SemaChecking.cpp revision 7ccc95bceebb2d1c8fbe277c9e33bde7dc1ccbb1
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    CheckMemsetArguments(TheCall);
324
325  return false;
326}
327
328bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
329  // Printf checking.
330  const FormatAttr *Format = NDecl->getAttr<FormatAttr>();
331  if (!Format)
332    return false;
333
334  const VarDecl *V = dyn_cast<VarDecl>(NDecl);
335  if (!V)
336    return false;
337
338  QualType Ty = V->getType();
339  if (!Ty->isBlockPointerType())
340    return false;
341
342  const bool b = Format->getType() == "scanf";
343  if (!b && !CheckablePrintfAttr(Format, TheCall))
344    return false;
345
346  bool HasVAListArg = Format->getFirstArg() == 0;
347  CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
348                            HasVAListArg ? 0 : Format->getFirstArg() - 1, !b);
349
350  return false;
351}
352
353/// SemaBuiltinAtomicOverloaded - We have a call to a function like
354/// __sync_fetch_and_add, which is an overloaded function based on the pointer
355/// type of its first argument.  The main ActOnCallExpr routines have already
356/// promoted the types of arguments because all of these calls are prototyped as
357/// void(...).
358///
359/// This function goes through and does final semantic checking for these
360/// builtins,
361ExprResult
362Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
363  CallExpr *TheCall = (CallExpr *)TheCallResult.get();
364  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
365  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
366
367  // Ensure that we have at least one argument to do type inference from.
368  if (TheCall->getNumArgs() < 1) {
369    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
370      << 0 << 1 << TheCall->getNumArgs()
371      << TheCall->getCallee()->getSourceRange();
372    return ExprError();
373  }
374
375  // Inspect the first argument of the atomic builtin.  This should always be
376  // a pointer type, whose element is an integral scalar or pointer type.
377  // Because it is a pointer type, we don't have to worry about any implicit
378  // casts here.
379  // FIXME: We don't allow floating point scalars as input.
380  Expr *FirstArg = TheCall->getArg(0);
381  if (!FirstArg->getType()->isPointerType()) {
382    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
383      << FirstArg->getType() << FirstArg->getSourceRange();
384    return ExprError();
385  }
386
387  QualType ValType =
388    FirstArg->getType()->getAs<PointerType>()->getPointeeType();
389  if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
390      !ValType->isBlockPointerType()) {
391    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
392      << FirstArg->getType() << FirstArg->getSourceRange();
393    return ExprError();
394  }
395
396  // The majority of builtins return a value, but a few have special return
397  // types, so allow them to override appropriately below.
398  QualType ResultType = ValType;
399
400  // We need to figure out which concrete builtin this maps onto.  For example,
401  // __sync_fetch_and_add with a 2 byte object turns into
402  // __sync_fetch_and_add_2.
403#define BUILTIN_ROW(x) \
404  { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
405    Builtin::BI##x##_8, Builtin::BI##x##_16 }
406
407  static const unsigned BuiltinIndices[][5] = {
408    BUILTIN_ROW(__sync_fetch_and_add),
409    BUILTIN_ROW(__sync_fetch_and_sub),
410    BUILTIN_ROW(__sync_fetch_and_or),
411    BUILTIN_ROW(__sync_fetch_and_and),
412    BUILTIN_ROW(__sync_fetch_and_xor),
413
414    BUILTIN_ROW(__sync_add_and_fetch),
415    BUILTIN_ROW(__sync_sub_and_fetch),
416    BUILTIN_ROW(__sync_and_and_fetch),
417    BUILTIN_ROW(__sync_or_and_fetch),
418    BUILTIN_ROW(__sync_xor_and_fetch),
419
420    BUILTIN_ROW(__sync_val_compare_and_swap),
421    BUILTIN_ROW(__sync_bool_compare_and_swap),
422    BUILTIN_ROW(__sync_lock_test_and_set),
423    BUILTIN_ROW(__sync_lock_release),
424    BUILTIN_ROW(__sync_swap)
425  };
426#undef BUILTIN_ROW
427
428  // Determine the index of the size.
429  unsigned SizeIndex;
430  switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
431  case 1: SizeIndex = 0; break;
432  case 2: SizeIndex = 1; break;
433  case 4: SizeIndex = 2; break;
434  case 8: SizeIndex = 3; break;
435  case 16: SizeIndex = 4; break;
436  default:
437    Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
438      << FirstArg->getType() << FirstArg->getSourceRange();
439    return ExprError();
440  }
441
442  // Each of these builtins has one pointer argument, followed by some number of
443  // values (0, 1 or 2) followed by a potentially empty varags list of stuff
444  // that we ignore.  Find out which row of BuiltinIndices to read from as well
445  // as the number of fixed args.
446  unsigned BuiltinID = FDecl->getBuiltinID();
447  unsigned BuiltinIndex, NumFixed = 1;
448  switch (BuiltinID) {
449  default: assert(0 && "Unknown overloaded atomic builtin!");
450  case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break;
451  case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break;
452  case Builtin::BI__sync_fetch_and_or:  BuiltinIndex = 2; break;
453  case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break;
454  case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break;
455
456  case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break;
457  case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break;
458  case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break;
459  case Builtin::BI__sync_or_and_fetch:  BuiltinIndex = 8; break;
460  case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break;
461
462  case Builtin::BI__sync_val_compare_and_swap:
463    BuiltinIndex = 10;
464    NumFixed = 2;
465    break;
466  case Builtin::BI__sync_bool_compare_and_swap:
467    BuiltinIndex = 11;
468    NumFixed = 2;
469    ResultType = Context.BoolTy;
470    break;
471  case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break;
472  case Builtin::BI__sync_lock_release:
473    BuiltinIndex = 13;
474    NumFixed = 0;
475    ResultType = Context.VoidTy;
476    break;
477  case Builtin::BI__sync_swap: BuiltinIndex = 14; break;
478  }
479
480  // Now that we know how many fixed arguments we expect, first check that we
481  // have at least that many.
482  if (TheCall->getNumArgs() < 1+NumFixed) {
483    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
484      << 0 << 1+NumFixed << TheCall->getNumArgs()
485      << TheCall->getCallee()->getSourceRange();
486    return ExprError();
487  }
488
489  // Get the decl for the concrete builtin from this, we can tell what the
490  // concrete integer type we should convert to is.
491  unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
492  const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
493  IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
494  FunctionDecl *NewBuiltinDecl =
495    cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
496                                           TUScope, false, DRE->getLocStart()));
497
498  // The first argument --- the pointer --- has a fixed type; we
499  // deduce the types of the rest of the arguments accordingly.  Walk
500  // the remaining arguments, converting them to the deduced value type.
501  for (unsigned i = 0; i != NumFixed; ++i) {
502    ExprResult Arg = TheCall->getArg(i+1);
503
504    // If the argument is an implicit cast, then there was a promotion due to
505    // "...", just remove it now.
506    if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) {
507      Arg = ICE->getSubExpr();
508      ICE->setSubExpr(0);
509      TheCall->setArg(i+1, Arg.get());
510    }
511
512    // GCC does an implicit conversion to the pointer or integer ValType.  This
513    // can fail in some cases (1i -> int**), check for this error case now.
514    CastKind Kind = CK_Invalid;
515    ExprValueKind VK = VK_RValue;
516    CXXCastPath BasePath;
517    Arg = CheckCastTypes(Arg.get()->getSourceRange(), ValType, Arg.take(), Kind, VK, BasePath);
518    if (Arg.isInvalid())
519      return ExprError();
520
521    // Okay, we have something that *can* be converted to the right type.  Check
522    // to see if there is a potentially weird extension going on here.  This can
523    // happen when you do an atomic operation on something like an char* and
524    // pass in 42.  The 42 gets converted to char.  This is even more strange
525    // for things like 45.123 -> char, etc.
526    // FIXME: Do this check.
527    Arg = ImpCastExprToType(Arg.take(), ValType, Kind, VK, &BasePath);
528    TheCall->setArg(i+1, Arg.get());
529  }
530
531  // Switch the DeclRefExpr to refer to the new decl.
532  DRE->setDecl(NewBuiltinDecl);
533  DRE->setType(NewBuiltinDecl->getType());
534
535  // Set the callee in the CallExpr.
536  // FIXME: This leaks the original parens and implicit casts.
537  ExprResult PromotedCall = UsualUnaryConversions(DRE);
538  if (PromotedCall.isInvalid())
539    return ExprError();
540  TheCall->setCallee(PromotedCall.take());
541
542  // Change the result type of the call to match the original value type. This
543  // is arbitrary, but the codegen for these builtins ins design to handle it
544  // gracefully.
545  TheCall->setType(ResultType);
546
547  return move(TheCallResult);
548}
549
550
551/// CheckObjCString - Checks that the argument to the builtin
552/// CFString constructor is correct
553/// Note: It might also make sense to do the UTF-16 conversion here (would
554/// simplify the backend).
555bool Sema::CheckObjCString(Expr *Arg) {
556  Arg = Arg->IgnoreParenCasts();
557  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
558
559  if (!Literal || Literal->isWide()) {
560    Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
561      << Arg->getSourceRange();
562    return true;
563  }
564
565  if (Literal->containsNonAsciiOrNull()) {
566    llvm::StringRef String = Literal->getString();
567    unsigned NumBytes = String.size();
568    llvm::SmallVector<UTF16, 128> ToBuf(NumBytes);
569    const UTF8 *FromPtr = (UTF8 *)String.data();
570    UTF16 *ToPtr = &ToBuf[0];
571
572    ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
573                                                 &ToPtr, ToPtr + NumBytes,
574                                                 strictConversion);
575    // Check for conversion failure.
576    if (Result != conversionOK)
577      Diag(Arg->getLocStart(),
578           diag::warn_cfstring_truncated) << Arg->getSourceRange();
579  }
580  return false;
581}
582
583/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
584/// Emit an error and return true on failure, return false on success.
585bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
586  Expr *Fn = TheCall->getCallee();
587  if (TheCall->getNumArgs() > 2) {
588    Diag(TheCall->getArg(2)->getLocStart(),
589         diag::err_typecheck_call_too_many_args)
590      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
591      << Fn->getSourceRange()
592      << SourceRange(TheCall->getArg(2)->getLocStart(),
593                     (*(TheCall->arg_end()-1))->getLocEnd());
594    return true;
595  }
596
597  if (TheCall->getNumArgs() < 2) {
598    return Diag(TheCall->getLocEnd(),
599      diag::err_typecheck_call_too_few_args_at_least)
600      << 0 /*function call*/ << 2 << TheCall->getNumArgs();
601  }
602
603  // Determine whether the current function is variadic or not.
604  BlockScopeInfo *CurBlock = getCurBlock();
605  bool isVariadic;
606  if (CurBlock)
607    isVariadic = CurBlock->TheDecl->isVariadic();
608  else if (FunctionDecl *FD = getCurFunctionDecl())
609    isVariadic = FD->isVariadic();
610  else
611    isVariadic = getCurMethodDecl()->isVariadic();
612
613  if (!isVariadic) {
614    Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
615    return true;
616  }
617
618  // Verify that the second argument to the builtin is the last argument of the
619  // current function or method.
620  bool SecondArgIsLastNamedArgument = false;
621  const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
622
623  if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
624    if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
625      // FIXME: This isn't correct for methods (results in bogus warning).
626      // Get the last formal in the current function.
627      const ParmVarDecl *LastArg;
628      if (CurBlock)
629        LastArg = *(CurBlock->TheDecl->param_end()-1);
630      else if (FunctionDecl *FD = getCurFunctionDecl())
631        LastArg = *(FD->param_end()-1);
632      else
633        LastArg = *(getCurMethodDecl()->param_end()-1);
634      SecondArgIsLastNamedArgument = PV == LastArg;
635    }
636  }
637
638  if (!SecondArgIsLastNamedArgument)
639    Diag(TheCall->getArg(1)->getLocStart(),
640         diag::warn_second_parameter_of_va_start_not_last_named_argument);
641  return false;
642}
643
644/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
645/// friends.  This is declared to take (...), so we have to check everything.
646bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
647  if (TheCall->getNumArgs() < 2)
648    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
649      << 0 << 2 << TheCall->getNumArgs()/*function call*/;
650  if (TheCall->getNumArgs() > 2)
651    return Diag(TheCall->getArg(2)->getLocStart(),
652                diag::err_typecheck_call_too_many_args)
653      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
654      << SourceRange(TheCall->getArg(2)->getLocStart(),
655                     (*(TheCall->arg_end()-1))->getLocEnd());
656
657  ExprResult OrigArg0 = TheCall->getArg(0);
658  ExprResult OrigArg1 = TheCall->getArg(1);
659
660  // Do standard promotions between the two arguments, returning their common
661  // type.
662  QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
663  if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
664    return true;
665
666  // Make sure any conversions are pushed back into the call; this is
667  // type safe since unordered compare builtins are declared as "_Bool
668  // foo(...)".
669  TheCall->setArg(0, OrigArg0.get());
670  TheCall->setArg(1, OrigArg1.get());
671
672  if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
673    return false;
674
675  // If the common type isn't a real floating type, then the arguments were
676  // invalid for this operation.
677  if (!Res->isRealFloatingType())
678    return Diag(OrigArg0.get()->getLocStart(),
679                diag::err_typecheck_call_invalid_ordered_compare)
680      << OrigArg0.get()->getType() << OrigArg1.get()->getType()
681      << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
682
683  return false;
684}
685
686/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
687/// __builtin_isnan and friends.  This is declared to take (...), so we have
688/// to check everything. We expect the last argument to be a floating point
689/// value.
690bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
691  if (TheCall->getNumArgs() < NumArgs)
692    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
693      << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
694  if (TheCall->getNumArgs() > NumArgs)
695    return Diag(TheCall->getArg(NumArgs)->getLocStart(),
696                diag::err_typecheck_call_too_many_args)
697      << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
698      << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
699                     (*(TheCall->arg_end()-1))->getLocEnd());
700
701  Expr *OrigArg = TheCall->getArg(NumArgs-1);
702
703  if (OrigArg->isTypeDependent())
704    return false;
705
706  // This operation requires a non-_Complex floating-point number.
707  if (!OrigArg->getType()->isRealFloatingType())
708    return Diag(OrigArg->getLocStart(),
709                diag::err_typecheck_call_invalid_unary_fp)
710      << OrigArg->getType() << OrigArg->getSourceRange();
711
712  // If this is an implicit conversion from float -> double, remove it.
713  if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
714    Expr *CastArg = Cast->getSubExpr();
715    if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
716      assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
717             "promotion from float to double is the only expected cast here");
718      Cast->setSubExpr(0);
719      TheCall->setArg(NumArgs-1, CastArg);
720      OrigArg = CastArg;
721    }
722  }
723
724  return false;
725}
726
727/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
728// This is declared to take (...), so we have to check everything.
729ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
730  if (TheCall->getNumArgs() < 2)
731    return ExprError(Diag(TheCall->getLocEnd(),
732                          diag::err_typecheck_call_too_few_args_at_least)
733      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
734      << TheCall->getSourceRange());
735
736  // Determine which of the following types of shufflevector we're checking:
737  // 1) unary, vector mask: (lhs, mask)
738  // 2) binary, vector mask: (lhs, rhs, mask)
739  // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
740  QualType resType = TheCall->getArg(0)->getType();
741  unsigned numElements = 0;
742
743  if (!TheCall->getArg(0)->isTypeDependent() &&
744      !TheCall->getArg(1)->isTypeDependent()) {
745    QualType LHSType = TheCall->getArg(0)->getType();
746    QualType RHSType = TheCall->getArg(1)->getType();
747
748    if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
749      Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
750        << SourceRange(TheCall->getArg(0)->getLocStart(),
751                       TheCall->getArg(1)->getLocEnd());
752      return ExprError();
753    }
754
755    numElements = LHSType->getAs<VectorType>()->getNumElements();
756    unsigned numResElements = TheCall->getNumArgs() - 2;
757
758    // Check to see if we have a call with 2 vector arguments, the unary shuffle
759    // with mask.  If so, verify that RHS is an integer vector type with the
760    // same number of elts as lhs.
761    if (TheCall->getNumArgs() == 2) {
762      if (!RHSType->hasIntegerRepresentation() ||
763          RHSType->getAs<VectorType>()->getNumElements() != numElements)
764        Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
765          << SourceRange(TheCall->getArg(1)->getLocStart(),
766                         TheCall->getArg(1)->getLocEnd());
767      numResElements = numElements;
768    }
769    else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
770      Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
771        << SourceRange(TheCall->getArg(0)->getLocStart(),
772                       TheCall->getArg(1)->getLocEnd());
773      return ExprError();
774    } else if (numElements != numResElements) {
775      QualType eltType = LHSType->getAs<VectorType>()->getElementType();
776      resType = Context.getVectorType(eltType, numResElements,
777                                      VectorType::GenericVector);
778    }
779  }
780
781  for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
782    if (TheCall->getArg(i)->isTypeDependent() ||
783        TheCall->getArg(i)->isValueDependent())
784      continue;
785
786    llvm::APSInt Result(32);
787    if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
788      return ExprError(Diag(TheCall->getLocStart(),
789                  diag::err_shufflevector_nonconstant_argument)
790                << TheCall->getArg(i)->getSourceRange());
791
792    if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
793      return ExprError(Diag(TheCall->getLocStart(),
794                  diag::err_shufflevector_argument_too_large)
795               << TheCall->getArg(i)->getSourceRange());
796  }
797
798  llvm::SmallVector<Expr*, 32> exprs;
799
800  for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
801    exprs.push_back(TheCall->getArg(i));
802    TheCall->setArg(i, 0);
803  }
804
805  return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
806                                            exprs.size(), resType,
807                                            TheCall->getCallee()->getLocStart(),
808                                            TheCall->getRParenLoc()));
809}
810
811/// SemaBuiltinPrefetch - Handle __builtin_prefetch.
812// This is declared to take (const void*, ...) and can take two
813// optional constant int args.
814bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
815  unsigned NumArgs = TheCall->getNumArgs();
816
817  if (NumArgs > 3)
818    return Diag(TheCall->getLocEnd(),
819             diag::err_typecheck_call_too_many_args_at_most)
820             << 0 /*function call*/ << 3 << NumArgs
821             << TheCall->getSourceRange();
822
823  // Argument 0 is checked for us and the remaining arguments must be
824  // constant integers.
825  for (unsigned i = 1; i != NumArgs; ++i) {
826    Expr *Arg = TheCall->getArg(i);
827
828    llvm::APSInt Result;
829    if (SemaBuiltinConstantArg(TheCall, i, Result))
830      return true;
831
832    // FIXME: gcc issues a warning and rewrites these to 0. These
833    // seems especially odd for the third argument since the default
834    // is 3.
835    if (i == 1) {
836      if (Result.getLimitedValue() > 1)
837        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
838             << "0" << "1" << Arg->getSourceRange();
839    } else {
840      if (Result.getLimitedValue() > 3)
841        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
842            << "0" << "3" << Arg->getSourceRange();
843    }
844  }
845
846  return false;
847}
848
849/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
850/// TheCall is a constant expression.
851bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
852                                  llvm::APSInt &Result) {
853  Expr *Arg = TheCall->getArg(ArgNum);
854  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
855  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
856
857  if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
858
859  if (!Arg->isIntegerConstantExpr(Result, Context))
860    return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
861                << FDecl->getDeclName() <<  Arg->getSourceRange();
862
863  return false;
864}
865
866/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
867/// int type). This simply type checks that type is one of the defined
868/// constants (0-3).
869// For compatibility check 0-3, llvm only handles 0 and 2.
870bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
871  llvm::APSInt Result;
872
873  // Check constant-ness first.
874  if (SemaBuiltinConstantArg(TheCall, 1, Result))
875    return true;
876
877  Expr *Arg = TheCall->getArg(1);
878  if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
879    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
880             << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
881  }
882
883  return false;
884}
885
886/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
887/// This checks that val is a constant 1.
888bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
889  Expr *Arg = TheCall->getArg(1);
890  llvm::APSInt Result;
891
892  // TODO: This is less than ideal. Overload this to take a value.
893  if (SemaBuiltinConstantArg(TheCall, 1, Result))
894    return true;
895
896  if (Result != 1)
897    return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
898             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
899
900  return false;
901}
902
903// Handle i > 1 ? "x" : "y", recursively.
904bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall,
905                                  bool HasVAListArg,
906                                  unsigned format_idx, unsigned firstDataArg,
907                                  bool isPrintf) {
908 tryAgain:
909  if (E->isTypeDependent() || E->isValueDependent())
910    return false;
911
912  E = E->IgnoreParens();
913
914  switch (E->getStmtClass()) {
915  case Stmt::BinaryConditionalOperatorClass:
916  case Stmt::ConditionalOperatorClass: {
917    const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E);
918    return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg,
919                                  format_idx, firstDataArg, isPrintf)
920        && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg,
921                                  format_idx, firstDataArg, isPrintf);
922  }
923
924  case Stmt::IntegerLiteralClass:
925    // Technically -Wformat-nonliteral does not warn about this case.
926    // The behavior of printf and friends in this case is implementation
927    // dependent.  Ideally if the format string cannot be null then
928    // it should have a 'nonnull' attribute in the function prototype.
929    return true;
930
931  case Stmt::ImplicitCastExprClass: {
932    E = cast<ImplicitCastExpr>(E)->getSubExpr();
933    goto tryAgain;
934  }
935
936  case Stmt::OpaqueValueExprClass:
937    if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
938      E = src;
939      goto tryAgain;
940    }
941    return false;
942
943  case Stmt::PredefinedExprClass:
944    // While __func__, etc., are technically not string literals, they
945    // cannot contain format specifiers and thus are not a security
946    // liability.
947    return true;
948
949  case Stmt::DeclRefExprClass: {
950    const DeclRefExpr *DR = cast<DeclRefExpr>(E);
951
952    // As an exception, do not flag errors for variables binding to
953    // const string literals.
954    if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
955      bool isConstant = false;
956      QualType T = DR->getType();
957
958      if (const ArrayType *AT = Context.getAsArrayType(T)) {
959        isConstant = AT->getElementType().isConstant(Context);
960      } else if (const PointerType *PT = T->getAs<PointerType>()) {
961        isConstant = T.isConstant(Context) &&
962                     PT->getPointeeType().isConstant(Context);
963      }
964
965      if (isConstant) {
966        if (const Expr *Init = VD->getAnyInitializer())
967          return SemaCheckStringLiteral(Init, TheCall,
968                                        HasVAListArg, format_idx, firstDataArg,
969                                        isPrintf);
970      }
971
972      // For vprintf* functions (i.e., HasVAListArg==true), we add a
973      // special check to see if the format string is a function parameter
974      // of the function calling the printf function.  If the function
975      // has an attribute indicating it is a printf-like function, then we
976      // should suppress warnings concerning non-literals being used in a call
977      // to a vprintf function.  For example:
978      //
979      // void
980      // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
981      //      va_list ap;
982      //      va_start(ap, fmt);
983      //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
984      //      ...
985      //
986      //
987      //  FIXME: We don't have full attribute support yet, so just check to see
988      //    if the argument is a DeclRefExpr that references a parameter.  We'll
989      //    add proper support for checking the attribute later.
990      if (HasVAListArg)
991        if (isa<ParmVarDecl>(VD))
992          return true;
993    }
994
995    return false;
996  }
997
998  case Stmt::CallExprClass: {
999    const CallExpr *CE = cast<CallExpr>(E);
1000    if (const ImplicitCastExpr *ICE
1001          = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
1002      if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
1003        if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
1004          if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
1005            unsigned ArgIndex = FA->getFormatIdx();
1006            const Expr *Arg = CE->getArg(ArgIndex - 1);
1007
1008            return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg,
1009                                          format_idx, firstDataArg, isPrintf);
1010          }
1011        }
1012      }
1013    }
1014
1015    return false;
1016  }
1017  case Stmt::ObjCStringLiteralClass:
1018  case Stmt::StringLiteralClass: {
1019    const StringLiteral *StrE = NULL;
1020
1021    if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1022      StrE = ObjCFExpr->getString();
1023    else
1024      StrE = cast<StringLiteral>(E);
1025
1026    if (StrE) {
1027      CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx,
1028                        firstDataArg, isPrintf);
1029      return true;
1030    }
1031
1032    return false;
1033  }
1034
1035  default:
1036    return false;
1037  }
1038}
1039
1040void
1041Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1042                            const Expr * const *ExprArgs,
1043                            SourceLocation CallSiteLoc) {
1044  for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1045                                  e = NonNull->args_end();
1046       i != e; ++i) {
1047    const Expr *ArgExpr = ExprArgs[*i];
1048    if (ArgExpr->isNullPointerConstant(Context,
1049                                       Expr::NPC_ValueDependentIsNotNull))
1050      Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1051  }
1052}
1053
1054/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar
1055/// functions) for correct use of format strings.
1056void
1057Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg,
1058                                unsigned format_idx, unsigned firstDataArg,
1059                                bool isPrintf) {
1060
1061  const Expr *Fn = TheCall->getCallee();
1062
1063  // The way the format attribute works in GCC, the implicit this argument
1064  // of member functions is counted. However, it doesn't appear in our own
1065  // lists, so decrement format_idx in that case.
1066  if (isa<CXXMemberCallExpr>(TheCall)) {
1067    const CXXMethodDecl *method_decl =
1068      dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl());
1069    if (method_decl && method_decl->isInstance()) {
1070      // Catch a format attribute mistakenly referring to the object argument.
1071      if (format_idx == 0)
1072        return;
1073      --format_idx;
1074      if(firstDataArg != 0)
1075        --firstDataArg;
1076    }
1077  }
1078
1079  // CHECK: printf/scanf-like function is called with no format string.
1080  if (format_idx >= TheCall->getNumArgs()) {
1081    Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string)
1082      << Fn->getSourceRange();
1083    return;
1084  }
1085
1086  const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts();
1087
1088  // CHECK: format string is not a string literal.
1089  //
1090  // Dynamically generated format strings are difficult to
1091  // automatically vet at compile time.  Requiring that format strings
1092  // are string literals: (1) permits the checking of format strings by
1093  // the compiler and thereby (2) can practically remove the source of
1094  // many format string exploits.
1095
1096  // Format string can be either ObjC string (e.g. @"%d") or
1097  // C string (e.g. "%d")
1098  // ObjC string uses the same format specifiers as C string, so we can use
1099  // the same format string checking logic for both ObjC and C strings.
1100  if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx,
1101                             firstDataArg, isPrintf))
1102    return;  // Literal format string found, check done!
1103
1104  // If there are no arguments specified, warn with -Wformat-security, otherwise
1105  // warn only with -Wformat-nonliteral.
1106  if (TheCall->getNumArgs() == format_idx+1)
1107    Diag(TheCall->getArg(format_idx)->getLocStart(),
1108         diag::warn_format_nonliteral_noargs)
1109      << OrigFormatExpr->getSourceRange();
1110  else
1111    Diag(TheCall->getArg(format_idx)->getLocStart(),
1112         diag::warn_format_nonliteral)
1113           << OrigFormatExpr->getSourceRange();
1114}
1115
1116namespace {
1117class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1118protected:
1119  Sema &S;
1120  const StringLiteral *FExpr;
1121  const Expr *OrigFormatExpr;
1122  const unsigned FirstDataArg;
1123  const unsigned NumDataArgs;
1124  const bool IsObjCLiteral;
1125  const char *Beg; // Start of format string.
1126  const bool HasVAListArg;
1127  const CallExpr *TheCall;
1128  unsigned FormatIdx;
1129  llvm::BitVector CoveredArgs;
1130  bool usesPositionalArgs;
1131  bool atFirstArg;
1132public:
1133  CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1134                     const Expr *origFormatExpr, unsigned firstDataArg,
1135                     unsigned numDataArgs, bool isObjCLiteral,
1136                     const char *beg, bool hasVAListArg,
1137                     const CallExpr *theCall, unsigned formatIdx)
1138    : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1139      FirstDataArg(firstDataArg),
1140      NumDataArgs(numDataArgs),
1141      IsObjCLiteral(isObjCLiteral), Beg(beg),
1142      HasVAListArg(hasVAListArg),
1143      TheCall(theCall), FormatIdx(formatIdx),
1144      usesPositionalArgs(false), atFirstArg(true) {
1145        CoveredArgs.resize(numDataArgs);
1146        CoveredArgs.reset();
1147      }
1148
1149  void DoneProcessing();
1150
1151  void HandleIncompleteSpecifier(const char *startSpecifier,
1152                                 unsigned specifierLen);
1153
1154  virtual void HandleInvalidPosition(const char *startSpecifier,
1155                                     unsigned specifierLen,
1156                                     analyze_format_string::PositionContext p);
1157
1158  virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
1159
1160  void HandleNullChar(const char *nullCharacter);
1161
1162protected:
1163  bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
1164                                        const char *startSpec,
1165                                        unsigned specifierLen,
1166                                        const char *csStart, unsigned csLen);
1167
1168  SourceRange getFormatStringRange();
1169  CharSourceRange getSpecifierRange(const char *startSpecifier,
1170                                    unsigned specifierLen);
1171  SourceLocation getLocationOfByte(const char *x);
1172
1173  const Expr *getDataArg(unsigned i) const;
1174
1175  bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
1176                    const analyze_format_string::ConversionSpecifier &CS,
1177                    const char *startSpecifier, unsigned specifierLen,
1178                    unsigned argIndex);
1179};
1180}
1181
1182SourceRange CheckFormatHandler::getFormatStringRange() {
1183  return OrigFormatExpr->getSourceRange();
1184}
1185
1186CharSourceRange CheckFormatHandler::
1187getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
1188  SourceLocation Start = getLocationOfByte(startSpecifier);
1189  SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
1190
1191  // Advance the end SourceLocation by one due to half-open ranges.
1192  End = End.getFileLocWithOffset(1);
1193
1194  return CharSourceRange::getCharRange(Start, End);
1195}
1196
1197SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
1198  return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
1199}
1200
1201void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
1202                                                   unsigned specifierLen){
1203  SourceLocation Loc = getLocationOfByte(startSpecifier);
1204  S.Diag(Loc, diag::warn_printf_incomplete_specifier)
1205    << getSpecifierRange(startSpecifier, specifierLen);
1206}
1207
1208void
1209CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
1210                                     analyze_format_string::PositionContext p) {
1211  SourceLocation Loc = getLocationOfByte(startPos);
1212  S.Diag(Loc, diag::warn_format_invalid_positional_specifier)
1213    << (unsigned) p << getSpecifierRange(startPos, posLen);
1214}
1215
1216void CheckFormatHandler::HandleZeroPosition(const char *startPos,
1217                                            unsigned posLen) {
1218  SourceLocation Loc = getLocationOfByte(startPos);
1219  S.Diag(Loc, diag::warn_format_zero_positional_specifier)
1220    << getSpecifierRange(startPos, posLen);
1221}
1222
1223void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
1224  if (!IsObjCLiteral) {
1225    // The presence of a null character is likely an error.
1226    S.Diag(getLocationOfByte(nullCharacter),
1227           diag::warn_printf_format_string_contains_null_char)
1228      << getFormatStringRange();
1229  }
1230}
1231
1232const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
1233  return TheCall->getArg(FirstDataArg + i);
1234}
1235
1236void CheckFormatHandler::DoneProcessing() {
1237    // Does the number of data arguments exceed the number of
1238    // format conversions in the format string?
1239  if (!HasVAListArg) {
1240      // Find any arguments that weren't covered.
1241    CoveredArgs.flip();
1242    signed notCoveredArg = CoveredArgs.find_first();
1243    if (notCoveredArg >= 0) {
1244      assert((unsigned)notCoveredArg < NumDataArgs);
1245      S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(),
1246             diag::warn_printf_data_arg_not_used)
1247      << getFormatStringRange();
1248    }
1249  }
1250}
1251
1252bool
1253CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
1254                                                     SourceLocation Loc,
1255                                                     const char *startSpec,
1256                                                     unsigned specifierLen,
1257                                                     const char *csStart,
1258                                                     unsigned csLen) {
1259
1260  bool keepGoing = true;
1261  if (argIndex < NumDataArgs) {
1262    // Consider the argument coverered, even though the specifier doesn't
1263    // make sense.
1264    CoveredArgs.set(argIndex);
1265  }
1266  else {
1267    // If argIndex exceeds the number of data arguments we
1268    // don't issue a warning because that is just a cascade of warnings (and
1269    // they may have intended '%%' anyway). We don't want to continue processing
1270    // the format string after this point, however, as we will like just get
1271    // gibberish when trying to match arguments.
1272    keepGoing = false;
1273  }
1274
1275  S.Diag(Loc, diag::warn_format_invalid_conversion)
1276    << llvm::StringRef(csStart, csLen)
1277    << getSpecifierRange(startSpec, specifierLen);
1278
1279  return keepGoing;
1280}
1281
1282bool
1283CheckFormatHandler::CheckNumArgs(
1284  const analyze_format_string::FormatSpecifier &FS,
1285  const analyze_format_string::ConversionSpecifier &CS,
1286  const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
1287
1288  if (argIndex >= NumDataArgs) {
1289    if (FS.usesPositionalArg())  {
1290      S.Diag(getLocationOfByte(CS.getStart()),
1291             diag::warn_printf_positional_arg_exceeds_data_args)
1292      << (argIndex+1) << NumDataArgs
1293      << getSpecifierRange(startSpecifier, specifierLen);
1294    }
1295    else {
1296      S.Diag(getLocationOfByte(CS.getStart()),
1297             diag::warn_printf_insufficient_data_args)
1298      << getSpecifierRange(startSpecifier, specifierLen);
1299    }
1300
1301    return false;
1302  }
1303  return true;
1304}
1305
1306//===--- CHECK: Printf format string checking ------------------------------===//
1307
1308namespace {
1309class CheckPrintfHandler : public CheckFormatHandler {
1310public:
1311  CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
1312                     const Expr *origFormatExpr, unsigned firstDataArg,
1313                     unsigned numDataArgs, bool isObjCLiteral,
1314                     const char *beg, bool hasVAListArg,
1315                     const CallExpr *theCall, unsigned formatIdx)
1316  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1317                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1318                       theCall, formatIdx) {}
1319
1320
1321  bool HandleInvalidPrintfConversionSpecifier(
1322                                      const analyze_printf::PrintfSpecifier &FS,
1323                                      const char *startSpecifier,
1324                                      unsigned specifierLen);
1325
1326  bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
1327                             const char *startSpecifier,
1328                             unsigned specifierLen);
1329
1330  bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
1331                    const char *startSpecifier, unsigned specifierLen);
1332  void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
1333                           const analyze_printf::OptionalAmount &Amt,
1334                           unsigned type,
1335                           const char *startSpecifier, unsigned specifierLen);
1336  void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1337                  const analyze_printf::OptionalFlag &flag,
1338                  const char *startSpecifier, unsigned specifierLen);
1339  void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
1340                         const analyze_printf::OptionalFlag &ignoredFlag,
1341                         const analyze_printf::OptionalFlag &flag,
1342                         const char *startSpecifier, unsigned specifierLen);
1343};
1344}
1345
1346bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
1347                                      const analyze_printf::PrintfSpecifier &FS,
1348                                      const char *startSpecifier,
1349                                      unsigned specifierLen) {
1350  const analyze_printf::PrintfConversionSpecifier &CS =
1351    FS.getConversionSpecifier();
1352
1353  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1354                                          getLocationOfByte(CS.getStart()),
1355                                          startSpecifier, specifierLen,
1356                                          CS.getStart(), CS.getLength());
1357}
1358
1359bool CheckPrintfHandler::HandleAmount(
1360                               const analyze_format_string::OptionalAmount &Amt,
1361                               unsigned k, const char *startSpecifier,
1362                               unsigned specifierLen) {
1363
1364  if (Amt.hasDataArgument()) {
1365    if (!HasVAListArg) {
1366      unsigned argIndex = Amt.getArgIndex();
1367      if (argIndex >= NumDataArgs) {
1368        S.Diag(getLocationOfByte(Amt.getStart()),
1369               diag::warn_printf_asterisk_missing_arg)
1370          << k << getSpecifierRange(startSpecifier, specifierLen);
1371        // Don't do any more checking.  We will just emit
1372        // spurious errors.
1373        return false;
1374      }
1375
1376      // Type check the data argument.  It should be an 'int'.
1377      // Although not in conformance with C99, we also allow the argument to be
1378      // an 'unsigned int' as that is a reasonably safe case.  GCC also
1379      // doesn't emit a warning for that case.
1380      CoveredArgs.set(argIndex);
1381      const Expr *Arg = getDataArg(argIndex);
1382      QualType T = Arg->getType();
1383
1384      const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context);
1385      assert(ATR.isValid());
1386
1387      if (!ATR.matchesType(S.Context, T)) {
1388        S.Diag(getLocationOfByte(Amt.getStart()),
1389               diag::warn_printf_asterisk_wrong_type)
1390          << k
1391          << ATR.getRepresentativeType(S.Context) << T
1392          << getSpecifierRange(startSpecifier, specifierLen)
1393          << Arg->getSourceRange();
1394        // Don't do any more checking.  We will just emit
1395        // spurious errors.
1396        return false;
1397      }
1398    }
1399  }
1400  return true;
1401}
1402
1403void CheckPrintfHandler::HandleInvalidAmount(
1404                                      const analyze_printf::PrintfSpecifier &FS,
1405                                      const analyze_printf::OptionalAmount &Amt,
1406                                      unsigned type,
1407                                      const char *startSpecifier,
1408                                      unsigned specifierLen) {
1409  const analyze_printf::PrintfConversionSpecifier &CS =
1410    FS.getConversionSpecifier();
1411  switch (Amt.getHowSpecified()) {
1412  case analyze_printf::OptionalAmount::Constant:
1413    S.Diag(getLocationOfByte(Amt.getStart()),
1414        diag::warn_printf_nonsensical_optional_amount)
1415      << type
1416      << CS.toString()
1417      << getSpecifierRange(startSpecifier, specifierLen)
1418      << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
1419          Amt.getConstantLength()));
1420    break;
1421
1422  default:
1423    S.Diag(getLocationOfByte(Amt.getStart()),
1424        diag::warn_printf_nonsensical_optional_amount)
1425      << type
1426      << CS.toString()
1427      << getSpecifierRange(startSpecifier, specifierLen);
1428    break;
1429  }
1430}
1431
1432void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1433                                    const analyze_printf::OptionalFlag &flag,
1434                                    const char *startSpecifier,
1435                                    unsigned specifierLen) {
1436  // Warn about pointless flag with a fixit removal.
1437  const analyze_printf::PrintfConversionSpecifier &CS =
1438    FS.getConversionSpecifier();
1439  S.Diag(getLocationOfByte(flag.getPosition()),
1440      diag::warn_printf_nonsensical_flag)
1441    << flag.toString() << CS.toString()
1442    << getSpecifierRange(startSpecifier, specifierLen)
1443    << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1));
1444}
1445
1446void CheckPrintfHandler::HandleIgnoredFlag(
1447                                const analyze_printf::PrintfSpecifier &FS,
1448                                const analyze_printf::OptionalFlag &ignoredFlag,
1449                                const analyze_printf::OptionalFlag &flag,
1450                                const char *startSpecifier,
1451                                unsigned specifierLen) {
1452  // Warn about ignored flag with a fixit removal.
1453  S.Diag(getLocationOfByte(ignoredFlag.getPosition()),
1454      diag::warn_printf_ignored_flag)
1455    << ignoredFlag.toString() << flag.toString()
1456    << getSpecifierRange(startSpecifier, specifierLen)
1457    << FixItHint::CreateRemoval(getSpecifierRange(
1458        ignoredFlag.getPosition(), 1));
1459}
1460
1461bool
1462CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
1463                                            &FS,
1464                                          const char *startSpecifier,
1465                                          unsigned specifierLen) {
1466
1467  using namespace analyze_format_string;
1468  using namespace analyze_printf;
1469  const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
1470
1471  if (FS.consumesDataArgument()) {
1472    if (atFirstArg) {
1473        atFirstArg = false;
1474        usesPositionalArgs = FS.usesPositionalArg();
1475    }
1476    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1477      // Cannot mix-and-match positional and non-positional arguments.
1478      S.Diag(getLocationOfByte(CS.getStart()),
1479             diag::warn_format_mix_positional_nonpositional_args)
1480        << getSpecifierRange(startSpecifier, specifierLen);
1481      return false;
1482    }
1483  }
1484
1485  // First check if the field width, precision, and conversion specifier
1486  // have matching data arguments.
1487  if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
1488                    startSpecifier, specifierLen)) {
1489    return false;
1490  }
1491
1492  if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
1493                    startSpecifier, specifierLen)) {
1494    return false;
1495  }
1496
1497  if (!CS.consumesDataArgument()) {
1498    // FIXME: Technically specifying a precision or field width here
1499    // makes no sense.  Worth issuing a warning at some point.
1500    return true;
1501  }
1502
1503  // Consume the argument.
1504  unsigned argIndex = FS.getArgIndex();
1505  if (argIndex < NumDataArgs) {
1506    // The check to see if the argIndex is valid will come later.
1507    // We set the bit here because we may exit early from this
1508    // function if we encounter some other error.
1509    CoveredArgs.set(argIndex);
1510  }
1511
1512  // Check for using an Objective-C specific conversion specifier
1513  // in a non-ObjC literal.
1514  if (!IsObjCLiteral && CS.isObjCArg()) {
1515    return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
1516                                                  specifierLen);
1517  }
1518
1519  // Check for invalid use of field width
1520  if (!FS.hasValidFieldWidth()) {
1521    HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
1522        startSpecifier, specifierLen);
1523  }
1524
1525  // Check for invalid use of precision
1526  if (!FS.hasValidPrecision()) {
1527    HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
1528        startSpecifier, specifierLen);
1529  }
1530
1531  // Check each flag does not conflict with any other component.
1532  if (!FS.hasValidThousandsGroupingPrefix())
1533    HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
1534  if (!FS.hasValidLeadingZeros())
1535    HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
1536  if (!FS.hasValidPlusPrefix())
1537    HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
1538  if (!FS.hasValidSpacePrefix())
1539    HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
1540  if (!FS.hasValidAlternativeForm())
1541    HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
1542  if (!FS.hasValidLeftJustified())
1543    HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
1544
1545  // Check that flags are not ignored by another flag
1546  if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
1547    HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
1548        startSpecifier, specifierLen);
1549  if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
1550    HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
1551            startSpecifier, specifierLen);
1552
1553  // Check the length modifier is valid with the given conversion specifier.
1554  const LengthModifier &LM = FS.getLengthModifier();
1555  if (!FS.hasValidLengthModifier())
1556    S.Diag(getLocationOfByte(LM.getStart()),
1557        diag::warn_format_nonsensical_length)
1558      << LM.toString() << CS.toString()
1559      << getSpecifierRange(startSpecifier, specifierLen)
1560      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1561          LM.getLength()));
1562
1563  // Are we using '%n'?
1564  if (CS.getKind() == ConversionSpecifier::nArg) {
1565    // Issue a warning about this being a possible security issue.
1566    S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back)
1567      << getSpecifierRange(startSpecifier, specifierLen);
1568    // Continue checking the other format specifiers.
1569    return true;
1570  }
1571
1572  // The remaining checks depend on the data arguments.
1573  if (HasVAListArg)
1574    return true;
1575
1576  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1577    return false;
1578
1579  // Now type check the data expression that matches the
1580  // format specifier.
1581  const Expr *Ex = getDataArg(argIndex);
1582  const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context);
1583  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
1584    // Check if we didn't match because of an implicit cast from a 'char'
1585    // or 'short' to an 'int'.  This is done because printf is a varargs
1586    // function.
1587    if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex))
1588      if (ICE->getType() == S.Context.IntTy) {
1589        // All further checking is done on the subexpression.
1590        Ex = ICE->getSubExpr();
1591        if (ATR.matchesType(S.Context, Ex->getType()))
1592          return true;
1593      }
1594
1595    // We may be able to offer a FixItHint if it is a supported type.
1596    PrintfSpecifier fixedFS = FS;
1597    bool success = fixedFS.fixType(Ex->getType());
1598
1599    if (success) {
1600      // Get the fix string from the fixed format specifier
1601      llvm::SmallString<128> buf;
1602      llvm::raw_svector_ostream os(buf);
1603      fixedFS.toString(os);
1604
1605      // FIXME: getRepresentativeType() perhaps should return a string
1606      // instead of a QualType to better handle when the representative
1607      // type is 'wint_t' (which is defined in the system headers).
1608      S.Diag(getLocationOfByte(CS.getStart()),
1609          diag::warn_printf_conversion_argument_type_mismatch)
1610        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1611        << getSpecifierRange(startSpecifier, specifierLen)
1612        << Ex->getSourceRange()
1613        << FixItHint::CreateReplacement(
1614            getSpecifierRange(startSpecifier, specifierLen),
1615            os.str());
1616    }
1617    else {
1618      S.Diag(getLocationOfByte(CS.getStart()),
1619             diag::warn_printf_conversion_argument_type_mismatch)
1620        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1621        << getSpecifierRange(startSpecifier, specifierLen)
1622        << Ex->getSourceRange();
1623    }
1624  }
1625
1626  return true;
1627}
1628
1629//===--- CHECK: Scanf format string checking ------------------------------===//
1630
1631namespace {
1632class CheckScanfHandler : public CheckFormatHandler {
1633public:
1634  CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
1635                    const Expr *origFormatExpr, unsigned firstDataArg,
1636                    unsigned numDataArgs, bool isObjCLiteral,
1637                    const char *beg, bool hasVAListArg,
1638                    const CallExpr *theCall, unsigned formatIdx)
1639  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1640                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1641                       theCall, formatIdx) {}
1642
1643  bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
1644                            const char *startSpecifier,
1645                            unsigned specifierLen);
1646
1647  bool HandleInvalidScanfConversionSpecifier(
1648          const analyze_scanf::ScanfSpecifier &FS,
1649          const char *startSpecifier,
1650          unsigned specifierLen);
1651
1652  void HandleIncompleteScanList(const char *start, const char *end);
1653};
1654}
1655
1656void CheckScanfHandler::HandleIncompleteScanList(const char *start,
1657                                                 const char *end) {
1658  S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete)
1659    << getSpecifierRange(start, end - start);
1660}
1661
1662bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
1663                                        const analyze_scanf::ScanfSpecifier &FS,
1664                                        const char *startSpecifier,
1665                                        unsigned specifierLen) {
1666
1667  const analyze_scanf::ScanfConversionSpecifier &CS =
1668    FS.getConversionSpecifier();
1669
1670  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1671                                          getLocationOfByte(CS.getStart()),
1672                                          startSpecifier, specifierLen,
1673                                          CS.getStart(), CS.getLength());
1674}
1675
1676bool CheckScanfHandler::HandleScanfSpecifier(
1677                                       const analyze_scanf::ScanfSpecifier &FS,
1678                                       const char *startSpecifier,
1679                                       unsigned specifierLen) {
1680
1681  using namespace analyze_scanf;
1682  using namespace analyze_format_string;
1683
1684  const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
1685
1686  // Handle case where '%' and '*' don't consume an argument.  These shouldn't
1687  // be used to decide if we are using positional arguments consistently.
1688  if (FS.consumesDataArgument()) {
1689    if (atFirstArg) {
1690      atFirstArg = false;
1691      usesPositionalArgs = FS.usesPositionalArg();
1692    }
1693    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1694      // Cannot mix-and-match positional and non-positional arguments.
1695      S.Diag(getLocationOfByte(CS.getStart()),
1696             diag::warn_format_mix_positional_nonpositional_args)
1697        << getSpecifierRange(startSpecifier, specifierLen);
1698      return false;
1699    }
1700  }
1701
1702  // Check if the field with is non-zero.
1703  const OptionalAmount &Amt = FS.getFieldWidth();
1704  if (Amt.getHowSpecified() == OptionalAmount::Constant) {
1705    if (Amt.getConstantAmount() == 0) {
1706      const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
1707                                                   Amt.getConstantLength());
1708      S.Diag(getLocationOfByte(Amt.getStart()),
1709             diag::warn_scanf_nonzero_width)
1710        << R << FixItHint::CreateRemoval(R);
1711    }
1712  }
1713
1714  if (!FS.consumesDataArgument()) {
1715    // FIXME: Technically specifying a precision or field width here
1716    // makes no sense.  Worth issuing a warning at some point.
1717    return true;
1718  }
1719
1720  // Consume the argument.
1721  unsigned argIndex = FS.getArgIndex();
1722  if (argIndex < NumDataArgs) {
1723      // The check to see if the argIndex is valid will come later.
1724      // We set the bit here because we may exit early from this
1725      // function if we encounter some other error.
1726    CoveredArgs.set(argIndex);
1727  }
1728
1729  // Check the length modifier is valid with the given conversion specifier.
1730  const LengthModifier &LM = FS.getLengthModifier();
1731  if (!FS.hasValidLengthModifier()) {
1732    S.Diag(getLocationOfByte(LM.getStart()),
1733           diag::warn_format_nonsensical_length)
1734      << LM.toString() << CS.toString()
1735      << getSpecifierRange(startSpecifier, specifierLen)
1736      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1737                                                    LM.getLength()));
1738  }
1739
1740  // The remaining checks depend on the data arguments.
1741  if (HasVAListArg)
1742    return true;
1743
1744  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1745    return false;
1746
1747  // FIXME: Check that the argument type matches the format specifier.
1748
1749  return true;
1750}
1751
1752void Sema::CheckFormatString(const StringLiteral *FExpr,
1753                             const Expr *OrigFormatExpr,
1754                             const CallExpr *TheCall, bool HasVAListArg,
1755                             unsigned format_idx, unsigned firstDataArg,
1756                             bool isPrintf) {
1757
1758  // CHECK: is the format string a wide literal?
1759  if (FExpr->isWide()) {
1760    Diag(FExpr->getLocStart(),
1761         diag::warn_format_string_is_wide_literal)
1762    << OrigFormatExpr->getSourceRange();
1763    return;
1764  }
1765
1766  // Str - The format string.  NOTE: this is NOT null-terminated!
1767  llvm::StringRef StrRef = FExpr->getString();
1768  const char *Str = StrRef.data();
1769  unsigned StrLen = StrRef.size();
1770
1771  // CHECK: empty format string?
1772  if (StrLen == 0) {
1773    Diag(FExpr->getLocStart(), diag::warn_empty_format_string)
1774    << OrigFormatExpr->getSourceRange();
1775    return;
1776  }
1777
1778  if (isPrintf) {
1779    CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1780                         TheCall->getNumArgs() - firstDataArg,
1781                         isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1782                         HasVAListArg, TheCall, format_idx);
1783
1784    if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen))
1785      H.DoneProcessing();
1786  }
1787  else {
1788    CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1789                        TheCall->getNumArgs() - firstDataArg,
1790                        isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1791                        HasVAListArg, TheCall, format_idx);
1792
1793    if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen))
1794      H.DoneProcessing();
1795  }
1796}
1797
1798//===--- CHECK: Standard memory functions ---------------------------------===//
1799
1800/// \brief Check for dangerous or invalid arguments to memset().
1801///
1802/// This issues warnings on known problematic or dangerous or unspecified
1803/// arguments to the standard 'memset' function call.
1804///
1805/// \param Call The call expression to diagnose.
1806void Sema::CheckMemsetArguments(const CallExpr *Call) {
1807  assert(Call->getNumArgs() == 3 && "Unexpected number of arguments to memset");
1808  const Expr *Dest = Call->getArg(0)->IgnoreParenImpCasts();
1809
1810  QualType DestTy = Dest->getType();
1811  if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
1812    QualType PointeeTy = DestPtrTy->getPointeeType();
1813    if (!PointeeTy->isPODType()) {
1814      DiagRuntimeBehavior(
1815        Dest->getExprLoc(), Dest,
1816        PDiag(diag::warn_non_pod_memset)
1817          << PointeeTy << Call->getCallee()->getSourceRange());
1818
1819      SourceRange ArgRange = Call->getArg(0)->getSourceRange();
1820      DiagRuntimeBehavior(
1821        Dest->getExprLoc(), Dest,
1822        PDiag(diag::note_non_pod_memset_silence)
1823          << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
1824    }
1825  }
1826}
1827
1828//===--- CHECK: Return Address of Stack Variable --------------------------===//
1829
1830static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars);
1831static Expr *EvalAddr(Expr* E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars);
1832
1833/// CheckReturnStackAddr - Check if a return statement returns the address
1834///   of a stack variable.
1835void
1836Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
1837                           SourceLocation ReturnLoc) {
1838
1839  Expr *stackE = 0;
1840  llvm::SmallVector<DeclRefExpr *, 8> refVars;
1841
1842  // Perform checking for returned stack addresses, local blocks,
1843  // label addresses or references to temporaries.
1844  if (lhsType->isPointerType() || lhsType->isBlockPointerType()) {
1845    stackE = EvalAddr(RetValExp, refVars);
1846  } else if (lhsType->isReferenceType()) {
1847    stackE = EvalVal(RetValExp, refVars);
1848  }
1849
1850  if (stackE == 0)
1851    return; // Nothing suspicious was found.
1852
1853  SourceLocation diagLoc;
1854  SourceRange diagRange;
1855  if (refVars.empty()) {
1856    diagLoc = stackE->getLocStart();
1857    diagRange = stackE->getSourceRange();
1858  } else {
1859    // We followed through a reference variable. 'stackE' contains the
1860    // problematic expression but we will warn at the return statement pointing
1861    // at the reference variable. We will later display the "trail" of
1862    // reference variables using notes.
1863    diagLoc = refVars[0]->getLocStart();
1864    diagRange = refVars[0]->getSourceRange();
1865  }
1866
1867  if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
1868    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
1869                                             : diag::warn_ret_stack_addr)
1870     << DR->getDecl()->getDeclName() << diagRange;
1871  } else if (isa<BlockExpr>(stackE)) { // local block.
1872    Diag(diagLoc, diag::err_ret_local_block) << diagRange;
1873  } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
1874    Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
1875  } else { // local temporary.
1876    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
1877                                             : diag::warn_ret_local_temp_addr)
1878     << diagRange;
1879  }
1880
1881  // Display the "trail" of reference variables that we followed until we
1882  // found the problematic expression using notes.
1883  for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
1884    VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
1885    // If this var binds to another reference var, show the range of the next
1886    // var, otherwise the var binds to the problematic expression, in which case
1887    // show the range of the expression.
1888    SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
1889                                  : stackE->getSourceRange();
1890    Diag(VD->getLocation(), diag::note_ref_var_local_bind)
1891      << VD->getDeclName() << range;
1892  }
1893}
1894
1895/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
1896///  check if the expression in a return statement evaluates to an address
1897///  to a location on the stack, a local block, an address of a label, or a
1898///  reference to local temporary. The recursion is used to traverse the
1899///  AST of the return expression, with recursion backtracking when we
1900///  encounter a subexpression that (1) clearly does not lead to one of the
1901///  above problematic expressions (2) is something we cannot determine leads to
1902///  a problematic expression based on such local checking.
1903///
1904///  Both EvalAddr and EvalVal follow through reference variables to evaluate
1905///  the expression that they point to. Such variables are added to the
1906///  'refVars' vector so that we know what the reference variable "trail" was.
1907///
1908///  EvalAddr processes expressions that are pointers that are used as
1909///  references (and not L-values).  EvalVal handles all other values.
1910///  At the base case of the recursion is a check for the above problematic
1911///  expressions.
1912///
1913///  This implementation handles:
1914///
1915///   * pointer-to-pointer casts
1916///   * implicit conversions from array references to pointers
1917///   * taking the address of fields
1918///   * arbitrary interplay between "&" and "*" operators
1919///   * pointer arithmetic from an address of a stack variable
1920///   * taking the address of an array element where the array is on the stack
1921static Expr *EvalAddr(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) {
1922  if (E->isTypeDependent())
1923      return NULL;
1924
1925  // We should only be called for evaluating pointer expressions.
1926  assert((E->getType()->isAnyPointerType() ||
1927          E->getType()->isBlockPointerType() ||
1928          E->getType()->isObjCQualifiedIdType()) &&
1929         "EvalAddr only works on pointers");
1930
1931  E = E->IgnoreParens();
1932
1933  // Our "symbolic interpreter" is just a dispatch off the currently
1934  // viewed AST node.  We then recursively traverse the AST by calling
1935  // EvalAddr and EvalVal appropriately.
1936  switch (E->getStmtClass()) {
1937  case Stmt::DeclRefExprClass: {
1938    DeclRefExpr *DR = cast<DeclRefExpr>(E);
1939
1940    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
1941      // If this is a reference variable, follow through to the expression that
1942      // it points to.
1943      if (V->hasLocalStorage() &&
1944          V->getType()->isReferenceType() && V->hasInit()) {
1945        // Add the reference variable to the "trail".
1946        refVars.push_back(DR);
1947        return EvalAddr(V->getInit(), refVars);
1948      }
1949
1950    return NULL;
1951  }
1952
1953  case Stmt::UnaryOperatorClass: {
1954    // The only unary operator that make sense to handle here
1955    // is AddrOf.  All others don't make sense as pointers.
1956    UnaryOperator *U = cast<UnaryOperator>(E);
1957
1958    if (U->getOpcode() == UO_AddrOf)
1959      return EvalVal(U->getSubExpr(), refVars);
1960    else
1961      return NULL;
1962  }
1963
1964  case Stmt::BinaryOperatorClass: {
1965    // Handle pointer arithmetic.  All other binary operators are not valid
1966    // in this context.
1967    BinaryOperator *B = cast<BinaryOperator>(E);
1968    BinaryOperatorKind op = B->getOpcode();
1969
1970    if (op != BO_Add && op != BO_Sub)
1971      return NULL;
1972
1973    Expr *Base = B->getLHS();
1974
1975    // Determine which argument is the real pointer base.  It could be
1976    // the RHS argument instead of the LHS.
1977    if (!Base->getType()->isPointerType()) Base = B->getRHS();
1978
1979    assert (Base->getType()->isPointerType());
1980    return EvalAddr(Base, refVars);
1981  }
1982
1983  // For conditional operators we need to see if either the LHS or RHS are
1984  // valid DeclRefExpr*s.  If one of them is valid, we return it.
1985  case Stmt::ConditionalOperatorClass: {
1986    ConditionalOperator *C = cast<ConditionalOperator>(E);
1987
1988    // Handle the GNU extension for missing LHS.
1989    if (Expr *lhsExpr = C->getLHS()) {
1990    // In C++, we can have a throw-expression, which has 'void' type.
1991      if (!lhsExpr->getType()->isVoidType())
1992        if (Expr* LHS = EvalAddr(lhsExpr, refVars))
1993          return LHS;
1994    }
1995
1996    // In C++, we can have a throw-expression, which has 'void' type.
1997    if (C->getRHS()->getType()->isVoidType())
1998      return NULL;
1999
2000    return EvalAddr(C->getRHS(), refVars);
2001  }
2002
2003  case Stmt::BlockExprClass:
2004    if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
2005      return E; // local block.
2006    return NULL;
2007
2008  case Stmt::AddrLabelExprClass:
2009    return E; // address of label.
2010
2011  // For casts, we need to handle conversions from arrays to
2012  // pointer values, and pointer-to-pointer conversions.
2013  case Stmt::ImplicitCastExprClass:
2014  case Stmt::CStyleCastExprClass:
2015  case Stmt::CXXFunctionalCastExprClass: {
2016    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
2017    QualType T = SubExpr->getType();
2018
2019    if (SubExpr->getType()->isPointerType() ||
2020        SubExpr->getType()->isBlockPointerType() ||
2021        SubExpr->getType()->isObjCQualifiedIdType())
2022      return EvalAddr(SubExpr, refVars);
2023    else if (T->isArrayType())
2024      return EvalVal(SubExpr, refVars);
2025    else
2026      return 0;
2027  }
2028
2029  // C++ casts.  For dynamic casts, static casts, and const casts, we
2030  // are always converting from a pointer-to-pointer, so we just blow
2031  // through the cast.  In the case the dynamic cast doesn't fail (and
2032  // return NULL), we take the conservative route and report cases
2033  // where we return the address of a stack variable.  For Reinterpre
2034  // FIXME: The comment about is wrong; we're not always converting
2035  // from pointer to pointer. I'm guessing that this code should also
2036  // handle references to objects.
2037  case Stmt::CXXStaticCastExprClass:
2038  case Stmt::CXXDynamicCastExprClass:
2039  case Stmt::CXXConstCastExprClass:
2040  case Stmt::CXXReinterpretCastExprClass: {
2041      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
2042      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
2043        return EvalAddr(S, refVars);
2044      else
2045        return NULL;
2046  }
2047
2048  // Everything else: we simply don't reason about them.
2049  default:
2050    return NULL;
2051  }
2052}
2053
2054
2055///  EvalVal - This function is complements EvalAddr in the mutual recursion.
2056///   See the comments for EvalAddr for more details.
2057static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) {
2058do {
2059  // We should only be called for evaluating non-pointer expressions, or
2060  // expressions with a pointer type that are not used as references but instead
2061  // are l-values (e.g., DeclRefExpr with a pointer type).
2062
2063  // Our "symbolic interpreter" is just a dispatch off the currently
2064  // viewed AST node.  We then recursively traverse the AST by calling
2065  // EvalAddr and EvalVal appropriately.
2066
2067  E = E->IgnoreParens();
2068  switch (E->getStmtClass()) {
2069  case Stmt::ImplicitCastExprClass: {
2070    ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
2071    if (IE->getValueKind() == VK_LValue) {
2072      E = IE->getSubExpr();
2073      continue;
2074    }
2075    return NULL;
2076  }
2077
2078  case Stmt::DeclRefExprClass: {
2079    // When we hit a DeclRefExpr we are looking at code that refers to a
2080    // variable's name. If it's not a reference variable we check if it has
2081    // local storage within the function, and if so, return the expression.
2082    DeclRefExpr *DR = cast<DeclRefExpr>(E);
2083
2084    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2085      if (V->hasLocalStorage()) {
2086        if (!V->getType()->isReferenceType())
2087          return DR;
2088
2089        // Reference variable, follow through to the expression that
2090        // it points to.
2091        if (V->hasInit()) {
2092          // Add the reference variable to the "trail".
2093          refVars.push_back(DR);
2094          return EvalVal(V->getInit(), refVars);
2095        }
2096      }
2097
2098    return NULL;
2099  }
2100
2101  case Stmt::UnaryOperatorClass: {
2102    // The only unary operator that make sense to handle here
2103    // is Deref.  All others don't resolve to a "name."  This includes
2104    // handling all sorts of rvalues passed to a unary operator.
2105    UnaryOperator *U = cast<UnaryOperator>(E);
2106
2107    if (U->getOpcode() == UO_Deref)
2108      return EvalAddr(U->getSubExpr(), refVars);
2109
2110    return NULL;
2111  }
2112
2113  case Stmt::ArraySubscriptExprClass: {
2114    // Array subscripts are potential references to data on the stack.  We
2115    // retrieve the DeclRefExpr* for the array variable if it indeed
2116    // has local storage.
2117    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars);
2118  }
2119
2120  case Stmt::ConditionalOperatorClass: {
2121    // For conditional operators we need to see if either the LHS or RHS are
2122    // non-NULL Expr's.  If one is non-NULL, we return it.
2123    ConditionalOperator *C = cast<ConditionalOperator>(E);
2124
2125    // Handle the GNU extension for missing LHS.
2126    if (Expr *lhsExpr = C->getLHS())
2127      if (Expr *LHS = EvalVal(lhsExpr, refVars))
2128        return LHS;
2129
2130    return EvalVal(C->getRHS(), refVars);
2131  }
2132
2133  // Accesses to members are potential references to data on the stack.
2134  case Stmt::MemberExprClass: {
2135    MemberExpr *M = cast<MemberExpr>(E);
2136
2137    // Check for indirect access.  We only want direct field accesses.
2138    if (M->isArrow())
2139      return NULL;
2140
2141    // Check whether the member type is itself a reference, in which case
2142    // we're not going to refer to the member, but to what the member refers to.
2143    if (M->getMemberDecl()->getType()->isReferenceType())
2144      return NULL;
2145
2146    return EvalVal(M->getBase(), refVars);
2147  }
2148
2149  default:
2150    // Check that we don't return or take the address of a reference to a
2151    // temporary. This is only useful in C++.
2152    if (!E->isTypeDependent() && E->isRValue())
2153      return E;
2154
2155    // Everything else: we simply don't reason about them.
2156    return NULL;
2157  }
2158} while (true);
2159}
2160
2161//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
2162
2163/// Check for comparisons of floating point operands using != and ==.
2164/// Issue a warning if these are no self-comparisons, as they are not likely
2165/// to do what the programmer intended.
2166void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) {
2167  bool EmitWarning = true;
2168
2169  Expr* LeftExprSansParen = lex->IgnoreParenImpCasts();
2170  Expr* RightExprSansParen = rex->IgnoreParenImpCasts();
2171
2172  // Special case: check for x == x (which is OK).
2173  // Do not emit warnings for such cases.
2174  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
2175    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
2176      if (DRL->getDecl() == DRR->getDecl())
2177        EmitWarning = false;
2178
2179
2180  // Special case: check for comparisons against literals that can be exactly
2181  //  represented by APFloat.  In such cases, do not emit a warning.  This
2182  //  is a heuristic: often comparison against such literals are used to
2183  //  detect if a value in a variable has not changed.  This clearly can
2184  //  lead to false negatives.
2185  if (EmitWarning) {
2186    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
2187      if (FLL->isExact())
2188        EmitWarning = false;
2189    } else
2190      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
2191        if (FLR->isExact())
2192          EmitWarning = false;
2193    }
2194  }
2195
2196  // Check for comparisons with builtin types.
2197  if (EmitWarning)
2198    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
2199      if (CL->isBuiltinCall(Context))
2200        EmitWarning = false;
2201
2202  if (EmitWarning)
2203    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
2204      if (CR->isBuiltinCall(Context))
2205        EmitWarning = false;
2206
2207  // Emit the diagnostic.
2208  if (EmitWarning)
2209    Diag(loc, diag::warn_floatingpoint_eq)
2210      << lex->getSourceRange() << rex->getSourceRange();
2211}
2212
2213//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
2214//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
2215
2216namespace {
2217
2218/// Structure recording the 'active' range of an integer-valued
2219/// expression.
2220struct IntRange {
2221  /// The number of bits active in the int.
2222  unsigned Width;
2223
2224  /// True if the int is known not to have negative values.
2225  bool NonNegative;
2226
2227  IntRange(unsigned Width, bool NonNegative)
2228    : Width(Width), NonNegative(NonNegative)
2229  {}
2230
2231  /// Returns the range of the bool type.
2232  static IntRange forBoolType() {
2233    return IntRange(1, true);
2234  }
2235
2236  /// Returns the range of an opaque value of the given integral type.
2237  static IntRange forValueOfType(ASTContext &C, QualType T) {
2238    return forValueOfCanonicalType(C,
2239                          T->getCanonicalTypeInternal().getTypePtr());
2240  }
2241
2242  /// Returns the range of an opaque value of a canonical integral type.
2243  static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
2244    assert(T->isCanonicalUnqualified());
2245
2246    if (const VectorType *VT = dyn_cast<VectorType>(T))
2247      T = VT->getElementType().getTypePtr();
2248    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2249      T = CT->getElementType().getTypePtr();
2250
2251    // For enum types, use the known bit width of the enumerators.
2252    if (const EnumType *ET = dyn_cast<EnumType>(T)) {
2253      EnumDecl *Enum = ET->getDecl();
2254      if (!Enum->isDefinition())
2255        return IntRange(C.getIntWidth(QualType(T, 0)), false);
2256
2257      unsigned NumPositive = Enum->getNumPositiveBits();
2258      unsigned NumNegative = Enum->getNumNegativeBits();
2259
2260      return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
2261    }
2262
2263    const BuiltinType *BT = cast<BuiltinType>(T);
2264    assert(BT->isInteger());
2265
2266    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2267  }
2268
2269  /// Returns the "target" range of a canonical integral type, i.e.
2270  /// the range of values expressible in the type.
2271  ///
2272  /// This matches forValueOfCanonicalType except that enums have the
2273  /// full range of their type, not the range of their enumerators.
2274  static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
2275    assert(T->isCanonicalUnqualified());
2276
2277    if (const VectorType *VT = dyn_cast<VectorType>(T))
2278      T = VT->getElementType().getTypePtr();
2279    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2280      T = CT->getElementType().getTypePtr();
2281    if (const EnumType *ET = dyn_cast<EnumType>(T))
2282      T = ET->getDecl()->getIntegerType().getTypePtr();
2283
2284    const BuiltinType *BT = cast<BuiltinType>(T);
2285    assert(BT->isInteger());
2286
2287    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2288  }
2289
2290  /// Returns the supremum of two ranges: i.e. their conservative merge.
2291  static IntRange join(IntRange L, IntRange R) {
2292    return IntRange(std::max(L.Width, R.Width),
2293                    L.NonNegative && R.NonNegative);
2294  }
2295
2296  /// Returns the infinum of two ranges: i.e. their aggressive merge.
2297  static IntRange meet(IntRange L, IntRange R) {
2298    return IntRange(std::min(L.Width, R.Width),
2299                    L.NonNegative || R.NonNegative);
2300  }
2301};
2302
2303IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) {
2304  if (value.isSigned() && value.isNegative())
2305    return IntRange(value.getMinSignedBits(), false);
2306
2307  if (value.getBitWidth() > MaxWidth)
2308    value = value.trunc(MaxWidth);
2309
2310  // isNonNegative() just checks the sign bit without considering
2311  // signedness.
2312  return IntRange(value.getActiveBits(), true);
2313}
2314
2315IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
2316                       unsigned MaxWidth) {
2317  if (result.isInt())
2318    return GetValueRange(C, result.getInt(), MaxWidth);
2319
2320  if (result.isVector()) {
2321    IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
2322    for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
2323      IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
2324      R = IntRange::join(R, El);
2325    }
2326    return R;
2327  }
2328
2329  if (result.isComplexInt()) {
2330    IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
2331    IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
2332    return IntRange::join(R, I);
2333  }
2334
2335  // This can happen with lossless casts to intptr_t of "based" lvalues.
2336  // Assume it might use arbitrary bits.
2337  // FIXME: The only reason we need to pass the type in here is to get
2338  // the sign right on this one case.  It would be nice if APValue
2339  // preserved this.
2340  assert(result.isLValue());
2341  return IntRange(MaxWidth, Ty->isUnsignedIntegerType());
2342}
2343
2344/// Pseudo-evaluate the given integer expression, estimating the
2345/// range of values it might take.
2346///
2347/// \param MaxWidth - the width to which the value will be truncated
2348IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
2349  E = E->IgnoreParens();
2350
2351  // Try a full evaluation first.
2352  Expr::EvalResult result;
2353  if (E->Evaluate(result, C))
2354    return GetValueRange(C, result.Val, E->getType(), MaxWidth);
2355
2356  // I think we only want to look through implicit casts here; if the
2357  // user has an explicit widening cast, we should treat the value as
2358  // being of the new, wider type.
2359  if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
2360    if (CE->getCastKind() == CK_NoOp)
2361      return GetExprRange(C, CE->getSubExpr(), MaxWidth);
2362
2363    IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
2364
2365    bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
2366
2367    // Assume that non-integer casts can span the full range of the type.
2368    if (!isIntegerCast)
2369      return OutputTypeRange;
2370
2371    IntRange SubRange
2372      = GetExprRange(C, CE->getSubExpr(),
2373                     std::min(MaxWidth, OutputTypeRange.Width));
2374
2375    // Bail out if the subexpr's range is as wide as the cast type.
2376    if (SubRange.Width >= OutputTypeRange.Width)
2377      return OutputTypeRange;
2378
2379    // Otherwise, we take the smaller width, and we're non-negative if
2380    // either the output type or the subexpr is.
2381    return IntRange(SubRange.Width,
2382                    SubRange.NonNegative || OutputTypeRange.NonNegative);
2383  }
2384
2385  if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
2386    // If we can fold the condition, just take that operand.
2387    bool CondResult;
2388    if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
2389      return GetExprRange(C, CondResult ? CO->getTrueExpr()
2390                                        : CO->getFalseExpr(),
2391                          MaxWidth);
2392
2393    // Otherwise, conservatively merge.
2394    IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
2395    IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
2396    return IntRange::join(L, R);
2397  }
2398
2399  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
2400    switch (BO->getOpcode()) {
2401
2402    // Boolean-valued operations are single-bit and positive.
2403    case BO_LAnd:
2404    case BO_LOr:
2405    case BO_LT:
2406    case BO_GT:
2407    case BO_LE:
2408    case BO_GE:
2409    case BO_EQ:
2410    case BO_NE:
2411      return IntRange::forBoolType();
2412
2413    // The type of these compound assignments is the type of the LHS,
2414    // so the RHS is not necessarily an integer.
2415    case BO_MulAssign:
2416    case BO_DivAssign:
2417    case BO_RemAssign:
2418    case BO_AddAssign:
2419    case BO_SubAssign:
2420      return IntRange::forValueOfType(C, E->getType());
2421
2422    // Operations with opaque sources are black-listed.
2423    case BO_PtrMemD:
2424    case BO_PtrMemI:
2425      return IntRange::forValueOfType(C, E->getType());
2426
2427    // Bitwise-and uses the *infinum* of the two source ranges.
2428    case BO_And:
2429    case BO_AndAssign:
2430      return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
2431                            GetExprRange(C, BO->getRHS(), MaxWidth));
2432
2433    // Left shift gets black-listed based on a judgement call.
2434    case BO_Shl:
2435      // ...except that we want to treat '1 << (blah)' as logically
2436      // positive.  It's an important idiom.
2437      if (IntegerLiteral *I
2438            = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
2439        if (I->getValue() == 1) {
2440          IntRange R = IntRange::forValueOfType(C, E->getType());
2441          return IntRange(R.Width, /*NonNegative*/ true);
2442        }
2443      }
2444      // fallthrough
2445
2446    case BO_ShlAssign:
2447      return IntRange::forValueOfType(C, E->getType());
2448
2449    // Right shift by a constant can narrow its left argument.
2450    case BO_Shr:
2451    case BO_ShrAssign: {
2452      IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2453
2454      // If the shift amount is a positive constant, drop the width by
2455      // that much.
2456      llvm::APSInt shift;
2457      if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
2458          shift.isNonNegative()) {
2459        unsigned zext = shift.getZExtValue();
2460        if (zext >= L.Width)
2461          L.Width = (L.NonNegative ? 0 : 1);
2462        else
2463          L.Width -= zext;
2464      }
2465
2466      return L;
2467    }
2468
2469    // Comma acts as its right operand.
2470    case BO_Comma:
2471      return GetExprRange(C, BO->getRHS(), MaxWidth);
2472
2473    // Black-list pointer subtractions.
2474    case BO_Sub:
2475      if (BO->getLHS()->getType()->isPointerType())
2476        return IntRange::forValueOfType(C, E->getType());
2477      // fallthrough
2478
2479    default:
2480      break;
2481    }
2482
2483    // Treat every other operator as if it were closed on the
2484    // narrowest type that encompasses both operands.
2485    IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2486    IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
2487    return IntRange::join(L, R);
2488  }
2489
2490  if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
2491    switch (UO->getOpcode()) {
2492    // Boolean-valued operations are white-listed.
2493    case UO_LNot:
2494      return IntRange::forBoolType();
2495
2496    // Operations with opaque sources are black-listed.
2497    case UO_Deref:
2498    case UO_AddrOf: // should be impossible
2499      return IntRange::forValueOfType(C, E->getType());
2500
2501    default:
2502      return GetExprRange(C, UO->getSubExpr(), MaxWidth);
2503    }
2504  }
2505
2506  if (dyn_cast<OffsetOfExpr>(E)) {
2507    IntRange::forValueOfType(C, E->getType());
2508  }
2509
2510  FieldDecl *BitField = E->getBitField();
2511  if (BitField) {
2512    llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C);
2513    unsigned BitWidth = BitWidthAP.getZExtValue();
2514
2515    return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType());
2516  }
2517
2518  return IntRange::forValueOfType(C, E->getType());
2519}
2520
2521IntRange GetExprRange(ASTContext &C, Expr *E) {
2522  return GetExprRange(C, E, C.getIntWidth(E->getType()));
2523}
2524
2525/// Checks whether the given value, which currently has the given
2526/// source semantics, has the same value when coerced through the
2527/// target semantics.
2528bool IsSameFloatAfterCast(const llvm::APFloat &value,
2529                          const llvm::fltSemantics &Src,
2530                          const llvm::fltSemantics &Tgt) {
2531  llvm::APFloat truncated = value;
2532
2533  bool ignored;
2534  truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
2535  truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
2536
2537  return truncated.bitwiseIsEqual(value);
2538}
2539
2540/// Checks whether the given value, which currently has the given
2541/// source semantics, has the same value when coerced through the
2542/// target semantics.
2543///
2544/// The value might be a vector of floats (or a complex number).
2545bool IsSameFloatAfterCast(const APValue &value,
2546                          const llvm::fltSemantics &Src,
2547                          const llvm::fltSemantics &Tgt) {
2548  if (value.isFloat())
2549    return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
2550
2551  if (value.isVector()) {
2552    for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
2553      if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
2554        return false;
2555    return true;
2556  }
2557
2558  assert(value.isComplexFloat());
2559  return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
2560          IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
2561}
2562
2563void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
2564
2565static bool IsZero(Sema &S, Expr *E) {
2566  // Suppress cases where we are comparing against an enum constant.
2567  if (const DeclRefExpr *DR =
2568      dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
2569    if (isa<EnumConstantDecl>(DR->getDecl()))
2570      return false;
2571
2572  // Suppress cases where the '0' value is expanded from a macro.
2573  if (E->getLocStart().isMacroID())
2574    return false;
2575
2576  llvm::APSInt Value;
2577  return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
2578}
2579
2580static bool HasEnumType(Expr *E) {
2581  // Strip off implicit integral promotions.
2582  while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
2583    if (ICE->getCastKind() != CK_IntegralCast &&
2584        ICE->getCastKind() != CK_NoOp)
2585      break;
2586    E = ICE->getSubExpr();
2587  }
2588
2589  return E->getType()->isEnumeralType();
2590}
2591
2592void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
2593  BinaryOperatorKind op = E->getOpcode();
2594  if (E->isValueDependent())
2595    return;
2596
2597  if (op == BO_LT && IsZero(S, E->getRHS())) {
2598    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2599      << "< 0" << "false" << HasEnumType(E->getLHS())
2600      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2601  } else if (op == BO_GE && IsZero(S, E->getRHS())) {
2602    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2603      << ">= 0" << "true" << HasEnumType(E->getLHS())
2604      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2605  } else if (op == BO_GT && IsZero(S, E->getLHS())) {
2606    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2607      << "0 >" << "false" << HasEnumType(E->getRHS())
2608      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2609  } else if (op == BO_LE && IsZero(S, E->getLHS())) {
2610    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2611      << "0 <=" << "true" << HasEnumType(E->getRHS())
2612      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2613  }
2614}
2615
2616/// Analyze the operands of the given comparison.  Implements the
2617/// fallback case from AnalyzeComparison.
2618void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
2619  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
2620  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
2621}
2622
2623/// \brief Implements -Wsign-compare.
2624///
2625/// \param lex the left-hand expression
2626/// \param rex the right-hand expression
2627/// \param OpLoc the location of the joining operator
2628/// \param BinOpc binary opcode or 0
2629void AnalyzeComparison(Sema &S, BinaryOperator *E) {
2630  // The type the comparison is being performed in.
2631  QualType T = E->getLHS()->getType();
2632  assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
2633         && "comparison with mismatched types");
2634
2635  // We don't do anything special if this isn't an unsigned integral
2636  // comparison:  we're only interested in integral comparisons, and
2637  // signed comparisons only happen in cases we don't care to warn about.
2638  //
2639  // We also don't care about value-dependent expressions or expressions
2640  // whose result is a constant.
2641  if (!T->hasUnsignedIntegerRepresentation()
2642      || E->isValueDependent() || E->isIntegerConstantExpr(S.Context))
2643    return AnalyzeImpConvsInComparison(S, E);
2644
2645  Expr *lex = E->getLHS()->IgnoreParenImpCasts();
2646  Expr *rex = E->getRHS()->IgnoreParenImpCasts();
2647
2648  // Check to see if one of the (unmodified) operands is of different
2649  // signedness.
2650  Expr *signedOperand, *unsignedOperand;
2651  if (lex->getType()->hasSignedIntegerRepresentation()) {
2652    assert(!rex->getType()->hasSignedIntegerRepresentation() &&
2653           "unsigned comparison between two signed integer expressions?");
2654    signedOperand = lex;
2655    unsignedOperand = rex;
2656  } else if (rex->getType()->hasSignedIntegerRepresentation()) {
2657    signedOperand = rex;
2658    unsignedOperand = lex;
2659  } else {
2660    CheckTrivialUnsignedComparison(S, E);
2661    return AnalyzeImpConvsInComparison(S, E);
2662  }
2663
2664  // Otherwise, calculate the effective range of the signed operand.
2665  IntRange signedRange = GetExprRange(S.Context, signedOperand);
2666
2667  // Go ahead and analyze implicit conversions in the operands.  Note
2668  // that we skip the implicit conversions on both sides.
2669  AnalyzeImplicitConversions(S, lex, E->getOperatorLoc());
2670  AnalyzeImplicitConversions(S, rex, E->getOperatorLoc());
2671
2672  // If the signed range is non-negative, -Wsign-compare won't fire,
2673  // but we should still check for comparisons which are always true
2674  // or false.
2675  if (signedRange.NonNegative)
2676    return CheckTrivialUnsignedComparison(S, E);
2677
2678  // For (in)equality comparisons, if the unsigned operand is a
2679  // constant which cannot collide with a overflowed signed operand,
2680  // then reinterpreting the signed operand as unsigned will not
2681  // change the result of the comparison.
2682  if (E->isEqualityOp()) {
2683    unsigned comparisonWidth = S.Context.getIntWidth(T);
2684    IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
2685
2686    // We should never be unable to prove that the unsigned operand is
2687    // non-negative.
2688    assert(unsignedRange.NonNegative && "unsigned range includes negative?");
2689
2690    if (unsignedRange.Width < comparisonWidth)
2691      return;
2692  }
2693
2694  S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison)
2695    << lex->getType() << rex->getType()
2696    << lex->getSourceRange() << rex->getSourceRange();
2697}
2698
2699/// Analyzes an attempt to assign the given value to a bitfield.
2700///
2701/// Returns true if there was something fishy about the attempt.
2702bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
2703                               SourceLocation InitLoc) {
2704  assert(Bitfield->isBitField());
2705  if (Bitfield->isInvalidDecl())
2706    return false;
2707
2708  // White-list bool bitfields.
2709  if (Bitfield->getType()->isBooleanType())
2710    return false;
2711
2712  // Ignore value- or type-dependent expressions.
2713  if (Bitfield->getBitWidth()->isValueDependent() ||
2714      Bitfield->getBitWidth()->isTypeDependent() ||
2715      Init->isValueDependent() ||
2716      Init->isTypeDependent())
2717    return false;
2718
2719  Expr *OriginalInit = Init->IgnoreParenImpCasts();
2720
2721  llvm::APSInt Width(32);
2722  Expr::EvalResult InitValue;
2723  if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) ||
2724      !OriginalInit->Evaluate(InitValue, S.Context) ||
2725      !InitValue.Val.isInt())
2726    return false;
2727
2728  const llvm::APSInt &Value = InitValue.Val.getInt();
2729  unsigned OriginalWidth = Value.getBitWidth();
2730  unsigned FieldWidth = Width.getZExtValue();
2731
2732  if (OriginalWidth <= FieldWidth)
2733    return false;
2734
2735  llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
2736
2737  // It's fairly common to write values into signed bitfields
2738  // that, if sign-extended, would end up becoming a different
2739  // value.  We don't want to warn about that.
2740  if (Value.isSigned() && Value.isNegative())
2741    TruncatedValue = TruncatedValue.sext(OriginalWidth);
2742  else
2743    TruncatedValue = TruncatedValue.zext(OriginalWidth);
2744
2745  if (Value == TruncatedValue)
2746    return false;
2747
2748  std::string PrettyValue = Value.toString(10);
2749  std::string PrettyTrunc = TruncatedValue.toString(10);
2750
2751  S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
2752    << PrettyValue << PrettyTrunc << OriginalInit->getType()
2753    << Init->getSourceRange();
2754
2755  return true;
2756}
2757
2758/// Analyze the given simple or compound assignment for warning-worthy
2759/// operations.
2760void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
2761  // Just recurse on the LHS.
2762  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
2763
2764  // We want to recurse on the RHS as normal unless we're assigning to
2765  // a bitfield.
2766  if (FieldDecl *Bitfield = E->getLHS()->getBitField()) {
2767    if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
2768                                  E->getOperatorLoc())) {
2769      // Recurse, ignoring any implicit conversions on the RHS.
2770      return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
2771                                        E->getOperatorLoc());
2772    }
2773  }
2774
2775  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
2776}
2777
2778/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
2779void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
2780                     SourceLocation CContext, unsigned diag) {
2781  S.Diag(E->getExprLoc(), diag)
2782    << SourceType << T << E->getSourceRange() << SourceRange(CContext);
2783}
2784
2785/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
2786void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext,
2787                     unsigned diag) {
2788  DiagnoseImpCast(S, E, E->getType(), T, CContext, diag);
2789}
2790
2791/// Diagnose an implicit cast from a literal expression. Also attemps to supply
2792/// fixit hints when the cast wouldn't lose information to simply write the
2793/// expression with the expected type.
2794void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
2795                                    SourceLocation CContext) {
2796  // Emit the primary warning first, then try to emit a fixit hint note if
2797  // reasonable.
2798  S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
2799    << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext);
2800
2801  const llvm::APFloat &Value = FL->getValue();
2802
2803  // Don't attempt to fix PPC double double literals.
2804  if (&Value.getSemantics() == &llvm::APFloat::PPCDoubleDouble)
2805    return;
2806
2807  // Try to convert this exactly to an 64-bit integer. FIXME: It would be
2808  // nice to support arbitrarily large integers here.
2809  bool isExact = false;
2810  uint64_t IntegerPart;
2811  if (Value.convertToInteger(&IntegerPart, 64, /*isSigned=*/true,
2812                             llvm::APFloat::rmTowardZero, &isExact)
2813      != llvm::APFloat::opOK || !isExact)
2814    return;
2815
2816  llvm::APInt IntegerValue(64, IntegerPart, /*isSigned=*/true);
2817
2818  std::string LiteralValue = IntegerValue.toString(10, /*isSigned=*/true);
2819  S.Diag(FL->getExprLoc(), diag::note_fix_integral_float_as_integer)
2820    << FixItHint::CreateReplacement(FL->getSourceRange(), LiteralValue);
2821}
2822
2823std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
2824  if (!Range.Width) return "0";
2825
2826  llvm::APSInt ValueInRange = Value;
2827  ValueInRange.setIsSigned(!Range.NonNegative);
2828  ValueInRange = ValueInRange.trunc(Range.Width);
2829  return ValueInRange.toString(10);
2830}
2831
2832static bool isFromSystemMacro(Sema &S, SourceLocation loc) {
2833  SourceManager &smgr = S.Context.getSourceManager();
2834  return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc));
2835}
2836
2837void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
2838                             SourceLocation CC, bool *ICContext = 0) {
2839  if (E->isTypeDependent() || E->isValueDependent()) return;
2840
2841  const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
2842  const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
2843  if (Source == Target) return;
2844  if (Target->isDependentType()) return;
2845
2846  // If the conversion context location is invalid don't complain.
2847  // We also don't want to emit a warning if the issue occurs from the
2848  // instantiation of a system macro.  The problem is that 'getSpellingLoc()'
2849  // is slow, so we delay this check as long as possible.  Once we detect
2850  // we are in that scenario, we just return.
2851  if (CC.isInvalid())
2852    return;
2853
2854  // Never diagnose implicit casts to bool.
2855  if (Target->isSpecificBuiltinType(BuiltinType::Bool))
2856    return;
2857
2858  // Strip vector types.
2859  if (isa<VectorType>(Source)) {
2860    if (!isa<VectorType>(Target)) {
2861      if (isFromSystemMacro(S, CC))
2862        return;
2863      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
2864    }
2865
2866    Source = cast<VectorType>(Source)->getElementType().getTypePtr();
2867    Target = cast<VectorType>(Target)->getElementType().getTypePtr();
2868  }
2869
2870  // Strip complex types.
2871  if (isa<ComplexType>(Source)) {
2872    if (!isa<ComplexType>(Target)) {
2873      if (isFromSystemMacro(S, CC))
2874        return;
2875
2876      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
2877    }
2878
2879    Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
2880    Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
2881  }
2882
2883  const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
2884  const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
2885
2886  // If the source is floating point...
2887  if (SourceBT && SourceBT->isFloatingPoint()) {
2888    // ...and the target is floating point...
2889    if (TargetBT && TargetBT->isFloatingPoint()) {
2890      // ...then warn if we're dropping FP rank.
2891
2892      // Builtin FP kinds are ordered by increasing FP rank.
2893      if (SourceBT->getKind() > TargetBT->getKind()) {
2894        // Don't warn about float constants that are precisely
2895        // representable in the target type.
2896        Expr::EvalResult result;
2897        if (E->Evaluate(result, S.Context)) {
2898          // Value might be a float, a float vector, or a float complex.
2899          if (IsSameFloatAfterCast(result.Val,
2900                   S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
2901                   S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
2902            return;
2903        }
2904
2905        if (isFromSystemMacro(S, CC))
2906          return;
2907
2908        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
2909      }
2910      return;
2911    }
2912
2913    // If the target is integral, always warn.
2914    if ((TargetBT && TargetBT->isInteger())) {
2915      if (isFromSystemMacro(S, CC))
2916        return;
2917
2918      Expr *InnerE = E->IgnoreParenImpCasts();
2919      if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
2920        DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
2921      } else {
2922        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
2923      }
2924    }
2925
2926    return;
2927  }
2928
2929  if (!Source->isIntegerType() || !Target->isIntegerType())
2930    return;
2931
2932  IntRange SourceRange = GetExprRange(S.Context, E);
2933  IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
2934
2935  if (SourceRange.Width > TargetRange.Width) {
2936    // If the source is a constant, use a default-on diagnostic.
2937    // TODO: this should happen for bitfield stores, too.
2938    llvm::APSInt Value(32);
2939    if (E->isIntegerConstantExpr(Value, S.Context)) {
2940      if (isFromSystemMacro(S, CC))
2941        return;
2942
2943      std::string PrettySourceValue = Value.toString(10);
2944      std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
2945
2946      S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant)
2947        << PrettySourceValue << PrettyTargetValue
2948        << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC);
2949      return;
2950    }
2951
2952    // People want to build with -Wshorten-64-to-32 and not -Wconversion
2953    // and by god we'll let them.
2954
2955    if (isFromSystemMacro(S, CC))
2956      return;
2957
2958    if (SourceRange.Width == 64 && TargetRange.Width == 32)
2959      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32);
2960    return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
2961  }
2962
2963  if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
2964      (!TargetRange.NonNegative && SourceRange.NonNegative &&
2965       SourceRange.Width == TargetRange.Width)) {
2966
2967    if (isFromSystemMacro(S, CC))
2968      return;
2969
2970    unsigned DiagID = diag::warn_impcast_integer_sign;
2971
2972    // Traditionally, gcc has warned about this under -Wsign-compare.
2973    // We also want to warn about it in -Wconversion.
2974    // So if -Wconversion is off, use a completely identical diagnostic
2975    // in the sign-compare group.
2976    // The conditional-checking code will
2977    if (ICContext) {
2978      DiagID = diag::warn_impcast_integer_sign_conditional;
2979      *ICContext = true;
2980    }
2981
2982    return DiagnoseImpCast(S, E, T, CC, DiagID);
2983  }
2984
2985  // Diagnose conversions between different enumeration types.
2986  // In C, we pretend that the type of an EnumConstantDecl is its enumeration
2987  // type, to give us better diagnostics.
2988  QualType SourceType = E->getType();
2989  if (!S.getLangOptions().CPlusPlus) {
2990    if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
2991      if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
2992        EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
2993        SourceType = S.Context.getTypeDeclType(Enum);
2994        Source = S.Context.getCanonicalType(SourceType).getTypePtr();
2995      }
2996  }
2997
2998  if (const EnumType *SourceEnum = Source->getAs<EnumType>())
2999    if (const EnumType *TargetEnum = Target->getAs<EnumType>())
3000      if ((SourceEnum->getDecl()->getIdentifier() ||
3001           SourceEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3002          (TargetEnum->getDecl()->getIdentifier() ||
3003           TargetEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3004          SourceEnum != TargetEnum) {
3005        if (isFromSystemMacro(S, CC))
3006          return;
3007
3008        return DiagnoseImpCast(S, E, SourceType, T, CC,
3009                               diag::warn_impcast_different_enum_types);
3010      }
3011
3012  return;
3013}
3014
3015void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T);
3016
3017void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
3018                             SourceLocation CC, bool &ICContext) {
3019  E = E->IgnoreParenImpCasts();
3020
3021  if (isa<ConditionalOperator>(E))
3022    return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T);
3023
3024  AnalyzeImplicitConversions(S, E, CC);
3025  if (E->getType() != T)
3026    return CheckImplicitConversion(S, E, T, CC, &ICContext);
3027  return;
3028}
3029
3030void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) {
3031  SourceLocation CC = E->getQuestionLoc();
3032
3033  AnalyzeImplicitConversions(S, E->getCond(), CC);
3034
3035  bool Suspicious = false;
3036  CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
3037  CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
3038
3039  // If -Wconversion would have warned about either of the candidates
3040  // for a signedness conversion to the context type...
3041  if (!Suspicious) return;
3042
3043  // ...but it's currently ignored...
3044  if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
3045                                 CC))
3046    return;
3047
3048  // ...and -Wsign-compare isn't...
3049  if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional, CC))
3050    return;
3051
3052  // ...then check whether it would have warned about either of the
3053  // candidates for a signedness conversion to the condition type.
3054  if (E->getType() != T) {
3055    Suspicious = false;
3056    CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
3057                            E->getType(), CC, &Suspicious);
3058    if (!Suspicious)
3059      CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
3060                              E->getType(), CC, &Suspicious);
3061    if (!Suspicious)
3062      return;
3063  }
3064
3065  // If so, emit a diagnostic under -Wsign-compare.
3066  Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts();
3067  Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts();
3068  S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional)
3069    << lex->getType() << rex->getType()
3070    << lex->getSourceRange() << rex->getSourceRange();
3071}
3072
3073/// AnalyzeImplicitConversions - Find and report any interesting
3074/// implicit conversions in the given expression.  There are a couple
3075/// of competing diagnostics here, -Wconversion and -Wsign-compare.
3076void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
3077  QualType T = OrigE->getType();
3078  Expr *E = OrigE->IgnoreParenImpCasts();
3079
3080  // For conditional operators, we analyze the arguments as if they
3081  // were being fed directly into the output.
3082  if (isa<ConditionalOperator>(E)) {
3083    ConditionalOperator *CO = cast<ConditionalOperator>(E);
3084    CheckConditionalOperator(S, CO, T);
3085    return;
3086  }
3087
3088  // Go ahead and check any implicit conversions we might have skipped.
3089  // The non-canonical typecheck is just an optimization;
3090  // CheckImplicitConversion will filter out dead implicit conversions.
3091  if (E->getType() != T)
3092    CheckImplicitConversion(S, E, T, CC);
3093
3094  // Now continue drilling into this expression.
3095
3096  // Skip past explicit casts.
3097  if (isa<ExplicitCastExpr>(E)) {
3098    E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
3099    return AnalyzeImplicitConversions(S, E, CC);
3100  }
3101
3102  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
3103    // Do a somewhat different check with comparison operators.
3104    if (BO->isComparisonOp())
3105      return AnalyzeComparison(S, BO);
3106
3107    // And with assignments and compound assignments.
3108    if (BO->isAssignmentOp())
3109      return AnalyzeAssignment(S, BO);
3110  }
3111
3112  // These break the otherwise-useful invariant below.  Fortunately,
3113  // we don't really need to recurse into them, because any internal
3114  // expressions should have been analyzed already when they were
3115  // built into statements.
3116  if (isa<StmtExpr>(E)) return;
3117
3118  // Don't descend into unevaluated contexts.
3119  if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
3120
3121  // Now just recurse over the expression's children.
3122  CC = E->getExprLoc();
3123  for (Stmt::child_range I = E->children(); I; ++I)
3124    AnalyzeImplicitConversions(S, cast<Expr>(*I), CC);
3125}
3126
3127} // end anonymous namespace
3128
3129/// Diagnoses "dangerous" implicit conversions within the given
3130/// expression (which is a full expression).  Implements -Wconversion
3131/// and -Wsign-compare.
3132///
3133/// \param CC the "context" location of the implicit conversion, i.e.
3134///   the most location of the syntactic entity requiring the implicit
3135///   conversion
3136void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
3137  // Don't diagnose in unevaluated contexts.
3138  if (ExprEvalContexts.back().Context == Sema::Unevaluated)
3139    return;
3140
3141  // Don't diagnose for value- or type-dependent expressions.
3142  if (E->isTypeDependent() || E->isValueDependent())
3143    return;
3144
3145  // This is not the right CC for (e.g.) a variable initialization.
3146  AnalyzeImplicitConversions(*this, E, CC);
3147}
3148
3149void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
3150                                       FieldDecl *BitField,
3151                                       Expr *Init) {
3152  (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
3153}
3154
3155/// CheckParmsForFunctionDef - Check that the parameters of the given
3156/// function are appropriate for the definition of a function. This
3157/// takes care of any checks that cannot be performed on the
3158/// declaration itself, e.g., that the types of each of the function
3159/// parameters are complete.
3160bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
3161                                    bool CheckParameterNames) {
3162  bool HasInvalidParm = false;
3163  for (; P != PEnd; ++P) {
3164    ParmVarDecl *Param = *P;
3165
3166    // C99 6.7.5.3p4: the parameters in a parameter type list in a
3167    // function declarator that is part of a function definition of
3168    // that function shall not have incomplete type.
3169    //
3170    // This is also C++ [dcl.fct]p6.
3171    if (!Param->isInvalidDecl() &&
3172        RequireCompleteType(Param->getLocation(), Param->getType(),
3173                               diag::err_typecheck_decl_incomplete_type)) {
3174      Param->setInvalidDecl();
3175      HasInvalidParm = true;
3176    }
3177
3178    // C99 6.9.1p5: If the declarator includes a parameter type list, the
3179    // declaration of each parameter shall include an identifier.
3180    if (CheckParameterNames &&
3181        Param->getIdentifier() == 0 &&
3182        !Param->isImplicit() &&
3183        !getLangOptions().CPlusPlus)
3184      Diag(Param->getLocation(), diag::err_parameter_name_omitted);
3185
3186    // C99 6.7.5.3p12:
3187    //   If the function declarator is not part of a definition of that
3188    //   function, parameters may have incomplete type and may use the [*]
3189    //   notation in their sequences of declarator specifiers to specify
3190    //   variable length array types.
3191    QualType PType = Param->getOriginalType();
3192    if (const ArrayType *AT = Context.getAsArrayType(PType)) {
3193      if (AT->getSizeModifier() == ArrayType::Star) {
3194        // FIXME: This diagnosic should point the the '[*]' if source-location
3195        // information is added for it.
3196        Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
3197      }
3198    }
3199  }
3200
3201  return HasInvalidParm;
3202}
3203
3204/// CheckCastAlign - Implements -Wcast-align, which warns when a
3205/// pointer cast increases the alignment requirements.
3206void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
3207  // This is actually a lot of work to potentially be doing on every
3208  // cast; don't do it if we're ignoring -Wcast_align (as is the default).
3209  if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
3210                                          TRange.getBegin())
3211        == Diagnostic::Ignored)
3212    return;
3213
3214  // Ignore dependent types.
3215  if (T->isDependentType() || Op->getType()->isDependentType())
3216    return;
3217
3218  // Require that the destination be a pointer type.
3219  const PointerType *DestPtr = T->getAs<PointerType>();
3220  if (!DestPtr) return;
3221
3222  // If the destination has alignment 1, we're done.
3223  QualType DestPointee = DestPtr->getPointeeType();
3224  if (DestPointee->isIncompleteType()) return;
3225  CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
3226  if (DestAlign.isOne()) return;
3227
3228  // Require that the source be a pointer type.
3229  const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
3230  if (!SrcPtr) return;
3231  QualType SrcPointee = SrcPtr->getPointeeType();
3232
3233  // Whitelist casts from cv void*.  We already implicitly
3234  // whitelisted casts to cv void*, since they have alignment 1.
3235  // Also whitelist casts involving incomplete types, which implicitly
3236  // includes 'void'.
3237  if (SrcPointee->isIncompleteType()) return;
3238
3239  CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
3240  if (SrcAlign >= DestAlign) return;
3241
3242  Diag(TRange.getBegin(), diag::warn_cast_align)
3243    << Op->getType() << T
3244    << static_cast<unsigned>(SrcAlign.getQuantity())
3245    << static_cast<unsigned>(DestAlign.getQuantity())
3246    << TRange << Op->getSourceRange();
3247}
3248
3249static void CheckArrayAccess_Check(Sema &S,
3250                                   const clang::ArraySubscriptExpr *E) {
3251  const Expr *BaseExpr = E->getBase()->IgnoreParenImpCasts();
3252  const ConstantArrayType *ArrayTy =
3253    S.Context.getAsConstantArrayType(BaseExpr->getType());
3254  if (!ArrayTy)
3255    return;
3256
3257  const Expr *IndexExpr = E->getIdx();
3258  if (IndexExpr->isValueDependent())
3259    return;
3260  llvm::APSInt index;
3261  if (!IndexExpr->isIntegerConstantExpr(index, S.Context))
3262    return;
3263
3264  if (index.isUnsigned() || !index.isNegative()) {
3265    llvm::APInt size = ArrayTy->getSize();
3266    if (!size.isStrictlyPositive())
3267      return;
3268    if (size.getBitWidth() > index.getBitWidth())
3269      index = index.sext(size.getBitWidth());
3270    else if (size.getBitWidth() < index.getBitWidth())
3271      size = size.sext(index.getBitWidth());
3272
3273    if (index.slt(size))
3274      return;
3275
3276    S.DiagRuntimeBehavior(E->getBase()->getLocStart(), BaseExpr,
3277                          S.PDiag(diag::warn_array_index_exceeds_bounds)
3278                            << index.toString(10, true)
3279                            << size.toString(10, true)
3280                            << IndexExpr->getSourceRange());
3281  } else {
3282    S.DiagRuntimeBehavior(E->getBase()->getLocStart(), BaseExpr,
3283                          S.PDiag(diag::warn_array_index_precedes_bounds)
3284                            << index.toString(10, true)
3285                            << IndexExpr->getSourceRange());
3286  }
3287
3288  const NamedDecl *ND = NULL;
3289  if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
3290    ND = dyn_cast<NamedDecl>(DRE->getDecl());
3291  if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
3292    ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
3293  if (ND)
3294    S.DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
3295                          S.PDiag(diag::note_array_index_out_of_bounds)
3296                            << ND->getDeclName());
3297}
3298
3299void Sema::CheckArrayAccess(const Expr *expr) {
3300  while (true) {
3301    expr = expr->IgnoreParens();
3302    switch (expr->getStmtClass()) {
3303      case Stmt::ArraySubscriptExprClass:
3304        CheckArrayAccess_Check(*this, cast<ArraySubscriptExpr>(expr));
3305        return;
3306      case Stmt::ConditionalOperatorClass: {
3307        const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
3308        if (const Expr *lhs = cond->getLHS())
3309          CheckArrayAccess(lhs);
3310        if (const Expr *rhs = cond->getRHS())
3311          CheckArrayAccess(rhs);
3312        return;
3313      }
3314      default:
3315        return;
3316    }
3317  }
3318}
3319