "symbolic interpreter" is just a dispatch off the currently
2521  // viewed AST node.  We then recursively traverse the AST by calling
2522  // EvalAddr and EvalVal appropriately.
2523  switch (E->getStmtClass()) {
2524  case Stmt::DeclRefExprClass: {
2525    DeclRefExpr *DR = cast<DeclRefExpr>(E);
2526
2527    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2528      // If this is a reference variable, follow through to the expression that
2529      // it points to.
2530      if (V->hasLocalStorage() &&
2531          V->getType()->isReferenceType() && V->hasInit()) {
2532        // Add the reference variable to the "trail".
2533        refVars.push_back(DR);
2534        return EvalAddr(V->getInit(), refVars);
2535      }
2536
2537    return NULL;
2538  }
2539
2540  case Stmt::UnaryOperatorClass: {
2541    // The only unary operator that make sense to handle here
2542    // is AddrOf.  All others don't make sense as pointers.
2543    UnaryOperator *U = cast<UnaryOperator>(E);
2544
2545    if (U->getOpcode() == UO_AddrOf)
2546      return EvalVal(U->getSubExpr(), refVars);
2547    else
2548      return NULL;
2549  }
2550
2551  case Stmt::BinaryOperatorClass: {
2552    // Handle pointer arithmetic.  All other binary operators are not valid
2553    // in this context.
2554    BinaryOperator *B = cast<BinaryOperator>(E);
2555    BinaryOperatorKind op = B->getOpcode();
2556
2557    if (op != BO_Add && op != BO_Sub)
2558      return NULL;
2559
2560    Expr *Base = B->getLHS();
2561
2562    // Determine which argument is the real pointer base.  It could be
2563    // the RHS argument instead of the LHS.
2564    if (!Base->getType()->isPointerType()) Base = B->getRHS();
2565
2566    assert (Base->getType()->isPointerType());
2567    return EvalAddr(Base, refVars);
2568  }
2569
2570  // For conditional operators we need to see if either the LHS or RHS are
2571  // valid DeclRefExpr*s.  If one of them is valid, we return it.
2572  case Stmt::ConditionalOperatorClass: {
2573    ConditionalOperator *C = cast<ConditionalOperator>(E);
2574
2575    // Handle the GNU extension for missing LHS.
2576    if (Expr *lhsExpr = C->getLHS()) {
2577    // In C++, we can have a throw-expression, which has 'void' type.
2578      if (!lhsExpr->getType()->isVoidType())
2579        if (Expr* LHS = EvalAddr(lhsExpr, refVars))
2580          return LHS;
2581    }
2582
2583    // In C++, we can have a throw-expression, which has 'void' type.
2584    if (C->getRHS()->getType()->isVoidType())
2585      return NULL;
2586
2587    return EvalAddr(C->getRHS(), refVars);
2588  }
2589
2590  case Stmt::BlockExprClass:
2591    if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
2592      return E; // local block.
2593    return NULL;
2594
2595  case Stmt::AddrLabelExprClass:
2596    return E; // address of label.
2597
2598  // For casts, we need to handle conversions from arrays to
2599  // pointer values, and pointer-to-pointer conversions.
2600  case Stmt::ImplicitCastExprClass:
2601  case Stmt::CStyleCastExprClass:
2602  case Stmt::CXXFunctionalCastExprClass:
2603  case Stmt::ObjCBridgedCastExprClass: {
2604    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
2605    QualType T = SubExpr->getType();
2606
2607    if (SubExpr->getType()->isPointerType() ||
2608        SubExpr->getType()->isBlockPointerType() ||
2609        SubExpr->getType()->isObjCQualifiedIdType())
2610      return EvalAddr(SubExpr, refVars);
2611    else if (T->isArrayType())
2612      return EvalVal(SubExpr, refVars);
2613    else
2614      return 0;
2615  }
2616
2617  // C++ casts.  For dynamic casts, static casts, and const casts, we
2618  // are always converting from a pointer-to-pointer, so we just blow
2619  // through the cast.  In the case the dynamic cast doesn't fail (and
2620  // return NULL), we take the conservative route and report cases
2621  // where we return the address of a stack variable.  For Reinterpre
2622  // FIXME: The comment about is wrong; we're not always converting
2623  // from pointer to pointer. I'm guessing that this code should also
2624  // handle references to objects.
2625  case Stmt::CXXStaticCastExprClass:
2626  case Stmt::CXXDynamicCastExprClass:
2627  case Stmt::CXXConstCastExprClass:
2628  case Stmt::CXXReinterpretCastExprClass: {
2629      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
2630      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
2631        return EvalAddr(S, refVars);
2632      else
2633        return NULL;
2634  }
2635
2636  case Stmt::MaterializeTemporaryExprClass:
2637    if (Expr *Result = EvalAddr(
2638                         cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
2639                                refVars))
2640      return Result;
2641
2642    return E;
2643
2644  // Everything else: we simply don't reason about them.
2645  default:
2646    return NULL;
2647  }
2648}
2649
2650
2651///  EvalVal - This function is complements EvalAddr in the mutual recursion.
2652///   See the comments for EvalAddr for more details.
2653static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) {
2654do {
2655  // We should only be called for evaluating non-pointer expressions, or
2656  // expressions with a pointer type that are not used as references but instead
2657  // are l-values (e.g., DeclRefExpr with a pointer type).
2658
2659  // Our "symbolic interpreter" is just a dispatch off the currently
2660  // viewed AST node.  We then recursively traverse the AST by calling
2661  // EvalAddr and EvalVal appropriately.
2662
2663  E = E->IgnoreParens();
2664  switch (E->getStmtClass()) {
2665  case Stmt::ImplicitCastExprClass: {
2666    ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
2667    if (IE->getValueKind() == VK_LValue) {
2668      E = IE->getSubExpr();
2669      continue;
2670    }
2671    return NULL;
2672  }
2673
2674  case Stmt::DeclRefExprClass: {
2675    // When we hit a DeclRefExpr we are looking at code that refers to a
2676    // variable's name. If it's not a reference variable we check if it has
2677    // local storage within the function, and if so, return the expression.
2678    DeclRefExpr *DR = cast<DeclRefExpr>(E);
2679
2680    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2681      if (V->hasLocalStorage()) {
2682        if (!V->getType()->isReferenceType())
2683          return DR;
2684
2685        // Reference variable, follow through to the expression that
2686        // it points to.
2687        if (V->hasInit()) {
2688          // Add the reference variable to the "trail".
2689          refVars.push_back(DR);
2690          return EvalVal(V->getInit(), refVars);
2691        }
2692      }
2693
2694    return NULL;
2695  }
2696
2697  case Stmt::UnaryOperatorClass: {
2698    // The only unary operator that make sense to handle here
2699    // is Deref.  All others don't resolve to a "name."  This includes
2700    // handling all sorts of rvalues passed to a unary operator.
2701    UnaryOperator *U = cast<UnaryOperator>(E);
2702
2703    if (U->getOpcode() == UO_Deref)
2704      return EvalAddr(U->getSubExpr(), refVars);
2705
2706    return NULL;
2707  }
2708
2709  case Stmt::ArraySubscriptExprClass: {
2710    // Array subscripts are potential references to data on the stack.  We
2711    // retrieve the DeclRefExpr* for the array variable if it indeed
2712    // has local storage.
2713    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars);
2714  }
2715
2716  case Stmt::ConditionalOperatorClass: {
2717    // For conditional operators we need to see if either the LHS or RHS are
2718    // non-NULL Expr's.  If one is non-NULL, we return it.
2719    ConditionalOperator *C = cast<ConditionalOperator>(E);
2720
2721    // Handle the GNU extension for missing LHS.
2722    if (Expr *lhsExpr = C->getLHS())
2723      if (Expr *LHS = EvalVal(lhsExpr, refVars))
2724        return LHS;
2725
2726    return EvalVal(C->getRHS(), refVars);
2727  }
2728
2729  // Accesses to members are potential references to data on the stack.
2730  case Stmt::MemberExprClass: {
2731    MemberExpr *M = cast<MemberExpr>(E);
2732
2733    // Check for indirect access.  We only want direct field accesses.
2734    if (M->isArrow())
2735      return NULL;
2736
2737    // Check whether the member type is itself a reference, in which case
2738    // we're not going to refer to the member, but to what the member refers to.
2739    if (M->getMemberDecl()->getType()->isReferenceType())
2740      return NULL;
2741
2742    return EvalVal(M->getBase(), refVars);
2743  }
2744
2745  case Stmt::MaterializeTemporaryExprClass:
2746    if (Expr *Result = EvalVal(
2747                          cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
2748                               refVars))
2749      return Result;
2750
2751    return E;
2752
2753  default:
2754    // Check that we don't return or take the address of a reference to a
2755    // temporary. This is only useful in C++.
2756    if (!E->isTypeDependent() && E->isRValue())
2757      return E;
2758
2759    // Everything else: we simply don't reason about them.
2760    return NULL;
2761  }
2762} while (true);
2763}
2764
2765//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
2766
2767/// Check for comparisons of floating point operands using != and ==.
2768/// Issue a warning if these are no self-comparisons, as they are not likely
2769/// to do what the programmer intended.
2770void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) {
2771  bool EmitWarning = true;
2772
2773  Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts();
2774  Expr* RightExprSansParen = RHS->IgnoreParenImpCasts();
2775
2776  // Special case: check for x == x (which is OK).
2777  // Do not emit warnings for such cases.
2778  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
2779    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
2780      if (DRL->getDecl() == DRR->getDecl())
2781        EmitWarning = false;
2782
2783
2784  // Special case: check for comparisons against literals that can be exactly
2785  //  represented by APFloat.  In such cases, do not emit a warning.  This
2786  //  is a heuristic: often comparison against such literals are used to
2787  //  detect if a value in a variable has not changed.  This clearly can
2788  //  lead to false negatives.
2789  if (EmitWarning) {
2790    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
2791      if (FLL->isExact())
2792        EmitWarning = false;
2793    } else
2794      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
2795        if (FLR->isExact())
2796          EmitWarning = false;
2797    }
2798  }
2799
2800  // Check for comparisons with builtin types.
2801  if (EmitWarning)
2802    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
2803      if (CL->isBuiltinCall(Context))
2804        EmitWarning = false;
2805
2806  if (EmitWarning)
2807    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
2808      if (CR->isBuiltinCall(Context))
2809        EmitWarning = false;
2810
2811  // Emit the diagnostic.
2812  if (EmitWarning)
2813    Diag(Loc, diag::warn_floatingpoint_eq)
2814      << LHS->getSourceRange() << RHS->getSourceRange();
2815}
2816
2817//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
2818//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
2819
2820namespace {
2821
2822/// Structure recording the 'active' range of an integer-valued
2823/// expression.
2824struct IntRange {
2825  /// The number of bits active in the int.
2826  unsigned Width;
2827
2828  /// True if the int is known not to have negative values.
2829  bool NonNegative;
2830
2831  IntRange(unsigned Width, bool NonNegative)
2832    : Width(Width), NonNegative(NonNegative)
2833  {}
2834
2835  /// Returns the range of the bool type.
2836  static IntRange forBoolType() {
2837    return IntRange(1, true);
2838  }
2839
2840  /// Returns the range of an opaque value of the given integral type.
2841  static IntRange forValueOfType(ASTContext &C, QualType T) {
2842    return forValueOfCanonicalType(C,
2843                          T->getCanonicalTypeInternal().getTypePtr());
2844  }
2845
2846  /// Returns the range of an opaque value of a canonical integral type.
2847  static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
2848    assert(T->isCanonicalUnqualified());
2849
2850    if (const VectorType *VT = dyn_cast<VectorType>(T))
2851      T = VT->getElementType().getTypePtr();
2852    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2853      T = CT->getElementType().getTypePtr();
2854
2855    // For enum types, use the known bit width of the enumerators.
2856    if (const EnumType *ET = dyn_cast<EnumType>(T)) {
2857      EnumDecl *Enum = ET->getDecl();
2858      if (!Enum->isCompleteDefinition())
2859        return IntRange(C.getIntWidth(QualType(T, 0)), false);
2860
2861      unsigned NumPositive = Enum->getNumPositiveBits();
2862      unsigned NumNegative = Enum->getNumNegativeBits();
2863
2864      return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
2865    }
2866
2867    const BuiltinType *BT = cast<BuiltinType>(T);
2868    assert(BT->isInteger());
2869
2870    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2871  }
2872
2873  /// Returns the "target" range of a canonical integral type, i.e.
2874  /// the range of values expressible in the type.
2875  ///
2876  /// This matches forValueOfCanonicalType except that enums have the
2877  /// full range of their type, not the range of their enumerators.
2878  static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
2879    assert(T->isCanonicalUnqualified());
2880
2881    if (const VectorType *VT = dyn_cast<VectorType>(T))
2882      T = VT->getElementType().getTypePtr();
2883    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2884      T = CT->getElementType().getTypePtr();
2885    if (const EnumType *ET = dyn_cast<EnumType>(T))
2886      T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr();
2887
2888    const BuiltinType *BT = cast<BuiltinType>(T);
2889    assert(BT->isInteger());
2890
2891    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2892  }
2893
2894  /// Returns the supremum of two ranges: i.e. their conservative merge.
2895  static IntRange join(IntRange L, IntRange R) {
2896    return IntRange(std::max(L.Width, R.Width),
2897                    L.NonNegative && R.NonNegative);
2898  }
2899
2900  /// Returns the infinum of two ranges: i.e. their aggressive merge.
2901  static IntRange meet(IntRange L, IntRange R) {
2902    return IntRange(std::min(L.Width, R.Width),
2903                    L.NonNegative || R.NonNegative);
2904  }
2905};
2906
2907IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) {
2908  if (value.isSigned() && value.isNegative())
2909    return IntRange(value.getMinSignedBits(), false);
2910
2911  if (value.getBitWidth() > MaxWidth)
2912    value = value.trunc(MaxWidth);
2913
2914  // isNonNegative() just checks the sign bit without considering
2915  // signedness.
2916  return IntRange(value.getActiveBits(), true);
2917}
2918
2919IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
2920                       unsigned MaxWidth) {
2921  if (result.isInt())
2922    return GetValueRange(C, result.getInt(), MaxWidth);
2923
2924  if (result.isVector()) {
2925    IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
2926    for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
2927      IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
2928      R = IntRange::join(R, El);
2929    }
2930    return R;
2931  }
2932
2933  if (result.isComplexInt()) {
2934    IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
2935    IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
2936    return IntRange::join(R, I);
2937  }
2938
2939  // This can happen with lossless casts to intptr_t of "based" lvalues.
2940  // Assume it might use arbitrary bits.
2941  // FIXME: The only reason we need to pass the type in here is to get
2942  // the sign right on this one case.  It would be nice if APValue
2943  // preserved this.
2944  assert(result.isLValue());
2945  return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType());
2946}
2947
2948/// Pseudo-evaluate the given integer expression, estimating the
2949/// range of values it might take.
2950///
2951/// \param MaxWidth - the width to which the value will be truncated
2952IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
2953  E = E->IgnoreParens();
2954
2955  // Try a full evaluation first.
2956  Expr::EvalResult result;
2957  if (E->Evaluate(result, C))
2958    return GetValueRange(C, result.Val, E->getType(), MaxWidth);
2959
2960  // I think we only want to look through implicit casts here; if the
2961  // user has an explicit widening cast, we should treat the value as
2962  // being of the new, wider type.
2963  if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
2964    if (CE->getCastKind() == CK_NoOp)
2965      return GetExprRange(C, CE->getSubExpr(), MaxWidth);
2966
2967    IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
2968
2969    bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
2970
2971    // Assume that non-integer casts can span the full range of the type.
2972    if (!isIntegerCast)
2973      return OutputTypeRange;
2974
2975    IntRange SubRange
2976      = GetExprRange(C, CE->getSubExpr(),
2977                     std::min(MaxWidth, OutputTypeRange.Width));
2978
2979    // Bail out if the subexpr's range is as wide as the cast type.
2980    if (SubRange.Width >= OutputTypeRange.Width)
2981      return OutputTypeRange;
2982
2983    // Otherwise, we take the smaller width, and we're non-negative if
2984    // either the output type or the subexpr is.
2985    return IntRange(SubRange.Width,
2986                    SubRange.NonNegative || OutputTypeRange.NonNegative);
2987  }
2988
2989  if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
2990    // If we can fold the condition, just take that operand.
2991    bool CondResult;
2992    if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
2993      return GetExprRange(C, CondResult ? CO->getTrueExpr()
2994                                        : CO->getFalseExpr(),
2995                          MaxWidth);
2996
2997    // Otherwise, conservatively merge.
2998    IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
2999    IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
3000    return IntRange::join(L, R);
3001  }
3002
3003  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
3004    switch (BO->getOpcode()) {
3005
3006    // Boolean-valued operations are single-bit and positive.
3007    case BO_LAnd:
3008    case BO_LOr:
3009    case BO_LT:
3010    case BO_GT:
3011    case BO_LE:
3012    case BO_GE:
3013    case BO_EQ:
3014    case BO_NE:
3015      return IntRange::forBoolType();
3016
3017    // The type of the assignments is the type of the LHS, so the RHS
3018    // is not necessarily the same type.
3019    case BO_MulAssign:
3020    case BO_DivAssign:
3021    case BO_RemAssign:
3022    case BO_AddAssign:
3023    case BO_SubAssign:
3024    case BO_XorAssign:
3025    case BO_OrAssign:
3026      // TODO: bitfields?
3027      return IntRange::forValueOfType(C, E->getType());
3028
3029    // Simple assignments just pass through the RHS, which will have
3030    // been coerced to the LHS type.
3031    case BO_Assign:
3032      // TODO: bitfields?
3033      return GetExprRange(C, BO->getRHS(), MaxWidth);
3034
3035    // Operations with opaque sources are black-listed.
3036    case BO_PtrMemD:
3037    case BO_PtrMemI:
3038      return IntRange::forValueOfType(C, E->getType());
3039
3040    // Bitwise-and uses the *infinum* of the two source ranges.
3041    case BO_And:
3042    case BO_AndAssign:
3043      return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
3044                            GetExprRange(C, BO->getRHS(), MaxWidth));
3045
3046    // Left shift gets black-listed based on a judgement call.
3047    case BO_Shl:
3048      // ...except that we want to treat '1 << (blah)' as logically
3049      // positive.  It's an important idiom.
3050      if (IntegerLiteral *I
3051            = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
3052        if (I->getValue() == 1) {
3053          IntRange R = IntRange::forValueOfType(C, E->getType());
3054          return IntRange(R.Width, /*NonNegative*/ true);
3055        }
3056      }
3057      // fallthrough
3058
3059    case BO_ShlAssign:
3060      return IntRange::forValueOfType(C, E->getType());
3061
3062    // Right shift by a constant can narrow its left argument.
3063    case BO_Shr:
3064    case BO_ShrAssign: {
3065      IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
3066
3067      // If the shift amount is a positive constant, drop the width by
3068      // that much.
3069      llvm::APSInt shift;
3070      if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
3071          shift.isNonNegative()) {
3072        unsigned zext = shift.getZExtValue();
3073        if (zext >= L.Width)
3074          L.Width = (L.NonNegative ? 0 : 1);
3075        else
3076          L.Width -= zext;
3077      }
3078
3079      return L;
3080    }
3081
3082    // Comma acts as its right operand.
3083    case BO_Comma:
3084      return GetExprRange(C, BO->getRHS(), MaxWidth);
3085
3086    // Black-list pointer subtractions.
3087    case BO_Sub:
3088      if (BO->getLHS()->getType()->isPointerType())
3089        return IntRange::forValueOfType(C, E->getType());
3090      break;
3091
3092    // The width of a division result is mostly determined by the size
3093    // of the LHS.
3094    case BO_Div: {
3095      // Don't 'pre-truncate' the operands.
3096      unsigned opWidth = C.getIntWidth(E->getType());
3097      IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
3098
3099      // If the divisor is constant, use that.
3100      llvm::APSInt divisor;
3101      if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) {
3102        unsigned log2 = divisor.logBase2(); // floor(log_2(divisor))
3103        if (log2 >= L.Width)
3104          L.Width = (L.NonNegative ? 0 : 1);
3105        else
3106          L.Width = std::min(L.Width - log2, MaxWidth);
3107        return L;
3108      }
3109
3110      // Otherwise, just use the LHS's width.
3111      IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
3112      return IntRange(L.Width, L.NonNegative && R.NonNegative);
3113    }
3114
3115    // The result of a remainder can't be larger than the result of
3116    // either side.
3117    case BO_Rem: {
3118      // Don't 'pre-truncate' the operands.
3119      unsigned opWidth = C.getIntWidth(E->getType());
3120      IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
3121      IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
3122
3123      IntRange meet = IntRange::meet(L, R);
3124      meet.Width = std::min(meet.Width, MaxWidth);
3125      return meet;
3126    }
3127
3128    // The default behavior is okay for these.
3129    case BO_Mul:
3130    case BO_Add:
3131    case BO_Xor:
3132    case BO_Or:
3133      break;
3134    }
3135
3136    // The default case is to treat the operation as if it were closed
3137    // on the narrowest type that encompasses both operands.
3138    IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
3139    IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
3140    return IntRange::join(L, R);
3141  }
3142
3143  if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
3144    switch (UO->getOpcode()) {
3145    // Boolean-valued operations are white-listed.
3146    case UO_LNot:
3147      return IntRange::forBoolType();
3148
3149    // Operations with opaque sources are black-listed.
3150    case UO_Deref:
3151    case UO_AddrOf: // should be impossible
3152      return IntRange::forValueOfType(C, E->getType());
3153
3154    default:
3155      return GetExprRange(C, UO->getSubExpr(), MaxWidth);
3156    }
3157  }
3158
3159  if (dyn_cast<OffsetOfExpr>(E)) {
3160    IntRange::forValueOfType(C, E->getType());
3161  }
3162
3163  if (FieldDecl *BitField = E->getBitField())
3164    return IntRange(BitField->getBitWidthValue(C),
3165                    BitField->getType()->isUnsignedIntegerOrEnumerationType());
3166
3167  return IntRange::forValueOfType(C, E->getType());
3168}
3169
3170IntRange GetExprRange(ASTContext &C, Expr *E) {
3171  return GetExprRange(C, E, C.getIntWidth(E->getType()));
3172}
3173
3174/// Checks whether the given value, which currently has the given
3175/// source semantics, has the same value when coerced through the
3176/// target semantics.
3177bool IsSameFloatAfterCast(const llvm::APFloat &value,
3178                          const llvm::fltSemantics &Src,
3179                          const llvm::fltSemantics &Tgt) {
3180  llvm::APFloat truncated = value;
3181
3182  bool ignored;
3183  truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
3184  truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
3185
3186  return truncated.bitwiseIsEqual(value);
3187}
3188
3189/// Checks whether the given value, which currently has the given
3190/// source semantics, has the same value when coerced through the
3191/// target semantics.
3192///
3193/// The value might be a vector of floats (or a complex number).
3194bool IsSameFloatAfterCast(const APValue &value,
3195                          const llvm::fltSemantics &Src,
3196                          const llvm::fltSemantics &Tgt) {
3197  if (value.isFloat())
3198    return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
3199
3200  if (value.isVector()) {
3201    for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
3202      if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
3203        return false;
3204    return true;
3205  }
3206
3207  assert(value.isComplexFloat());
3208  return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
3209          IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
3210}
3211
3212void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
3213
3214static bool IsZero(Sema &S, Expr *E) {
3215  // Suppress cases where we are comparing against an enum constant.
3216  if (const DeclRefExpr *DR =
3217      dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
3218    if (isa<EnumConstantDecl>(DR->getDecl()))
3219      return false;
3220
3221  // Suppress cases where the '0' value is expanded from a macro.
3222  if (E->getLocStart().isMacroID())
3223    return false;
3224
3225  llvm::APSInt Value;
3226  return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
3227}
3228
3229static bool HasEnumType(Expr *E) {
3230  // Strip off implicit integral promotions.
3231  while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
3232    if (ICE->getCastKind() != CK_IntegralCast &&
3233        ICE->getCastKind() != CK_NoOp)
3234      break;
3235    E = ICE->getSubExpr();
3236  }
3237
3238  return E->getType()->isEnumeralType();
3239}
3240
3241void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
3242  BinaryOperatorKind op = E->getOpcode();
3243  if (E->isValueDependent())
3244    return;
3245
3246  if (op == BO_LT && IsZero(S, E->getRHS())) {
3247    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
3248      << "< 0" << "false" << HasEnumType(E->getLHS())
3249      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3250  } else if (op == BO_GE && IsZero(S, E->getRHS())) {
3251    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
3252      << ">= 0" << "true" << HasEnumType(E->getLHS())
3253      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3254  } else if (op == BO_GT && IsZero(S, E->getLHS())) {
3255    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
3256      << "0 >" << "false" << HasEnumType(E->getRHS())
3257      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3258  } else if (op == BO_LE && IsZero(S, E->getLHS())) {
3259    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
3260      << "0 <=" << "true" << HasEnumType(E->getRHS())
3261      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
3262  }
3263}
3264
3265/// Analyze the operands of the given comparison.  Implements the
3266/// fallback case from AnalyzeComparison.
3267void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
3268  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
3269  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
3270}
3271
3272/// \brief Implements -Wsign-compare.
3273///
3274/// \param E the binary operator to check for warnings
3275void AnalyzeComparison(Sema &S, BinaryOperator *E) {
3276  // The type the comparison is being performed in.
3277  QualType T = E->getLHS()->getType();
3278  assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
3279         && "comparison with mismatched types");
3280
3281  // We don't do anything special if this isn't an unsigned integral
3282  // comparison:  we're only interested in integral comparisons, and
3283  // signed comparisons only happen in cases we don't care to warn about.
3284  //
3285  // We also don't care about value-dependent expressions or expressions
3286  // whose result is a constant.
3287  if (!T->hasUnsignedIntegerRepresentation()
3288      || E->isValueDependent() || E->isIntegerConstantExpr(S.Context))
3289    return AnalyzeImpConvsInComparison(S, E);
3290
3291  Expr *LHS = E->getLHS()->IgnoreParenImpCasts();
3292  Expr *RHS = E->getRHS()->IgnoreParenImpCasts();
3293
3294  // Check to see if one of the (unmodified) operands is of different
3295  // signedness.
3296  Expr *signedOperand, *unsignedOperand;
3297  if (LHS->getType()->hasSignedIntegerRepresentation()) {
3298    assert(!RHS->getType()->hasSignedIntegerRepresentation() &&
3299           "unsigned comparison between two signed integer expressions?");
3300    signedOperand = LHS;
3301    unsignedOperand = RHS;
3302  } else if (RHS->getType()->hasSignedIntegerRepresentation()) {
3303    signedOperand = RHS;
3304    unsignedOperand = LHS;
3305  } else {
3306    CheckTrivialUnsignedComparison(S, E);
3307    return AnalyzeImpConvsInComparison(S, E);
3308  }
3309
3310  // Otherwise, calculate the effective range of the signed operand.
3311  IntRange signedRange = GetExprRange(S.Context, signedOperand);
3312
3313  // Go ahead and analyze implicit conversions in the operands.  Note
3314  // that we skip the implicit conversions on both sides.
3315  AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc());
3316  AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc());
3317
3318  // If the signed range is non-negative, -Wsign-compare won't fire,
3319  // but we should still check for comparisons which are always true
3320  // or false.
3321  if (signedRange.NonNegative)
3322    return CheckTrivialUnsignedComparison(S, E);
3323
3324  // For (in)equality comparisons, if the unsigned operand is a
3325  // constant which cannot collide with a overflowed signed operand,
3326  // then reinterpreting the signed operand as unsigned will not
3327  // change the result of the comparison.
3328  if (E->isEqualityOp()) {
3329    unsigned comparisonWidth = S.Context.getIntWidth(T);
3330    IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
3331
3332    // We should never be unable to prove that the unsigned operand is
3333    // non-negative.
3334    assert(unsignedRange.NonNegative && "unsigned range includes negative?");
3335
3336    if (unsignedRange.Width < comparisonWidth)
3337      return;
3338  }
3339
3340  S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison)
3341    << LHS->getType() << RHS->getType()
3342    << LHS->getSourceRange() << RHS->getSourceRange();
3343}
3344
3345/// Analyzes an attempt to assign the given value to a bitfield.
3346///
3347/// Returns true if there was something fishy about the attempt.
3348bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
3349                               SourceLocation InitLoc) {
3350  assert(Bitfield->isBitField());
3351  if (Bitfield->isInvalidDecl())
3352    return false;
3353
3354  // White-list bool bitfields.
3355  if (Bitfield->getType()->isBooleanType())
3356    return false;
3357
3358  // Ignore value- or type-dependent expressions.
3359  if (Bitfield->getBitWidth()->isValueDependent() ||
3360      Bitfield->getBitWidth()->isTypeDependent() ||
3361      Init->isValueDependent() ||
3362      Init->isTypeDependent())
3363    return false;
3364
3365  Expr *OriginalInit = Init->IgnoreParenImpCasts();
3366
3367  Expr::EvalResult InitValue;
3368  if (!OriginalInit->Evaluate(InitValue, S.Context) ||
3369      !InitValue.Val.isInt())
3370    return false;
3371
3372  const llvm::APSInt &Value = InitValue.Val.getInt();
3373  unsigned OriginalWidth = Value.getBitWidth();
3374  unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context);
3375
3376  if (OriginalWidth <= FieldWidth)
3377    return false;
3378
3379  llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
3380
3381  // It's fairly common to write values into signed bitfields
3382  // that, if sign-extended, would end up becoming a different
3383  // value.  We don't want to warn about that.
3384  if (Value.isSigned() && Value.isNegative())
3385    TruncatedValue = TruncatedValue.sext(OriginalWidth);
3386  else
3387    TruncatedValue = TruncatedValue.zext(OriginalWidth);
3388
3389  if (Value == TruncatedValue)
3390    return false;
3391
3392  std::string PrettyValue = Value.toString(10);
3393  std::string PrettyTrunc = TruncatedValue.toString(10);
3394
3395  S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
3396    << PrettyValue << PrettyTrunc << OriginalInit->getType()
3397    << Init->getSourceRange();
3398
3399  return true;
3400}
3401
3402/// Analyze the given simple or compound assignment for warning-worthy
3403/// operations.
3404void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
3405  // Just recurse on the LHS.
3406  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
3407
3408  // We want to recurse on the RHS as normal unless we're assigning to
3409  // a bitfield.
3410  if (FieldDecl *Bitfield = E->getLHS()->getBitField()) {
3411    if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
3412                                  E->getOperatorLoc())) {
3413      // Recurse, ignoring any implicit conversions on the RHS.
3414      return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
3415                                        E->getOperatorLoc());
3416    }
3417  }
3418
3419  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
3420}
3421
3422/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
3423void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
3424                     SourceLocation CContext, unsigned diag) {
3425  S.Diag(E->getExprLoc(), diag)
3426    << SourceType << T << E->getSourceRange() << SourceRange(CContext);
3427}
3428
3429/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
3430void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext,
3431                     unsigned diag) {
3432  DiagnoseImpCast(S, E, E->getType(), T, CContext, diag);
3433}
3434
3435/// Diagnose an implicit cast from a literal expression. Does not warn when the
3436/// cast wouldn't lose information.
3437void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
3438                                    SourceLocation CContext) {
3439  // Try to convert the literal exactly to an integer. If we can, don't warn.
3440  bool isExact = false;
3441  const llvm::APFloat &Value = FL->getValue();
3442  llvm::APSInt IntegerValue(S.Context.getIntWidth(T),
3443                            T->hasUnsignedIntegerRepresentation());
3444  if (Value.convertToInteger(IntegerValue,
3445                             llvm::APFloat::rmTowardZero, &isExact)
3446      == llvm::APFloat::opOK && isExact)
3447    return;
3448
3449  S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
3450    << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext);
3451}
3452
3453std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
3454  if (!Range.Width) return "0";
3455
3456  llvm::APSInt ValueInRange = Value;
3457  ValueInRange.setIsSigned(!Range.NonNegative);
3458  ValueInRange = ValueInRange.trunc(Range.Width);
3459  return ValueInRange.toString(10);
3460}
3461
3462static bool isFromSystemMacro(Sema &S, SourceLocation loc) {
3463  SourceManager &smgr = S.Context.getSourceManager();
3464  return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc));
3465}
3466
3467void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
3468                             SourceLocation CC, bool *ICContext = 0) {
3469  if (E->isTypeDependent() || E->isValueDependent()) return;
3470
3471  const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
3472  const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
3473  if (Source == Target) return;
3474  if (Target->isDependentType()) return;
3475
3476  // If the conversion context location is invalid don't complain. We also
3477  // don't want to emit a warning if the issue occurs from the expansion of
3478  // a system macro. The problem is that 'getSpellingLoc()' is slow, so we
3479  // delay this check as long as possible. Once we detect we are in that
3480  // scenario, we just return.
3481  if (CC.isInvalid())
3482    return;
3483
3484  // Diagnose implicit casts to bool.
3485  if (Target->isSpecificBuiltinType(BuiltinType::Bool)) {
3486    if (isa<StringLiteral>(E))
3487      // Warn on string literal to bool.  Checks for string literals in logical
3488      // expressions, for instances, assert(0 && "error here"), is prevented
3489      // by a check in AnalyzeImplicitConversions().
3490      return DiagnoseImpCast(S, E, T, CC,
3491                             diag::warn_impcast_string_literal_to_bool);
3492    return; // Other casts to bool are not checked.
3493  }
3494
3495  // Strip vector types.
3496  if (isa<VectorType>(Source)) {
3497    if (!isa<VectorType>(Target)) {
3498      if (isFromSystemMacro(S, CC))
3499        return;
3500      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
3501    }
3502
3503    // If the vector cast is cast between two vectors of the same size, it is
3504    // a bitcast, not a conversion.
3505    if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target))
3506      return;
3507
3508    Source = cast<VectorType>(Source)->getElementType().getTypePtr();
3509    Target = cast<VectorType>(Target)->getElementType().getTypePtr();
3510  }
3511
3512  // Strip complex types.
3513  if (isa<ComplexType>(Source)) {
3514    if (!isa<ComplexType>(Target)) {
3515      if (isFromSystemMacro(S, CC))
3516        return;
3517
3518      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
3519    }
3520
3521    Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
3522    Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
3523  }
3524
3525  const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
3526  const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
3527
3528  // If the source is floating point...
3529  if (SourceBT && SourceBT->isFloatingPoint()) {
3530    // ...and the target is floating point...
3531    if (TargetBT && TargetBT->isFloatingPoint()) {
3532      // ...then warn if we're dropping FP rank.
3533
3534      // Builtin FP kinds are ordered by increasing FP rank.
3535      if (SourceBT->getKind() > TargetBT->getKind()) {
3536        // Don't warn about float constants that are precisely
3537        // representable in the target type.
3538        Expr::EvalResult result;
3539        if (E->Evaluate(result, S.Context)) {
3540          // Value might be a float, a float vector, or a float complex.
3541          if (IsSameFloatAfterCast(result.Val,
3542                   S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
3543                   S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
3544            return;
3545        }
3546
3547        if (isFromSystemMacro(S, CC))
3548          return;
3549
3550        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
3551      }
3552      return;
3553    }
3554
3555    // If the target is integral, always warn.
3556    if ((TargetBT && TargetBT->isInteger())) {
3557      if (isFromSystemMacro(S, CC))
3558        return;
3559
3560      Expr *InnerE = E->IgnoreParenImpCasts();
3561      // We also want to warn on, e.g., "int i = -1.234"
3562      if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE))
3563        if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus)
3564          InnerE = UOp->getSubExpr()->IgnoreParenImpCasts();
3565
3566      if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
3567        DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
3568      } else {
3569        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
3570      }
3571    }
3572
3573    return;
3574  }
3575
3576  if (!Source->isIntegerType() || !Target->isIntegerType())
3577    return;
3578
3579  if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)
3580           == Expr::NPCK_GNUNull) && Target->isIntegerType()) {
3581    S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer)
3582        << E->getSourceRange() << clang::SourceRange(CC);
3583    return;
3584  }
3585
3586  IntRange SourceRange = GetExprRange(S.Context, E);
3587  IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
3588
3589  if (SourceRange.Width > TargetRange.Width) {
3590    // If the source is a constant, use a default-on diagnostic.
3591    // TODO: this should happen for bitfield stores, too.
3592    llvm::APSInt Value(32);
3593    if (E->isIntegerConstantExpr(Value, S.Context)) {
3594      if (isFromSystemMacro(S, CC))
3595        return;
3596
3597      std::string PrettySourceValue = Value.toString(10);
3598      std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
3599
3600      S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant)
3601        << PrettySourceValue << PrettyTargetValue
3602        << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC);
3603      return;
3604    }
3605
3606    // People want to build with -Wshorten-64-to-32 and not -Wconversion.
3607    if (isFromSystemMacro(S, CC))
3608      return;
3609
3610    if (SourceRange.Width == 64 && TargetRange.Width == 32)
3611      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32);
3612    return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
3613  }
3614
3615  if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
3616      (!TargetRange.NonNegative && SourceRange.NonNegative &&
3617       SourceRange.Width == TargetRange.Width)) {
3618
3619    if (isFromSystemMacro(S, CC))
3620      return;
3621
3622    unsigned DiagID = diag::warn_impcast_integer_sign;
3623
3624    // Traditionally, gcc has warned about this under -Wsign-compare.
3625    // We also want to warn about it in -Wconversion.
3626    // So if -Wconversion is off, use a completely identical diagnostic
3627    // in the sign-compare group.
3628    // The conditional-checking code will
3629    if (ICContext) {
3630      DiagID = diag::warn_impcast_integer_sign_conditional;
3631      *ICContext = true;
3632    }
3633
3634    return DiagnoseImpCast(S, E, T, CC, DiagID);
3635  }
3636
3637  // Diagnose conversions between different enumeration types.
3638  // In C, we pretend that the type of an EnumConstantDecl is its enumeration
3639  // type, to give us better diagnostics.
3640  QualType SourceType = E->getType();
3641  if (!S.getLangOptions().CPlusPlus) {
3642    if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
3643      if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
3644        EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
3645        SourceType = S.Context.getTypeDeclType(Enum);
3646        Source = S.Context.getCanonicalType(SourceType).getTypePtr();
3647      }
3648  }
3649
3650  if (const EnumType *SourceEnum = Source->getAs<EnumType>())
3651    if (const EnumType *TargetEnum = Target->getAs<EnumType>())
3652      if ((SourceEnum->getDecl()->getIdentifier() ||
3653           SourceEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3654          (TargetEnum->getDecl()->getIdentifier() ||
3655           TargetEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3656          SourceEnum != TargetEnum) {
3657        if (isFromSystemMacro(S, CC))
3658          return;
3659
3660        return DiagnoseImpCast(S, E, SourceType, T, CC,
3661                               diag::warn_impcast_different_enum_types);
3662      }
3663
3664  return;
3665}
3666
3667void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T);
3668
3669void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
3670                             SourceLocation CC, bool &ICContext) {
3671  E = E->IgnoreParenImpCasts();
3672
3673  if (isa<ConditionalOperator>(E))
3674    return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T);
3675
3676  AnalyzeImplicitConversions(S, E, CC);
3677  if (E->getType() != T)
3678    return CheckImplicitConversion(S, E, T, CC, &ICContext);
3679  return;
3680}
3681
3682void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) {
3683  SourceLocation CC = E->getQuestionLoc();
3684
3685  AnalyzeImplicitConversions(S, E->getCond(), CC);
3686
3687  bool Suspicious = false;
3688  CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
3689  CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
3690
3691  // If -Wconversion would have warned about either of the candidates
3692  // for a signedness conversion to the context type...
3693  if (!Suspicious) return;
3694
3695  // ...but it's currently ignored...
3696  if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
3697                                 CC))
3698    return;
3699
3700  // ...then check whether it would have warned about either of the
3701  // candidates for a signedness conversion to the condition type.
3702  if (E->getType() == T) return;
3703
3704  Suspicious = false;
3705  CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
3706                          E->getType(), CC, &Suspicious);
3707  if (!Suspicious)
3708    CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
3709                            E->getType(), CC, &Suspicious);
3710}
3711
3712/// AnalyzeImplicitConversions - Find and report any interesting
3713/// implicit conversions in the given expression.  There are a couple
3714/// of competing diagnostics here, -Wconversion and -Wsign-compare.
3715void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
3716  QualType T = OrigE->getType();
3717  Expr *E = OrigE->IgnoreParenImpCasts();
3718
3719  if (E->isTypeDependent() || E->isValueDependent())
3720    return;
3721
3722  // For conditional operators, we analyze the arguments as if they
3723  // were being fed directly into the output.
3724  if (isa<ConditionalOperator>(E)) {
3725    ConditionalOperator *CO = cast<ConditionalOperator>(E);
3726    CheckConditionalOperator(S, CO, T);
3727    return;
3728  }
3729
3730  // Go ahead and check any implicit conversions we might have skipped.
3731  // The non-canonical typecheck is just an optimization;
3732  // CheckImplicitConversion will filter out dead implicit conversions.
3733  if (E->getType() != T)
3734    CheckImplicitConversion(S, E, T, CC);
3735
3736  // Now continue drilling into this expression.
3737
3738  // Skip past explicit casts.
3739  if (isa<ExplicitCastExpr>(E)) {
3740    E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
3741    return AnalyzeImplicitConversions(S, E, CC);
3742  }
3743
3744  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
3745    // Do a somewhat different check with comparison operators.
3746    if (BO->isComparisonOp())
3747      return AnalyzeComparison(S, BO);
3748
3749    // And with assignments and compound assignments.
3750    if (BO->isAssignmentOp())
3751      return AnalyzeAssignment(S, BO);
3752  }
3753
3754  // These break the otherwise-useful invariant below.  Fortunately,
3755  // we don't really need to recurse into them, because any internal
3756  // expressions should have been analyzed already when they were
3757  // built into statements.
3758  if (isa<StmtExpr>(E)) return;
3759
3760  // Don't descend into unevaluated contexts.
3761  if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
3762
3763  // Now just recurse over the expression's children.
3764  CC = E->getExprLoc();
3765  BinaryOperator *BO = dyn_cast<BinaryOperator>(E);
3766  bool IsLogicalOperator = BO && BO->isLogicalOp();
3767  for (Stmt::child_range I = E->children(); I; ++I) {
3768    Expr *ChildExpr = cast<Expr>(*I);
3769    if (IsLogicalOperator &&
3770        isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts()))
3771      // Ignore checking string literals that are in logical operators.
3772      continue;
3773    AnalyzeImplicitConversions(S, ChildExpr, CC);
3774  }
3775}
3776
3777} // end anonymous namespace
3778
3779/// Diagnoses "dangerous" implicit conversions within the given
3780/// expression (which is a full expression).  Implements -Wconversion
3781/// and -Wsign-compare.
3782///
3783/// \param CC the "context" location of the implicit conversion, i.e.
3784///   the most location of the syntactic entity requiring the implicit
3785///   conversion
3786void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
3787  // Don't diagnose in unevaluated contexts.
3788  if (ExprEvalContexts.back().Context == Sema::Unevaluated)
3789    return;
3790
3791  // Don't diagnose for value- or type-dependent expressions.
3792  if (E->isTypeDependent() || E->isValueDependent())
3793    return;
3794
3795  // Check for array bounds violations in cases where the check isn't triggered
3796  // elsewhere for other Expr types (like BinaryOperators), e.g. when an
3797  // ArraySubscriptExpr is on the RHS of a variable initialization.
3798  CheckArrayAccess(E);
3799
3800  // This is not the right CC for (e.g.) a variable initialization.
3801  AnalyzeImplicitConversions(*this, E, CC);
3802}
3803
3804void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
3805                                       FieldDecl *BitField,
3806                                       Expr *Init) {
3807  (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
3808}
3809
3810/// CheckParmsForFunctionDef - Check that the parameters of the given
3811/// function are appropriate for the definition of a function. This
3812/// takes care of any checks that cannot be performed on the
3813/// declaration itself, e.g., that the types of each of the function
3814/// parameters are complete.
3815bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
3816                                    bool CheckParameterNames) {
3817  bool HasInvalidParm = false;
3818  for (; P != PEnd; ++P) {
3819    ParmVarDecl *Param = *P;
3820
3821    // C99 6.7.5.3p4: the parameters in a parameter type list in a
3822    // function declarator that is part of a function definition of
3823    // that function shall not have incomplete type.
3824    //
3825    // This is also C++ [dcl.fct]p6.
3826    if (!Param->isInvalidDecl() &&
3827        RequireCompleteType(Param->getLocation(), Param->getType(),
3828                               diag::err_typecheck_decl_incomplete_type)) {
3829      Param->setInvalidDecl();
3830      HasInvalidParm = true;
3831    }
3832
3833    // C99 6.9.1p5: If the declarator includes a parameter type list, the
3834    // declaration of each parameter shall include an identifier.
3835    if (CheckParameterNames &&
3836        Param->getIdentifier() == 0 &&
3837        !Param->isImplicit() &&
3838        !getLangOptions().CPlusPlus)
3839      Diag(Param->getLocation(), diag::err_parameter_name_omitted);
3840
3841    // C99 6.7.5.3p12:
3842    //   If the function declarator is not part of a definition of that
3843    //   function, parameters may have incomplete type and may use the [*]
3844    //   notation in their sequences of declarator specifiers to specify
3845    //   variable length array types.
3846    QualType PType = Param->getOriginalType();
3847    if (const ArrayType *AT = Context.getAsArrayType(PType)) {
3848      if (AT->getSizeModifier() == ArrayType::Star) {
3849        // FIXME: This diagnosic should point the the '[*]' if source-location
3850        // information is added for it.
3851        Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
3852      }
3853    }
3854  }
3855
3856  return HasInvalidParm;
3857}
3858
3859/// CheckCastAlign - Implements -Wcast-align, which warns when a
3860/// pointer cast increases the alignment requirements.
3861void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
3862  // This is actually a lot of work to potentially be doing on every
3863  // cast; don't do it if we're ignoring -Wcast_align (as is the default).
3864  if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
3865                                          TRange.getBegin())
3866        == DiagnosticsEngine::Ignored)
3867    return;
3868
3869  // Ignore dependent types.
3870  if (T->isDependentType() || Op->getType()->isDependentType())
3871    return;
3872
3873  // Require that the destination be a pointer type.
3874  const PointerType *DestPtr = T->getAs<PointerType>();
3875  if (!DestPtr) return;
3876
3877  // If the destination has alignment 1, we're done.
3878  QualType DestPointee = DestPtr->getPointeeType();
3879  if (DestPointee->isIncompleteType()) return;
3880  CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
3881  if (DestAlign.isOne()) return;
3882
3883  // Require that the source be a pointer type.
3884  const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
3885  if (!SrcPtr) return;
3886  QualType SrcPointee = SrcPtr->getPointeeType();
3887
3888  // Whitelist casts from cv void*.  We already implicitly
3889  // whitelisted casts to cv void*, since they have alignment 1.
3890  // Also whitelist casts involving incomplete types, which implicitly
3891  // includes 'void'.
3892  if (SrcPointee->isIncompleteType()) return;
3893
3894  CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
3895  if (SrcAlign >= DestAlign) return;
3896
3897  Diag(TRange.getBegin(), diag::warn_cast_align)
3898    << Op->getType() << T
3899    << static_cast<unsigned>(SrcAlign.getQuantity())
3900    << static_cast<unsigned>(DestAlign.getQuantity())
3901    << TRange << Op->getSourceRange();
3902}
3903
3904static const Type* getElementType(const Expr *BaseExpr) {
3905  const Type* EltType = BaseExpr->getType().getTypePtr();
3906  if (EltType->isAnyPointerType())
3907    return EltType->getPointeeType().getTypePtr();
3908  else if (EltType->isArrayType())
3909    return EltType->getBaseElementTypeUnsafe();
3910  return EltType;
3911}
3912
3913/// \brief Check whether this array fits the idiom of a size-one tail padded
3914/// array member of a struct.
3915///
3916/// We avoid emitting out-of-bounds access warnings for such arrays as they are
3917/// commonly used to emulate flexible arrays in C89 code.
3918static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size,
3919                                    const NamedDecl *ND) {
3920  if (Size != 1 || !ND) return false;
3921
3922  const FieldDecl *FD = dyn_cast<FieldDecl>(ND);
3923  if (!FD) return false;
3924
3925  // Don't consider sizes resulting from macro expansions or template argument
3926  // substitution to form C89 tail-padded arrays.
3927  ConstantArrayTypeLoc TL =
3928    cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc());
3929  const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr());
3930  if (!SizeExpr || SizeExpr->getExprLoc().isMacroID())
3931    return false;
3932
3933  const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext());
3934  if (!RD || !RD->isStruct())
3935    return false;
3936
3937  // See if this is the last field decl in the record.
3938  const Decl *D = FD;
3939  while ((D = D->getNextDeclInContext()))
3940    if (isa<FieldDecl>(D))
3941      return false;
3942  return true;
3943}
3944
3945void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr,
3946                            bool isSubscript, bool AllowOnePastEnd) {
3947  const Type* EffectiveType = getElementType(BaseExpr);
3948  BaseExpr = BaseExpr->IgnoreParenCasts();
3949  IndexExpr = IndexExpr->IgnoreParenCasts();
3950
3951  const ConstantArrayType *ArrayTy =
3952    Context.getAsConstantArrayType(BaseExpr->getType());
3953  if (!ArrayTy)
3954    return;
3955
3956  if (IndexExpr->isValueDependent())
3957    return;
3958  llvm::APSInt index;
3959  if (!IndexExpr->isIntegerConstantExpr(index, Context))
3960    return;
3961
3962  const NamedDecl *ND = NULL;
3963  if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
3964    ND = dyn_cast<NamedDecl>(DRE->getDecl());
3965  if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
3966    ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
3967
3968  if (index.isUnsigned() || !index.isNegative()) {
3969    llvm::APInt size = ArrayTy->getSize();
3970    if (!size.isStrictlyPositive())
3971      return;
3972
3973    const Type* BaseType = getElementType(BaseExpr);
3974    if (BaseType != EffectiveType) {
3975      // Make sure we're comparing apples to apples when comparing index to size
3976      uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType);
3977      uint64_t array_typesize = Context.getTypeSize(BaseType);
3978      // Handle ptrarith_typesize being zero, such as when casting to void*
3979      if (!ptrarith_typesize) ptrarith_typesize = 1;
3980      if (ptrarith_typesize != array_typesize) {
3981        // There's a cast to a different size type involved
3982        uint64_t ratio = array_typesize / ptrarith_typesize;
3983        // TODO: Be smarter about handling cases where array_typesize is not a
3984        // multiple of ptrarith_typesize
3985        if (ptrarith_typesize * ratio == array_typesize)
3986          size *= llvm::APInt(size.getBitWidth(), ratio);
3987      }
3988    }
3989
3990    if (size.getBitWidth() > index.getBitWidth())
3991      index = index.sext(size.getBitWidth());
3992    else if (size.getBitWidth() < index.getBitWidth())
3993      size = size.sext(index.getBitWidth());
3994
3995    // For array subscripting the index must be less than size, but for pointer
3996    // arithmetic also allow the index (offset) to be equal to size since
3997    // computing the next address after the end of the array is legal and
3998    // commonly done e.g. in C++ iterators and range-based for loops.
3999    if (AllowOnePastEnd ? index.sle(size) : index.slt(size))
4000      return;
4001
4002    // Also don't warn for arrays of size 1 which are members of some
4003    // structure. These are often used to approximate flexible arrays in C89
4004    // code.
4005    if (IsTailPaddedMemberArray(*this, size, ND))
4006      return;
4007
4008    unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds;
4009    if (isSubscript)
4010      DiagID = diag::warn_array_index_exceeds_bounds;
4011
4012    DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
4013                        PDiag(DiagID) << index.toString(10, true)
4014                          << size.toString(10, true)
4015                          << (unsigned)size.getLimitedValue(~0U)
4016                          << IndexExpr->getSourceRange());
4017  } else {
4018    unsigned DiagID = diag::warn_array_index_precedes_bounds;
4019    if (!isSubscript) {
4020      DiagID = diag::warn_ptr_arith_precedes_bounds;
4021      if (index.isNegative()) index = -index;
4022    }
4023
4024    DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
4025                        PDiag(DiagID) << index.toString(10, true)
4026                          << IndexExpr->getSourceRange());
4027  }
4028
4029  if (ND)
4030    DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
4031                        PDiag(diag::note_array_index_out_of_bounds)
4032                          << ND->getDeclName());
4033}
4034
4035void Sema::CheckArrayAccess(const Expr *expr) {
4036  int AllowOnePastEnd = 0;
4037  while (expr) {
4038    expr = expr->IgnoreParenImpCasts();
4039    switch (expr->getStmtClass()) {
4040      case Stmt::ArraySubscriptExprClass: {
4041        const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr);
4042        CheckArrayAccess(ASE->getBase(), ASE->getIdx(), true,
4043                         AllowOnePastEnd > 0);
4044        return;
4045      }
4046      case Stmt::UnaryOperatorClass: {
4047        // Only unwrap the * and & unary operators
4048        const UnaryOperator *UO = cast<UnaryOperator>(expr);
4049        expr = UO->getSubExpr();
4050        switch (UO->getOpcode()) {
4051          case UO_AddrOf:
4052            AllowOnePastEnd++;
4053            break;
4054          case UO_Deref:
4055            AllowOnePastEnd--;
4056            break;
4057          default:
4058            return;
4059        }
4060        break;
4061      }
4062      case Stmt::ConditionalOperatorClass: {
4063        const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
4064        if (const Expr *lhs = cond->getLHS())
4065          CheckArrayAccess(lhs);
4066        if (const Expr *rhs = cond->getRHS())
4067          CheckArrayAccess(rhs);
4068        return;
4069      }
4070      default:
4071        return;
4072    }
4073  }
4074}
4075
4076//===--- CHECK: Objective-C retain cycles ----------------------------------//
4077
4078namespace {
4079  struct RetainCycleOwner {
4080    RetainCycleOwner() : Variable(0), Indirect(false) {}
4081    VarDecl *Variable;
4082    SourceRange Range;
4083    SourceLocation Loc;
4084    bool Indirect;
4085
4086    void setLocsFrom(Expr *e) {
4087      Loc = e->getExprLoc();
4088      Range = e->getSourceRange();
4089    }
4090  };
4091}
4092
4093/// Consider whether capturing the given variable can possibly lead to
4094/// a retain cycle.
4095static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) {
4096  // In ARC, it's captured strongly iff the variable has __strong
4097  // lifetime.  In MRR, it's captured strongly if the variable is
4098  // __block and has an appropriate type.
4099  if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
4100    return false;
4101
4102  owner.Variable = var;
4103  owner.setLocsFrom(ref);
4104  return true;
4105}
4106
4107static bool findRetainCycleOwner(Expr *e, RetainCycleOwner &owner) {
4108  while (true) {
4109    e = e->IgnoreParens();
4110    if (CastExpr *cast = dyn_cast<CastExpr>(e)) {
4111      switch (cast->getCastKind()) {
4112      case CK_BitCast:
4113      case CK_LValueBitCast:
4114      case CK_LValueToRValue:
4115      case CK_ARCReclaimReturnedObject:
4116        e = cast->getSubExpr();
4117        continue;
4118
4119      case CK_GetObjCProperty: {
4120        // Bail out if this isn't a strong explicit property.
4121        const ObjCPropertyRefExpr *pre = cast->getSubExpr()->getObjCProperty();
4122        if (pre->isImplicitProperty()) return false;
4123        ObjCPropertyDecl *property = pre->getExplicitProperty();
4124        if (!property->isRetaining() &&
4125            !(property->getPropertyIvarDecl() &&
4126              property->getPropertyIvarDecl()->getType()
4127                .getObjCLifetime() == Qualifiers::OCL_Strong))
4128          return false;
4129
4130        owner.Indirect = true;
4131        e = const_cast<Expr*>(pre->getBase());
4132        continue;
4133      }
4134
4135      default:
4136        return false;
4137      }
4138    }
4139
4140    if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) {
4141      ObjCIvarDecl *ivar = ref->getDecl();
4142      if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
4143        return false;
4144
4145      // Try to find a retain cycle in the base.
4146      if (!findRetainCycleOwner(ref->getBase(), owner))
4147        return false;
4148
4149      if (ref->isFreeIvar()) owner.setLocsFrom(ref);
4150      owner.Indirect = true;
4151      return true;
4152    }
4153
4154    if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) {
4155      VarDecl *var = dyn_cast<VarDecl>(ref->getDecl());
4156      if (!var) return false;
4157      return considerVariable(var, ref, owner);
4158    }
4159
4160    if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) {
4161      owner.Variable = ref->getDecl();
4162      owner.setLocsFrom(ref);
4163      return true;
4164    }
4165
4166    if (MemberExpr *member = dyn_cast<MemberExpr>(e)) {
4167      if (member->isArrow()) return false;
4168
4169      // Don't count this as an indirect ownership.
4170      e = member->getBase();
4171      continue;
4172    }
4173
4174    // Array ivars?
4175
4176    return false;
4177  }
4178}
4179
4180namespace {
4181  struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> {
4182    FindCaptureVisitor(ASTContext &Context, VarDecl *variable)
4183      : EvaluatedExprVisitor<FindCaptureVisitor>(Context),
4184        Variable(variable), Capturer(0) {}
4185
4186    VarDecl *Variable;
4187    Expr *Capturer;
4188
4189    void VisitDeclRefExpr(DeclRefExpr *ref) {
4190      if (ref->getDecl() == Variable && !Capturer)
4191        Capturer = ref;
4192    }
4193
4194    void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) {
4195      if (ref->getDecl() == Variable && !Capturer)
4196        Capturer = ref;
4197    }
4198
4199    void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) {
4200      if (Capturer) return;
4201      Visit(ref->getBase());
4202      if (Capturer && ref->isFreeIvar())
4203        Capturer = ref;
4204    }
4205
4206    void VisitBlockExpr(BlockExpr *block) {
4207      // Look inside nested blocks
4208      if (block->getBlockDecl()->capturesVariable(Variable))
4209        Visit(block->getBlockDecl()->getBody());
4210    }
4211  };
4212}
4213
4214/// Check whether the given argument is a block which captures a
4215/// variable.
4216static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) {
4217  assert(owner.Variable && owner.Loc.isValid());
4218
4219  e = e->IgnoreParenCasts();
4220  BlockExpr *block = dyn_cast<BlockExpr>(e);
4221  if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable))
4222    return 0;
4223
4224  FindCaptureVisitor visitor(S.Context, owner.Variable);
4225  visitor.Visit(block->getBlockDecl()->getBody());
4226  return visitor.Capturer;
4227}
4228
4229static void diagnoseRetainCycle(Sema &S, Expr *capturer,
4230                                RetainCycleOwner &owner) {
4231  assert(capturer);
4232  assert(owner.Variable && owner.Loc.isValid());
4233
4234  S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle)
4235    << owner.Variable << capturer->getSourceRange();
4236  S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner)
4237    << owner.Indirect << owner.Range;
4238}
4239
4240/// Check for a keyword selector that starts with the word 'add' or
4241/// 'set'.
4242static bool isSetterLikeSelector(Selector sel) {
4243  if (sel.isUnarySelector()) return false;
4244
4245  StringRef str = sel.getNameForSlot(0);
4246  while (!str.empty() && str.front() == '_') str = str.substr(1);
4247  if (str.startswith("set") || str.startswith("add"))
4248    str = str.substr(3);
4249  else
4250    return false;
4251
4252  if (str.empty()) return true;
4253  return !islower(str.front());
4254}
4255
4256/// Check a message send to see if it's likely to cause a retain cycle.
4257void Sema::checkRetainCycles(ObjCMessageExpr *msg) {
4258  // Only check instance methods whose selector looks like a setter.
4259  if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector()))
4260    return;
4261
4262  // Try to find a variable that the receiver is strongly owned by.
4263  RetainCycleOwner owner;
4264  if (msg->getReceiverKind() == ObjCMessageExpr::Instance) {
4265    if (!findRetainCycleOwner(msg->getInstanceReceiver(), owner))
4266      return;
4267  } else {
4268    assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance);
4269    owner.Variable = getCurMethodDecl()->getSelfDecl();
4270    owner.Loc = msg->getSuperLoc();
4271    owner.Range = msg->getSuperLoc();
4272  }
4273
4274  // Check whether the receiver is captured by any of the arguments.
4275  for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i)
4276    if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner))
4277      return diagnoseRetainCycle(*this, capturer, owner);
4278}
4279
4280/// Check a property assign to see if it's likely to cause a retain cycle.
4281void Sema::checkRetainCycles(Expr *receiver, Expr *argument) {
4282  RetainCycleOwner owner;
4283  if (!findRetainCycleOwner(receiver, owner))
4284    return;
4285
4286  if (Expr *capturer = findCapturingExpr(*this, argument, owner))
4287    diagnoseRetainCycle(*this, capturer, owner);
4288}
4289
4290bool Sema::checkUnsafeAssigns(SourceLocation Loc,
4291                              QualType LHS, Expr *RHS) {
4292  Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime();
4293  if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone)
4294    return false;
4295  // strip off any implicit cast added to get to the one arc-specific
4296  while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
4297    if (cast->getCastKind() == CK_ARCConsumeObject) {
4298      Diag(Loc, diag::warn_arc_retained_assign)
4299        << (LT == Qualifiers::OCL_ExplicitNone)
4300        << RHS->getSourceRange();
4301      return true;
4302    }
4303    RHS = cast->getSubExpr();
4304  }
4305  return false;
4306}
4307
4308void Sema::checkUnsafeExprAssigns(SourceLocation Loc,
4309                              Expr *LHS, Expr *RHS) {
4310  QualType LHSType = LHS->getType();
4311  if (checkUnsafeAssigns(Loc, LHSType, RHS))
4312    return;
4313  Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime();
4314  // FIXME. Check for other life times.
4315  if (LT != Qualifiers::OCL_None)
4316    return;
4317
4318  if (ObjCPropertyRefExpr *PRE = dyn_cast<ObjCPropertyRefExpr>(LHS)) {
4319    if (PRE->isImplicitProperty())
4320      return;
4321    const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
4322    if (!PD)
4323      return;
4324
4325    unsigned Attributes = PD->getPropertyAttributes();
4326    if (Attributes & ObjCPropertyDecl::OBJC_PR_assign)
4327      while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
4328        if (cast->getCastKind() == CK_ARCConsumeObject) {
4329          Diag(Loc, diag::warn_arc_retained_property_assign)
4330          << RHS->getSourceRange();
4331          return;
4332        }
4333        RHS = cast->getSubExpr();
4334      }
4335  }
4336}
4337