SemaChecking.cpp revision 2a053a33707eba9e642adedb557b1a6aace103b5
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 handling
322  if (FnInfo->isStr("memset") &&
323      FDecl->getLinkage() == ExternalLinkage &&
324      (!getLangOptions().CPlusPlus || FDecl->isExternC()))
325    CheckMemsetArguments(TheCall);
326
327  return false;
328}
329
330bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
331  // Printf checking.
332  const FormatAttr *Format = NDecl->getAttr<FormatAttr>();
333  if (!Format)
334    return false;
335
336  const VarDecl *V = dyn_cast<VarDecl>(NDecl);
337  if (!V)
338    return false;
339
340  QualType Ty = V->getType();
341  if (!Ty->isBlockPointerType())
342    return false;
343
344  const bool b = Format->getType() == "scanf";
345  if (!b && !CheckablePrintfAttr(Format, TheCall))
346    return false;
347
348  bool HasVAListArg = Format->getFirstArg() == 0;
349  CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
350                            HasVAListArg ? 0 : Format->getFirstArg() - 1, !b);
351
352  return false;
353}
354
355/// SemaBuiltinAtomicOverloaded - We have a call to a function like
356/// __sync_fetch_and_add, which is an overloaded function based on the pointer
357/// type of its first argument.  The main ActOnCallExpr routines have already
358/// promoted the types of arguments because all of these calls are prototyped as
359/// void(...).
360///
361/// This function goes through and does final semantic checking for these
362/// builtins,
363ExprResult
364Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
365  CallExpr *TheCall = (CallExpr *)TheCallResult.get();
366  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
367  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
368
369  // Ensure that we have at least one argument to do type inference from.
370  if (TheCall->getNumArgs() < 1) {
371    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
372      << 0 << 1 << TheCall->getNumArgs()
373      << TheCall->getCallee()->getSourceRange();
374    return ExprError();
375  }
376
377  // Inspect the first argument of the atomic builtin.  This should always be
378  // a pointer type, whose element is an integral scalar or pointer type.
379  // Because it is a pointer type, we don't have to worry about any implicit
380  // casts here.
381  // FIXME: We don't allow floating point scalars as input.
382  Expr *FirstArg = TheCall->getArg(0);
383  if (!FirstArg->getType()->isPointerType()) {
384    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
385      << FirstArg->getType() << FirstArg->getSourceRange();
386    return ExprError();
387  }
388
389  QualType ValType =
390    FirstArg->getType()->getAs<PointerType>()->getPointeeType();
391  if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
392      !ValType->isBlockPointerType()) {
393    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
394      << FirstArg->getType() << FirstArg->getSourceRange();
395    return ExprError();
396  }
397
398  // The majority of builtins return a value, but a few have special return
399  // types, so allow them to override appropriately below.
400  QualType ResultType = ValType;
401
402  // We need to figure out which concrete builtin this maps onto.  For example,
403  // __sync_fetch_and_add with a 2 byte object turns into
404  // __sync_fetch_and_add_2.
405#define BUILTIN_ROW(x) \
406  { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
407    Builtin::BI##x##_8, Builtin::BI##x##_16 }
408
409  static const unsigned BuiltinIndices[][5] = {
410    BUILTIN_ROW(__sync_fetch_and_add),
411    BUILTIN_ROW(__sync_fetch_and_sub),
412    BUILTIN_ROW(__sync_fetch_and_or),
413    BUILTIN_ROW(__sync_fetch_and_and),
414    BUILTIN_ROW(__sync_fetch_and_xor),
415
416    BUILTIN_ROW(__sync_add_and_fetch),
417    BUILTIN_ROW(__sync_sub_and_fetch),
418    BUILTIN_ROW(__sync_and_and_fetch),
419    BUILTIN_ROW(__sync_or_and_fetch),
420    BUILTIN_ROW(__sync_xor_and_fetch),
421
422    BUILTIN_ROW(__sync_val_compare_and_swap),
423    BUILTIN_ROW(__sync_bool_compare_and_swap),
424    BUILTIN_ROW(__sync_lock_test_and_set),
425    BUILTIN_ROW(__sync_lock_release),
426    BUILTIN_ROW(__sync_swap)
427  };
428#undef BUILTIN_ROW
429
430  // Determine the index of the size.
431  unsigned SizeIndex;
432  switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
433  case 1: SizeIndex = 0; break;
434  case 2: SizeIndex = 1; break;
435  case 4: SizeIndex = 2; break;
436  case 8: SizeIndex = 3; break;
437  case 16: SizeIndex = 4; break;
438  default:
439    Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
440      << FirstArg->getType() << FirstArg->getSourceRange();
441    return ExprError();
442  }
443
444  // Each of these builtins has one pointer argument, followed by some number of
445  // values (0, 1 or 2) followed by a potentially empty varags list of stuff
446  // that we ignore.  Find out which row of BuiltinIndices to read from as well
447  // as the number of fixed args.
448  unsigned BuiltinID = FDecl->getBuiltinID();
449  unsigned BuiltinIndex, NumFixed = 1;
450  switch (BuiltinID) {
451  default: assert(0 && "Unknown overloaded atomic builtin!");
452  case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break;
453  case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break;
454  case Builtin::BI__sync_fetch_and_or:  BuiltinIndex = 2; break;
455  case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break;
456  case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break;
457
458  case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break;
459  case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break;
460  case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break;
461  case Builtin::BI__sync_or_and_fetch:  BuiltinIndex = 8; break;
462  case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break;
463
464  case Builtin::BI__sync_val_compare_and_swap:
465    BuiltinIndex = 10;
466    NumFixed = 2;
467    break;
468  case Builtin::BI__sync_bool_compare_and_swap:
469    BuiltinIndex = 11;
470    NumFixed = 2;
471    ResultType = Context.BoolTy;
472    break;
473  case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break;
474  case Builtin::BI__sync_lock_release:
475    BuiltinIndex = 13;
476    NumFixed = 0;
477    ResultType = Context.VoidTy;
478    break;
479  case Builtin::BI__sync_swap: BuiltinIndex = 14; break;
480  }
481
482  // Now that we know how many fixed arguments we expect, first check that we
483  // have at least that many.
484  if (TheCall->getNumArgs() < 1+NumFixed) {
485    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
486      << 0 << 1+NumFixed << TheCall->getNumArgs()
487      << TheCall->getCallee()->getSourceRange();
488    return ExprError();
489  }
490
491  // Get the decl for the concrete builtin from this, we can tell what the
492  // concrete integer type we should convert to is.
493  unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
494  const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
495  IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
496  FunctionDecl *NewBuiltinDecl =
497    cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
498                                           TUScope, false, DRE->getLocStart()));
499
500  // The first argument --- the pointer --- has a fixed type; we
501  // deduce the types of the rest of the arguments accordingly.  Walk
502  // the remaining arguments, converting them to the deduced value type.
503  for (unsigned i = 0; i != NumFixed; ++i) {
504    ExprResult Arg = TheCall->getArg(i+1);
505
506    // If the argument is an implicit cast, then there was a promotion due to
507    // "...", just remove it now.
508    if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) {
509      Arg = ICE->getSubExpr();
510      ICE->setSubExpr(0);
511      TheCall->setArg(i+1, Arg.get());
512    }
513
514    // GCC does an implicit conversion to the pointer or integer ValType.  This
515    // can fail in some cases (1i -> int**), check for this error case now.
516    CastKind Kind = CK_Invalid;
517    ExprValueKind VK = VK_RValue;
518    CXXCastPath BasePath;
519    Arg = CheckCastTypes(Arg.get()->getSourceRange(), ValType, Arg.take(), Kind, VK, BasePath);
520    if (Arg.isInvalid())
521      return ExprError();
522
523    // Okay, we have something that *can* be converted to the right type.  Check
524    // to see if there is a potentially weird extension going on here.  This can
525    // happen when you do an atomic operation on something like an char* and
526    // pass in 42.  The 42 gets converted to char.  This is even more strange
527    // for things like 45.123 -> char, etc.
528    // FIXME: Do this check.
529    Arg = ImpCastExprToType(Arg.take(), ValType, Kind, VK, &BasePath);
530    TheCall->setArg(i+1, Arg.get());
531  }
532
533  // Switch the DeclRefExpr to refer to the new decl.
534  DRE->setDecl(NewBuiltinDecl);
535  DRE->setType(NewBuiltinDecl->getType());
536
537  // Set the callee in the CallExpr.
538  // FIXME: This leaks the original parens and implicit casts.
539  ExprResult PromotedCall = UsualUnaryConversions(DRE);
540  if (PromotedCall.isInvalid())
541    return ExprError();
542  TheCall->setCallee(PromotedCall.take());
543
544  // Change the result type of the call to match the original value type. This
545  // is arbitrary, but the codegen for these builtins ins design to handle it
546  // gracefully.
547  TheCall->setType(ResultType);
548
549  return move(TheCallResult);
550}
551
552
553/// CheckObjCString - Checks that the argument to the builtin
554/// CFString constructor is correct
555/// Note: It might also make sense to do the UTF-16 conversion here (would
556/// simplify the backend).
557bool Sema::CheckObjCString(Expr *Arg) {
558  Arg = Arg->IgnoreParenCasts();
559  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
560
561  if (!Literal || Literal->isWide()) {
562    Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
563      << Arg->getSourceRange();
564    return true;
565  }
566
567  if (Literal->containsNonAsciiOrNull()) {
568    llvm::StringRef String = Literal->getString();
569    unsigned NumBytes = String.size();
570    llvm::SmallVector<UTF16, 128> ToBuf(NumBytes);
571    const UTF8 *FromPtr = (UTF8 *)String.data();
572    UTF16 *ToPtr = &ToBuf[0];
573
574    ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
575                                                 &ToPtr, ToPtr + NumBytes,
576                                                 strictConversion);
577    // Check for conversion failure.
578    if (Result != conversionOK)
579      Diag(Arg->getLocStart(),
580           diag::warn_cfstring_truncated) << Arg->getSourceRange();
581  }
582  return false;
583}
584
585/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
586/// Emit an error and return true on failure, return false on success.
587bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
588  Expr *Fn = TheCall->getCallee();
589  if (TheCall->getNumArgs() > 2) {
590    Diag(TheCall->getArg(2)->getLocStart(),
591         diag::err_typecheck_call_too_many_args)
592      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
593      << Fn->getSourceRange()
594      << SourceRange(TheCall->getArg(2)->getLocStart(),
595                     (*(TheCall->arg_end()-1))->getLocEnd());
596    return true;
597  }
598
599  if (TheCall->getNumArgs() < 2) {
600    return Diag(TheCall->getLocEnd(),
601      diag::err_typecheck_call_too_few_args_at_least)
602      << 0 /*function call*/ << 2 << TheCall->getNumArgs();
603  }
604
605  // Determine whether the current function is variadic or not.
606  BlockScopeInfo *CurBlock = getCurBlock();
607  bool isVariadic;
608  if (CurBlock)
609    isVariadic = CurBlock->TheDecl->isVariadic();
610  else if (FunctionDecl *FD = getCurFunctionDecl())
611    isVariadic = FD->isVariadic();
612  else
613    isVariadic = getCurMethodDecl()->isVariadic();
614
615  if (!isVariadic) {
616    Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
617    return true;
618  }
619
620  // Verify that the second argument to the builtin is the last argument of the
621  // current function or method.
622  bool SecondArgIsLastNamedArgument = false;
623  const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
624
625  if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
626    if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
627      // FIXME: This isn't correct for methods (results in bogus warning).
628      // Get the last formal in the current function.
629      const ParmVarDecl *LastArg;
630      if (CurBlock)
631        LastArg = *(CurBlock->TheDecl->param_end()-1);
632      else if (FunctionDecl *FD = getCurFunctionDecl())
633        LastArg = *(FD->param_end()-1);
634      else
635        LastArg = *(getCurMethodDecl()->param_end()-1);
636      SecondArgIsLastNamedArgument = PV == LastArg;
637    }
638  }
639
640  if (!SecondArgIsLastNamedArgument)
641    Diag(TheCall->getArg(1)->getLocStart(),
642         diag::warn_second_parameter_of_va_start_not_last_named_argument);
643  return false;
644}
645
646/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
647/// friends.  This is declared to take (...), so we have to check everything.
648bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
649  if (TheCall->getNumArgs() < 2)
650    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
651      << 0 << 2 << TheCall->getNumArgs()/*function call*/;
652  if (TheCall->getNumArgs() > 2)
653    return Diag(TheCall->getArg(2)->getLocStart(),
654                diag::err_typecheck_call_too_many_args)
655      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
656      << SourceRange(TheCall->getArg(2)->getLocStart(),
657                     (*(TheCall->arg_end()-1))->getLocEnd());
658
659  ExprResult OrigArg0 = TheCall->getArg(0);
660  ExprResult OrigArg1 = TheCall->getArg(1);
661
662  // Do standard promotions between the two arguments, returning their common
663  // type.
664  QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
665  if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
666    return true;
667
668  // Make sure any conversions are pushed back into the call; this is
669  // type safe since unordered compare builtins are declared as "_Bool
670  // foo(...)".
671  TheCall->setArg(0, OrigArg0.get());
672  TheCall->setArg(1, OrigArg1.get());
673
674  if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
675    return false;
676
677  // If the common type isn't a real floating type, then the arguments were
678  // invalid for this operation.
679  if (!Res->isRealFloatingType())
680    return Diag(OrigArg0.get()->getLocStart(),
681                diag::err_typecheck_call_invalid_ordered_compare)
682      << OrigArg0.get()->getType() << OrigArg1.get()->getType()
683      << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
684
685  return false;
686}
687
688/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
689/// __builtin_isnan and friends.  This is declared to take (...), so we have
690/// to check everything. We expect the last argument to be a floating point
691/// value.
692bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
693  if (TheCall->getNumArgs() < NumArgs)
694    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
695      << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
696  if (TheCall->getNumArgs() > NumArgs)
697    return Diag(TheCall->getArg(NumArgs)->getLocStart(),
698                diag::err_typecheck_call_too_many_args)
699      << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
700      << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
701                     (*(TheCall->arg_end()-1))->getLocEnd());
702
703  Expr *OrigArg = TheCall->getArg(NumArgs-1);
704
705  if (OrigArg->isTypeDependent())
706    return false;
707
708  // This operation requires a non-_Complex floating-point number.
709  if (!OrigArg->getType()->isRealFloatingType())
710    return Diag(OrigArg->getLocStart(),
711                diag::err_typecheck_call_invalid_unary_fp)
712      << OrigArg->getType() << OrigArg->getSourceRange();
713
714  // If this is an implicit conversion from float -> double, remove it.
715  if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
716    Expr *CastArg = Cast->getSubExpr();
717    if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
718      assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
719             "promotion from float to double is the only expected cast here");
720      Cast->setSubExpr(0);
721      TheCall->setArg(NumArgs-1, CastArg);
722      OrigArg = CastArg;
723    }
724  }
725
726  return false;
727}
728
729/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
730// This is declared to take (...), so we have to check everything.
731ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
732  if (TheCall->getNumArgs() < 2)
733    return ExprError(Diag(TheCall->getLocEnd(),
734                          diag::err_typecheck_call_too_few_args_at_least)
735      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
736      << TheCall->getSourceRange());
737
738  // Determine which of the following types of shufflevector we're checking:
739  // 1) unary, vector mask: (lhs, mask)
740  // 2) binary, vector mask: (lhs, rhs, mask)
741  // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
742  QualType resType = TheCall->getArg(0)->getType();
743  unsigned numElements = 0;
744
745  if (!TheCall->getArg(0)->isTypeDependent() &&
746      !TheCall->getArg(1)->isTypeDependent()) {
747    QualType LHSType = TheCall->getArg(0)->getType();
748    QualType RHSType = TheCall->getArg(1)->getType();
749
750    if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
751      Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
752        << SourceRange(TheCall->getArg(0)->getLocStart(),
753                       TheCall->getArg(1)->getLocEnd());
754      return ExprError();
755    }
756
757    numElements = LHSType->getAs<VectorType>()->getNumElements();
758    unsigned numResElements = TheCall->getNumArgs() - 2;
759
760    // Check to see if we have a call with 2 vector arguments, the unary shuffle
761    // with mask.  If so, verify that RHS is an integer vector type with the
762    // same number of elts as lhs.
763    if (TheCall->getNumArgs() == 2) {
764      if (!RHSType->hasIntegerRepresentation() ||
765          RHSType->getAs<VectorType>()->getNumElements() != numElements)
766        Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
767          << SourceRange(TheCall->getArg(1)->getLocStart(),
768                         TheCall->getArg(1)->getLocEnd());
769      numResElements = numElements;
770    }
771    else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
772      Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
773        << SourceRange(TheCall->getArg(0)->getLocStart(),
774                       TheCall->getArg(1)->getLocEnd());
775      return ExprError();
776    } else if (numElements != numResElements) {
777      QualType eltType = LHSType->getAs<VectorType>()->getElementType();
778      resType = Context.getVectorType(eltType, numResElements,
779                                      VectorType::GenericVector);
780    }
781  }
782
783  for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
784    if (TheCall->getArg(i)->isTypeDependent() ||
785        TheCall->getArg(i)->isValueDependent())
786      continue;
787
788    llvm::APSInt Result(32);
789    if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
790      return ExprError(Diag(TheCall->getLocStart(),
791                  diag::err_shufflevector_nonconstant_argument)
792                << TheCall->getArg(i)->getSourceRange());
793
794    if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
795      return ExprError(Diag(TheCall->getLocStart(),
796                  diag::err_shufflevector_argument_too_large)
797               << TheCall->getArg(i)->getSourceRange());
798  }
799
800  llvm::SmallVector<Expr*, 32> exprs;
801
802  for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
803    exprs.push_back(TheCall->getArg(i));
804    TheCall->setArg(i, 0);
805  }
806
807  return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
808                                            exprs.size(), resType,
809                                            TheCall->getCallee()->getLocStart(),
810                                            TheCall->getRParenLoc()));
811}
812
813/// SemaBuiltinPrefetch - Handle __builtin_prefetch.
814// This is declared to take (const void*, ...) and can take two
815// optional constant int args.
816bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
817  unsigned NumArgs = TheCall->getNumArgs();
818
819  if (NumArgs > 3)
820    return Diag(TheCall->getLocEnd(),
821             diag::err_typecheck_call_too_many_args_at_most)
822             << 0 /*function call*/ << 3 << NumArgs
823             << TheCall->getSourceRange();
824
825  // Argument 0 is checked for us and the remaining arguments must be
826  // constant integers.
827  for (unsigned i = 1; i != NumArgs; ++i) {
828    Expr *Arg = TheCall->getArg(i);
829
830    llvm::APSInt Result;
831    if (SemaBuiltinConstantArg(TheCall, i, Result))
832      return true;
833
834    // FIXME: gcc issues a warning and rewrites these to 0. These
835    // seems especially odd for the third argument since the default
836    // is 3.
837    if (i == 1) {
838      if (Result.getLimitedValue() > 1)
839        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
840             << "0" << "1" << Arg->getSourceRange();
841    } else {
842      if (Result.getLimitedValue() > 3)
843        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
844            << "0" << "3" << Arg->getSourceRange();
845    }
846  }
847
848  return false;
849}
850
851/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
852/// TheCall is a constant expression.
853bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
854                                  llvm::APSInt &Result) {
855  Expr *Arg = TheCall->getArg(ArgNum);
856  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
857  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
858
859  if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
860
861  if (!Arg->isIntegerConstantExpr(Result, Context))
862    return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
863                << FDecl->getDeclName() <<  Arg->getSourceRange();
864
865  return false;
866}
867
868/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
869/// int type). This simply type checks that type is one of the defined
870/// constants (0-3).
871// For compatibility check 0-3, llvm only handles 0 and 2.
872bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
873  llvm::APSInt Result;
874
875  // Check constant-ness first.
876  if (SemaBuiltinConstantArg(TheCall, 1, Result))
877    return true;
878
879  Expr *Arg = TheCall->getArg(1);
880  if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
881    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
882             << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
883  }
884
885  return false;
886}
887
888/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
889/// This checks that val is a constant 1.
890bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
891  Expr *Arg = TheCall->getArg(1);
892  llvm::APSInt Result;
893
894  // TODO: This is less than ideal. Overload this to take a value.
895  if (SemaBuiltinConstantArg(TheCall, 1, Result))
896    return true;
897
898  if (Result != 1)
899    return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
900             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
901
902  return false;
903}
904
905// Handle i > 1 ? "x" : "y", recursively.
906bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall,
907                                  bool HasVAListArg,
908                                  unsigned format_idx, unsigned firstDataArg,
909                                  bool isPrintf) {
910 tryAgain:
911  if (E->isTypeDependent() || E->isValueDependent())
912    return false;
913
914  E = E->IgnoreParens();
915
916  switch (E->getStmtClass()) {
917  case Stmt::BinaryConditionalOperatorClass:
918  case Stmt::ConditionalOperatorClass: {
919    const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E);
920    return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg,
921                                  format_idx, firstDataArg, isPrintf)
922        && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg,
923                                  format_idx, firstDataArg, isPrintf);
924  }
925
926  case Stmt::IntegerLiteralClass:
927    // Technically -Wformat-nonliteral does not warn about this case.
928    // The behavior of printf and friends in this case is implementation
929    // dependent.  Ideally if the format string cannot be null then
930    // it should have a 'nonnull' attribute in the function prototype.
931    return true;
932
933  case Stmt::ImplicitCastExprClass: {
934    E = cast<ImplicitCastExpr>(E)->getSubExpr();
935    goto tryAgain;
936  }
937
938  case Stmt::OpaqueValueExprClass:
939    if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
940      E = src;
941      goto tryAgain;
942    }
943    return false;
944
945  case Stmt::PredefinedExprClass:
946    // While __func__, etc., are technically not string literals, they
947    // cannot contain format specifiers and thus are not a security
948    // liability.
949    return true;
950
951  case Stmt::DeclRefExprClass: {
952    const DeclRefExpr *DR = cast<DeclRefExpr>(E);
953
954    // As an exception, do not flag errors for variables binding to
955    // const string literals.
956    if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
957      bool isConstant = false;
958      QualType T = DR->getType();
959
960      if (const ArrayType *AT = Context.getAsArrayType(T)) {
961        isConstant = AT->getElementType().isConstant(Context);
962      } else if (const PointerType *PT = T->getAs<PointerType>()) {
963        isConstant = T.isConstant(Context) &&
964                     PT->getPointeeType().isConstant(Context);
965      }
966
967      if (isConstant) {
968        if (const Expr *Init = VD->getAnyInitializer())
969          return SemaCheckStringLiteral(Init, TheCall,
970                                        HasVAListArg, format_idx, firstDataArg,
971                                        isPrintf);
972      }
973
974      // For vprintf* functions (i.e., HasVAListArg==true), we add a
975      // special check to see if the format string is a function parameter
976      // of the function calling the printf function.  If the function
977      // has an attribute indicating it is a printf-like function, then we
978      // should suppress warnings concerning non-literals being used in a call
979      // to a vprintf function.  For example:
980      //
981      // void
982      // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
983      //      va_list ap;
984      //      va_start(ap, fmt);
985      //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
986      //      ...
987      //
988      //
989      //  FIXME: We don't have full attribute support yet, so just check to see
990      //    if the argument is a DeclRefExpr that references a parameter.  We'll
991      //    add proper support for checking the attribute later.
992      if (HasVAListArg)
993        if (isa<ParmVarDecl>(VD))
994          return true;
995    }
996
997    return false;
998  }
999
1000  case Stmt::CallExprClass: {
1001    const CallExpr *CE = cast<CallExpr>(E);
1002    if (const ImplicitCastExpr *ICE
1003          = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
1004      if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
1005        if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
1006          if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
1007            unsigned ArgIndex = FA->getFormatIdx();
1008            const Expr *Arg = CE->getArg(ArgIndex - 1);
1009
1010            return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg,
1011                                          format_idx, firstDataArg, isPrintf);
1012          }
1013        }
1014      }
1015    }
1016
1017    return false;
1018  }
1019  case Stmt::ObjCStringLiteralClass:
1020  case Stmt::StringLiteralClass: {
1021    const StringLiteral *StrE = NULL;
1022
1023    if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1024      StrE = ObjCFExpr->getString();
1025    else
1026      StrE = cast<StringLiteral>(E);
1027
1028    if (StrE) {
1029      CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx,
1030                        firstDataArg, isPrintf);
1031      return true;
1032    }
1033
1034    return false;
1035  }
1036
1037  default:
1038    return false;
1039  }
1040}
1041
1042void
1043Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1044                            const Expr * const *ExprArgs,
1045                            SourceLocation CallSiteLoc) {
1046  for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1047                                  e = NonNull->args_end();
1048       i != e; ++i) {
1049    const Expr *ArgExpr = ExprArgs[*i];
1050    if (ArgExpr->isNullPointerConstant(Context,
1051                                       Expr::NPC_ValueDependentIsNotNull))
1052      Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1053  }
1054}
1055
1056/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar
1057/// functions) for correct use of format strings.
1058void
1059Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg,
1060                                unsigned format_idx, unsigned firstDataArg,
1061                                bool isPrintf) {
1062
1063  const Expr *Fn = TheCall->getCallee();
1064
1065  // The way the format attribute works in GCC, the implicit this argument
1066  // of member functions is counted. However, it doesn't appear in our own
1067  // lists, so decrement format_idx in that case.
1068  if (isa<CXXMemberCallExpr>(TheCall)) {
1069    const CXXMethodDecl *method_decl =
1070      dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl());
1071    if (method_decl && method_decl->isInstance()) {
1072      // Catch a format attribute mistakenly referring to the object argument.
1073      if (format_idx == 0)
1074        return;
1075      --format_idx;
1076      if(firstDataArg != 0)
1077        --firstDataArg;
1078    }
1079  }
1080
1081  // CHECK: printf/scanf-like function is called with no format string.
1082  if (format_idx >= TheCall->getNumArgs()) {
1083    Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string)
1084      << Fn->getSourceRange();
1085    return;
1086  }
1087
1088  const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts();
1089
1090  // CHECK: format string is not a string literal.
1091  //
1092  // Dynamically generated format strings are difficult to
1093  // automatically vet at compile time.  Requiring that format strings
1094  // are string literals: (1) permits the checking of format strings by
1095  // the compiler and thereby (2) can practically remove the source of
1096  // many format string exploits.
1097
1098  // Format string can be either ObjC string (e.g. @"%d") or
1099  // C string (e.g. "%d")
1100  // ObjC string uses the same format specifiers as C string, so we can use
1101  // the same format string checking logic for both ObjC and C strings.
1102  if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx,
1103                             firstDataArg, isPrintf))
1104    return;  // Literal format string found, check done!
1105
1106  // If there are no arguments specified, warn with -Wformat-security, otherwise
1107  // warn only with -Wformat-nonliteral.
1108  if (TheCall->getNumArgs() == format_idx+1)
1109    Diag(TheCall->getArg(format_idx)->getLocStart(),
1110         diag::warn_format_nonliteral_noargs)
1111      << OrigFormatExpr->getSourceRange();
1112  else
1113    Diag(TheCall->getArg(format_idx)->getLocStart(),
1114         diag::warn_format_nonliteral)
1115           << OrigFormatExpr->getSourceRange();
1116}
1117
1118namespace {
1119class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1120protected:
1121  Sema &S;
1122  const StringLiteral *FExpr;
1123  const Expr *OrigFormatExpr;
1124  const unsigned FirstDataArg;
1125  const unsigned NumDataArgs;
1126  const bool IsObjCLiteral;
1127  const char *Beg; // Start of format string.
1128  const bool HasVAListArg;
1129  const CallExpr *TheCall;
1130  unsigned FormatIdx;
1131  llvm::BitVector CoveredArgs;
1132  bool usesPositionalArgs;
1133  bool atFirstArg;
1134public:
1135  CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1136                     const Expr *origFormatExpr, unsigned firstDataArg,
1137                     unsigned numDataArgs, bool isObjCLiteral,
1138                     const char *beg, bool hasVAListArg,
1139                     const CallExpr *theCall, unsigned formatIdx)
1140    : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1141      FirstDataArg(firstDataArg),
1142      NumDataArgs(numDataArgs),
1143      IsObjCLiteral(isObjCLiteral), Beg(beg),
1144      HasVAListArg(hasVAListArg),
1145      TheCall(theCall), FormatIdx(formatIdx),
1146      usesPositionalArgs(false), atFirstArg(true) {
1147        CoveredArgs.resize(numDataArgs);
1148        CoveredArgs.reset();
1149      }
1150
1151  void DoneProcessing();
1152
1153  void HandleIncompleteSpecifier(const char *startSpecifier,
1154                                 unsigned specifierLen);
1155
1156  virtual void HandleInvalidPosition(const char *startSpecifier,
1157                                     unsigned specifierLen,
1158                                     analyze_format_string::PositionContext p);
1159
1160  virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
1161
1162  void HandleNullChar(const char *nullCharacter);
1163
1164protected:
1165  bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
1166                                        const char *startSpec,
1167                                        unsigned specifierLen,
1168                                        const char *csStart, unsigned csLen);
1169
1170  SourceRange getFormatStringRange();
1171  CharSourceRange getSpecifierRange(const char *startSpecifier,
1172                                    unsigned specifierLen);
1173  SourceLocation getLocationOfByte(const char *x);
1174
1175  const Expr *getDataArg(unsigned i) const;
1176
1177  bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
1178                    const analyze_format_string::ConversionSpecifier &CS,
1179                    const char *startSpecifier, unsigned specifierLen,
1180                    unsigned argIndex);
1181};
1182}
1183
1184SourceRange CheckFormatHandler::getFormatStringRange() {
1185  return OrigFormatExpr->getSourceRange();
1186}
1187
1188CharSourceRange CheckFormatHandler::
1189getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
1190  SourceLocation Start = getLocationOfByte(startSpecifier);
1191  SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
1192
1193  // Advance the end SourceLocation by one due to half-open ranges.
1194  End = End.getFileLocWithOffset(1);
1195
1196  return CharSourceRange::getCharRange(Start, End);
1197}
1198
1199SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
1200  return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
1201}
1202
1203void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
1204                                                   unsigned specifierLen){
1205  SourceLocation Loc = getLocationOfByte(startSpecifier);
1206  S.Diag(Loc, diag::warn_printf_incomplete_specifier)
1207    << getSpecifierRange(startSpecifier, specifierLen);
1208}
1209
1210void
1211CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
1212                                     analyze_format_string::PositionContext p) {
1213  SourceLocation Loc = getLocationOfByte(startPos);
1214  S.Diag(Loc, diag::warn_format_invalid_positional_specifier)
1215    << (unsigned) p << getSpecifierRange(startPos, posLen);
1216}
1217
1218void CheckFormatHandler::HandleZeroPosition(const char *startPos,
1219                                            unsigned posLen) {
1220  SourceLocation Loc = getLocationOfByte(startPos);
1221  S.Diag(Loc, diag::warn_format_zero_positional_specifier)
1222    << getSpecifierRange(startPos, posLen);
1223}
1224
1225void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
1226  if (!IsObjCLiteral) {
1227    // The presence of a null character is likely an error.
1228    S.Diag(getLocationOfByte(nullCharacter),
1229           diag::warn_printf_format_string_contains_null_char)
1230      << getFormatStringRange();
1231  }
1232}
1233
1234const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
1235  return TheCall->getArg(FirstDataArg + i);
1236}
1237
1238void CheckFormatHandler::DoneProcessing() {
1239    // Does the number of data arguments exceed the number of
1240    // format conversions in the format string?
1241  if (!HasVAListArg) {
1242      // Find any arguments that weren't covered.
1243    CoveredArgs.flip();
1244    signed notCoveredArg = CoveredArgs.find_first();
1245    if (notCoveredArg >= 0) {
1246      assert((unsigned)notCoveredArg < NumDataArgs);
1247      S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(),
1248             diag::warn_printf_data_arg_not_used)
1249      << getFormatStringRange();
1250    }
1251  }
1252}
1253
1254bool
1255CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
1256                                                     SourceLocation Loc,
1257                                                     const char *startSpec,
1258                                                     unsigned specifierLen,
1259                                                     const char *csStart,
1260                                                     unsigned csLen) {
1261
1262  bool keepGoing = true;
1263  if (argIndex < NumDataArgs) {
1264    // Consider the argument coverered, even though the specifier doesn't
1265    // make sense.
1266    CoveredArgs.set(argIndex);
1267  }
1268  else {
1269    // If argIndex exceeds the number of data arguments we
1270    // don't issue a warning because that is just a cascade of warnings (and
1271    // they may have intended '%%' anyway). We don't want to continue processing
1272    // the format string after this point, however, as we will like just get
1273    // gibberish when trying to match arguments.
1274    keepGoing = false;
1275  }
1276
1277  S.Diag(Loc, diag::warn_format_invalid_conversion)
1278    << llvm::StringRef(csStart, csLen)
1279    << getSpecifierRange(startSpec, specifierLen);
1280
1281  return keepGoing;
1282}
1283
1284bool
1285CheckFormatHandler::CheckNumArgs(
1286  const analyze_format_string::FormatSpecifier &FS,
1287  const analyze_format_string::ConversionSpecifier &CS,
1288  const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
1289
1290  if (argIndex >= NumDataArgs) {
1291    if (FS.usesPositionalArg())  {
1292      S.Diag(getLocationOfByte(CS.getStart()),
1293             diag::warn_printf_positional_arg_exceeds_data_args)
1294      << (argIndex+1) << NumDataArgs
1295      << getSpecifierRange(startSpecifier, specifierLen);
1296    }
1297    else {
1298      S.Diag(getLocationOfByte(CS.getStart()),
1299             diag::warn_printf_insufficient_data_args)
1300      << getSpecifierRange(startSpecifier, specifierLen);
1301    }
1302
1303    return false;
1304  }
1305  return true;
1306}
1307
1308//===--- CHECK: Printf format string checking ------------------------------===//
1309
1310namespace {
1311class CheckPrintfHandler : public CheckFormatHandler {
1312public:
1313  CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
1314                     const Expr *origFormatExpr, unsigned firstDataArg,
1315                     unsigned numDataArgs, bool isObjCLiteral,
1316                     const char *beg, bool hasVAListArg,
1317                     const CallExpr *theCall, unsigned formatIdx)
1318  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1319                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1320                       theCall, formatIdx) {}
1321
1322
1323  bool HandleInvalidPrintfConversionSpecifier(
1324                                      const analyze_printf::PrintfSpecifier &FS,
1325                                      const char *startSpecifier,
1326                                      unsigned specifierLen);
1327
1328  bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
1329                             const char *startSpecifier,
1330                             unsigned specifierLen);
1331
1332  bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
1333                    const char *startSpecifier, unsigned specifierLen);
1334  void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
1335                           const analyze_printf::OptionalAmount &Amt,
1336                           unsigned type,
1337                           const char *startSpecifier, unsigned specifierLen);
1338  void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1339                  const analyze_printf::OptionalFlag &flag,
1340                  const char *startSpecifier, unsigned specifierLen);
1341  void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
1342                         const analyze_printf::OptionalFlag &ignoredFlag,
1343                         const analyze_printf::OptionalFlag &flag,
1344                         const char *startSpecifier, unsigned specifierLen);
1345};
1346}
1347
1348bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
1349                                      const analyze_printf::PrintfSpecifier &FS,
1350                                      const char *startSpecifier,
1351                                      unsigned specifierLen) {
1352  const analyze_printf::PrintfConversionSpecifier &CS =
1353    FS.getConversionSpecifier();
1354
1355  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1356                                          getLocationOfByte(CS.getStart()),
1357                                          startSpecifier, specifierLen,
1358                                          CS.getStart(), CS.getLength());
1359}
1360
1361bool CheckPrintfHandler::HandleAmount(
1362                               const analyze_format_string::OptionalAmount &Amt,
1363                               unsigned k, const char *startSpecifier,
1364                               unsigned specifierLen) {
1365
1366  if (Amt.hasDataArgument()) {
1367    if (!HasVAListArg) {
1368      unsigned argIndex = Amt.getArgIndex();
1369      if (argIndex >= NumDataArgs) {
1370        S.Diag(getLocationOfByte(Amt.getStart()),
1371               diag::warn_printf_asterisk_missing_arg)
1372          << k << getSpecifierRange(startSpecifier, specifierLen);
1373        // Don't do any more checking.  We will just emit
1374        // spurious errors.
1375        return false;
1376      }
1377
1378      // Type check the data argument.  It should be an 'int'.
1379      // Although not in conformance with C99, we also allow the argument to be
1380      // an 'unsigned int' as that is a reasonably safe case.  GCC also
1381      // doesn't emit a warning for that case.
1382      CoveredArgs.set(argIndex);
1383      const Expr *Arg = getDataArg(argIndex);
1384      QualType T = Arg->getType();
1385
1386      const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context);
1387      assert(ATR.isValid());
1388
1389      if (!ATR.matchesType(S.Context, T)) {
1390        S.Diag(getLocationOfByte(Amt.getStart()),
1391               diag::warn_printf_asterisk_wrong_type)
1392          << k
1393          << ATR.getRepresentativeType(S.Context) << T
1394          << getSpecifierRange(startSpecifier, specifierLen)
1395          << Arg->getSourceRange();
1396        // Don't do any more checking.  We will just emit
1397        // spurious errors.
1398        return false;
1399      }
1400    }
1401  }
1402  return true;
1403}
1404
1405void CheckPrintfHandler::HandleInvalidAmount(
1406                                      const analyze_printf::PrintfSpecifier &FS,
1407                                      const analyze_printf::OptionalAmount &Amt,
1408                                      unsigned type,
1409                                      const char *startSpecifier,
1410                                      unsigned specifierLen) {
1411  const analyze_printf::PrintfConversionSpecifier &CS =
1412    FS.getConversionSpecifier();
1413  switch (Amt.getHowSpecified()) {
1414  case analyze_printf::OptionalAmount::Constant:
1415    S.Diag(getLocationOfByte(Amt.getStart()),
1416        diag::warn_printf_nonsensical_optional_amount)
1417      << type
1418      << CS.toString()
1419      << getSpecifierRange(startSpecifier, specifierLen)
1420      << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
1421          Amt.getConstantLength()));
1422    break;
1423
1424  default:
1425    S.Diag(getLocationOfByte(Amt.getStart()),
1426        diag::warn_printf_nonsensical_optional_amount)
1427      << type
1428      << CS.toString()
1429      << getSpecifierRange(startSpecifier, specifierLen);
1430    break;
1431  }
1432}
1433
1434void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1435                                    const analyze_printf::OptionalFlag &flag,
1436                                    const char *startSpecifier,
1437                                    unsigned specifierLen) {
1438  // Warn about pointless flag with a fixit removal.
1439  const analyze_printf::PrintfConversionSpecifier &CS =
1440    FS.getConversionSpecifier();
1441  S.Diag(getLocationOfByte(flag.getPosition()),
1442      diag::warn_printf_nonsensical_flag)
1443    << flag.toString() << CS.toString()
1444    << getSpecifierRange(startSpecifier, specifierLen)
1445    << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1));
1446}
1447
1448void CheckPrintfHandler::HandleIgnoredFlag(
1449                                const analyze_printf::PrintfSpecifier &FS,
1450                                const analyze_printf::OptionalFlag &ignoredFlag,
1451                                const analyze_printf::OptionalFlag &flag,
1452                                const char *startSpecifier,
1453                                unsigned specifierLen) {
1454  // Warn about ignored flag with a fixit removal.
1455  S.Diag(getLocationOfByte(ignoredFlag.getPosition()),
1456      diag::warn_printf_ignored_flag)
1457    << ignoredFlag.toString() << flag.toString()
1458    << getSpecifierRange(startSpecifier, specifierLen)
1459    << FixItHint::CreateRemoval(getSpecifierRange(
1460        ignoredFlag.getPosition(), 1));
1461}
1462
1463bool
1464CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
1465                                            &FS,
1466                                          const char *startSpecifier,
1467                                          unsigned specifierLen) {
1468
1469  using namespace analyze_format_string;
1470  using namespace analyze_printf;
1471  const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
1472
1473  if (FS.consumesDataArgument()) {
1474    if (atFirstArg) {
1475        atFirstArg = false;
1476        usesPositionalArgs = FS.usesPositionalArg();
1477    }
1478    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1479      // Cannot mix-and-match positional and non-positional arguments.
1480      S.Diag(getLocationOfByte(CS.getStart()),
1481             diag::warn_format_mix_positional_nonpositional_args)
1482        << getSpecifierRange(startSpecifier, specifierLen);
1483      return false;
1484    }
1485  }
1486
1487  // First check if the field width, precision, and conversion specifier
1488  // have matching data arguments.
1489  if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
1490                    startSpecifier, specifierLen)) {
1491    return false;
1492  }
1493
1494  if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
1495                    startSpecifier, specifierLen)) {
1496    return false;
1497  }
1498
1499  if (!CS.consumesDataArgument()) {
1500    // FIXME: Technically specifying a precision or field width here
1501    // makes no sense.  Worth issuing a warning at some point.
1502    return true;
1503  }
1504
1505  // Consume the argument.
1506  unsigned argIndex = FS.getArgIndex();
1507  if (argIndex < NumDataArgs) {
1508    // The check to see if the argIndex is valid will come later.
1509    // We set the bit here because we may exit early from this
1510    // function if we encounter some other error.
1511    CoveredArgs.set(argIndex);
1512  }
1513
1514  // Check for using an Objective-C specific conversion specifier
1515  // in a non-ObjC literal.
1516  if (!IsObjCLiteral && CS.isObjCArg()) {
1517    return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
1518                                                  specifierLen);
1519  }
1520
1521  // Check for invalid use of field width
1522  if (!FS.hasValidFieldWidth()) {
1523    HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
1524        startSpecifier, specifierLen);
1525  }
1526
1527  // Check for invalid use of precision
1528  if (!FS.hasValidPrecision()) {
1529    HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
1530        startSpecifier, specifierLen);
1531  }
1532
1533  // Check each flag does not conflict with any other component.
1534  if (!FS.hasValidThousandsGroupingPrefix())
1535    HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
1536  if (!FS.hasValidLeadingZeros())
1537    HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
1538  if (!FS.hasValidPlusPrefix())
1539    HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
1540  if (!FS.hasValidSpacePrefix())
1541    HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
1542  if (!FS.hasValidAlternativeForm())
1543    HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
1544  if (!FS.hasValidLeftJustified())
1545    HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
1546
1547  // Check that flags are not ignored by another flag
1548  if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
1549    HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
1550        startSpecifier, specifierLen);
1551  if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
1552    HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
1553            startSpecifier, specifierLen);
1554
1555  // Check the length modifier is valid with the given conversion specifier.
1556  const LengthModifier &LM = FS.getLengthModifier();
1557  if (!FS.hasValidLengthModifier())
1558    S.Diag(getLocationOfByte(LM.getStart()),
1559        diag::warn_format_nonsensical_length)
1560      << LM.toString() << CS.toString()
1561      << getSpecifierRange(startSpecifier, specifierLen)
1562      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1563          LM.getLength()));
1564
1565  // Are we using '%n'?
1566  if (CS.getKind() == ConversionSpecifier::nArg) {
1567    // Issue a warning about this being a possible security issue.
1568    S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back)
1569      << getSpecifierRange(startSpecifier, specifierLen);
1570    // Continue checking the other format specifiers.
1571    return true;
1572  }
1573
1574  // The remaining checks depend on the data arguments.
1575  if (HasVAListArg)
1576    return true;
1577
1578  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1579    return false;
1580
1581  // Now type check the data expression that matches the
1582  // format specifier.
1583  const Expr *Ex = getDataArg(argIndex);
1584  const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context);
1585  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
1586    // Check if we didn't match because of an implicit cast from a 'char'
1587    // or 'short' to an 'int'.  This is done because printf is a varargs
1588    // function.
1589    if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex))
1590      if (ICE->getType() == S.Context.IntTy) {
1591        // All further checking is done on the subexpression.
1592        Ex = ICE->getSubExpr();
1593        if (ATR.matchesType(S.Context, Ex->getType()))
1594          return true;
1595      }
1596
1597    // We may be able to offer a FixItHint if it is a supported type.
1598    PrintfSpecifier fixedFS = FS;
1599    bool success = fixedFS.fixType(Ex->getType());
1600
1601    if (success) {
1602      // Get the fix string from the fixed format specifier
1603      llvm::SmallString<128> buf;
1604      llvm::raw_svector_ostream os(buf);
1605      fixedFS.toString(os);
1606
1607      // FIXME: getRepresentativeType() perhaps should return a string
1608      // instead of a QualType to better handle when the representative
1609      // type is 'wint_t' (which is defined in the system headers).
1610      S.Diag(getLocationOfByte(CS.getStart()),
1611          diag::warn_printf_conversion_argument_type_mismatch)
1612        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1613        << getSpecifierRange(startSpecifier, specifierLen)
1614        << Ex->getSourceRange()
1615        << FixItHint::CreateReplacement(
1616            getSpecifierRange(startSpecifier, specifierLen),
1617            os.str());
1618    }
1619    else {
1620      S.Diag(getLocationOfByte(CS.getStart()),
1621             diag::warn_printf_conversion_argument_type_mismatch)
1622        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1623        << getSpecifierRange(startSpecifier, specifierLen)
1624        << Ex->getSourceRange();
1625    }
1626  }
1627
1628  return true;
1629}
1630
1631//===--- CHECK: Scanf format string checking ------------------------------===//
1632
1633namespace {
1634class CheckScanfHandler : public CheckFormatHandler {
1635public:
1636  CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
1637                    const Expr *origFormatExpr, unsigned firstDataArg,
1638                    unsigned numDataArgs, bool isObjCLiteral,
1639                    const char *beg, bool hasVAListArg,
1640                    const CallExpr *theCall, unsigned formatIdx)
1641  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1642                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1643                       theCall, formatIdx) {}
1644
1645  bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
1646                            const char *startSpecifier,
1647                            unsigned specifierLen);
1648
1649  bool HandleInvalidScanfConversionSpecifier(
1650          const analyze_scanf::ScanfSpecifier &FS,
1651          const char *startSpecifier,
1652          unsigned specifierLen);
1653
1654  void HandleIncompleteScanList(const char *start, const char *end);
1655};
1656}
1657
1658void CheckScanfHandler::HandleIncompleteScanList(const char *start,
1659                                                 const char *end) {
1660  S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete)
1661    << getSpecifierRange(start, end - start);
1662}
1663
1664bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
1665                                        const analyze_scanf::ScanfSpecifier &FS,
1666                                        const char *startSpecifier,
1667                                        unsigned specifierLen) {
1668
1669  const analyze_scanf::ScanfConversionSpecifier &CS =
1670    FS.getConversionSpecifier();
1671
1672  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1673                                          getLocationOfByte(CS.getStart()),
1674                                          startSpecifier, specifierLen,
1675                                          CS.getStart(), CS.getLength());
1676}
1677
1678bool CheckScanfHandler::HandleScanfSpecifier(
1679                                       const analyze_scanf::ScanfSpecifier &FS,
1680                                       const char *startSpecifier,
1681                                       unsigned specifierLen) {
1682
1683  using namespace analyze_scanf;
1684  using namespace analyze_format_string;
1685
1686  const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
1687
1688  // Handle case where '%' and '*' don't consume an argument.  These shouldn't
1689  // be used to decide if we are using positional arguments consistently.
1690  if (FS.consumesDataArgument()) {
1691    if (atFirstArg) {
1692      atFirstArg = false;
1693      usesPositionalArgs = FS.usesPositionalArg();
1694    }
1695    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1696      // Cannot mix-and-match positional and non-positional arguments.
1697      S.Diag(getLocationOfByte(CS.getStart()),
1698             diag::warn_format_mix_positional_nonpositional_args)
1699        << getSpecifierRange(startSpecifier, specifierLen);
1700      return false;
1701    }
1702  }
1703
1704  // Check if the field with is non-zero.
1705  const OptionalAmount &Amt = FS.getFieldWidth();
1706  if (Amt.getHowSpecified() == OptionalAmount::Constant) {
1707    if (Amt.getConstantAmount() == 0) {
1708      const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
1709                                                   Amt.getConstantLength());
1710      S.Diag(getLocationOfByte(Amt.getStart()),
1711             diag::warn_scanf_nonzero_width)
1712        << R << FixItHint::CreateRemoval(R);
1713    }
1714  }
1715
1716  if (!FS.consumesDataArgument()) {
1717    // FIXME: Technically specifying a precision or field width here
1718    // makes no sense.  Worth issuing a warning at some point.
1719    return true;
1720  }
1721
1722  // Consume the argument.
1723  unsigned argIndex = FS.getArgIndex();
1724  if (argIndex < NumDataArgs) {
1725      // The check to see if the argIndex is valid will come later.
1726      // We set the bit here because we may exit early from this
1727      // function if we encounter some other error.
1728    CoveredArgs.set(argIndex);
1729  }
1730
1731  // Check the length modifier is valid with the given conversion specifier.
1732  const LengthModifier &LM = FS.getLengthModifier();
1733  if (!FS.hasValidLengthModifier()) {
1734    S.Diag(getLocationOfByte(LM.getStart()),
1735           diag::warn_format_nonsensical_length)
1736      << LM.toString() << CS.toString()
1737      << getSpecifierRange(startSpecifier, specifierLen)
1738      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1739                                                    LM.getLength()));
1740  }
1741
1742  // The remaining checks depend on the data arguments.
1743  if (HasVAListArg)
1744    return true;
1745
1746  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1747    return false;
1748
1749  // FIXME: Check that the argument type matches the format specifier.
1750
1751  return true;
1752}
1753
1754void Sema::CheckFormatString(const StringLiteral *FExpr,
1755                             const Expr *OrigFormatExpr,
1756                             const CallExpr *TheCall, bool HasVAListArg,
1757                             unsigned format_idx, unsigned firstDataArg,
1758                             bool isPrintf) {
1759
1760  // CHECK: is the format string a wide literal?
1761  if (FExpr->isWide()) {
1762    Diag(FExpr->getLocStart(),
1763         diag::warn_format_string_is_wide_literal)
1764    << OrigFormatExpr->getSourceRange();
1765    return;
1766  }
1767
1768  // Str - The format string.  NOTE: this is NOT null-terminated!
1769  llvm::StringRef StrRef = FExpr->getString();
1770  const char *Str = StrRef.data();
1771  unsigned StrLen = StrRef.size();
1772
1773  // CHECK: empty format string?
1774  if (StrLen == 0) {
1775    Diag(FExpr->getLocStart(), diag::warn_empty_format_string)
1776    << OrigFormatExpr->getSourceRange();
1777    return;
1778  }
1779
1780  if (isPrintf) {
1781    CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1782                         TheCall->getNumArgs() - firstDataArg,
1783                         isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1784                         HasVAListArg, TheCall, format_idx);
1785
1786    if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen))
1787      H.DoneProcessing();
1788  }
1789  else {
1790    CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1791                        TheCall->getNumArgs() - firstDataArg,
1792                        isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1793                        HasVAListArg, TheCall, format_idx);
1794
1795    if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen))
1796      H.DoneProcessing();
1797  }
1798}
1799
1800//===--- CHECK: Standard memory functions ---------------------------------===//
1801
1802/// \brief Determine whether the given type is a dynamic class type (e.g.,
1803/// whether it has a vtable).
1804static bool isDynamicClassType(QualType T) {
1805  if (CXXRecordDecl *Record = T->getAsCXXRecordDecl())
1806    if (CXXRecordDecl *Definition = Record->getDefinition())
1807      if (Definition->isDynamicClass())
1808        return true;
1809
1810  return false;
1811}
1812
1813/// \brief Check for dangerous or invalid arguments to memset().
1814///
1815/// This issues warnings on known problematic or dangerous or unspecified
1816/// arguments to the standard 'memset' function call.
1817///
1818/// \param Call The call expression to diagnose.
1819void Sema::CheckMemsetArguments(const CallExpr *Call) {
1820  // It is possible to have a non-standard definition of memset.  Validate
1821  // we have the proper number of arguments, and if not, abort further
1822  // checking.
1823  if (Call->getNumArgs() != 3)
1824    return;
1825
1826  const Expr *Dest = Call->getArg(0)->IgnoreParenImpCasts();
1827
1828  QualType DestTy = Dest->getType();
1829  if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
1830    QualType PointeeTy = DestPtrTy->getPointeeType();
1831    if (PointeeTy->isVoidType())
1832      return;
1833
1834    unsigned DiagID = 0;
1835    // Always complain about dynamic classes.
1836    if (isDynamicClassType(PointeeTy))
1837      DiagID = diag::warn_dyn_class_memset;
1838    // Check the C++11 POD definition regardless of language mode; it is more
1839    // relaxed than earlier definitions and we don't want spurious warnings.
1840    else if (!PointeeTy->isCXX11PODType())
1841      DiagID = diag::warn_non_pod_memset;
1842    else
1843      return;
1844
1845    DiagRuntimeBehavior(
1846      Dest->getExprLoc(), Dest,
1847      PDiag(DiagID)
1848        << PointeeTy << Call->getCallee()->getSourceRange());
1849
1850    SourceRange ArgRange = Call->getArg(0)->getSourceRange();
1851    DiagRuntimeBehavior(
1852      Dest->getExprLoc(), Dest,
1853      PDiag(diag::note_non_pod_memset_silence)
1854        << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
1855  }
1856}
1857
1858//===--- CHECK: Return Address of Stack Variable --------------------------===//
1859
1860static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars);
1861static Expr *EvalAddr(Expr* E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars);
1862
1863/// CheckReturnStackAddr - Check if a return statement returns the address
1864///   of a stack variable.
1865void
1866Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
1867                           SourceLocation ReturnLoc) {
1868
1869  Expr *stackE = 0;
1870  llvm::SmallVector<DeclRefExpr *, 8> refVars;
1871
1872  // Perform checking for returned stack addresses, local blocks,
1873  // label addresses or references to temporaries.
1874  if (lhsType->isPointerType() || lhsType->isBlockPointerType()) {
1875    stackE = EvalAddr(RetValExp, refVars);
1876  } else if (lhsType->isReferenceType()) {
1877    stackE = EvalVal(RetValExp, refVars);
1878  }
1879
1880  if (stackE == 0)
1881    return; // Nothing suspicious was found.
1882
1883  SourceLocation diagLoc;
1884  SourceRange diagRange;
1885  if (refVars.empty()) {
1886    diagLoc = stackE->getLocStart();
1887    diagRange = stackE->getSourceRange();
1888  } else {
1889    // We followed through a reference variable. 'stackE' contains the
1890    // problematic expression but we will warn at the return statement pointing
1891    // at the reference variable. We will later display the "trail" of
1892    // reference variables using notes.
1893    diagLoc = refVars[0]->getLocStart();
1894    diagRange = refVars[0]->getSourceRange();
1895  }
1896
1897  if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
1898    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
1899                                             : diag::warn_ret_stack_addr)
1900     << DR->getDecl()->getDeclName() << diagRange;
1901  } else if (isa<BlockExpr>(stackE)) { // local block.
1902    Diag(diagLoc, diag::err_ret_local_block) << diagRange;
1903  } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
1904    Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
1905  } else { // local temporary.
1906    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
1907                                             : diag::warn_ret_local_temp_addr)
1908     << diagRange;
1909  }
1910
1911  // Display the "trail" of reference variables that we followed until we
1912  // found the problematic expression using notes.
1913  for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
1914    VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
1915    // If this var binds to another reference var, show the range of the next
1916    // var, otherwise the var binds to the problematic expression, in which case
1917    // show the range of the expression.
1918    SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
1919                                  : stackE->getSourceRange();
1920    Diag(VD->getLocation(), diag::note_ref_var_local_bind)
1921      << VD->getDeclName() << range;
1922  }
1923}
1924
1925/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
1926///  check if the expression in a return statement evaluates to an address
1927///  to a location on the stack, a local block, an address of a label, or a
1928///  reference to local temporary. The recursion is used to traverse the
1929///  AST of the return expression, with recursion backtracking when we
1930///  encounter a subexpression that (1) clearly does not lead to one of the
1931///  above problematic expressions (2) is something we cannot determine leads to
1932///  a problematic expression based on such local checking.
1933///
1934///  Both EvalAddr and EvalVal follow through reference variables to evaluate
1935///  the expression that they point to. Such variables are added to the
1936///  'refVars' vector so that we know what the reference variable "trail" was.
1937///
1938///  EvalAddr processes expressions that are pointers that are used as
1939///  references (and not L-values).  EvalVal handles all other values.
1940///  At the base case of the recursion is a check for the above problematic
1941///  expressions.
1942///
1943///  This implementation handles:
1944///
1945///   * pointer-to-pointer casts
1946///   * implicit conversions from array references to pointers
1947///   * taking the address of fields
1948///   * arbitrary interplay between "&" and "*" operators
1949///   * pointer arithmetic from an address of a stack variable
1950///   * taking the address of an array element where the array is on the stack
1951static Expr *EvalAddr(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) {
1952  if (E->isTypeDependent())
1953      return NULL;
1954
1955  // We should only be called for evaluating pointer expressions.
1956  assert((E->getType()->isAnyPointerType() ||
1957          E->getType()->isBlockPointerType() ||
1958          E->getType()->isObjCQualifiedIdType()) &&
1959         "EvalAddr only works on pointers");
1960
1961  E = E->IgnoreParens();
1962
1963  // Our "symbolic interpreter" is just a dispatch off the currently
1964  // viewed AST node.  We then recursively traverse the AST by calling
1965  // EvalAddr and EvalVal appropriately.
1966  switch (E->getStmtClass()) {
1967  case Stmt::DeclRefExprClass: {
1968    DeclRefExpr *DR = cast<DeclRefExpr>(E);
1969
1970    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
1971      // If this is a reference variable, follow through to the expression that
1972      // it points to.
1973      if (V->hasLocalStorage() &&
1974          V->getType()->isReferenceType() && V->hasInit()) {
1975        // Add the reference variable to the "trail".
1976        refVars.push_back(DR);
1977        return EvalAddr(V->getInit(), refVars);
1978      }
1979
1980    return NULL;
1981  }
1982
1983  case Stmt::UnaryOperatorClass: {
1984    // The only unary operator that make sense to handle here
1985    // is AddrOf.  All others don't make sense as pointers.
1986    UnaryOperator *U = cast<UnaryOperator>(E);
1987
1988    if (U->getOpcode() == UO_AddrOf)
1989      return EvalVal(U->getSubExpr(), refVars);
1990    else
1991      return NULL;
1992  }
1993
1994  case Stmt::BinaryOperatorClass: {
1995    // Handle pointer arithmetic.  All other binary operators are not valid
1996    // in this context.
1997    BinaryOperator *B = cast<BinaryOperator>(E);
1998    BinaryOperatorKind op = B->getOpcode();
1999
2000    if (op != BO_Add && op != BO_Sub)
2001      return NULL;
2002
2003    Expr *Base = B->getLHS();
2004
2005    // Determine which argument is the real pointer base.  It could be
2006    // the RHS argument instead of the LHS.
2007    if (!Base->getType()->isPointerType()) Base = B->getRHS();
2008
2009    assert (Base->getType()->isPointerType());
2010    return EvalAddr(Base, refVars);
2011  }
2012
2013  // For conditional operators we need to see if either the LHS or RHS are
2014  // valid DeclRefExpr*s.  If one of them is valid, we return it.
2015  case Stmt::ConditionalOperatorClass: {
2016    ConditionalOperator *C = cast<ConditionalOperator>(E);
2017
2018    // Handle the GNU extension for missing LHS.
2019    if (Expr *lhsExpr = C->getLHS()) {
2020    // In C++, we can have a throw-expression, which has 'void' type.
2021      if (!lhsExpr->getType()->isVoidType())
2022        if (Expr* LHS = EvalAddr(lhsExpr, refVars))
2023          return LHS;
2024    }
2025
2026    // In C++, we can have a throw-expression, which has 'void' type.
2027    if (C->getRHS()->getType()->isVoidType())
2028      return NULL;
2029
2030    return EvalAddr(C->getRHS(), refVars);
2031  }
2032
2033  case Stmt::BlockExprClass:
2034    if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
2035      return E; // local block.
2036    return NULL;
2037
2038  case Stmt::AddrLabelExprClass:
2039    return E; // address of label.
2040
2041  // For casts, we need to handle conversions from arrays to
2042  // pointer values, and pointer-to-pointer conversions.
2043  case Stmt::ImplicitCastExprClass:
2044  case Stmt::CStyleCastExprClass:
2045  case Stmt::CXXFunctionalCastExprClass: {
2046    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
2047    QualType T = SubExpr->getType();
2048
2049    if (SubExpr->getType()->isPointerType() ||
2050        SubExpr->getType()->isBlockPointerType() ||
2051        SubExpr->getType()->isObjCQualifiedIdType())
2052      return EvalAddr(SubExpr, refVars);
2053    else if (T->isArrayType())
2054      return EvalVal(SubExpr, refVars);
2055    else
2056      return 0;
2057  }
2058
2059  // C++ casts.  For dynamic casts, static casts, and const casts, we
2060  // are always converting from a pointer-to-pointer, so we just blow
2061  // through the cast.  In the case the dynamic cast doesn't fail (and
2062  // return NULL), we take the conservative route and report cases
2063  // where we return the address of a stack variable.  For Reinterpre
2064  // FIXME: The comment about is wrong; we're not always converting
2065  // from pointer to pointer. I'm guessing that this code should also
2066  // handle references to objects.
2067  case Stmt::CXXStaticCastExprClass:
2068  case Stmt::CXXDynamicCastExprClass:
2069  case Stmt::CXXConstCastExprClass:
2070  case Stmt::CXXReinterpretCastExprClass: {
2071      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
2072      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
2073        return EvalAddr(S, refVars);
2074      else
2075        return NULL;
2076  }
2077
2078  // Everything else: we simply don't reason about them.
2079  default:
2080    return NULL;
2081  }
2082}
2083
2084
2085///  EvalVal - This function is complements EvalAddr in the mutual recursion.
2086///   See the comments for EvalAddr for more details.
2087static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) {
2088do {
2089  // We should only be called for evaluating non-pointer expressions, or
2090  // expressions with a pointer type that are not used as references but instead
2091  // are l-values (e.g., DeclRefExpr with a pointer type).
2092
2093  // Our "symbolic interpreter" is just a dispatch off the currently
2094  // viewed AST node.  We then recursively traverse the AST by calling
2095  // EvalAddr and EvalVal appropriately.
2096
2097  E = E->IgnoreParens();
2098  switch (E->getStmtClass()) {
2099  case Stmt::ImplicitCastExprClass: {
2100    ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
2101    if (IE->getValueKind() == VK_LValue) {
2102      E = IE->getSubExpr();
2103      continue;
2104    }
2105    return NULL;
2106  }
2107
2108  case Stmt::DeclRefExprClass: {
2109    // When we hit a DeclRefExpr we are looking at code that refers to a
2110    // variable's name. If it's not a reference variable we check if it has
2111    // local storage within the function, and if so, return the expression.
2112    DeclRefExpr *DR = cast<DeclRefExpr>(E);
2113
2114    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2115      if (V->hasLocalStorage()) {
2116        if (!V->getType()->isReferenceType())
2117          return DR;
2118
2119        // Reference variable, follow through to the expression that
2120        // it points to.
2121        if (V->hasInit()) {
2122          // Add the reference variable to the "trail".
2123          refVars.push_back(DR);
2124          return EvalVal(V->getInit(), refVars);
2125        }
2126      }
2127
2128    return NULL;
2129  }
2130
2131  case Stmt::UnaryOperatorClass: {
2132    // The only unary operator that make sense to handle here
2133    // is Deref.  All others don't resolve to a "name."  This includes
2134    // handling all sorts of rvalues passed to a unary operator.
2135    UnaryOperator *U = cast<UnaryOperator>(E);
2136
2137    if (U->getOpcode() == UO_Deref)
2138      return EvalAddr(U->getSubExpr(), refVars);
2139
2140    return NULL;
2141  }
2142
2143  case Stmt::ArraySubscriptExprClass: {
2144    // Array subscripts are potential references to data on the stack.  We
2145    // retrieve the DeclRefExpr* for the array variable if it indeed
2146    // has local storage.
2147    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars);
2148  }
2149
2150  case Stmt::ConditionalOperatorClass: {
2151    // For conditional operators we need to see if either the LHS or RHS are
2152    // non-NULL Expr's.  If one is non-NULL, we return it.
2153    ConditionalOperator *C = cast<ConditionalOperator>(E);
2154
2155    // Handle the GNU extension for missing LHS.
2156    if (Expr *lhsExpr = C->getLHS())
2157      if (Expr *LHS = EvalVal(lhsExpr, refVars))
2158        return LHS;
2159
2160    return EvalVal(C->getRHS(), refVars);
2161  }
2162
2163  // Accesses to members are potential references to data on the stack.
2164  case Stmt::MemberExprClass: {
2165    MemberExpr *M = cast<MemberExpr>(E);
2166
2167    // Check for indirect access.  We only want direct field accesses.
2168    if (M->isArrow())
2169      return NULL;
2170
2171    // Check whether the member type is itself a reference, in which case
2172    // we're not going to refer to the member, but to what the member refers to.
2173    if (M->getMemberDecl()->getType()->isReferenceType())
2174      return NULL;
2175
2176    return EvalVal(M->getBase(), refVars);
2177  }
2178
2179  default:
2180    // Check that we don't return or take the address of a reference to a
2181    // temporary. This is only useful in C++.
2182    if (!E->isTypeDependent() && E->isRValue())
2183      return E;
2184
2185    // Everything else: we simply don't reason about them.
2186    return NULL;
2187  }
2188} while (true);
2189}
2190
2191//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
2192
2193/// Check for comparisons of floating point operands using != and ==.
2194/// Issue a warning if these are no self-comparisons, as they are not likely
2195/// to do what the programmer intended.
2196void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) {
2197  bool EmitWarning = true;
2198
2199  Expr* LeftExprSansParen = lex->IgnoreParenImpCasts();
2200  Expr* RightExprSansParen = rex->IgnoreParenImpCasts();
2201
2202  // Special case: check for x == x (which is OK).
2203  // Do not emit warnings for such cases.
2204  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
2205    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
2206      if (DRL->getDecl() == DRR->getDecl())
2207        EmitWarning = false;
2208
2209
2210  // Special case: check for comparisons against literals that can be exactly
2211  //  represented by APFloat.  In such cases, do not emit a warning.  This
2212  //  is a heuristic: often comparison against such literals are used to
2213  //  detect if a value in a variable has not changed.  This clearly can
2214  //  lead to false negatives.
2215  if (EmitWarning) {
2216    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
2217      if (FLL->isExact())
2218        EmitWarning = false;
2219    } else
2220      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
2221        if (FLR->isExact())
2222          EmitWarning = false;
2223    }
2224  }
2225
2226  // Check for comparisons with builtin types.
2227  if (EmitWarning)
2228    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
2229      if (CL->isBuiltinCall(Context))
2230        EmitWarning = false;
2231
2232  if (EmitWarning)
2233    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
2234      if (CR->isBuiltinCall(Context))
2235        EmitWarning = false;
2236
2237  // Emit the diagnostic.
2238  if (EmitWarning)
2239    Diag(loc, diag::warn_floatingpoint_eq)
2240      << lex->getSourceRange() << rex->getSourceRange();
2241}
2242
2243//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
2244//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
2245
2246namespace {
2247
2248/// Structure recording the 'active' range of an integer-valued
2249/// expression.
2250struct IntRange {
2251  /// The number of bits active in the int.
2252  unsigned Width;
2253
2254  /// True if the int is known not to have negative values.
2255  bool NonNegative;
2256
2257  IntRange(unsigned Width, bool NonNegative)
2258    : Width(Width), NonNegative(NonNegative)
2259  {}
2260
2261  /// Returns the range of the bool type.
2262  static IntRange forBoolType() {
2263    return IntRange(1, true);
2264  }
2265
2266  /// Returns the range of an opaque value of the given integral type.
2267  static IntRange forValueOfType(ASTContext &C, QualType T) {
2268    return forValueOfCanonicalType(C,
2269                          T->getCanonicalTypeInternal().getTypePtr());
2270  }
2271
2272  /// Returns the range of an opaque value of a canonical integral type.
2273  static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
2274    assert(T->isCanonicalUnqualified());
2275
2276    if (const VectorType *VT = dyn_cast<VectorType>(T))
2277      T = VT->getElementType().getTypePtr();
2278    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2279      T = CT->getElementType().getTypePtr();
2280
2281    // For enum types, use the known bit width of the enumerators.
2282    if (const EnumType *ET = dyn_cast<EnumType>(T)) {
2283      EnumDecl *Enum = ET->getDecl();
2284      if (!Enum->isDefinition())
2285        return IntRange(C.getIntWidth(QualType(T, 0)), false);
2286
2287      unsigned NumPositive = Enum->getNumPositiveBits();
2288      unsigned NumNegative = Enum->getNumNegativeBits();
2289
2290      return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
2291    }
2292
2293    const BuiltinType *BT = cast<BuiltinType>(T);
2294    assert(BT->isInteger());
2295
2296    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2297  }
2298
2299  /// Returns the "target" range of a canonical integral type, i.e.
2300  /// the range of values expressible in the type.
2301  ///
2302  /// This matches forValueOfCanonicalType except that enums have the
2303  /// full range of their type, not the range of their enumerators.
2304  static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
2305    assert(T->isCanonicalUnqualified());
2306
2307    if (const VectorType *VT = dyn_cast<VectorType>(T))
2308      T = VT->getElementType().getTypePtr();
2309    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2310      T = CT->getElementType().getTypePtr();
2311    if (const EnumType *ET = dyn_cast<EnumType>(T))
2312      T = ET->getDecl()->getIntegerType().getTypePtr();
2313
2314    const BuiltinType *BT = cast<BuiltinType>(T);
2315    assert(BT->isInteger());
2316
2317    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2318  }
2319
2320  /// Returns the supremum of two ranges: i.e. their conservative merge.
2321  static IntRange join(IntRange L, IntRange R) {
2322    return IntRange(std::max(L.Width, R.Width),
2323                    L.NonNegative && R.NonNegative);
2324  }
2325
2326  /// Returns the infinum of two ranges: i.e. their aggressive merge.
2327  static IntRange meet(IntRange L, IntRange R) {
2328    return IntRange(std::min(L.Width, R.Width),
2329                    L.NonNegative || R.NonNegative);
2330  }
2331};
2332
2333IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) {
2334  if (value.isSigned() && value.isNegative())
2335    return IntRange(value.getMinSignedBits(), false);
2336
2337  if (value.getBitWidth() > MaxWidth)
2338    value = value.trunc(MaxWidth);
2339
2340  // isNonNegative() just checks the sign bit without considering
2341  // signedness.
2342  return IntRange(value.getActiveBits(), true);
2343}
2344
2345IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
2346                       unsigned MaxWidth) {
2347  if (result.isInt())
2348    return GetValueRange(C, result.getInt(), MaxWidth);
2349
2350  if (result.isVector()) {
2351    IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
2352    for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
2353      IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
2354      R = IntRange::join(R, El);
2355    }
2356    return R;
2357  }
2358
2359  if (result.isComplexInt()) {
2360    IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
2361    IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
2362    return IntRange::join(R, I);
2363  }
2364
2365  // This can happen with lossless casts to intptr_t of "based" lvalues.
2366  // Assume it might use arbitrary bits.
2367  // FIXME: The only reason we need to pass the type in here is to get
2368  // the sign right on this one case.  It would be nice if APValue
2369  // preserved this.
2370  assert(result.isLValue());
2371  return IntRange(MaxWidth, Ty->isUnsignedIntegerType());
2372}
2373
2374/// Pseudo-evaluate the given integer expression, estimating the
2375/// range of values it might take.
2376///
2377/// \param MaxWidth - the width to which the value will be truncated
2378IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
2379  E = E->IgnoreParens();
2380
2381  // Try a full evaluation first.
2382  Expr::EvalResult result;
2383  if (E->Evaluate(result, C))
2384    return GetValueRange(C, result.Val, E->getType(), MaxWidth);
2385
2386  // I think we only want to look through implicit casts here; if the
2387  // user has an explicit widening cast, we should treat the value as
2388  // being of the new, wider type.
2389  if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
2390    if (CE->getCastKind() == CK_NoOp)
2391      return GetExprRange(C, CE->getSubExpr(), MaxWidth);
2392
2393    IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
2394
2395    bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
2396
2397    // Assume that non-integer casts can span the full range of the type.
2398    if (!isIntegerCast)
2399      return OutputTypeRange;
2400
2401    IntRange SubRange
2402      = GetExprRange(C, CE->getSubExpr(),
2403                     std::min(MaxWidth, OutputTypeRange.Width));
2404
2405    // Bail out if the subexpr's range is as wide as the cast type.
2406    if (SubRange.Width >= OutputTypeRange.Width)
2407      return OutputTypeRange;
2408
2409    // Otherwise, we take the smaller width, and we're non-negative if
2410    // either the output type or the subexpr is.
2411    return IntRange(SubRange.Width,
2412                    SubRange.NonNegative || OutputTypeRange.NonNegative);
2413  }
2414
2415  if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
2416    // If we can fold the condition, just take that operand.
2417    bool CondResult;
2418    if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
2419      return GetExprRange(C, CondResult ? CO->getTrueExpr()
2420                                        : CO->getFalseExpr(),
2421                          MaxWidth);
2422
2423    // Otherwise, conservatively merge.
2424    IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
2425    IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
2426    return IntRange::join(L, R);
2427  }
2428
2429  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
2430    switch (BO->getOpcode()) {
2431
2432    // Boolean-valued operations are single-bit and positive.
2433    case BO_LAnd:
2434    case BO_LOr:
2435    case BO_LT:
2436    case BO_GT:
2437    case BO_LE:
2438    case BO_GE:
2439    case BO_EQ:
2440    case BO_NE:
2441      return IntRange::forBoolType();
2442
2443    // The type of these compound assignments is the type of the LHS,
2444    // so the RHS is not necessarily an integer.
2445    case BO_MulAssign:
2446    case BO_DivAssign:
2447    case BO_RemAssign:
2448    case BO_AddAssign:
2449    case BO_SubAssign:
2450      return IntRange::forValueOfType(C, E->getType());
2451
2452    // Operations with opaque sources are black-listed.
2453    case BO_PtrMemD:
2454    case BO_PtrMemI:
2455      return IntRange::forValueOfType(C, E->getType());
2456
2457    // Bitwise-and uses the *infinum* of the two source ranges.
2458    case BO_And:
2459    case BO_AndAssign:
2460      return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
2461                            GetExprRange(C, BO->getRHS(), MaxWidth));
2462
2463    // Left shift gets black-listed based on a judgement call.
2464    case BO_Shl:
2465      // ...except that we want to treat '1 << (blah)' as logically
2466      // positive.  It's an important idiom.
2467      if (IntegerLiteral *I
2468            = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
2469        if (I->getValue() == 1) {
2470          IntRange R = IntRange::forValueOfType(C, E->getType());
2471          return IntRange(R.Width, /*NonNegative*/ true);
2472        }
2473      }
2474      // fallthrough
2475
2476    case BO_ShlAssign:
2477      return IntRange::forValueOfType(C, E->getType());
2478
2479    // Right shift by a constant can narrow its left argument.
2480    case BO_Shr:
2481    case BO_ShrAssign: {
2482      IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2483
2484      // If the shift amount is a positive constant, drop the width by
2485      // that much.
2486      llvm::APSInt shift;
2487      if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
2488          shift.isNonNegative()) {
2489        unsigned zext = shift.getZExtValue();
2490        if (zext >= L.Width)
2491          L.Width = (L.NonNegative ? 0 : 1);
2492        else
2493          L.Width -= zext;
2494      }
2495
2496      return L;
2497    }
2498
2499    // Comma acts as its right operand.
2500    case BO_Comma:
2501      return GetExprRange(C, BO->getRHS(), MaxWidth);
2502
2503    // Black-list pointer subtractions.
2504    case BO_Sub:
2505      if (BO->getLHS()->getType()->isPointerType())
2506        return IntRange::forValueOfType(C, E->getType());
2507      // fallthrough
2508
2509    default:
2510      break;
2511    }
2512
2513    // Treat every other operator as if it were closed on the
2514    // narrowest type that encompasses both operands.
2515    IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2516    IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
2517    return IntRange::join(L, R);
2518  }
2519
2520  if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
2521    switch (UO->getOpcode()) {
2522    // Boolean-valued operations are white-listed.
2523    case UO_LNot:
2524      return IntRange::forBoolType();
2525
2526    // Operations with opaque sources are black-listed.
2527    case UO_Deref:
2528    case UO_AddrOf: // should be impossible
2529      return IntRange::forValueOfType(C, E->getType());
2530
2531    default:
2532      return GetExprRange(C, UO->getSubExpr(), MaxWidth);
2533    }
2534  }
2535
2536  if (dyn_cast<OffsetOfExpr>(E)) {
2537    IntRange::forValueOfType(C, E->getType());
2538  }
2539
2540  FieldDecl *BitField = E->getBitField();
2541  if (BitField) {
2542    llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C);
2543    unsigned BitWidth = BitWidthAP.getZExtValue();
2544
2545    return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType());
2546  }
2547
2548  return IntRange::forValueOfType(C, E->getType());
2549}
2550
2551IntRange GetExprRange(ASTContext &C, Expr *E) {
2552  return GetExprRange(C, E, C.getIntWidth(E->getType()));
2553}
2554
2555/// Checks whether the given value, which currently has the given
2556/// source semantics, has the same value when coerced through the
2557/// target semantics.
2558bool IsSameFloatAfterCast(const llvm::APFloat &value,
2559                          const llvm::fltSemantics &Src,
2560                          const llvm::fltSemantics &Tgt) {
2561  llvm::APFloat truncated = value;
2562
2563  bool ignored;
2564  truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
2565  truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
2566
2567  return truncated.bitwiseIsEqual(value);
2568}
2569
2570/// Checks whether the given value, which currently has the given
2571/// source semantics, has the same value when coerced through the
2572/// target semantics.
2573///
2574/// The value might be a vector of floats (or a complex number).
2575bool IsSameFloatAfterCast(const APValue &value,
2576                          const llvm::fltSemantics &Src,
2577                          const llvm::fltSemantics &Tgt) {
2578  if (value.isFloat())
2579    return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
2580
2581  if (value.isVector()) {
2582    for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
2583      if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
2584        return false;
2585    return true;
2586  }
2587
2588  assert(value.isComplexFloat());
2589  return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
2590          IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
2591}
2592
2593void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
2594
2595static bool IsZero(Sema &S, Expr *E) {
2596  // Suppress cases where we are comparing against an enum constant.
2597  if (const DeclRefExpr *DR =
2598      dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
2599    if (isa<EnumConstantDecl>(DR->getDecl()))
2600      return false;
2601
2602  // Suppress cases where the '0' value is expanded from a macro.
2603  if (E->getLocStart().isMacroID())
2604    return false;
2605
2606  llvm::APSInt Value;
2607  return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
2608}
2609
2610static bool HasEnumType(Expr *E) {
2611  // Strip off implicit integral promotions.
2612  while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
2613    if (ICE->getCastKind() != CK_IntegralCast &&
2614        ICE->getCastKind() != CK_NoOp)
2615      break;
2616    E = ICE->getSubExpr();
2617  }
2618
2619  return E->getType()->isEnumeralType();
2620}
2621
2622void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
2623  BinaryOperatorKind op = E->getOpcode();
2624  if (E->isValueDependent())
2625    return;
2626
2627  if (op == BO_LT && IsZero(S, E->getRHS())) {
2628    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2629      << "< 0" << "false" << HasEnumType(E->getLHS())
2630      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2631  } else if (op == BO_GE && IsZero(S, E->getRHS())) {
2632    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2633      << ">= 0" << "true" << HasEnumType(E->getLHS())
2634      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2635  } else if (op == BO_GT && IsZero(S, E->getLHS())) {
2636    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2637      << "0 >" << "false" << HasEnumType(E->getRHS())
2638      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2639  } else if (op == BO_LE && IsZero(S, E->getLHS())) {
2640    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2641      << "0 <=" << "true" << HasEnumType(E->getRHS())
2642      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2643  }
2644}
2645
2646/// Analyze the operands of the given comparison.  Implements the
2647/// fallback case from AnalyzeComparison.
2648void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
2649  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
2650  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
2651}
2652
2653/// \brief Implements -Wsign-compare.
2654///
2655/// \param lex the left-hand expression
2656/// \param rex the right-hand expression
2657/// \param OpLoc the location of the joining operator
2658/// \param BinOpc binary opcode or 0
2659void AnalyzeComparison(Sema &S, BinaryOperator *E) {
2660  // The type the comparison is being performed in.
2661  QualType T = E->getLHS()->getType();
2662  assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
2663         && "comparison with mismatched types");
2664
2665  // We don't do anything special if this isn't an unsigned integral
2666  // comparison:  we're only interested in integral comparisons, and
2667  // signed comparisons only happen in cases we don't care to warn about.
2668  //
2669  // We also don't care about value-dependent expressions or expressions
2670  // whose result is a constant.
2671  if (!T->hasUnsignedIntegerRepresentation()
2672      || E->isValueDependent() || E->isIntegerConstantExpr(S.Context))
2673    return AnalyzeImpConvsInComparison(S, E);
2674
2675  Expr *lex = E->getLHS()->IgnoreParenImpCasts();
2676  Expr *rex = E->getRHS()->IgnoreParenImpCasts();
2677
2678  // Check to see if one of the (unmodified) operands is of different
2679  // signedness.
2680  Expr *signedOperand, *unsignedOperand;
2681  if (lex->getType()->hasSignedIntegerRepresentation()) {
2682    assert(!rex->getType()->hasSignedIntegerRepresentation() &&
2683           "unsigned comparison between two signed integer expressions?");
2684    signedOperand = lex;
2685    unsignedOperand = rex;
2686  } else if (rex->getType()->hasSignedIntegerRepresentation()) {
2687    signedOperand = rex;
2688    unsignedOperand = lex;
2689  } else {
2690    CheckTrivialUnsignedComparison(S, E);
2691    return AnalyzeImpConvsInComparison(S, E);
2692  }
2693
2694  // Otherwise, calculate the effective range of the signed operand.
2695  IntRange signedRange = GetExprRange(S.Context, signedOperand);
2696
2697  // Go ahead and analyze implicit conversions in the operands.  Note
2698  // that we skip the implicit conversions on both sides.
2699  AnalyzeImplicitConversions(S, lex, E->getOperatorLoc());
2700  AnalyzeImplicitConversions(S, rex, E->getOperatorLoc());
2701
2702  // If the signed range is non-negative, -Wsign-compare won't fire,
2703  // but we should still check for comparisons which are always true
2704  // or false.
2705  if (signedRange.NonNegative)
2706    return CheckTrivialUnsignedComparison(S, E);
2707
2708  // For (in)equality comparisons, if the unsigned operand is a
2709  // constant which cannot collide with a overflowed signed operand,
2710  // then reinterpreting the signed operand as unsigned will not
2711  // change the result of the comparison.
2712  if (E->isEqualityOp()) {
2713    unsigned comparisonWidth = S.Context.getIntWidth(T);
2714    IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
2715
2716    // We should never be unable to prove that the unsigned operand is
2717    // non-negative.
2718    assert(unsignedRange.NonNegative && "unsigned range includes negative?");
2719
2720    if (unsignedRange.Width < comparisonWidth)
2721      return;
2722  }
2723
2724  S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison)
2725    << lex->getType() << rex->getType()
2726    << lex->getSourceRange() << rex->getSourceRange();
2727}
2728
2729/// Analyzes an attempt to assign the given value to a bitfield.
2730///
2731/// Returns true if there was something fishy about the attempt.
2732bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
2733                               SourceLocation InitLoc) {
2734  assert(Bitfield->isBitField());
2735  if (Bitfield->isInvalidDecl())
2736    return false;
2737
2738  // White-list bool bitfields.
2739  if (Bitfield->getType()->isBooleanType())
2740    return false;
2741
2742  // Ignore value- or type-dependent expressions.
2743  if (Bitfield->getBitWidth()->isValueDependent() ||
2744      Bitfield->getBitWidth()->isTypeDependent() ||
2745      Init->isValueDependent() ||
2746      Init->isTypeDependent())
2747    return false;
2748
2749  Expr *OriginalInit = Init->IgnoreParenImpCasts();
2750
2751  llvm::APSInt Width(32);
2752  Expr::EvalResult InitValue;
2753  if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) ||
2754      !OriginalInit->Evaluate(InitValue, S.Context) ||
2755      !InitValue.Val.isInt())
2756    return false;
2757
2758  const llvm::APSInt &Value = InitValue.Val.getInt();
2759  unsigned OriginalWidth = Value.getBitWidth();
2760  unsigned FieldWidth = Width.getZExtValue();
2761
2762  if (OriginalWidth <= FieldWidth)
2763    return false;
2764
2765  llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
2766
2767  // It's fairly common to write values into signed bitfields
2768  // that, if sign-extended, would end up becoming a different
2769  // value.  We don't want to warn about that.
2770  if (Value.isSigned() && Value.isNegative())
2771    TruncatedValue = TruncatedValue.sext(OriginalWidth);
2772  else
2773    TruncatedValue = TruncatedValue.zext(OriginalWidth);
2774
2775  if (Value == TruncatedValue)
2776    return false;
2777
2778  std::string PrettyValue = Value.toString(10);
2779  std::string PrettyTrunc = TruncatedValue.toString(10);
2780
2781  S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
2782    << PrettyValue << PrettyTrunc << OriginalInit->getType()
2783    << Init->getSourceRange();
2784
2785  return true;
2786}
2787
2788/// Analyze the given simple or compound assignment for warning-worthy
2789/// operations.
2790void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
2791  // Just recurse on the LHS.
2792  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
2793
2794  // We want to recurse on the RHS as normal unless we're assigning to
2795  // a bitfield.
2796  if (FieldDecl *Bitfield = E->getLHS()->getBitField()) {
2797    if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
2798                                  E->getOperatorLoc())) {
2799      // Recurse, ignoring any implicit conversions on the RHS.
2800      return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
2801                                        E->getOperatorLoc());
2802    }
2803  }
2804
2805  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
2806}
2807
2808/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
2809void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
2810                     SourceLocation CContext, unsigned diag) {
2811  S.Diag(E->getExprLoc(), diag)
2812    << SourceType << T << E->getSourceRange() << SourceRange(CContext);
2813}
2814
2815/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
2816void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext,
2817                     unsigned diag) {
2818  DiagnoseImpCast(S, E, E->getType(), T, CContext, diag);
2819}
2820
2821/// Diagnose an implicit cast from a literal expression. Also attemps to supply
2822/// fixit hints when the cast wouldn't lose information to simply write the
2823/// expression with the expected type.
2824void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
2825                                    SourceLocation CContext) {
2826  // Emit the primary warning first, then try to emit a fixit hint note if
2827  // reasonable.
2828  S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
2829    << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext);
2830
2831  const llvm::APFloat &Value = FL->getValue();
2832
2833  // Don't attempt to fix PPC double double literals.
2834  if (&Value.getSemantics() == &llvm::APFloat::PPCDoubleDouble)
2835    return;
2836
2837  // Try to convert this exactly to an 64-bit integer. FIXME: It would be
2838  // nice to support arbitrarily large integers here.
2839  bool isExact = false;
2840  uint64_t IntegerPart;
2841  if (Value.convertToInteger(&IntegerPart, 64, /*isSigned=*/true,
2842                             llvm::APFloat::rmTowardZero, &isExact)
2843      != llvm::APFloat::opOK || !isExact)
2844    return;
2845
2846  llvm::APInt IntegerValue(64, IntegerPart, /*isSigned=*/true);
2847
2848  std::string LiteralValue = IntegerValue.toString(10, /*isSigned=*/true);
2849  S.Diag(FL->getExprLoc(), diag::note_fix_integral_float_as_integer)
2850    << FixItHint::CreateReplacement(FL->getSourceRange(), LiteralValue);
2851}
2852
2853std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
2854  if (!Range.Width) return "0";
2855
2856  llvm::APSInt ValueInRange = Value;
2857  ValueInRange.setIsSigned(!Range.NonNegative);
2858  ValueInRange = ValueInRange.trunc(Range.Width);
2859  return ValueInRange.toString(10);
2860}
2861
2862static bool isFromSystemMacro(Sema &S, SourceLocation loc) {
2863  SourceManager &smgr = S.Context.getSourceManager();
2864  return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc));
2865}
2866
2867void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
2868                             SourceLocation CC, bool *ICContext = 0) {
2869  if (E->isTypeDependent() || E->isValueDependent()) return;
2870
2871  const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
2872  const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
2873  if (Source == Target) return;
2874  if (Target->isDependentType()) return;
2875
2876  // If the conversion context location is invalid don't complain.
2877  // We also don't want to emit a warning if the issue occurs from the
2878  // instantiation of a system macro.  The problem is that 'getSpellingLoc()'
2879  // is slow, so we delay this check as long as possible.  Once we detect
2880  // we are in that scenario, we just return.
2881  if (CC.isInvalid())
2882    return;
2883
2884  // Never diagnose implicit casts to bool.
2885  if (Target->isSpecificBuiltinType(BuiltinType::Bool))
2886    return;
2887
2888  // Strip vector types.
2889  if (isa<VectorType>(Source)) {
2890    if (!isa<VectorType>(Target)) {
2891      if (isFromSystemMacro(S, CC))
2892        return;
2893      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
2894    }
2895
2896    Source = cast<VectorType>(Source)->getElementType().getTypePtr();
2897    Target = cast<VectorType>(Target)->getElementType().getTypePtr();
2898  }
2899
2900  // Strip complex types.
2901  if (isa<ComplexType>(Source)) {
2902    if (!isa<ComplexType>(Target)) {
2903      if (isFromSystemMacro(S, CC))
2904        return;
2905
2906      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
2907    }
2908
2909    Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
2910    Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
2911  }
2912
2913  const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
2914  const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
2915
2916  // If the source is floating point...
2917  if (SourceBT && SourceBT->isFloatingPoint()) {
2918    // ...and the target is floating point...
2919    if (TargetBT && TargetBT->isFloatingPoint()) {
2920      // ...then warn if we're dropping FP rank.
2921
2922      // Builtin FP kinds are ordered by increasing FP rank.
2923      if (SourceBT->getKind() > TargetBT->getKind()) {
2924        // Don't warn about float constants that are precisely
2925        // representable in the target type.
2926        Expr::EvalResult result;
2927        if (E->Evaluate(result, S.Context)) {
2928          // Value might be a float, a float vector, or a float complex.
2929          if (IsSameFloatAfterCast(result.Val,
2930                   S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
2931                   S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
2932            return;
2933        }
2934
2935        if (isFromSystemMacro(S, CC))
2936          return;
2937
2938        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
2939      }
2940      return;
2941    }
2942
2943    // If the target is integral, always warn.
2944    if ((TargetBT && TargetBT->isInteger())) {
2945      if (isFromSystemMacro(S, CC))
2946        return;
2947
2948      Expr *InnerE = E->IgnoreParenImpCasts();
2949      if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
2950        DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
2951      } else {
2952        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
2953      }
2954    }
2955
2956    return;
2957  }
2958
2959  if (!Source->isIntegerType() || !Target->isIntegerType())
2960    return;
2961
2962  IntRange SourceRange = GetExprRange(S.Context, E);
2963  IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
2964
2965  if (SourceRange.Width > TargetRange.Width) {
2966    // If the source is a constant, use a default-on diagnostic.
2967    // TODO: this should happen for bitfield stores, too.
2968    llvm::APSInt Value(32);
2969    if (E->isIntegerConstantExpr(Value, S.Context)) {
2970      if (isFromSystemMacro(S, CC))
2971        return;
2972
2973      std::string PrettySourceValue = Value.toString(10);
2974      std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
2975
2976      S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant)
2977        << PrettySourceValue << PrettyTargetValue
2978        << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC);
2979      return;
2980    }
2981
2982    // People want to build with -Wshorten-64-to-32 and not -Wconversion
2983    // and by god we'll let them.
2984
2985    if (isFromSystemMacro(S, CC))
2986      return;
2987
2988    if (SourceRange.Width == 64 && TargetRange.Width == 32)
2989      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32);
2990    return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
2991  }
2992
2993  if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
2994      (!TargetRange.NonNegative && SourceRange.NonNegative &&
2995       SourceRange.Width == TargetRange.Width)) {
2996
2997    if (isFromSystemMacro(S, CC))
2998      return;
2999
3000    unsigned DiagID = diag::warn_impcast_integer_sign;
3001
3002    // Traditionally, gcc has warned about this under -Wsign-compare.
3003    // We also want to warn about it in -Wconversion.
3004    // So if -Wconversion is off, use a completely identical diagnostic
3005    // in the sign-compare group.
3006    // The conditional-checking code will
3007    if (ICContext) {
3008      DiagID = diag::warn_impcast_integer_sign_conditional;
3009      *ICContext = true;
3010    }
3011
3012    return DiagnoseImpCast(S, E, T, CC, DiagID);
3013  }
3014
3015  // Diagnose conversions between different enumeration types.
3016  // In C, we pretend that the type of an EnumConstantDecl is its enumeration
3017  // type, to give us better diagnostics.
3018  QualType SourceType = E->getType();
3019  if (!S.getLangOptions().CPlusPlus) {
3020    if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
3021      if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
3022        EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
3023        SourceType = S.Context.getTypeDeclType(Enum);
3024        Source = S.Context.getCanonicalType(SourceType).getTypePtr();
3025      }
3026  }
3027
3028  if (const EnumType *SourceEnum = Source->getAs<EnumType>())
3029    if (const EnumType *TargetEnum = Target->getAs<EnumType>())
3030      if ((SourceEnum->getDecl()->getIdentifier() ||
3031           SourceEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3032          (TargetEnum->getDecl()->getIdentifier() ||
3033           TargetEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3034          SourceEnum != TargetEnum) {
3035        if (isFromSystemMacro(S, CC))
3036          return;
3037
3038        return DiagnoseImpCast(S, E, SourceType, T, CC,
3039                               diag::warn_impcast_different_enum_types);
3040      }
3041
3042  return;
3043}
3044
3045void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T);
3046
3047void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
3048                             SourceLocation CC, bool &ICContext) {
3049  E = E->IgnoreParenImpCasts();
3050
3051  if (isa<ConditionalOperator>(E))
3052    return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T);
3053
3054  AnalyzeImplicitConversions(S, E, CC);
3055  if (E->getType() != T)
3056    return CheckImplicitConversion(S, E, T, CC, &ICContext);
3057  return;
3058}
3059
3060void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) {
3061  SourceLocation CC = E->getQuestionLoc();
3062
3063  AnalyzeImplicitConversions(S, E->getCond(), CC);
3064
3065  bool Suspicious = false;
3066  CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
3067  CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
3068
3069  // If -Wconversion would have warned about either of the candidates
3070  // for a signedness conversion to the context type...
3071  if (!Suspicious) return;
3072
3073  // ...but it's currently ignored...
3074  if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
3075                                 CC))
3076    return;
3077
3078  // ...and -Wsign-compare isn't...
3079  if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional, CC))
3080    return;
3081
3082  // ...then check whether it would have warned about either of the
3083  // candidates for a signedness conversion to the condition type.
3084  if (E->getType() != T) {
3085    Suspicious = false;
3086    CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
3087                            E->getType(), CC, &Suspicious);
3088    if (!Suspicious)
3089      CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
3090                              E->getType(), CC, &Suspicious);
3091    if (!Suspicious)
3092      return;
3093  }
3094
3095  // If so, emit a diagnostic under -Wsign-compare.
3096  Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts();
3097  Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts();
3098  S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional)
3099    << lex->getType() << rex->getType()
3100    << lex->getSourceRange() << rex->getSourceRange();
3101}
3102
3103/// AnalyzeImplicitConversions - Find and report any interesting
3104/// implicit conversions in the given expression.  There are a couple
3105/// of competing diagnostics here, -Wconversion and -Wsign-compare.
3106void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
3107  QualType T = OrigE->getType();
3108  Expr *E = OrigE->IgnoreParenImpCasts();
3109
3110  // For conditional operators, we analyze the arguments as if they
3111  // were being fed directly into the output.
3112  if (isa<ConditionalOperator>(E)) {
3113    ConditionalOperator *CO = cast<ConditionalOperator>(E);
3114    CheckConditionalOperator(S, CO, T);
3115    return;
3116  }
3117
3118  // Go ahead and check any implicit conversions we might have skipped.
3119  // The non-canonical typecheck is just an optimization;
3120  // CheckImplicitConversion will filter out dead implicit conversions.
3121  if (E->getType() != T)
3122    CheckImplicitConversion(S, E, T, CC);
3123
3124  // Now continue drilling into this expression.
3125
3126  // Skip past explicit casts.
3127  if (isa<ExplicitCastExpr>(E)) {
3128    E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
3129    return AnalyzeImplicitConversions(S, E, CC);
3130  }
3131
3132  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
3133    // Do a somewhat different check with comparison operators.
3134    if (BO->isComparisonOp())
3135      return AnalyzeComparison(S, BO);
3136
3137    // And with assignments and compound assignments.
3138    if (BO->isAssignmentOp())
3139      return AnalyzeAssignment(S, BO);
3140  }
3141
3142  // These break the otherwise-useful invariant below.  Fortunately,
3143  // we don't really need to recurse into them, because any internal
3144  // expressions should have been analyzed already when they were
3145  // built into statements.
3146  if (isa<StmtExpr>(E)) return;
3147
3148  // Don't descend into unevaluated contexts.
3149  if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
3150
3151  // Now just recurse over the expression's children.
3152  CC = E->getExprLoc();
3153  for (Stmt::child_range I = E->children(); I; ++I)
3154    AnalyzeImplicitConversions(S, cast<Expr>(*I), CC);
3155}
3156
3157} // end anonymous namespace
3158
3159/// Diagnoses "dangerous" implicit conversions within the given
3160/// expression (which is a full expression).  Implements -Wconversion
3161/// and -Wsign-compare.
3162///
3163/// \param CC the "context" location of the implicit conversion, i.e.
3164///   the most location of the syntactic entity requiring the implicit
3165///   conversion
3166void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
3167  // Don't diagnose in unevaluated contexts.
3168  if (ExprEvalContexts.back().Context == Sema::Unevaluated)
3169    return;
3170
3171  // Don't diagnose for value- or type-dependent expressions.
3172  if (E->isTypeDependent() || E->isValueDependent())
3173    return;
3174
3175  // This is not the right CC for (e.g.) a variable initialization.
3176  AnalyzeImplicitConversions(*this, E, CC);
3177}
3178
3179void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
3180                                       FieldDecl *BitField,
3181                                       Expr *Init) {
3182  (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
3183}
3184
3185/// CheckParmsForFunctionDef - Check that the parameters of the given
3186/// function are appropriate for the definition of a function. This
3187/// takes care of any checks that cannot be performed on the
3188/// declaration itself, e.g., that the types of each of the function
3189/// parameters are complete.
3190bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
3191                                    bool CheckParameterNames) {
3192  bool HasInvalidParm = false;
3193  for (; P != PEnd; ++P) {
3194    ParmVarDecl *Param = *P;
3195
3196    // C99 6.7.5.3p4: the parameters in a parameter type list in a
3197    // function declarator that is part of a function definition of
3198    // that function shall not have incomplete type.
3199    //
3200    // This is also C++ [dcl.fct]p6.
3201    if (!Param->isInvalidDecl() &&
3202        RequireCompleteType(Param->getLocation(), Param->getType(),
3203                               diag::err_typecheck_decl_incomplete_type)) {
3204      Param->setInvalidDecl();
3205      HasInvalidParm = true;
3206    }
3207
3208    // C99 6.9.1p5: If the declarator includes a parameter type list, the
3209    // declaration of each parameter shall include an identifier.
3210    if (CheckParameterNames &&
3211        Param->getIdentifier() == 0 &&
3212        !Param->isImplicit() &&
3213        !getLangOptions().CPlusPlus)
3214      Diag(Param->getLocation(), diag::err_parameter_name_omitted);
3215
3216    // C99 6.7.5.3p12:
3217    //   If the function declarator is not part of a definition of that
3218    //   function, parameters may have incomplete type and may use the [*]
3219    //   notation in their sequences of declarator specifiers to specify
3220    //   variable length array types.
3221    QualType PType = Param->getOriginalType();
3222    if (const ArrayType *AT = Context.getAsArrayType(PType)) {
3223      if (AT->getSizeModifier() == ArrayType::Star) {
3224        // FIXME: This diagnosic should point the the '[*]' if source-location
3225        // information is added for it.
3226        Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
3227      }
3228    }
3229  }
3230
3231  return HasInvalidParm;
3232}
3233
3234/// CheckCastAlign - Implements -Wcast-align, which warns when a
3235/// pointer cast increases the alignment requirements.
3236void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
3237  // This is actually a lot of work to potentially be doing on every
3238  // cast; don't do it if we're ignoring -Wcast_align (as is the default).
3239  if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
3240                                          TRange.getBegin())
3241        == Diagnostic::Ignored)
3242    return;
3243
3244  // Ignore dependent types.
3245  if (T->isDependentType() || Op->getType()->isDependentType())
3246    return;
3247
3248  // Require that the destination be a pointer type.
3249  const PointerType *DestPtr = T->getAs<PointerType>();
3250  if (!DestPtr) return;
3251
3252  // If the destination has alignment 1, we're done.
3253  QualType DestPointee = DestPtr->getPointeeType();
3254  if (DestPointee->isIncompleteType()) return;
3255  CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
3256  if (DestAlign.isOne()) return;
3257
3258  // Require that the source be a pointer type.
3259  const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
3260  if (!SrcPtr) return;
3261  QualType SrcPointee = SrcPtr->getPointeeType();
3262
3263  // Whitelist casts from cv void*.  We already implicitly
3264  // whitelisted casts to cv void*, since they have alignment 1.
3265  // Also whitelist casts involving incomplete types, which implicitly
3266  // includes 'void'.
3267  if (SrcPointee->isIncompleteType()) return;
3268
3269  CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
3270  if (SrcAlign >= DestAlign) return;
3271
3272  Diag(TRange.getBegin(), diag::warn_cast_align)
3273    << Op->getType() << T
3274    << static_cast<unsigned>(SrcAlign.getQuantity())
3275    << static_cast<unsigned>(DestAlign.getQuantity())
3276    << TRange << Op->getSourceRange();
3277}
3278
3279static void CheckArrayAccess_Check(Sema &S,
3280                                   const clang::ArraySubscriptExpr *E) {
3281  const Expr *BaseExpr = E->getBase()->IgnoreParenImpCasts();
3282  const ConstantArrayType *ArrayTy =
3283    S.Context.getAsConstantArrayType(BaseExpr->getType());
3284  if (!ArrayTy)
3285    return;
3286
3287  const Expr *IndexExpr = E->getIdx();
3288  if (IndexExpr->isValueDependent())
3289    return;
3290  llvm::APSInt index;
3291  if (!IndexExpr->isIntegerConstantExpr(index, S.Context))
3292    return;
3293
3294  if (index.isUnsigned() || !index.isNegative()) {
3295    llvm::APInt size = ArrayTy->getSize();
3296    if (!size.isStrictlyPositive())
3297      return;
3298    if (size.getBitWidth() > index.getBitWidth())
3299      index = index.sext(size.getBitWidth());
3300    else if (size.getBitWidth() < index.getBitWidth())
3301      size = size.sext(index.getBitWidth());
3302
3303    if (index.slt(size))
3304      return;
3305
3306    S.DiagRuntimeBehavior(E->getBase()->getLocStart(), BaseExpr,
3307                          S.PDiag(diag::warn_array_index_exceeds_bounds)
3308                            << index.toString(10, true)
3309                            << size.toString(10, true)
3310                            << IndexExpr->getSourceRange());
3311  } else {
3312    S.DiagRuntimeBehavior(E->getBase()->getLocStart(), BaseExpr,
3313                          S.PDiag(diag::warn_array_index_precedes_bounds)
3314                            << index.toString(10, true)
3315                            << IndexExpr->getSourceRange());
3316  }
3317
3318  const NamedDecl *ND = NULL;
3319  if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
3320    ND = dyn_cast<NamedDecl>(DRE->getDecl());
3321  if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
3322    ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
3323  if (ND)
3324    S.DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
3325                          S.PDiag(diag::note_array_index_out_of_bounds)
3326                            << ND->getDeclName());
3327}
3328
3329void Sema::CheckArrayAccess(const Expr *expr) {
3330  while (true) {
3331    expr = expr->IgnoreParens();
3332    switch (expr->getStmtClass()) {
3333      case Stmt::ArraySubscriptExprClass:
3334        CheckArrayAccess_Check(*this, cast<ArraySubscriptExpr>(expr));
3335        return;
3336      case Stmt::ConditionalOperatorClass: {
3337        const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
3338        if (const Expr *lhs = cond->getLHS())
3339          CheckArrayAccess(lhs);
3340        if (const Expr *rhs = cond->getRHS())
3341          CheckArrayAccess(rhs);
3342        return;
3343      }
3344      default:
3345        return;
3346    }
3347  }
3348}
3349