SemaChecking.cpp revision 5e9ebb3c0fb554d9285aa99c470abdf283272bd9
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/Sema.h"
16#include "clang/Sema/SemaInternal.h"
17#include "clang/Sema/ScopeInfo.h"
18#include "clang/Analysis/Analyses/FormatString.h"
19#include "clang/AST/ASTContext.h"
20#include "clang/AST/CharUnits.h"
21#include "clang/AST/DeclCXX.h"
22#include "clang/AST/DeclObjC.h"
23#include "clang/AST/ExprCXX.h"
24#include "clang/AST/ExprObjC.h"
25#include "clang/AST/DeclObjC.h"
26#include "clang/AST/StmtCXX.h"
27#include "clang/AST/StmtObjC.h"
28#include "clang/Lex/Preprocessor.h"
29#include "llvm/ADT/BitVector.h"
30#include "llvm/ADT/STLExtras.h"
31#include "llvm/Support/raw_ostream.h"
32#include "clang/Basic/TargetBuiltins.h"
33#include "clang/Basic/TargetInfo.h"
34#include "clang/Basic/ConvertUTF.h"
35#include <limits>
36using namespace clang;
37using namespace sema;
38
39SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL,
40                                                    unsigned ByteNo) const {
41  return SL->getLocationOfByte(ByteNo, PP.getSourceManager(),
42                               PP.getLangOptions(), PP.getTargetInfo());
43}
44
45
46/// CheckablePrintfAttr - does a function call have a "printf" attribute
47/// and arguments that merit checking?
48bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) {
49  if (Format->getType() == "printf") return true;
50  if (Format->getType() == "printf0") {
51    // printf0 allows null "format" string; if so don't check format/args
52    unsigned format_idx = Format->getFormatIdx() - 1;
53    // Does the index refer to the implicit object argument?
54    if (isa<CXXMemberCallExpr>(TheCall)) {
55      if (format_idx == 0)
56        return false;
57      --format_idx;
58    }
59    if (format_idx < TheCall->getNumArgs()) {
60      Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts();
61      if (!Format->isNullPointerConstant(Context,
62                                         Expr::NPC_ValueDependentIsNull))
63        return true;
64    }
65  }
66  return false;
67}
68
69/// Checks that a call expression's argument count is the desired number.
70/// This is useful when doing custom type-checking.  Returns true on error.
71static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) {
72  unsigned argCount = call->getNumArgs();
73  if (argCount == desiredArgCount) return false;
74
75  if (argCount < desiredArgCount)
76    return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args)
77        << 0 /*function call*/ << desiredArgCount << argCount
78        << call->getSourceRange();
79
80  // Highlight all the excess arguments.
81  SourceRange range(call->getArg(desiredArgCount)->getLocStart(),
82                    call->getArg(argCount - 1)->getLocEnd());
83
84  return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args)
85    << 0 /*function call*/ << desiredArgCount << argCount
86    << call->getArg(1)->getSourceRange();
87}
88
89ExprResult
90Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
91  ExprResult TheCallResult(Owned(TheCall));
92
93  // Find out if any arguments are required to be integer constant expressions.
94  unsigned ICEArguments = 0;
95  ASTContext::GetBuiltinTypeError Error;
96  Context.GetBuiltinType(BuiltinID, Error, &ICEArguments);
97  if (Error != ASTContext::GE_None)
98    ICEArguments = 0;  // Don't diagnose previously diagnosed errors.
99
100  // If any arguments are required to be ICE's, check and diagnose.
101  for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) {
102    // Skip arguments not required to be ICE's.
103    if ((ICEArguments & (1 << ArgNo)) == 0) continue;
104
105    llvm::APSInt Result;
106    if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
107      return true;
108    ICEArguments &= ~(1 << ArgNo);
109  }
110
111  switch (BuiltinID) {
112  case Builtin::BI__builtin___CFStringMakeConstantString:
113    assert(TheCall->getNumArgs() == 1 &&
114           "Wrong # arguments to builtin CFStringMakeConstantString");
115    if (CheckObjCString(TheCall->getArg(0)))
116      return ExprError();
117    break;
118  case Builtin::BI__builtin_stdarg_start:
119  case Builtin::BI__builtin_va_start:
120    if (SemaBuiltinVAStart(TheCall))
121      return ExprError();
122    break;
123  case Builtin::BI__builtin_isgreater:
124  case Builtin::BI__builtin_isgreaterequal:
125  case Builtin::BI__builtin_isless:
126  case Builtin::BI__builtin_islessequal:
127  case Builtin::BI__builtin_islessgreater:
128  case Builtin::BI__builtin_isunordered:
129    if (SemaBuiltinUnorderedCompare(TheCall))
130      return ExprError();
131    break;
132  case Builtin::BI__builtin_fpclassify:
133    if (SemaBuiltinFPClassification(TheCall, 6))
134      return ExprError();
135    break;
136  case Builtin::BI__builtin_isfinite:
137  case Builtin::BI__builtin_isinf:
138  case Builtin::BI__builtin_isinf_sign:
139  case Builtin::BI__builtin_isnan:
140  case Builtin::BI__builtin_isnormal:
141    if (SemaBuiltinFPClassification(TheCall, 1))
142      return ExprError();
143    break;
144  case Builtin::BI__builtin_shufflevector:
145    return SemaBuiltinShuffleVector(TheCall);
146    // TheCall will be freed by the smart pointer here, but that's fine, since
147    // SemaBuiltinShuffleVector guts it, but then doesn't release it.
148  case Builtin::BI__builtin_prefetch:
149    if (SemaBuiltinPrefetch(TheCall))
150      return ExprError();
151    break;
152  case Builtin::BI__builtin_object_size:
153    if (SemaBuiltinObjectSize(TheCall))
154      return ExprError();
155    break;
156  case Builtin::BI__builtin_longjmp:
157    if (SemaBuiltinLongjmp(TheCall))
158      return ExprError();
159    break;
160
161  case Builtin::BI__builtin_classify_type:
162    if (checkArgCount(*this, TheCall, 1)) return true;
163    TheCall->setType(Context.IntTy);
164    break;
165  case Builtin::BI__builtin_constant_p:
166    if (checkArgCount(*this, TheCall, 1)) return true;
167    TheCall->setType(Context.IntTy);
168    break;
169  case Builtin::BI__sync_fetch_and_add:
170  case Builtin::BI__sync_fetch_and_sub:
171  case Builtin::BI__sync_fetch_and_or:
172  case Builtin::BI__sync_fetch_and_and:
173  case Builtin::BI__sync_fetch_and_xor:
174  case Builtin::BI__sync_add_and_fetch:
175  case Builtin::BI__sync_sub_and_fetch:
176  case Builtin::BI__sync_and_and_fetch:
177  case Builtin::BI__sync_or_and_fetch:
178  case Builtin::BI__sync_xor_and_fetch:
179  case Builtin::BI__sync_val_compare_and_swap:
180  case Builtin::BI__sync_bool_compare_and_swap:
181  case Builtin::BI__sync_lock_test_and_set:
182  case Builtin::BI__sync_lock_release:
183  case Builtin::BI__sync_swap:
184    return SemaBuiltinAtomicOverloaded(move(TheCallResult));
185  }
186
187  // Since the target specific builtins for each arch overlap, only check those
188  // of the arch we are compiling for.
189  if (BuiltinID >= Builtin::FirstTSBuiltin) {
190    switch (Context.Target.getTriple().getArch()) {
191      case llvm::Triple::arm:
192      case llvm::Triple::thumb:
193        if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall))
194          return ExprError();
195        break;
196      default:
197        break;
198    }
199  }
200
201  return move(TheCallResult);
202}
203
204// Get the valid immediate range for the specified NEON type code.
205static unsigned RFT(unsigned t, bool shift = false) {
206  bool quad = t & 0x10;
207
208  switch (t & 0x7) {
209    case 0: // i8
210      return shift ? 7 : (8 << (int)quad) - 1;
211    case 1: // i16
212      return shift ? 15 : (4 << (int)quad) - 1;
213    case 2: // i32
214      return shift ? 31 : (2 << (int)quad) - 1;
215    case 3: // i64
216      return shift ? 63 : (1 << (int)quad) - 1;
217    case 4: // f32
218      assert(!shift && "cannot shift float types!");
219      return (2 << (int)quad) - 1;
220    case 5: // poly8
221      return shift ? 7 : (8 << (int)quad) - 1;
222    case 6: // poly16
223      return shift ? 15 : (4 << (int)quad) - 1;
224    case 7: // float16
225      assert(!shift && "cannot shift float types!");
226      return (4 << (int)quad) - 1;
227  }
228  return 0;
229}
230
231bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
232  llvm::APSInt Result;
233
234  unsigned mask = 0;
235  unsigned TV = 0;
236  switch (BuiltinID) {
237#define GET_NEON_OVERLOAD_CHECK
238#include "clang/Basic/arm_neon.inc"
239#undef GET_NEON_OVERLOAD_CHECK
240  }
241
242  // For NEON intrinsics which are overloaded on vector element type, validate
243  // the immediate which specifies which variant to emit.
244  if (mask) {
245    unsigned ArgNo = TheCall->getNumArgs()-1;
246    if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
247      return true;
248
249    TV = Result.getLimitedValue(32);
250    if ((TV > 31) || (mask & (1 << TV)) == 0)
251      return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code)
252        << TheCall->getArg(ArgNo)->getSourceRange();
253  }
254
255  // For NEON intrinsics which take an immediate value as part of the
256  // instruction, range check them here.
257  unsigned i = 0, l = 0, u = 0;
258  switch (BuiltinID) {
259  default: return false;
260  case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break;
261  case ARM::BI__builtin_arm_usat: i = 1; u = 31; break;
262  case ARM::BI__builtin_arm_vcvtr_f:
263  case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break;
264#define GET_NEON_IMMEDIATE_CHECK
265#include "clang/Basic/arm_neon.inc"
266#undef GET_NEON_IMMEDIATE_CHECK
267  };
268
269  // Check that the immediate argument is actually a constant.
270  if (SemaBuiltinConstantArg(TheCall, i, Result))
271    return true;
272
273  // Range check against the upper/lower values for this isntruction.
274  unsigned Val = Result.getZExtValue();
275  if (Val < l || Val > (u + l))
276    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
277      << l << u+l << TheCall->getArg(i)->getSourceRange();
278
279  // FIXME: VFP Intrinsics should error if VFP not present.
280  return false;
281}
282
283/// CheckFunctionCall - Check a direct function call for various correctness
284/// and safety properties not strictly enforced by the C type system.
285bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) {
286  // Get the IdentifierInfo* for the called function.
287  IdentifierInfo *FnInfo = FDecl->getIdentifier();
288
289  // None of the checks below are needed for functions that don't have
290  // simple names (e.g., C++ conversion functions).
291  if (!FnInfo)
292    return false;
293
294  // FIXME: This mechanism should be abstracted to be less fragile and
295  // more efficient. For example, just map function ids to custom
296  // handlers.
297
298  // Printf and scanf checking.
299  for (specific_attr_iterator<FormatAttr>
300         i = FDecl->specific_attr_begin<FormatAttr>(),
301         e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) {
302
303    const FormatAttr *Format = *i;
304    const bool b = Format->getType() == "scanf";
305    if (b || CheckablePrintfAttr(Format, TheCall)) {
306      bool HasVAListArg = Format->getFirstArg() == 0;
307      CheckPrintfScanfArguments(TheCall, HasVAListArg,
308                                Format->getFormatIdx() - 1,
309                                HasVAListArg ? 0 : Format->getFirstArg() - 1,
310                                !b);
311    }
312  }
313
314  for (specific_attr_iterator<NonNullAttr>
315         i = FDecl->specific_attr_begin<NonNullAttr>(),
316         e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) {
317    CheckNonNullArguments(*i, TheCall->getArgs(),
318                          TheCall->getCallee()->getLocStart());
319  }
320
321  // Memset/memcpy/memmove handling
322  if (FDecl->getLinkage() == ExternalLinkage &&
323      (!getLangOptions().CPlusPlus || FDecl->isExternC())) {
324    if (FnInfo->isStr("memset") || FnInfo->isStr("memcpy") ||
325        FnInfo->isStr("memmove"))
326      CheckMemsetcpymoveArguments(TheCall, FnInfo);
327  }
328
329  return false;
330}
331
332bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
333  // Printf checking.
334  const FormatAttr *Format = NDecl->getAttr<FormatAttr>();
335  if (!Format)
336    return false;
337
338  const VarDecl *V = dyn_cast<VarDecl>(NDecl);
339  if (!V)
340    return false;
341
342  QualType Ty = V->getType();
343  if (!Ty->isBlockPointerType())
344    return false;
345
346  const bool b = Format->getType() == "scanf";
347  if (!b && !CheckablePrintfAttr(Format, TheCall))
348    return false;
349
350  bool HasVAListArg = Format->getFirstArg() == 0;
351  CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
352                            HasVAListArg ? 0 : Format->getFirstArg() - 1, !b);
353
354  return false;
355}
356
357/// SemaBuiltinAtomicOverloaded - We have a call to a function like
358/// __sync_fetch_and_add, which is an overloaded function based on the pointer
359/// type of its first argument.  The main ActOnCallExpr routines have already
360/// promoted the types of arguments because all of these calls are prototyped as
361/// void(...).
362///
363/// This function goes through and does final semantic checking for these
364/// builtins,
365ExprResult
366Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
367  CallExpr *TheCall = (CallExpr *)TheCallResult.get();
368  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
369  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
370
371  // Ensure that we have at least one argument to do type inference from.
372  if (TheCall->getNumArgs() < 1) {
373    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
374      << 0 << 1 << TheCall->getNumArgs()
375      << TheCall->getCallee()->getSourceRange();
376    return ExprError();
377  }
378
379  // Inspect the first argument of the atomic builtin.  This should always be
380  // a pointer type, whose element is an integral scalar or pointer type.
381  // Because it is a pointer type, we don't have to worry about any implicit
382  // casts here.
383  // FIXME: We don't allow floating point scalars as input.
384  Expr *FirstArg = TheCall->getArg(0);
385  if (!FirstArg->getType()->isPointerType()) {
386    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
387      << FirstArg->getType() << FirstArg->getSourceRange();
388    return ExprError();
389  }
390
391  QualType ValType =
392    FirstArg->getType()->getAs<PointerType>()->getPointeeType();
393  if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
394      !ValType->isBlockPointerType()) {
395    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
396      << FirstArg->getType() << FirstArg->getSourceRange();
397    return ExprError();
398  }
399
400  // The majority of builtins return a value, but a few have special return
401  // types, so allow them to override appropriately below.
402  QualType ResultType = ValType;
403
404  // We need to figure out which concrete builtin this maps onto.  For example,
405  // __sync_fetch_and_add with a 2 byte object turns into
406  // __sync_fetch_and_add_2.
407#define BUILTIN_ROW(x) \
408  { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
409    Builtin::BI##x##_8, Builtin::BI##x##_16 }
410
411  static const unsigned BuiltinIndices[][5] = {
412    BUILTIN_ROW(__sync_fetch_and_add),
413    BUILTIN_ROW(__sync_fetch_and_sub),
414    BUILTIN_ROW(__sync_fetch_and_or),
415    BUILTIN_ROW(__sync_fetch_and_and),
416    BUILTIN_ROW(__sync_fetch_and_xor),
417
418    BUILTIN_ROW(__sync_add_and_fetch),
419    BUILTIN_ROW(__sync_sub_and_fetch),
420    BUILTIN_ROW(__sync_and_and_fetch),
421    BUILTIN_ROW(__sync_or_and_fetch),
422    BUILTIN_ROW(__sync_xor_and_fetch),
423
424    BUILTIN_ROW(__sync_val_compare_and_swap),
425    BUILTIN_ROW(__sync_bool_compare_and_swap),
426    BUILTIN_ROW(__sync_lock_test_and_set),
427    BUILTIN_ROW(__sync_lock_release),
428    BUILTIN_ROW(__sync_swap)
429  };
430#undef BUILTIN_ROW
431
432  // Determine the index of the size.
433  unsigned SizeIndex;
434  switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
435  case 1: SizeIndex = 0; break;
436  case 2: SizeIndex = 1; break;
437  case 4: SizeIndex = 2; break;
438  case 8: SizeIndex = 3; break;
439  case 16: SizeIndex = 4; break;
440  default:
441    Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
442      << FirstArg->getType() << FirstArg->getSourceRange();
443    return ExprError();
444  }
445
446  // Each of these builtins has one pointer argument, followed by some number of
447  // values (0, 1 or 2) followed by a potentially empty varags list of stuff
448  // that we ignore.  Find out which row of BuiltinIndices to read from as well
449  // as the number of fixed args.
450  unsigned BuiltinID = FDecl->getBuiltinID();
451  unsigned BuiltinIndex, NumFixed = 1;
452  switch (BuiltinID) {
453  default: assert(0 && "Unknown overloaded atomic builtin!");
454  case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break;
455  case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break;
456  case Builtin::BI__sync_fetch_and_or:  BuiltinIndex = 2; break;
457  case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break;
458  case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break;
459
460  case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break;
461  case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break;
462  case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break;
463  case Builtin::BI__sync_or_and_fetch:  BuiltinIndex = 8; break;
464  case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break;
465
466  case Builtin::BI__sync_val_compare_and_swap:
467    BuiltinIndex = 10;
468    NumFixed = 2;
469    break;
470  case Builtin::BI__sync_bool_compare_and_swap:
471    BuiltinIndex = 11;
472    NumFixed = 2;
473    ResultType = Context.BoolTy;
474    break;
475  case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break;
476  case Builtin::BI__sync_lock_release:
477    BuiltinIndex = 13;
478    NumFixed = 0;
479    ResultType = Context.VoidTy;
480    break;
481  case Builtin::BI__sync_swap: BuiltinIndex = 14; break;
482  }
483
484  // Now that we know how many fixed arguments we expect, first check that we
485  // have at least that many.
486  if (TheCall->getNumArgs() < 1+NumFixed) {
487    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
488      << 0 << 1+NumFixed << TheCall->getNumArgs()
489      << TheCall->getCallee()->getSourceRange();
490    return ExprError();
491  }
492
493  // Get the decl for the concrete builtin from this, we can tell what the
494  // concrete integer type we should convert to is.
495  unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
496  const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
497  IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
498  FunctionDecl *NewBuiltinDecl =
499    cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
500                                           TUScope, false, DRE->getLocStart()));
501
502  // The first argument --- the pointer --- has a fixed type; we
503  // deduce the types of the rest of the arguments accordingly.  Walk
504  // the remaining arguments, converting them to the deduced value type.
505  for (unsigned i = 0; i != NumFixed; ++i) {
506    ExprResult Arg = TheCall->getArg(i+1);
507
508    // If the argument is an implicit cast, then there was a promotion due to
509    // "...", just remove it now.
510    if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) {
511      Arg = ICE->getSubExpr();
512      ICE->setSubExpr(0);
513      TheCall->setArg(i+1, Arg.get());
514    }
515
516    // GCC does an implicit conversion to the pointer or integer ValType.  This
517    // can fail in some cases (1i -> int**), check for this error case now.
518    CastKind Kind = CK_Invalid;
519    ExprValueKind VK = VK_RValue;
520    CXXCastPath BasePath;
521    Arg = CheckCastTypes(Arg.get()->getSourceRange(), ValType, Arg.take(), Kind, VK, BasePath);
522    if (Arg.isInvalid())
523      return ExprError();
524
525    // Okay, we have something that *can* be converted to the right type.  Check
526    // to see if there is a potentially weird extension going on here.  This can
527    // happen when you do an atomic operation on something like an char* and
528    // pass in 42.  The 42 gets converted to char.  This is even more strange
529    // for things like 45.123 -> char, etc.
530    // FIXME: Do this check.
531    Arg = ImpCastExprToType(Arg.take(), ValType, Kind, VK, &BasePath);
532    TheCall->setArg(i+1, Arg.get());
533  }
534
535  // Switch the DeclRefExpr to refer to the new decl.
536  DRE->setDecl(NewBuiltinDecl);
537  DRE->setType(NewBuiltinDecl->getType());
538
539  // Set the callee in the CallExpr.
540  // FIXME: This leaks the original parens and implicit casts.
541  ExprResult PromotedCall = UsualUnaryConversions(DRE);
542  if (PromotedCall.isInvalid())
543    return ExprError();
544  TheCall->setCallee(PromotedCall.take());
545
546  // Change the result type of the call to match the original value type. This
547  // is arbitrary, but the codegen for these builtins ins design to handle it
548  // gracefully.
549  TheCall->setType(ResultType);
550
551  return move(TheCallResult);
552}
553
554
555/// CheckObjCString - Checks that the argument to the builtin
556/// CFString constructor is correct
557/// Note: It might also make sense to do the UTF-16 conversion here (would
558/// simplify the backend).
559bool Sema::CheckObjCString(Expr *Arg) {
560  Arg = Arg->IgnoreParenCasts();
561  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
562
563  if (!Literal || Literal->isWide()) {
564    Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
565      << Arg->getSourceRange();
566    return true;
567  }
568
569  if (Literal->containsNonAsciiOrNull()) {
570    llvm::StringRef String = Literal->getString();
571    unsigned NumBytes = String.size();
572    llvm::SmallVector<UTF16, 128> ToBuf(NumBytes);
573    const UTF8 *FromPtr = (UTF8 *)String.data();
574    UTF16 *ToPtr = &ToBuf[0];
575
576    ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
577                                                 &ToPtr, ToPtr + NumBytes,
578                                                 strictConversion);
579    // Check for conversion failure.
580    if (Result != conversionOK)
581      Diag(Arg->getLocStart(),
582           diag::warn_cfstring_truncated) << Arg->getSourceRange();
583  }
584  return false;
585}
586
587/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
588/// Emit an error and return true on failure, return false on success.
589bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
590  Expr *Fn = TheCall->getCallee();
591  if (TheCall->getNumArgs() > 2) {
592    Diag(TheCall->getArg(2)->getLocStart(),
593         diag::err_typecheck_call_too_many_args)
594      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
595      << Fn->getSourceRange()
596      << SourceRange(TheCall->getArg(2)->getLocStart(),
597                     (*(TheCall->arg_end()-1))->getLocEnd());
598    return true;
599  }
600
601  if (TheCall->getNumArgs() < 2) {
602    return Diag(TheCall->getLocEnd(),
603      diag::err_typecheck_call_too_few_args_at_least)
604      << 0 /*function call*/ << 2 << TheCall->getNumArgs();
605  }
606
607  // Determine whether the current function is variadic or not.
608  BlockScopeInfo *CurBlock = getCurBlock();
609  bool isVariadic;
610  if (CurBlock)
611    isVariadic = CurBlock->TheDecl->isVariadic();
612  else if (FunctionDecl *FD = getCurFunctionDecl())
613    isVariadic = FD->isVariadic();
614  else
615    isVariadic = getCurMethodDecl()->isVariadic();
616
617  if (!isVariadic) {
618    Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
619    return true;
620  }
621
622  // Verify that the second argument to the builtin is the last argument of the
623  // current function or method.
624  bool SecondArgIsLastNamedArgument = false;
625  const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
626
627  if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
628    if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
629      // FIXME: This isn't correct for methods (results in bogus warning).
630      // Get the last formal in the current function.
631      const ParmVarDecl *LastArg;
632      if (CurBlock)
633        LastArg = *(CurBlock->TheDecl->param_end()-1);
634      else if (FunctionDecl *FD = getCurFunctionDecl())
635        LastArg = *(FD->param_end()-1);
636      else
637        LastArg = *(getCurMethodDecl()->param_end()-1);
638      SecondArgIsLastNamedArgument = PV == LastArg;
639    }
640  }
641
642  if (!SecondArgIsLastNamedArgument)
643    Diag(TheCall->getArg(1)->getLocStart(),
644         diag::warn_second_parameter_of_va_start_not_last_named_argument);
645  return false;
646}
647
648/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
649/// friends.  This is declared to take (...), so we have to check everything.
650bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
651  if (TheCall->getNumArgs() < 2)
652    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
653      << 0 << 2 << TheCall->getNumArgs()/*function call*/;
654  if (TheCall->getNumArgs() > 2)
655    return Diag(TheCall->getArg(2)->getLocStart(),
656                diag::err_typecheck_call_too_many_args)
657      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
658      << SourceRange(TheCall->getArg(2)->getLocStart(),
659                     (*(TheCall->arg_end()-1))->getLocEnd());
660
661  ExprResult OrigArg0 = TheCall->getArg(0);
662  ExprResult OrigArg1 = TheCall->getArg(1);
663
664  // Do standard promotions between the two arguments, returning their common
665  // type.
666  QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
667  if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
668    return true;
669
670  // Make sure any conversions are pushed back into the call; this is
671  // type safe since unordered compare builtins are declared as "_Bool
672  // foo(...)".
673  TheCall->setArg(0, OrigArg0.get());
674  TheCall->setArg(1, OrigArg1.get());
675
676  if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
677    return false;
678
679  // If the common type isn't a real floating type, then the arguments were
680  // invalid for this operation.
681  if (!Res->isRealFloatingType())
682    return Diag(OrigArg0.get()->getLocStart(),
683                diag::err_typecheck_call_invalid_ordered_compare)
684      << OrigArg0.get()->getType() << OrigArg1.get()->getType()
685      << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
686
687  return false;
688}
689
690/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
691/// __builtin_isnan and friends.  This is declared to take (...), so we have
692/// to check everything. We expect the last argument to be a floating point
693/// value.
694bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
695  if (TheCall->getNumArgs() < NumArgs)
696    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
697      << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
698  if (TheCall->getNumArgs() > NumArgs)
699    return Diag(TheCall->getArg(NumArgs)->getLocStart(),
700                diag::err_typecheck_call_too_many_args)
701      << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
702      << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
703                     (*(TheCall->arg_end()-1))->getLocEnd());
704
705  Expr *OrigArg = TheCall->getArg(NumArgs-1);
706
707  if (OrigArg->isTypeDependent())
708    return false;
709
710  // This operation requires a non-_Complex floating-point number.
711  if (!OrigArg->getType()->isRealFloatingType())
712    return Diag(OrigArg->getLocStart(),
713                diag::err_typecheck_call_invalid_unary_fp)
714      << OrigArg->getType() << OrigArg->getSourceRange();
715
716  // If this is an implicit conversion from float -> double, remove it.
717  if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
718    Expr *CastArg = Cast->getSubExpr();
719    if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
720      assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
721             "promotion from float to double is the only expected cast here");
722      Cast->setSubExpr(0);
723      TheCall->setArg(NumArgs-1, CastArg);
724      OrigArg = CastArg;
725    }
726  }
727
728  return false;
729}
730
731/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
732// This is declared to take (...), so we have to check everything.
733ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
734  if (TheCall->getNumArgs() < 2)
735    return ExprError(Diag(TheCall->getLocEnd(),
736                          diag::err_typecheck_call_too_few_args_at_least)
737      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
738      << TheCall->getSourceRange());
739
740  // Determine which of the following types of shufflevector we're checking:
741  // 1) unary, vector mask: (lhs, mask)
742  // 2) binary, vector mask: (lhs, rhs, mask)
743  // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
744  QualType resType = TheCall->getArg(0)->getType();
745  unsigned numElements = 0;
746
747  if (!TheCall->getArg(0)->isTypeDependent() &&
748      !TheCall->getArg(1)->isTypeDependent()) {
749    QualType LHSType = TheCall->getArg(0)->getType();
750    QualType RHSType = TheCall->getArg(1)->getType();
751
752    if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
753      Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
754        << SourceRange(TheCall->getArg(0)->getLocStart(),
755                       TheCall->getArg(1)->getLocEnd());
756      return ExprError();
757    }
758
759    numElements = LHSType->getAs<VectorType>()->getNumElements();
760    unsigned numResElements = TheCall->getNumArgs() - 2;
761
762    // Check to see if we have a call with 2 vector arguments, the unary shuffle
763    // with mask.  If so, verify that RHS is an integer vector type with the
764    // same number of elts as lhs.
765    if (TheCall->getNumArgs() == 2) {
766      if (!RHSType->hasIntegerRepresentation() ||
767          RHSType->getAs<VectorType>()->getNumElements() != numElements)
768        Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
769          << SourceRange(TheCall->getArg(1)->getLocStart(),
770                         TheCall->getArg(1)->getLocEnd());
771      numResElements = numElements;
772    }
773    else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
774      Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
775        << SourceRange(TheCall->getArg(0)->getLocStart(),
776                       TheCall->getArg(1)->getLocEnd());
777      return ExprError();
778    } else if (numElements != numResElements) {
779      QualType eltType = LHSType->getAs<VectorType>()->getElementType();
780      resType = Context.getVectorType(eltType, numResElements,
781                                      VectorType::GenericVector);
782    }
783  }
784
785  for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
786    if (TheCall->getArg(i)->isTypeDependent() ||
787        TheCall->getArg(i)->isValueDependent())
788      continue;
789
790    llvm::APSInt Result(32);
791    if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
792      return ExprError(Diag(TheCall->getLocStart(),
793                  diag::err_shufflevector_nonconstant_argument)
794                << TheCall->getArg(i)->getSourceRange());
795
796    if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
797      return ExprError(Diag(TheCall->getLocStart(),
798                  diag::err_shufflevector_argument_too_large)
799               << TheCall->getArg(i)->getSourceRange());
800  }
801
802  llvm::SmallVector<Expr*, 32> exprs;
803
804  for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
805    exprs.push_back(TheCall->getArg(i));
806    TheCall->setArg(i, 0);
807  }
808
809  return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
810                                            exprs.size(), resType,
811                                            TheCall->getCallee()->getLocStart(),
812                                            TheCall->getRParenLoc()));
813}
814
815/// SemaBuiltinPrefetch - Handle __builtin_prefetch.
816// This is declared to take (const void*, ...) and can take two
817// optional constant int args.
818bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
819  unsigned NumArgs = TheCall->getNumArgs();
820
821  if (NumArgs > 3)
822    return Diag(TheCall->getLocEnd(),
823             diag::err_typecheck_call_too_many_args_at_most)
824             << 0 /*function call*/ << 3 << NumArgs
825             << TheCall->getSourceRange();
826
827  // Argument 0 is checked for us and the remaining arguments must be
828  // constant integers.
829  for (unsigned i = 1; i != NumArgs; ++i) {
830    Expr *Arg = TheCall->getArg(i);
831
832    llvm::APSInt Result;
833    if (SemaBuiltinConstantArg(TheCall, i, Result))
834      return true;
835
836    // FIXME: gcc issues a warning and rewrites these to 0. These
837    // seems especially odd for the third argument since the default
838    // is 3.
839    if (i == 1) {
840      if (Result.getLimitedValue() > 1)
841        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
842             << "0" << "1" << Arg->getSourceRange();
843    } else {
844      if (Result.getLimitedValue() > 3)
845        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
846            << "0" << "3" << Arg->getSourceRange();
847    }
848  }
849
850  return false;
851}
852
853/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
854/// TheCall is a constant expression.
855bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
856                                  llvm::APSInt &Result) {
857  Expr *Arg = TheCall->getArg(ArgNum);
858  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
859  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
860
861  if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
862
863  if (!Arg->isIntegerConstantExpr(Result, Context))
864    return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
865                << FDecl->getDeclName() <<  Arg->getSourceRange();
866
867  return false;
868}
869
870/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
871/// int type). This simply type checks that type is one of the defined
872/// constants (0-3).
873// For compatibility check 0-3, llvm only handles 0 and 2.
874bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
875  llvm::APSInt Result;
876
877  // Check constant-ness first.
878  if (SemaBuiltinConstantArg(TheCall, 1, Result))
879    return true;
880
881  Expr *Arg = TheCall->getArg(1);
882  if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
883    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
884             << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
885  }
886
887  return false;
888}
889
890/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
891/// This checks that val is a constant 1.
892bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
893  Expr *Arg = TheCall->getArg(1);
894  llvm::APSInt Result;
895
896  // TODO: This is less than ideal. Overload this to take a value.
897  if (SemaBuiltinConstantArg(TheCall, 1, Result))
898    return true;
899
900  if (Result != 1)
901    return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
902             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
903
904  return false;
905}
906
907// Handle i > 1 ? "x" : "y", recursively.
908bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall,
909                                  bool HasVAListArg,
910                                  unsigned format_idx, unsigned firstDataArg,
911                                  bool isPrintf) {
912 tryAgain:
913  if (E->isTypeDependent() || E->isValueDependent())
914    return false;
915
916  E = E->IgnoreParens();
917
918  switch (E->getStmtClass()) {
919  case Stmt::BinaryConditionalOperatorClass:
920  case Stmt::ConditionalOperatorClass: {
921    const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E);
922    return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg,
923                                  format_idx, firstDataArg, isPrintf)
924        && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg,
925                                  format_idx, firstDataArg, isPrintf);
926  }
927
928  case Stmt::IntegerLiteralClass:
929    // Technically -Wformat-nonliteral does not warn about this case.
930    // The behavior of printf and friends in this case is implementation
931    // dependent.  Ideally if the format string cannot be null then
932    // it should have a 'nonnull' attribute in the function prototype.
933    return true;
934
935  case Stmt::ImplicitCastExprClass: {
936    E = cast<ImplicitCastExpr>(E)->getSubExpr();
937    goto tryAgain;
938  }
939
940  case Stmt::OpaqueValueExprClass:
941    if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
942      E = src;
943      goto tryAgain;
944    }
945    return false;
946
947  case Stmt::PredefinedExprClass:
948    // While __func__, etc., are technically not string literals, they
949    // cannot contain format specifiers and thus are not a security
950    // liability.
951    return true;
952
953  case Stmt::DeclRefExprClass: {
954    const DeclRefExpr *DR = cast<DeclRefExpr>(E);
955
956    // As an exception, do not flag errors for variables binding to
957    // const string literals.
958    if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
959      bool isConstant = false;
960      QualType T = DR->getType();
961
962      if (const ArrayType *AT = Context.getAsArrayType(T)) {
963        isConstant = AT->getElementType().isConstant(Context);
964      } else if (const PointerType *PT = T->getAs<PointerType>()) {
965        isConstant = T.isConstant(Context) &&
966                     PT->getPointeeType().isConstant(Context);
967      }
968
969      if (isConstant) {
970        if (const Expr *Init = VD->getAnyInitializer())
971          return SemaCheckStringLiteral(Init, TheCall,
972                                        HasVAListArg, format_idx, firstDataArg,
973                                        isPrintf);
974      }
975
976      // For vprintf* functions (i.e., HasVAListArg==true), we add a
977      // special check to see if the format string is a function parameter
978      // of the function calling the printf function.  If the function
979      // has an attribute indicating it is a printf-like function, then we
980      // should suppress warnings concerning non-literals being used in a call
981      // to a vprintf function.  For example:
982      //
983      // void
984      // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
985      //      va_list ap;
986      //      va_start(ap, fmt);
987      //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
988      //      ...
989      //
990      //
991      //  FIXME: We don't have full attribute support yet, so just check to see
992      //    if the argument is a DeclRefExpr that references a parameter.  We'll
993      //    add proper support for checking the attribute later.
994      if (HasVAListArg)
995        if (isa<ParmVarDecl>(VD))
996          return true;
997    }
998
999    return false;
1000  }
1001
1002  case Stmt::CallExprClass: {
1003    const CallExpr *CE = cast<CallExpr>(E);
1004    if (const ImplicitCastExpr *ICE
1005          = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
1006      if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
1007        if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
1008          if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
1009            unsigned ArgIndex = FA->getFormatIdx();
1010            const Expr *Arg = CE->getArg(ArgIndex - 1);
1011
1012            return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg,
1013                                          format_idx, firstDataArg, isPrintf);
1014          }
1015        }
1016      }
1017    }
1018
1019    return false;
1020  }
1021  case Stmt::ObjCStringLiteralClass:
1022  case Stmt::StringLiteralClass: {
1023    const StringLiteral *StrE = NULL;
1024
1025    if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1026      StrE = ObjCFExpr->getString();
1027    else
1028      StrE = cast<StringLiteral>(E);
1029
1030    if (StrE) {
1031      CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx,
1032                        firstDataArg, isPrintf);
1033      return true;
1034    }
1035
1036    return false;
1037  }
1038
1039  default:
1040    return false;
1041  }
1042}
1043
1044void
1045Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1046                            const Expr * const *ExprArgs,
1047                            SourceLocation CallSiteLoc) {
1048  for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1049                                  e = NonNull->args_end();
1050       i != e; ++i) {
1051    const Expr *ArgExpr = ExprArgs[*i];
1052    if (ArgExpr->isNullPointerConstant(Context,
1053                                       Expr::NPC_ValueDependentIsNotNull))
1054      Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1055  }
1056}
1057
1058/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar
1059/// functions) for correct use of format strings.
1060void
1061Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg,
1062                                unsigned format_idx, unsigned firstDataArg,
1063                                bool isPrintf) {
1064
1065  const Expr *Fn = TheCall->getCallee();
1066
1067  // The way the format attribute works in GCC, the implicit this argument
1068  // of member functions is counted. However, it doesn't appear in our own
1069  // lists, so decrement format_idx in that case.
1070  if (isa<CXXMemberCallExpr>(TheCall)) {
1071    const CXXMethodDecl *method_decl =
1072      dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl());
1073    if (method_decl && method_decl->isInstance()) {
1074      // Catch a format attribute mistakenly referring to the object argument.
1075      if (format_idx == 0)
1076        return;
1077      --format_idx;
1078      if(firstDataArg != 0)
1079        --firstDataArg;
1080    }
1081  }
1082
1083  // CHECK: printf/scanf-like function is called with no format string.
1084  if (format_idx >= TheCall->getNumArgs()) {
1085    Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string)
1086      << Fn->getSourceRange();
1087    return;
1088  }
1089
1090  const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts();
1091
1092  // CHECK: format string is not a string literal.
1093  //
1094  // Dynamically generated format strings are difficult to
1095  // automatically vet at compile time.  Requiring that format strings
1096  // are string literals: (1) permits the checking of format strings by
1097  // the compiler and thereby (2) can practically remove the source of
1098  // many format string exploits.
1099
1100  // Format string can be either ObjC string (e.g. @"%d") or
1101  // C string (e.g. "%d")
1102  // ObjC string uses the same format specifiers as C string, so we can use
1103  // the same format string checking logic for both ObjC and C strings.
1104  if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx,
1105                             firstDataArg, isPrintf))
1106    return;  // Literal format string found, check done!
1107
1108  // If there are no arguments specified, warn with -Wformat-security, otherwise
1109  // warn only with -Wformat-nonliteral.
1110  if (TheCall->getNumArgs() == format_idx+1)
1111    Diag(TheCall->getArg(format_idx)->getLocStart(),
1112         diag::warn_format_nonliteral_noargs)
1113      << OrigFormatExpr->getSourceRange();
1114  else
1115    Diag(TheCall->getArg(format_idx)->getLocStart(),
1116         diag::warn_format_nonliteral)
1117           << OrigFormatExpr->getSourceRange();
1118}
1119
1120namespace {
1121class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1122protected:
1123  Sema &S;
1124  const StringLiteral *FExpr;
1125  const Expr *OrigFormatExpr;
1126  const unsigned FirstDataArg;
1127  const unsigned NumDataArgs;
1128  const bool IsObjCLiteral;
1129  const char *Beg; // Start of format string.
1130  const bool HasVAListArg;
1131  const CallExpr *TheCall;
1132  unsigned FormatIdx;
1133  llvm::BitVector CoveredArgs;
1134  bool usesPositionalArgs;
1135  bool atFirstArg;
1136public:
1137  CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1138                     const Expr *origFormatExpr, unsigned firstDataArg,
1139                     unsigned numDataArgs, bool isObjCLiteral,
1140                     const char *beg, bool hasVAListArg,
1141                     const CallExpr *theCall, unsigned formatIdx)
1142    : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1143      FirstDataArg(firstDataArg),
1144      NumDataArgs(numDataArgs),
1145      IsObjCLiteral(isObjCLiteral), Beg(beg),
1146      HasVAListArg(hasVAListArg),
1147      TheCall(theCall), FormatIdx(formatIdx),
1148      usesPositionalArgs(false), atFirstArg(true) {
1149        CoveredArgs.resize(numDataArgs);
1150        CoveredArgs.reset();
1151      }
1152
1153  void DoneProcessing();
1154
1155  void HandleIncompleteSpecifier(const char *startSpecifier,
1156                                 unsigned specifierLen);
1157
1158  virtual void HandleInvalidPosition(const char *startSpecifier,
1159                                     unsigned specifierLen,
1160                                     analyze_format_string::PositionContext p);
1161
1162  virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
1163
1164  void HandleNullChar(const char *nullCharacter);
1165
1166protected:
1167  bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
1168                                        const char *startSpec,
1169                                        unsigned specifierLen,
1170                                        const char *csStart, unsigned csLen);
1171
1172  SourceRange getFormatStringRange();
1173  CharSourceRange getSpecifierRange(const char *startSpecifier,
1174                                    unsigned specifierLen);
1175  SourceLocation getLocationOfByte(const char *x);
1176
1177  const Expr *getDataArg(unsigned i) const;
1178
1179  bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
1180                    const analyze_format_string::ConversionSpecifier &CS,
1181                    const char *startSpecifier, unsigned specifierLen,
1182                    unsigned argIndex);
1183};
1184}
1185
1186SourceRange CheckFormatHandler::getFormatStringRange() {
1187  return OrigFormatExpr->getSourceRange();
1188}
1189
1190CharSourceRange CheckFormatHandler::
1191getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
1192  SourceLocation Start = getLocationOfByte(startSpecifier);
1193  SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
1194
1195  // Advance the end SourceLocation by one due to half-open ranges.
1196  End = End.getFileLocWithOffset(1);
1197
1198  return CharSourceRange::getCharRange(Start, End);
1199}
1200
1201SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
1202  return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
1203}
1204
1205void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
1206                                                   unsigned specifierLen){
1207  SourceLocation Loc = getLocationOfByte(startSpecifier);
1208  S.Diag(Loc, diag::warn_printf_incomplete_specifier)
1209    << getSpecifierRange(startSpecifier, specifierLen);
1210}
1211
1212void
1213CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
1214                                     analyze_format_string::PositionContext p) {
1215  SourceLocation Loc = getLocationOfByte(startPos);
1216  S.Diag(Loc, diag::warn_format_invalid_positional_specifier)
1217    << (unsigned) p << getSpecifierRange(startPos, posLen);
1218}
1219
1220void CheckFormatHandler::HandleZeroPosition(const char *startPos,
1221                                            unsigned posLen) {
1222  SourceLocation Loc = getLocationOfByte(startPos);
1223  S.Diag(Loc, diag::warn_format_zero_positional_specifier)
1224    << getSpecifierRange(startPos, posLen);
1225}
1226
1227void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
1228  if (!IsObjCLiteral) {
1229    // The presence of a null character is likely an error.
1230    S.Diag(getLocationOfByte(nullCharacter),
1231           diag::warn_printf_format_string_contains_null_char)
1232      << getFormatStringRange();
1233  }
1234}
1235
1236const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
1237  return TheCall->getArg(FirstDataArg + i);
1238}
1239
1240void CheckFormatHandler::DoneProcessing() {
1241    // Does the number of data arguments exceed the number of
1242    // format conversions in the format string?
1243  if (!HasVAListArg) {
1244      // Find any arguments that weren't covered.
1245    CoveredArgs.flip();
1246    signed notCoveredArg = CoveredArgs.find_first();
1247    if (notCoveredArg >= 0) {
1248      assert((unsigned)notCoveredArg < NumDataArgs);
1249      S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(),
1250             diag::warn_printf_data_arg_not_used)
1251      << getFormatStringRange();
1252    }
1253  }
1254}
1255
1256bool
1257CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
1258                                                     SourceLocation Loc,
1259                                                     const char *startSpec,
1260                                                     unsigned specifierLen,
1261                                                     const char *csStart,
1262                                                     unsigned csLen) {
1263
1264  bool keepGoing = true;
1265  if (argIndex < NumDataArgs) {
1266    // Consider the argument coverered, even though the specifier doesn't
1267    // make sense.
1268    CoveredArgs.set(argIndex);
1269  }
1270  else {
1271    // If argIndex exceeds the number of data arguments we
1272    // don't issue a warning because that is just a cascade of warnings (and
1273    // they may have intended '%%' anyway). We don't want to continue processing
1274    // the format string after this point, however, as we will like just get
1275    // gibberish when trying to match arguments.
1276    keepGoing = false;
1277  }
1278
1279  S.Diag(Loc, diag::warn_format_invalid_conversion)
1280    << llvm::StringRef(csStart, csLen)
1281    << getSpecifierRange(startSpec, specifierLen);
1282
1283  return keepGoing;
1284}
1285
1286bool
1287CheckFormatHandler::CheckNumArgs(
1288  const analyze_format_string::FormatSpecifier &FS,
1289  const analyze_format_string::ConversionSpecifier &CS,
1290  const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
1291
1292  if (argIndex >= NumDataArgs) {
1293    if (FS.usesPositionalArg())  {
1294      S.Diag(getLocationOfByte(CS.getStart()),
1295             diag::warn_printf_positional_arg_exceeds_data_args)
1296      << (argIndex+1) << NumDataArgs
1297      << getSpecifierRange(startSpecifier, specifierLen);
1298    }
1299    else {
1300      S.Diag(getLocationOfByte(CS.getStart()),
1301             diag::warn_printf_insufficient_data_args)
1302      << getSpecifierRange(startSpecifier, specifierLen);
1303    }
1304
1305    return false;
1306  }
1307  return true;
1308}
1309
1310//===--- CHECK: Printf format string checking ------------------------------===//
1311
1312namespace {
1313class CheckPrintfHandler : public CheckFormatHandler {
1314public:
1315  CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
1316                     const Expr *origFormatExpr, unsigned firstDataArg,
1317                     unsigned numDataArgs, bool isObjCLiteral,
1318                     const char *beg, bool hasVAListArg,
1319                     const CallExpr *theCall, unsigned formatIdx)
1320  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1321                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1322                       theCall, formatIdx) {}
1323
1324
1325  bool HandleInvalidPrintfConversionSpecifier(
1326                                      const analyze_printf::PrintfSpecifier &FS,
1327                                      const char *startSpecifier,
1328                                      unsigned specifierLen);
1329
1330  bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
1331                             const char *startSpecifier,
1332                             unsigned specifierLen);
1333
1334  bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
1335                    const char *startSpecifier, unsigned specifierLen);
1336  void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
1337                           const analyze_printf::OptionalAmount &Amt,
1338                           unsigned type,
1339                           const char *startSpecifier, unsigned specifierLen);
1340  void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1341                  const analyze_printf::OptionalFlag &flag,
1342                  const char *startSpecifier, unsigned specifierLen);
1343  void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
1344                         const analyze_printf::OptionalFlag &ignoredFlag,
1345                         const analyze_printf::OptionalFlag &flag,
1346                         const char *startSpecifier, unsigned specifierLen);
1347};
1348}
1349
1350bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
1351                                      const analyze_printf::PrintfSpecifier &FS,
1352                                      const char *startSpecifier,
1353                                      unsigned specifierLen) {
1354  const analyze_printf::PrintfConversionSpecifier &CS =
1355    FS.getConversionSpecifier();
1356
1357  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1358                                          getLocationOfByte(CS.getStart()),
1359                                          startSpecifier, specifierLen,
1360                                          CS.getStart(), CS.getLength());
1361}
1362
1363bool CheckPrintfHandler::HandleAmount(
1364                               const analyze_format_string::OptionalAmount &Amt,
1365                               unsigned k, const char *startSpecifier,
1366                               unsigned specifierLen) {
1367
1368  if (Amt.hasDataArgument()) {
1369    if (!HasVAListArg) {
1370      unsigned argIndex = Amt.getArgIndex();
1371      if (argIndex >= NumDataArgs) {
1372        S.Diag(getLocationOfByte(Amt.getStart()),
1373               diag::warn_printf_asterisk_missing_arg)
1374          << k << getSpecifierRange(startSpecifier, specifierLen);
1375        // Don't do any more checking.  We will just emit
1376        // spurious errors.
1377        return false;
1378      }
1379
1380      // Type check the data argument.  It should be an 'int'.
1381      // Although not in conformance with C99, we also allow the argument to be
1382      // an 'unsigned int' as that is a reasonably safe case.  GCC also
1383      // doesn't emit a warning for that case.
1384      CoveredArgs.set(argIndex);
1385      const Expr *Arg = getDataArg(argIndex);
1386      QualType T = Arg->getType();
1387
1388      const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context);
1389      assert(ATR.isValid());
1390
1391      if (!ATR.matchesType(S.Context, T)) {
1392        S.Diag(getLocationOfByte(Amt.getStart()),
1393               diag::warn_printf_asterisk_wrong_type)
1394          << k
1395          << ATR.getRepresentativeType(S.Context) << T
1396          << getSpecifierRange(startSpecifier, specifierLen)
1397          << Arg->getSourceRange();
1398        // Don't do any more checking.  We will just emit
1399        // spurious errors.
1400        return false;
1401      }
1402    }
1403  }
1404  return true;
1405}
1406
1407void CheckPrintfHandler::HandleInvalidAmount(
1408                                      const analyze_printf::PrintfSpecifier &FS,
1409                                      const analyze_printf::OptionalAmount &Amt,
1410                                      unsigned type,
1411                                      const char *startSpecifier,
1412                                      unsigned specifierLen) {
1413  const analyze_printf::PrintfConversionSpecifier &CS =
1414    FS.getConversionSpecifier();
1415  switch (Amt.getHowSpecified()) {
1416  case analyze_printf::OptionalAmount::Constant:
1417    S.Diag(getLocationOfByte(Amt.getStart()),
1418        diag::warn_printf_nonsensical_optional_amount)
1419      << type
1420      << CS.toString()
1421      << getSpecifierRange(startSpecifier, specifierLen)
1422      << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
1423          Amt.getConstantLength()));
1424    break;
1425
1426  default:
1427    S.Diag(getLocationOfByte(Amt.getStart()),
1428        diag::warn_printf_nonsensical_optional_amount)
1429      << type
1430      << CS.toString()
1431      << getSpecifierRange(startSpecifier, specifierLen);
1432    break;
1433  }
1434}
1435
1436void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1437                                    const analyze_printf::OptionalFlag &flag,
1438                                    const char *startSpecifier,
1439                                    unsigned specifierLen) {
1440  // Warn about pointless flag with a fixit removal.
1441  const analyze_printf::PrintfConversionSpecifier &CS =
1442    FS.getConversionSpecifier();
1443  S.Diag(getLocationOfByte(flag.getPosition()),
1444      diag::warn_printf_nonsensical_flag)
1445    << flag.toString() << CS.toString()
1446    << getSpecifierRange(startSpecifier, specifierLen)
1447    << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1));
1448}
1449
1450void CheckPrintfHandler::HandleIgnoredFlag(
1451                                const analyze_printf::PrintfSpecifier &FS,
1452                                const analyze_printf::OptionalFlag &ignoredFlag,
1453                                const analyze_printf::OptionalFlag &flag,
1454                                const char *startSpecifier,
1455                                unsigned specifierLen) {
1456  // Warn about ignored flag with a fixit removal.
1457  S.Diag(getLocationOfByte(ignoredFlag.getPosition()),
1458      diag::warn_printf_ignored_flag)
1459    << ignoredFlag.toString() << flag.toString()
1460    << getSpecifierRange(startSpecifier, specifierLen)
1461    << FixItHint::CreateRemoval(getSpecifierRange(
1462        ignoredFlag.getPosition(), 1));
1463}
1464
1465bool
1466CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
1467                                            &FS,
1468                                          const char *startSpecifier,
1469                                          unsigned specifierLen) {
1470
1471  using namespace analyze_format_string;
1472  using namespace analyze_printf;
1473  const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
1474
1475  if (FS.consumesDataArgument()) {
1476    if (atFirstArg) {
1477        atFirstArg = false;
1478        usesPositionalArgs = FS.usesPositionalArg();
1479    }
1480    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1481      // Cannot mix-and-match positional and non-positional arguments.
1482      S.Diag(getLocationOfByte(CS.getStart()),
1483             diag::warn_format_mix_positional_nonpositional_args)
1484        << getSpecifierRange(startSpecifier, specifierLen);
1485      return false;
1486    }
1487  }
1488
1489  // First check if the field width, precision, and conversion specifier
1490  // have matching data arguments.
1491  if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
1492                    startSpecifier, specifierLen)) {
1493    return false;
1494  }
1495
1496  if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
1497                    startSpecifier, specifierLen)) {
1498    return false;
1499  }
1500
1501  if (!CS.consumesDataArgument()) {
1502    // FIXME: Technically specifying a precision or field width here
1503    // makes no sense.  Worth issuing a warning at some point.
1504    return true;
1505  }
1506
1507  // Consume the argument.
1508  unsigned argIndex = FS.getArgIndex();
1509  if (argIndex < NumDataArgs) {
1510    // The check to see if the argIndex is valid will come later.
1511    // We set the bit here because we may exit early from this
1512    // function if we encounter some other error.
1513    CoveredArgs.set(argIndex);
1514  }
1515
1516  // Check for using an Objective-C specific conversion specifier
1517  // in a non-ObjC literal.
1518  if (!IsObjCLiteral && CS.isObjCArg()) {
1519    return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
1520                                                  specifierLen);
1521  }
1522
1523  // Check for invalid use of field width
1524  if (!FS.hasValidFieldWidth()) {
1525    HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
1526        startSpecifier, specifierLen);
1527  }
1528
1529  // Check for invalid use of precision
1530  if (!FS.hasValidPrecision()) {
1531    HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
1532        startSpecifier, specifierLen);
1533  }
1534
1535  // Check each flag does not conflict with any other component.
1536  if (!FS.hasValidThousandsGroupingPrefix())
1537    HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
1538  if (!FS.hasValidLeadingZeros())
1539    HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
1540  if (!FS.hasValidPlusPrefix())
1541    HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
1542  if (!FS.hasValidSpacePrefix())
1543    HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
1544  if (!FS.hasValidAlternativeForm())
1545    HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
1546  if (!FS.hasValidLeftJustified())
1547    HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
1548
1549  // Check that flags are not ignored by another flag
1550  if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
1551    HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
1552        startSpecifier, specifierLen);
1553  if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
1554    HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
1555            startSpecifier, specifierLen);
1556
1557  // Check the length modifier is valid with the given conversion specifier.
1558  const LengthModifier &LM = FS.getLengthModifier();
1559  if (!FS.hasValidLengthModifier())
1560    S.Diag(getLocationOfByte(LM.getStart()),
1561        diag::warn_format_nonsensical_length)
1562      << LM.toString() << CS.toString()
1563      << getSpecifierRange(startSpecifier, specifierLen)
1564      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1565          LM.getLength()));
1566
1567  // Are we using '%n'?
1568  if (CS.getKind() == ConversionSpecifier::nArg) {
1569    // Issue a warning about this being a possible security issue.
1570    S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back)
1571      << getSpecifierRange(startSpecifier, specifierLen);
1572    // Continue checking the other format specifiers.
1573    return true;
1574  }
1575
1576  // The remaining checks depend on the data arguments.
1577  if (HasVAListArg)
1578    return true;
1579
1580  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1581    return false;
1582
1583  // Now type check the data expression that matches the
1584  // format specifier.
1585  const Expr *Ex = getDataArg(argIndex);
1586  const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context);
1587  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
1588    // Check if we didn't match because of an implicit cast from a 'char'
1589    // or 'short' to an 'int'.  This is done because printf is a varargs
1590    // function.
1591    if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex))
1592      if (ICE->getType() == S.Context.IntTy) {
1593        // All further checking is done on the subexpression.
1594        Ex = ICE->getSubExpr();
1595        if (ATR.matchesType(S.Context, Ex->getType()))
1596          return true;
1597      }
1598
1599    // We may be able to offer a FixItHint if it is a supported type.
1600    PrintfSpecifier fixedFS = FS;
1601    bool success = fixedFS.fixType(Ex->getType());
1602
1603    if (success) {
1604      // Get the fix string from the fixed format specifier
1605      llvm::SmallString<128> buf;
1606      llvm::raw_svector_ostream os(buf);
1607      fixedFS.toString(os);
1608
1609      // FIXME: getRepresentativeType() perhaps should return a string
1610      // instead of a QualType to better handle when the representative
1611      // type is 'wint_t' (which is defined in the system headers).
1612      S.Diag(getLocationOfByte(CS.getStart()),
1613          diag::warn_printf_conversion_argument_type_mismatch)
1614        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1615        << getSpecifierRange(startSpecifier, specifierLen)
1616        << Ex->getSourceRange()
1617        << FixItHint::CreateReplacement(
1618            getSpecifierRange(startSpecifier, specifierLen),
1619            os.str());
1620    }
1621    else {
1622      S.Diag(getLocationOfByte(CS.getStart()),
1623             diag::warn_printf_conversion_argument_type_mismatch)
1624        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1625        << getSpecifierRange(startSpecifier, specifierLen)
1626        << Ex->getSourceRange();
1627    }
1628  }
1629
1630  return true;
1631}
1632
1633//===--- CHECK: Scanf format string checking ------------------------------===//
1634
1635namespace {
1636class CheckScanfHandler : public CheckFormatHandler {
1637public:
1638  CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
1639                    const Expr *origFormatExpr, unsigned firstDataArg,
1640                    unsigned numDataArgs, bool isObjCLiteral,
1641                    const char *beg, bool hasVAListArg,
1642                    const CallExpr *theCall, unsigned formatIdx)
1643  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1644                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1645                       theCall, formatIdx) {}
1646
1647  bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
1648                            const char *startSpecifier,
1649                            unsigned specifierLen);
1650
1651  bool HandleInvalidScanfConversionSpecifier(
1652          const analyze_scanf::ScanfSpecifier &FS,
1653          const char *startSpecifier,
1654          unsigned specifierLen);
1655
1656  void HandleIncompleteScanList(const char *start, const char *end);
1657};
1658}
1659
1660void CheckScanfHandler::HandleIncompleteScanList(const char *start,
1661                                                 const char *end) {
1662  S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete)
1663    << getSpecifierRange(start, end - start);
1664}
1665
1666bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
1667                                        const analyze_scanf::ScanfSpecifier &FS,
1668                                        const char *startSpecifier,
1669                                        unsigned specifierLen) {
1670
1671  const analyze_scanf::ScanfConversionSpecifier &CS =
1672    FS.getConversionSpecifier();
1673
1674  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1675                                          getLocationOfByte(CS.getStart()),
1676                                          startSpecifier, specifierLen,
1677                                          CS.getStart(), CS.getLength());
1678}
1679
1680bool CheckScanfHandler::HandleScanfSpecifier(
1681                                       const analyze_scanf::ScanfSpecifier &FS,
1682                                       const char *startSpecifier,
1683                                       unsigned specifierLen) {
1684
1685  using namespace analyze_scanf;
1686  using namespace analyze_format_string;
1687
1688  const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
1689
1690  // Handle case where '%' and '*' don't consume an argument.  These shouldn't
1691  // be used to decide if we are using positional arguments consistently.
1692  if (FS.consumesDataArgument()) {
1693    if (atFirstArg) {
1694      atFirstArg = false;
1695      usesPositionalArgs = FS.usesPositionalArg();
1696    }
1697    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1698      // Cannot mix-and-match positional and non-positional arguments.
1699      S.Diag(getLocationOfByte(CS.getStart()),
1700             diag::warn_format_mix_positional_nonpositional_args)
1701        << getSpecifierRange(startSpecifier, specifierLen);
1702      return false;
1703    }
1704  }
1705
1706  // Check if the field with is non-zero.
1707  const OptionalAmount &Amt = FS.getFieldWidth();
1708  if (Amt.getHowSpecified() == OptionalAmount::Constant) {
1709    if (Amt.getConstantAmount() == 0) {
1710      const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
1711                                                   Amt.getConstantLength());
1712      S.Diag(getLocationOfByte(Amt.getStart()),
1713             diag::warn_scanf_nonzero_width)
1714        << R << FixItHint::CreateRemoval(R);
1715    }
1716  }
1717
1718  if (!FS.consumesDataArgument()) {
1719    // FIXME: Technically specifying a precision or field width here
1720    // makes no sense.  Worth issuing a warning at some point.
1721    return true;
1722  }
1723
1724  // Consume the argument.
1725  unsigned argIndex = FS.getArgIndex();
1726  if (argIndex < NumDataArgs) {
1727      // The check to see if the argIndex is valid will come later.
1728      // We set the bit here because we may exit early from this
1729      // function if we encounter some other error.
1730    CoveredArgs.set(argIndex);
1731  }
1732
1733  // Check the length modifier is valid with the given conversion specifier.
1734  const LengthModifier &LM = FS.getLengthModifier();
1735  if (!FS.hasValidLengthModifier()) {
1736    S.Diag(getLocationOfByte(LM.getStart()),
1737           diag::warn_format_nonsensical_length)
1738      << LM.toString() << CS.toString()
1739      << getSpecifierRange(startSpecifier, specifierLen)
1740      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1741                                                    LM.getLength()));
1742  }
1743
1744  // The remaining checks depend on the data arguments.
1745  if (HasVAListArg)
1746    return true;
1747
1748  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1749    return false;
1750
1751  // FIXME: Check that the argument type matches the format specifier.
1752
1753  return true;
1754}
1755
1756void Sema::CheckFormatString(const StringLiteral *FExpr,
1757                             const Expr *OrigFormatExpr,
1758                             const CallExpr *TheCall, bool HasVAListArg,
1759                             unsigned format_idx, unsigned firstDataArg,
1760                             bool isPrintf) {
1761
1762  // CHECK: is the format string a wide literal?
1763  if (FExpr->isWide()) {
1764    Diag(FExpr->getLocStart(),
1765         diag::warn_format_string_is_wide_literal)
1766    << OrigFormatExpr->getSourceRange();
1767    return;
1768  }
1769
1770  // Str - The format string.  NOTE: this is NOT null-terminated!
1771  llvm::StringRef StrRef = FExpr->getString();
1772  const char *Str = StrRef.data();
1773  unsigned StrLen = StrRef.size();
1774
1775  // CHECK: empty format string?
1776  if (StrLen == 0) {
1777    Diag(FExpr->getLocStart(), diag::warn_empty_format_string)
1778    << OrigFormatExpr->getSourceRange();
1779    return;
1780  }
1781
1782  if (isPrintf) {
1783    CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1784                         TheCall->getNumArgs() - firstDataArg,
1785                         isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1786                         HasVAListArg, TheCall, format_idx);
1787
1788    if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen))
1789      H.DoneProcessing();
1790  }
1791  else {
1792    CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1793                        TheCall->getNumArgs() - firstDataArg,
1794                        isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1795                        HasVAListArg, TheCall, format_idx);
1796
1797    if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen))
1798      H.DoneProcessing();
1799  }
1800}
1801
1802//===--- CHECK: Standard memory functions ---------------------------------===//
1803
1804/// \brief Determine whether the given type is a dynamic class type (e.g.,
1805/// whether it has a vtable).
1806static bool isDynamicClassType(QualType T) {
1807  if (CXXRecordDecl *Record = T->getAsCXXRecordDecl())
1808    if (CXXRecordDecl *Definition = Record->getDefinition())
1809      if (Definition->isDynamicClass())
1810        return true;
1811
1812  return false;
1813}
1814
1815/// \brief Check for dangerous or invalid arguments to memset().
1816///
1817/// This issues warnings on known problematic or dangerous or unspecified
1818/// arguments to the standard 'memset', 'memcpy', and 'memmove' function calls.
1819///
1820/// \param Call The call expression to diagnose.
1821void Sema::CheckMemsetcpymoveArguments(const CallExpr *Call,
1822                                       const IdentifierInfo *FnName) {
1823  // It is possible to have a non-standard definition of memset.  Validate
1824  // we have the proper number of arguments, and if not, abort further
1825  // checking.
1826  if (Call->getNumArgs() != 3)
1827    return;
1828
1829  unsigned LastArg = FnName->isStr("memset")? 1 : 2;
1830  for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) {
1831    const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts();
1832
1833    QualType DestTy = Dest->getType();
1834    if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
1835      QualType PointeeTy = DestPtrTy->getPointeeType();
1836      if (PointeeTy->isVoidType())
1837        continue;
1838
1839      unsigned DiagID = 0;
1840      // Always complain about dynamic classes.
1841      if (isDynamicClassType(PointeeTy))
1842        DiagID = diag::warn_dyn_class_memaccess;
1843      // Check the C++11 POD definition regardless of language mode; it is more
1844      // relaxed than earlier definitions and we don't want spurious warnings.
1845      else if (!PointeeTy->isCXX11PODType())
1846        DiagID = diag::warn_non_pod_memaccess;
1847      else
1848        continue;
1849
1850      DiagRuntimeBehavior(
1851        Dest->getExprLoc(), Dest,
1852        PDiag(DiagID)
1853          << ArgIdx << FnName << PointeeTy
1854          << Call->getCallee()->getSourceRange());
1855
1856      SourceRange ArgRange = Call->getArg(0)->getSourceRange();
1857      DiagRuntimeBehavior(
1858        Dest->getExprLoc(), Dest,
1859        PDiag(diag::note_non_pod_memaccess_silence)
1860          << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
1861      break;
1862    }
1863  }
1864}
1865
1866//===--- CHECK: Return Address of Stack Variable --------------------------===//
1867
1868static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars);
1869static Expr *EvalAddr(Expr* E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars);
1870
1871/// CheckReturnStackAddr - Check if a return statement returns the address
1872///   of a stack variable.
1873void
1874Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
1875                           SourceLocation ReturnLoc) {
1876
1877  Expr *stackE = 0;
1878  llvm::SmallVector<DeclRefExpr *, 8> refVars;
1879
1880  // Perform checking for returned stack addresses, local blocks,
1881  // label addresses or references to temporaries.
1882  if (lhsType->isPointerType() || lhsType->isBlockPointerType()) {
1883    stackE = EvalAddr(RetValExp, refVars);
1884  } else if (lhsType->isReferenceType()) {
1885    stackE = EvalVal(RetValExp, refVars);
1886  }
1887
1888  if (stackE == 0)
1889    return; // Nothing suspicious was found.
1890
1891  SourceLocation diagLoc;
1892  SourceRange diagRange;
1893  if (refVars.empty()) {
1894    diagLoc = stackE->getLocStart();
1895    diagRange = stackE->getSourceRange();
1896  } else {
1897    // We followed through a reference variable. 'stackE' contains the
1898    // problematic expression but we will warn at the return statement pointing
1899    // at the reference variable. We will later display the "trail" of
1900    // reference variables using notes.
1901    diagLoc = refVars[0]->getLocStart();
1902    diagRange = refVars[0]->getSourceRange();
1903  }
1904
1905  if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
1906    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
1907                                             : diag::warn_ret_stack_addr)
1908     << DR->getDecl()->getDeclName() << diagRange;
1909  } else if (isa<BlockExpr>(stackE)) { // local block.
1910    Diag(diagLoc, diag::err_ret_local_block) << diagRange;
1911  } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
1912    Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
1913  } else { // local temporary.
1914    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
1915                                             : diag::warn_ret_local_temp_addr)
1916     << diagRange;
1917  }
1918
1919  // Display the "trail" of reference variables that we followed until we
1920  // found the problematic expression using notes.
1921  for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
1922    VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
1923    // If this var binds to another reference var, show the range of the next
1924    // var, otherwise the var binds to the problematic expression, in which case
1925    // show the range of the expression.
1926    SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
1927                                  : stackE->getSourceRange();
1928    Diag(VD->getLocation(), diag::note_ref_var_local_bind)
1929      << VD->getDeclName() << range;
1930  }
1931}
1932
1933/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
1934///  check if the expression in a return statement evaluates to an address
1935///  to a location on the stack, a local block, an address of a label, or a
1936///  reference to local temporary. The recursion is used to traverse the
1937///  AST of the return expression, with recursion backtracking when we
1938///  encounter a subexpression that (1) clearly does not lead to one of the
1939///  above problematic expressions (2) is something we cannot determine leads to
1940///  a problematic expression based on such local checking.
1941///
1942///  Both EvalAddr and EvalVal follow through reference variables to evaluate
1943///  the expression that they point to. Such variables are added to the
1944///  'refVars' vector so that we know what the reference variable "trail" was.
1945///
1946///  EvalAddr processes expressions that are pointers that are used as
1947///  references (and not L-values).  EvalVal handles all other values.
1948///  At the base case of the recursion is a check for the above problematic
1949///  expressions.
1950///
1951///  This implementation handles:
1952///
1953///   * pointer-to-pointer casts
1954///   * implicit conversions from array references to pointers
1955///   * taking the address of fields
1956///   * arbitrary interplay between "&" and "*" operators
1957///   * pointer arithmetic from an address of a stack variable
1958///   * taking the address of an array element where the array is on the stack
1959static Expr *EvalAddr(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) {
1960  if (E->isTypeDependent())
1961      return NULL;
1962
1963  // We should only be called for evaluating pointer expressions.
1964  assert((E->getType()->isAnyPointerType() ||
1965          E->getType()->isBlockPointerType() ||
1966          E->getType()->isObjCQualifiedIdType()) &&
1967         "EvalAddr only works on pointers");
1968
1969  E = E->IgnoreParens();
1970
1971  // Our "symbolic interpreter" is just a dispatch off the currently
1972  // viewed AST node.  We then recursively traverse the AST by calling
1973  // EvalAddr and EvalVal appropriately.
1974  switch (E->getStmtClass()) {
1975  case Stmt::DeclRefExprClass: {
1976    DeclRefExpr *DR = cast<DeclRefExpr>(E);
1977
1978    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
1979      // If this is a reference variable, follow through to the expression that
1980      // it points to.
1981      if (V->hasLocalStorage() &&
1982          V->getType()->isReferenceType() && V->hasInit()) {
1983        // Add the reference variable to the "trail".
1984        refVars.push_back(DR);
1985        return EvalAddr(V->getInit(), refVars);
1986      }
1987
1988    return NULL;
1989  }
1990
1991  case Stmt::UnaryOperatorClass: {
1992    // The only unary operator that make sense to handle here
1993    // is AddrOf.  All others don't make sense as pointers.
1994    UnaryOperator *U = cast<UnaryOperator>(E);
1995
1996    if (U->getOpcode() == UO_AddrOf)
1997      return EvalVal(U->getSubExpr(), refVars);
1998    else
1999      return NULL;
2000  }
2001
2002  case Stmt::BinaryOperatorClass: {
2003    // Handle pointer arithmetic.  All other binary operators are not valid
2004    // in this context.
2005    BinaryOperator *B = cast<BinaryOperator>(E);
2006    BinaryOperatorKind op = B->getOpcode();
2007
2008    if (op != BO_Add && op != BO_Sub)
2009      return NULL;
2010
2011    Expr *Base = B->getLHS();
2012
2013    // Determine which argument is the real pointer base.  It could be
2014    // the RHS argument instead of the LHS.
2015    if (!Base->getType()->isPointerType()) Base = B->getRHS();
2016
2017    assert (Base->getType()->isPointerType());
2018    return EvalAddr(Base, refVars);
2019  }
2020
2021  // For conditional operators we need to see if either the LHS or RHS are
2022  // valid DeclRefExpr*s.  If one of them is valid, we return it.
2023  case Stmt::ConditionalOperatorClass: {
2024    ConditionalOperator *C = cast<ConditionalOperator>(E);
2025
2026    // Handle the GNU extension for missing LHS.
2027    if (Expr *lhsExpr = C->getLHS()) {
2028    // In C++, we can have a throw-expression, which has 'void' type.
2029      if (!lhsExpr->getType()->isVoidType())
2030        if (Expr* LHS = EvalAddr(lhsExpr, refVars))
2031          return LHS;
2032    }
2033
2034    // In C++, we can have a throw-expression, which has 'void' type.
2035    if (C->getRHS()->getType()->isVoidType())
2036      return NULL;
2037
2038    return EvalAddr(C->getRHS(), refVars);
2039  }
2040
2041  case Stmt::BlockExprClass:
2042    if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
2043      return E; // local block.
2044    return NULL;
2045
2046  case Stmt::AddrLabelExprClass:
2047    return E; // address of label.
2048
2049  // For casts, we need to handle conversions from arrays to
2050  // pointer values, and pointer-to-pointer conversions.
2051  case Stmt::ImplicitCastExprClass:
2052  case Stmt::CStyleCastExprClass:
2053  case Stmt::CXXFunctionalCastExprClass: {
2054    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
2055    QualType T = SubExpr->getType();
2056
2057    if (SubExpr->getType()->isPointerType() ||
2058        SubExpr->getType()->isBlockPointerType() ||
2059        SubExpr->getType()->isObjCQualifiedIdType())
2060      return EvalAddr(SubExpr, refVars);
2061    else if (T->isArrayType())
2062      return EvalVal(SubExpr, refVars);
2063    else
2064      return 0;
2065  }
2066
2067  // C++ casts.  For dynamic casts, static casts, and const casts, we
2068  // are always converting from a pointer-to-pointer, so we just blow
2069  // through the cast.  In the case the dynamic cast doesn't fail (and
2070  // return NULL), we take the conservative route and report cases
2071  // where we return the address of a stack variable.  For Reinterpre
2072  // FIXME: The comment about is wrong; we're not always converting
2073  // from pointer to pointer. I'm guessing that this code should also
2074  // handle references to objects.
2075  case Stmt::CXXStaticCastExprClass:
2076  case Stmt::CXXDynamicCastExprClass:
2077  case Stmt::CXXConstCastExprClass:
2078  case Stmt::CXXReinterpretCastExprClass: {
2079      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
2080      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
2081        return EvalAddr(S, refVars);
2082      else
2083        return NULL;
2084  }
2085
2086  // Everything else: we simply don't reason about them.
2087  default:
2088    return NULL;
2089  }
2090}
2091
2092
2093///  EvalVal - This function is complements EvalAddr in the mutual recursion.
2094///   See the comments for EvalAddr for more details.
2095static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) {
2096do {
2097  // We should only be called for evaluating non-pointer expressions, or
2098  // expressions with a pointer type that are not used as references but instead
2099  // are l-values (e.g., DeclRefExpr with a pointer type).
2100
2101  // Our "symbolic interpreter" is just a dispatch off the currently
2102  // viewed AST node.  We then recursively traverse the AST by calling
2103  // EvalAddr and EvalVal appropriately.
2104
2105  E = E->IgnoreParens();
2106  switch (E->getStmtClass()) {
2107  case Stmt::ImplicitCastExprClass: {
2108    ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
2109    if (IE->getValueKind() == VK_LValue) {
2110      E = IE->getSubExpr();
2111      continue;
2112    }
2113    return NULL;
2114  }
2115
2116  case Stmt::DeclRefExprClass: {
2117    // When we hit a DeclRefExpr we are looking at code that refers to a
2118    // variable's name. If it's not a reference variable we check if it has
2119    // local storage within the function, and if so, return the expression.
2120    DeclRefExpr *DR = cast<DeclRefExpr>(E);
2121
2122    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2123      if (V->hasLocalStorage()) {
2124        if (!V->getType()->isReferenceType())
2125          return DR;
2126
2127        // Reference variable, follow through to the expression that
2128        // it points to.
2129        if (V->hasInit()) {
2130          // Add the reference variable to the "trail".
2131          refVars.push_back(DR);
2132          return EvalVal(V->getInit(), refVars);
2133        }
2134      }
2135
2136    return NULL;
2137  }
2138
2139  case Stmt::UnaryOperatorClass: {
2140    // The only unary operator that make sense to handle here
2141    // is Deref.  All others don't resolve to a "name."  This includes
2142    // handling all sorts of rvalues passed to a unary operator.
2143    UnaryOperator *U = cast<UnaryOperator>(E);
2144
2145    if (U->getOpcode() == UO_Deref)
2146      return EvalAddr(U->getSubExpr(), refVars);
2147
2148    return NULL;
2149  }
2150
2151  case Stmt::ArraySubscriptExprClass: {
2152    // Array subscripts are potential references to data on the stack.  We
2153    // retrieve the DeclRefExpr* for the array variable if it indeed
2154    // has local storage.
2155    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars);
2156  }
2157
2158  case Stmt::ConditionalOperatorClass: {
2159    // For conditional operators we need to see if either the LHS or RHS are
2160    // non-NULL Expr's.  If one is non-NULL, we return it.
2161    ConditionalOperator *C = cast<ConditionalOperator>(E);
2162
2163    // Handle the GNU extension for missing LHS.
2164    if (Expr *lhsExpr = C->getLHS())
2165      if (Expr *LHS = EvalVal(lhsExpr, refVars))
2166        return LHS;
2167
2168    return EvalVal(C->getRHS(), refVars);
2169  }
2170
2171  // Accesses to members are potential references to data on the stack.
2172  case Stmt::MemberExprClass: {
2173    MemberExpr *M = cast<MemberExpr>(E);
2174
2175    // Check for indirect access.  We only want direct field accesses.
2176    if (M->isArrow())
2177      return NULL;
2178
2179    // Check whether the member type is itself a reference, in which case
2180    // we're not going to refer to the member, but to what the member refers to.
2181    if (M->getMemberDecl()->getType()->isReferenceType())
2182      return NULL;
2183
2184    return EvalVal(M->getBase(), refVars);
2185  }
2186
2187  default:
2188    // Check that we don't return or take the address of a reference to a
2189    // temporary. This is only useful in C++.
2190    if (!E->isTypeDependent() && E->isRValue())
2191      return E;
2192
2193    // Everything else: we simply don't reason about them.
2194    return NULL;
2195  }
2196} while (true);
2197}
2198
2199//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
2200
2201/// Check for comparisons of floating point operands using != and ==.
2202/// Issue a warning if these are no self-comparisons, as they are not likely
2203/// to do what the programmer intended.
2204void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) {
2205  bool EmitWarning = true;
2206
2207  Expr* LeftExprSansParen = lex->IgnoreParenImpCasts();
2208  Expr* RightExprSansParen = rex->IgnoreParenImpCasts();
2209
2210  // Special case: check for x == x (which is OK).
2211  // Do not emit warnings for such cases.
2212  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
2213    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
2214      if (DRL->getDecl() == DRR->getDecl())
2215        EmitWarning = false;
2216
2217
2218  // Special case: check for comparisons against literals that can be exactly
2219  //  represented by APFloat.  In such cases, do not emit a warning.  This
2220  //  is a heuristic: often comparison against such literals are used to
2221  //  detect if a value in a variable has not changed.  This clearly can
2222  //  lead to false negatives.
2223  if (EmitWarning) {
2224    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
2225      if (FLL->isExact())
2226        EmitWarning = false;
2227    } else
2228      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
2229        if (FLR->isExact())
2230          EmitWarning = false;
2231    }
2232  }
2233
2234  // Check for comparisons with builtin types.
2235  if (EmitWarning)
2236    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
2237      if (CL->isBuiltinCall(Context))
2238        EmitWarning = false;
2239
2240  if (EmitWarning)
2241    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
2242      if (CR->isBuiltinCall(Context))
2243        EmitWarning = false;
2244
2245  // Emit the diagnostic.
2246  if (EmitWarning)
2247    Diag(loc, diag::warn_floatingpoint_eq)
2248      << lex->getSourceRange() << rex->getSourceRange();
2249}
2250
2251//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
2252//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
2253
2254namespace {
2255
2256/// Structure recording the 'active' range of an integer-valued
2257/// expression.
2258struct IntRange {
2259  /// The number of bits active in the int.
2260  unsigned Width;
2261
2262  /// True if the int is known not to have negative values.
2263  bool NonNegative;
2264
2265  IntRange(unsigned Width, bool NonNegative)
2266    : Width(Width), NonNegative(NonNegative)
2267  {}
2268
2269  /// Returns the range of the bool type.
2270  static IntRange forBoolType() {
2271    return IntRange(1, true);
2272  }
2273
2274  /// Returns the range of an opaque value of the given integral type.
2275  static IntRange forValueOfType(ASTContext &C, QualType T) {
2276    return forValueOfCanonicalType(C,
2277                          T->getCanonicalTypeInternal().getTypePtr());
2278  }
2279
2280  /// Returns the range of an opaque value of a canonical integral type.
2281  static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
2282    assert(T->isCanonicalUnqualified());
2283
2284    if (const VectorType *VT = dyn_cast<VectorType>(T))
2285      T = VT->getElementType().getTypePtr();
2286    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2287      T = CT->getElementType().getTypePtr();
2288
2289    // For enum types, use the known bit width of the enumerators.
2290    if (const EnumType *ET = dyn_cast<EnumType>(T)) {
2291      EnumDecl *Enum = ET->getDecl();
2292      if (!Enum->isDefinition())
2293        return IntRange(C.getIntWidth(QualType(T, 0)), false);
2294
2295      unsigned NumPositive = Enum->getNumPositiveBits();
2296      unsigned NumNegative = Enum->getNumNegativeBits();
2297
2298      return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
2299    }
2300
2301    const BuiltinType *BT = cast<BuiltinType>(T);
2302    assert(BT->isInteger());
2303
2304    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2305  }
2306
2307  /// Returns the "target" range of a canonical integral type, i.e.
2308  /// the range of values expressible in the type.
2309  ///
2310  /// This matches forValueOfCanonicalType except that enums have the
2311  /// full range of their type, not the range of their enumerators.
2312  static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
2313    assert(T->isCanonicalUnqualified());
2314
2315    if (const VectorType *VT = dyn_cast<VectorType>(T))
2316      T = VT->getElementType().getTypePtr();
2317    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2318      T = CT->getElementType().getTypePtr();
2319    if (const EnumType *ET = dyn_cast<EnumType>(T))
2320      T = ET->getDecl()->getIntegerType().getTypePtr();
2321
2322    const BuiltinType *BT = cast<BuiltinType>(T);
2323    assert(BT->isInteger());
2324
2325    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2326  }
2327
2328  /// Returns the supremum of two ranges: i.e. their conservative merge.
2329  static IntRange join(IntRange L, IntRange R) {
2330    return IntRange(std::max(L.Width, R.Width),
2331                    L.NonNegative && R.NonNegative);
2332  }
2333
2334  /// Returns the infinum of two ranges: i.e. their aggressive merge.
2335  static IntRange meet(IntRange L, IntRange R) {
2336    return IntRange(std::min(L.Width, R.Width),
2337                    L.NonNegative || R.NonNegative);
2338  }
2339};
2340
2341IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) {
2342  if (value.isSigned() && value.isNegative())
2343    return IntRange(value.getMinSignedBits(), false);
2344
2345  if (value.getBitWidth() > MaxWidth)
2346    value = value.trunc(MaxWidth);
2347
2348  // isNonNegative() just checks the sign bit without considering
2349  // signedness.
2350  return IntRange(value.getActiveBits(), true);
2351}
2352
2353IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
2354                       unsigned MaxWidth) {
2355  if (result.isInt())
2356    return GetValueRange(C, result.getInt(), MaxWidth);
2357
2358  if (result.isVector()) {
2359    IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
2360    for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
2361      IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
2362      R = IntRange::join(R, El);
2363    }
2364    return R;
2365  }
2366
2367  if (result.isComplexInt()) {
2368    IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
2369    IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
2370    return IntRange::join(R, I);
2371  }
2372
2373  // This can happen with lossless casts to intptr_t of "based" lvalues.
2374  // Assume it might use arbitrary bits.
2375  // FIXME: The only reason we need to pass the type in here is to get
2376  // the sign right on this one case.  It would be nice if APValue
2377  // preserved this.
2378  assert(result.isLValue());
2379  return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType());
2380}
2381
2382/// Pseudo-evaluate the given integer expression, estimating the
2383/// range of values it might take.
2384///
2385/// \param MaxWidth - the width to which the value will be truncated
2386IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
2387  E = E->IgnoreParens();
2388
2389  // Try a full evaluation first.
2390  Expr::EvalResult result;
2391  if (E->Evaluate(result, C))
2392    return GetValueRange(C, result.Val, E->getType(), MaxWidth);
2393
2394  // I think we only want to look through implicit casts here; if the
2395  // user has an explicit widening cast, we should treat the value as
2396  // being of the new, wider type.
2397  if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
2398    if (CE->getCastKind() == CK_NoOp)
2399      return GetExprRange(C, CE->getSubExpr(), MaxWidth);
2400
2401    IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
2402
2403    bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
2404
2405    // Assume that non-integer casts can span the full range of the type.
2406    if (!isIntegerCast)
2407      return OutputTypeRange;
2408
2409    IntRange SubRange
2410      = GetExprRange(C, CE->getSubExpr(),
2411                     std::min(MaxWidth, OutputTypeRange.Width));
2412
2413    // Bail out if the subexpr's range is as wide as the cast type.
2414    if (SubRange.Width >= OutputTypeRange.Width)
2415      return OutputTypeRange;
2416
2417    // Otherwise, we take the smaller width, and we're non-negative if
2418    // either the output type or the subexpr is.
2419    return IntRange(SubRange.Width,
2420                    SubRange.NonNegative || OutputTypeRange.NonNegative);
2421  }
2422
2423  if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
2424    // If we can fold the condition, just take that operand.
2425    bool CondResult;
2426    if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
2427      return GetExprRange(C, CondResult ? CO->getTrueExpr()
2428                                        : CO->getFalseExpr(),
2429                          MaxWidth);
2430
2431    // Otherwise, conservatively merge.
2432    IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
2433    IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
2434    return IntRange::join(L, R);
2435  }
2436
2437  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
2438    switch (BO->getOpcode()) {
2439
2440    // Boolean-valued operations are single-bit and positive.
2441    case BO_LAnd:
2442    case BO_LOr:
2443    case BO_LT:
2444    case BO_GT:
2445    case BO_LE:
2446    case BO_GE:
2447    case BO_EQ:
2448    case BO_NE:
2449      return IntRange::forBoolType();
2450
2451    // The type of these compound assignments is the type of the LHS,
2452    // so the RHS is not necessarily an integer.
2453    case BO_MulAssign:
2454    case BO_DivAssign:
2455    case BO_RemAssign:
2456    case BO_AddAssign:
2457    case BO_SubAssign:
2458      return IntRange::forValueOfType(C, E->getType());
2459
2460    // Operations with opaque sources are black-listed.
2461    case BO_PtrMemD:
2462    case BO_PtrMemI:
2463      return IntRange::forValueOfType(C, E->getType());
2464
2465    // Bitwise-and uses the *infinum* of the two source ranges.
2466    case BO_And:
2467    case BO_AndAssign:
2468      return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
2469                            GetExprRange(C, BO->getRHS(), MaxWidth));
2470
2471    // Left shift gets black-listed based on a judgement call.
2472    case BO_Shl:
2473      // ...except that we want to treat '1 << (blah)' as logically
2474      // positive.  It's an important idiom.
2475      if (IntegerLiteral *I
2476            = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
2477        if (I->getValue() == 1) {
2478          IntRange R = IntRange::forValueOfType(C, E->getType());
2479          return IntRange(R.Width, /*NonNegative*/ true);
2480        }
2481      }
2482      // fallthrough
2483
2484    case BO_ShlAssign:
2485      return IntRange::forValueOfType(C, E->getType());
2486
2487    // Right shift by a constant can narrow its left argument.
2488    case BO_Shr:
2489    case BO_ShrAssign: {
2490      IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2491
2492      // If the shift amount is a positive constant, drop the width by
2493      // that much.
2494      llvm::APSInt shift;
2495      if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
2496          shift.isNonNegative()) {
2497        unsigned zext = shift.getZExtValue();
2498        if (zext >= L.Width)
2499          L.Width = (L.NonNegative ? 0 : 1);
2500        else
2501          L.Width -= zext;
2502      }
2503
2504      return L;
2505    }
2506
2507    // Comma acts as its right operand.
2508    case BO_Comma:
2509      return GetExprRange(C, BO->getRHS(), MaxWidth);
2510
2511    // Black-list pointer subtractions.
2512    case BO_Sub:
2513      if (BO->getLHS()->getType()->isPointerType())
2514        return IntRange::forValueOfType(C, E->getType());
2515      // fallthrough
2516
2517    default:
2518      break;
2519    }
2520
2521    // Treat every other operator as if it were closed on the
2522    // narrowest type that encompasses both operands.
2523    IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2524    IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
2525    return IntRange::join(L, R);
2526  }
2527
2528  if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
2529    switch (UO->getOpcode()) {
2530    // Boolean-valued operations are white-listed.
2531    case UO_LNot:
2532      return IntRange::forBoolType();
2533
2534    // Operations with opaque sources are black-listed.
2535    case UO_Deref:
2536    case UO_AddrOf: // should be impossible
2537      return IntRange::forValueOfType(C, E->getType());
2538
2539    default:
2540      return GetExprRange(C, UO->getSubExpr(), MaxWidth);
2541    }
2542  }
2543
2544  if (dyn_cast<OffsetOfExpr>(E)) {
2545    IntRange::forValueOfType(C, E->getType());
2546  }
2547
2548  FieldDecl *BitField = E->getBitField();
2549  if (BitField) {
2550    llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C);
2551    unsigned BitWidth = BitWidthAP.getZExtValue();
2552
2553    return IntRange(BitWidth,
2554                    BitField->getType()->isUnsignedIntegerOrEnumerationType());
2555  }
2556
2557  return IntRange::forValueOfType(C, E->getType());
2558}
2559
2560IntRange GetExprRange(ASTContext &C, Expr *E) {
2561  return GetExprRange(C, E, C.getIntWidth(E->getType()));
2562}
2563
2564/// Checks whether the given value, which currently has the given
2565/// source semantics, has the same value when coerced through the
2566/// target semantics.
2567bool IsSameFloatAfterCast(const llvm::APFloat &value,
2568                          const llvm::fltSemantics &Src,
2569                          const llvm::fltSemantics &Tgt) {
2570  llvm::APFloat truncated = value;
2571
2572  bool ignored;
2573  truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
2574  truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
2575
2576  return truncated.bitwiseIsEqual(value);
2577}
2578
2579/// Checks whether the given value, which currently has the given
2580/// source semantics, has the same value when coerced through the
2581/// target semantics.
2582///
2583/// The value might be a vector of floats (or a complex number).
2584bool IsSameFloatAfterCast(const APValue &value,
2585                          const llvm::fltSemantics &Src,
2586                          const llvm::fltSemantics &Tgt) {
2587  if (value.isFloat())
2588    return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
2589
2590  if (value.isVector()) {
2591    for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
2592      if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
2593        return false;
2594    return true;
2595  }
2596
2597  assert(value.isComplexFloat());
2598  return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
2599          IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
2600}
2601
2602void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
2603
2604static bool IsZero(Sema &S, Expr *E) {
2605  // Suppress cases where we are comparing against an enum constant.
2606  if (const DeclRefExpr *DR =
2607      dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
2608    if (isa<EnumConstantDecl>(DR->getDecl()))
2609      return false;
2610
2611  // Suppress cases where the '0' value is expanded from a macro.
2612  if (E->getLocStart().isMacroID())
2613    return false;
2614
2615  llvm::APSInt Value;
2616  return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
2617}
2618
2619static bool HasEnumType(Expr *E) {
2620  // Strip off implicit integral promotions.
2621  while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
2622    if (ICE->getCastKind() != CK_IntegralCast &&
2623        ICE->getCastKind() != CK_NoOp)
2624      break;
2625    E = ICE->getSubExpr();
2626  }
2627
2628  return E->getType()->isEnumeralType();
2629}
2630
2631void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
2632  BinaryOperatorKind op = E->getOpcode();
2633  if (E->isValueDependent())
2634    return;
2635
2636  if (op == BO_LT && IsZero(S, E->getRHS())) {
2637    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2638      << "< 0" << "false" << HasEnumType(E->getLHS())
2639      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2640  } else if (op == BO_GE && IsZero(S, E->getRHS())) {
2641    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2642      << ">= 0" << "true" << HasEnumType(E->getLHS())
2643      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2644  } else if (op == BO_GT && IsZero(S, E->getLHS())) {
2645    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2646      << "0 >" << "false" << HasEnumType(E->getRHS())
2647      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2648  } else if (op == BO_LE && IsZero(S, E->getLHS())) {
2649    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2650      << "0 <=" << "true" << HasEnumType(E->getRHS())
2651      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2652  }
2653}
2654
2655/// Analyze the operands of the given comparison.  Implements the
2656/// fallback case from AnalyzeComparison.
2657void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
2658  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
2659  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
2660}
2661
2662/// \brief Implements -Wsign-compare.
2663///
2664/// \param lex the left-hand expression
2665/// \param rex the right-hand expression
2666/// \param OpLoc the location of the joining operator
2667/// \param BinOpc binary opcode or 0
2668void AnalyzeComparison(Sema &S, BinaryOperator *E) {
2669  // The type the comparison is being performed in.
2670  QualType T = E->getLHS()->getType();
2671  assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
2672         && "comparison with mismatched types");
2673
2674  // We don't do anything special if this isn't an unsigned integral
2675  // comparison:  we're only interested in integral comparisons, and
2676  // signed comparisons only happen in cases we don't care to warn about.
2677  //
2678  // We also don't care about value-dependent expressions or expressions
2679  // whose result is a constant.
2680  if (!T->hasUnsignedIntegerRepresentation()
2681      || E->isValueDependent() || E->isIntegerConstantExpr(S.Context))
2682    return AnalyzeImpConvsInComparison(S, E);
2683
2684  Expr *lex = E->getLHS()->IgnoreParenImpCasts();
2685  Expr *rex = E->getRHS()->IgnoreParenImpCasts();
2686
2687  // Check to see if one of the (unmodified) operands is of different
2688  // signedness.
2689  Expr *signedOperand, *unsignedOperand;
2690  if (lex->getType()->hasSignedIntegerRepresentation()) {
2691    assert(!rex->getType()->hasSignedIntegerRepresentation() &&
2692           "unsigned comparison between two signed integer expressions?");
2693    signedOperand = lex;
2694    unsignedOperand = rex;
2695  } else if (rex->getType()->hasSignedIntegerRepresentation()) {
2696    signedOperand = rex;
2697    unsignedOperand = lex;
2698  } else {
2699    CheckTrivialUnsignedComparison(S, E);
2700    return AnalyzeImpConvsInComparison(S, E);
2701  }
2702
2703  // Otherwise, calculate the effective range of the signed operand.
2704  IntRange signedRange = GetExprRange(S.Context, signedOperand);
2705
2706  // Go ahead and analyze implicit conversions in the operands.  Note
2707  // that we skip the implicit conversions on both sides.
2708  AnalyzeImplicitConversions(S, lex, E->getOperatorLoc());
2709  AnalyzeImplicitConversions(S, rex, E->getOperatorLoc());
2710
2711  // If the signed range is non-negative, -Wsign-compare won't fire,
2712  // but we should still check for comparisons which are always true
2713  // or false.
2714  if (signedRange.NonNegative)
2715    return CheckTrivialUnsignedComparison(S, E);
2716
2717  // For (in)equality comparisons, if the unsigned operand is a
2718  // constant which cannot collide with a overflowed signed operand,
2719  // then reinterpreting the signed operand as unsigned will not
2720  // change the result of the comparison.
2721  if (E->isEqualityOp()) {
2722    unsigned comparisonWidth = S.Context.getIntWidth(T);
2723    IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
2724
2725    // We should never be unable to prove that the unsigned operand is
2726    // non-negative.
2727    assert(unsignedRange.NonNegative && "unsigned range includes negative?");
2728
2729    if (unsignedRange.Width < comparisonWidth)
2730      return;
2731  }
2732
2733  S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison)
2734    << lex->getType() << rex->getType()
2735    << lex->getSourceRange() << rex->getSourceRange();
2736}
2737
2738/// Analyzes an attempt to assign the given value to a bitfield.
2739///
2740/// Returns true if there was something fishy about the attempt.
2741bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
2742                               SourceLocation InitLoc) {
2743  assert(Bitfield->isBitField());
2744  if (Bitfield->isInvalidDecl())
2745    return false;
2746
2747  // White-list bool bitfields.
2748  if (Bitfield->getType()->isBooleanType())
2749    return false;
2750
2751  // Ignore value- or type-dependent expressions.
2752  if (Bitfield->getBitWidth()->isValueDependent() ||
2753      Bitfield->getBitWidth()->isTypeDependent() ||
2754      Init->isValueDependent() ||
2755      Init->isTypeDependent())
2756    return false;
2757
2758  Expr *OriginalInit = Init->IgnoreParenImpCasts();
2759
2760  llvm::APSInt Width(32);
2761  Expr::EvalResult InitValue;
2762  if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) ||
2763      !OriginalInit->Evaluate(InitValue, S.Context) ||
2764      !InitValue.Val.isInt())
2765    return false;
2766
2767  const llvm::APSInt &Value = InitValue.Val.getInt();
2768  unsigned OriginalWidth = Value.getBitWidth();
2769  unsigned FieldWidth = Width.getZExtValue();
2770
2771  if (OriginalWidth <= FieldWidth)
2772    return false;
2773
2774  llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
2775
2776  // It's fairly common to write values into signed bitfields
2777  // that, if sign-extended, would end up becoming a different
2778  // value.  We don't want to warn about that.
2779  if (Value.isSigned() && Value.isNegative())
2780    TruncatedValue = TruncatedValue.sext(OriginalWidth);
2781  else
2782    TruncatedValue = TruncatedValue.zext(OriginalWidth);
2783
2784  if (Value == TruncatedValue)
2785    return false;
2786
2787  std::string PrettyValue = Value.toString(10);
2788  std::string PrettyTrunc = TruncatedValue.toString(10);
2789
2790  S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
2791    << PrettyValue << PrettyTrunc << OriginalInit->getType()
2792    << Init->getSourceRange();
2793
2794  return true;
2795}
2796
2797/// Analyze the given simple or compound assignment for warning-worthy
2798/// operations.
2799void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
2800  // Just recurse on the LHS.
2801  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
2802
2803  // We want to recurse on the RHS as normal unless we're assigning to
2804  // a bitfield.
2805  if (FieldDecl *Bitfield = E->getLHS()->getBitField()) {
2806    if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
2807                                  E->getOperatorLoc())) {
2808      // Recurse, ignoring any implicit conversions on the RHS.
2809      return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
2810                                        E->getOperatorLoc());
2811    }
2812  }
2813
2814  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
2815}
2816
2817/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
2818void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
2819                     SourceLocation CContext, unsigned diag) {
2820  S.Diag(E->getExprLoc(), diag)
2821    << SourceType << T << E->getSourceRange() << SourceRange(CContext);
2822}
2823
2824/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
2825void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext,
2826                     unsigned diag) {
2827  DiagnoseImpCast(S, E, E->getType(), T, CContext, diag);
2828}
2829
2830/// Diagnose an implicit cast from a literal expression. Also attemps to supply
2831/// fixit hints when the cast wouldn't lose information to simply write the
2832/// expression with the expected type.
2833void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
2834                                    SourceLocation CContext) {
2835  // Emit the primary warning first, then try to emit a fixit hint note if
2836  // reasonable.
2837  S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
2838    << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext);
2839
2840  const llvm::APFloat &Value = FL->getValue();
2841
2842  // Don't attempt to fix PPC double double literals.
2843  if (&Value.getSemantics() == &llvm::APFloat::PPCDoubleDouble)
2844    return;
2845
2846  // Try to convert this exactly to an 64-bit integer. FIXME: It would be
2847  // nice to support arbitrarily large integers here.
2848  bool isExact = false;
2849  uint64_t IntegerPart;
2850  if (Value.convertToInteger(&IntegerPart, 64, /*isSigned=*/true,
2851                             llvm::APFloat::rmTowardZero, &isExact)
2852      != llvm::APFloat::opOK || !isExact)
2853    return;
2854
2855  llvm::APInt IntegerValue(64, IntegerPart, /*isSigned=*/true);
2856
2857  std::string LiteralValue = IntegerValue.toString(10, /*isSigned=*/true);
2858  S.Diag(FL->getExprLoc(), diag::note_fix_integral_float_as_integer)
2859    << FixItHint::CreateReplacement(FL->getSourceRange(), LiteralValue);
2860}
2861
2862std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
2863  if (!Range.Width) return "0";
2864
2865  llvm::APSInt ValueInRange = Value;
2866  ValueInRange.setIsSigned(!Range.NonNegative);
2867  ValueInRange = ValueInRange.trunc(Range.Width);
2868  return ValueInRange.toString(10);
2869}
2870
2871static bool isFromSystemMacro(Sema &S, SourceLocation loc) {
2872  SourceManager &smgr = S.Context.getSourceManager();
2873  return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc));
2874}
2875
2876void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
2877                             SourceLocation CC, bool *ICContext = 0) {
2878  if (E->isTypeDependent() || E->isValueDependent()) return;
2879
2880  const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
2881  const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
2882  if (Source == Target) return;
2883  if (Target->isDependentType()) return;
2884
2885  // If the conversion context location is invalid don't complain.
2886  // We also don't want to emit a warning if the issue occurs from the
2887  // instantiation of a system macro.  The problem is that 'getSpellingLoc()'
2888  // is slow, so we delay this check as long as possible.  Once we detect
2889  // we are in that scenario, we just return.
2890  if (CC.isInvalid())
2891    return;
2892
2893  // Never diagnose implicit casts to bool.
2894  if (Target->isSpecificBuiltinType(BuiltinType::Bool))
2895    return;
2896
2897  // Strip vector types.
2898  if (isa<VectorType>(Source)) {
2899    if (!isa<VectorType>(Target)) {
2900      if (isFromSystemMacro(S, CC))
2901        return;
2902      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
2903    }
2904
2905    Source = cast<VectorType>(Source)->getElementType().getTypePtr();
2906    Target = cast<VectorType>(Target)->getElementType().getTypePtr();
2907  }
2908
2909  // Strip complex types.
2910  if (isa<ComplexType>(Source)) {
2911    if (!isa<ComplexType>(Target)) {
2912      if (isFromSystemMacro(S, CC))
2913        return;
2914
2915      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
2916    }
2917
2918    Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
2919    Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
2920  }
2921
2922  const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
2923  const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
2924
2925  // If the source is floating point...
2926  if (SourceBT && SourceBT->isFloatingPoint()) {
2927    // ...and the target is floating point...
2928    if (TargetBT && TargetBT->isFloatingPoint()) {
2929      // ...then warn if we're dropping FP rank.
2930
2931      // Builtin FP kinds are ordered by increasing FP rank.
2932      if (SourceBT->getKind() > TargetBT->getKind()) {
2933        // Don't warn about float constants that are precisely
2934        // representable in the target type.
2935        Expr::EvalResult result;
2936        if (E->Evaluate(result, S.Context)) {
2937          // Value might be a float, a float vector, or a float complex.
2938          if (IsSameFloatAfterCast(result.Val,
2939                   S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
2940                   S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
2941            return;
2942        }
2943
2944        if (isFromSystemMacro(S, CC))
2945          return;
2946
2947        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
2948      }
2949      return;
2950    }
2951
2952    // If the target is integral, always warn.
2953    if ((TargetBT && TargetBT->isInteger())) {
2954      if (isFromSystemMacro(S, CC))
2955        return;
2956
2957      Expr *InnerE = E->IgnoreParenImpCasts();
2958      if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
2959        DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
2960      } else {
2961        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
2962      }
2963    }
2964
2965    return;
2966  }
2967
2968  if (!Source->isIntegerType() || !Target->isIntegerType())
2969    return;
2970
2971  IntRange SourceRange = GetExprRange(S.Context, E);
2972  IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
2973
2974  if (SourceRange.Width > TargetRange.Width) {
2975    // If the source is a constant, use a default-on diagnostic.
2976    // TODO: this should happen for bitfield stores, too.
2977    llvm::APSInt Value(32);
2978    if (E->isIntegerConstantExpr(Value, S.Context)) {
2979      if (isFromSystemMacro(S, CC))
2980        return;
2981
2982      std::string PrettySourceValue = Value.toString(10);
2983      std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
2984
2985      S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant)
2986        << PrettySourceValue << PrettyTargetValue
2987        << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC);
2988      return;
2989    }
2990
2991    // People want to build with -Wshorten-64-to-32 and not -Wconversion
2992    // and by god we'll let them.
2993
2994    if (isFromSystemMacro(S, CC))
2995      return;
2996
2997    if (SourceRange.Width == 64 && TargetRange.Width == 32)
2998      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32);
2999    return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
3000  }
3001
3002  if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
3003      (!TargetRange.NonNegative && SourceRange.NonNegative &&
3004       SourceRange.Width == TargetRange.Width)) {
3005
3006    if (isFromSystemMacro(S, CC))
3007      return;
3008
3009    unsigned DiagID = diag::warn_impcast_integer_sign;
3010
3011    // Traditionally, gcc has warned about this under -Wsign-compare.
3012    // We also want to warn about it in -Wconversion.
3013    // So if -Wconversion is off, use a completely identical diagnostic
3014    // in the sign-compare group.
3015    // The conditional-checking code will
3016    if (ICContext) {
3017      DiagID = diag::warn_impcast_integer_sign_conditional;
3018      *ICContext = true;
3019    }
3020
3021    return DiagnoseImpCast(S, E, T, CC, DiagID);
3022  }
3023
3024  // Diagnose conversions between different enumeration types.
3025  // In C, we pretend that the type of an EnumConstantDecl is its enumeration
3026  // type, to give us better diagnostics.
3027  QualType SourceType = E->getType();
3028  if (!S.getLangOptions().CPlusPlus) {
3029    if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
3030      if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
3031        EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
3032        SourceType = S.Context.getTypeDeclType(Enum);
3033        Source = S.Context.getCanonicalType(SourceType).getTypePtr();
3034      }
3035  }
3036
3037  if (const EnumType *SourceEnum = Source->getAs<EnumType>())
3038    if (const EnumType *TargetEnum = Target->getAs<EnumType>())
3039      if ((SourceEnum->getDecl()->getIdentifier() ||
3040           SourceEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3041          (TargetEnum->getDecl()->getIdentifier() ||
3042           TargetEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3043          SourceEnum != TargetEnum) {
3044        if (isFromSystemMacro(S, CC))
3045          return;
3046
3047        return DiagnoseImpCast(S, E, SourceType, T, CC,
3048                               diag::warn_impcast_different_enum_types);
3049      }
3050
3051  return;
3052}
3053
3054void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T);
3055
3056void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
3057                             SourceLocation CC, bool &ICContext) {
3058  E = E->IgnoreParenImpCasts();
3059
3060  if (isa<ConditionalOperator>(E))
3061    return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T);
3062
3063  AnalyzeImplicitConversions(S, E, CC);
3064  if (E->getType() != T)
3065    return CheckImplicitConversion(S, E, T, CC, &ICContext);
3066  return;
3067}
3068
3069void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) {
3070  SourceLocation CC = E->getQuestionLoc();
3071
3072  AnalyzeImplicitConversions(S, E->getCond(), CC);
3073
3074  bool Suspicious = false;
3075  CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
3076  CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
3077
3078  // If -Wconversion would have warned about either of the candidates
3079  // for a signedness conversion to the context type...
3080  if (!Suspicious) return;
3081
3082  // ...but it's currently ignored...
3083  if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
3084                                 CC))
3085    return;
3086
3087  // ...and -Wsign-compare isn't...
3088  if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional, CC))
3089    return;
3090
3091  // ...then check whether it would have warned about either of the
3092  // candidates for a signedness conversion to the condition type.
3093  if (E->getType() != T) {
3094    Suspicious = false;
3095    CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
3096                            E->getType(), CC, &Suspicious);
3097    if (!Suspicious)
3098      CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
3099                              E->getType(), CC, &Suspicious);
3100    if (!Suspicious)
3101      return;
3102  }
3103
3104  // If so, emit a diagnostic under -Wsign-compare.
3105  Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts();
3106  Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts();
3107  S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional)
3108    << lex->getType() << rex->getType()
3109    << lex->getSourceRange() << rex->getSourceRange();
3110}
3111
3112/// AnalyzeImplicitConversions - Find and report any interesting
3113/// implicit conversions in the given expression.  There are a couple
3114/// of competing diagnostics here, -Wconversion and -Wsign-compare.
3115void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
3116  QualType T = OrigE->getType();
3117  Expr *E = OrigE->IgnoreParenImpCasts();
3118
3119  // For conditional operators, we analyze the arguments as if they
3120  // were being fed directly into the output.
3121  if (isa<ConditionalOperator>(E)) {
3122    ConditionalOperator *CO = cast<ConditionalOperator>(E);
3123    CheckConditionalOperator(S, CO, T);
3124    return;
3125  }
3126
3127  // Go ahead and check any implicit conversions we might have skipped.
3128  // The non-canonical typecheck is just an optimization;
3129  // CheckImplicitConversion will filter out dead implicit conversions.
3130  if (E->getType() != T)
3131    CheckImplicitConversion(S, E, T, CC);
3132
3133  // Now continue drilling into this expression.
3134
3135  // Skip past explicit casts.
3136  if (isa<ExplicitCastExpr>(E)) {
3137    E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
3138    return AnalyzeImplicitConversions(S, E, CC);
3139  }
3140
3141  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
3142    // Do a somewhat different check with comparison operators.
3143    if (BO->isComparisonOp())
3144      return AnalyzeComparison(S, BO);
3145
3146    // And with assignments and compound assignments.
3147    if (BO->isAssignmentOp())
3148      return AnalyzeAssignment(S, BO);
3149  }
3150
3151  // These break the otherwise-useful invariant below.  Fortunately,
3152  // we don't really need to recurse into them, because any internal
3153  // expressions should have been analyzed already when they were
3154  // built into statements.
3155  if (isa<StmtExpr>(E)) return;
3156
3157  // Don't descend into unevaluated contexts.
3158  if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
3159
3160  // Now just recurse over the expression's children.
3161  CC = E->getExprLoc();
3162  for (Stmt::child_range I = E->children(); I; ++I)
3163    AnalyzeImplicitConversions(S, cast<Expr>(*I), CC);
3164}
3165
3166} // end anonymous namespace
3167
3168/// Diagnoses "dangerous" implicit conversions within the given
3169/// expression (which is a full expression).  Implements -Wconversion
3170/// and -Wsign-compare.
3171///
3172/// \param CC the "context" location of the implicit conversion, i.e.
3173///   the most location of the syntactic entity requiring the implicit
3174///   conversion
3175void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
3176  // Don't diagnose in unevaluated contexts.
3177  if (ExprEvalContexts.back().Context == Sema::Unevaluated)
3178    return;
3179
3180  // Don't diagnose for value- or type-dependent expressions.
3181  if (E->isTypeDependent() || E->isValueDependent())
3182    return;
3183
3184  // This is not the right CC for (e.g.) a variable initialization.
3185  AnalyzeImplicitConversions(*this, E, CC);
3186}
3187
3188void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
3189                                       FieldDecl *BitField,
3190                                       Expr *Init) {
3191  (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
3192}
3193
3194/// CheckParmsForFunctionDef - Check that the parameters of the given
3195/// function are appropriate for the definition of a function. This
3196/// takes care of any checks that cannot be performed on the
3197/// declaration itself, e.g., that the types of each of the function
3198/// parameters are complete.
3199bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
3200                                    bool CheckParameterNames) {
3201  bool HasInvalidParm = false;
3202  for (; P != PEnd; ++P) {
3203    ParmVarDecl *Param = *P;
3204
3205    // C99 6.7.5.3p4: the parameters in a parameter type list in a
3206    // function declarator that is part of a function definition of
3207    // that function shall not have incomplete type.
3208    //
3209    // This is also C++ [dcl.fct]p6.
3210    if (!Param->isInvalidDecl() &&
3211        RequireCompleteType(Param->getLocation(), Param->getType(),
3212                               diag::err_typecheck_decl_incomplete_type)) {
3213      Param->setInvalidDecl();
3214      HasInvalidParm = true;
3215    }
3216
3217    // C99 6.9.1p5: If the declarator includes a parameter type list, the
3218    // declaration of each parameter shall include an identifier.
3219    if (CheckParameterNames &&
3220        Param->getIdentifier() == 0 &&
3221        !Param->isImplicit() &&
3222        !getLangOptions().CPlusPlus)
3223      Diag(Param->getLocation(), diag::err_parameter_name_omitted);
3224
3225    // C99 6.7.5.3p12:
3226    //   If the function declarator is not part of a definition of that
3227    //   function, parameters may have incomplete type and may use the [*]
3228    //   notation in their sequences of declarator specifiers to specify
3229    //   variable length array types.
3230    QualType PType = Param->getOriginalType();
3231    if (const ArrayType *AT = Context.getAsArrayType(PType)) {
3232      if (AT->getSizeModifier() == ArrayType::Star) {
3233        // FIXME: This diagnosic should point the the '[*]' if source-location
3234        // information is added for it.
3235        Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
3236      }
3237    }
3238  }
3239
3240  return HasInvalidParm;
3241}
3242
3243/// CheckCastAlign - Implements -Wcast-align, which warns when a
3244/// pointer cast increases the alignment requirements.
3245void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
3246  // This is actually a lot of work to potentially be doing on every
3247  // cast; don't do it if we're ignoring -Wcast_align (as is the default).
3248  if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
3249                                          TRange.getBegin())
3250        == Diagnostic::Ignored)
3251    return;
3252
3253  // Ignore dependent types.
3254  if (T->isDependentType() || Op->getType()->isDependentType())
3255    return;
3256
3257  // Require that the destination be a pointer type.
3258  const PointerType *DestPtr = T->getAs<PointerType>();
3259  if (!DestPtr) return;
3260
3261  // If the destination has alignment 1, we're done.
3262  QualType DestPointee = DestPtr->getPointeeType();
3263  if (DestPointee->isIncompleteType()) return;
3264  CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
3265  if (DestAlign.isOne()) return;
3266
3267  // Require that the source be a pointer type.
3268  const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
3269  if (!SrcPtr) return;
3270  QualType SrcPointee = SrcPtr->getPointeeType();
3271
3272  // Whitelist casts from cv void*.  We already implicitly
3273  // whitelisted casts to cv void*, since they have alignment 1.
3274  // Also whitelist casts involving incomplete types, which implicitly
3275  // includes 'void'.
3276  if (SrcPointee->isIncompleteType()) return;
3277
3278  CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
3279  if (SrcAlign >= DestAlign) return;
3280
3281  Diag(TRange.getBegin(), diag::warn_cast_align)
3282    << Op->getType() << T
3283    << static_cast<unsigned>(SrcAlign.getQuantity())
3284    << static_cast<unsigned>(DestAlign.getQuantity())
3285    << TRange << Op->getSourceRange();
3286}
3287
3288static void CheckArrayAccess_Check(Sema &S,
3289                                   const clang::ArraySubscriptExpr *E) {
3290  const Expr *BaseExpr = E->getBase()->IgnoreParenImpCasts();
3291  const ConstantArrayType *ArrayTy =
3292    S.Context.getAsConstantArrayType(BaseExpr->getType());
3293  if (!ArrayTy)
3294    return;
3295
3296  const Expr *IndexExpr = E->getIdx();
3297  if (IndexExpr->isValueDependent())
3298    return;
3299  llvm::APSInt index;
3300  if (!IndexExpr->isIntegerConstantExpr(index, S.Context))
3301    return;
3302
3303  if (index.isUnsigned() || !index.isNegative()) {
3304    llvm::APInt size = ArrayTy->getSize();
3305    if (!size.isStrictlyPositive())
3306      return;
3307    if (size.getBitWidth() > index.getBitWidth())
3308      index = index.sext(size.getBitWidth());
3309    else if (size.getBitWidth() < index.getBitWidth())
3310      size = size.sext(index.getBitWidth());
3311
3312    if (index.slt(size))
3313      return;
3314
3315    S.DiagRuntimeBehavior(E->getBase()->getLocStart(), BaseExpr,
3316                          S.PDiag(diag::warn_array_index_exceeds_bounds)
3317                            << index.toString(10, true)
3318                            << size.toString(10, true)
3319                            << IndexExpr->getSourceRange());
3320  } else {
3321    S.DiagRuntimeBehavior(E->getBase()->getLocStart(), BaseExpr,
3322                          S.PDiag(diag::warn_array_index_precedes_bounds)
3323                            << index.toString(10, true)
3324                            << IndexExpr->getSourceRange());
3325  }
3326
3327  const NamedDecl *ND = NULL;
3328  if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
3329    ND = dyn_cast<NamedDecl>(DRE->getDecl());
3330  if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
3331    ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
3332  if (ND)
3333    S.DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
3334                          S.PDiag(diag::note_array_index_out_of_bounds)
3335                            << ND->getDeclName());
3336}
3337
3338void Sema::CheckArrayAccess(const Expr *expr) {
3339  while (true) {
3340    expr = expr->IgnoreParens();
3341    switch (expr->getStmtClass()) {
3342      case Stmt::ArraySubscriptExprClass:
3343        CheckArrayAccess_Check(*this, cast<ArraySubscriptExpr>(expr));
3344        return;
3345      case Stmt::ConditionalOperatorClass: {
3346        const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
3347        if (const Expr *lhs = cond->getLHS())
3348          CheckArrayAccess(lhs);
3349        if (const Expr *rhs = cond->getRHS())
3350          CheckArrayAccess(rhs);
3351        return;
3352      }
3353      default:
3354        return;
3355    }
3356  }
3357}
3358